From bb206caff85c6a34b13c983f249183f69234c0fa Mon Sep 17 00:00:00 2001
From: jinyaoguo <384606266@qq.com>
Date: Sat, 15 Nov 2025 01:14:43 -0500
Subject: [PATCH] Add OOB seed files

---
 contrib/largeNbDicts/largeNbDicts.c           | 1085 +++
 contrib/match_finders/zstd_edist.c            |  558 ++
 contrib/pzstd/Pzstd.cpp                       |  626 ++
 contrib/seekable_format/zstdseek_decompress.c |  600 ++
 doc/educational_decoder/zstd_decompress.c     | 2323 +++++
 lib/common/fse.h                              |  625 ++
 lib/common/threading.c                        |  182 +
 lib/compress/fse_compress.c                   |  625 ++
 lib/compress/hist.c                           |  191 +
 lib/compress/huf_compress.c                   | 1465 +++
 lib/compress/zstd_compress.c                  | 7851 +++++++++++++++++
 lib/compress/zstd_compress_sequences.c        |  442 +
 lib/compress/zstd_double_fast.c               |  778 ++
 lib/compress/zstd_lazy.c                      | 2199 +++++
 lib/compress/zstd_ldm.c                       |  745 ++
 lib/compress/zstd_opt.c                       | 1580 ++++
 lib/compress/zstdmt_compress.c                | 1923 ++++
 lib/decompress/huf_decompress.c               | 1945 ++++
 lib/decompress/zstd_decompress.c              | 2410 +++++
 lib/decompress/zstd_decompress_block.c        | 2217 +++++
 lib/deprecated/zbuff_compress.c               |  167 +
 lib/dictBuilder/divsufsort.c                  | 1913 ++++
 lib/dictBuilder/fastcover.c                   |  765 ++
 lib/dictBuilder/zdict.c                       | 1137 +++
 lib/legacy/zstd_v02.c                         | 3465 ++++++++
 lib/legacy/zstd_v03.c                         | 3105 +++++++
 lib/legacy/zstd_v04.c                         | 3598 ++++++++
 lib/legacy/zstd_v05.c                         | 4005 +++++++++
 lib/legacy/zstd_v06.c                         | 4110 +++++++++
 lib/legacy/zstd_v07.c                         | 4490 ++++++++++
 programs/benchzstd.c                          | 1270 +++
 programs/dibio.c                              |  440 +
 programs/fileio.c                             | 3473 ++++++++
 programs/fileio_asyncio.c                     |  663 ++
 programs/lorem.c                              |  285 +
 programs/util.c                               | 1730 ++++
 programs/zstdcli.c                            | 1668 ++++
 37 files changed, 66654 insertions(+)
 create mode 100644 contrib/largeNbDicts/largeNbDicts.c
 create mode 100644 contrib/match_finders/zstd_edist.c
 create mode 100644 contrib/pzstd/Pzstd.cpp
 create mode 100644 contrib/seekable_format/zstdseek_decompress.c
 create mode 100644 doc/educational_decoder/zstd_decompress.c
 create mode 100644 lib/common/fse.h
 create mode 100644 lib/common/threading.c
 create mode 100644 lib/compress/fse_compress.c
 create mode 100644 lib/compress/hist.c
 create mode 100644 lib/compress/huf_compress.c
 create mode 100644 lib/compress/zstd_compress.c
 create mode 100644 lib/compress/zstd_compress_sequences.c
 create mode 100644 lib/compress/zstd_double_fast.c
 create mode 100644 lib/compress/zstd_lazy.c
 create mode 100644 lib/compress/zstd_ldm.c
 create mode 100644 lib/compress/zstd_opt.c
 create mode 100644 lib/compress/zstdmt_compress.c
 create mode 100644 lib/decompress/huf_decompress.c
 create mode 100644 lib/decompress/zstd_decompress.c
 create mode 100644 lib/decompress/zstd_decompress_block.c
 create mode 100644 lib/deprecated/zbuff_compress.c
 create mode 100644 lib/dictBuilder/divsufsort.c
 create mode 100644 lib/dictBuilder/fastcover.c
 create mode 100644 lib/dictBuilder/zdict.c
 create mode 100644 lib/legacy/zstd_v02.c
 create mode 100644 lib/legacy/zstd_v03.c
 create mode 100644 lib/legacy/zstd_v04.c
 create mode 100644 lib/legacy/zstd_v05.c
 create mode 100644 lib/legacy/zstd_v06.c
 create mode 100644 lib/legacy/zstd_v07.c
 create mode 100644 programs/benchzstd.c
 create mode 100644 programs/dibio.c
 create mode 100644 programs/fileio.c
 create mode 100644 programs/fileio_asyncio.c
 create mode 100644 programs/lorem.c
 create mode 100644 programs/util.c
 create mode 100644 programs/zstdcli.c

diff --git a/contrib/largeNbDicts/largeNbDicts.c b/contrib/largeNbDicts/largeNbDicts.c
new file mode 100644
index 0000000..6502f12
--- /dev/null
+++ b/contrib/largeNbDicts/largeNbDicts.c
@@ -0,0 +1,1085 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* largeNbDicts
+ * This is a benchmark test tool
+ * dedicated to the specific case of dictionary decompression
+ * using a very large nb of dictionaries
+ * thus suffering latency from lots of cache misses.
+ * It's created in a bid to investigate performance and find optimizations. */
+
+
+/*---  Dependencies  ---*/
+
+#include <stddef.h>   /* size_t */
+#include <stdlib.h>   /* malloc, free, abort, qsort*/
+#include <stdio.h>    /* fprintf */
+#include <limits.h>   /* UINT_MAX */
+#include <assert.h>   /* assert */
+
+#include "util.h"
+#include "benchfn.h"
+#define ZSTD_STATIC_LINKING_ONLY
+#include "zstd.h"
+#include "zdict.h"
+
+
+/*---  Constants  --- */
+
+#define KB  *(1<<10)
+#define MB  *(1<<20)
+
+#define BLOCKSIZE_DEFAULT 0  /* no slicing into blocks */
+#define DICTSIZE  (4 KB)
+#define CLEVEL_DEFAULT 3
+#define DICT_LOAD_METHOD ZSTD_dlm_byCopy
+
+#define BENCH_TIME_DEFAULT_S   6
+#define RUN_TIME_DEFAULT_MS    1000
+#define BENCH_TIME_DEFAULT_MS (BENCH_TIME_DEFAULT_S * RUN_TIME_DEFAULT_MS)
+
+#define DISPLAY_LEVEL_DEFAULT 3
+
+#define BENCH_SIZE_MAX (1200 MB)
+
+
+/*---  Macros  ---*/
+
+#define CONTROL(c)   { if (!(c)) abort(); }
+#undef MIN
+#define MIN(a,b)     ((a) < (b) ? (a) : (b))
+
+
+/*---  Display Macros  ---*/
+
+#define DISPLAY(...)         fprintf(stdout, __VA_ARGS__)
+#define DISPLAYLEVEL(l, ...) { if (g_displayLevel>=l) { DISPLAY(__VA_ARGS__); } }
+static int g_displayLevel = DISPLAY_LEVEL_DEFAULT;   /* 0 : no display,  1: errors,  2 : + result + interaction + warnings,  3 : + progression,  4 : + information */
+
+
+/*---  buffer_t  ---*/
+
+typedef struct {
+    void* ptr;
+    size_t size;
+    size_t capacity;
+} buffer_t;
+
+static const buffer_t kBuffNull = { NULL, 0, 0 };
+
+/* @return : kBuffNull if any error */
+static buffer_t createBuffer(size_t capacity)
+{
+    assert(capacity > 0);
+    void* const ptr = malloc(capacity);
+    if (ptr==NULL) return kBuffNull;
+
+    buffer_t buffer;
+    buffer.ptr = ptr;
+    buffer.capacity = capacity;
+    buffer.size = 0;
+    return buffer;
+}
+
+static void freeBuffer(buffer_t buff)
+{
+    free(buff.ptr);
+}
+
+
+static void fillBuffer_fromHandle(buffer_t* buff, FILE* f)
+{
+    size_t const readSize = fread(buff->ptr, 1, buff->capacity, f);
+    buff->size = readSize;
+}
+
+
+/* @return : kBuffNull if any error */
+static buffer_t createBuffer_fromFile(const char* fileName)
+{
+    U64 const fileSize = UTIL_getFileSize(fileName);
+    size_t const bufferSize = (size_t) fileSize;
+
+    if (fileSize == UTIL_FILESIZE_UNKNOWN) return kBuffNull;
+    assert((U64)bufferSize == fileSize);   /* check overflow */
+
+    {   FILE* const f = fopen(fileName, "rb");
+        if (f == NULL) return kBuffNull;
+
+        buffer_t buff = createBuffer(bufferSize);
+        CONTROL(buff.ptr != NULL);
+
+        fillBuffer_fromHandle(&buff, f);
+        CONTROL(buff.size == buff.capacity);
+
+        fclose(f);   /* do nothing specific if fclose() fails */
+        return buff;
+    }
+}
+
+
+/* @return : kBuffNull if any error */
+static buffer_t
+createDictionaryBuffer(const char* dictionaryName,
+                       const void* srcBuffer,
+                       const size_t* srcBlockSizes, size_t nbBlocks,
+                       size_t requestedDictSize)
+{
+    if (dictionaryName) {
+        DISPLAYLEVEL(3, "loading dictionary %s \n", dictionaryName);
+        return createBuffer_fromFile(dictionaryName);  /* note : result might be kBuffNull */
+
+    } else {
+
+        DISPLAYLEVEL(3, "creating dictionary, of target size %u bytes \n",
+                        (unsigned)requestedDictSize);
+        void* const dictBuffer = malloc(requestedDictSize);
+        CONTROL(dictBuffer != NULL);
+
+        assert(nbBlocks <= UINT_MAX);
+        size_t const dictSize = ZDICT_trainFromBuffer(dictBuffer, requestedDictSize,
+                                                      srcBuffer,
+                                                      srcBlockSizes, (unsigned)nbBlocks);
+        CONTROL(!ZSTD_isError(dictSize));
+
+        buffer_t result;
+        result.ptr = dictBuffer;
+        result.capacity = requestedDictSize;
+        result.size = dictSize;
+        return result;
+    }
+}
+
+/*! BMK_loadFiles() :
+ *  Loads `buffer`, with content from files listed within `fileNamesTable`.
+ *  Fills `buffer` entirely.
+ * @return : 0 on success, !=0 on error */
+static int loadFiles(void* buffer, size_t bufferSize,
+                     size_t* fileSizes,
+                     const char* const * fileNamesTable, unsigned nbFiles)
+{
+    size_t pos = 0, totalSize = 0;
+
+    for (unsigned n=0; n<nbFiles; n++) {
+        U64 fileSize = UTIL_getFileSize(fileNamesTable[n]);
+        if (UTIL_isDirectory(fileNamesTable[n])) {
+            fileSizes[n] = 0;
+            continue;
+        }
+        if (fileSize == UTIL_FILESIZE_UNKNOWN) {
+            fileSizes[n] = 0;
+            continue;
+        }
+
+        FILE* const f = fopen(fileNamesTable[n], "rb");
+        assert(f!=NULL);
+
+        assert(pos <= bufferSize);
+        assert(fileSize <= bufferSize - pos);
+
+        {   size_t const readSize = fread(((char*)buffer)+pos, 1, (size_t)fileSize, f);
+            assert(readSize == fileSize);
+            pos += readSize;
+        }
+        fileSizes[n] = (size_t)fileSize;
+        totalSize += (size_t)fileSize;
+        fclose(f);
+    }
+
+    assert(totalSize == bufferSize);
+    return 0;
+}
+
+
+
+/*---  slice_collection_t  ---*/
+
+typedef struct {
+    void** slicePtrs;
+    size_t* capacities;
+    size_t nbSlices;
+} slice_collection_t;
+
+static const slice_collection_t kNullCollection = { NULL, NULL, 0 };
+
+static void freeSliceCollection(slice_collection_t collection)
+{
+    free(collection.slicePtrs);
+    free(collection.capacities);
+}
+
+/* shrinkSizes() :
+ * downsizes sizes of slices within collection, according to `newSizes`.
+ * every `newSizes` entry must be <= than its corresponding collection size */
+void shrinkSizes(slice_collection_t collection,
+                 const size_t* newSizes)  /* presumed same size as collection */
+{
+    size_t const nbSlices = collection.nbSlices;
+    for (size_t blockNb = 0; blockNb < nbSlices; blockNb++) {
+        assert(newSizes[blockNb] <= collection.capacities[blockNb]);
+        collection.capacities[blockNb] = newSizes[blockNb];
+    }
+}
+
+
+/* splitSlices() :
+ * nbSlices : if == 0, nbSlices is automatically determined from srcSlices and blockSize.
+ *            otherwise, creates exactly nbSlices slices,
+ *            by either truncating input (when smaller)
+ *            or repeating input from beginning */
+static slice_collection_t
+splitSlices(slice_collection_t srcSlices, size_t blockSize, size_t nbSlices)
+{
+    if (blockSize==0) blockSize = (size_t)(-1);   /* means "do not cut" */
+    size_t nbSrcBlocks = 0;
+    for (size_t ssnb=0; ssnb < srcSlices.nbSlices; ssnb++) {
+        size_t pos = 0;
+        while (pos <= srcSlices.capacities[ssnb]) {
+            nbSrcBlocks++;
+            pos += blockSize;
+        }
+    }
+
+    if (nbSlices == 0) nbSlices = nbSrcBlocks;
+
+    void** const sliceTable = (void**)malloc(nbSlices * sizeof(*sliceTable));
+    size_t* const capacities = (size_t*)malloc(nbSlices * sizeof(*capacities));
+    if (sliceTable == NULL || capacities == NULL) {
+        free(sliceTable);
+        free(capacities);
+        return kNullCollection;
+    }
+
+    size_t ssnb = 0;
+    for (size_t sliceNb=0; sliceNb < nbSlices; ) {
+        ssnb = (ssnb + 1) % srcSlices.nbSlices;
+        size_t pos = 0;
+        char* const ptr = (char*)srcSlices.slicePtrs[ssnb];
+        while (pos < srcSlices.capacities[ssnb] && sliceNb < nbSlices) {
+            size_t const size = MIN(blockSize, srcSlices.capacities[ssnb] - pos);
+            sliceTable[sliceNb] = ptr + pos;
+            capacities[sliceNb] = size;
+            sliceNb++;
+            pos += blockSize;
+        }
+    }
+
+    slice_collection_t result;
+    result.nbSlices = nbSlices;
+    result.slicePtrs = sliceTable;
+    result.capacities = capacities;
+    return result;
+}
+
+
+static size_t sliceCollection_totalCapacity(slice_collection_t sc)
+{
+    size_t totalSize = 0;
+    for (size_t n=0; n<sc.nbSlices; n++)
+        totalSize += sc.capacities[n];
+    return totalSize;
+}
+
+
+/* ---  buffer collection  --- */
+
+typedef struct {
+    buffer_t buffer;
+    slice_collection_t slices;
+} buffer_collection_t;
+
+
+static void freeBufferCollection(buffer_collection_t bc)
+{
+    freeBuffer(bc.buffer);
+    freeSliceCollection(bc.slices);
+}
+
+
+static buffer_collection_t
+createBufferCollection_fromSliceCollectionSizes(slice_collection_t sc)
+{
+    size_t const bufferSize = sliceCollection_totalCapacity(sc);
+
+    buffer_t buffer = createBuffer(bufferSize);
+    CONTROL(buffer.ptr != NULL);
+
+    size_t const nbSlices = sc.nbSlices;
+    void** const slices = (void**)malloc(nbSlices * sizeof(*slices));
+    CONTROL(slices != NULL);
+
+    size_t* const capacities = (size_t*)malloc(nbSlices * sizeof(*capacities));
+    CONTROL(capacities != NULL);
+
+    char* const ptr = (char*)buffer.ptr;
+    size_t pos = 0;
+    for (size_t n=0; n < nbSlices; n++) {
+        capacities[n] = sc.capacities[n];
+        slices[n] = ptr + pos;
+        pos += capacities[n];
+    }
+
+    buffer_collection_t result;
+    result.buffer = buffer;
+    result.slices.nbSlices = nbSlices;
+    result.slices.capacities = capacities;
+    result.slices.slicePtrs = slices;
+    return result;
+}
+
+static buffer_collection_t
+createBufferCollection_fromSliceCollection(slice_collection_t sc)
+{
+    size_t const bufferSize = sliceCollection_totalCapacity(sc);
+
+    buffer_t buffer = createBuffer(bufferSize);
+    CONTROL(buffer.ptr != NULL);
+
+    size_t const nbSlices = sc.nbSlices;
+    void** const slices = (void**)malloc(nbSlices * sizeof(*slices));
+    CONTROL(slices != NULL);
+
+    size_t* const capacities = (size_t*)malloc(nbSlices * sizeof(*capacities));
+    CONTROL(capacities != NULL);
+
+    char* const ptr = (char*)buffer.ptr;
+    size_t pos = 0;
+    for (size_t n=0; n < nbSlices; n++) {
+        capacities[n] = sc.capacities[n];
+        slices[n] = ptr + pos;
+        pos += capacities[n];
+    }
+
+    for (size_t i = 0; i < nbSlices; i++) {
+        memcpy(slices[i], sc.slicePtrs[i], sc.capacities[i]);
+        capacities[i] = sc.capacities[i];
+    }
+
+    buffer_collection_t result;
+    result.buffer = buffer;
+    result.slices.nbSlices = nbSlices;
+    result.slices.capacities = capacities;
+    result.slices.slicePtrs = slices;
+
+    return result;
+}
+
+/* @return : kBuffNull if any error */
+static buffer_collection_t
+createBufferCollection_fromFiles(const char* const * fileNamesTable, unsigned nbFiles)
+{
+    U64 const totalSizeToLoad = UTIL_getTotalFileSize(fileNamesTable, nbFiles);
+    assert(totalSizeToLoad != UTIL_FILESIZE_UNKNOWN);
+    assert(totalSizeToLoad <= BENCH_SIZE_MAX);
+    size_t const loadedSize = (size_t)totalSizeToLoad;
+    assert(loadedSize > 0);
+    void* const srcBuffer = malloc(loadedSize);
+    assert(srcBuffer != NULL);
+
+    assert(nbFiles > 0);
+    size_t* const fileSizes = (size_t*)calloc(nbFiles, sizeof(*fileSizes));
+    assert(fileSizes != NULL);
+
+    /* Load input buffer */
+    int const errorCode = loadFiles(srcBuffer, loadedSize,
+                                    fileSizes,
+                                    fileNamesTable, nbFiles);
+    assert(errorCode == 0);
+
+    void** sliceTable = (void**)malloc(nbFiles * sizeof(*sliceTable));
+    assert(sliceTable != NULL);
+
+    char* const ptr = (char*)srcBuffer;
+    size_t pos = 0;
+    unsigned fileNb = 0;
+    for ( ; (pos < loadedSize) && (fileNb < nbFiles); fileNb++) {
+        sliceTable[fileNb] = ptr + pos;
+        pos += fileSizes[fileNb];
+    }
+    assert(pos == loadedSize);
+    assert(fileNb == nbFiles);
+
+
+    buffer_t buffer;
+    buffer.ptr = srcBuffer;
+    buffer.capacity = loadedSize;
+    buffer.size = loadedSize;
+
+    slice_collection_t slices;
+    slices.slicePtrs = sliceTable;
+    slices.capacities = fileSizes;
+    slices.nbSlices = nbFiles;
+
+    buffer_collection_t bc;
+    bc.buffer = buffer;
+    bc.slices = slices;
+    return bc;
+}
+
+
+
+
+/*---  ddict_collection_t  ---*/
+
+typedef struct {
+    ZSTD_DDict** ddicts;
+    size_t nbDDict;
+} ddict_collection_t;
+
+typedef struct {
+    ZSTD_CDict** cdicts;
+    size_t nbCDict;
+} cdict_collection_t;
+
+static const cdict_collection_t kNullCDictCollection = { NULL, 0 };
+
+static void freeCDictCollection(cdict_collection_t cdictc)
+{
+    for (size_t dictNb=0; dictNb < cdictc.nbCDict; dictNb++) {
+        ZSTD_freeCDict(cdictc.cdicts[dictNb]);
+    }
+    free(cdictc.cdicts);
+}
+
+/* returns .buffers=NULL if operation fails */
+static cdict_collection_t createCDictCollection(const void* dictBuffer, size_t dictSize, size_t nbCDict, ZSTD_dictContentType_e dictContentType, ZSTD_CCtx_params* cctxParams)
+{
+    ZSTD_CDict** const cdicts = malloc(nbCDict * sizeof(ZSTD_CDict*));
+    if (cdicts==NULL) return kNullCDictCollection;
+    for (size_t dictNb=0; dictNb < nbCDict; dictNb++) {
+        cdicts[dictNb] = ZSTD_createCDict_advanced2(dictBuffer, dictSize, DICT_LOAD_METHOD, dictContentType, cctxParams, ZSTD_defaultCMem);
+        CONTROL(cdicts[dictNb] != NULL);
+    }
+    cdict_collection_t cdictc;
+    cdictc.cdicts = cdicts;
+    cdictc.nbCDict = nbCDict;
+    return cdictc;
+}
+
+static const ddict_collection_t kNullDDictCollection = { NULL, 0 };
+
+static void freeDDictCollection(ddict_collection_t ddictc)
+{
+    for (size_t dictNb=0; dictNb < ddictc.nbDDict; dictNb++) {
+        ZSTD_freeDDict(ddictc.ddicts[dictNb]);
+    }
+    free(ddictc.ddicts);
+}
+
+/* returns .buffers=NULL if operation fails */
+static ddict_collection_t createDDictCollection(const void* dictBuffer, size_t dictSize, size_t nbDDict)
+{
+    ZSTD_DDict** const ddicts = malloc(nbDDict * sizeof(ZSTD_DDict*));
+    assert(ddicts != NULL);
+    if (ddicts==NULL) return kNullDDictCollection;
+    for (size_t dictNb=0; dictNb < nbDDict; dictNb++) {
+        ddicts[dictNb] = ZSTD_createDDict(dictBuffer, dictSize);
+        assert(ddicts[dictNb] != NULL);
+    }
+    ddict_collection_t ddictc;
+    ddictc.ddicts = ddicts;
+    ddictc.nbDDict = nbDDict;
+    return ddictc;
+}
+
+
+/* mess with addresses, so that linear scanning dictionaries != linear address scanning */
+void shuffleCDictionaries(cdict_collection_t dicts)
+{
+    size_t const nbDicts = dicts.nbCDict;
+    for (size_t r=0; r<nbDicts; r++) {
+        size_t const d = (size_t)rand() % nbDicts;
+        ZSTD_CDict* tmpd = dicts.cdicts[d];
+        dicts.cdicts[d] = dicts.cdicts[r];
+        dicts.cdicts[r] = tmpd;
+    }
+    for (size_t r=0; r<nbDicts; r++) {
+        size_t const d1 = (size_t)rand() % nbDicts;
+        size_t const d2 = (size_t)rand() % nbDicts;
+        ZSTD_CDict* tmpd = dicts.cdicts[d1];
+        dicts.cdicts[d1] = dicts.cdicts[d2];
+        dicts.cdicts[d2] = tmpd;
+    }
+}
+
+/* mess with addresses, so that linear scanning dictionaries != linear address scanning */
+void shuffleDDictionaries(ddict_collection_t dicts)
+{
+    size_t const nbDicts = dicts.nbDDict;
+    for (size_t r=0; r<nbDicts; r++) {
+        size_t const d = (size_t)rand() % nbDicts;
+        ZSTD_DDict* tmpd = dicts.ddicts[d];
+        dicts.ddicts[d] = dicts.ddicts[r];
+        dicts.ddicts[r] = tmpd;
+    }
+    for (size_t r=0; r<nbDicts; r++) {
+        size_t const d1 = (size_t)rand() % nbDicts;
+        size_t const d2 = (size_t)rand() % nbDicts;
+        ZSTD_DDict* tmpd = dicts.ddicts[d1];
+        dicts.ddicts[d1] = dicts.ddicts[d2];
+        dicts.ddicts[d2] = tmpd;
+    }
+}
+
+
+/* ---   Compression  --- */
+
+/* compressBlocks() :
+ * @return : total compressed size of all blocks,
+ *        or 0 if error.
+ */
+static size_t compressBlocks(size_t* cSizes,   /* optional (can be NULL). If present, must contain at least nbBlocks fields */
+                             slice_collection_t dstBlockBuffers,
+                             slice_collection_t srcBlockBuffers,
+                             ZSTD_CDict* cdict, int cLevel)
+{
+    size_t const nbBlocks = srcBlockBuffers.nbSlices;
+    assert(dstBlockBuffers.nbSlices == srcBlockBuffers.nbSlices);
+
+    ZSTD_CCtx* const cctx = ZSTD_createCCtx();
+    assert(cctx != NULL);
+
+    size_t totalCSize = 0;
+    for (size_t blockNb=0; blockNb < nbBlocks; blockNb++) {
+        size_t cBlockSize;
+        if (cdict == NULL) {
+            cBlockSize = ZSTD_compressCCtx(cctx,
+                            dstBlockBuffers.slicePtrs[blockNb], dstBlockBuffers.capacities[blockNb],
+                            srcBlockBuffers.slicePtrs[blockNb], srcBlockBuffers.capacities[blockNb],
+                            cLevel);
+        } else {
+            cBlockSize = ZSTD_compress_usingCDict(cctx,
+                            dstBlockBuffers.slicePtrs[blockNb], dstBlockBuffers.capacities[blockNb],
+                            srcBlockBuffers.slicePtrs[blockNb], srcBlockBuffers.capacities[blockNb],
+                            cdict);
+        }
+        CONTROL(!ZSTD_isError(cBlockSize));
+        if (cSizes) cSizes[blockNb] = cBlockSize;
+        totalCSize += cBlockSize;
+    }
+    return totalCSize;
+}
+
+
+/* ---  Benchmark  --- */
+
+typedef struct {
+    ZSTD_CCtx* cctx;
+    size_t nbDicts;
+    size_t dictNb;
+    cdict_collection_t dictionaries;
+} compressInstructions;
+
+compressInstructions createCompressInstructions(cdict_collection_t dictionaries, ZSTD_CCtx_params* cctxParams)
+{
+    compressInstructions ci;
+    ci.cctx = ZSTD_createCCtx();
+    CONTROL(ci.cctx != NULL);
+    if (cctxParams)
+      ZSTD_CCtx_setParametersUsingCCtxParams(ci.cctx, cctxParams);
+    ci.nbDicts = dictionaries.nbCDict;
+    ci.dictNb = 0;
+    ci.dictionaries = dictionaries;
+    return ci;
+}
+
+void freeCompressInstructions(compressInstructions ci)
+{
+    ZSTD_freeCCtx(ci.cctx);
+}
+
+typedef struct {
+    ZSTD_DCtx* dctx;
+    size_t nbDicts;
+    size_t dictNb;
+    ddict_collection_t dictionaries;
+} decompressInstructions;
+
+decompressInstructions createDecompressInstructions(ddict_collection_t dictionaries)
+{
+    decompressInstructions di;
+    di.dctx = ZSTD_createDCtx();
+    assert(di.dctx != NULL);
+    di.nbDicts = dictionaries.nbDDict;
+    di.dictNb = 0;
+    di.dictionaries = dictionaries;
+    return di;
+}
+
+void freeDecompressInstructions(decompressInstructions di)
+{
+    ZSTD_freeDCtx(di.dctx);
+}
+
+/* benched function */
+size_t compress(const void* src, size_t srcSize, void* dst, size_t dstCapacity, void* payload)
+{
+    compressInstructions* const ci = (compressInstructions*) payload;
+    (void)dstCapacity;
+
+    ZSTD_CCtx_refCDict(ci->cctx, ci->dictionaries.cdicts[ci->dictNb]);
+    ZSTD_compress2(ci->cctx,
+            dst, srcSize,
+            src, srcSize);
+
+    ci->dictNb = ci->dictNb + 1;
+    if (ci->dictNb >= ci->nbDicts) ci->dictNb = 0;
+
+    return srcSize;
+}
+
+/* benched function */
+size_t decompress(const void* src, size_t srcSize, void* dst, size_t dstCapacity, void* payload)
+{
+    decompressInstructions* const di = (decompressInstructions*) payload;
+
+    size_t const result = ZSTD_decompress_usingDDict(di->dctx,
+                                        dst, dstCapacity,
+                                        src, srcSize,
+                                        di->dictionaries.ddicts[di->dictNb]);
+
+    di->dictNb = di->dictNb + 1;
+    if (di->dictNb >= di->nbDicts) di->dictNb = 0;
+
+    return result;
+}
+
+typedef enum {
+  fastest = 0,
+  median = 1,
+} metricAggregatePref_e;
+
+/* compareFunction() :
+ * Sort input in decreasing order when used with qsort() */
+int compareFunction(const void *a, const void *b)
+{
+  double x = *(const double *)a;
+  double y = *(const double *)b;
+  if (x < y)
+    return 1;
+  else if (x > y)
+    return -1;
+  return 0;
+}
+
+double aggregateData(double *data, size_t size,
+                     metricAggregatePref_e metricAggregatePref)
+{
+  qsort(data, size, sizeof(*data), compareFunction);
+  if (metricAggregatePref == fastest)
+    return data[0];
+  else /* median */
+    return (data[(size - 1) / 2] + data[size / 2]) / 2;
+}
+
+static int benchMem(slice_collection_t dstBlocks, slice_collection_t srcBlocks,
+                    ddict_collection_t ddictionaries,
+                    cdict_collection_t cdictionaries, unsigned nbRounds,
+                    int benchCompression, const char *exeName,
+                    ZSTD_CCtx_params *cctxParams,
+                    metricAggregatePref_e metricAggregatePref)
+{
+    assert(dstBlocks.nbSlices == srcBlocks.nbSlices);
+    if (benchCompression) assert(cctxParams);
+
+    unsigned const ms_per_round = RUN_TIME_DEFAULT_MS;
+    unsigned const total_time_ms = nbRounds * ms_per_round;
+
+    double *const speedPerRound = (double *)malloc(nbRounds * sizeof(double));
+
+    BMK_timedFnState_t* const benchState =
+            BMK_createTimedFnState(total_time_ms, ms_per_round);
+
+    decompressInstructions di = createDecompressInstructions(ddictionaries);
+    compressInstructions ci =
+        createCompressInstructions(cdictionaries, cctxParams);
+    void* payload = benchCompression ? (void*)&ci : (void*)&di;
+    BMK_benchParams_t const bp = {
+        .benchFn = benchCompression ? compress : decompress,
+        .benchPayload = payload,
+        .initFn = NULL,
+        .initPayload = NULL,
+        .errorFn = ZSTD_isError,
+        .blockCount = dstBlocks.nbSlices,
+        .srcBuffers = (const void* const*) srcBlocks.slicePtrs,
+        .srcSizes = srcBlocks.capacities,
+        .dstBuffers = dstBlocks.slicePtrs,
+        .dstCapacities = dstBlocks.capacities,
+        .blockResults = NULL
+    };
+
+    size_t roundNb = 0;
+    for (;;) {
+        BMK_runOutcome_t const outcome = BMK_benchTimedFn(benchState, bp);
+        CONTROL(BMK_isSuccessful_runOutcome(outcome));
+
+        BMK_runTime_t const result = BMK_extract_runTime(outcome);
+        double const dTime_ns = result.nanoSecPerRun;
+        double const dTime_sec = (double)dTime_ns / 1000000000;
+        size_t const srcSize = result.sumOfReturn;
+        double const speed_MBps = (double)srcSize / dTime_sec / (1 MB);
+        speedPerRound[roundNb] = speed_MBps;
+        if (benchCompression)
+            DISPLAY("Compression Speed : %.1f MB/s \r", speed_MBps);
+        else
+            DISPLAY("Decompression Speed : %.1f MB/s \r", speed_MBps);
+
+        fflush(stdout);
+        if (BMK_isCompleted_TimedFn(benchState)) break;
+        roundNb++;
+    }
+    DISPLAY("\n");
+    /* BMK_benchTimedFn may not run exactly nbRounds iterations */
+    double speedAggregated =
+        aggregateData(speedPerRound, roundNb + 1, metricAggregatePref);
+    if (metricAggregatePref == fastest)
+      DISPLAY("Fastest Speed : %.1f MB/s \n", speedAggregated);
+    else
+      DISPLAY("Median Speed : %.1f MB/s \n", speedAggregated);
+
+    char* csvFileName = malloc(strlen(exeName) + 5);
+    strcpy(csvFileName, exeName);
+    strcat(csvFileName, ".csv");
+    FILE* csvFile = fopen(csvFileName, "r");
+    if (!csvFile) {
+        csvFile = fopen(csvFileName, "wt");
+        assert(csvFile);
+        fprintf(csvFile, "%s\n", exeName);
+        /* Print table headers */
+        fprintf(
+            csvFile,
+            "Compression/Decompression,Level,nbDicts,dictAttachPref,metricAggregatePref,Speed\n");
+    } else {
+        fclose(csvFile);
+        csvFile = fopen(csvFileName, "at");
+        assert(csvFile);
+    }
+
+    int cLevel = -1;
+    int dictAttachPref = -1;
+    if (benchCompression) {
+      ZSTD_CCtxParams_getParameter(cctxParams, ZSTD_c_compressionLevel,
+                                   &cLevel);
+      ZSTD_CCtxParams_getParameter(cctxParams, ZSTD_c_forceAttachDict,
+                                   &dictAttachPref);
+    }
+    fprintf(csvFile, "%s,%d,%ld,%d,%d,%.1f\n",
+            benchCompression ? "Compression" : "Decompression", cLevel,
+            benchCompression ? ci.nbDicts : di.nbDicts, dictAttachPref,
+            metricAggregatePref, speedAggregated);
+    fclose(csvFile);
+    free(csvFileName);
+
+    freeDecompressInstructions(di);
+    freeCompressInstructions(ci);
+    BMK_freeTimedFnState(benchState);
+
+    return 0;   /* success */
+}
+
+
+/*! bench() :
+ *  fileName : file to load for benchmarking purpose
+ *  dictionary : optional (can be NULL), file to load as dictionary,
+ *              if none provided : will be calculated on the fly by the program.
+ * @return : 0 is success, 1+ otherwise */
+int bench(const char **fileNameTable, unsigned nbFiles, const char *dictionary,
+          size_t blockSize, int clevel, unsigned nbDictMax, unsigned nbBlocks,
+          unsigned nbRounds, int benchCompression,
+          ZSTD_dictContentType_e dictContentType, ZSTD_CCtx_params *cctxParams,
+          const char *exeName, metricAggregatePref_e metricAggregatePref)
+{
+    int result = 0;
+
+    DISPLAYLEVEL(3, "loading %u files... \n", nbFiles);
+    buffer_collection_t const srcs = createBufferCollection_fromFiles(fileNameTable, nbFiles);
+    CONTROL(srcs.buffer.ptr != NULL);
+    buffer_t srcBuffer = srcs.buffer;
+    size_t const srcSize = srcBuffer.size;
+    DISPLAYLEVEL(3, "created src buffer of size %.1f MB \n",
+                    (double)srcSize / (1 MB));
+
+    slice_collection_t const srcSlices = splitSlices(srcs.slices, blockSize, nbBlocks);
+    nbBlocks = (unsigned)(srcSlices.nbSlices);
+    DISPLAYLEVEL(3, "split input into %u blocks ", nbBlocks);
+    if (blockSize)
+        DISPLAYLEVEL(3, "of max size %u bytes ", (unsigned)blockSize);
+    DISPLAYLEVEL(3, "\n");
+    size_t const totalSrcSlicesSize = sliceCollection_totalCapacity(srcSlices);
+
+
+    size_t* const dstCapacities = malloc(nbBlocks * sizeof(*dstCapacities));
+    CONTROL(dstCapacities != NULL);
+    size_t dstBufferCapacity = 0;
+    for (size_t bnb=0; bnb<nbBlocks; bnb++) {
+        dstCapacities[bnb] = ZSTD_compressBound(srcSlices.capacities[bnb]);
+        dstBufferCapacity += dstCapacities[bnb];
+    }
+
+    buffer_t dstBuffer = createBuffer(dstBufferCapacity);
+    CONTROL(dstBuffer.ptr != NULL);
+
+    void** const sliceTable = malloc(nbBlocks * sizeof(*sliceTable));
+    CONTROL(sliceTable != NULL);
+
+    {   char* const ptr = dstBuffer.ptr;
+        size_t pos = 0;
+        for (size_t snb=0; snb < nbBlocks; snb++) {
+            sliceTable[snb] = ptr + pos;
+            pos += dstCapacities[snb];
+    }   }
+
+    slice_collection_t dstSlices;
+    dstSlices.capacities = dstCapacities;
+    dstSlices.slicePtrs = sliceTable;
+    dstSlices.nbSlices = nbBlocks;
+
+
+    /* dictionary determination */
+    buffer_t const dictBuffer = createDictionaryBuffer(dictionary,
+                                srcs.buffer.ptr,
+                                srcSlices.capacities, srcSlices.nbSlices,
+                                DICTSIZE);
+    CONTROL(dictBuffer.ptr != NULL);
+
+    ZSTD_CDict* const cdict = ZSTD_createCDict_advanced2(dictBuffer.ptr, dictBuffer.size, DICT_LOAD_METHOD, dictContentType, cctxParams, ZSTD_defaultCMem);
+    CONTROL(cdict != NULL);
+
+    size_t const cTotalSizeNoDict = compressBlocks(NULL, dstSlices, srcSlices, NULL, clevel);
+    CONTROL(cTotalSizeNoDict != 0);
+    DISPLAYLEVEL(3, "compressing at level %u without dictionary : Ratio=%.2f  (%u bytes) \n",
+                    clevel,
+                    (double)totalSrcSlicesSize / (double)cTotalSizeNoDict, (unsigned)cTotalSizeNoDict);
+
+    size_t* const cSizes = malloc(nbBlocks * sizeof(size_t));
+    CONTROL(cSizes != NULL);
+
+    size_t const cTotalSize = compressBlocks(cSizes, dstSlices, srcSlices, cdict, clevel);
+    CONTROL(cTotalSize != 0);
+    DISPLAYLEVEL(3, "compressed using a %u bytes dictionary : Ratio=%.2f  (%u bytes) \n",
+                    (unsigned)dictBuffer.size,
+                    (double)totalSrcSlicesSize / (double)cTotalSize, (unsigned)cTotalSize);
+
+    /* now dstSlices contain the real compressed size of each block, instead of the maximum capacity */
+    shrinkSizes(dstSlices, cSizes);
+
+    unsigned const nbDicts = nbDictMax ? nbDictMax : nbBlocks;
+
+    cdict_collection_t const cdictionaries = createCDictCollection(dictBuffer.ptr, dictBuffer.size, nbDicts, dictContentType, cctxParams);
+    CONTROL(cdictionaries.cdicts != NULL);
+
+    ddict_collection_t const ddictionaries = createDDictCollection(dictBuffer.ptr, dictBuffer.size, nbDicts);
+    CONTROL(ddictionaries.ddicts != NULL);
+
+    if (benchCompression) {
+        size_t const dictMem = ZSTD_sizeof_CDict(cdictionaries.cdicts[0]);
+        size_t const allDictMem = dictMem * nbDicts;
+        DISPLAYLEVEL(3, "generating %u dictionaries, using %.1f MB of memory \n",
+                        nbDicts, (double)allDictMem / (1 MB));
+
+        shuffleCDictionaries(cdictionaries);
+
+        buffer_collection_t resultCollection = createBufferCollection_fromSliceCollection(srcSlices);
+        CONTROL(resultCollection.buffer.ptr != NULL);
+
+        result = benchMem(dstSlices, resultCollection.slices, ddictionaries,
+                          cdictionaries, nbRounds, benchCompression, exeName,
+                          cctxParams, metricAggregatePref);
+
+        freeBufferCollection(resultCollection);
+    } else {
+        size_t const dictMem = ZSTD_estimateDDictSize(dictBuffer.size, DICT_LOAD_METHOD);
+        size_t const allDictMem = dictMem * nbDicts;
+        DISPLAYLEVEL(3, "generating %u dictionaries, using %.1f MB of memory \n",
+                        nbDicts, (double)allDictMem / (1 MB));
+
+        shuffleDDictionaries(ddictionaries);
+
+        buffer_collection_t resultCollection = createBufferCollection_fromSliceCollectionSizes(srcSlices);
+        CONTROL(resultCollection.buffer.ptr != NULL);
+
+        result = benchMem(resultCollection.slices, dstSlices, ddictionaries,
+                          cdictionaries, nbRounds, benchCompression, exeName,
+                          NULL, metricAggregatePref);
+
+        freeBufferCollection(resultCollection);
+    }
+
+    /* free all heap objects in reverse order */
+    freeCDictCollection(cdictionaries);
+    freeDDictCollection(ddictionaries);
+    free(cSizes);
+    ZSTD_freeCDict(cdict);
+    freeBuffer(dictBuffer);
+    freeSliceCollection(dstSlices);
+    freeBuffer(dstBuffer);
+    freeSliceCollection(srcSlices);
+    freeBufferCollection(srcs);
+
+    return result;
+}
+
+
+
+/* ---  Command Line  --- */
+
+/*! readU32FromChar() :
+ * @return : unsigned integer value read from input in `char` format.
+ *  allows and interprets K, KB, KiB, M, MB and MiB suffix.
+ *  Will also modify `*stringPtr`, advancing it to position where it stopped reading.
+ *  Note : function will exit() program if digit sequence overflows */
+static unsigned readU32FromChar(const char** stringPtr)
+{
+    unsigned result = 0;
+    while ((**stringPtr >='0') && (**stringPtr <='9')) {
+        unsigned const max = (((unsigned)(-1)) / 10) - 1;
+        assert(result <= max);   /* check overflow */
+        result *= 10, result += (unsigned)**stringPtr - '0', (*stringPtr)++ ;
+    }
+    if ((**stringPtr=='K') || (**stringPtr=='M')) {
+        unsigned const maxK = ((unsigned)(-1)) >> 10;
+        assert(result <= maxK);   /* check overflow */
+        result <<= 10;
+        if (**stringPtr=='M') {
+            assert(result <= maxK);   /* check overflow */
+            result <<= 10;
+        }
+        (*stringPtr)++;  /* skip `K` or `M` */
+        if (**stringPtr=='i') (*stringPtr)++;
+        if (**stringPtr=='B') (*stringPtr)++;
+    }
+    return result;
+}
+
+/** longCommandWArg() :
+ *  check if *stringPtr is the same as longCommand.
+ *  If yes, @return 1 and advances *stringPtr to the position which immediately follows longCommand.
+ * @return 0 and doesn't modify *stringPtr otherwise.
+ */
+static int longCommandWArg(const char** stringPtr, const char* longCommand)
+{
+    size_t const comSize = strlen(longCommand);
+    int const result = !strncmp(*stringPtr, longCommand, comSize);
+    if (result) *stringPtr += comSize;
+    return result;
+}
+
+
+int usage(const char* exeName)
+{
+    DISPLAY (" \n");
+    DISPLAY (" %s [Options] filename(s) \n", exeName);
+    DISPLAY (" \n");
+    DISPLAY ("Options : \n");
+    DISPLAY ("-z          : benchmark compression (default) \n");
+    DISPLAY ("-d          : benchmark decompression \n");
+    DISPLAY ("-r          : recursively load all files in subdirectories (default: off) \n");
+    DISPLAY ("-B#         : split input into blocks of size # (default: no split) \n");
+    DISPLAY ("-#          : use compression level # (default: %u) \n", CLEVEL_DEFAULT);
+    DISPLAY ("-D #        : use # as a dictionary (default: create one) \n");
+    DISPLAY ("-i#         : nb benchmark rounds (default: %u) \n", BENCH_TIME_DEFAULT_S);
+    DISPLAY ("-p#         : print speed for all rounds 0=fastest 1=median (default: 0) \n");
+    DISPLAY ("--nbBlocks=#: use # blocks for bench (default: one per file) \n");
+    DISPLAY ("--nbDicts=# : create # dictionaries for bench (default: one per block) \n");
+    DISPLAY ("-h          : help (this text) \n");
+    DISPLAY (" \n");
+    DISPLAY ("Advanced Options (see zstd.h for documentation) : \n");
+    DISPLAY ("--dedicated-dict-search\n");
+    DISPLAY ("--dict-content-type=#\n");
+    DISPLAY ("--dict-attach-pref=#\n");
+    return 0;
+}
+
+int bad_usage(const char* exeName)
+{
+    DISPLAY (" bad usage : \n");
+    usage(exeName);
+    return 1;
+}
+
+int main (int argc, const char** argv)
+{
+    int recursiveMode = 0;
+    int benchCompression = 1;
+    int dedicatedDictSearch = 0;
+    unsigned nbRounds = BENCH_TIME_DEFAULT_S;
+    const char* const exeName = argv[0];
+
+    if (argc < 2) return bad_usage(exeName);
+
+    const char** nameTable = (const char**)malloc((size_t)argc * sizeof(const char*));
+    assert(nameTable != NULL);
+    unsigned nameIdx = 0;
+
+    const char* dictionary = NULL;
+    int cLevel = CLEVEL_DEFAULT;
+    size_t blockSize = BLOCKSIZE_DEFAULT;
+    unsigned nbDicts = 0;  /* determine nbDicts automatically: 1 dictionary per block */
+    unsigned nbBlocks = 0; /* determine nbBlocks automatically, from source and blockSize */
+    ZSTD_dictContentType_e dictContentType = ZSTD_dct_auto;
+    ZSTD_dictAttachPref_e dictAttachPref = ZSTD_dictDefaultAttach;
+    ZSTD_ParamSwitch_e prefetchCDictTables = ZSTD_ps_auto;
+    metricAggregatePref_e metricAggregatePref = fastest;
+
+    for (int argNb = 1; argNb < argc ; argNb++) {
+        const char* argument = argv[argNb];
+        if (!strcmp(argument, "-h")) { free(nameTable); return usage(exeName); }
+        if (!strcmp(argument, "-d")) { benchCompression = 0; continue; }
+        if (!strcmp(argument, "-z")) { benchCompression = 1; continue; }
+        if (!strcmp(argument, "-r")) { recursiveMode = 1; continue; }
+        if (!strcmp(argument, "-D")) { argNb++; assert(argNb < argc); dictionary = argv[argNb]; continue; }
+        if (longCommandWArg(&argument, "-i")) { nbRounds = readU32FromChar(&argument); continue; }
+        if (longCommandWArg(&argument, "-p")) { metricAggregatePref = (int)readU32FromChar(&argument); continue;}
+        if (longCommandWArg(&argument, "--dictionary=")) { dictionary = argument; continue; }
+        if (longCommandWArg(&argument, "-B")) { blockSize = readU32FromChar(&argument); continue; }
+        if (longCommandWArg(&argument, "--blockSize=")) { blockSize = readU32FromChar(&argument); continue; }
+        if (longCommandWArg(&argument, "--nbDicts=")) { nbDicts = readU32FromChar(&argument); continue; }
+        if (longCommandWArg(&argument, "--nbBlocks=")) { nbBlocks = readU32FromChar(&argument); continue; }
+        if (longCommandWArg(&argument, "--clevel=")) { cLevel = (int)readU32FromChar(&argument); continue; }
+        if (longCommandWArg(&argument, "--dedicated-dict-search")) { dedicatedDictSearch = 1; continue; }
+        if (longCommandWArg(&argument, "--dict-content-type=")) { dictContentType = (int)readU32FromChar(&argument); continue; }
+        if (longCommandWArg(&argument, "--dict-attach-pref=")) { dictAttachPref = (int)readU32FromChar(&argument); continue; }
+        if (longCommandWArg(&argument, "--prefetch-cdict-tables=")) { prefetchCDictTables = (int)readU32FromChar(&argument); continue; }
+        if (longCommandWArg(&argument, "-")) { cLevel = (int)readU32FromChar(&argument); continue; }
+        /* anything that's not a command is a filename */
+        nameTable[nameIdx++] = argument;
+    }
+
+    FileNamesTable* filenameTable;
+
+    if (recursiveMode) {
+#ifndef UTIL_HAS_CREATEFILELIST
+        assert(0);   /* missing capability, do not run */
+#endif
+        filenameTable = UTIL_createExpandedFNT(nameTable, nameIdx, 1 /* follow_links */);
+    } else {
+        filenameTable = UTIL_assembleFileNamesTable(nameTable, nameIdx, NULL);
+        nameTable = NULL;  /* UTIL_createFileNamesTable() takes ownership of nameTable */
+    }
+
+    ZSTD_CCtx_params* cctxParams = ZSTD_createCCtxParams();
+    ZSTD_CCtxParams_init(cctxParams, cLevel);
+    ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_enableDedicatedDictSearch, dedicatedDictSearch);
+    ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_nbWorkers, 0);
+    ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_forceAttachDict, dictAttachPref);
+    ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_prefetchCDictTables, prefetchCDictTables);
+
+    int result =
+        bench(filenameTable->fileNames, (unsigned)filenameTable->tableSize,
+              dictionary, blockSize, cLevel, nbDicts, nbBlocks, nbRounds,
+              benchCompression, dictContentType, cctxParams, exeName,
+              metricAggregatePref);
+
+    UTIL_freeFileNamesTable(filenameTable);
+    free(nameTable);
+    ZSTD_freeCCtxParams(cctxParams);
+
+    return result;
+}
diff --git a/contrib/match_finders/zstd_edist.c b/contrib/match_finders/zstd_edist.c
new file mode 100644
index 0000000..d685cdd
--- /dev/null
+++ b/contrib/match_finders/zstd_edist.c
@@ -0,0 +1,558 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/*-*************************************
+*  Dependencies
+***************************************/
+
+/* Currently relies on qsort when combining contiguous matches. This can probably 
+ * be avoided but would require changes to the algorithm. The qsort is far from 
+ * the bottleneck in this algorithm even for medium sized files so it's probably 
+ * not worth trying to address */ 
+#include <stdlib.h>
+#include <assert.h>
+
+#include "zstd_edist.h"
+#include "mem.h"
+
+/*-*************************************
+*  Constants
+***************************************/
+
+/* Just a sential for the entries of the diagonal matrix */
+#define ZSTD_EDIST_DIAG_MAX (S32)(1 << 30)
+
+/* How large should a snake be to be considered a 'big' snake. 
+ * For an explanation of what a 'snake' is with respect to the 
+ * edit distance matrix, see the linked paper in zstd_edist.h */
+#define ZSTD_EDIST_SNAKE_THRESH 20
+
+/* After how many iterations should we start to use the heuristic
+ * based on 'big' snakes */
+#define ZSTD_EDIST_SNAKE_ITER_THRESH 200
+
+/* After how many iterations should be just give up and take 
+ * the best available edit script for this round */ 
+#define ZSTD_EDIST_EXPENSIVE_THRESH 1024
+
+/*-*************************************
+*  Structures
+***************************************/
+
+typedef struct {
+    U32 dictIdx;
+    U32 srcIdx;
+    U32 matchLength;
+} ZSTD_eDist_match;
+
+typedef struct {
+    const BYTE* dict;
+    const BYTE* src;
+    size_t dictSize;
+    size_t srcSize;
+    S32* forwardDiag;            /* Entries of the forward diagonal stored here */
+    S32* backwardDiag;           /* Entries of the backward diagonal stored here.
+                                  *   Note: this buffer and the 'forwardDiag' buffer 
+                                  *   are contiguous. See the ZSTD_eDist_genSequences */
+    ZSTD_eDist_match* matches;   /* Accumulate matches of length 1 in this buffer. 
+                                  *   In a subsequence post-processing step, we combine 
+                                  *   contiguous matches. */
+    U32 nbMatches;
+} ZSTD_eDist_state;
+
+typedef struct {
+    S32 dictMid;           /* The mid diagonal for the dictionary */
+    S32 srcMid;            /* The mid diagonal for the source */ 
+    int lowUseHeuristics;  /* Should we use heuristics for the low part */
+    int highUseHeuristics; /* Should we use heuristics for the high part */ 
+} ZSTD_eDist_partition;
+
+/*-*************************************
+*  Internal
+***************************************/
+
+static void ZSTD_eDist_diag(ZSTD_eDist_state* state,
+                    ZSTD_eDist_partition* partition,
+                    S32 dictLow, S32 dictHigh, S32 srcLow, 
+                    S32 srcHigh, int useHeuristics)
+{
+    S32* const forwardDiag = state->forwardDiag;
+    S32* const backwardDiag = state->backwardDiag;
+    const BYTE* const dict = state->dict;
+    const BYTE* const src = state->src;
+
+    S32 const diagMin = dictLow - srcHigh;
+    S32 const diagMax = dictHigh - srcLow;
+    S32 const forwardMid = dictLow - srcLow;
+    S32 const backwardMid = dictHigh - srcHigh;
+
+    S32 forwardMin = forwardMid;
+    S32 forwardMax = forwardMid;
+    S32 backwardMin = backwardMid;
+    S32 backwardMax = backwardMid;
+    int odd = (forwardMid - backwardMid) & 1;
+    U32 iterations;
+
+    forwardDiag[forwardMid] = dictLow;
+    backwardDiag[backwardMid] = dictHigh;
+
+    /* Main loop for updating diag entries. Unless useHeuristics is 
+     * set to false, this loop will run until it finds the minimal 
+     * edit script */ 
+    for (iterations = 1;;iterations++) {
+        S32 diag;
+        int bigSnake = 0;
+        
+        if (forwardMin > diagMin) {
+            forwardMin--;
+            forwardDiag[forwardMin - 1] = -1;
+        } else {
+            forwardMin++;
+        }
+
+        if (forwardMax < diagMax) {
+            forwardMax++;
+            forwardDiag[forwardMax + 1] = -1;
+        } else {
+            forwardMax--;
+        }
+
+        for (diag = forwardMax; diag >= forwardMin; diag -= 2) {
+            S32 dictIdx;
+            S32 srcIdx;
+            S32 low = forwardDiag[diag - 1];
+            S32 high = forwardDiag[diag + 1];
+            S32 dictIdx0 = low < high ? high : low + 1;
+
+            for (dictIdx = dictIdx0, srcIdx = dictIdx0 - diag;
+                dictIdx < dictHigh && srcIdx < srcHigh && dict[dictIdx] == src[srcIdx];
+                dictIdx++, srcIdx++) continue;
+
+            if (dictIdx - dictIdx0 > ZSTD_EDIST_SNAKE_THRESH)
+                bigSnake = 1;
+
+            forwardDiag[diag] = dictIdx;
+
+            if (odd && backwardMin <= diag && diag <= backwardMax && backwardDiag[diag] <= dictIdx) {
+                partition->dictMid = dictIdx;
+                partition->srcMid = srcIdx;
+                partition->lowUseHeuristics = 0;
+                partition->highUseHeuristics = 0;
+                return;
+            }
+        }
+
+        if (backwardMin > diagMin) {
+            backwardMin--;
+            backwardDiag[backwardMin - 1] = ZSTD_EDIST_DIAG_MAX;
+        } else {
+            backwardMin++;
+        }
+
+        if (backwardMax < diagMax) {
+            backwardMax++;
+            backwardDiag[backwardMax + 1] = ZSTD_EDIST_DIAG_MAX;
+        } else {
+            backwardMax--;
+        }
+
+
+        for (diag = backwardMax; diag >= backwardMin; diag -= 2) {
+            S32 dictIdx;
+            S32 srcIdx;
+            S32 low = backwardDiag[diag - 1];
+            S32 high = backwardDiag[diag + 1];
+            S32 dictIdx0 = low < high ? low : high - 1;
+
+            for (dictIdx = dictIdx0, srcIdx = dictIdx0 - diag;
+                dictLow < dictIdx && srcLow < srcIdx && dict[dictIdx - 1] == src[srcIdx - 1];
+                dictIdx--, srcIdx--) continue;
+
+            if (dictIdx0 - dictIdx > ZSTD_EDIST_SNAKE_THRESH)
+                bigSnake = 1;
+
+            backwardDiag[diag] = dictIdx;
+
+            if (!odd && forwardMin <= diag && diag <= forwardMax && dictIdx <= forwardDiag[diag]) {
+                partition->dictMid = dictIdx;
+                partition->srcMid = srcIdx;
+                partition->lowUseHeuristics = 0;
+                partition->highUseHeuristics = 0;
+                return;
+            }
+        }
+
+        if (!useHeuristics)
+            continue;
+
+        /* Everything under this point is a heuristic. Using these will 
+         * substantially speed up the match finding. In some cases, taking 
+         * the total match finding time from several minutes to seconds.
+         * Of course, the caveat is that the edit script found may no longer 
+         * be optimal */ 
+
+        /* Big snake heuristic */ 
+        if (iterations > ZSTD_EDIST_SNAKE_ITER_THRESH && bigSnake) {
+            {
+                S32 best = 0;
+                
+                for (diag = forwardMax; diag >= forwardMin; diag -= 2) {
+                    S32 diagDiag = diag - forwardMid;
+                    S32 dictIdx = forwardDiag[diag];
+                    S32 srcIdx = dictIdx - diag;
+                    S32 v = (dictIdx - dictLow) * 2 - diagDiag;
+
+                    if (v > 12 * (iterations + (diagDiag < 0 ? -diagDiag : diagDiag))) {
+                        if (v > best 
+                          && dictLow + ZSTD_EDIST_SNAKE_THRESH <= dictIdx && dictIdx <= dictHigh
+                          && srcLow + ZSTD_EDIST_SNAKE_THRESH <= srcIdx && srcIdx <= srcHigh) {
+                            S32 k;
+                            for (k = 1; dict[dictIdx - k] == src[srcIdx - k]; k++) {
+                                if (k == ZSTD_EDIST_SNAKE_THRESH) {
+                                    best = v;
+                                    partition->dictMid = dictIdx;
+                                    partition->srcMid = srcIdx;
+                                    break;
+                                }
+                            }
+                        }
+                    }
+                }
+
+                if (best > 0) {
+                    partition->lowUseHeuristics = 0;
+                    partition->highUseHeuristics = 1;
+                    return;
+                }
+            }
+
+            {
+                S32 best = 0;
+
+                for (diag = backwardMax; diag >= backwardMin; diag -= 2) {
+                    S32 diagDiag = diag - backwardMid;
+                    S32 dictIdx = backwardDiag[diag];
+                    S32 srcIdx = dictIdx - diag;
+                    S32 v = (dictHigh - dictIdx) * 2 + diagDiag;
+
+                    if (v > 12 * (iterations + (diagDiag < 0 ? -diagDiag : diagDiag))) {
+                        if (v > best 
+                          && dictLow < dictIdx && dictIdx <= dictHigh - ZSTD_EDIST_SNAKE_THRESH
+                          && srcLow < srcIdx && srcIdx <= srcHigh - ZSTD_EDIST_SNAKE_THRESH) {
+                            int k;
+                            for (k = 0; dict[dictIdx + k] == src[srcIdx + k]; k++) {
+                                if (k == ZSTD_EDIST_SNAKE_THRESH - 1) { 
+                                    best = v;
+                                    partition->dictMid = dictIdx;
+                                    partition->srcMid = srcIdx;
+                                    break; 
+                                }
+                            }
+                        }
+                    }
+                }
+
+                if (best > 0) {
+                    partition->lowUseHeuristics = 1;
+                    partition->highUseHeuristics = 0;
+                    return;
+                }
+            }
+        }
+
+        /* More general 'too expensive' heuristic */ 
+        if (iterations >= ZSTD_EDIST_EXPENSIVE_THRESH) {
+            S32 forwardDictSrcBest;
+            S32 forwardDictBest = 0;
+            S32 backwardDictSrcBest;
+            S32 backwardDictBest = 0;
+
+            forwardDictSrcBest = -1;
+            for (diag = forwardMax; diag >= forwardMin; diag -= 2) {
+                S32 dictIdx = MIN(forwardDiag[diag], dictHigh);
+                S32 srcIdx = dictIdx - diag;
+
+                if (srcHigh < srcIdx) {
+                    dictIdx = srcHigh + diag;
+                    srcIdx = srcHigh;
+                }
+
+                if (forwardDictSrcBest < dictIdx + srcIdx) {
+                    forwardDictSrcBest = dictIdx + srcIdx;
+                    forwardDictBest = dictIdx;
+                }
+            }
+
+            backwardDictSrcBest = ZSTD_EDIST_DIAG_MAX;
+            for (diag = backwardMax; diag >= backwardMin; diag -= 2) {
+                S32 dictIdx = MAX(dictLow, backwardDiag[diag]);
+                S32 srcIdx = dictIdx - diag;
+
+                if (srcIdx < srcLow) {
+                    dictIdx = srcLow + diag;
+                    srcIdx = srcLow;
+                }
+
+                if (dictIdx + srcIdx < backwardDictSrcBest) {
+                    backwardDictSrcBest = dictIdx + srcIdx;
+                    backwardDictBest = dictIdx;
+                }
+            }
+
+            if ((dictHigh + srcHigh) - backwardDictSrcBest < forwardDictSrcBest - (dictLow + srcLow)) {
+                partition->dictMid = forwardDictBest;
+                partition->srcMid = forwardDictSrcBest - forwardDictBest;
+                partition->lowUseHeuristics = 0;
+                partition->highUseHeuristics = 1;
+            } else {
+                partition->dictMid = backwardDictBest;
+                partition->srcMid = backwardDictSrcBest - backwardDictBest;
+                partition->lowUseHeuristics = 1;
+                partition->highUseHeuristics = 0;
+            }
+            return;
+        }
+    }
+}
+
+static void ZSTD_eDist_insertMatch(ZSTD_eDist_state* state, 
+                    S32 const dictIdx, S32 const srcIdx)
+{
+    state->matches[state->nbMatches].dictIdx = dictIdx;
+    state->matches[state->nbMatches].srcIdx = srcIdx;
+    state->matches[state->nbMatches].matchLength = 1;
+    state->nbMatches++;
+}
+
+static int ZSTD_eDist_compare(ZSTD_eDist_state* state,
+                    S32 dictLow, S32 dictHigh, S32 srcLow,
+                    S32 srcHigh, int useHeuristics)
+{
+    const BYTE* const dict = state->dict;
+    const BYTE* const src = state->src;
+
+    /* Found matches while traversing from the low end */ 
+    while (dictLow < dictHigh && srcLow < srcHigh && dict[dictLow] == src[srcLow]) {
+        ZSTD_eDist_insertMatch(state, dictLow, srcLow);
+        dictLow++;
+        srcLow++;
+    }
+
+    /* Found matches while traversing from the high end */
+    while (dictLow < dictHigh && srcLow < srcHigh && dict[dictHigh - 1] == src[srcHigh - 1]) {
+        ZSTD_eDist_insertMatch(state, dictHigh - 1, srcHigh - 1);
+        dictHigh--;
+        srcHigh--;
+    }
+    
+    /* If the low and high end end up touching. If we wanted to make 
+     * note of the differences like most diffing algorithms do, we would 
+     * do so here. In our case, we're only concerned with matches 
+     * Note: if you wanted to find the edit distance of the algorithm, 
+     *   you could just accumulate the cost for an insertion/deletion 
+     *   below. */ 
+    if (dictLow == dictHigh) {
+        while (srcLow < srcHigh) {
+            /* Reaching this point means inserting src[srcLow] into 
+             * the current position of dict */ 
+            srcLow++;
+        }
+    } else if (srcLow == srcHigh) {
+        while (dictLow < dictHigh) {
+            /* Reaching this point means deleting dict[dictLow] from 
+             * the current position of dict */ 
+            dictLow++;
+        }
+    } else {
+        ZSTD_eDist_partition partition;
+        partition.dictMid = 0;
+        partition.srcMid = 0;
+        ZSTD_eDist_diag(state, &partition, dictLow, dictHigh, 
+            srcLow, srcHigh, useHeuristics);
+        if (ZSTD_eDist_compare(state, dictLow, partition.dictMid, 
+          srcLow, partition.srcMid, partition.lowUseHeuristics))
+            return 1;
+        if (ZSTD_eDist_compare(state, partition.dictMid, dictHigh,
+          partition.srcMid, srcHigh, partition.highUseHeuristics))
+            return 1;
+    }
+
+    return 0;
+}
+
+static int ZSTD_eDist_matchComp(const void* p, const void* q)
+{
+    S32 const l = ((ZSTD_eDist_match*)p)->srcIdx;
+    S32 const r = ((ZSTD_eDist_match*)q)->srcIdx;
+    return (l - r);
+}
+
+/* The matches from the approach above will all be of the form 
+ * (dictIdx, srcIdx, 1). This method combines contiguous matches 
+ * of length MINMATCH or greater. Matches less than MINMATCH 
+ * are discarded */ 
+static void ZSTD_eDist_combineMatches(ZSTD_eDist_state* state)
+{
+    /* Create a new buffer to put the combined matches into 
+     * and memcpy to state->matches after */ 
+    ZSTD_eDist_match* combinedMatches = 
+        ZSTD_malloc(state->nbMatches * sizeof(ZSTD_eDist_match), 
+        ZSTD_defaultCMem);
+
+    U32 nbCombinedMatches = 1;
+    size_t i;
+
+    /* Make sure that the srcIdx and dictIdx are in sorted order.
+     * The combination step won't work otherwise */ 
+    qsort(state->matches, state->nbMatches, sizeof(ZSTD_eDist_match), ZSTD_eDist_matchComp);
+
+    memcpy(combinedMatches, state->matches, sizeof(ZSTD_eDist_match));
+    for (i = 1; i < state->nbMatches; i++) {
+        ZSTD_eDist_match const match = state->matches[i];
+        ZSTD_eDist_match const combinedMatch = 
+            combinedMatches[nbCombinedMatches - 1];
+        if (combinedMatch.srcIdx + combinedMatch.matchLength == match.srcIdx && 
+          combinedMatch.dictIdx + combinedMatch.matchLength == match.dictIdx) {
+            combinedMatches[nbCombinedMatches - 1].matchLength++;
+        } else {
+            /* Discard matches that are less than MINMATCH */
+            if (combinedMatches[nbCombinedMatches - 1].matchLength < MINMATCH) {
+                nbCombinedMatches--;
+            }
+
+            memcpy(combinedMatches + nbCombinedMatches, 
+                state->matches + i, sizeof(ZSTD_eDist_match));
+            nbCombinedMatches++;
+        }
+    }
+    memcpy(state->matches, combinedMatches, nbCombinedMatches * sizeof(ZSTD_eDist_match));
+    state->nbMatches = nbCombinedMatches;
+    ZSTD_free(combinedMatches, ZSTD_defaultCMem);
+}
+
+static size_t ZSTD_eDist_convertMatchesToSequences(ZSTD_Sequence* sequences, 
+    ZSTD_eDist_state* state)
+{
+    const ZSTD_eDist_match* matches = state->matches;
+    size_t const nbMatches = state->nbMatches;
+    size_t const dictSize = state->dictSize;
+    size_t nbSequences = 0;
+    size_t i;
+    for (i = 0; i < nbMatches; i++) {
+        ZSTD_eDist_match const match = matches[i];
+        U32 const litLength = !i ? match.srcIdx : 
+            match.srcIdx - (matches[i - 1].srcIdx + matches[i - 1].matchLength);
+        U32 const offset = (match.srcIdx + dictSize) - match.dictIdx;
+        U32 const matchLength = match.matchLength;
+        sequences[nbSequences].offset = offset;
+        sequences[nbSequences].litLength = litLength;
+        sequences[nbSequences].matchLength = matchLength;
+        nbSequences++;
+    }
+    return nbSequences;
+}
+
+/*-*************************************
+*  Internal utils
+***************************************/
+
+static size_t ZSTD_eDist_hamingDist(const BYTE* const a,
+                        const BYTE* const b, size_t n)
+{
+    size_t i;
+    size_t dist = 0;
+    for (i = 0; i < n; i++)
+        dist += a[i] != b[i];
+    return dist; 
+}
+
+/* This is a pretty naive recursive implementation that should only
+ * be used for quick tests obviously. Don't try and run this on a 
+ * GB file or something. There are faster implementations. Use those
+ * if you need to run it for large files. */
+static size_t ZSTD_eDist_levenshteinDist(const BYTE* const s,
+                        size_t const sn, const BYTE* const t,
+                        size_t const tn)
+{
+    size_t a, b, c;
+
+    if (!sn)
+        return tn;
+    if (!tn)
+        return sn;
+    
+    if (s[sn - 1] == t[tn - 1])
+        return ZSTD_eDist_levenshteinDist(
+            s, sn - 1, t, tn - 1);
+    
+    a = ZSTD_eDist_levenshteinDist(s, sn - 1, t, tn - 1);
+    b = ZSTD_eDist_levenshteinDist(s, sn, t, tn - 1);
+    c = ZSTD_eDist_levenshteinDist(s, sn - 1, t, tn);
+
+    if (a > b)
+        a = b;
+    if (a > c)
+        a = c;
+    
+    return a + 1;
+}
+
+static void ZSTD_eDist_validateMatches(ZSTD_eDist_match* matches,
+                        size_t const nbMatches, const BYTE* const dict,
+                        size_t const dictSize, const BYTE* const src,
+                        size_t const srcSize)
+{
+    size_t i;
+    for (i = 0; i < nbMatches; i++) {
+        ZSTD_eDist_match match = matches[i];
+        U32 const dictIdx = match.dictIdx;
+        U32 const srcIdx = match.srcIdx;
+        U32 const matchLength = match.matchLength;
+        
+        assert(dictIdx + matchLength < dictSize);
+        assert(srcIdx + matchLength < srcSize);
+        assert(!memcmp(dict + dictIdx, src + srcIdx, matchLength));
+    }
+}
+
+/*-*************************************
+*  API
+***************************************/
+
+size_t ZSTD_eDist_genSequences(ZSTD_Sequence* sequences, 
+                        const void* dict, size_t dictSize,
+                        const void* src, size_t srcSize,
+                        int useHeuristics)
+{
+    size_t const nbDiags = dictSize + srcSize + 3;
+    S32* buffer = ZSTD_malloc(nbDiags * 2 * sizeof(S32), ZSTD_defaultCMem);
+    ZSTD_eDist_state state;
+    size_t nbSequences = 0;
+
+    state.dict = (const BYTE*)dict;
+    state.src = (const BYTE*)src;
+    state.dictSize = dictSize;
+    state.srcSize = srcSize;
+    state.forwardDiag = buffer;
+    state.backwardDiag = buffer + nbDiags;
+    state.forwardDiag += srcSize + 1;
+    state.backwardDiag += srcSize + 1;
+    state.matches = ZSTD_malloc(srcSize * sizeof(ZSTD_eDist_match), ZSTD_defaultCMem);
+    state.nbMatches = 0;
+
+    ZSTD_eDist_compare(&state, 0, dictSize, 0, srcSize, 1);
+    ZSTD_eDist_combineMatches(&state);
+    nbSequences = ZSTD_eDist_convertMatchesToSequences(sequences, &state);
+
+    ZSTD_free(buffer, ZSTD_defaultCMem);
+    ZSTD_free(state.matches, ZSTD_defaultCMem);
+
+    return nbSequences;
+}
diff --git a/contrib/pzstd/Pzstd.cpp b/contrib/pzstd/Pzstd.cpp
new file mode 100644
index 0000000..048a006
--- /dev/null
+++ b/contrib/pzstd/Pzstd.cpp
@@ -0,0 +1,626 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ */
+#include "platform.h"   /* Large Files support, SET_BINARY_MODE */
+#include "Pzstd.h"
+#include "SkippableFrame.h"
+#include "utils/FileSystem.h"
+#include "utils/Portability.h"
+#include "utils/Range.h"
+#include "utils/ScopeGuard.h"
+#include "utils/ThreadPool.h"
+#include "utils/WorkQueue.h"
+
+#include <algorithm>
+#include <chrono>
+#include <cinttypes>
+#include <cstddef>
+#include <cstdio>
+#include <memory>
+#include <string>
+
+
+namespace pzstd {
+
+namespace {
+#ifdef _WIN32
+const std::string nullOutput = "nul";
+#else
+const std::string nullOutput = "/dev/null";
+#endif
+}
+
+using std::size_t;
+
+static std::uintmax_t fileSizeOrZero(const std::string &file) {
+  if (file == "-") {
+    return 0;
+  }
+  std::error_code ec;
+  auto size = file_size(file, ec);
+  if (ec) {
+    size = 0;
+  }
+  return size;
+}
+
+static std::uint64_t handleOneInput(const Options &options,
+                             const std::string &inputFile,
+                             FILE* inputFd,
+                             const std::string &outputFile,
+                             FILE* outputFd,
+                             SharedState& state) {
+  auto inputSize = fileSizeOrZero(inputFile);
+  // WorkQueue outlives ThreadPool so in the case of error we are certain
+  // we don't accidentally try to call push() on it after it is destroyed
+  WorkQueue<std::shared_ptr<BufferWorkQueue>> outs{options.numThreads + 1};
+  std::uint64_t bytesRead;
+  std::uint64_t bytesWritten;
+  {
+    // Initialize the (de)compression thread pool with numThreads
+    ThreadPool executor(options.numThreads);
+    // Run the reader thread on an extra thread
+    ThreadPool readExecutor(1);
+    if (!options.decompress) {
+      // Add a job that reads the input and starts all the compression jobs
+      readExecutor.add(
+          [&state, &outs, &executor, inputFd, inputSize, &options, &bytesRead] {
+            bytesRead = asyncCompressChunks(
+                state,
+                outs,
+                executor,
+                inputFd,
+                inputSize,
+                options.numThreads,
+                options.determineParameters());
+          });
+      // Start writing
+      bytesWritten = writeFile(state, outs, outputFd, options.decompress);
+    } else {
+      // Add a job that reads the input and starts all the decompression jobs
+      readExecutor.add([&state, &outs, &executor, inputFd, &bytesRead] {
+        bytesRead = asyncDecompressFrames(state, outs, executor, inputFd);
+      });
+      // Start writing
+      bytesWritten = writeFile(state, outs, outputFd, options.decompress);
+    }
+  }
+  if (!state.errorHolder.hasError()) {
+    std::string inputFileName = inputFile == "-" ? "stdin" : inputFile;
+    std::string outputFileName = outputFile == "-" ? "stdout" : outputFile;
+    if (!options.decompress) {
+      double ratio = static_cast<double>(bytesWritten) /
+                     static_cast<double>(bytesRead + !bytesRead);
+      state.log(kLogInfo, "%-20s :%6.2f%%   (%6" PRIu64 " => %6" PRIu64
+                   " bytes, %s)\n",
+                   inputFileName.c_str(), ratio * 100, bytesRead, bytesWritten,
+                   outputFileName.c_str());
+    } else {
+      state.log(kLogInfo, "%-20s: %" PRIu64 " bytes \n",
+                   inputFileName.c_str(),bytesWritten);
+    }
+  }
+  return bytesWritten;
+}
+
+static FILE *openInputFile(const std::string &inputFile,
+                           ErrorHolder &errorHolder) {
+  if (inputFile == "-") {
+    SET_BINARY_MODE(stdin);
+    return stdin;
+  }
+  // Check if input file is a directory
+  {
+    std::error_code ec;
+    if (is_directory(inputFile, ec)) {
+      errorHolder.setError("Output file is a directory -- ignored");
+      return nullptr;
+    }
+  }
+  auto inputFd = std::fopen(inputFile.c_str(), "rb");
+  if (!errorHolder.check(inputFd != nullptr, "Failed to open input file")) {
+    return nullptr;
+  }
+  return inputFd;
+}
+
+static FILE *openOutputFile(const Options &options,
+                            const std::string &outputFile,
+                            SharedState& state) {
+  if (outputFile == "-") {
+    SET_BINARY_MODE(stdout);
+    return stdout;
+  }
+  // Check if the output file exists and then open it
+  if (!options.overwrite && outputFile != nullOutput) {
+    auto outputFd = std::fopen(outputFile.c_str(), "rb");
+    if (outputFd != nullptr) {
+      std::fclose(outputFd);
+      if (!state.log.logsAt(kLogInfo)) {
+        state.errorHolder.setError("Output file exists");
+        return nullptr;
+      }
+      state.log(
+          kLogInfo,
+          "pzstd: %s already exists; do you wish to overwrite (y/n) ? ",
+          outputFile.c_str());
+      int c = getchar();
+      if (c != 'y' && c != 'Y') {
+        state.errorHolder.setError("Not overwritten");
+        return nullptr;
+      }
+    }
+  }
+  auto outputFd = std::fopen(outputFile.c_str(), "wb");
+  if (!state.errorHolder.check(
+          outputFd != nullptr, "Failed to open output file")) {
+    return nullptr;
+  }
+  return outputFd;
+}
+
+int pzstdMain(const Options &options) {
+  int returnCode = 0;
+  SharedState state(options);
+  for (const auto& input : options.inputFiles) {
+    // Setup the shared state
+    auto printErrorGuard = makeScopeGuard([&] {
+      if (state.errorHolder.hasError()) {
+        returnCode = 1;
+        state.log(kLogError, "pzstd: %s: %s.\n", input.c_str(),
+                  state.errorHolder.getError().c_str());
+      }
+    });
+    // Open the input file
+    auto inputFd = openInputFile(input, state.errorHolder);
+    if (inputFd == nullptr) {
+      continue;
+    }
+    auto closeInputGuard = makeScopeGuard([&] { std::fclose(inputFd); });
+    // Open the output file
+    auto outputFile = options.getOutputFile(input);
+    if (!state.errorHolder.check(outputFile != "",
+                           "Input file does not have extension .zst")) {
+      continue;
+    }
+    auto outputFd = openOutputFile(options, outputFile, state);
+    if (outputFd == nullptr) {
+      continue;
+    }
+    auto closeOutputGuard = makeScopeGuard([&] { std::fclose(outputFd); });
+    // (de)compress the file
+    handleOneInput(options, input, inputFd, outputFile, outputFd, state);
+    if (state.errorHolder.hasError()) {
+      continue;
+    }
+    // Delete the input file if necessary
+    if (!options.keepSource) {
+      // Be sure that we are done and have written everything before we delete
+      if (!state.errorHolder.check(std::fclose(inputFd) == 0,
+                             "Failed to close input file")) {
+        continue;
+      }
+      closeInputGuard.dismiss();
+      if (!state.errorHolder.check(std::fclose(outputFd) == 0,
+                             "Failed to close output file")) {
+        continue;
+      }
+      closeOutputGuard.dismiss();
+      if (std::remove(input.c_str()) != 0) {
+        state.errorHolder.setError("Failed to remove input file");
+        continue;
+      }
+    }
+  }
+  // Returns 1 if any of the files failed to (de)compress.
+  return returnCode;
+}
+
+/// Construct a `ZSTD_inBuffer` that points to the data in `buffer`.
+static ZSTD_inBuffer makeZstdInBuffer(const Buffer& buffer) {
+  return ZSTD_inBuffer{buffer.data(), buffer.size(), 0};
+}
+
+/**
+ * Advance `buffer` and `inBuffer` by the amount of data read, as indicated by
+ * `inBuffer.pos`.
+ */
+void advance(Buffer& buffer, ZSTD_inBuffer& inBuffer) {
+  auto pos = inBuffer.pos;
+  inBuffer.src = static_cast<const unsigned char*>(inBuffer.src) + pos;
+  inBuffer.size -= pos;
+  inBuffer.pos = 0;
+  return buffer.advance(pos);
+}
+
+/// Construct a `ZSTD_outBuffer` that points to the data in `buffer`.
+static ZSTD_outBuffer makeZstdOutBuffer(Buffer& buffer) {
+  return ZSTD_outBuffer{buffer.data(), buffer.size(), 0};
+}
+
+/**
+ * Split `buffer` and advance `outBuffer` by the amount of data written, as
+ * indicated by `outBuffer.pos`.
+ */
+Buffer split(Buffer& buffer, ZSTD_outBuffer& outBuffer) {
+  auto pos = outBuffer.pos;
+  outBuffer.dst = static_cast<unsigned char*>(outBuffer.dst) + pos;
+  outBuffer.size -= pos;
+  outBuffer.pos = 0;
+  return buffer.splitAt(pos);
+}
+
+/**
+ * Stream chunks of input from `in`, compress it, and stream it out to `out`.
+ *
+ * @param state        The shared state
+ * @param in           Queue that we `pop()` input buffers from
+ * @param out          Queue that we `push()` compressed output buffers to
+ * @param maxInputSize An upper bound on the size of the input
+ */
+static void compress(
+    SharedState& state,
+    std::shared_ptr<BufferWorkQueue> in,
+    std::shared_ptr<BufferWorkQueue> out,
+    size_t maxInputSize) {
+  auto& errorHolder = state.errorHolder;
+  auto guard = makeScopeGuard([&] {
+    in->finish();
+    out->finish();
+  });
+  // Initialize the CCtx
+  auto ctx = state.cStreamPool->get();
+  if (!errorHolder.check(ctx != nullptr, "Failed to allocate ZSTD_CStream")) {
+    return;
+  }
+  {
+    auto err = ZSTD_CCtx_reset(ctx.get(), ZSTD_reset_session_only);
+    if (!errorHolder.check(!ZSTD_isError(err), ZSTD_getErrorName(err))) {
+      return;
+    }
+  }
+
+  // Allocate space for the result
+  auto outBuffer = Buffer(ZSTD_compressBound(maxInputSize));
+  auto zstdOutBuffer = makeZstdOutBuffer(outBuffer);
+  {
+    Buffer inBuffer;
+    // Read a buffer in from the input queue
+    while (in->pop(inBuffer) && !errorHolder.hasError()) {
+      auto zstdInBuffer = makeZstdInBuffer(inBuffer);
+      // Compress the whole buffer and send it to the output queue
+      while (!inBuffer.empty() && !errorHolder.hasError()) {
+        if (!errorHolder.check(
+                !outBuffer.empty(), "ZSTD_compressBound() was too small")) {
+          return;
+        }
+        // Compress
+        auto err =
+            ZSTD_compressStream(ctx.get(), &zstdOutBuffer, &zstdInBuffer);
+        if (!errorHolder.check(!ZSTD_isError(err), ZSTD_getErrorName(err))) {
+          return;
+        }
+        // Split the compressed data off outBuffer and pass to the output queue
+        out->push(split(outBuffer, zstdOutBuffer));
+        // Forget about the data we already compressed
+        advance(inBuffer, zstdInBuffer);
+      }
+    }
+  }
+  // Write the epilog
+  size_t bytesLeft;
+  do {
+    if (!errorHolder.check(
+            !outBuffer.empty(), "ZSTD_compressBound() was too small")) {
+      return;
+    }
+    bytesLeft = ZSTD_endStream(ctx.get(), &zstdOutBuffer);
+    if (!errorHolder.check(
+            !ZSTD_isError(bytesLeft), ZSTD_getErrorName(bytesLeft))) {
+      return;
+    }
+    out->push(split(outBuffer, zstdOutBuffer));
+  } while (bytesLeft != 0 && !errorHolder.hasError());
+}
+
+/**
+ * Calculates how large each independently compressed frame should be.
+ *
+ * @param size       The size of the source if known, 0 otherwise
+ * @param numThreads The number of threads available to run compression jobs on
+ * @param params     The zstd parameters to be used for compression
+ */
+static size_t calculateStep(
+    std::uintmax_t size,
+    size_t numThreads,
+    const ZSTD_parameters &params) {
+  (void)size;
+  (void)numThreads;
+  // Not validated to work correctly for window logs > 23.
+  // It will definitely fail if windowLog + 2 is >= 4GB because
+  // the skippable frame can only store sizes up to 4GB.
+  assert(params.cParams.windowLog <= 23);
+  return size_t{1} << (params.cParams.windowLog + 2);
+}
+
+namespace {
+enum class FileStatus { Continue, Done, Error };
+/// Determines the status of the file descriptor `fd`.
+FileStatus fileStatus(FILE* fd) {
+  if (std::feof(fd)) {
+    return FileStatus::Done;
+  } else if (std::ferror(fd)) {
+    return FileStatus::Error;
+  }
+  return FileStatus::Continue;
+}
+} // anonymous namespace
+
+/**
+ * Reads `size` data in chunks of `chunkSize` and puts it into `queue`.
+ * Will read less if an error or EOF occurs.
+ * Returns the status of the file after all of the reads have occurred.
+ */
+static FileStatus
+readData(BufferWorkQueue& queue, size_t chunkSize, size_t size, FILE* fd,
+         std::uint64_t *totalBytesRead) {
+  Buffer buffer(size);
+  while (!buffer.empty()) {
+    auto bytesRead =
+        std::fread(buffer.data(), 1, std::min(chunkSize, buffer.size()), fd);
+    *totalBytesRead += bytesRead;
+    queue.push(buffer.splitAt(bytesRead));
+    auto status = fileStatus(fd);
+    if (status != FileStatus::Continue) {
+      return status;
+    }
+  }
+  return FileStatus::Continue;
+}
+
+std::uint64_t asyncCompressChunks(
+    SharedState& state,
+    WorkQueue<std::shared_ptr<BufferWorkQueue>>& chunks,
+    ThreadPool& executor,
+    FILE* fd,
+    std::uintmax_t size,
+    size_t numThreads,
+    ZSTD_parameters params) {
+  auto chunksGuard = makeScopeGuard([&] { chunks.finish(); });
+  std::uint64_t bytesRead = 0;
+
+  // Break the input up into chunks of size `step` and compress each chunk
+  // independently.
+  size_t step = calculateStep(size, numThreads, params);
+  state.log(kLogDebug, "Chosen frame size: %zu\n", step);
+  auto status = FileStatus::Continue;
+  while (status == FileStatus::Continue && !state.errorHolder.hasError()) {
+    // Make a new input queue that we will put the chunk's input data into.
+    auto in = std::make_shared<BufferWorkQueue>();
+    auto inGuard = makeScopeGuard([&] { in->finish(); });
+    // Make a new output queue that compress will put the compressed data into.
+    auto out = std::make_shared<BufferWorkQueue>();
+    // Start compression in the thread pool
+    executor.add([&state, in, out, step] {
+      return compress(
+          state, std::move(in), std::move(out), step);
+    });
+    // Pass the output queue to the writer thread.
+    chunks.push(std::move(out));
+    state.log(kLogVerbose, "%s\n", "Starting a new frame");
+    // Fill the input queue for the compression job we just started
+    status = readData(*in, ZSTD_CStreamInSize(), step, fd, &bytesRead);
+  }
+  state.errorHolder.check(status != FileStatus::Error, "Error reading input");
+  return bytesRead;
+}
+
+/**
+ * Decompress a frame, whose data is streamed into `in`, and stream the output
+ * to `out`.
+ *
+ * @param state        The shared state
+ * @param in           Queue that we `pop()` input buffers from. It contains
+ *                      exactly one compressed frame.
+ * @param out          Queue that we `push()` decompressed output buffers to
+ */
+static void decompress(
+    SharedState& state,
+    std::shared_ptr<BufferWorkQueue> in,
+    std::shared_ptr<BufferWorkQueue> out) {
+  auto& errorHolder = state.errorHolder;
+  auto guard = makeScopeGuard([&] {
+    in->finish();
+    out->finish();
+  });
+  // Initialize the DCtx
+  auto ctx = state.dStreamPool->get();
+  if (!errorHolder.check(ctx != nullptr, "Failed to allocate ZSTD_DStream")) {
+    return;
+  }
+  {
+    auto err = ZSTD_DCtx_reset(ctx.get(), ZSTD_reset_session_only);
+    if (!errorHolder.check(!ZSTD_isError(err), ZSTD_getErrorName(err))) {
+      return;
+    }
+  }
+
+  const size_t outSize = ZSTD_DStreamOutSize();
+  Buffer inBuffer;
+  size_t returnCode = 0;
+  // Read a buffer in from the input queue
+  while (in->pop(inBuffer) && !errorHolder.hasError()) {
+    auto zstdInBuffer = makeZstdInBuffer(inBuffer);
+    // Decompress the whole buffer and send it to the output queue
+    while (!inBuffer.empty() && !errorHolder.hasError()) {
+      // Allocate a buffer with at least outSize bytes.
+      Buffer outBuffer(outSize);
+      auto zstdOutBuffer = makeZstdOutBuffer(outBuffer);
+      // Decompress
+      returnCode =
+          ZSTD_decompressStream(ctx.get(), &zstdOutBuffer, &zstdInBuffer);
+      if (!errorHolder.check(
+              !ZSTD_isError(returnCode), ZSTD_getErrorName(returnCode))) {
+        return;
+      }
+      // Pass the buffer with the decompressed data to the output queue
+      out->push(split(outBuffer, zstdOutBuffer));
+      // Advance past the input we already read
+      advance(inBuffer, zstdInBuffer);
+      if (returnCode == 0) {
+        // The frame is over, prepare to (maybe) start a new frame
+        ZSTD_initDStream(ctx.get());
+      }
+    }
+  }
+  if (!errorHolder.check(returnCode <= 1, "Incomplete block")) {
+    return;
+  }
+  // We've given ZSTD_decompressStream all of our data, but there may still
+  // be data to read.
+  while (returnCode == 1) {
+    // Allocate a buffer with at least outSize bytes.
+    Buffer outBuffer(outSize);
+    auto zstdOutBuffer = makeZstdOutBuffer(outBuffer);
+    // Pass in no input.
+    ZSTD_inBuffer zstdInBuffer{nullptr, 0, 0};
+    // Decompress
+    returnCode =
+        ZSTD_decompressStream(ctx.get(), &zstdOutBuffer, &zstdInBuffer);
+    if (!errorHolder.check(
+            !ZSTD_isError(returnCode), ZSTD_getErrorName(returnCode))) {
+      return;
+    }
+    // Pass the buffer with the decompressed data to the output queue
+    out->push(split(outBuffer, zstdOutBuffer));
+  }
+}
+
+std::uint64_t asyncDecompressFrames(
+    SharedState& state,
+    WorkQueue<std::shared_ptr<BufferWorkQueue>>& frames,
+    ThreadPool& executor,
+    FILE* fd) {
+  auto framesGuard = makeScopeGuard([&] { frames.finish(); });
+  std::uint64_t totalBytesRead = 0;
+
+  // Split the source up into its component frames.
+  // If we find our recognized skippable frame we know the next frames size
+  // which means that we can decompress each standard frame in independently.
+  // Otherwise, we will decompress using only one decompression task.
+  const size_t chunkSize = ZSTD_DStreamInSize();
+  auto status = FileStatus::Continue;
+  while (status == FileStatus::Continue && !state.errorHolder.hasError()) {
+    // Make a new input queue that we will put the frames's bytes into.
+    auto in = std::make_shared<BufferWorkQueue>();
+    auto inGuard = makeScopeGuard([&] { in->finish(); });
+    // Make a output queue that decompress will put the decompressed data into
+    auto out = std::make_shared<BufferWorkQueue>();
+
+    size_t frameSize;
+    {
+      // Calculate the size of the next frame.
+      // frameSize is 0 if the frame info can't be decoded.
+      Buffer buffer(SkippableFrame::kSize);
+      auto bytesRead = std::fread(buffer.data(), 1, buffer.size(), fd);
+      totalBytesRead += bytesRead;
+      status = fileStatus(fd);
+      if (bytesRead == 0 && status != FileStatus::Continue) {
+        break;
+      }
+      buffer.subtract(buffer.size() - bytesRead);
+      frameSize = SkippableFrame::tryRead(buffer.range());
+      in->push(std::move(buffer));
+    }
+    if (frameSize == 0) {
+      // We hit a non SkippableFrame, so this will be the last job.
+      // Make sure that we don't use too much memory
+      in->setMaxSize(64);
+      out->setMaxSize(64);
+    }
+    // Start decompression in the thread pool
+    executor.add([&state, in, out] {
+      return decompress(state, std::move(in), std::move(out));
+    });
+    // Pass the output queue to the writer thread
+    frames.push(std::move(out));
+    if (frameSize == 0) {
+      // We hit a non SkippableFrame ==> not compressed by pzstd or corrupted
+      // Pass the rest of the source to this decompression task
+      state.log(kLogVerbose, "%s\n",
+          "Input not in pzstd format, falling back to serial decompression");
+      while (status == FileStatus::Continue && !state.errorHolder.hasError()) {
+        status = readData(*in, chunkSize, chunkSize, fd, &totalBytesRead);
+      }
+      break;
+    }
+    state.log(kLogVerbose, "Decompressing a frame of size %zu", frameSize);
+    // Fill the input queue for the decompression job we just started
+    status = readData(*in, chunkSize, frameSize, fd, &totalBytesRead);
+  }
+  state.errorHolder.check(status != FileStatus::Error, "Error reading input");
+  return totalBytesRead;
+}
+
+/// Write `data` to `fd`, returns true iff success.
+static bool writeData(ByteRange data, FILE* fd) {
+  while (!data.empty()) {
+    data.advance(std::fwrite(data.begin(), 1, data.size(), fd));
+    if (std::ferror(fd)) {
+      return false;
+    }
+  }
+  return true;
+}
+
+std::uint64_t writeFile(
+    SharedState& state,
+    WorkQueue<std::shared_ptr<BufferWorkQueue>>& outs,
+    FILE* outputFd,
+    bool decompress) {
+  auto& errorHolder = state.errorHolder;
+  auto outsFinishGuard = makeScopeGuard([&outs] { outs.finish(); });
+  auto lineClearGuard = makeScopeGuard([&state] {
+    state.log.clear(kLogInfo);
+  });
+  std::uint64_t bytesWritten = 0;
+  std::shared_ptr<BufferWorkQueue> out;
+  // Grab the output queue for each decompression job (in order).
+  while (outs.pop(out)) {
+    auto outFinishGuard = makeScopeGuard([&out] { out->finish(); });
+    if (errorHolder.hasError()) {
+      continue;
+    }
+    if (!decompress) {
+      // If we are compressing and want to write skippable frames we can't
+      // start writing before compression is done because we need to know the
+      // compressed size.
+      // Wait for the compressed size to be available and write skippable frame
+      assert(uint64_t(out->size()) < uint64_t(1) << 32);
+      SkippableFrame frame(uint32_t(out->size()));
+      if (!writeData(frame.data(), outputFd)) {
+        errorHolder.setError("Failed to write output");
+        return bytesWritten;
+      }
+      bytesWritten += frame.kSize;
+    }
+    // For each chunk of the frame: Pop it from the queue and write it
+    Buffer buffer;
+    while (out->pop(buffer) && !errorHolder.hasError()) {
+      if (!writeData(buffer.range(), outputFd)) {
+        errorHolder.setError("Failed to write output");
+        return bytesWritten;
+      }
+      bytesWritten += buffer.size();
+      state.log.update(kLogInfo, "Written: %u MB   ",
+                static_cast<std::uint32_t>(bytesWritten >> 20));
+    }
+  }
+  return bytesWritten;
+}
+}
diff --git a/contrib/seekable_format/zstdseek_decompress.c b/contrib/seekable_format/zstdseek_decompress.c
new file mode 100644
index 0000000..ab9088a
--- /dev/null
+++ b/contrib/seekable_format/zstdseek_decompress.c
@@ -0,0 +1,600 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* *********************************************************
+*  Turn on Large Files support (>4GB) for 32-bit Linux/Unix
+***********************************************************/
+#if !defined(__64BIT__) || defined(__MINGW32__)       /* No point defining Large file for 64 bit but MinGW-w64 requires it */
+#  if !defined(_FILE_OFFSET_BITS)
+#    define _FILE_OFFSET_BITS 64                      /* turn off_t into a 64-bit type for ftello, fseeko */
+#  endif
+#  if !defined(_LARGEFILE_SOURCE)                     /* obsolete macro, replaced with _FILE_OFFSET_BITS */
+#    define _LARGEFILE_SOURCE 1                       /* Large File Support extension (LFS) - fseeko, ftello */
+#  endif
+#  if defined(_AIX) || defined(__hpux)
+#    define _LARGE_FILES                              /* Large file support on 32-bits AIX and HP-UX */
+#  endif
+#endif
+
+/* ************************************************************
+*  Detect POSIX version
+*  PLATFORM_POSIX_VERSION = 0 for non-Unix e.g. Windows
+*  PLATFORM_POSIX_VERSION = 1 for Unix-like but non-POSIX
+*  PLATFORM_POSIX_VERSION > 1 is equal to found _POSIX_VERSION
+*  Value of PLATFORM_POSIX_VERSION can be forced on command line
+***************************************************************/
+#ifndef PLATFORM_POSIX_VERSION
+
+#  if (defined(__APPLE__) && defined(__MACH__)) || defined(__SVR4) || defined(_AIX) || defined(__hpux) /* POSIX.1-2001 (SUSv3) conformant */ \
+     || defined(__DragonFly__) || defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)  /* BSD distros */
+     /* exception rule : force posix version to 200112L,
+      * note: it's better to use unistd.h's _POSIX_VERSION whenever possible */
+#    define PLATFORM_POSIX_VERSION 200112L
+
+/* try to determine posix version through official unistd.h's _POSIX_VERSION (https://pubs.opengroup.org/onlinepubs/7908799/xsh/unistd.h.html).
+ * note : there is no simple way to know in advance if <unistd.h> is present or not on target system,
+ * Posix specification mandates its presence and its content, but target system must respect this spec.
+ * It's necessary to _not_ #include <unistd.h> whenever target OS is not unix-like
+ * otherwise it will block preprocessing stage.
+ * The following list of build macros tries to "guess" if target OS is likely unix-like, and therefore can #include <unistd.h>
+ */
+#  elif !defined(_WIN32) \
+     && ( defined(__unix__) || defined(__unix) \
+       || defined(__midipix__) || defined(__VMS) || defined(__HAIKU__) )
+
+#    if defined(__linux__) || defined(__linux) || defined(__CYGWIN__)
+#      ifndef _POSIX_C_SOURCE
+#        define _POSIX_C_SOURCE 200809L  /* feature test macro : https://www.gnu.org/software/libc/manual/html_node/Feature-Test-Macros.html */
+#      endif
+#    endif
+#    include <unistd.h>  /* declares _POSIX_VERSION */
+#    if defined(_POSIX_VERSION)  /* POSIX compliant */
+#      define PLATFORM_POSIX_VERSION _POSIX_VERSION
+#    else
+#      define PLATFORM_POSIX_VERSION 1
+#    endif
+
+#    ifdef __UCLIBC__
+#     ifndef __USE_MISC
+#      define __USE_MISC /* enable st_mtim on uclibc */
+#     endif
+#    endif
+
+#  else  /* non-unix target platform (like Windows) */
+#    define PLATFORM_POSIX_VERSION 0
+#  endif
+
+#endif   /* PLATFORM_POSIX_VERSION */
+
+
+/* ************************************************************
+* Avoid fseek()'s 2GiB barrier with MSVC, macOS, *BSD, MinGW
+***************************************************************/
+#if defined(LIBC_NO_FSEEKO)
+/* Some older libc implementations don't include these functions (e.g. Bionic < 24) */
+#   define LONG_SEEK fseek
+#elif defined(_MSC_VER) && _MSC_VER >= 1400
+#   define LONG_SEEK _fseeki64
+#elif !defined(__64BIT__) && (PLATFORM_POSIX_VERSION >= 200112L) /* No point defining Large file for 64 bit */
+#   define LONG_SEEK fseeko
+#elif defined(__MINGW32__) && !defined(__STRICT_ANSI__) && !defined(__NO_MINGW_LFS) && defined(__MSVCRT__)
+#   define LONG_SEEK fseeko64
+#elif defined(_WIN32) && !defined(__DJGPP__)
+#   include <windows.h>
+    static int LONG_SEEK(FILE* file, __int64 offset, int origin) {
+        LARGE_INTEGER off;
+        DWORD method;
+        off.QuadPart = offset;
+        if (origin == SEEK_END)
+            method = FILE_END;
+        else if (origin == SEEK_CUR)
+            method = FILE_CURRENT;
+        else
+            method = FILE_BEGIN;
+
+        if (SetFilePointerEx((HANDLE) _get_osfhandle(_fileno(file)), off, NULL, method))
+            return 0;
+        else
+            return -1;
+    }
+#else
+#   define LONG_SEEK fseek
+#endif
+
+#include <stdlib.h>  /* malloc, free */
+#include <stdio.h>   /* FILE* */
+#include <limits.h>  /* UNIT_MAX */
+#include <assert.h>
+
+#define XXH_STATIC_LINKING_ONLY
+#include "xxhash.h"
+
+#define ZSTD_STATIC_LINKING_ONLY
+#include "zstd.h"
+#include "zstd_errors.h"
+#include "mem.h"
+#include "zstd_seekable.h"
+
+#undef ERROR
+#define ERROR(name) ((size_t)-ZSTD_error_##name)
+
+#define CHECK_IO(f) { int const errcod = (f); if (errcod < 0) return ERROR(seekableIO); }
+
+#undef MIN
+#undef MAX
+#define MIN(a, b) ((a) < (b) ? (a) : (b))
+#define MAX(a, b) ((a) > (b) ? (a) : (b))
+
+#define ZSTD_SEEKABLE_NO_OUTPUT_PROGRESS_MAX 16
+
+/* Special-case callbacks for FILE* and in-memory modes, so that we can treat
+ * them the same way as the advanced API */
+static int ZSTD_seekable_read_FILE(void* opaque, void* buffer, size_t n)
+{
+    size_t const result = fread(buffer, 1, n, (FILE*)opaque);
+    if (result != n) {
+        return -1;
+    }
+    return 0;
+}
+
+static int ZSTD_seekable_seek_FILE(void* opaque, long long offset, int origin)
+{
+    int const ret = LONG_SEEK((FILE*)opaque, offset, origin);
+    if (ret) return ret;
+    return fflush((FILE*)opaque);
+}
+
+typedef struct {
+    const void *ptr;
+    size_t size;
+    size_t pos;
+} buffWrapper_t;
+
+static int ZSTD_seekable_read_buff(void* opaque, void* buffer, size_t n)
+{
+    buffWrapper_t* const buff = (buffWrapper_t*)opaque;
+    assert(buff != NULL);
+    if (buff->pos + n > buff->size) return -1;
+    memcpy(buffer, (const BYTE*)buff->ptr + buff->pos, n);
+    buff->pos += n;
+    return 0;
+}
+
+static int ZSTD_seekable_seek_buff(void* opaque, long long offset, int origin)
+{
+    buffWrapper_t* const buff = (buffWrapper_t*) opaque;
+    unsigned long long newOffset;
+    assert(buff != NULL);
+    switch (origin) {
+    case SEEK_SET:
+        assert(offset >= 0);
+        newOffset = (unsigned long long)offset;
+        break;
+    case SEEK_CUR:
+        newOffset = (unsigned long long)((long long)buff->pos + offset);
+        break;
+    case SEEK_END:
+        newOffset = (unsigned long long)((long long)buff->size + offset);
+        break;
+    default:
+        assert(0);  /* not possible */
+    }
+    if (newOffset > buff->size) {
+        return -1;
+    }
+    buff->pos = newOffset;
+    return 0;
+}
+
+typedef struct {
+    U64 cOffset;
+    U64 dOffset;
+    U32 checksum;
+} seekEntry_t;
+
+struct ZSTD_seekTable_s {
+    seekEntry_t* entries;
+    size_t tableLen;
+
+    int checksumFlag;
+};
+
+#define SEEKABLE_BUFF_SIZE ZSTD_BLOCKSIZE_MAX
+
+struct ZSTD_seekable_s {
+    ZSTD_DStream* dstream;
+    ZSTD_seekTable seekTable;
+    ZSTD_seekable_customFile src;
+
+    U64 decompressedOffset;
+    U32 curFrame;
+
+    BYTE inBuff[SEEKABLE_BUFF_SIZE]; /* need to do our own input buffering */
+    BYTE outBuff[SEEKABLE_BUFF_SIZE]; /* so we can efficiently decompress the
+                                         starts of chunks before we get to the
+                                         desired section */
+    ZSTD_inBuffer in; /* maintain continuity across ZSTD_seekable_decompress operations */
+    buffWrapper_t buffWrapper; /* for `src.opaque` in in-memory mode */
+
+    XXH64_state_t xxhState;
+};
+
+ZSTD_seekable* ZSTD_seekable_create(void)
+{
+    ZSTD_seekable* const zs = (ZSTD_seekable*)malloc(sizeof(ZSTD_seekable));
+    if (zs == NULL) return NULL;
+
+    /* also initializes stage to zsds_init */
+    memset(zs, 0, sizeof(*zs));
+
+    zs->dstream = ZSTD_createDStream();
+    if (zs->dstream == NULL) {
+        free(zs);
+        return NULL;
+    }
+
+    return zs;
+}
+
+size_t ZSTD_seekable_free(ZSTD_seekable* zs)
+{
+    if (zs == NULL) return 0; /* support free on null */
+    ZSTD_freeDStream(zs->dstream);
+    free(zs->seekTable.entries);
+    free(zs);
+    return 0;
+}
+
+ZSTD_seekTable* ZSTD_seekTable_create_fromSeekable(const ZSTD_seekable* zs)
+{
+    assert(zs != NULL);
+    if (zs->seekTable.entries == NULL) return NULL;
+    ZSTD_seekTable* const st = (ZSTD_seekTable*)malloc(sizeof(ZSTD_seekTable));
+    if (st==NULL) return NULL;
+
+    st->checksumFlag = zs->seekTable.checksumFlag;
+    st->tableLen = zs->seekTable.tableLen;
+
+    /* Allocate an extra entry at the end to match logic of initial allocation */
+    size_t const entriesSize = sizeof(seekEntry_t) * (zs->seekTable.tableLen + 1);
+    seekEntry_t* const entries = (seekEntry_t*)malloc(entriesSize);
+    if (entries==NULL) {
+        free(st);
+        return NULL;
+    }
+
+    memcpy(entries, zs->seekTable.entries, entriesSize);
+    st->entries = entries;
+    return st;
+}
+
+size_t ZSTD_seekTable_free(ZSTD_seekTable* st)
+{
+    if (st == NULL) return 0; /* support free on null */
+    free(st->entries);
+    free(st);
+    return 0;
+}
+
+/** ZSTD_seekable_offsetToFrameIndex() :
+ *  Performs a binary search to find the last frame with a decompressed offset
+ *  <= pos
+ *  @return : the frame's index */
+unsigned ZSTD_seekable_offsetToFrameIndex(const ZSTD_seekable* zs, unsigned long long pos)
+{
+    return ZSTD_seekTable_offsetToFrameIndex(&zs->seekTable, pos);
+}
+
+unsigned ZSTD_seekTable_offsetToFrameIndex(const ZSTD_seekTable* st, unsigned long long pos)
+{
+    U32 lo = 0;
+    U32 hi = (U32)st->tableLen;
+    assert(st->tableLen <= UINT_MAX);
+
+    if (pos >= st->entries[st->tableLen].dOffset) {
+        return (unsigned)st->tableLen;
+    }
+
+    while (lo + 1 < hi) {
+        U32 const mid = lo + ((hi - lo) >> 1);
+        if (st->entries[mid].dOffset <= pos) {
+            lo = mid;
+        } else {
+            hi = mid;
+        }
+    }
+    return lo;
+}
+
+unsigned ZSTD_seekable_getNumFrames(const ZSTD_seekable* zs)
+{
+    return ZSTD_seekTable_getNumFrames(&zs->seekTable);
+}
+
+unsigned ZSTD_seekTable_getNumFrames(const ZSTD_seekTable* st)
+{
+    assert(st->tableLen <= UINT_MAX);
+    return (unsigned)st->tableLen;
+}
+
+unsigned long long ZSTD_seekable_getFrameCompressedOffset(const ZSTD_seekable* zs, unsigned frameIndex)
+{
+    return ZSTD_seekTable_getFrameCompressedOffset(&zs->seekTable, frameIndex);
+}
+
+unsigned long long ZSTD_seekTable_getFrameCompressedOffset(const ZSTD_seekTable* st, unsigned frameIndex)
+{
+    if (frameIndex >= st->tableLen) return ZSTD_SEEKABLE_FRAMEINDEX_TOOLARGE;
+    return st->entries[frameIndex].cOffset;
+}
+
+unsigned long long ZSTD_seekable_getFrameDecompressedOffset(const ZSTD_seekable* zs, unsigned frameIndex)
+{
+    return ZSTD_seekTable_getFrameDecompressedOffset(&zs->seekTable, frameIndex);
+}
+
+unsigned long long ZSTD_seekTable_getFrameDecompressedOffset(const ZSTD_seekTable* st, unsigned frameIndex)
+{
+    if (frameIndex >= st->tableLen) return ZSTD_SEEKABLE_FRAMEINDEX_TOOLARGE;
+    return st->entries[frameIndex].dOffset;
+}
+
+size_t ZSTD_seekable_getFrameCompressedSize(const ZSTD_seekable* zs, unsigned frameIndex)
+{
+    return ZSTD_seekTable_getFrameCompressedSize(&zs->seekTable, frameIndex);
+}
+
+size_t ZSTD_seekTable_getFrameCompressedSize(const ZSTD_seekTable* st, unsigned frameIndex)
+{
+    if (frameIndex >= st->tableLen) return ERROR(frameIndex_tooLarge);
+    return st->entries[frameIndex + 1].cOffset -
+           st->entries[frameIndex].cOffset;
+}
+
+size_t ZSTD_seekable_getFrameDecompressedSize(const ZSTD_seekable* zs, unsigned frameIndex)
+{
+    return ZSTD_seekTable_getFrameDecompressedSize(&zs->seekTable, frameIndex);
+}
+
+size_t ZSTD_seekTable_getFrameDecompressedSize(const ZSTD_seekTable* st, unsigned frameIndex)
+{
+    if (frameIndex > st->tableLen) return ERROR(frameIndex_tooLarge);
+    return st->entries[frameIndex + 1].dOffset -
+           st->entries[frameIndex].dOffset;
+}
+
+static size_t ZSTD_seekable_loadSeekTable(ZSTD_seekable* zs)
+{
+    int checksumFlag;
+    ZSTD_seekable_customFile src = zs->src;
+    /* read the footer, fixed size */
+    CHECK_IO(src.seek(src.opaque, -(int)ZSTD_seekTableFooterSize, SEEK_END));
+    CHECK_IO(src.read(src.opaque, zs->inBuff, ZSTD_seekTableFooterSize));
+
+    if (MEM_readLE32(zs->inBuff + 5) != ZSTD_SEEKABLE_MAGICNUMBER) {
+        return ERROR(prefix_unknown);
+    }
+
+    {   BYTE const sfd = zs->inBuff[4];
+        checksumFlag = sfd >> 7;
+
+        /* check reserved bits */
+        if ((sfd >> 2) & 0x1f) {
+            return ERROR(corruption_detected);
+    }   }
+
+    {   U32 const numFrames = MEM_readLE32(zs->inBuff);
+        U32 const sizePerEntry = 8 + (checksumFlag?4:0);
+        U32 const tableSize = sizePerEntry * numFrames;
+        U32 const frameSize = tableSize + ZSTD_seekTableFooterSize + ZSTD_SKIPPABLEHEADERSIZE;
+
+        U32 remaining = frameSize - ZSTD_seekTableFooterSize; /* don't need to re-read footer */
+        {   U32 const toRead = MIN(remaining, SEEKABLE_BUFF_SIZE);
+            CHECK_IO(src.seek(src.opaque, -(S64)frameSize, SEEK_END));
+            CHECK_IO(src.read(src.opaque, zs->inBuff, toRead));
+            remaining -= toRead;
+        }
+
+        if (MEM_readLE32(zs->inBuff) != (ZSTD_MAGIC_SKIPPABLE_START | 0xE)) {
+            return ERROR(prefix_unknown);
+        }
+        if (MEM_readLE32(zs->inBuff+4) + ZSTD_SKIPPABLEHEADERSIZE != frameSize) {
+            return ERROR(prefix_unknown);
+        }
+
+        {   /* Allocate an extra entry at the end so that we can do size
+             * computations on the last element without special case */
+            seekEntry_t* const entries = (seekEntry_t*)malloc(sizeof(seekEntry_t) * (numFrames + 1));
+
+            U32 idx = 0;
+            U32 pos = 8;
+
+            U64 cOffset = 0;
+            U64 dOffset = 0;
+
+            if (entries == NULL) return ERROR(memory_allocation);
+
+            /* compute cumulative positions */
+            for (; idx < numFrames; idx++) {
+                if (pos + sizePerEntry > SEEKABLE_BUFF_SIZE) {
+                    U32 const offset = SEEKABLE_BUFF_SIZE - pos;
+                    U32 const toRead = MIN(remaining, SEEKABLE_BUFF_SIZE - offset);
+                    memmove(zs->inBuff, zs->inBuff + pos, offset); /* move any data we haven't read yet */
+                    CHECK_IO(src.read(src.opaque, zs->inBuff+offset, toRead));
+                    remaining -= toRead;
+                    pos = 0;
+                }
+                entries[idx].cOffset = cOffset;
+                entries[idx].dOffset = dOffset;
+
+                cOffset += MEM_readLE32(zs->inBuff + pos);
+                pos += 4;
+                dOffset += MEM_readLE32(zs->inBuff + pos);
+                pos += 4;
+                if (checksumFlag) {
+                    entries[idx].checksum = MEM_readLE32(zs->inBuff + pos);
+                    pos += 4;
+                }
+            }
+            entries[numFrames].cOffset = cOffset;
+            entries[numFrames].dOffset = dOffset;
+
+            zs->seekTable.entries = entries;
+            zs->seekTable.tableLen = numFrames;
+            zs->seekTable.checksumFlag = checksumFlag;
+            return 0;
+        }
+    }
+}
+
+size_t ZSTD_seekable_initBuff(ZSTD_seekable* zs, const void* src, size_t srcSize)
+{
+    zs->buffWrapper = (buffWrapper_t){src, srcSize, 0};
+    {   ZSTD_seekable_customFile srcFile = {&zs->buffWrapper,
+                                            &ZSTD_seekable_read_buff,
+                                            &ZSTD_seekable_seek_buff};
+        return ZSTD_seekable_initAdvanced(zs, srcFile); }
+}
+
+size_t ZSTD_seekable_initFile(ZSTD_seekable* zs, FILE* src)
+{
+    ZSTD_seekable_customFile srcFile = {src, &ZSTD_seekable_read_FILE,
+                                        &ZSTD_seekable_seek_FILE};
+    return ZSTD_seekable_initAdvanced(zs, srcFile);
+}
+
+size_t ZSTD_seekable_initAdvanced(ZSTD_seekable* zs, ZSTD_seekable_customFile src)
+{
+    zs->src = src;
+
+    {   const size_t seekTableInit = ZSTD_seekable_loadSeekTable(zs);
+        if (ZSTD_isError(seekTableInit)) return seekTableInit; }
+
+    zs->decompressedOffset = (U64)-1;
+    zs->curFrame = (U32)-1;
+
+    {   const size_t dstreamInit = ZSTD_initDStream(zs->dstream);
+        if (ZSTD_isError(dstreamInit)) return dstreamInit; }
+    return 0;
+}
+
+size_t ZSTD_seekable_decompress(ZSTD_seekable* zs, void* dst, size_t len, unsigned long long offset)
+{
+    unsigned long long const eos = zs->seekTable.entries[zs->seekTable.tableLen].dOffset;
+    if (offset + len > eos) {
+        len = eos - offset;
+    }
+
+    U32 targetFrame = ZSTD_seekable_offsetToFrameIndex(zs, offset);
+    U32 noOutputProgressCount = 0;
+    size_t srcBytesRead = 0;
+    do {
+        /* check if we can continue from a previous decompress job */
+        if (targetFrame != zs->curFrame || offset < zs->decompressedOffset) {
+            zs->decompressedOffset = zs->seekTable.entries[targetFrame].dOffset;
+            zs->curFrame = targetFrame;
+
+            assert(zs->seekTable.entries[targetFrame].cOffset < LLONG_MAX);
+            CHECK_IO(zs->src.seek(zs->src.opaque,
+                                  (long long)zs->seekTable.entries[targetFrame].cOffset,
+                                  SEEK_SET));
+            zs->in = (ZSTD_inBuffer){zs->inBuff, 0, 0};
+            XXH64_reset(&zs->xxhState, 0);
+            ZSTD_DCtx_reset(zs->dstream, ZSTD_reset_session_only);
+            if (zs->buffWrapper.size && srcBytesRead > zs->buffWrapper.size) {
+                return ERROR(seekableIO);
+            }
+        }
+
+        while (zs->decompressedOffset < offset + len) {
+            size_t toRead;
+            ZSTD_outBuffer outTmp;
+            size_t prevOutPos;
+            size_t prevInPos;
+            size_t forwardProgress;
+            if (zs->decompressedOffset < offset) {
+                /* dummy decompressions until we get to the target offset */
+                outTmp = (ZSTD_outBuffer){zs->outBuff, (size_t) (MIN(SEEKABLE_BUFF_SIZE, offset - zs->decompressedOffset)), 0};
+            } else {
+                outTmp = (ZSTD_outBuffer){dst, len, (size_t) (zs->decompressedOffset - offset)};
+            }
+
+            prevOutPos = outTmp.pos;
+            prevInPos = zs->in.pos;
+            toRead = ZSTD_decompressStream(zs->dstream, &outTmp, &zs->in);
+            if (ZSTD_isError(toRead)) {
+                return toRead;
+            }
+
+            if (zs->seekTable.checksumFlag) {
+                XXH64_update(&zs->xxhState, (BYTE*)outTmp.dst + prevOutPos,
+                             outTmp.pos - prevOutPos);
+            }
+            forwardProgress = outTmp.pos - prevOutPos;
+            if (forwardProgress == 0) {
+                if (noOutputProgressCount++ > ZSTD_SEEKABLE_NO_OUTPUT_PROGRESS_MAX) {
+                    return ERROR(seekableIO);
+                }
+            } else {
+                noOutputProgressCount = 0;
+            }
+            zs->decompressedOffset += forwardProgress;
+            srcBytesRead += zs->in.pos - prevInPos;
+
+            if (toRead == 0) {
+                /* frame complete */
+
+                /* verify checksum */
+                if (zs->seekTable.checksumFlag &&
+                    (XXH64_digest(&zs->xxhState) & 0xFFFFFFFFU) !=
+                            zs->seekTable.entries[targetFrame].checksum) {
+                    return ERROR(corruption_detected);
+                }
+
+                if (zs->decompressedOffset < offset + len) {
+                    /* go back to the start and force a reset of the stream */
+                    targetFrame = ZSTD_seekable_offsetToFrameIndex(zs, zs->decompressedOffset);
+                    /* in this case it will fail later with corruption_detected, since last block does not have checksum */
+                    assert(targetFrame != zs->seekTable.tableLen);
+                }
+                break;
+            }
+
+            /* read in more data if we're done with this buffer */
+            if (zs->in.pos == zs->in.size) {
+                toRead = MIN(toRead, SEEKABLE_BUFF_SIZE);
+                CHECK_IO(zs->src.read(zs->src.opaque, zs->inBuff, toRead));
+                zs->in.size = toRead;
+                zs->in.pos = 0;
+            }
+        }  /* while (zs->decompressedOffset < offset + len) */
+    } while (zs->decompressedOffset != offset + len);
+
+    return len;
+}
+
+size_t ZSTD_seekable_decompressFrame(ZSTD_seekable* zs, void* dst, size_t dstSize, unsigned frameIndex)
+{
+    if (frameIndex >= zs->seekTable.tableLen) {
+        return ERROR(frameIndex_tooLarge);
+    }
+
+    {   size_t const decompressedSize =
+                zs->seekTable.entries[frameIndex + 1].dOffset -
+                zs->seekTable.entries[frameIndex].dOffset;
+        if (dstSize < decompressedSize) {
+            return ERROR(dstSize_tooSmall);
+        }
+        return ZSTD_seekable_decompress(
+                zs, dst, decompressedSize,
+                zs->seekTable.entries[frameIndex].dOffset);
+    }
+}
diff --git a/doc/educational_decoder/zstd_decompress.c b/doc/educational_decoder/zstd_decompress.c
new file mode 100644
index 0000000..839e085
--- /dev/null
+++ b/doc/educational_decoder/zstd_decompress.c
@@ -0,0 +1,2323 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/// Zstandard educational decoder implementation
+/// See https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md
+
+#include <stdint.h>   // uint8_t, etc.
+#include <stdlib.h>   // malloc, free, exit
+#include <stdio.h>    // fprintf
+#include <string.h>   // memset, memcpy
+#include "zstd_decompress.h"
+
+
+/******* IMPORTANT CONSTANTS *********************************************/
+
+// Zstandard frame
+// "Magic_Number
+// 4 Bytes, little-endian format. Value : 0xFD2FB528"
+#define ZSTD_MAGIC_NUMBER 0xFD2FB528U
+
+// The size of `Block_Content` is limited by `Block_Maximum_Size`,
+#define ZSTD_BLOCK_SIZE_MAX ((size_t)128 * 1024)
+
+// literal blocks can't be larger than their block
+#define MAX_LITERALS_SIZE ZSTD_BLOCK_SIZE_MAX
+
+
+/******* UTILITY MACROS AND TYPES *********************************************/
+#define MAX(a, b) ((a) > (b) ? (a) : (b))
+#define MIN(a, b) ((a) < (b) ? (a) : (b))
+
+#if defined(ZDEC_NO_MESSAGE)
+#define MESSAGE(...)
+#else
+#define MESSAGE(...)  fprintf(stderr, "" __VA_ARGS__)
+#endif
+
+/// This decoder calls exit(1) when it encounters an error, however a production
+/// library should propagate error codes
+#define ERROR(s)                                                               \
+    do {                                                                       \
+        MESSAGE("Error: %s\n", s);                                     \
+        exit(1);                                                               \
+    } while (0)
+#define INP_SIZE()                                                             \
+    ERROR("Input buffer smaller than it should be or input is "                \
+          "corrupted")
+#define OUT_SIZE() ERROR("Output buffer too small for output")
+#define CORRUPTION() ERROR("Corruption detected while decompressing")
+#define BAD_ALLOC() ERROR("Memory allocation error")
+#define IMPOSSIBLE() ERROR("An impossibility has occurred")
+
+typedef uint8_t  u8;
+typedef uint16_t u16;
+typedef uint32_t u32;
+typedef uint64_t u64;
+
+typedef int8_t  i8;
+typedef int16_t i16;
+typedef int32_t i32;
+typedef int64_t i64;
+/******* END UTILITY MACROS AND TYPES *****************************************/
+
+/******* IMPLEMENTATION PRIMITIVE PROTOTYPES **********************************/
+/// The implementations for these functions can be found at the bottom of this
+/// file.  They implement low-level functionality needed for the higher level
+/// decompression functions.
+
+/*** IO STREAM OPERATIONS *************/
+
+/// ostream_t/istream_t are used to wrap the pointers/length data passed into
+/// ZSTD_decompress, so that all IO operations are safely bounds checked
+/// They are written/read forward, and reads are treated as little-endian
+/// They should be used opaquely to ensure safety
+typedef struct {
+    u8 *ptr;
+    size_t len;
+} ostream_t;
+
+typedef struct {
+    const u8 *ptr;
+    size_t len;
+
+    // Input often reads a few bits at a time, so maintain an internal offset
+    int bit_offset;
+} istream_t;
+
+/// The following two functions are the only ones that allow the istream to be
+/// non-byte aligned
+
+/// Reads `num` bits from a bitstream, and updates the internal offset
+static inline u64 IO_read_bits(istream_t *const in, const int num_bits);
+/// Backs-up the stream by `num` bits so they can be read again
+static inline void IO_rewind_bits(istream_t *const in, const int num_bits);
+/// If the remaining bits in a byte will be unused, advance to the end of the
+/// byte
+static inline void IO_align_stream(istream_t *const in);
+
+/// Write the given byte into the output stream
+static inline void IO_write_byte(ostream_t *const out, u8 symb);
+
+/// Returns the number of bytes left to be read in this stream.  The stream must
+/// be byte aligned.
+static inline size_t IO_istream_len(const istream_t *const in);
+
+/// Advances the stream by `len` bytes, and returns a pointer to the chunk that
+/// was skipped.  The stream must be byte aligned.
+static inline const u8 *IO_get_read_ptr(istream_t *const in, size_t len);
+/// Advances the stream by `len` bytes, and returns a pointer to the chunk that
+/// was skipped so it can be written to.
+static inline u8 *IO_get_write_ptr(ostream_t *const out, size_t len);
+
+/// Advance the inner state by `len` bytes.  The stream must be byte aligned.
+static inline void IO_advance_input(istream_t *const in, size_t len);
+
+/// Returns an `ostream_t` constructed from the given pointer and length.
+static inline ostream_t IO_make_ostream(u8 *out, size_t len);
+/// Returns an `istream_t` constructed from the given pointer and length.
+static inline istream_t IO_make_istream(const u8 *in, size_t len);
+
+/// Returns an `istream_t` with the same base as `in`, and length `len`.
+/// Then, advance `in` to account for the consumed bytes.
+/// `in` must be byte aligned.
+static inline istream_t IO_make_sub_istream(istream_t *const in, size_t len);
+/*** END IO STREAM OPERATIONS *********/
+
+/*** BITSTREAM OPERATIONS *************/
+/// Read `num` bits (up to 64) from `src + offset`, where `offset` is in bits,
+/// and return them interpreted as a little-endian unsigned integer.
+static inline u64 read_bits_LE(const u8 *src, const int num_bits,
+                               const size_t offset);
+
+/// Read bits from the end of a HUF or FSE bitstream.  `offset` is in bits, so
+/// it updates `offset` to `offset - bits`, and then reads `bits` bits from
+/// `src + offset`.  If the offset becomes negative, the extra bits at the
+/// bottom are filled in with `0` bits instead of reading from before `src`.
+static inline u64 STREAM_read_bits(const u8 *src, const int bits,
+                                   i64 *const offset);
+/*** END BITSTREAM OPERATIONS *********/
+
+/*** BIT COUNTING OPERATIONS **********/
+/// Returns the index of the highest set bit in `num`, or `-1` if `num == 0`
+static inline int highest_set_bit(const u64 num);
+/*** END BIT COUNTING OPERATIONS ******/
+
+/*** HUFFMAN PRIMITIVES ***************/
+// Table decode method uses exponential memory, so we need to limit depth
+#define HUF_MAX_BITS (16)
+
+// Limit the maximum number of symbols to 256 so we can store a symbol in a byte
+#define HUF_MAX_SYMBS (256)
+
+/// Structure containing all tables necessary for efficient Huffman decoding
+typedef struct {
+    u8 *symbols;
+    u8 *num_bits;
+    int max_bits;
+} HUF_dtable;
+
+/// Decode a single symbol and read in enough bits to refresh the state
+static inline u8 HUF_decode_symbol(const HUF_dtable *const dtable,
+                                   u16 *const state, const u8 *const src,
+                                   i64 *const offset);
+/// Read in a full state's worth of bits to initialize it
+static inline void HUF_init_state(const HUF_dtable *const dtable,
+                                  u16 *const state, const u8 *const src,
+                                  i64 *const offset);
+
+/// Decompresses a single Huffman stream, returns the number of bytes decoded.
+/// `src_len` must be the exact length of the Huffman-coded block.
+static size_t HUF_decompress_1stream(const HUF_dtable *const dtable,
+                                     ostream_t *const out, istream_t *const in);
+/// Same as previous but decodes 4 streams, formatted as in the Zstandard
+/// specification.
+/// `src_len` must be the exact length of the Huffman-coded block.
+static size_t HUF_decompress_4stream(const HUF_dtable *const dtable,
+                                     ostream_t *const out, istream_t *const in);
+
+/// Initialize a Huffman decoding table using the table of bit counts provided
+static void HUF_init_dtable(HUF_dtable *const table, const u8 *const bits,
+                            const int num_symbs);
+/// Initialize a Huffman decoding table using the table of weights provided
+/// Weights follow the definition provided in the Zstandard specification
+static void HUF_init_dtable_usingweights(HUF_dtable *const table,
+                                         const u8 *const weights,
+                                         const int num_symbs);
+
+/// Free the malloc'ed parts of a decoding table
+static void HUF_free_dtable(HUF_dtable *const dtable);
+/*** END HUFFMAN PRIMITIVES ***********/
+
+/*** FSE PRIMITIVES *******************/
+/// For more description of FSE see
+/// https://github.com/Cyan4973/FiniteStateEntropy/
+
+// FSE table decoding uses exponential memory, so limit the maximum accuracy
+#define FSE_MAX_ACCURACY_LOG (15)
+// Limit the maximum number of symbols so they can be stored in a single byte
+#define FSE_MAX_SYMBS (256)
+
+/// The tables needed to decode FSE encoded streams
+typedef struct {
+    u8 *symbols;
+    u8 *num_bits;
+    u16 *new_state_base;
+    int accuracy_log;
+} FSE_dtable;
+
+/// Return the symbol for the current state
+static inline u8 FSE_peek_symbol(const FSE_dtable *const dtable,
+                                 const u16 state);
+/// Read the number of bits necessary to update state, update, and shift offset
+/// back to reflect the bits read
+static inline void FSE_update_state(const FSE_dtable *const dtable,
+                                    u16 *const state, const u8 *const src,
+                                    i64 *const offset);
+
+/// Combine peek and update: decode a symbol and update the state
+static inline u8 FSE_decode_symbol(const FSE_dtable *const dtable,
+                                   u16 *const state, const u8 *const src,
+                                   i64 *const offset);
+
+/// Read bits from the stream to initialize the state and shift offset back
+static inline void FSE_init_state(const FSE_dtable *const dtable,
+                                  u16 *const state, const u8 *const src,
+                                  i64 *const offset);
+
+/// Decompress two interleaved bitstreams (e.g. compressed Huffman weights)
+/// using an FSE decoding table.  `src_len` must be the exact length of the
+/// block.
+static size_t FSE_decompress_interleaved2(const FSE_dtable *const dtable,
+                                          ostream_t *const out,
+                                          istream_t *const in);
+
+/// Initialize a decoding table using normalized frequencies.
+static void FSE_init_dtable(FSE_dtable *const dtable,
+                            const i16 *const norm_freqs, const int num_symbs,
+                            const int accuracy_log);
+
+/// Decode an FSE header as defined in the Zstandard format specification and
+/// use the decoded frequencies to initialize a decoding table.
+static void FSE_decode_header(FSE_dtable *const dtable, istream_t *const in,
+                                const int max_accuracy_log);
+
+/// Initialize an FSE table that will always return the same symbol and consume
+/// 0 bits per symbol, to be used for RLE mode in sequence commands
+static void FSE_init_dtable_rle(FSE_dtable *const dtable, const u8 symb);
+
+/// Free the malloc'ed parts of a decoding table
+static void FSE_free_dtable(FSE_dtable *const dtable);
+/*** END FSE PRIMITIVES ***************/
+
+/******* END IMPLEMENTATION PRIMITIVE PROTOTYPES ******************************/
+
+/******* ZSTD HELPER STRUCTS AND PROTOTYPES ***********************************/
+
+/// A small structure that can be reused in various places that need to access
+/// frame header information
+typedef struct {
+    // The size of window that we need to be able to contiguously store for
+    // references
+    size_t window_size;
+    // The total output size of this compressed frame
+    size_t frame_content_size;
+
+    // The dictionary id if this frame uses one
+    u32 dictionary_id;
+
+    // Whether or not the content of this frame has a checksum
+    int content_checksum_flag;
+    // Whether or not the output for this frame is in a single segment
+    int single_segment_flag;
+} frame_header_t;
+
+/// The context needed to decode blocks in a frame
+typedef struct {
+    frame_header_t header;
+
+    // The total amount of data available for backreferences, to determine if an
+    // offset too large to be correct
+    size_t current_total_output;
+
+    const u8 *dict_content;
+    size_t dict_content_len;
+
+    // Entropy encoding tables so they can be repeated by future blocks instead
+    // of retransmitting
+    HUF_dtable literals_dtable;
+    FSE_dtable ll_dtable;
+    FSE_dtable ml_dtable;
+    FSE_dtable of_dtable;
+
+    // The last 3 offsets for the special "repeat offsets".
+    u64 previous_offsets[3];
+} frame_context_t;
+
+/// The decoded contents of a dictionary so that it doesn't have to be repeated
+/// for each frame that uses it
+struct dictionary_s {
+    // Entropy tables
+    HUF_dtable literals_dtable;
+    FSE_dtable ll_dtable;
+    FSE_dtable ml_dtable;
+    FSE_dtable of_dtable;
+
+    // Raw content for backreferences
+    u8 *content;
+    size_t content_size;
+
+    // Offset history to prepopulate the frame's history
+    u64 previous_offsets[3];
+
+    u32 dictionary_id;
+};
+
+/// A tuple containing the parts necessary to decode and execute a ZSTD sequence
+/// command
+typedef struct {
+    u32 literal_length;
+    u32 match_length;
+    u32 offset;
+} sequence_command_t;
+
+/// The decoder works top-down, starting at the high level like Zstd frames, and
+/// working down to lower more technical levels such as blocks, literals, and
+/// sequences.  The high-level functions roughly follow the outline of the
+/// format specification:
+/// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md
+
+/// Before the implementation of each high-level function declared here, the
+/// prototypes for their helper functions are defined and explained
+
+/// Decode a single Zstd frame, or error if the input is not a valid frame.
+/// Accepts a dict argument, which may be NULL indicating no dictionary.
+/// See
+/// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#frame-concatenation
+static void decode_frame(ostream_t *const out, istream_t *const in,
+                         const dictionary_t *const dict);
+
+// Decode data in a compressed block
+static void decompress_block(frame_context_t *const ctx, ostream_t *const out,
+                             istream_t *const in);
+
+// Decode the literals section of a block
+static size_t decode_literals(frame_context_t *const ctx, istream_t *const in,
+                              u8 **const literals);
+
+// Decode the sequences part of a block
+static size_t decode_sequences(frame_context_t *const ctx, istream_t *const in,
+                               sequence_command_t **const sequences);
+
+// Execute the decoded sequences on the literals block
+static void execute_sequences(frame_context_t *const ctx, ostream_t *const out,
+                              const u8 *const literals,
+                              const size_t literals_len,
+                              const sequence_command_t *const sequences,
+                              const size_t num_sequences);
+
+// Copies literals and returns the total literal length that was copied
+static u32 copy_literals(const size_t seq, istream_t *litstream,
+                         ostream_t *const out);
+
+// Given an offset code from a sequence command (either an actual offset value
+// or an index for previous offset), computes the correct offset and updates
+// the offset history
+static size_t compute_offset(sequence_command_t seq, u64 *const offset_hist);
+
+// Given an offset, match length, and total output, as well as the frame
+// context for the dictionary, determines if the dictionary is used and
+// executes the copy operation
+static void execute_match_copy(frame_context_t *const ctx, size_t offset,
+                              size_t match_length, size_t total_output,
+                              ostream_t *const out);
+
+/******* END ZSTD HELPER STRUCTS AND PROTOTYPES *******************************/
+
+size_t ZSTD_decompress(void *const dst, const size_t dst_len,
+                       const void *const src, const size_t src_len) {
+    dictionary_t* const uninit_dict = create_dictionary();
+    size_t const decomp_size = ZSTD_decompress_with_dict(dst, dst_len, src,
+                                                         src_len, uninit_dict);
+    free_dictionary(uninit_dict);
+    return decomp_size;
+}
+
+size_t ZSTD_decompress_with_dict(void *const dst, const size_t dst_len,
+                                 const void *const src, const size_t src_len,
+                                 dictionary_t* parsed_dict) {
+
+    istream_t in = IO_make_istream(src, src_len);
+    ostream_t out = IO_make_ostream(dst, dst_len);
+
+    // "A content compressed by Zstandard is transformed into a Zstandard frame.
+    // Multiple frames can be appended into a single file or stream. A frame is
+    // totally independent, has a defined beginning and end, and a set of
+    // parameters which tells the decoder how to decompress it."
+
+    /* this decoder assumes decompression of a single frame */
+    decode_frame(&out, &in, parsed_dict);
+
+    return (size_t)(out.ptr - (u8 *)dst);
+}
+
+/******* FRAME DECODING ******************************************************/
+
+static void decode_data_frame(ostream_t *const out, istream_t *const in,
+                              const dictionary_t *const dict);
+static void init_frame_context(frame_context_t *const context,
+                               istream_t *const in,
+                               const dictionary_t *const dict);
+static void free_frame_context(frame_context_t *const context);
+static void parse_frame_header(frame_header_t *const header,
+                               istream_t *const in);
+static void frame_context_apply_dict(frame_context_t *const ctx,
+                                     const dictionary_t *const dict);
+
+static void decompress_data(frame_context_t *const ctx, ostream_t *const out,
+                            istream_t *const in);
+
+static void decode_frame(ostream_t *const out, istream_t *const in,
+                         const dictionary_t *const dict) {
+    const u32 magic_number = (u32)IO_read_bits(in, 32);
+    if (magic_number == ZSTD_MAGIC_NUMBER) {
+        // ZSTD frame
+        decode_data_frame(out, in, dict);
+
+        return;
+    }
+
+    // not a real frame or a skippable frame
+    ERROR("Tried to decode non-ZSTD frame");
+}
+
+/// Decode a frame that contains compressed data.  Not all frames do as there
+/// are skippable frames.
+/// See
+/// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#general-structure-of-zstandard-frame-format
+static void decode_data_frame(ostream_t *const out, istream_t *const in,
+                              const dictionary_t *const dict) {
+    frame_context_t ctx;
+
+    // Initialize the context that needs to be carried from block to block
+    init_frame_context(&ctx, in, dict);
+
+    if (ctx.header.frame_content_size != 0 &&
+        ctx.header.frame_content_size > out->len) {
+        OUT_SIZE();
+    }
+
+    decompress_data(&ctx, out, in);
+
+    free_frame_context(&ctx);
+}
+
+/// Takes the information provided in the header and dictionary, and initializes
+/// the context for this frame
+static void init_frame_context(frame_context_t *const context,
+                               istream_t *const in,
+                               const dictionary_t *const dict) {
+    // Most fields in context are correct when initialized to 0
+    memset(context, 0, sizeof(frame_context_t));
+
+    // Parse data from the frame header
+    parse_frame_header(&context->header, in);
+
+    // Set up the offset history for the repeat offset commands
+    context->previous_offsets[0] = 1;
+    context->previous_offsets[1] = 4;
+    context->previous_offsets[2] = 8;
+
+    // Apply details from the dict if it exists
+    frame_context_apply_dict(context, dict);
+}
+
+static void free_frame_context(frame_context_t *const context) {
+    HUF_free_dtable(&context->literals_dtable);
+
+    FSE_free_dtable(&context->ll_dtable);
+    FSE_free_dtable(&context->ml_dtable);
+    FSE_free_dtable(&context->of_dtable);
+
+    memset(context, 0, sizeof(frame_context_t));
+}
+
+static void parse_frame_header(frame_header_t *const header,
+                               istream_t *const in) {
+    // "The first header's byte is called the Frame_Header_Descriptor. It tells
+    // which other fields are present. Decoding this byte is enough to tell the
+    // size of Frame_Header.
+    //
+    // Bit number   Field name
+    // 7-6  Frame_Content_Size_flag
+    // 5    Single_Segment_flag
+    // 4    Unused_bit
+    // 3    Reserved_bit
+    // 2    Content_Checksum_flag
+    // 1-0  Dictionary_ID_flag"
+    const u8 descriptor = (u8)IO_read_bits(in, 8);
+
+    // decode frame header descriptor into flags
+    const u8 frame_content_size_flag = descriptor >> 6;
+    const u8 single_segment_flag = (descriptor >> 5) & 1;
+    const u8 reserved_bit = (descriptor >> 3) & 1;
+    const u8 content_checksum_flag = (descriptor >> 2) & 1;
+    const u8 dictionary_id_flag = descriptor & 3;
+
+    if (reserved_bit != 0) {
+        CORRUPTION();
+    }
+
+    header->single_segment_flag = single_segment_flag;
+    header->content_checksum_flag = content_checksum_flag;
+
+    // decode window size
+    if (!single_segment_flag) {
+        // "Provides guarantees on maximum back-reference distance that will be
+        // used within compressed data. This information is important for
+        // decoders to allocate enough memory.
+        //
+        // Bit numbers  7-3         2-0
+        // Field name   Exponent    Mantissa"
+        u8 window_descriptor = (u8)IO_read_bits(in, 8);
+        u8 exponent = window_descriptor >> 3;
+        u8 mantissa = window_descriptor & 7;
+
+        // Use the algorithm from the specification to compute window size
+        // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#window_descriptor
+        size_t window_base = (size_t)1 << (10 + exponent);
+        size_t window_add = (window_base / 8) * mantissa;
+        header->window_size = window_base + window_add;
+    }
+
+    // decode dictionary id if it exists
+    if (dictionary_id_flag) {
+        // "This is a variable size field, which contains the ID of the
+        // dictionary required to properly decode the frame. Note that this
+        // field is optional. When it's not present, it's up to the caller to
+        // make sure it uses the correct dictionary. Format is little-endian."
+        const int bytes_array[] = {0, 1, 2, 4};
+        const int bytes = bytes_array[dictionary_id_flag];
+
+        header->dictionary_id = (u32)IO_read_bits(in, bytes * 8);
+    } else {
+        header->dictionary_id = 0;
+    }
+
+    // decode frame content size if it exists
+    if (single_segment_flag || frame_content_size_flag) {
+        // "This is the original (uncompressed) size. This information is
+        // optional. The Field_Size is provided according to value of
+        // Frame_Content_Size_flag. The Field_Size can be equal to 0 (not
+        // present), 1, 2, 4 or 8 bytes. Format is little-endian."
+        //
+        // if frame_content_size_flag == 0 but single_segment_flag is set, we
+        // still have a 1 byte field
+        const int bytes_array[] = {1, 2, 4, 8};
+        const int bytes = bytes_array[frame_content_size_flag];
+
+        header->frame_content_size = IO_read_bits(in, bytes * 8);
+        if (bytes == 2) {
+            // "When Field_Size is 2, the offset of 256 is added."
+            header->frame_content_size += 256;
+        }
+    } else {
+        header->frame_content_size = 0;
+    }
+
+    if (single_segment_flag) {
+        // "The Window_Descriptor byte is optional. It is absent when
+        // Single_Segment_flag is set. In this case, the maximum back-reference
+        // distance is the content size itself, which can be any value from 1 to
+        // 2^64-1 bytes (16 EB)."
+        header->window_size = header->frame_content_size;
+    }
+}
+
+/// Decompress the data from a frame block by block
+static void decompress_data(frame_context_t *const ctx, ostream_t *const out,
+                            istream_t *const in) {
+    // "A frame encapsulates one or multiple blocks. Each block can be
+    // compressed or not, and has a guaranteed maximum content size, which
+    // depends on frame parameters. Unlike frames, each block depends on
+    // previous blocks for proper decoding. However, each block can be
+    // decompressed without waiting for its successor, allowing streaming
+    // operations."
+    int last_block = 0;
+    do {
+        // "Last_Block
+        //
+        // The lowest bit signals if this block is the last one. Frame ends
+        // right after this block.
+        //
+        // Block_Type and Block_Size
+        //
+        // The next 2 bits represent the Block_Type, while the remaining 21 bits
+        // represent the Block_Size. Format is little-endian."
+        last_block = (int)IO_read_bits(in, 1);
+        const int block_type = (int)IO_read_bits(in, 2);
+        const size_t block_len = IO_read_bits(in, 21);
+
+        switch (block_type) {
+        case 0: {
+            // "Raw_Block - this is an uncompressed block. Block_Size is the
+            // number of bytes to read and copy."
+            const u8 *const read_ptr = IO_get_read_ptr(in, block_len);
+            u8 *const write_ptr = IO_get_write_ptr(out, block_len);
+
+            // Copy the raw data into the output
+            memcpy(write_ptr, read_ptr, block_len);
+
+            ctx->current_total_output += block_len;
+            break;
+        }
+        case 1: {
+            // "RLE_Block - this is a single byte, repeated N times. In which
+            // case, Block_Size is the size to regenerate, while the
+            // "compressed" block is just 1 byte (the byte to repeat)."
+            const u8 *const read_ptr = IO_get_read_ptr(in, 1);
+            u8 *const write_ptr = IO_get_write_ptr(out, block_len);
+
+            // Copy `block_len` copies of `read_ptr[0]` to the output
+            memset(write_ptr, read_ptr[0], block_len);
+
+            ctx->current_total_output += block_len;
+            break;
+        }
+        case 2: {
+            // "Compressed_Block - this is a Zstandard compressed block,
+            // detailed in another section of this specification. Block_Size is
+            // the compressed size.
+
+            // Create a sub-stream for the block
+            istream_t block_stream = IO_make_sub_istream(in, block_len);
+            decompress_block(ctx, out, &block_stream);
+            break;
+        }
+        case 3:
+            // "Reserved - this is not a block. This value cannot be used with
+            // current version of this specification."
+            CORRUPTION();
+            break;
+        default:
+            IMPOSSIBLE();
+        }
+    } while (!last_block);
+
+    if (ctx->header.content_checksum_flag) {
+        // This program does not support checking the checksum, so skip over it
+        // if it's present
+        IO_advance_input(in, 4);
+    }
+}
+/******* END FRAME DECODING ***************************************************/
+
+/******* BLOCK DECOMPRESSION **************************************************/
+static void decompress_block(frame_context_t *const ctx, ostream_t *const out,
+                             istream_t *const in) {
+    // "A compressed block consists of 2 sections :
+    //
+    // Literals_Section
+    // Sequences_Section"
+
+
+    // Part 1: decode the literals block
+    u8 *literals = NULL;
+    const size_t literals_size = decode_literals(ctx, in, &literals);
+
+    // Part 2: decode the sequences block
+    sequence_command_t *sequences = NULL;
+    const size_t num_sequences =
+        decode_sequences(ctx, in, &sequences);
+
+    // Part 3: combine literals and sequence commands to generate output
+    execute_sequences(ctx, out, literals, literals_size, sequences,
+                      num_sequences);
+    free(literals);
+    free(sequences);
+}
+/******* END BLOCK DECOMPRESSION **********************************************/
+
+/******* LITERALS DECODING ****************************************************/
+static size_t decode_literals_simple(istream_t *const in, u8 **const literals,
+                                     const int block_type,
+                                     const int size_format);
+static size_t decode_literals_compressed(frame_context_t *const ctx,
+                                         istream_t *const in,
+                                         u8 **const literals,
+                                         const int block_type,
+                                         const int size_format);
+static void decode_huf_table(HUF_dtable *const dtable, istream_t *const in);
+static void fse_decode_hufweights(ostream_t *weights, istream_t *const in,
+                                    int *const num_symbs);
+
+static size_t decode_literals(frame_context_t *const ctx, istream_t *const in,
+                              u8 **const literals) {
+    // "Literals can be stored uncompressed or compressed using Huffman prefix
+    // codes. When compressed, an optional tree description can be present,
+    // followed by 1 or 4 streams."
+    //
+    // "Literals_Section_Header
+    //
+    // Header is in charge of describing how literals are packed. It's a
+    // byte-aligned variable-size bitfield, ranging from 1 to 5 bytes, using
+    // little-endian convention."
+    //
+    // "Literals_Block_Type
+    //
+    // This field uses 2 lowest bits of first byte, describing 4 different block
+    // types"
+    //
+    // size_format takes between 1 and 2 bits
+    int block_type = (int)IO_read_bits(in, 2);
+    int size_format = (int)IO_read_bits(in, 2);
+
+    if (block_type <= 1) {
+        // Raw or RLE literals block
+        return decode_literals_simple(in, literals, block_type,
+                                      size_format);
+    } else {
+        // Huffman compressed literals
+        return decode_literals_compressed(ctx, in, literals, block_type,
+                                          size_format);
+    }
+}
+
+/// Decodes literals blocks in raw or RLE form
+static size_t decode_literals_simple(istream_t *const in, u8 **const literals,
+                                     const int block_type,
+                                     const int size_format) {
+    size_t size;
+    switch (size_format) {
+    // These cases are in the form ?0
+    // In this case, the ? bit is actually part of the size field
+    case 0:
+    case 2:
+        // "Size_Format uses 1 bit. Regenerated_Size uses 5 bits (0-31)."
+        IO_rewind_bits(in, 1);
+        size = IO_read_bits(in, 5);
+        break;
+    case 1:
+        // "Size_Format uses 2 bits. Regenerated_Size uses 12 bits (0-4095)."
+        size = IO_read_bits(in, 12);
+        break;
+    case 3:
+        // "Size_Format uses 2 bits. Regenerated_Size uses 20 bits (0-1048575)."
+        size = IO_read_bits(in, 20);
+        break;
+    default:
+        // Size format is in range 0-3
+        IMPOSSIBLE();
+    }
+
+    if (size > MAX_LITERALS_SIZE) {
+        CORRUPTION();
+    }
+
+    *literals = malloc(size);
+    if (!*literals) {
+        BAD_ALLOC();
+    }
+
+    switch (block_type) {
+    case 0: {
+        // "Raw_Literals_Block - Literals are stored uncompressed."
+        const u8 *const read_ptr = IO_get_read_ptr(in, size);
+        memcpy(*literals, read_ptr, size);
+        break;
+    }
+    case 1: {
+        // "RLE_Literals_Block - Literals consist of a single byte value repeated N times."
+        const u8 *const read_ptr = IO_get_read_ptr(in, 1);
+        memset(*literals, read_ptr[0], size);
+        break;
+    }
+    default:
+        IMPOSSIBLE();
+    }
+
+    return size;
+}
+
+/// Decodes Huffman compressed literals
+static size_t decode_literals_compressed(frame_context_t *const ctx,
+                                         istream_t *const in,
+                                         u8 **const literals,
+                                         const int block_type,
+                                         const int size_format) {
+    size_t regenerated_size, compressed_size;
+    // Only size_format=0 has 1 stream, so default to 4
+    int num_streams = 4;
+    switch (size_format) {
+    case 0:
+        // "A single stream. Both Compressed_Size and Regenerated_Size use 10
+        // bits (0-1023)."
+        num_streams = 1;
+    // Fall through as it has the same size format
+        /* fallthrough */
+    case 1:
+        // "4 streams. Both Compressed_Size and Regenerated_Size use 10 bits
+        // (0-1023)."
+        regenerated_size = IO_read_bits(in, 10);
+        compressed_size = IO_read_bits(in, 10);
+        break;
+    case 2:
+        // "4 streams. Both Compressed_Size and Regenerated_Size use 14 bits
+        // (0-16383)."
+        regenerated_size = IO_read_bits(in, 14);
+        compressed_size = IO_read_bits(in, 14);
+        break;
+    case 3:
+        // "4 streams. Both Compressed_Size and Regenerated_Size use 18 bits
+        // (0-262143)."
+        regenerated_size = IO_read_bits(in, 18);
+        compressed_size = IO_read_bits(in, 18);
+        break;
+    default:
+        // Impossible
+        IMPOSSIBLE();
+    }
+    if (regenerated_size > MAX_LITERALS_SIZE) {
+        CORRUPTION();
+    }
+
+    *literals = malloc(regenerated_size);
+    if (!*literals) {
+        BAD_ALLOC();
+    }
+
+    ostream_t lit_stream = IO_make_ostream(*literals, regenerated_size);
+    istream_t huf_stream = IO_make_sub_istream(in, compressed_size);
+
+    if (block_type == 2) {
+        // Decode the provided Huffman table
+        // "This section is only present when Literals_Block_Type type is
+        // Compressed_Literals_Block (2)."
+
+        HUF_free_dtable(&ctx->literals_dtable);
+        decode_huf_table(&ctx->literals_dtable, &huf_stream);
+    } else {
+        // If the previous Huffman table is being repeated, ensure it exists
+        if (!ctx->literals_dtable.symbols) {
+            CORRUPTION();
+        }
+    }
+
+    size_t symbols_decoded;
+    if (num_streams == 1) {
+        symbols_decoded = HUF_decompress_1stream(&ctx->literals_dtable, &lit_stream, &huf_stream);
+    } else {
+        symbols_decoded = HUF_decompress_4stream(&ctx->literals_dtable, &lit_stream, &huf_stream);
+    }
+
+    if (symbols_decoded != regenerated_size) {
+        CORRUPTION();
+    }
+
+    return regenerated_size;
+}
+
+// Decode the Huffman table description
+static void decode_huf_table(HUF_dtable *const dtable, istream_t *const in) {
+    // "All literal values from zero (included) to last present one (excluded)
+    // are represented by Weight with values from 0 to Max_Number_of_Bits."
+
+    // "This is a single byte value (0-255), which describes how to decode the list of weights."
+    const u8 header = IO_read_bits(in, 8);
+
+    u8 weights[HUF_MAX_SYMBS];
+    memset(weights, 0, sizeof(weights));
+
+    int num_symbs;
+
+    if (header >= 128) {
+        // "This is a direct representation, where each Weight is written
+        // directly as a 4 bits field (0-15). The full representation occupies
+        // ((Number_of_Symbols+1)/2) bytes, meaning it uses a last full byte
+        // even if Number_of_Symbols is odd. Number_of_Symbols = headerByte -
+        // 127"
+        num_symbs = header - 127;
+        const size_t bytes = (num_symbs + 1) / 2;
+
+        const u8 *const weight_src = IO_get_read_ptr(in, bytes);
+
+        for (int i = 0; i < num_symbs; i++) {
+            // "They are encoded forward, 2
+            // weights to a byte with the first weight taking the top four bits
+            // and the second taking the bottom four (e.g. the following
+            // operations could be used to read the weights: Weight[0] =
+            // (Byte[0] >> 4), Weight[1] = (Byte[0] & 0xf), etc.)."
+            if (i % 2 == 0) {
+                weights[i] = weight_src[i / 2] >> 4;
+            } else {
+                weights[i] = weight_src[i / 2] & 0xf;
+            }
+        }
+    } else {
+        // The weights are FSE encoded, decode them before we can construct the
+        // table
+        istream_t fse_stream = IO_make_sub_istream(in, header);
+        ostream_t weight_stream = IO_make_ostream(weights, HUF_MAX_SYMBS);
+        fse_decode_hufweights(&weight_stream, &fse_stream, &num_symbs);
+    }
+
+    // Construct the table using the decoded weights
+    HUF_init_dtable_usingweights(dtable, weights, num_symbs);
+}
+
+static void fse_decode_hufweights(ostream_t *weights, istream_t *const in,
+                                    int *const num_symbs) {
+    const int MAX_ACCURACY_LOG = 7;
+
+    FSE_dtable dtable;
+
+    // "An FSE bitstream starts by a header, describing probabilities
+    // distribution. It will create a Decoding Table. For a list of Huffman
+    // weights, maximum accuracy is 7 bits."
+    FSE_decode_header(&dtable, in, MAX_ACCURACY_LOG);
+
+    // Decode the weights
+    *num_symbs = FSE_decompress_interleaved2(&dtable, weights, in);
+
+    FSE_free_dtable(&dtable);
+}
+/******* END LITERALS DECODING ************************************************/
+
+/******* SEQUENCE DECODING ****************************************************/
+/// The combination of FSE states needed to decode sequences
+typedef struct {
+    FSE_dtable ll_table;
+    FSE_dtable of_table;
+    FSE_dtable ml_table;
+
+    u16 ll_state;
+    u16 of_state;
+    u16 ml_state;
+} sequence_states_t;
+
+/// Different modes to signal to decode_seq_tables what to do
+typedef enum {
+    seq_literal_length = 0,
+    seq_offset = 1,
+    seq_match_length = 2,
+} seq_part_t;
+
+typedef enum {
+    seq_predefined = 0,
+    seq_rle = 1,
+    seq_fse = 2,
+    seq_repeat = 3,
+} seq_mode_t;
+
+/// The predefined FSE distribution tables for `seq_predefined` mode
+static const i16 SEQ_LITERAL_LENGTH_DEFAULT_DIST[36] = {
+    4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1,  1,  2,  2,
+    2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1, -1, -1, -1, -1};
+static const i16 SEQ_OFFSET_DEFAULT_DIST[29] = {
+    1, 1, 1, 1, 1, 1, 2, 2, 2, 1,  1,  1,  1,  1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, -1, -1, -1, -1, -1};
+static const i16 SEQ_MATCH_LENGTH_DEFAULT_DIST[53] = {
+    1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1,  1,  1,  1,  1,  1,  1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  1,  1,  1,  1,  1,  1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, -1, -1, -1, -1, -1, -1};
+
+/// The sequence decoding baseline and number of additional bits to read/add
+/// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#the-codes-for-literals-lengths-match-lengths-and-offsets
+static const u32 SEQ_LITERAL_LENGTH_BASELINES[36] = {
+    0,  1,  2,   3,   4,   5,    6,    7,    8,    9,     10,    11,
+    12, 13, 14,  15,  16,  18,   20,   22,   24,   28,    32,    40,
+    48, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768, 65536};
+static const u8 SEQ_LITERAL_LENGTH_EXTRA_BITS[36] = {
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  0,  0,  0,  0,  1,  1,
+    1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16};
+
+static const u32 SEQ_MATCH_LENGTH_BASELINES[53] = {
+    3,  4,   5,   6,   7,    8,    9,    10,   11,    12,    13,   14, 15, 16,
+    17, 18,  19,  20,  21,   22,   23,   24,   25,    26,    27,   28, 29, 30,
+    31, 32,  33,  34,  35,   37,   39,   41,   43,    47,    51,   59, 67, 83,
+    99, 131, 259, 515, 1027, 2051, 4099, 8195, 16387, 32771, 65539};
+static const u8 SEQ_MATCH_LENGTH_EXTRA_BITS[53] = {
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  0,  0,  0,  0,  0,  0, 0,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  0,  0,  0,  1,  1,  1, 1,
+    2, 2, 3, 3, 4, 4, 5, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16};
+
+/// Offset decoding is simpler so we just need a maximum code value
+static const u8 SEQ_MAX_CODES[3] = {35, (u8)-1, 52};
+
+static void decompress_sequences(frame_context_t *const ctx,
+                                 istream_t *const in,
+                                 sequence_command_t *const sequences,
+                                 const size_t num_sequences);
+static sequence_command_t decode_sequence(sequence_states_t *const state,
+                                          const u8 *const src,
+                                          i64 *const offset,
+                                          int lastSequence);
+static void decode_seq_table(FSE_dtable *const table, istream_t *const in,
+                               const seq_part_t type, const seq_mode_t mode);
+
+static size_t decode_sequences(frame_context_t *const ctx, istream_t *in,
+                               sequence_command_t **const sequences) {
+    // "A compressed block is a succession of sequences . A sequence is a
+    // literal copy command, followed by a match copy command. A literal copy
+    // command specifies a length. It is the number of bytes to be copied (or
+    // extracted) from the literal section. A match copy command specifies an
+    // offset and a length. The offset gives the position to copy from, which
+    // can be within a previous block."
+
+    size_t num_sequences;
+
+    // "Number_of_Sequences
+    //
+    // This is a variable size field using between 1 and 3 bytes. Let's call its
+    // first byte byte0."
+    u8 header = IO_read_bits(in, 8);
+    if (header < 128) {
+        // "Number_of_Sequences = byte0 . Uses 1 byte."
+        num_sequences = header;
+    } else if (header < 255) {
+        // "Number_of_Sequences = ((byte0-128) << 8) + byte1 . Uses 2 bytes."
+        num_sequences = ((header - 128) << 8) + IO_read_bits(in, 8);
+    } else {
+        // "Number_of_Sequences = byte1 + (byte2<<8) + 0x7F00 . Uses 3 bytes."
+        num_sequences = IO_read_bits(in, 16) + 0x7F00;
+    }
+
+    if (num_sequences == 0) {
+        // "There are no sequences. The sequence section stops there."
+        *sequences = NULL;
+        return 0;
+    }
+
+    *sequences = malloc(num_sequences * sizeof(sequence_command_t));
+    if (!*sequences) {
+        BAD_ALLOC();
+    }
+
+    decompress_sequences(ctx, in, *sequences, num_sequences);
+    return num_sequences;
+}
+
+/// Decompress the FSE encoded sequence commands
+static void decompress_sequences(frame_context_t *const ctx, istream_t *in,
+                                 sequence_command_t *const sequences,
+                                 const size_t num_sequences) {
+    // "The Sequences_Section regroup all symbols required to decode commands.
+    // There are 3 symbol types : literals lengths, offsets and match lengths.
+    // They are encoded together, interleaved, in a single bitstream."
+
+    // "Symbol compression modes
+    //
+    // This is a single byte, defining the compression mode of each symbol
+    // type."
+    //
+    // Bit number : Field name
+    // 7-6        : Literals_Lengths_Mode
+    // 5-4        : Offsets_Mode
+    // 3-2        : Match_Lengths_Mode
+    // 1-0        : Reserved
+    u8 compression_modes = IO_read_bits(in, 8);
+
+    if ((compression_modes & 3) != 0) {
+        // Reserved bits set
+        CORRUPTION();
+    }
+
+    // "Following the header, up to 3 distribution tables can be described. When
+    // present, they are in this order :
+    //
+    // Literals lengths
+    // Offsets
+    // Match Lengths"
+    // Update the tables we have stored in the context
+    decode_seq_table(&ctx->ll_dtable, in, seq_literal_length,
+                     (compression_modes >> 6) & 3);
+
+    decode_seq_table(&ctx->of_dtable, in, seq_offset,
+                     (compression_modes >> 4) & 3);
+
+    decode_seq_table(&ctx->ml_dtable, in, seq_match_length,
+                     (compression_modes >> 2) & 3);
+
+
+    sequence_states_t states;
+
+    // Initialize the decoding tables
+    {
+        states.ll_table = ctx->ll_dtable;
+        states.of_table = ctx->of_dtable;
+        states.ml_table = ctx->ml_dtable;
+    }
+
+    const size_t len = IO_istream_len(in);
+    const u8 *const src = IO_get_read_ptr(in, len);
+
+    // "After writing the last bit containing information, the compressor writes
+    // a single 1-bit and then fills the byte with 0-7 0 bits of padding."
+    const int padding = 8 - highest_set_bit(src[len - 1]);
+    // The offset starts at the end because FSE streams are read backwards
+    i64 bit_offset = (i64)(len * 8 - (size_t)padding);
+
+    // "The bitstream starts with initial state values, each using the required
+    // number of bits in their respective accuracy, decoded previously from
+    // their normalized distribution.
+    //
+    // It starts by Literals_Length_State, followed by Offset_State, and finally
+    // Match_Length_State."
+    FSE_init_state(&states.ll_table, &states.ll_state, src, &bit_offset);
+    FSE_init_state(&states.of_table, &states.of_state, src, &bit_offset);
+    FSE_init_state(&states.ml_table, &states.ml_state, src, &bit_offset);
+
+    for (size_t i = 0; i < num_sequences; i++) {
+        // Decode sequences one by one
+        sequences[i] = decode_sequence(&states, src, &bit_offset, i==num_sequences-1);
+    }
+
+    if (bit_offset != 0) {
+        CORRUPTION();
+    }
+}
+
+// Decode a single sequence and update the state
+static sequence_command_t decode_sequence(sequence_states_t *const states,
+                                          const u8 *const src,
+                                          i64 *const offset,
+                                          int lastSequence) {
+    // "Each symbol is a code in its own context, which specifies Baseline and
+    // Number_of_Bits to add. Codes are FSE compressed, and interleaved with raw
+    // additional bits in the same bitstream."
+
+    // Decode symbols, but don't update states
+    const u8 of_code = FSE_peek_symbol(&states->of_table, states->of_state);
+    const u8 ll_code = FSE_peek_symbol(&states->ll_table, states->ll_state);
+    const u8 ml_code = FSE_peek_symbol(&states->ml_table, states->ml_state);
+
+    // Offset doesn't need a max value as it's not decoded using a table
+    if (ll_code > SEQ_MAX_CODES[seq_literal_length] ||
+        ml_code > SEQ_MAX_CODES[seq_match_length]) {
+        CORRUPTION();
+    }
+
+    // Read the interleaved bits
+    sequence_command_t seq;
+    // "Decoding starts by reading the Number_of_Bits required to decode Offset.
+    // It then does the same for Match_Length, and then for Literals_Length."
+    seq.offset = ((u32)1 << of_code) + STREAM_read_bits(src, of_code, offset);
+
+    seq.match_length =
+        SEQ_MATCH_LENGTH_BASELINES[ml_code] +
+        STREAM_read_bits(src, SEQ_MATCH_LENGTH_EXTRA_BITS[ml_code], offset);
+
+    seq.literal_length =
+        SEQ_LITERAL_LENGTH_BASELINES[ll_code] +
+        STREAM_read_bits(src, SEQ_LITERAL_LENGTH_EXTRA_BITS[ll_code], offset);
+
+    // "If it is not the last sequence in the block, the next operation is to
+    // update states. Using the rules pre-calculated in the decoding tables,
+    // Literals_Length_State is updated, followed by Match_Length_State, and
+    // then Offset_State."
+    // If the stream is complete don't read bits to update state
+    if (!lastSequence) {
+        FSE_update_state(&states->ll_table, &states->ll_state, src, offset);
+        FSE_update_state(&states->ml_table, &states->ml_state, src, offset);
+        FSE_update_state(&states->of_table, &states->of_state, src, offset);
+    }
+
+    return seq;
+}
+
+/// Given a sequence part and table mode, decode the FSE distribution
+/// Errors if the mode is `seq_repeat` without a pre-existing table in `table`
+static void decode_seq_table(FSE_dtable *const table, istream_t *const in,
+                             const seq_part_t type, const seq_mode_t mode) {
+    // Constant arrays indexed by seq_part_t
+    const i16 *const default_distributions[] = {SEQ_LITERAL_LENGTH_DEFAULT_DIST,
+                                                SEQ_OFFSET_DEFAULT_DIST,
+                                                SEQ_MATCH_LENGTH_DEFAULT_DIST};
+    const size_t default_distribution_lengths[] = {36, 29, 53};
+    const size_t default_distribution_accuracies[] = {6, 5, 6};
+
+    const size_t max_accuracies[] = {9, 8, 9};
+
+    if (mode != seq_repeat) {
+        // Free old one before overwriting
+        FSE_free_dtable(table);
+    }
+
+    switch (mode) {
+    case seq_predefined: {
+        // "Predefined_Mode : uses a predefined distribution table."
+        const i16 *distribution = default_distributions[type];
+        const size_t symbs = default_distribution_lengths[type];
+        const size_t accuracy_log = default_distribution_accuracies[type];
+
+        FSE_init_dtable(table, distribution, symbs, accuracy_log);
+        break;
+    }
+    case seq_rle: {
+        // "RLE_Mode : it's a single code, repeated Number_of_Sequences times."
+        const u8 symb = IO_get_read_ptr(in, 1)[0];
+        FSE_init_dtable_rle(table, symb);
+        break;
+    }
+    case seq_fse: {
+        // "FSE_Compressed_Mode : standard FSE compression. A distribution table
+        // will be present "
+        FSE_decode_header(table, in, max_accuracies[type]);
+        break;
+    }
+    case seq_repeat:
+        // "Repeat_Mode : reuse distribution table from previous compressed
+        // block."
+        // Nothing to do here, table will be unchanged
+        if (!table->symbols) {
+            // This mode is invalid if we don't already have a table
+            CORRUPTION();
+        }
+        break;
+    default:
+        // Impossible, as mode is from 0-3
+        IMPOSSIBLE();
+        break;
+    }
+
+}
+/******* END SEQUENCE DECODING ************************************************/
+
+/******* SEQUENCE EXECUTION ***************************************************/
+static void execute_sequences(frame_context_t *const ctx, ostream_t *const out,
+                              const u8 *const literals,
+                              const size_t literals_len,
+                              const sequence_command_t *const sequences,
+                              const size_t num_sequences) {
+    istream_t litstream = IO_make_istream(literals, literals_len);
+
+    u64 *const offset_hist = ctx->previous_offsets;
+    size_t total_output = ctx->current_total_output;
+
+    for (size_t i = 0; i < num_sequences; i++) {
+        const sequence_command_t seq = sequences[i];
+        {
+            const u32 literals_size = copy_literals(seq.literal_length, &litstream, out);
+            total_output += literals_size;
+        }
+
+        size_t const offset = compute_offset(seq, offset_hist);
+
+        size_t const match_length = seq.match_length;
+
+        execute_match_copy(ctx, offset, match_length, total_output, out);
+
+        total_output += match_length;
+    }
+
+    // Copy any leftover literals
+    {
+        size_t len = IO_istream_len(&litstream);
+        copy_literals(len, &litstream, out);
+        total_output += len;
+    }
+
+    ctx->current_total_output = total_output;
+}
+
+static u32 copy_literals(const size_t literal_length, istream_t *litstream,
+                         ostream_t *const out) {
+    // If the sequence asks for more literals than are left, the
+    // sequence must be corrupted
+    if (literal_length > IO_istream_len(litstream)) {
+        CORRUPTION();
+    }
+
+    u8 *const write_ptr = IO_get_write_ptr(out, literal_length);
+    const u8 *const read_ptr =
+         IO_get_read_ptr(litstream, literal_length);
+    // Copy literals to output
+    memcpy(write_ptr, read_ptr, literal_length);
+
+    return literal_length;
+}
+
+static size_t compute_offset(sequence_command_t seq, u64 *const offset_hist) {
+    size_t offset;
+    // Offsets are special, we need to handle the repeat offsets
+    if (seq.offset <= 3) {
+        // "The first 3 values define a repeated offset and we will call
+        // them Repeated_Offset1, Repeated_Offset2, and Repeated_Offset3.
+        // They are sorted in recency order, with Repeated_Offset1 meaning
+        // 'most recent one'".
+
+        // Use 0 indexing for the array
+        u32 idx = seq.offset - 1;
+        if (seq.literal_length == 0) {
+            // "There is an exception though, when current sequence's
+            // literals length is 0. In this case, repeated offsets are
+            // shifted by one, so Repeated_Offset1 becomes Repeated_Offset2,
+            // Repeated_Offset2 becomes Repeated_Offset3, and
+            // Repeated_Offset3 becomes Repeated_Offset1 - 1_byte."
+            idx++;
+        }
+
+        if (idx == 0) {
+            offset = offset_hist[0];
+        } else {
+            // If idx == 3 then literal length was 0 and the offset was 3,
+            // as per the exception listed above
+            offset = idx < 3 ? offset_hist[idx] : offset_hist[0] - 1;
+
+            // If idx == 1 we don't need to modify offset_hist[2], since
+            // we're using the second-most recent code
+            if (idx > 1) {
+                offset_hist[2] = offset_hist[1];
+            }
+            offset_hist[1] = offset_hist[0];
+            offset_hist[0] = offset;
+        }
+    } else {
+        // When it's not a repeat offset:
+        // "if (Offset_Value > 3) offset = Offset_Value - 3;"
+        offset = seq.offset - 3;
+
+        // Shift back history
+        offset_hist[2] = offset_hist[1];
+        offset_hist[1] = offset_hist[0];
+        offset_hist[0] = offset;
+    }
+    return offset;
+}
+
+static void execute_match_copy(frame_context_t *const ctx, size_t offset,
+                              size_t match_length, size_t total_output,
+                              ostream_t *const out) {
+    u8 *write_ptr = IO_get_write_ptr(out, match_length);
+    if (total_output <= ctx->header.window_size) {
+        // In this case offset might go back into the dictionary
+        if (offset > total_output + ctx->dict_content_len) {
+            // The offset goes beyond even the dictionary
+            CORRUPTION();
+        }
+
+        if (offset > total_output) {
+            // "The rest of the dictionary is its content. The content act
+            // as a "past" in front of data to compress or decompress, so it
+            // can be referenced in sequence commands."
+            const size_t dict_copy =
+                MIN(offset - total_output, match_length);
+            const size_t dict_offset =
+                ctx->dict_content_len - (offset - total_output);
+
+            memcpy(write_ptr, ctx->dict_content + dict_offset, dict_copy);
+            write_ptr += dict_copy;
+            match_length -= dict_copy;
+        }
+    } else if (offset > ctx->header.window_size) {
+        CORRUPTION();
+    }
+
+    // We must copy byte by byte because the match length might be larger
+    // than the offset
+    // ex: if the output so far was "abc", a command with offset=3 and
+    // match_length=6 would produce "abcabcabc" as the new output
+    for (size_t j = 0; j < match_length; j++) {
+        *write_ptr = *(write_ptr - offset);
+        write_ptr++;
+    }
+}
+/******* END SEQUENCE EXECUTION ***********************************************/
+
+/******* OUTPUT SIZE COUNTING *************************************************/
+/// Get the decompressed size of an input stream so memory can be allocated in
+/// advance.
+/// This implementation assumes `src` points to a single ZSTD-compressed frame
+size_t ZSTD_get_decompressed_size(const void *src, const size_t src_len) {
+    istream_t in = IO_make_istream(src, src_len);
+
+    // get decompressed size from ZSTD frame header
+    {
+        const u32 magic_number = (u32)IO_read_bits(&in, 32);
+
+        if (magic_number == ZSTD_MAGIC_NUMBER) {
+            // ZSTD frame
+            frame_header_t header;
+            parse_frame_header(&header, &in);
+
+            if (header.frame_content_size == 0 && !header.single_segment_flag) {
+                // Content size not provided, we can't tell
+                return (size_t)-1;
+            }
+
+            return header.frame_content_size;
+        } else {
+            // not a real frame or skippable frame
+            ERROR("ZSTD frame magic number did not match");
+        }
+    }
+}
+/******* END OUTPUT SIZE COUNTING *********************************************/
+
+/******* DICTIONARY PARSING ***************************************************/
+dictionary_t* create_dictionary(void) {
+    dictionary_t* const dict = calloc(1, sizeof(dictionary_t));
+    if (!dict) {
+        BAD_ALLOC();
+    }
+    return dict;
+}
+
+/// Free an allocated dictionary
+void free_dictionary(dictionary_t *const dict) {
+    HUF_free_dtable(&dict->literals_dtable);
+    FSE_free_dtable(&dict->ll_dtable);
+    FSE_free_dtable(&dict->of_dtable);
+    FSE_free_dtable(&dict->ml_dtable);
+
+    free(dict->content);
+
+    memset(dict, 0, sizeof(dictionary_t));
+
+    free(dict);
+}
+
+
+#if !defined(ZDEC_NO_DICTIONARY)
+#define DICT_SIZE_ERROR() ERROR("Dictionary size cannot be less than 8 bytes")
+#define NULL_SRC() ERROR("Tried to create dictionary with pointer to null src");
+
+static void init_dictionary_content(dictionary_t *const dict,
+                                    istream_t *const in);
+
+void parse_dictionary(dictionary_t *const dict, const void *src,
+                             size_t src_len) {
+    const u8 *byte_src = (const u8 *)src;
+    memset(dict, 0, sizeof(dictionary_t));
+    if (src == NULL) { /* cannot initialize dictionary with null src */
+        NULL_SRC();
+    }
+    if (src_len < 8) {
+        DICT_SIZE_ERROR();
+    }
+
+    istream_t in = IO_make_istream(byte_src, src_len);
+
+    const u32 magic_number = IO_read_bits(&in, 32);
+    if (magic_number != 0xEC30A437) {
+        // raw content dict
+        IO_rewind_bits(&in, 32);
+        init_dictionary_content(dict, &in);
+        return;
+    }
+
+    dict->dictionary_id = IO_read_bits(&in, 32);
+
+    // "Entropy_Tables : following the same format as the tables in compressed
+    // blocks. They are stored in following order : Huffman tables for literals,
+    // FSE table for offsets, FSE table for match lengths, and FSE table for
+    // literals lengths. It's finally followed by 3 offset values, populating
+    // recent offsets (instead of using {1,4,8}), stored in order, 4-bytes
+    // little-endian each, for a total of 12 bytes. Each recent offset must have
+    // a value < dictionary size."
+    decode_huf_table(&dict->literals_dtable, &in);
+    decode_seq_table(&dict->of_dtable, &in, seq_offset, seq_fse);
+    decode_seq_table(&dict->ml_dtable, &in, seq_match_length, seq_fse);
+    decode_seq_table(&dict->ll_dtable, &in, seq_literal_length, seq_fse);
+
+    // Read in the previous offset history
+    dict->previous_offsets[0] = IO_read_bits(&in, 32);
+    dict->previous_offsets[1] = IO_read_bits(&in, 32);
+    dict->previous_offsets[2] = IO_read_bits(&in, 32);
+
+    // Ensure the provided offsets aren't too large
+    // "Each recent offset must have a value < dictionary size."
+    for (int i = 0; i < 3; i++) {
+        if (dict->previous_offsets[i] > src_len) {
+            ERROR("Dictionary corrupted");
+        }
+    }
+
+    // "Content : The rest of the dictionary is its content. The content act as
+    // a "past" in front of data to compress or decompress, so it can be
+    // referenced in sequence commands."
+    init_dictionary_content(dict, &in);
+}
+
+static void init_dictionary_content(dictionary_t *const dict,
+                                    istream_t *const in) {
+    // Copy in the content
+    dict->content_size = IO_istream_len(in);
+    dict->content = malloc(dict->content_size);
+    if (!dict->content) {
+        BAD_ALLOC();
+    }
+
+    const u8 *const content = IO_get_read_ptr(in, dict->content_size);
+
+    memcpy(dict->content, content, dict->content_size);
+}
+
+static void HUF_copy_dtable(HUF_dtable *const dst,
+                            const HUF_dtable *const src) {
+    if (src->max_bits == 0) {
+        memset(dst, 0, sizeof(HUF_dtable));
+        return;
+    }
+
+    const size_t size = (size_t)1 << src->max_bits;
+    dst->max_bits = src->max_bits;
+
+    dst->symbols = malloc(size);
+    dst->num_bits = malloc(size);
+    if (!dst->symbols || !dst->num_bits) {
+        BAD_ALLOC();
+    }
+
+    memcpy(dst->symbols, src->symbols, size);
+    memcpy(dst->num_bits, src->num_bits, size);
+}
+
+static void FSE_copy_dtable(FSE_dtable *const dst, const FSE_dtable *const src) {
+    if (src->accuracy_log == 0) {
+        memset(dst, 0, sizeof(FSE_dtable));
+        return;
+    }
+
+    size_t size = (size_t)1 << src->accuracy_log;
+    dst->accuracy_log = src->accuracy_log;
+
+    dst->symbols = malloc(size);
+    dst->num_bits = malloc(size);
+    dst->new_state_base = malloc(size * sizeof(u16));
+    if (!dst->symbols || !dst->num_bits || !dst->new_state_base) {
+        BAD_ALLOC();
+    }
+
+    memcpy(dst->symbols, src->symbols, size);
+    memcpy(dst->num_bits, src->num_bits, size);
+    memcpy(dst->new_state_base, src->new_state_base, size * sizeof(u16));
+}
+
+/// A dictionary acts as initializing values for the frame context before
+/// decompression, so we implement it by applying it's predetermined
+/// tables and content to the context before beginning decompression
+static void frame_context_apply_dict(frame_context_t *const ctx,
+                                     const dictionary_t *const dict) {
+    // If the content pointer is NULL then it must be an empty dict
+    if (!dict || !dict->content)
+        return;
+
+    // If the requested dictionary_id is non-zero, the correct dictionary must
+    // be present
+    if (ctx->header.dictionary_id != 0 &&
+        ctx->header.dictionary_id != dict->dictionary_id) {
+        ERROR("Wrong dictionary provided");
+    }
+
+    // Copy the dict content to the context for references during sequence
+    // execution
+    ctx->dict_content = dict->content;
+    ctx->dict_content_len = dict->content_size;
+
+    // If it's a formatted dict copy the precomputed tables in so they can
+    // be used in the table repeat modes
+    if (dict->dictionary_id != 0) {
+        // Deep copy the entropy tables so they can be freed independently of
+        // the dictionary struct
+        HUF_copy_dtable(&ctx->literals_dtable, &dict->literals_dtable);
+        FSE_copy_dtable(&ctx->ll_dtable, &dict->ll_dtable);
+        FSE_copy_dtable(&ctx->of_dtable, &dict->of_dtable);
+        FSE_copy_dtable(&ctx->ml_dtable, &dict->ml_dtable);
+
+        // Copy the repeated offsets
+        memcpy(ctx->previous_offsets, dict->previous_offsets,
+               sizeof(ctx->previous_offsets));
+    }
+}
+
+#else  // ZDEC_NO_DICTIONARY is defined
+
+static void frame_context_apply_dict(frame_context_t *const ctx,
+                                     const dictionary_t *const dict) {
+    (void)ctx;
+    if (dict && dict->content) ERROR("dictionary not supported");
+}
+
+#endif
+/******* END DICTIONARY PARSING ***********************************************/
+
+/******* IO STREAM OPERATIONS *************************************************/
+
+/// Reads `num` bits from a bitstream, and updates the internal offset
+static inline u64 IO_read_bits(istream_t *const in, const int num_bits) {
+    if (num_bits > 64 || num_bits <= 0) {
+        ERROR("Attempt to read an invalid number of bits");
+    }
+
+    const size_t bytes = (num_bits + in->bit_offset + 7) / 8;
+    const size_t full_bytes = (num_bits + in->bit_offset) / 8;
+    if (bytes > in->len) {
+        INP_SIZE();
+    }
+
+    const u64 result = read_bits_LE(in->ptr, num_bits, in->bit_offset);
+
+    in->bit_offset = (num_bits + in->bit_offset) % 8;
+    in->ptr += full_bytes;
+    in->len -= full_bytes;
+
+    return result;
+}
+
+/// If a non-zero number of bits have been read from the current byte, advance
+/// the offset to the next byte
+static inline void IO_rewind_bits(istream_t *const in, int num_bits) {
+    if (num_bits < 0) {
+        ERROR("Attempting to rewind stream by a negative number of bits");
+    }
+
+    // move the offset back by `num_bits` bits
+    const int new_offset = in->bit_offset - num_bits;
+    // determine the number of whole bytes we have to rewind, rounding up to an
+    // integer number (e.g. if `new_offset == -5`, `bytes == 1`)
+    const i64 bytes = -(new_offset - 7) / 8;
+
+    in->ptr -= bytes;
+    in->len += bytes;
+    // make sure the resulting `bit_offset` is positive, as mod in C does not
+    // convert numbers from negative to positive (e.g. -22 % 8 == -6)
+    in->bit_offset = ((new_offset % 8) + 8) % 8;
+}
+
+/// If the remaining bits in a byte will be unused, advance to the end of the
+/// byte
+static inline void IO_align_stream(istream_t *const in) {
+    if (in->bit_offset != 0) {
+        if (in->len == 0) {
+            INP_SIZE();
+        }
+        in->ptr++;
+        in->len--;
+        in->bit_offset = 0;
+    }
+}
+
+/// Write the given byte into the output stream
+static inline void IO_write_byte(ostream_t *const out, u8 symb) {
+    if (out->len == 0) {
+        OUT_SIZE();
+    }
+
+    out->ptr[0] = symb;
+    out->ptr++;
+    out->len--;
+}
+
+/// Returns the number of bytes left to be read in this stream.  The stream must
+/// be byte aligned.
+static inline size_t IO_istream_len(const istream_t *const in) {
+    return in->len;
+}
+
+/// Returns a pointer where `len` bytes can be read, and advances the internal
+/// state.  The stream must be byte aligned.
+static inline const u8 *IO_get_read_ptr(istream_t *const in, size_t len) {
+    if (len > in->len) {
+        INP_SIZE();
+    }
+    if (in->bit_offset != 0) {
+        ERROR("Attempting to operate on a non-byte aligned stream");
+    }
+    const u8 *const ptr = in->ptr;
+    in->ptr += len;
+    in->len -= len;
+
+    return ptr;
+}
+/// Returns a pointer to write `len` bytes to, and advances the internal state
+static inline u8 *IO_get_write_ptr(ostream_t *const out, size_t len) {
+    if (len > out->len) {
+        OUT_SIZE();
+    }
+    u8 *const ptr = out->ptr;
+    out->ptr += len;
+    out->len -= len;
+
+    return ptr;
+}
+
+/// Advance the inner state by `len` bytes
+static inline void IO_advance_input(istream_t *const in, size_t len) {
+    if (len > in->len) {
+         INP_SIZE();
+    }
+    if (in->bit_offset != 0) {
+        ERROR("Attempting to operate on a non-byte aligned stream");
+    }
+
+    in->ptr += len;
+    in->len -= len;
+}
+
+/// Returns an `ostream_t` constructed from the given pointer and length
+static inline ostream_t IO_make_ostream(u8 *out, size_t len) {
+    return (ostream_t) { out, len };
+}
+
+/// Returns an `istream_t` constructed from the given pointer and length
+static inline istream_t IO_make_istream(const u8 *in, size_t len) {
+    return (istream_t) { in, len, 0 };
+}
+
+/// Returns an `istream_t` with the same base as `in`, and length `len`
+/// Then, advance `in` to account for the consumed bytes
+/// `in` must be byte aligned
+static inline istream_t IO_make_sub_istream(istream_t *const in, size_t len) {
+    // Consume `len` bytes of the parent stream
+    const u8 *const ptr = IO_get_read_ptr(in, len);
+
+    // Make a substream using the pointer to those `len` bytes
+    return IO_make_istream(ptr, len);
+}
+/******* END IO STREAM OPERATIONS *********************************************/
+
+/******* BITSTREAM OPERATIONS *************************************************/
+/// Read `num` bits (up to 64) from `src + offset`, where `offset` is in bits
+static inline u64 read_bits_LE(const u8 *src, const int num_bits,
+                               const size_t offset) {
+    if (num_bits > 64) {
+        ERROR("Attempt to read an invalid number of bits");
+    }
+
+    // Skip over bytes that aren't in range
+    src += offset / 8;
+    size_t bit_offset = offset % 8;
+    u64 res = 0;
+
+    int shift = 0;
+    int left = num_bits;
+    while (left > 0) {
+        u64 mask = left >= 8 ? 0xff : (((u64)1 << left) - 1);
+        // Read the next byte, shift it to account for the offset, and then mask
+        // out the top part if we don't need all the bits
+        res += (((u64)*src++ >> bit_offset) & mask) << shift;
+        shift += 8 - bit_offset;
+        left -= 8 - bit_offset;
+        bit_offset = 0;
+    }
+
+    return res;
+}
+
+/// Read bits from the end of a HUF or FSE bitstream.  `offset` is in bits, so
+/// it updates `offset` to `offset - bits`, and then reads `bits` bits from
+/// `src + offset`.  If the offset becomes negative, the extra bits at the
+/// bottom are filled in with `0` bits instead of reading from before `src`.
+static inline u64 STREAM_read_bits(const u8 *const src, const int bits,
+                                   i64 *const offset) {
+    *offset = *offset - bits;
+    size_t actual_off = *offset;
+    size_t actual_bits = bits;
+    // Don't actually read bits from before the start of src, so if `*offset <
+    // 0` fix actual_off and actual_bits to reflect the quantity to read
+    if (*offset < 0) {
+        actual_bits += *offset;
+        actual_off = 0;
+    }
+    u64 res = read_bits_LE(src, actual_bits, actual_off);
+
+    if (*offset < 0) {
+        // Fill in the bottom "overflowed" bits with 0's
+        res = -*offset >= 64 ? 0 : (res << -*offset);
+    }
+    return res;
+}
+/******* END BITSTREAM OPERATIONS *********************************************/
+
+/******* BIT COUNTING OPERATIONS **********************************************/
+/// Returns `x`, where `2^x` is the largest power of 2 less than or equal to
+/// `num`, or `-1` if `num == 0`.
+static inline int highest_set_bit(const u64 num) {
+    for (int i = 63; i >= 0; i--) {
+        if (((u64)1 << i) <= num) {
+            return i;
+        }
+    }
+    return -1;
+}
+/******* END BIT COUNTING OPERATIONS ******************************************/
+
+/******* HUFFMAN PRIMITIVES ***************************************************/
+static inline u8 HUF_decode_symbol(const HUF_dtable *const dtable,
+                                   u16 *const state, const u8 *const src,
+                                   i64 *const offset) {
+    // Look up the symbol and number of bits to read
+    const u8 symb = dtable->symbols[*state];
+    const u8 bits = dtable->num_bits[*state];
+    const u16 rest = STREAM_read_bits(src, bits, offset);
+    // Shift `bits` bits out of the state, keeping the low order bits that
+    // weren't necessary to determine this symbol.  Then add in the new bits
+    // read from the stream.
+    *state = ((*state << bits) + rest) & (((u16)1 << dtable->max_bits) - 1);
+
+    return symb;
+}
+
+static inline void HUF_init_state(const HUF_dtable *const dtable,
+                                  u16 *const state, const u8 *const src,
+                                  i64 *const offset) {
+    // Read in a full `dtable->max_bits` bits to initialize the state
+    const u8 bits = dtable->max_bits;
+    *state = STREAM_read_bits(src, bits, offset);
+}
+
+static size_t HUF_decompress_1stream(const HUF_dtable *const dtable,
+                                     ostream_t *const out,
+                                     istream_t *const in) {
+    const size_t len = IO_istream_len(in);
+    if (len == 0) {
+        INP_SIZE();
+    }
+    const u8 *const src = IO_get_read_ptr(in, len);
+
+    // "Each bitstream must be read backward, that is starting from the end down
+    // to the beginning. Therefore it's necessary to know the size of each
+    // bitstream.
+    //
+    // It's also necessary to know exactly which bit is the latest. This is
+    // detected by a final bit flag : the highest bit of latest byte is a
+    // final-bit-flag. Consequently, a last byte of 0 is not possible. And the
+    // final-bit-flag itself is not part of the useful bitstream. Hence, the
+    // last byte contains between 0 and 7 useful bits."
+    const int padding = 8 - highest_set_bit(src[len - 1]);
+
+    // Offset starts at the end because HUF streams are read backwards
+    i64 bit_offset = len * 8 - padding;
+    u16 state;
+
+    HUF_init_state(dtable, &state, src, &bit_offset);
+
+    size_t symbols_written = 0;
+    while (bit_offset > -dtable->max_bits) {
+        // Iterate over the stream, decoding one symbol at a time
+        IO_write_byte(out, HUF_decode_symbol(dtable, &state, src, &bit_offset));
+        symbols_written++;
+    }
+    // "The process continues up to reading the required number of symbols per
+    // stream. If a bitstream is not entirely and exactly consumed, hence
+    // reaching exactly its beginning position with all bits consumed, the
+    // decoding process is considered faulty."
+
+    // When all symbols have been decoded, the final state value shouldn't have
+    // any data from the stream, so it should have "read" dtable->max_bits from
+    // before the start of `src`
+    // Therefore `offset`, the edge to start reading new bits at, should be
+    // dtable->max_bits before the start of the stream
+    if (bit_offset != -dtable->max_bits) {
+        CORRUPTION();
+    }
+
+    return symbols_written;
+}
+
+static size_t HUF_decompress_4stream(const HUF_dtable *const dtable,
+                                     ostream_t *const out, istream_t *const in) {
+    // "Compressed size is provided explicitly : in the 4-streams variant,
+    // bitstreams are preceded by 3 unsigned little-endian 16-bits values. Each
+    // value represents the compressed size of one stream, in order. The last
+    // stream size is deducted from total compressed size and from previously
+    // decoded stream sizes"
+    const size_t csize1 = IO_read_bits(in, 16);
+    const size_t csize2 = IO_read_bits(in, 16);
+    const size_t csize3 = IO_read_bits(in, 16);
+
+    istream_t in1 = IO_make_sub_istream(in, csize1);
+    istream_t in2 = IO_make_sub_istream(in, csize2);
+    istream_t in3 = IO_make_sub_istream(in, csize3);
+    istream_t in4 = IO_make_sub_istream(in, IO_istream_len(in));
+
+    size_t total_output = 0;
+    // Decode each stream independently for simplicity
+    // If we wanted to we could decode all 4 at the same time for speed,
+    // utilizing more execution units
+    total_output += HUF_decompress_1stream(dtable, out, &in1);
+    total_output += HUF_decompress_1stream(dtable, out, &in2);
+    total_output += HUF_decompress_1stream(dtable, out, &in3);
+    total_output += HUF_decompress_1stream(dtable, out, &in4);
+
+    return total_output;
+}
+
+/// Initializes a Huffman table using canonical Huffman codes
+/// For more explanation on canonical Huffman codes see
+/// https://www.cs.scranton.edu/~mccloske/courses/cmps340/huff_canonical_dec2015.html
+/// Codes within a level are allocated in symbol order (i.e. smaller symbols get
+/// earlier codes)
+static void HUF_init_dtable(HUF_dtable *const table, const u8 *const bits,
+                            const int num_symbs) {
+    memset(table, 0, sizeof(HUF_dtable));
+    if (num_symbs > HUF_MAX_SYMBS) {
+        ERROR("Too many symbols for Huffman");
+    }
+
+    u8 max_bits = 0;
+    u16 rank_count[HUF_MAX_BITS + 1];
+    memset(rank_count, 0, sizeof(rank_count));
+
+    // Count the number of symbols for each number of bits, and determine the
+    // depth of the tree
+    for (int i = 0; i < num_symbs; i++) {
+        if (bits[i] > HUF_MAX_BITS) {
+            ERROR("Huffman table depth too large");
+        }
+        max_bits = MAX(max_bits, bits[i]);
+        rank_count[bits[i]]++;
+    }
+
+    const size_t table_size = 1 << max_bits;
+    table->max_bits = max_bits;
+    table->symbols = malloc(table_size);
+    table->num_bits = malloc(table_size);
+
+    if (!table->symbols || !table->num_bits) {
+        free(table->symbols);
+        free(table->num_bits);
+        BAD_ALLOC();
+    }
+
+    // "Symbols are sorted by Weight. Within same Weight, symbols keep natural
+    // order. Symbols with a Weight of zero are removed. Then, starting from
+    // lowest weight, prefix codes are distributed in order."
+
+    u32 rank_idx[HUF_MAX_BITS + 1];
+    // Initialize the starting codes for each rank (number of bits)
+    rank_idx[max_bits] = 0;
+    for (int i = max_bits; i >= 1; i--) {
+        rank_idx[i - 1] = rank_idx[i] + rank_count[i] * (1 << (max_bits - i));
+        // The entire range takes the same number of bits so we can memset it
+        memset(&table->num_bits[rank_idx[i]], i, rank_idx[i - 1] - rank_idx[i]);
+    }
+
+    if (rank_idx[0] != table_size) {
+        CORRUPTION();
+    }
+
+    // Allocate codes and fill in the table
+    for (int i = 0; i < num_symbs; i++) {
+        if (bits[i] != 0) {
+            // Allocate a code for this symbol and set its range in the table
+            const u16 code = rank_idx[bits[i]];
+            // Since the code doesn't care about the bottom `max_bits - bits[i]`
+            // bits of state, it gets a range that spans all possible values of
+            // the lower bits
+            const u16 len = 1 << (max_bits - bits[i]);
+            memset(&table->symbols[code], i, len);
+            rank_idx[bits[i]] += len;
+        }
+    }
+}
+
+static void HUF_init_dtable_usingweights(HUF_dtable *const table,
+                                         const u8 *const weights,
+                                         const int num_symbs) {
+    // +1 because the last weight is not transmitted in the header
+    if (num_symbs + 1 > HUF_MAX_SYMBS) {
+        ERROR("Too many symbols for Huffman");
+    }
+
+    u8 bits[HUF_MAX_SYMBS];
+
+    u64 weight_sum = 0;
+    for (int i = 0; i < num_symbs; i++) {
+        // Weights are in the same range as bit count
+        if (weights[i] > HUF_MAX_BITS) {
+            CORRUPTION();
+        }
+        weight_sum += weights[i] > 0 ? (u64)1 << (weights[i] - 1) : 0;
+    }
+
+    // Find the first power of 2 larger than the sum
+    const int max_bits = highest_set_bit(weight_sum) + 1;
+    const u64 left_over = ((u64)1 << max_bits) - weight_sum;
+    // If the left over isn't a power of 2, the weights are invalid
+    if (left_over & (left_over - 1)) {
+        CORRUPTION();
+    }
+
+    // left_over is used to find the last weight as it's not transmitted
+    // by inverting 2^(weight - 1) we can determine the value of last_weight
+    const int last_weight = highest_set_bit(left_over) + 1;
+
+    for (int i = 0; i < num_symbs; i++) {
+        // "Number_of_Bits = Number_of_Bits ? Max_Number_of_Bits + 1 - Weight : 0"
+        bits[i] = weights[i] > 0 ? (max_bits + 1 - weights[i]) : 0;
+    }
+    bits[num_symbs] =
+        max_bits + 1 - last_weight; // Last weight is always non-zero
+
+    HUF_init_dtable(table, bits, num_symbs + 1);
+}
+
+static void HUF_free_dtable(HUF_dtable *const dtable) {
+    free(dtable->symbols);
+    free(dtable->num_bits);
+    memset(dtable, 0, sizeof(HUF_dtable));
+}
+/******* END HUFFMAN PRIMITIVES ***********************************************/
+
+/******* FSE PRIMITIVES *******************************************************/
+/// For more description of FSE see
+/// https://github.com/Cyan4973/FiniteStateEntropy/
+
+/// Allow a symbol to be decoded without updating state
+static inline u8 FSE_peek_symbol(const FSE_dtable *const dtable,
+                                 const u16 state) {
+    return dtable->symbols[state];
+}
+
+/// Consumes bits from the input and uses the current state to determine the
+/// next state
+static inline void FSE_update_state(const FSE_dtable *const dtable,
+                                    u16 *const state, const u8 *const src,
+                                    i64 *const offset) {
+    const u8 bits = dtable->num_bits[*state];
+    const u16 rest = STREAM_read_bits(src, bits, offset);
+    *state = dtable->new_state_base[*state] + rest;
+}
+
+/// Decodes a single FSE symbol and updates the offset
+static inline u8 FSE_decode_symbol(const FSE_dtable *const dtable,
+                                   u16 *const state, const u8 *const src,
+                                   i64 *const offset) {
+    const u8 symb = FSE_peek_symbol(dtable, *state);
+    FSE_update_state(dtable, state, src, offset);
+    return symb;
+}
+
+static inline void FSE_init_state(const FSE_dtable *const dtable,
+                                  u16 *const state, const u8 *const src,
+                                  i64 *const offset) {
+    // Read in a full `accuracy_log` bits to initialize the state
+    const u8 bits = dtable->accuracy_log;
+    *state = STREAM_read_bits(src, bits, offset);
+}
+
+static size_t FSE_decompress_interleaved2(const FSE_dtable *const dtable,
+                                          ostream_t *const out,
+                                          istream_t *const in) {
+    const size_t len = IO_istream_len(in);
+    if (len == 0) {
+        INP_SIZE();
+    }
+    const u8 *const src = IO_get_read_ptr(in, len);
+
+    // "Each bitstream must be read backward, that is starting from the end down
+    // to the beginning. Therefore it's necessary to know the size of each
+    // bitstream.
+    //
+    // It's also necessary to know exactly which bit is the latest. This is
+    // detected by a final bit flag : the highest bit of latest byte is a
+    // final-bit-flag. Consequently, a last byte of 0 is not possible. And the
+    // final-bit-flag itself is not part of the useful bitstream. Hence, the
+    // last byte contains between 0 and 7 useful bits."
+    const int padding = 8 - highest_set_bit(src[len - 1]);
+    i64 offset = len * 8 - padding;
+
+    u16 state1, state2;
+    // "The first state (State1) encodes the even indexed symbols, and the
+    // second (State2) encodes the odd indexes. State1 is initialized first, and
+    // then State2, and they take turns decoding a single symbol and updating
+    // their state."
+    FSE_init_state(dtable, &state1, src, &offset);
+    FSE_init_state(dtable, &state2, src, &offset);
+
+    // Decode until we overflow the stream
+    // Since we decode in reverse order, overflowing the stream is offset going
+    // negative
+    size_t symbols_written = 0;
+    while (1) {
+        // "The number of symbols to decode is determined by tracking bitStream
+        // overflow condition: If updating state after decoding a symbol would
+        // require more bits than remain in the stream, it is assumed the extra
+        // bits are 0. Then, the symbols for each of the final states are
+        // decoded and the process is complete."
+        IO_write_byte(out, FSE_decode_symbol(dtable, &state1, src, &offset));
+        symbols_written++;
+        if (offset < 0) {
+            // There's still a symbol to decode in state2
+            IO_write_byte(out, FSE_peek_symbol(dtable, state2));
+            symbols_written++;
+            break;
+        }
+
+        IO_write_byte(out, FSE_decode_symbol(dtable, &state2, src, &offset));
+        symbols_written++;
+        if (offset < 0) {
+            // There's still a symbol to decode in state1
+            IO_write_byte(out, FSE_peek_symbol(dtable, state1));
+            symbols_written++;
+            break;
+        }
+    }
+
+    return symbols_written;
+}
+
+static void FSE_init_dtable(FSE_dtable *const dtable,
+                            const i16 *const norm_freqs, const int num_symbs,
+                            const int accuracy_log) {
+    if (accuracy_log > FSE_MAX_ACCURACY_LOG) {
+        ERROR("FSE accuracy too large");
+    }
+    if (num_symbs > FSE_MAX_SYMBS) {
+        ERROR("Too many symbols for FSE");
+    }
+
+    dtable->accuracy_log = accuracy_log;
+
+    const size_t size = (size_t)1 << accuracy_log;
+    dtable->symbols = malloc(size * sizeof(u8));
+    dtable->num_bits = malloc(size * sizeof(u8));
+    dtable->new_state_base = malloc(size * sizeof(u16));
+
+    if (!dtable->symbols || !dtable->num_bits || !dtable->new_state_base) {
+        BAD_ALLOC();
+    }
+
+    // Used to determine how many bits need to be read for each state,
+    // and where the destination range should start
+    // Needs to be u16 because max value is 2 * max number of symbols,
+    // which can be larger than a byte can store
+    u16 state_desc[FSE_MAX_SYMBS];
+
+    // "Symbols are scanned in their natural order for "less than 1"
+    // probabilities. Symbols with this probability are being attributed a
+    // single cell, starting from the end of the table. These symbols define a
+    // full state reset, reading Accuracy_Log bits."
+    int high_threshold = size;
+    for (int s = 0; s < num_symbs; s++) {
+        // Scan for low probability symbols to put at the top
+        if (norm_freqs[s] == -1) {
+            dtable->symbols[--high_threshold] = s;
+            state_desc[s] = 1;
+        }
+    }
+
+    // "All remaining symbols are sorted in their natural order. Starting from
+    // symbol 0 and table position 0, each symbol gets attributed as many cells
+    // as its probability. Cell allocation is spread, not linear."
+    // Place the rest in the table
+    const u16 step = (size >> 1) + (size >> 3) + 3;
+    const u16 mask = size - 1;
+    u16 pos = 0;
+    for (int s = 0; s < num_symbs; s++) {
+        if (norm_freqs[s] <= 0) {
+            continue;
+        }
+
+        state_desc[s] = norm_freqs[s];
+
+        for (int i = 0; i < norm_freqs[s]; i++) {
+            // Give `norm_freqs[s]` states to symbol s
+            dtable->symbols[pos] = s;
+            // "A position is skipped if already occupied, typically by a "less
+            // than 1" probability symbol."
+            do {
+                pos = (pos + step) & mask;
+            } while (pos >=
+                     high_threshold);
+            // Note: no other collision checking is necessary as `step` is
+            // coprime to `size`, so the cycle will visit each position exactly
+            // once
+        }
+    }
+    if (pos != 0) {
+        CORRUPTION();
+    }
+
+    // Now we can fill baseline and num bits
+    for (size_t i = 0; i < size; i++) {
+        u8 symbol = dtable->symbols[i];
+        u16 next_state_desc = state_desc[symbol]++;
+        // Fills in the table appropriately, next_state_desc increases by symbol
+        // over time, decreasing number of bits
+        dtable->num_bits[i] = (u8)(accuracy_log - highest_set_bit(next_state_desc));
+        // Baseline increases until the bit threshold is passed, at which point
+        // it resets to 0
+        dtable->new_state_base[i] =
+            ((u16)next_state_desc << dtable->num_bits[i]) - size;
+    }
+}
+
+/// Decode an FSE header as defined in the Zstandard format specification and
+/// use the decoded frequencies to initialize a decoding table.
+static void FSE_decode_header(FSE_dtable *const dtable, istream_t *const in,
+                                const int max_accuracy_log) {
+    // "An FSE distribution table describes the probabilities of all symbols
+    // from 0 to the last present one (included) on a normalized scale of 1 <<
+    // Accuracy_Log .
+    //
+    // It's a bitstream which is read forward, in little-endian fashion. It's
+    // not necessary to know its exact size, since it will be discovered and
+    // reported by the decoding process.
+    if (max_accuracy_log > FSE_MAX_ACCURACY_LOG) {
+        ERROR("FSE accuracy too large");
+    }
+
+    // The bitstream starts by reporting on which scale it operates.
+    // Accuracy_Log = low4bits + 5. Note that maximum Accuracy_Log for literal
+    // and match lengths is 9, and for offsets is 8. Higher values are
+    // considered errors."
+    const int accuracy_log = 5 + IO_read_bits(in, 4);
+    if (accuracy_log > max_accuracy_log) {
+        ERROR("FSE accuracy too large");
+    }
+
+    // "Then follows each symbol value, from 0 to last present one. The number
+    // of bits used by each field is variable. It depends on :
+    //
+    // Remaining probabilities + 1 : example : Presuming an Accuracy_Log of 8,
+    // and presuming 100 probabilities points have already been distributed, the
+    // decoder may read any value from 0 to 255 - 100 + 1 == 156 (inclusive).
+    // Therefore, it must read log2sup(156) == 8 bits.
+    //
+    // Value decoded : small values use 1 less bit : example : Presuming values
+    // from 0 to 156 (inclusive) are possible, 255-156 = 99 values are remaining
+    // in an 8-bits field. They are used this way : first 99 values (hence from
+    // 0 to 98) use only 7 bits, values from 99 to 156 use 8 bits. "
+
+    i32 remaining = 1 << accuracy_log;
+    i16 frequencies[FSE_MAX_SYMBS];
+
+    int symb = 0;
+    while (remaining > 0 && symb < FSE_MAX_SYMBS) {
+        // Log of the number of possible values we could read
+        int bits = highest_set_bit(remaining + 1) + 1;
+
+        u16 val = IO_read_bits(in, bits);
+
+        // Try to mask out the lower bits to see if it qualifies for the "small
+        // value" threshold
+        const u16 lower_mask = ((u16)1 << (bits - 1)) - 1;
+        const u16 threshold = ((u16)1 << bits) - 1 - (remaining + 1);
+
+        if ((val & lower_mask) < threshold) {
+            IO_rewind_bits(in, 1);
+            val = val & lower_mask;
+        } else if (val > lower_mask) {
+            val = val - threshold;
+        }
+
+        // "Probability is obtained from Value decoded by following formula :
+        // Proba = value - 1"
+        const i16 proba = (i16)val - 1;
+
+        // "It means value 0 becomes negative probability -1. -1 is a special
+        // probability, which means "less than 1". Its effect on distribution
+        // table is described in next paragraph. For the purpose of calculating
+        // cumulated distribution, it counts as one."
+        remaining -= proba < 0 ? -proba : proba;
+
+        frequencies[symb] = proba;
+        symb++;
+
+        // "When a symbol has a probability of zero, it is followed by a 2-bits
+        // repeat flag. This repeat flag tells how many probabilities of zeroes
+        // follow the current one. It provides a number ranging from 0 to 3. If
+        // it is a 3, another 2-bits repeat flag follows, and so on."
+        if (proba == 0) {
+            // Read the next two bits to see how many more 0s
+            int repeat = IO_read_bits(in, 2);
+
+            while (1) {
+                for (int i = 0; i < repeat && symb < FSE_MAX_SYMBS; i++) {
+                    frequencies[symb++] = 0;
+                }
+                if (repeat == 3) {
+                    repeat = IO_read_bits(in, 2);
+                } else {
+                    break;
+                }
+            }
+        }
+    }
+    IO_align_stream(in);
+
+    // "When last symbol reaches cumulated total of 1 << Accuracy_Log, decoding
+    // is complete. If the last symbol makes cumulated total go above 1 <<
+    // Accuracy_Log, distribution is considered corrupted."
+    if (remaining != 0 || symb >= FSE_MAX_SYMBS) {
+        CORRUPTION();
+    }
+
+    // Initialize the decoding table using the determined weights
+    FSE_init_dtable(dtable, frequencies, symb, accuracy_log);
+}
+
+static void FSE_init_dtable_rle(FSE_dtable *const dtable, const u8 symb) {
+    dtable->symbols = malloc(sizeof(u8));
+    dtable->num_bits = malloc(sizeof(u8));
+    dtable->new_state_base = malloc(sizeof(u16));
+
+    if (!dtable->symbols || !dtable->num_bits || !dtable->new_state_base) {
+        BAD_ALLOC();
+    }
+
+    // This setup will always have a state of 0, always return symbol `symb`,
+    // and never consume any bits
+    dtable->symbols[0] = symb;
+    dtable->num_bits[0] = 0;
+    dtable->new_state_base[0] = 0;
+    dtable->accuracy_log = 0;
+}
+
+static void FSE_free_dtable(FSE_dtable *const dtable) {
+    free(dtable->symbols);
+    free(dtable->num_bits);
+    free(dtable->new_state_base);
+    memset(dtable, 0, sizeof(FSE_dtable));
+}
+/******* END FSE PRIMITIVES ***************************************************/
diff --git a/lib/common/fse.h b/lib/common/fse.h
new file mode 100644
index 0000000..b6c2a3e
--- /dev/null
+++ b/lib/common/fse.h
@@ -0,0 +1,625 @@
+/* ******************************************************************
+ * FSE : Finite State Entropy codec
+ * Public Prototypes declaration
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ * You can contact the author at :
+ * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+#ifndef FSE_H
+#define FSE_H
+
+
+/*-*****************************************
+*  Dependencies
+******************************************/
+#include "zstd_deps.h"    /* size_t, ptrdiff_t */
+
+/*-*****************************************
+*  FSE_PUBLIC_API : control library symbols visibility
+******************************************/
+#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)
+#  define FSE_PUBLIC_API __attribute__ ((visibility ("default")))
+#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1)   /* Visual expected */
+#  define FSE_PUBLIC_API __declspec(dllexport)
+#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)
+#  define FSE_PUBLIC_API __declspec(dllimport) /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define FSE_PUBLIC_API
+#endif
+
+/*------   Version   ------*/
+#define FSE_VERSION_MAJOR    0
+#define FSE_VERSION_MINOR    9
+#define FSE_VERSION_RELEASE  0
+
+#define FSE_LIB_VERSION FSE_VERSION_MAJOR.FSE_VERSION_MINOR.FSE_VERSION_RELEASE
+#define FSE_QUOTE(str) #str
+#define FSE_EXPAND_AND_QUOTE(str) FSE_QUOTE(str)
+#define FSE_VERSION_STRING FSE_EXPAND_AND_QUOTE(FSE_LIB_VERSION)
+
+#define FSE_VERSION_NUMBER  (FSE_VERSION_MAJOR *100*100 + FSE_VERSION_MINOR *100 + FSE_VERSION_RELEASE)
+FSE_PUBLIC_API unsigned FSE_versionNumber(void);   /**< library version number; to be used when checking dll version */
+
+
+/*-*****************************************
+*  Tool functions
+******************************************/
+FSE_PUBLIC_API size_t FSE_compressBound(size_t size);       /* maximum compressed size */
+
+/* Error Management */
+FSE_PUBLIC_API unsigned    FSE_isError(size_t code);        /* tells if a return value is an error code */
+FSE_PUBLIC_API const char* FSE_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+/*-*****************************************
+*  FSE detailed API
+******************************************/
+/*!
+FSE_compress() does the following:
+1. count symbol occurrence from source[] into table count[] (see hist.h)
+2. normalize counters so that sum(count[]) == Power_of_2 (2^tableLog)
+3. save normalized counters to memory buffer using writeNCount()
+4. build encoding table 'CTable' from normalized counters
+5. encode the data stream using encoding table 'CTable'
+
+FSE_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+/* *** COMPRESSION *** */
+
+/*! FSE_optimalTableLog():
+    dynamically downsize 'tableLog' when conditions are met.
+    It saves CPU time, by using smaller tables, while preserving or even improving compression ratio.
+    @return : recommended tableLog (necessarily <= 'maxTableLog') */
+FSE_PUBLIC_API unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
+
+/*! FSE_normalizeCount():
+    normalize counts so that sum(count[]) == Power_of_2 (2^tableLog)
+    'normalizedCounter' is a table of short, of minimum size (maxSymbolValue+1).
+    useLowProbCount is a boolean parameter which trades off compressed size for
+    faster header decoding. When it is set to 1, the compressed data will be slightly
+    smaller. And when it is set to 0, FSE_readNCount() and FSE_buildDTable() will be
+    faster. If you are compressing a small amount of data (< 2 KB) then useLowProbCount=0
+    is a good default, since header deserialization makes a big speed difference.
+    Otherwise, useLowProbCount=1 is a good default, since the speed difference is small.
+    @return : tableLog,
+              or an errorCode, which can be tested using FSE_isError() */
+FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog,
+                    const unsigned* count, size_t srcSize, unsigned maxSymbolValue, unsigned useLowProbCount);
+
+/*! FSE_NCountWriteBound():
+    Provides the maximum possible size of an FSE normalized table, given 'maxSymbolValue' and 'tableLog'.
+    Typically useful for allocation purpose. */
+FSE_PUBLIC_API size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSE_writeNCount():
+    Compactly save 'normalizedCounter' into 'buffer'.
+    @return : size of the compressed table,
+              or an errorCode, which can be tested using FSE_isError(). */
+FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize,
+                                 const short* normalizedCounter,
+                                 unsigned maxSymbolValue, unsigned tableLog);
+
+/*! Constructor and Destructor of FSE_CTable.
+    Note that FSE_CTable size depends on 'tableLog' and 'maxSymbolValue' */
+typedef unsigned FSE_CTable;   /* don't allocate that. It's only meant to be more restrictive than void* */
+
+/*! FSE_buildCTable():
+    Builds `ct`, which must be already allocated, using FSE_createCTable().
+    @return : 0, or an errorCode, which can be tested using FSE_isError() */
+FSE_PUBLIC_API size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSE_compress_usingCTable():
+    Compress `src` using `ct` into `dst` which must be already allocated.
+    @return : size of compressed data (<= `dstCapacity`),
+              or 0 if compressed data could not fit into `dst`,
+              or an errorCode, which can be tested using FSE_isError() */
+FSE_PUBLIC_API size_t FSE_compress_usingCTable (void* dst, size_t dstCapacity, const void* src, size_t srcSize, const FSE_CTable* ct);
+
+/*!
+Tutorial :
+----------
+The first step is to count all symbols. FSE_count() does this job very fast.
+Result will be saved into 'count', a table of unsigned int, which must be already allocated, and have 'maxSymbolValuePtr[0]+1' cells.
+'src' is a table of bytes of size 'srcSize'. All values within 'src' MUST be <= maxSymbolValuePtr[0]
+maxSymbolValuePtr[0] will be updated, with its real value (necessarily <= original value)
+FSE_count() will return the number of occurrence of the most frequent symbol.
+This can be used to know if there is a single symbol within 'src', and to quickly evaluate its compressibility.
+If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError()).
+
+The next step is to normalize the frequencies.
+FSE_normalizeCount() will ensure that sum of frequencies is == 2 ^'tableLog'.
+It also guarantees a minimum of 1 to any Symbol with frequency >= 1.
+You can use 'tableLog'==0 to mean "use default tableLog value".
+If you are unsure of which tableLog value to use, you can ask FSE_optimalTableLog(),
+which will provide the optimal valid tableLog given sourceSize, maxSymbolValue, and a user-defined maximum (0 means "default").
+
+The result of FSE_normalizeCount() will be saved into a table,
+called 'normalizedCounter', which is a table of signed short.
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValue+1' cells.
+The return value is tableLog if everything proceeded as expected.
+It is 0 if there is a single symbol within distribution.
+If there is an error (ex: invalid tableLog value), the function will return an ErrorCode (which can be tested using FSE_isError()).
+
+'normalizedCounter' can be saved in a compact manner to a memory area using FSE_writeNCount().
+'buffer' must be already allocated.
+For guaranteed success, buffer size must be at least FSE_headerBound().
+The result of the function is the number of bytes written into 'buffer'.
+If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError(); ex : buffer size too small).
+
+'normalizedCounter' can then be used to create the compression table 'CTable'.
+The space required by 'CTable' must be already allocated, using FSE_createCTable().
+You can then use FSE_buildCTable() to fill 'CTable'.
+If there is an error, both functions will return an ErrorCode (which can be tested using FSE_isError()).
+
+'CTable' can then be used to compress 'src', with FSE_compress_usingCTable().
+Similar to FSE_count(), the convention is that 'src' is assumed to be a table of char of size 'srcSize'
+The function returns the size of compressed data (without header), necessarily <= `dstCapacity`.
+If it returns '0', compressed data could not fit into 'dst'.
+If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError()).
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*! FSE_readNCount():
+    Read compactly saved 'normalizedCounter' from 'rBuffer'.
+    @return : size read from 'rBuffer',
+              or an errorCode, which can be tested using FSE_isError().
+              maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter,
+                           unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
+                           const void* rBuffer, size_t rBuffSize);
+
+/*! FSE_readNCount_bmi2():
+ * Same as FSE_readNCount() but pass bmi2=1 when your CPU supports BMI2 and 0 otherwise.
+ */
+FSE_PUBLIC_API size_t FSE_readNCount_bmi2(short* normalizedCounter,
+                           unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
+                           const void* rBuffer, size_t rBuffSize, int bmi2);
+
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSE_readNCount() if it was saved using FSE_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSE_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSE_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+The next step is to build the decompression tables 'FSE_DTable' from 'normalizedCounter'.
+This is performed by the function FSE_buildDTable().
+The space required by 'FSE_DTable' must be already allocated using FSE_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+`FSE_DTable` can then be used to decompress `cSrc`, with FSE_decompress_usingDTable().
+`cSrcSize` must be strictly correct, otherwise decompression will fail.
+FSE_decompress_usingDTable() result will tell how many bytes were regenerated (<=`dstCapacity`).
+If there is an error, the function will return an error code, which can be tested using FSE_isError(). (ex: dst buffer too small)
+*/
+
+#endif  /* FSE_H */
+
+
+#if defined(FSE_STATIC_LINKING_ONLY) && !defined(FSE_H_FSE_STATIC_LINKING_ONLY)
+#define FSE_H_FSE_STATIC_LINKING_ONLY
+#include "bitstream.h"
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) ((size) + ((size)>>7) + 4 /* fse states */ + sizeof(size_t) /* bitContainer */)
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of FSE_CTable/FSE_DTable using below macros */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<((maxTableLog)-1)) + (((maxSymbolValue)+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<(maxTableLog)))
+
+/* or use the size to malloc() space directly. Pay attention to alignment restrictions though */
+#define FSE_CTABLE_SIZE(maxTableLog, maxSymbolValue)   (FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(FSE_CTable))
+#define FSE_DTABLE_SIZE(maxTableLog)                   (FSE_DTABLE_SIZE_U32(maxTableLog) * sizeof(FSE_DTable))
+
+
+/* *****************************************
+ *  FSE advanced API
+ ***************************************** */
+
+unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
+/**< same as FSE_optimalTableLog(), which used `minus==2` */
+
+size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);
+/**< build a fake FSE_CTable, designed to compress always the same symbolValue */
+
+/* FSE_buildCTable_wksp() :
+ * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
+ * `wkspSize` must be >= `FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog)` of `unsigned`.
+ * See FSE_buildCTable_wksp() for breakdown of workspace usage.
+ */
+#define FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog) (((maxSymbolValue + 2) + (1ull << (tableLog)))/2 + sizeof(U64)/sizeof(U32) /* additional 8 bytes for potential table overwrite */)
+#define FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) (sizeof(unsigned) * FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog))
+size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
+
+#define FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) (sizeof(short) * (maxSymbolValue + 1) + (1ULL << maxTableLog) + 8)
+#define FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) ((FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) + sizeof(unsigned) - 1) / sizeof(unsigned))
+FSE_PUBLIC_API size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
+/**< Same as FSE_buildDTable(), using an externally allocated `workspace` produced with `FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxSymbolValue)` */
+
+#define FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) (FSE_DTABLE_SIZE_U32(maxTableLog) + 1 + FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) + (FSE_MAX_SYMBOL_VALUE + 1) / 2 + 1)
+#define FSE_DECOMPRESS_WKSP_SIZE(maxTableLog, maxSymbolValue) (FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(unsigned))
+size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2);
+/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DECOMPRESS_WKSP_SIZE_U32(maxLog, maxSymbolValue)`.
+ * Set bmi2 to 1 if your CPU supports BMI2 or 0 if it doesn't */
+
+typedef enum {
+   FSE_repeat_none,  /**< Cannot use the previous table */
+   FSE_repeat_check, /**< Can use the previous table but it must be checked */
+   FSE_repeat_valid  /**< Can use the previous table and it is assumed to be valid */
+ } FSE_repeat;
+
+/* *****************************************
+*  FSE symbol compression API
+*******************************************/
+/*!
+   This API consists of small unitary functions, which highly benefit from being inlined.
+   Hence their body are included in next section.
+*/
+typedef struct {
+    ptrdiff_t   value;
+    const void* stateTable;
+    const void* symbolTT;
+    unsigned    stateLog;
+} FSE_CState_t;
+
+static void FSE_initCState(FSE_CState_t* CStatePtr, const FSE_CTable* ct);
+
+static void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* CStatePtr, unsigned symbol);
+
+static void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* CStatePtr);
+
+/**<
+These functions are inner components of FSE_compress_usingCTable().
+They allow the creation of custom streams, mixing multiple tables and bit sources.
+
+A key property to keep in mind is that encoding and decoding are done **in reverse direction**.
+So the first symbol you will encode is the last you will decode, like a LIFO stack.
+
+You will need a few variables to track your CStream. They are :
+
+FSE_CTable    ct;         // Provided by FSE_buildCTable()
+BIT_CStream_t bitStream;  // bitStream tracking structure
+FSE_CState_t  state;      // State tracking structure (can have several)
+
+
+The first thing to do is to init bitStream and state.
+    size_t errorCode = BIT_initCStream(&bitStream, dstBuffer, maxDstSize);
+    FSE_initCState(&state, ct);
+
+Note that BIT_initCStream() can produce an error code, so its result should be tested, using FSE_isError();
+You can then encode your input data, byte after byte.
+FSE_encodeSymbol() outputs a maximum of 'tableLog' bits at a time.
+Remember decoding will be done in reverse direction.
+    FSE_encodeByte(&bitStream, &state, symbol);
+
+At any time, you can also add any bit sequence.
+Note : maximum allowed nbBits is 25, for compatibility with 32-bits decoders
+    BIT_addBits(&bitStream, bitField, nbBits);
+
+The above methods don't commit data to memory, they just store it into local register, for speed.
+Local register size is 64-bits on 64-bits systems, 32-bits on 32-bits systems (size_t).
+Writing data to memory is a manual operation, performed by the flushBits function.
+    BIT_flushBits(&bitStream);
+
+Your last FSE encoding operation shall be to flush your last state value(s).
+    FSE_flushState(&bitStream, &state);
+
+Finally, you must close the bitStream.
+The function returns the size of CStream in bytes.
+If data couldn't fit into dstBuffer, it will return a 0 ( == not compressible)
+If there is an error, it returns an errorCode (which can be tested using FSE_isError()).
+    size_t size = BIT_closeCStream(&bitStream);
+*/
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct {
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+/**<
+Let's now decompose FSE_decompress_usingDTable() into its unitary components.
+You will decode FSE-encoded symbols from the bitStream,
+and also any other bitFields you put in, **in reverse order**.
+
+You will need a few variables to track your bitStream. They are :
+
+BIT_DStream_t DStream;    // Stream context
+FSE_DState_t  DState;     // State context. Multiple ones are possible
+FSE_DTable*   DTablePtr;  // Decoding table, provided by FSE_buildDTable()
+
+The first thing to do is to init the bitStream.
+    errorCode = BIT_initDStream(&DStream, srcBuffer, srcSize);
+
+You should then retrieve your initial state(s)
+(in reverse flushing order if you have several ones) :
+    errorCode = FSE_initDState(&DState, &DStream, DTablePtr);
+
+You can then decode your data, symbol after symbol.
+For information the maximum number of bits read by FSE_decodeSymbol() is 'tableLog'.
+Keep in mind that symbols are decoded in reverse order, like a LIFO stack (last in, first out).
+    unsigned char symbol = FSE_decodeSymbol(&DState, &DStream);
+
+You can retrieve any bitfield you eventually stored into the bitStream (in reverse order)
+Note : maximum allowed nbBits is 25, for 32-bits compatibility
+    size_t bitField = BIT_readBits(&DStream, nbBits);
+
+All above operations only read from local register (which size depends on size_t).
+Refueling the register from memory is manually performed by the reload method.
+    endSignal = FSE_reloadDStream(&DStream);
+
+BIT_reloadDStream() result tells if there is still some more data to read from DStream.
+BIT_DStream_unfinished : there is still some data left into the DStream.
+BIT_DStream_endOfBuffer : Dstream reached end of buffer. Its container may no longer be completely filled.
+BIT_DStream_completed : Dstream reached its exact end, corresponding in general to decompression completed.
+BIT_DStream_tooFar : Dstream went too far. Decompression result is corrupted.
+
+When reaching end of buffer (BIT_DStream_endOfBuffer), progress slowly, notably if you decode multiple symbols per loop,
+to properly detect the exact end of stream.
+After each decoded symbol, check if DStream is fully consumed using this simple test :
+    BIT_reloadDStream(&DStream) >= BIT_DStream_completed
+
+When it's done, verify decompression is fully completed, by checking both DStream and the relevant states.
+Checking if DStream has reached its end is performed by :
+    BIT_endOfDStream(&DStream);
+Check also the states. There might be some symbols left there, if some high probability ones (>50%) are possible.
+    FSE_endOfDState(&DState);
+*/
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+typedef struct {
+    int deltaFindState;
+    U32 deltaNbBits;
+} FSE_symbolCompressionTransform; /* total 8 bytes */
+
+MEM_STATIC void FSE_initCState(FSE_CState_t* statePtr, const FSE_CTable* ct)
+{
+    const void* ptr = ct;
+    const U16* u16ptr = (const U16*) ptr;
+    const U32 tableLog = MEM_read16(ptr);
+    statePtr->value = (ptrdiff_t)1<<tableLog;
+    statePtr->stateTable = u16ptr+2;
+    statePtr->symbolTT = ct + 1 + (tableLog ? (1<<(tableLog-1)) : 1);
+    statePtr->stateLog = tableLog;
+}
+
+
+/*! FSE_initCState2() :
+*   Same as FSE_initCState(), but the first symbol to include (which will be the last to be read)
+*   uses the smallest state value possible, saving the cost of this symbol */
+MEM_STATIC void FSE_initCState2(FSE_CState_t* statePtr, const FSE_CTable* ct, U32 symbol)
+{
+    FSE_initCState(statePtr, ct);
+    {   const FSE_symbolCompressionTransform symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
+        const U16* stateTable = (const U16*)(statePtr->stateTable);
+        U32 nbBitsOut  = (U32)((symbolTT.deltaNbBits + (1<<15)) >> 16);
+        statePtr->value = (nbBitsOut << 16) - symbolTT.deltaNbBits;
+        statePtr->value = stateTable[(statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];
+    }
+}
+
+MEM_STATIC void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* statePtr, unsigned symbol)
+{
+    FSE_symbolCompressionTransform const symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
+    const U16* const stateTable = (const U16*)(statePtr->stateTable);
+    U32 const nbBitsOut  = (U32)((statePtr->value + symbolTT.deltaNbBits) >> 16);
+    BIT_addBits(bitC, (BitContainerType)statePtr->value, nbBitsOut);
+    statePtr->value = stateTable[ (statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];
+}
+
+MEM_STATIC void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* statePtr)
+{
+    BIT_addBits(bitC, (BitContainerType)statePtr->value, statePtr->stateLog);
+    BIT_flushBits(bitC);
+}
+
+
+/* FSE_getMaxNbBits() :
+ * Approximate maximum cost of a symbol, in bits.
+ * Fractional get rounded up (i.e. a symbol with a normalized frequency of 3 gives the same result as a frequency of 2)
+ * note 1 : assume symbolValue is valid (<= maxSymbolValue)
+ * note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
+MEM_STATIC U32 FSE_getMaxNbBits(const void* symbolTTPtr, U32 symbolValue)
+{
+    const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;
+    return (symbolTT[symbolValue].deltaNbBits + ((1<<16)-1)) >> 16;
+}
+
+/* FSE_bitCost() :
+ * Approximate symbol cost, as fractional value, using fixed-point format (accuracyLog fractional bits)
+ * note 1 : assume symbolValue is valid (<= maxSymbolValue)
+ * note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
+MEM_STATIC U32 FSE_bitCost(const void* symbolTTPtr, U32 tableLog, U32 symbolValue, U32 accuracyLog)
+{
+    const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;
+    U32 const minNbBits = symbolTT[symbolValue].deltaNbBits >> 16;
+    U32 const threshold = (minNbBits+1) << 16;
+    assert(tableLog < 16);
+    assert(accuracyLog < 31-tableLog);  /* ensure enough room for renormalization double shift */
+    {   U32 const tableSize = 1 << tableLog;
+        U32 const deltaFromThreshold = threshold - (symbolTT[symbolValue].deltaNbBits + tableSize);
+        U32 const normalizedDeltaFromThreshold = (deltaFromThreshold << accuracyLog) >> tableLog;   /* linear interpolation (very approximate) */
+        U32 const bitMultiplier = 1 << accuracyLog;
+        assert(symbolTT[symbolValue].deltaNbBits + tableSize <= threshold);
+        assert(normalizedDeltaFromThreshold <= bitMultiplier);
+        return (minNbBits+1)*bitMultiplier - normalizedDeltaFromThreshold;
+    }
+}
+
+
+/* ======    Decompression    ====== */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSE_DTableHeader* const DTableH = (const FSE_DTableHeader*)ptr;
+    DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_peekSymbol(const FSE_DState_t* DStatePtr)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC void FSE_updateState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.newState + lowBits;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+/*! FSE_decodeSymbolFast() :
+    unsafe, only works if no symbol has a probability > 50% */
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#ifndef FSE_MAX_MEMORY_USAGE
+#  define FSE_MAX_MEMORY_USAGE 14
+#endif
+#ifndef FSE_DEFAULT_MEMORY_USAGE
+#  define FSE_DEFAULT_MEMORY_USAGE 13
+#endif
+#if (FSE_DEFAULT_MEMORY_USAGE > FSE_MAX_MEMORY_USAGE)
+#  error "FSE_DEFAULT_MEMORY_USAGE must be <= FSE_MAX_MEMORY_USAGE"
+#endif
+
+/*!FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#ifndef FSE_MAX_SYMBOL_VALUE
+#  define FSE_MAX_SYMBOL_VALUE 255
+#endif
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+#define FSE_DECODE_TYPE FSE_decode_t
+
+
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#  error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+#define FSE_TABLESTEP(tableSize) (((tableSize)>>1) + ((tableSize)>>3) + 3)
+
+#endif /* FSE_STATIC_LINKING_ONLY */
diff --git a/lib/common/threading.c b/lib/common/threading.c
new file mode 100644
index 0000000..25bb8b9
--- /dev/null
+++ b/lib/common/threading.c
@@ -0,0 +1,182 @@
+/**
+ * Copyright (c) 2016 Tino Reichardt
+ * All rights reserved.
+ *
+ * You can contact the author at:
+ * - zstdmt source repository: https://github.com/mcmilk/zstdmt
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/**
+ * This file will hold wrapper for systems, which do not support pthreads
+ */
+
+#include "threading.h"
+
+/* create fake symbol to avoid empty translation unit warning */
+int g_ZSTD_threading_useless_symbol;
+
+#if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
+
+/**
+ * Windows minimalist Pthread Wrapper
+ */
+
+
+/* ===  Dependencies  === */
+#include <process.h>
+#include <errno.h>
+
+
+/* ===  Implementation  === */
+
+typedef struct {
+    void* (*start_routine)(void*);
+    void* arg;
+    int initialized;
+    ZSTD_pthread_cond_t initialized_cond;
+    ZSTD_pthread_mutex_t initialized_mutex;
+} ZSTD_thread_params_t;
+
+static unsigned __stdcall worker(void *arg)
+{
+    void* (*start_routine)(void*);
+    void* thread_arg;
+
+    /* Initialized thread_arg and start_routine and signal main thread that we don't need it
+     * to wait any longer.
+     */
+    {
+        ZSTD_thread_params_t*  thread_param = (ZSTD_thread_params_t*)arg;
+        thread_arg = thread_param->arg;
+        start_routine = thread_param->start_routine;
+
+        /* Signal main thread that we are running and do not depend on its memory anymore */
+        ZSTD_pthread_mutex_lock(&thread_param->initialized_mutex);
+        thread_param->initialized = 1;
+        ZSTD_pthread_cond_signal(&thread_param->initialized_cond);
+        ZSTD_pthread_mutex_unlock(&thread_param->initialized_mutex);
+    }
+
+    start_routine(thread_arg);
+
+    return 0;
+}
+
+int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
+            void* (*start_routine) (void*), void* arg)
+{
+    ZSTD_thread_params_t thread_param;
+    (void)unused;
+
+    if (thread==NULL) return -1;
+    *thread = NULL;
+
+    thread_param.start_routine = start_routine;
+    thread_param.arg = arg;
+    thread_param.initialized = 0;
+
+    /* Setup thread initialization synchronization */
+    if(ZSTD_pthread_cond_init(&thread_param.initialized_cond, NULL)) {
+        /* Should never happen on Windows */
+        return -1;
+    }
+    if(ZSTD_pthread_mutex_init(&thread_param.initialized_mutex, NULL)) {
+        /* Should never happen on Windows */
+        ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
+        return -1;
+    }
+
+    /* Spawn thread */
+    *thread = (HANDLE)_beginthreadex(NULL, 0, worker, &thread_param, 0, NULL);
+    if (*thread==NULL) {
+        ZSTD_pthread_mutex_destroy(&thread_param.initialized_mutex);
+        ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
+        return errno;
+    }
+
+    /* Wait for thread to be initialized */
+    ZSTD_pthread_mutex_lock(&thread_param.initialized_mutex);
+    while(!thread_param.initialized) {
+        ZSTD_pthread_cond_wait(&thread_param.initialized_cond, &thread_param.initialized_mutex);
+    }
+    ZSTD_pthread_mutex_unlock(&thread_param.initialized_mutex);
+    ZSTD_pthread_mutex_destroy(&thread_param.initialized_mutex);
+    ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
+
+    return 0;
+}
+
+int ZSTD_pthread_join(ZSTD_pthread_t thread)
+{
+    DWORD result;
+
+    if (!thread) return 0;
+
+    result = WaitForSingleObject(thread, INFINITE);
+    CloseHandle(thread);
+
+    switch (result) {
+    case WAIT_OBJECT_0:
+        return 0;
+    case WAIT_ABANDONED:
+        return EINVAL;
+    default:
+        return GetLastError();
+    }
+}
+
+#endif   /* ZSTD_MULTITHREAD */
+
+#if defined(ZSTD_MULTITHREAD) && DEBUGLEVEL >= 1 && !defined(_WIN32)
+
+#define ZSTD_DEPS_NEED_MALLOC
+#include "zstd_deps.h"
+
+int ZSTD_pthread_mutex_init(ZSTD_pthread_mutex_t* mutex, pthread_mutexattr_t const* attr)
+{
+    assert(mutex != NULL);
+    *mutex = (pthread_mutex_t*)ZSTD_malloc(sizeof(pthread_mutex_t));
+    if (!*mutex)
+        return 1;
+    return pthread_mutex_init(*mutex, attr);
+}
+
+int ZSTD_pthread_mutex_destroy(ZSTD_pthread_mutex_t* mutex)
+{
+    assert(mutex != NULL);
+    if (!*mutex)
+        return 0;
+    {
+        int const ret = pthread_mutex_destroy(*mutex);
+        ZSTD_free(*mutex);
+        return ret;
+    }
+}
+
+int ZSTD_pthread_cond_init(ZSTD_pthread_cond_t* cond, pthread_condattr_t const* attr)
+{
+    assert(cond != NULL);
+    *cond = (pthread_cond_t*)ZSTD_malloc(sizeof(pthread_cond_t));
+    if (!*cond)
+        return 1;
+    return pthread_cond_init(*cond, attr);
+}
+
+int ZSTD_pthread_cond_destroy(ZSTD_pthread_cond_t* cond)
+{
+    assert(cond != NULL);
+    if (!*cond)
+        return 0;
+    {
+        int const ret = pthread_cond_destroy(*cond);
+        ZSTD_free(*cond);
+        return ret;
+    }
+}
+
+#endif
diff --git a/lib/compress/fse_compress.c b/lib/compress/fse_compress.c
new file mode 100644
index 0000000..1ce3cf1
--- /dev/null
+++ b/lib/compress/fse_compress.c
@@ -0,0 +1,625 @@
+/* ******************************************************************
+ * FSE : Finite State Entropy encoder
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include "../common/compiler.h"
+#include "../common/mem.h"        /* U32, U16, etc. */
+#include "../common/debug.h"      /* assert, DEBUGLOG */
+#include "hist.h"       /* HIST_count_wksp */
+#include "../common/bitstream.h"
+#define FSE_STATIC_LINKING_ONLY
+#include "../common/fse.h"
+#include "../common/error_private.h"
+#define ZSTD_DEPS_NEED_MALLOC
+#define ZSTD_DEPS_NEED_MATH64
+#include "../common/zstd_deps.h"  /* ZSTD_memset */
+#include "../common/bits.h" /* ZSTD_highbit32 */
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_isError ERR_isError
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+/* Function templates */
+
+/* FSE_buildCTable_wksp() :
+ * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
+ * wkspSize should be sized to handle worst case situation, which is `1<<max_tableLog * sizeof(FSE_FUNCTION_TYPE)`
+ * workSpace must also be properly aligned with FSE_FUNCTION_TYPE requirements
+ */
+size_t FSE_buildCTable_wksp(FSE_CTable* ct,
+                      const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
+                            void* workSpace, size_t wkspSize)
+{
+    U32 const tableSize = 1 << tableLog;
+    U32 const tableMask = tableSize - 1;
+    void* const ptr = ct;
+    U16* const tableU16 = ( (U16*) ptr) + 2;
+    void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
+    FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
+    U32 const step = FSE_TABLESTEP(tableSize);
+    U32 const maxSV1 = maxSymbolValue+1;
+
+    U16* cumul = (U16*)workSpace;   /* size = maxSV1 */
+    FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSV1+1));  /* size = tableSize */
+
+    U32 highThreshold = tableSize-1;
+
+    assert(((size_t)workSpace & 1) == 0);  /* Must be 2 bytes-aligned */
+    if (FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) > wkspSize) return ERROR(tableLog_tooLarge);
+    /* CTable header */
+    tableU16[-2] = (U16) tableLog;
+    tableU16[-1] = (U16) maxSymbolValue;
+    assert(tableLog < 16);   /* required for threshold strategy to work */
+
+    /* For explanations on how to distribute symbol values over the table :
+     * https://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+
+     #ifdef __clang_analyzer__
+     ZSTD_memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize);   /* useless initialization, just to keep scan-build happy */
+     #endif
+
+    /* symbol start positions */
+    {   U32 u;
+        cumul[0] = 0;
+        for (u=1; u <= maxSV1; u++) {
+            if (normalizedCounter[u-1]==-1) {  /* Low proba symbol */
+                cumul[u] = cumul[u-1] + 1;
+                tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
+            } else {
+                assert(normalizedCounter[u-1] >= 0);
+                cumul[u] = cumul[u-1] + (U16)normalizedCounter[u-1];
+                assert(cumul[u] >= cumul[u-1]);  /* no overflow */
+        }   }
+        cumul[maxSV1] = (U16)(tableSize+1);
+    }
+
+    /* Spread symbols */
+    if (highThreshold == tableSize - 1) {
+        /* Case for no low prob count symbols. Lay down 8 bytes at a time
+         * to reduce branch misses since we are operating on a small block
+         */
+        BYTE* const spread = tableSymbol + tableSize; /* size = tableSize + 8 (may write beyond tableSize) */
+        {   U64 const add = 0x0101010101010101ull;
+            size_t pos = 0;
+            U64 sv = 0;
+            U32 s;
+            for (s=0; s<maxSV1; ++s, sv += add) {
+                int i;
+                int const n = normalizedCounter[s];
+                MEM_write64(spread + pos, sv);
+                for (i = 8; i < n; i += 8) {
+                    MEM_write64(spread + pos + i, sv);
+                }
+                assert(n>=0);
+                pos += (size_t)n;
+            }
+        }
+        /* Spread symbols across the table. Lack of lowprob symbols means that
+         * we don't need variable sized inner loop, so we can unroll the loop and
+         * reduce branch misses.
+         */
+        {   size_t position = 0;
+            size_t s;
+            size_t const unroll = 2; /* Experimentally determined optimal unroll */
+            assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
+            for (s = 0; s < (size_t)tableSize; s += unroll) {
+                size_t u;
+                for (u = 0; u < unroll; ++u) {
+                    size_t const uPosition = (position + (u * step)) & tableMask;
+                    tableSymbol[uPosition] = spread[s + u];
+                }
+                position = (position + (unroll * step)) & tableMask;
+            }
+            assert(position == 0);   /* Must have initialized all positions */
+        }
+    } else {
+        U32 position = 0;
+        U32 symbol;
+        for (symbol=0; symbol<maxSV1; symbol++) {
+            int nbOccurrences;
+            int const freq = normalizedCounter[symbol];
+            for (nbOccurrences=0; nbOccurrences<freq; nbOccurrences++) {
+                tableSymbol[position] = (FSE_FUNCTION_TYPE)symbol;
+                position = (position + step) & tableMask;
+                while (position > highThreshold)
+                    position = (position + step) & tableMask;   /* Low proba area */
+        }   }
+        assert(position==0);  /* Must have initialized all positions */
+    }
+
+    /* Build table */
+    {   U32 u; for (u=0; u<tableSize; u++) {
+        FSE_FUNCTION_TYPE s = tableSymbol[u];   /* note : static analyzer may not understand tableSymbol is properly initialized */
+        tableU16[cumul[s]++] = (U16) (tableSize+u);   /* TableU16 : sorted by symbol order; gives next state value */
+    }   }
+
+    /* Build Symbol Transformation Table */
+    {   unsigned total = 0;
+        unsigned s;
+        for (s=0; s<=maxSymbolValue; s++) {
+            switch (normalizedCounter[s])
+            {
+            case  0:
+                /* filling nonetheless, for compatibility with FSE_getMaxNbBits() */
+                symbolTT[s].deltaNbBits = ((tableLog+1) << 16) - (1<<tableLog);
+                break;
+
+            case -1:
+            case  1:
+                symbolTT[s].deltaNbBits = (tableLog << 16) - (1<<tableLog);
+                assert(total <= INT_MAX);
+                symbolTT[s].deltaFindState = (int)(total - 1);
+                total ++;
+                break;
+            default :
+                assert(normalizedCounter[s] > 1);
+                {   U32 const maxBitsOut = tableLog - ZSTD_highbit32 ((U32)normalizedCounter[s]-1);
+                    U32 const minStatePlus = (U32)normalizedCounter[s] << maxBitsOut;
+                    symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
+                    symbolTT[s].deltaFindState = (int)(total - (unsigned)normalizedCounter[s]);
+                    total +=  (unsigned)normalizedCounter[s];
+    }   }   }   }
+
+#if 0  /* debug : symbol costs */
+    DEBUGLOG(5, "\n --- table statistics : ");
+    {   U32 symbol;
+        for (symbol=0; symbol<=maxSymbolValue; symbol++) {
+            DEBUGLOG(5, "%3u: w=%3i,   maxBits=%u, fracBits=%.2f",
+                symbol, normalizedCounter[symbol],
+                FSE_getMaxNbBits(symbolTT, symbol),
+                (double)FSE_bitCost(symbolTT, tableLog, symbol, 8) / 256);
+    }   }
+#endif
+
+    return 0;
+}
+
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/*-**************************************************************
+*  FSE NCount encoding
+****************************************************************/
+size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
+{
+    size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog
+                                   + 4 /* bitCount initialized at 4 */
+                                   + 2 /* first two symbols may use one additional bit each */) / 8)
+                                   + 1 /* round up to whole nb bytes */
+                                   + 2 /* additional two bytes for bitstream flush */;
+    return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
+}
+
+static size_t
+FSE_writeNCount_generic (void* header, size_t headerBufferSize,
+                   const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
+                         unsigned writeIsSafe)
+{
+    BYTE* const ostart = (BYTE*) header;
+    BYTE* out = ostart;
+    BYTE* const oend = ostart + headerBufferSize;
+    int nbBits;
+    const int tableSize = 1 << tableLog;
+    int remaining;
+    int threshold;
+    U32 bitStream = 0;
+    int bitCount = 0;
+    unsigned symbol = 0;
+    unsigned const alphabetSize = maxSymbolValue + 1;
+    int previousIs0 = 0;
+
+    /* Table Size */
+    bitStream += (tableLog-FSE_MIN_TABLELOG) << bitCount;
+    bitCount  += 4;
+
+    /* Init */
+    remaining = tableSize+1;   /* +1 for extra accuracy */
+    threshold = tableSize;
+    nbBits = (int)tableLog+1;
+
+    while ((symbol < alphabetSize) && (remaining>1)) {  /* stops at 1 */
+        if (previousIs0) {
+            unsigned start = symbol;
+            while ((symbol < alphabetSize) && !normalizedCounter[symbol]) symbol++;
+            if (symbol == alphabetSize) break;   /* incorrect distribution */
+            while (symbol >= start+24) {
+                start+=24;
+                bitStream += 0xFFFFU << bitCount;
+                if ((!writeIsSafe) && (out > oend-2))
+                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                out[0] = (BYTE) bitStream;
+                out[1] = (BYTE)(bitStream>>8);
+                out+=2;
+                bitStream>>=16;
+            }
+            while (symbol >= start+3) {
+                start+=3;
+                bitStream += 3U << bitCount;
+                bitCount += 2;
+            }
+            bitStream += (symbol-start) << bitCount;
+            bitCount += 2;
+            if (bitCount>16) {
+                if ((!writeIsSafe) && (out > oend - 2))
+                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                out[0] = (BYTE)bitStream;
+                out[1] = (BYTE)(bitStream>>8);
+                out += 2;
+                bitStream >>= 16;
+                bitCount -= 16;
+        }   }
+        {   int count = normalizedCounter[symbol++];
+            int const max = (2*threshold-1) - remaining;
+            remaining -= count < 0 ? -count : count;
+            count++;   /* +1 for extra accuracy */
+            if (count>=threshold)
+                count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
+            bitStream += (U32)count << bitCount;
+            bitCount  += nbBits;
+            bitCount  -= (count<max);
+            previousIs0  = (count==1);
+            if (remaining<1) return ERROR(GENERIC);
+            while (remaining<threshold) { nbBits--; threshold>>=1; }
+        }
+        if (bitCount>16) {
+            if ((!writeIsSafe) && (out > oend - 2))
+                return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+            out[0] = (BYTE)bitStream;
+            out[1] = (BYTE)(bitStream>>8);
+            out += 2;
+            bitStream >>= 16;
+            bitCount -= 16;
+    }   }
+
+    if (remaining != 1)
+        return ERROR(GENERIC);  /* incorrect normalized distribution */
+    assert(symbol <= alphabetSize);
+
+    /* flush remaining bitStream */
+    if ((!writeIsSafe) && (out > oend - 2))
+        return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+    out[0] = (BYTE)bitStream;
+    out[1] = (BYTE)(bitStream>>8);
+    out+= (bitCount+7) /8;
+
+    assert(out >= ostart);
+    return (size_t)(out-ostart);
+}
+
+
+size_t FSE_writeNCount (void* buffer, size_t bufferSize,
+                  const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported */
+    if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
+
+    if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
+        return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 0);
+
+    return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1 /* write in buffer is safe */);
+}
+
+
+/*-**************************************************************
+*  FSE Compression Code
+****************************************************************/
+
+/* provides the minimum logSize to safely represent a distribution */
+static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
+{
+    U32 minBitsSrc = ZSTD_highbit32((U32)(srcSize)) + 1;
+    U32 minBitsSymbols = ZSTD_highbit32(maxSymbolValue) + 2;
+    U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
+    return minBits;
+}
+
+unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
+{
+    U32 maxBitsSrc = ZSTD_highbit32((U32)(srcSize - 1)) - minus;
+    U32 tableLog = maxTableLog;
+    U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
+    if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
+    if (maxBitsSrc < tableLog) tableLog = maxBitsSrc;   /* Accuracy can be reduced */
+    if (minBits > tableLog) tableLog = minBits;   /* Need a minimum to safely represent all symbol values */
+    if (tableLog < FSE_MIN_TABLELOG) tableLog = FSE_MIN_TABLELOG;
+    if (tableLog > FSE_MAX_TABLELOG) tableLog = FSE_MAX_TABLELOG;
+    return tableLog;
+}
+
+unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
+{
+    return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 2);
+}
+
+/* Secondary normalization method.
+   To be used when primary method fails. */
+
+static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue, short lowProbCount)
+{
+    short const NOT_YET_ASSIGNED = -2;
+    U32 s;
+    U32 distributed = 0;
+    U32 ToDistribute;
+
+    /* Init */
+    U32 const lowThreshold = (U32)(total >> tableLog);
+    U32 lowOne = (U32)((total * 3) >> (tableLog + 1));
+
+    for (s=0; s<=maxSymbolValue; s++) {
+        if (count[s] == 0) {
+            norm[s]=0;
+            continue;
+        }
+        if (count[s] <= lowThreshold) {
+            norm[s] = lowProbCount;
+            distributed++;
+            total -= count[s];
+            continue;
+        }
+        if (count[s] <= lowOne) {
+            norm[s] = 1;
+            distributed++;
+            total -= count[s];
+            continue;
+        }
+
+        norm[s]=NOT_YET_ASSIGNED;
+    }
+    ToDistribute = (1 << tableLog) - distributed;
+
+    if (ToDistribute == 0)
+        return 0;
+
+    if ((total / ToDistribute) > lowOne) {
+        /* risk of rounding to zero */
+        lowOne = (U32)((total * 3) / (ToDistribute * 2));
+        for (s=0; s<=maxSymbolValue; s++) {
+            if ((norm[s] == NOT_YET_ASSIGNED) && (count[s] <= lowOne)) {
+                norm[s] = 1;
+                distributed++;
+                total -= count[s];
+                continue;
+        }   }
+        ToDistribute = (1 << tableLog) - distributed;
+    }
+
+    if (distributed == maxSymbolValue+1) {
+        /* all values are pretty poor;
+           probably incompressible data (should have already been detected);
+           find max, then give all remaining points to max */
+        U32 maxV = 0, maxC = 0;
+        for (s=0; s<=maxSymbolValue; s++)
+            if (count[s] > maxC) { maxV=s; maxC=count[s]; }
+        norm[maxV] += (short)ToDistribute;
+        return 0;
+    }
+
+    if (total == 0) {
+        /* all of the symbols were low enough for the lowOne or lowThreshold */
+        for (s=0; ToDistribute > 0; s = (s+1)%(maxSymbolValue+1))
+            if (norm[s] > 0) { ToDistribute--; norm[s]++; }
+        return 0;
+    }
+
+    {   U64 const vStepLog = 62 - tableLog;
+        U64 const mid = (1ULL << (vStepLog-1)) - 1;
+        U64 const rStep = ZSTD_div64((((U64)1<<vStepLog) * ToDistribute) + mid, (U32)total);   /* scale on remaining */
+        U64 tmpTotal = mid;
+        for (s=0; s<=maxSymbolValue; s++) {
+            if (norm[s]==NOT_YET_ASSIGNED) {
+                U64 const end = tmpTotal + (count[s] * rStep);
+                U32 const sStart = (U32)(tmpTotal >> vStepLog);
+                U32 const sEnd = (U32)(end >> vStepLog);
+                U32 const weight = sEnd - sStart;
+                if (weight < 1)
+                    return ERROR(GENERIC);
+                norm[s] = (short)weight;
+                tmpTotal = end;
+    }   }   }
+
+    return 0;
+}
+
+size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
+                           const unsigned* count, size_t total,
+                           unsigned maxSymbolValue, unsigned useLowProbCount)
+{
+    /* Sanity checks */
+    if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
+    if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported size */
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported size */
+    if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC);   /* Too small tableLog, compression potentially impossible */
+
+    {   static U32 const rtbTable[] = {     0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
+        short const lowProbCount = useLowProbCount ? -1 : 1;
+        U64 const scale = 62 - tableLog;
+        U64 const step = ZSTD_div64((U64)1<<62, (U32)total);   /* <== here, one division ! */
+        U64 const vStep = 1ULL<<(scale-20);
+        int stillToDistribute = 1<<tableLog;
+        unsigned s;
+        unsigned largest=0;
+        short largestP=0;
+        U32 lowThreshold = (U32)(total >> tableLog);
+
+        for (s=0; s<=maxSymbolValue; s++) {
+            if (count[s] == total) return 0;   /* rle special case */
+            if (count[s] == 0) { normalizedCounter[s]=0; continue; }
+            if (count[s] <= lowThreshold) {
+                normalizedCounter[s] = lowProbCount;
+                stillToDistribute--;
+            } else {
+                short proba = (short)((count[s]*step) >> scale);
+                if (proba<8) {
+                    U64 restToBeat = vStep * rtbTable[proba];
+                    proba += (count[s]*step) - ((U64)proba<<scale) > restToBeat;
+                }
+                if (proba > largestP) { largestP=proba; largest=s; }
+                normalizedCounter[s] = proba;
+                stillToDistribute -= proba;
+        }   }
+        if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
+            /* corner case, need another normalization method */
+            size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue, lowProbCount);
+            if (FSE_isError(errorCode)) return errorCode;
+        }
+        else normalizedCounter[largest] += (short)stillToDistribute;
+    }
+
+#if 0
+    {   /* Print Table (debug) */
+        U32 s;
+        U32 nTotal = 0;
+        for (s=0; s<=maxSymbolValue; s++)
+            RAWLOG(2, "%3i: %4i \n", s, normalizedCounter[s]);
+        for (s=0; s<=maxSymbolValue; s++)
+            nTotal += abs(normalizedCounter[s]);
+        if (nTotal != (1U<<tableLog))
+            RAWLOG(2, "Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
+        getchar();
+    }
+#endif
+
+    return tableLog;
+}
+
+/* fake FSE_CTable, for rle input (always same symbol) */
+size_t FSE_buildCTable_rle (FSE_CTable* ct, BYTE symbolValue)
+{
+    void* ptr = ct;
+    U16* tableU16 = ( (U16*) ptr) + 2;
+    void* FSCTptr = (U32*)ptr + 2;
+    FSE_symbolCompressionTransform* symbolTT = (FSE_symbolCompressionTransform*) FSCTptr;
+
+    /* header */
+    tableU16[-2] = (U16) 0;
+    tableU16[-1] = (U16) symbolValue;
+
+    /* Build table */
+    tableU16[0] = 0;
+    tableU16[1] = 0;   /* just in case */
+
+    /* Build Symbol Transformation Table */
+    symbolTT[symbolValue].deltaNbBits = 0;
+    symbolTT[symbolValue].deltaFindState = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_compress_usingCTable_generic (void* dst, size_t dstSize,
+                           const void* src, size_t srcSize,
+                           const FSE_CTable* ct, const unsigned fast)
+{
+    const BYTE* const istart = (const BYTE*) src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip=iend;
+
+    BIT_CStream_t bitC;
+    FSE_CState_t CState1, CState2;
+
+    /* init */
+    if (srcSize <= 2) return 0;
+    { size_t const initError = BIT_initCStream(&bitC, dst, dstSize);
+      if (FSE_isError(initError)) return 0; /* not enough space available to write a bitstream */ }
+
+#define FSE_FLUSHBITS(s)  (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
+
+    if (srcSize & 1) {
+        FSE_initCState2(&CState1, ct, *--ip);
+        FSE_initCState2(&CState2, ct, *--ip);
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        FSE_FLUSHBITS(&bitC);
+    } else {
+        FSE_initCState2(&CState2, ct, *--ip);
+        FSE_initCState2(&CState1, ct, *--ip);
+    }
+
+    /* join to mod 4 */
+    srcSize -= 2;
+    if ((sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) && (srcSize & 2)) {  /* test bit 2 */
+        FSE_encodeSymbol(&bitC, &CState2, *--ip);
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        FSE_FLUSHBITS(&bitC);
+    }
+
+    /* 2 or 4 encoding per loop */
+    while ( ip>istart ) {
+
+        FSE_encodeSymbol(&bitC, &CState2, *--ip);
+
+        if (sizeof(bitC.bitContainer)*8 < FSE_MAX_TABLELOG*2+7 )   /* this test must be static */
+            FSE_FLUSHBITS(&bitC);
+
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+
+        if (sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) {  /* this test must be static */
+            FSE_encodeSymbol(&bitC, &CState2, *--ip);
+            FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        }
+
+        FSE_FLUSHBITS(&bitC);
+    }
+
+    FSE_flushCState(&bitC, &CState2);
+    FSE_flushCState(&bitC, &CState1);
+    return BIT_closeCStream(&bitC);
+}
+
+size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
+                           const void* src, size_t srcSize,
+                           const FSE_CTable* ct)
+{
+    unsigned const fast = (dstSize >= FSE_BLOCKBOUND(srcSize));
+
+    if (fast)
+        return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 1);
+    else
+        return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 0);
+}
+
+
+size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
+
+#endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/lib/compress/hist.c b/lib/compress/hist.c
new file mode 100644
index 0000000..4ccf9a9
--- /dev/null
+++ b/lib/compress/hist.c
@@ -0,0 +1,191 @@
+/* ******************************************************************
+ * hist : Histogram functions
+ * part of Finite State Entropy project
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* --- dependencies --- */
+#include "../common/mem.h"             /* U32, BYTE, etc. */
+#include "../common/debug.h"           /* assert, DEBUGLOG */
+#include "../common/error_private.h"   /* ERROR */
+#include "hist.h"
+
+
+/* --- Error management --- */
+unsigned HIST_isError(size_t code) { return ERR_isError(code); }
+
+/*-**************************************************************
+ *  Histogram functions
+ ****************************************************************/
+void HIST_add(unsigned* count, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const end = ip + srcSize;
+
+    while (ip<end) {
+        count[*ip++]++;
+    }
+}
+
+unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
+                           const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const end = ip + srcSize;
+    unsigned maxSymbolValue = *maxSymbolValuePtr;
+    unsigned largestCount=0;
+
+    ZSTD_memset(count, 0, (maxSymbolValue+1) * sizeof(*count));
+    if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
+
+    while (ip<end) {
+        assert(*ip <= maxSymbolValue);
+        count[*ip++]++;
+    }
+
+    while (!count[maxSymbolValue]) maxSymbolValue--;
+    *maxSymbolValuePtr = maxSymbolValue;
+
+    {   U32 s;
+        for (s=0; s<=maxSymbolValue; s++)
+            if (count[s] > largestCount) largestCount = count[s];
+    }
+
+    return largestCount;
+}
+
+typedef enum { trustInput, checkMaxSymbolValue } HIST_checkInput_e;
+
+/* HIST_count_parallel_wksp() :
+ * store histogram into 4 intermediate tables, recombined at the end.
+ * this design makes better use of OoO cpus,
+ * and is noticeably faster when some values are heavily repeated.
+ * But it needs some additional workspace for intermediate tables.
+ * `workSpace` must be a U32 table of size >= HIST_WKSP_SIZE_U32.
+ * @return : largest histogram frequency,
+ *           or an error code (notably when histogram's alphabet is larger than *maxSymbolValuePtr) */
+static size_t HIST_count_parallel_wksp(
+                                unsigned* count, unsigned* maxSymbolValuePtr,
+                                const void* source, size_t sourceSize,
+                                HIST_checkInput_e check,
+                                U32* const workSpace)
+{
+    const BYTE* ip = (const BYTE*)source;
+    const BYTE* const iend = ip+sourceSize;
+    size_t const countSize = (*maxSymbolValuePtr + 1) * sizeof(*count);
+    unsigned max=0;
+    U32* const Counting1 = workSpace;
+    U32* const Counting2 = Counting1 + 256;
+    U32* const Counting3 = Counting2 + 256;
+    U32* const Counting4 = Counting3 + 256;
+
+    /* safety checks */
+    assert(*maxSymbolValuePtr <= 255);
+    if (!sourceSize) {
+        ZSTD_memset(count, 0, countSize);
+        *maxSymbolValuePtr = 0;
+        return 0;
+    }
+    ZSTD_memset(workSpace, 0, 4*256*sizeof(unsigned));
+
+    /* by stripes of 16 bytes */
+    {   U32 cached = MEM_read32(ip); ip += 4;
+        while (ip < iend-15) {
+            U32 c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+        }
+        ip-=4;
+    }
+
+    /* finish last symbols */
+    while (ip<iend) Counting1[*ip++]++;
+
+    {   U32 s;
+        for (s=0; s<256; s++) {
+            Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
+            if (Counting1[s] > max) max = Counting1[s];
+    }   }
+
+    {   unsigned maxSymbolValue = 255;
+        while (!Counting1[maxSymbolValue]) maxSymbolValue--;
+        if (check && maxSymbolValue > *maxSymbolValuePtr) return ERROR(maxSymbolValue_tooSmall);
+        *maxSymbolValuePtr = maxSymbolValue;
+        ZSTD_memmove(count, Counting1, countSize);   /* in case count & Counting1 are overlapping */
+    }
+    return (size_t)max;
+}
+
+/* HIST_countFast_wksp() :
+ * Same as HIST_countFast(), but using an externally provided scratch buffer.
+ * `workSpace` is a writable buffer which must be 4-bytes aligned,
+ * `workSpaceSize` must be >= HIST_WKSP_SIZE
+ */
+size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                          const void* source, size_t sourceSize,
+                          void* workSpace, size_t workSpaceSize)
+{
+    if (sourceSize < 1500) /* heuristic threshold */
+        return HIST_count_simple(count, maxSymbolValuePtr, source, sourceSize);
+    if ((size_t)workSpace & 3) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (workSpaceSize < HIST_WKSP_SIZE) return ERROR(workSpace_tooSmall);
+    return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, trustInput, (U32*)workSpace);
+}
+
+/* HIST_count_wksp() :
+ * Same as HIST_count(), but using an externally provided scratch buffer.
+ * `workSpace` size must be table of >= HIST_WKSP_SIZE_U32 unsigned */
+size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                       const void* source, size_t sourceSize,
+                       void* workSpace, size_t workSpaceSize)
+{
+    if ((size_t)workSpace & 3) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (workSpaceSize < HIST_WKSP_SIZE) return ERROR(workSpace_tooSmall);
+    if (*maxSymbolValuePtr < 255)
+        return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, checkMaxSymbolValue, (U32*)workSpace);
+    *maxSymbolValuePtr = 255;
+    return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace, workSpaceSize);
+}
+
+#ifndef ZSTD_NO_UNUSED_FUNCTIONS
+/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
+size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
+                     const void* source, size_t sourceSize)
+{
+    unsigned tmpCounters[HIST_WKSP_SIZE_U32];
+    return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters, sizeof(tmpCounters));
+}
+
+size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
+                 const void* src, size_t srcSize)
+{
+    unsigned tmpCounters[HIST_WKSP_SIZE_U32];
+    return HIST_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters, sizeof(tmpCounters));
+}
+#endif
diff --git a/lib/compress/huf_compress.c b/lib/compress/huf_compress.c
new file mode 100644
index 0000000..6f2194e
--- /dev/null
+++ b/lib/compress/huf_compress.c
@@ -0,0 +1,1465 @@
+/* ******************************************************************
+ * Huffman encoder, part of New Generation Entropy library
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include "../common/zstd_deps.h"     /* ZSTD_memcpy, ZSTD_memset */
+#include "../common/compiler.h"
+#include "../common/bitstream.h"
+#include "hist.h"
+#define FSE_STATIC_LINKING_ONLY   /* FSE_optimalTableLog_internal */
+#include "../common/fse.h"        /* header compression */
+#include "../common/huf.h"
+#include "../common/error_private.h"
+#include "../common/bits.h"       /* ZSTD_highbit32 */
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_isError ERR_isError
+#define HUF_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Required declarations
+****************************************************************/
+typedef struct nodeElt_s {
+    U32 count;
+    U16 parent;
+    BYTE byte;
+    BYTE nbBits;
+} nodeElt;
+
+
+/* **************************************************************
+*  Debug Traces
+****************************************************************/
+
+#if DEBUGLEVEL >= 2
+
+static size_t showU32(const U32* arr, size_t size)
+{
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %u", arr[u]); (void)arr;
+    }
+    RAWLOG(6, " \n");
+    return size;
+}
+
+static size_t HUF_getNbBits(HUF_CElt elt);
+
+static size_t showCTableBits(const HUF_CElt* ctable, size_t size)
+{
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %zu", HUF_getNbBits(ctable[u])); (void)ctable;
+    }
+    RAWLOG(6, " \n");
+    return size;
+
+}
+
+static size_t showHNodeSymbols(const nodeElt* hnode, size_t size)
+{
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %u", hnode[u].byte); (void)hnode;
+    }
+    RAWLOG(6, " \n");
+    return size;
+}
+
+static size_t showHNodeBits(const nodeElt* hnode, size_t size)
+{
+    size_t u;
+    for (u=0; u<size; u++) {
+        RAWLOG(6, " %u", hnode[u].nbBits); (void)hnode;
+    }
+    RAWLOG(6, " \n");
+    return size;
+}
+
+#endif
+
+
+/* *******************************************************
+*  HUF : Huffman block compression
+*********************************************************/
+#define HUF_WORKSPACE_MAX_ALIGNMENT 8
+
+static void* HUF_alignUpWorkspace(void* workspace, size_t* workspaceSizePtr, size_t align)
+{
+    size_t const mask = align - 1;
+    size_t const rem = (size_t)workspace & mask;
+    size_t const add = (align - rem) & mask;
+    BYTE* const aligned = (BYTE*)workspace + add;
+    assert((align & (align - 1)) == 0); /* pow 2 */
+    assert(align <= HUF_WORKSPACE_MAX_ALIGNMENT);
+    if (*workspaceSizePtr >= add) {
+        assert(add < align);
+        assert(((size_t)aligned & mask) == 0);
+        *workspaceSizePtr -= add;
+        return aligned;
+    } else {
+        *workspaceSizePtr = 0;
+        return NULL;
+    }
+}
+
+
+/* HUF_compressWeights() :
+ * Same as FSE_compress(), but dedicated to huff0's weights compression.
+ * The use case needs much less stack memory.
+ * Note : all elements within weightTable are supposed to be <= HUF_TABLELOG_MAX.
+ */
+#define MAX_FSE_TABLELOG_FOR_HUFF_HEADER 6
+
+typedef struct {
+    FSE_CTable CTable[FSE_CTABLE_SIZE_U32(MAX_FSE_TABLELOG_FOR_HUFF_HEADER, HUF_TABLELOG_MAX)];
+    U32 scratchBuffer[FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(HUF_TABLELOG_MAX, MAX_FSE_TABLELOG_FOR_HUFF_HEADER)];
+    unsigned count[HUF_TABLELOG_MAX+1];
+    S16 norm[HUF_TABLELOG_MAX+1];
+} HUF_CompressWeightsWksp;
+
+static size_t
+HUF_compressWeights(void* dst, size_t dstSize,
+              const void* weightTable, size_t wtSize,
+                    void* workspace, size_t workspaceSize)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + dstSize;
+
+    unsigned maxSymbolValue = HUF_TABLELOG_MAX;
+    U32 tableLog = MAX_FSE_TABLELOG_FOR_HUFF_HEADER;
+    HUF_CompressWeightsWksp* wksp = (HUF_CompressWeightsWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, ZSTD_ALIGNOF(U32));
+
+    if (workspaceSize < sizeof(HUF_CompressWeightsWksp)) return ERROR(GENERIC);
+
+    /* init conditions */
+    if (wtSize <= 1) return 0;  /* Not compressible */
+
+    /* Scan input and build symbol stats */
+    {   unsigned const maxCount = HIST_count_simple(wksp->count, &maxSymbolValue, weightTable, wtSize);   /* never fails */
+        if (maxCount == wtSize) return 1;   /* only a single symbol in src : rle */
+        if (maxCount == 1) return 0;        /* each symbol present maximum once => not compressible */
+    }
+
+    tableLog = FSE_optimalTableLog(tableLog, wtSize, maxSymbolValue);
+    CHECK_F( FSE_normalizeCount(wksp->norm, tableLog, wksp->count, wtSize, maxSymbolValue, /* useLowProbCount */ 0) );
+
+    /* Write table description header */
+    {   CHECK_V_F(hSize, FSE_writeNCount(op, (size_t)(oend-op), wksp->norm, maxSymbolValue, tableLog) );
+        op += hSize;
+    }
+
+    /* Compress */
+    CHECK_F( FSE_buildCTable_wksp(wksp->CTable, wksp->norm, maxSymbolValue, tableLog, wksp->scratchBuffer, sizeof(wksp->scratchBuffer)) );
+    {   CHECK_V_F(cSize, FSE_compress_usingCTable(op, (size_t)(oend - op), weightTable, wtSize, wksp->CTable) );
+        if (cSize == 0) return 0;   /* not enough space for compressed data */
+        op += cSize;
+    }
+
+    return (size_t)(op-ostart);
+}
+
+static size_t HUF_getNbBits(HUF_CElt elt)
+{
+    return elt & 0xFF;
+}
+
+static size_t HUF_getNbBitsFast(HUF_CElt elt)
+{
+    return elt;
+}
+
+static size_t HUF_getValue(HUF_CElt elt)
+{
+    return elt & ~(size_t)0xFF;
+}
+
+static size_t HUF_getValueFast(HUF_CElt elt)
+{
+    return elt;
+}
+
+static void HUF_setNbBits(HUF_CElt* elt, size_t nbBits)
+{
+    assert(nbBits <= HUF_TABLELOG_ABSOLUTEMAX);
+    *elt = nbBits;
+}
+
+static void HUF_setValue(HUF_CElt* elt, size_t value)
+{
+    size_t const nbBits = HUF_getNbBits(*elt);
+    if (nbBits > 0) {
+        assert((value >> nbBits) == 0);
+        *elt |= value << (sizeof(HUF_CElt) * 8 - nbBits);
+    }
+}
+
+HUF_CTableHeader HUF_readCTableHeader(HUF_CElt const* ctable)
+{
+    HUF_CTableHeader header;
+    ZSTD_memcpy(&header, ctable, sizeof(header));
+    return header;
+}
+
+static void HUF_writeCTableHeader(HUF_CElt* ctable, U32 tableLog, U32 maxSymbolValue)
+{
+    HUF_CTableHeader header;
+    HUF_STATIC_ASSERT(sizeof(ctable[0]) == sizeof(header));
+    ZSTD_memset(&header, 0, sizeof(header));
+    assert(tableLog < 256);
+    header.tableLog = (BYTE)tableLog;
+    assert(maxSymbolValue < 256);
+    header.maxSymbolValue = (BYTE)maxSymbolValue;
+    ZSTD_memcpy(ctable, &header, sizeof(header));
+}
+
+typedef struct {
+    HUF_CompressWeightsWksp wksp;
+    BYTE bitsToWeight[HUF_TABLELOG_MAX + 1];   /* precomputed conversion table */
+    BYTE huffWeight[HUF_SYMBOLVALUE_MAX];
+} HUF_WriteCTableWksp;
+
+size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize,
+                            const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog,
+                            void* workspace, size_t workspaceSize)
+{
+    HUF_CElt const* const ct = CTable + 1;
+    BYTE* op = (BYTE*)dst;
+    U32 n;
+    HUF_WriteCTableWksp* wksp = (HUF_WriteCTableWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, ZSTD_ALIGNOF(U32));
+
+    HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE >= sizeof(HUF_WriteCTableWksp));
+
+    assert(HUF_readCTableHeader(CTable).maxSymbolValue == maxSymbolValue);
+    assert(HUF_readCTableHeader(CTable).tableLog == huffLog);
+
+    /* check conditions */
+    if (workspaceSize < sizeof(HUF_WriteCTableWksp)) return ERROR(GENERIC);
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
+
+    /* convert to weight */
+    wksp->bitsToWeight[0] = 0;
+    for (n=1; n<huffLog+1; n++)
+        wksp->bitsToWeight[n] = (BYTE)(huffLog + 1 - n);
+    for (n=0; n<maxSymbolValue; n++)
+        wksp->huffWeight[n] = wksp->bitsToWeight[HUF_getNbBits(ct[n])];
+
+    /* attempt weights compression by FSE */
+    if (maxDstSize < 1) return ERROR(dstSize_tooSmall);
+    {   CHECK_V_F(hSize, HUF_compressWeights(op+1, maxDstSize-1, wksp->huffWeight, maxSymbolValue, &wksp->wksp, sizeof(wksp->wksp)) );
+        if ((hSize>1) & (hSize < maxSymbolValue/2)) {   /* FSE compressed */
+            op[0] = (BYTE)hSize;
+            return hSize+1;
+    }   }
+
+    /* write raw values as 4-bits (max : 15) */
+    if (maxSymbolValue > (256-128)) return ERROR(GENERIC);   /* should not happen : likely means source cannot be compressed */
+    if (((maxSymbolValue+1)/2) + 1 > maxDstSize) return ERROR(dstSize_tooSmall);   /* not enough space within dst buffer */
+    op[0] = (BYTE)(128 /*special case*/ + (maxSymbolValue-1));
+    wksp->huffWeight[maxSymbolValue] = 0;   /* to be sure it doesn't cause msan issue in final combination */
+    for (n=0; n<maxSymbolValue; n+=2)
+        op[(n/2)+1] = (BYTE)((wksp->huffWeight[n] << 4) + wksp->huffWeight[n+1]);
+    return ((maxSymbolValue+1)/2) + 1;
+}
+
+
+size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* hasZeroWeights)
+{
+    BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];   /* init not required, even though some static analyzer may complain */
+    U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    U32 nbSymbols = 0;
+    HUF_CElt* const ct = CTable + 1;
+
+    /* get symbol weights */
+    CHECK_V_F(readSize, HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX+1, rankVal, &nbSymbols, &tableLog, src, srcSize));
+    *hasZeroWeights = (rankVal[0] > 0);
+
+    /* check result */
+    if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    if (nbSymbols > *maxSymbolValuePtr+1) return ERROR(maxSymbolValue_tooSmall);
+
+    *maxSymbolValuePtr = nbSymbols - 1;
+
+    HUF_writeCTableHeader(CTable, tableLog, *maxSymbolValuePtr);
+
+    /* Prepare base value per rank */
+    {   U32 n, nextRankStart = 0;
+        for (n=1; n<=tableLog; n++) {
+            U32 curr = nextRankStart;
+            nextRankStart += (rankVal[n] << (n-1));
+            rankVal[n] = curr;
+    }   }
+
+    /* fill nbBits */
+    {   U32 n; for (n=0; n<nbSymbols; n++) {
+            const U32 w = huffWeight[n];
+            HUF_setNbBits(ct + n, (BYTE)(tableLog + 1 - w) & -(w != 0));
+    }   }
+
+    /* fill val */
+    {   U16 nbPerRank[HUF_TABLELOG_MAX+2]  = {0};  /* support w=0=>n=tableLog+1 */
+        U16 valPerRank[HUF_TABLELOG_MAX+2] = {0};
+        { U32 n; for (n=0; n<nbSymbols; n++) nbPerRank[HUF_getNbBits(ct[n])]++; }
+        /* determine stating value per rank */
+        valPerRank[tableLog+1] = 0;   /* for w==0 */
+        {   U16 min = 0;
+            U32 n; for (n=tableLog; n>0; n--) {  /* start at n=tablelog <-> w=1 */
+                valPerRank[n] = min;     /* get starting value within each rank */
+                min += nbPerRank[n];
+                min >>= 1;
+        }   }
+        /* assign value within rank, symbol order */
+        { U32 n; for (n=0; n<nbSymbols; n++) HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++); }
+    }
+
+    return readSize;
+}
+
+U32 HUF_getNbBitsFromCTable(HUF_CElt const* CTable, U32 symbolValue)
+{
+    const HUF_CElt* const ct = CTable + 1;
+    assert(symbolValue <= HUF_SYMBOLVALUE_MAX);
+    if (symbolValue > HUF_readCTableHeader(CTable).maxSymbolValue)
+        return 0;
+    return (U32)HUF_getNbBits(ct[symbolValue]);
+}
+
+
+/**
+ * HUF_setMaxHeight():
+ * Try to enforce @targetNbBits on the Huffman tree described in @huffNode.
+ *
+ * It attempts to convert all nodes with nbBits > @targetNbBits
+ * to employ @targetNbBits instead. Then it adjusts the tree
+ * so that it remains a valid canonical Huffman tree.
+ *
+ * @pre               The sum of the ranks of each symbol == 2^largestBits,
+ *                    where largestBits == huffNode[lastNonNull].nbBits.
+ * @post              The sum of the ranks of each symbol == 2^largestBits,
+ *                    where largestBits is the return value (expected <= targetNbBits).
+ *
+ * @param huffNode    The Huffman tree modified in place to enforce targetNbBits.
+ *                    It's presumed sorted, from most frequent to rarest symbol.
+ * @param lastNonNull The symbol with the lowest count in the Huffman tree.
+ * @param targetNbBits  The allowed number of bits, which the Huffman tree
+ *                    may not respect. After this function the Huffman tree will
+ *                    respect targetNbBits.
+ * @return            The maximum number of bits of the Huffman tree after adjustment.
+ */
+static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 targetNbBits)
+{
+    const U32 largestBits = huffNode[lastNonNull].nbBits;
+    /* early exit : no elt > targetNbBits, so the tree is already valid. */
+    if (largestBits <= targetNbBits) return largestBits;
+
+    DEBUGLOG(5, "HUF_setMaxHeight (targetNbBits = %u)", targetNbBits);
+
+    /* there are several too large elements (at least >= 2) */
+    {   int totalCost = 0;
+        const U32 baseCost = 1 << (largestBits - targetNbBits);
+        int n = (int)lastNonNull;
+
+        /* Adjust any ranks > targetNbBits to targetNbBits.
+         * Compute totalCost, which is how far the sum of the ranks is
+         * we are over 2^largestBits after adjust the offending ranks.
+         */
+        while (huffNode[n].nbBits > targetNbBits) {
+            totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits));
+            huffNode[n].nbBits = (BYTE)targetNbBits;
+            n--;
+        }
+        /* n stops at huffNode[n].nbBits <= targetNbBits */
+        assert(huffNode[n].nbBits <= targetNbBits);
+        /* n end at index of smallest symbol using < targetNbBits */
+        while (huffNode[n].nbBits == targetNbBits) --n;
+
+        /* renorm totalCost from 2^largestBits to 2^targetNbBits
+         * note : totalCost is necessarily a multiple of baseCost */
+        assert(((U32)totalCost & (baseCost - 1)) == 0);
+        totalCost >>= (largestBits - targetNbBits);
+        assert(totalCost > 0);
+
+        /* repay normalized cost */
+        {   U32 const noSymbol = 0xF0F0F0F0;
+            U32 rankLast[HUF_TABLELOG_MAX+2];
+
+            /* Get pos of last (smallest = lowest cum. count) symbol per rank */
+            ZSTD_memset(rankLast, 0xF0, sizeof(rankLast));
+            {   U32 currentNbBits = targetNbBits;
+                int pos;
+                for (pos=n ; pos >= 0; pos--) {
+                    if (huffNode[pos].nbBits >= currentNbBits) continue;
+                    currentNbBits = huffNode[pos].nbBits;   /* < targetNbBits */
+                    rankLast[targetNbBits-currentNbBits] = (U32)pos;
+            }   }
+
+            while (totalCost > 0) {
+                /* Try to reduce the next power of 2 above totalCost because we
+                 * gain back half the rank.
+                 */
+                U32 nBitsToDecrease = ZSTD_highbit32((U32)totalCost) + 1;
+                assert(nBitsToDecrease <= HUF_TABLELOG_MAX+1);
+                for ( ; nBitsToDecrease > 1; nBitsToDecrease--) {
+                    U32 const highPos = rankLast[nBitsToDecrease];
+                    U32 const lowPos = rankLast[nBitsToDecrease-1];
+                    if (highPos == noSymbol) continue;
+                    /* Decrease highPos if no symbols of lowPos or if it is
+                     * not cheaper to remove 2 lowPos than highPos.
+                     */
+                    if (lowPos == noSymbol) break;
+                    {   U32 const highTotal = huffNode[highPos].count;
+                        U32 const lowTotal = 2 * huffNode[lowPos].count;
+                        if (highTotal <= lowTotal) break;
+                }   }
+                /* only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !) */
+                assert(rankLast[nBitsToDecrease] != noSymbol || nBitsToDecrease == 1);
+                /* HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary */
+                while ((nBitsToDecrease<=HUF_TABLELOG_MAX) && (rankLast[nBitsToDecrease] == noSymbol))
+                    nBitsToDecrease++;
+                assert(rankLast[nBitsToDecrease] != noSymbol);
+                /* Increase the number of bits to gain back half the rank cost. */
+                totalCost -= 1 << (nBitsToDecrease-1);
+                huffNode[rankLast[nBitsToDecrease]].nbBits++;
+
+                /* Fix up the new rank.
+                 * If the new rank was empty, this symbol is now its smallest.
+                 * Otherwise, this symbol will be the largest in the new rank so no adjustment.
+                 */
+                if (rankLast[nBitsToDecrease-1] == noSymbol)
+                    rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease];
+                /* Fix up the old rank.
+                 * If the symbol was at position 0, meaning it was the highest weight symbol in the tree,
+                 * it must be the only symbol in its rank, so the old rank now has no symbols.
+                 * Otherwise, since the Huffman nodes are sorted by count, the previous position is now
+                 * the smallest node in the rank. If the previous position belongs to a different rank,
+                 * then the rank is now empty.
+                 */
+                if (rankLast[nBitsToDecrease] == 0)    /* special case, reached largest symbol */
+                    rankLast[nBitsToDecrease] = noSymbol;
+                else {
+                    rankLast[nBitsToDecrease]--;
+                    if (huffNode[rankLast[nBitsToDecrease]].nbBits != targetNbBits-nBitsToDecrease)
+                        rankLast[nBitsToDecrease] = noSymbol;   /* this rank is now empty */
+                }
+            }   /* while (totalCost > 0) */
+
+            /* If we've removed too much weight, then we have to add it back.
+             * To avoid overshooting again, we only adjust the smallest rank.
+             * We take the largest nodes from the lowest rank 0 and move them
+             * to rank 1. There's guaranteed to be enough rank 0 symbols because
+             * TODO.
+             */
+            while (totalCost < 0) {  /* Sometimes, cost correction overshoot */
+                /* special case : no rank 1 symbol (using targetNbBits-1);
+                 * let's create one from largest rank 0 (using targetNbBits).
+                 */
+                if (rankLast[1] == noSymbol) {
+                    while (huffNode[n].nbBits == targetNbBits) n--;
+                    huffNode[n+1].nbBits--;
+                    assert(n >= 0);
+                    rankLast[1] = (U32)(n+1);
+                    totalCost++;
+                    continue;
+                }
+                huffNode[ rankLast[1] + 1 ].nbBits--;
+                rankLast[1]++;
+                totalCost ++;
+            }
+        }   /* repay normalized cost */
+    }   /* there are several too large elements (at least >= 2) */
+
+    return targetNbBits;
+}
+
+typedef struct {
+    U16 base;
+    U16 curr;
+} rankPos;
+
+typedef nodeElt huffNodeTable[2 * (HUF_SYMBOLVALUE_MAX + 1)];
+
+/* Number of buckets available for HUF_sort() */
+#define RANK_POSITION_TABLE_SIZE 192
+
+typedef struct {
+  huffNodeTable huffNodeTbl;
+  rankPos rankPosition[RANK_POSITION_TABLE_SIZE];
+} HUF_buildCTable_wksp_tables;
+
+/* RANK_POSITION_DISTINCT_COUNT_CUTOFF == Cutoff point in HUF_sort() buckets for which we use log2 bucketing.
+ * Strategy is to use as many buckets as possible for representing distinct
+ * counts while using the remainder to represent all "large" counts.
+ *
+ * To satisfy this requirement for 192 buckets, we can do the following:
+ * Let buckets 0-166 represent distinct counts of [0, 166]
+ * Let buckets 166 to 192 represent all remaining counts up to RANK_POSITION_MAX_COUNT_LOG using log2 bucketing.
+ */
+#define RANK_POSITION_MAX_COUNT_LOG 32
+#define RANK_POSITION_LOG_BUCKETS_BEGIN ((RANK_POSITION_TABLE_SIZE - 1) - RANK_POSITION_MAX_COUNT_LOG - 1 /* == 158 */)
+#define RANK_POSITION_DISTINCT_COUNT_CUTOFF (RANK_POSITION_LOG_BUCKETS_BEGIN + ZSTD_highbit32(RANK_POSITION_LOG_BUCKETS_BEGIN) /* == 166 */)
+
+/* Return the appropriate bucket index for a given count. See definition of
+ * RANK_POSITION_DISTINCT_COUNT_CUTOFF for explanation of bucketing strategy.
+ */
+static U32 HUF_getIndex(U32 const count) {
+    return (count < RANK_POSITION_DISTINCT_COUNT_CUTOFF)
+        ? count
+        : ZSTD_highbit32(count) + RANK_POSITION_LOG_BUCKETS_BEGIN;
+}
+
+/* Helper swap function for HUF_quickSortPartition() */
+static void HUF_swapNodes(nodeElt* a, nodeElt* b) {
+	nodeElt tmp = *a;
+	*a = *b;
+	*b = tmp;
+}
+
+/* Returns 0 if the huffNode array is not sorted by descending count */
+MEM_STATIC int HUF_isSorted(nodeElt huffNode[], U32 const maxSymbolValue1) {
+    U32 i;
+    for (i = 1; i < maxSymbolValue1; ++i) {
+        if (huffNode[i].count > huffNode[i-1].count) {
+            return 0;
+        }
+    }
+    return 1;
+}
+
+/* Insertion sort by descending order */
+HINT_INLINE void HUF_insertionSort(nodeElt huffNode[], int const low, int const high) {
+    int i;
+    int const size = high-low+1;
+    huffNode += low;
+    for (i = 1; i < size; ++i) {
+        nodeElt const key = huffNode[i];
+        int j = i - 1;
+        while (j >= 0 && huffNode[j].count < key.count) {
+            huffNode[j + 1] = huffNode[j];
+            j--;
+        }
+        huffNode[j + 1] = key;
+    }
+}
+
+/* Pivot helper function for quicksort. */
+static int HUF_quickSortPartition(nodeElt arr[], int const low, int const high) {
+    /* Simply select rightmost element as pivot. "Better" selectors like
+     * median-of-three don't experimentally appear to have any benefit.
+     */
+    U32 const pivot = arr[high].count;
+    int i = low - 1;
+    int j = low;
+    for ( ; j < high; j++) {
+        if (arr[j].count > pivot) {
+            i++;
+            HUF_swapNodes(&arr[i], &arr[j]);
+        }
+    }
+    HUF_swapNodes(&arr[i + 1], &arr[high]);
+    return i + 1;
+}
+
+/* Classic quicksort by descending with partially iterative calls
+ * to reduce worst case callstack size.
+ */
+static void HUF_simpleQuickSort(nodeElt arr[], int low, int high) {
+    int const kInsertionSortThreshold = 8;
+    if (high - low < kInsertionSortThreshold) {
+        HUF_insertionSort(arr, low, high);
+        return;
+    }
+    while (low < high) {
+        int const idx = HUF_quickSortPartition(arr, low, high);
+        if (idx - low < high - idx) {
+            HUF_simpleQuickSort(arr, low, idx - 1);
+            low = idx + 1;
+        } else {
+            HUF_simpleQuickSort(arr, idx + 1, high);
+            high = idx - 1;
+        }
+    }
+}
+
+/**
+ * HUF_sort():
+ * Sorts the symbols [0, maxSymbolValue] by count[symbol] in decreasing order.
+ * This is a typical bucket sorting strategy that uses either quicksort or insertion sort to sort each bucket.
+ *
+ * @param[out] huffNode       Sorted symbols by decreasing count. Only members `.count` and `.byte` are filled.
+ *                            Must have (maxSymbolValue + 1) entries.
+ * @param[in]  count          Histogram of the symbols.
+ * @param[in]  maxSymbolValue Maximum symbol value.
+ * @param      rankPosition   This is a scratch workspace. Must have RANK_POSITION_TABLE_SIZE entries.
+ */
+static void HUF_sort(nodeElt huffNode[], const unsigned count[], U32 const maxSymbolValue, rankPos rankPosition[]) {
+    U32 n;
+    U32 const maxSymbolValue1 = maxSymbolValue+1;
+
+    /* Compute base and set curr to base.
+     * For symbol s let lowerRank = HUF_getIndex(count[n]) and rank = lowerRank + 1.
+     * See HUF_getIndex to see bucketing strategy.
+     * We attribute each symbol to lowerRank's base value, because we want to know where
+     * each rank begins in the output, so for rank R we want to count ranks R+1 and above.
+     */
+    ZSTD_memset(rankPosition, 0, sizeof(*rankPosition) * RANK_POSITION_TABLE_SIZE);
+    for (n = 0; n < maxSymbolValue1; ++n) {
+        U32 lowerRank = HUF_getIndex(count[n]);
+        assert(lowerRank < RANK_POSITION_TABLE_SIZE - 1);
+        rankPosition[lowerRank].base++;
+    }
+
+    assert(rankPosition[RANK_POSITION_TABLE_SIZE - 1].base == 0);
+    /* Set up the rankPosition table */
+    for (n = RANK_POSITION_TABLE_SIZE - 1; n > 0; --n) {
+        rankPosition[n-1].base += rankPosition[n].base;
+        rankPosition[n-1].curr = rankPosition[n-1].base;
+    }
+
+    /* Insert each symbol into their appropriate bucket, setting up rankPosition table. */
+    for (n = 0; n < maxSymbolValue1; ++n) {
+        U32 const c = count[n];
+        U32 const r = HUF_getIndex(c) + 1;
+        U32 const pos = rankPosition[r].curr++;
+        assert(pos < maxSymbolValue1);
+        huffNode[pos].count = c;
+        huffNode[pos].byte  = (BYTE)n;
+    }
+
+    /* Sort each bucket. */
+    for (n = RANK_POSITION_DISTINCT_COUNT_CUTOFF; n < RANK_POSITION_TABLE_SIZE - 1; ++n) {
+        int const bucketSize = rankPosition[n].curr - rankPosition[n].base;
+        U32 const bucketStartIdx = rankPosition[n].base;
+        if (bucketSize > 1) {
+            assert(bucketStartIdx < maxSymbolValue1);
+            HUF_simpleQuickSort(huffNode + bucketStartIdx, 0, bucketSize-1);
+        }
+    }
+
+    assert(HUF_isSorted(huffNode, maxSymbolValue1));
+}
+
+
+/** HUF_buildCTable_wksp() :
+ *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
+ *  `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as sizeof(HUF_buildCTable_wksp_tables).
+ */
+#define STARTNODE (HUF_SYMBOLVALUE_MAX+1)
+
+/* HUF_buildTree():
+ * Takes the huffNode array sorted by HUF_sort() and builds an unlimited-depth Huffman tree.
+ *
+ * @param huffNode        The array sorted by HUF_sort(). Builds the Huffman tree in this array.
+ * @param maxSymbolValue  The maximum symbol value.
+ * @return                The smallest node in the Huffman tree (by count).
+ */
+static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
+{
+    nodeElt* const huffNode0 = huffNode - 1;
+    int nonNullRank;
+    int lowS, lowN;
+    int nodeNb = STARTNODE;
+    int n, nodeRoot;
+    DEBUGLOG(5, "HUF_buildTree (alphabet size = %u)", maxSymbolValue + 1);
+    /* init for parents */
+    nonNullRank = (int)maxSymbolValue;
+    while(huffNode[nonNullRank].count == 0) nonNullRank--;
+    lowS = nonNullRank; nodeRoot = nodeNb + lowS - 1; lowN = nodeNb;
+    huffNode[nodeNb].count = huffNode[lowS].count + huffNode[lowS-1].count;
+    huffNode[lowS].parent = huffNode[lowS-1].parent = (U16)nodeNb;
+    nodeNb++; lowS-=2;
+    for (n=nodeNb; n<=nodeRoot; n++) huffNode[n].count = (U32)(1U<<30);
+    huffNode0[0].count = (U32)(1U<<31);  /* fake entry, strong barrier */
+
+    /* create parents */
+    while (nodeNb <= nodeRoot) {
+        int const n1 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
+        int const n2 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
+        huffNode[nodeNb].count = huffNode[n1].count + huffNode[n2].count;
+        huffNode[n1].parent = huffNode[n2].parent = (U16)nodeNb;
+        nodeNb++;
+    }
+
+    /* distribute weights (unlimited tree height) */
+    huffNode[nodeRoot].nbBits = 0;
+    for (n=nodeRoot-1; n>=STARTNODE; n--)
+        huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
+    for (n=0; n<=nonNullRank; n++)
+        huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
+
+    DEBUGLOG(6, "Initial distribution of bits completed (%zu sorted symbols)", showHNodeBits(huffNode, maxSymbolValue+1));
+
+    return nonNullRank;
+}
+
+/**
+ * HUF_buildCTableFromTree():
+ * Build the CTable given the Huffman tree in huffNode.
+ *
+ * @param[out] CTable         The output Huffman CTable.
+ * @param      huffNode       The Huffman tree.
+ * @param      nonNullRank    The last and smallest node in the Huffman tree.
+ * @param      maxSymbolValue The maximum symbol value.
+ * @param      maxNbBits      The exact maximum number of bits used in the Huffman tree.
+ */
+static void HUF_buildCTableFromTree(HUF_CElt* CTable, nodeElt const* huffNode, int nonNullRank, U32 maxSymbolValue, U32 maxNbBits)
+{
+    HUF_CElt* const ct = CTable + 1;
+    /* fill result into ctable (val, nbBits) */
+    int n;
+    U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
+    U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
+    int const alphabetSize = (int)(maxSymbolValue + 1);
+    for (n=0; n<=nonNullRank; n++)
+        nbPerRank[huffNode[n].nbBits]++;
+    /* determine starting value per rank */
+    {   U16 min = 0;
+        for (n=(int)maxNbBits; n>0; n--) {
+            valPerRank[n] = min;      /* get starting value within each rank */
+            min += nbPerRank[n];
+            min >>= 1;
+    }   }
+    for (n=0; n<alphabetSize; n++)
+        HUF_setNbBits(ct + huffNode[n].byte, huffNode[n].nbBits);   /* push nbBits per symbol, symbol order */
+    for (n=0; n<alphabetSize; n++)
+        HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++);   /* assign value within rank, symbol order */
+
+    HUF_writeCTableHeader(CTable, maxNbBits, maxSymbolValue);
+}
+
+size_t
+HUF_buildCTable_wksp(HUF_CElt* CTable, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits,
+                     void* workSpace, size_t wkspSize)
+{
+    HUF_buildCTable_wksp_tables* const wksp_tables =
+        (HUF_buildCTable_wksp_tables*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(U32));
+    nodeElt* const huffNode0 = wksp_tables->huffNodeTbl;
+    nodeElt* const huffNode = huffNode0+1;
+    int nonNullRank;
+
+    HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE == sizeof(HUF_buildCTable_wksp_tables));
+
+    DEBUGLOG(5, "HUF_buildCTable_wksp (alphabet size = %u)", maxSymbolValue+1);
+
+    /* safety checks */
+    if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
+        return ERROR(workSpace_tooSmall);
+    if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX)
+        return ERROR(maxSymbolValue_tooLarge);
+    ZSTD_memset(huffNode0, 0, sizeof(huffNodeTable));
+
+    /* sort, decreasing order */
+    HUF_sort(huffNode, count, maxSymbolValue, wksp_tables->rankPosition);
+    DEBUGLOG(6, "sorted symbols completed (%zu symbols)", showHNodeSymbols(huffNode, maxSymbolValue+1));
+
+    /* build tree */
+    nonNullRank = HUF_buildTree(huffNode, maxSymbolValue);
+
+    /* determine and enforce maxTableLog */
+    maxNbBits = HUF_setMaxHeight(huffNode, (U32)nonNullRank, maxNbBits);
+    if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC);   /* check fit into table */
+
+    HUF_buildCTableFromTree(CTable, huffNode, nonNullRank, maxSymbolValue, maxNbBits);
+
+    return maxNbBits;
+}
+
+size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
+{
+    HUF_CElt const* ct = CTable + 1;
+    size_t nbBits = 0;
+    int s;
+    for (s = 0; s <= (int)maxSymbolValue; ++s) {
+        nbBits += HUF_getNbBits(ct[s]) * count[s];
+    }
+    return nbBits >> 3;
+}
+
+int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
+    HUF_CTableHeader header = HUF_readCTableHeader(CTable);
+    HUF_CElt const* ct = CTable + 1;
+    int bad = 0;
+    int s;
+
+    assert(header.tableLog <= HUF_TABLELOG_ABSOLUTEMAX);
+
+    if (header.maxSymbolValue < maxSymbolValue)
+        return 0;
+
+    for (s = 0; s <= (int)maxSymbolValue; ++s) {
+        bad |= (count[s] != 0) & (HUF_getNbBits(ct[s]) == 0);
+    }
+    return !bad;
+}
+
+size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
+
+/** HUF_CStream_t:
+ * Huffman uses its own BIT_CStream_t implementation.
+ * There are three major differences from BIT_CStream_t:
+ *   1. HUF_addBits() takes a HUF_CElt (size_t) which is
+ *      the pair (nbBits, value) in the format:
+ *      format:
+ *        - Bits [0, 4)            = nbBits
+ *        - Bits [4, 64 - nbBits)  = 0
+ *        - Bits [64 - nbBits, 64) = value
+ *   2. The bitContainer is built from the upper bits and
+ *      right shifted. E.g. to add a new value of N bits
+ *      you right shift the bitContainer by N, then or in
+ *      the new value into the N upper bits.
+ *   3. The bitstream has two bit containers. You can add
+ *      bits to the second container and merge them into
+ *      the first container.
+ */
+
+#define HUF_BITS_IN_CONTAINER (sizeof(size_t) * 8)
+
+typedef struct {
+    size_t bitContainer[2];
+    size_t bitPos[2];
+
+    BYTE* startPtr;
+    BYTE* ptr;
+    BYTE* endPtr;
+} HUF_CStream_t;
+
+/**! HUF_initCStream():
+ * Initializes the bitstream.
+ * @returns 0 or an error code.
+ */
+static size_t HUF_initCStream(HUF_CStream_t* bitC,
+                                  void* startPtr, size_t dstCapacity)
+{
+    ZSTD_memset(bitC, 0, sizeof(*bitC));
+    bitC->startPtr = (BYTE*)startPtr;
+    bitC->ptr = bitC->startPtr;
+    bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->bitContainer[0]);
+    if (dstCapacity <= sizeof(bitC->bitContainer[0])) return ERROR(dstSize_tooSmall);
+    return 0;
+}
+
+/*! HUF_addBits():
+ * Adds the symbol stored in HUF_CElt elt to the bitstream.
+ *
+ * @param elt   The element we're adding. This is a (nbBits, value) pair.
+ *              See the HUF_CStream_t docs for the format.
+ * @param idx   Insert into the bitstream at this idx.
+ * @param kFast This is a template parameter. If the bitstream is guaranteed
+ *              to have at least 4 unused bits after this call it may be 1,
+ *              otherwise it must be 0. HUF_addBits() is faster when fast is set.
+ */
+FORCE_INLINE_TEMPLATE void HUF_addBits(HUF_CStream_t* bitC, HUF_CElt elt, int idx, int kFast)
+{
+    assert(idx <= 1);
+    assert(HUF_getNbBits(elt) <= HUF_TABLELOG_ABSOLUTEMAX);
+    /* This is efficient on x86-64 with BMI2 because shrx
+     * only reads the low 6 bits of the register. The compiler
+     * knows this and elides the mask. When fast is set,
+     * every operation can use the same value loaded from elt.
+     */
+    bitC->bitContainer[idx] >>= HUF_getNbBits(elt);
+    bitC->bitContainer[idx] |= kFast ? HUF_getValueFast(elt) : HUF_getValue(elt);
+    /* We only read the low 8 bits of bitC->bitPos[idx] so it
+     * doesn't matter that the high bits have noise from the value.
+     */
+    bitC->bitPos[idx] += HUF_getNbBitsFast(elt);
+    assert((bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
+    /* The last 4-bits of elt are dirty if fast is set,
+     * so we must not be overwriting bits that have already been
+     * inserted into the bit container.
+     */
+#if DEBUGLEVEL >= 1
+    {
+        size_t const nbBits = HUF_getNbBits(elt);
+        size_t const dirtyBits = nbBits == 0 ? 0 : ZSTD_highbit32((U32)nbBits) + 1;
+        (void)dirtyBits;
+        /* Middle bits are 0. */
+        assert(((elt >> dirtyBits) << (dirtyBits + nbBits)) == 0);
+        /* We didn't overwrite any bits in the bit container. */
+        assert(!kFast || (bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
+        (void)dirtyBits;
+    }
+#endif
+}
+
+FORCE_INLINE_TEMPLATE void HUF_zeroIndex1(HUF_CStream_t* bitC)
+{
+    bitC->bitContainer[1] = 0;
+    bitC->bitPos[1] = 0;
+}
+
+/*! HUF_mergeIndex1() :
+ * Merges the bit container @ index 1 into the bit container @ index 0
+ * and zeros the bit container @ index 1.
+ */
+FORCE_INLINE_TEMPLATE void HUF_mergeIndex1(HUF_CStream_t* bitC)
+{
+    assert((bitC->bitPos[1] & 0xFF) < HUF_BITS_IN_CONTAINER);
+    bitC->bitContainer[0] >>= (bitC->bitPos[1] & 0xFF);
+    bitC->bitContainer[0] |= bitC->bitContainer[1];
+    bitC->bitPos[0] += bitC->bitPos[1];
+    assert((bitC->bitPos[0] & 0xFF) <= HUF_BITS_IN_CONTAINER);
+}
+
+/*! HUF_flushBits() :
+* Flushes the bits in the bit container @ index 0.
+*
+* @post bitPos will be < 8.
+* @param kFast If kFast is set then we must know a-priori that
+*              the bit container will not overflow.
+*/
+FORCE_INLINE_TEMPLATE void HUF_flushBits(HUF_CStream_t* bitC, int kFast)
+{
+    /* The upper bits of bitPos are noisy, so we must mask by 0xFF. */
+    size_t const nbBits = bitC->bitPos[0] & 0xFF;
+    size_t const nbBytes = nbBits >> 3;
+    /* The top nbBits bits of bitContainer are the ones we need. */
+    size_t const bitContainer = bitC->bitContainer[0] >> (HUF_BITS_IN_CONTAINER - nbBits);
+    /* Mask bitPos to account for the bytes we consumed. */
+    bitC->bitPos[0] &= 7;
+    assert(nbBits > 0);
+    assert(nbBits <= sizeof(bitC->bitContainer[0]) * 8);
+    assert(bitC->ptr <= bitC->endPtr);
+    MEM_writeLEST(bitC->ptr, bitContainer);
+    bitC->ptr += nbBytes;
+    assert(!kFast || bitC->ptr <= bitC->endPtr);
+    if (!kFast && bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
+    /* bitContainer doesn't need to be modified because the leftover
+     * bits are already the top bitPos bits. And we don't care about
+     * noise in the lower values.
+     */
+}
+
+/*! HUF_endMark()
+ * @returns The Huffman stream end mark: A 1-bit value = 1.
+ */
+static HUF_CElt HUF_endMark(void)
+{
+    HUF_CElt endMark;
+    HUF_setNbBits(&endMark, 1);
+    HUF_setValue(&endMark, 1);
+    return endMark;
+}
+
+/*! HUF_closeCStream() :
+ *  @return Size of CStream, in bytes,
+ *          or 0 if it could not fit into dstBuffer */
+static size_t HUF_closeCStream(HUF_CStream_t* bitC)
+{
+    HUF_addBits(bitC, HUF_endMark(), /* idx */ 0, /* kFast */ 0);
+    HUF_flushBits(bitC, /* kFast */ 0);
+    {
+        size_t const nbBits = bitC->bitPos[0] & 0xFF;
+        if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
+        return (size_t)(bitC->ptr - bitC->startPtr) + (nbBits > 0);
+    }
+}
+
+FORCE_INLINE_TEMPLATE void
+HUF_encodeSymbol(HUF_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable, int idx, int fast)
+{
+    HUF_addBits(bitCPtr, CTable[symbol], idx, fast);
+}
+
+FORCE_INLINE_TEMPLATE void
+HUF_compress1X_usingCTable_internal_body_loop(HUF_CStream_t* bitC,
+                                   const BYTE* ip, size_t srcSize,
+                                   const HUF_CElt* ct,
+                                   int kUnroll, int kFastFlush, int kLastFast)
+{
+    /* Join to kUnroll */
+    int n = (int)srcSize;
+    int rem = n % kUnroll;
+    if (rem > 0) {
+        for (; rem > 0; --rem) {
+            HUF_encodeSymbol(bitC, ip[--n], ct, 0, /* fast */ 0);
+        }
+        HUF_flushBits(bitC, kFastFlush);
+    }
+    assert(n % kUnroll == 0);
+
+    /* Join to 2 * kUnroll */
+    if (n % (2 * kUnroll)) {
+        int u;
+        for (u = 1; u < kUnroll; ++u) {
+            HUF_encodeSymbol(bitC, ip[n - u], ct, 0, 1);
+        }
+        HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, 0, kLastFast);
+        HUF_flushBits(bitC, kFastFlush);
+        n -= kUnroll;
+    }
+    assert(n % (2 * kUnroll) == 0);
+
+    for (; n>0; n-= 2 * kUnroll) {
+        /* Encode kUnroll symbols into the bitstream @ index 0. */
+        int u;
+        for (u = 1; u < kUnroll; ++u) {
+            HUF_encodeSymbol(bitC, ip[n - u], ct, /* idx */ 0, /* fast */ 1);
+        }
+        HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, /* idx */ 0, /* fast */ kLastFast);
+        HUF_flushBits(bitC, kFastFlush);
+        /* Encode kUnroll symbols into the bitstream @ index 1.
+         * This allows us to start filling the bit container
+         * without any data dependencies.
+         */
+        HUF_zeroIndex1(bitC);
+        for (u = 1; u < kUnroll; ++u) {
+            HUF_encodeSymbol(bitC, ip[n - kUnroll - u], ct, /* idx */ 1, /* fast */ 1);
+        }
+        HUF_encodeSymbol(bitC, ip[n - kUnroll - kUnroll], ct, /* idx */ 1, /* fast */ kLastFast);
+        /* Merge bitstream @ index 1 into the bitstream @ index 0 */
+        HUF_mergeIndex1(bitC);
+        HUF_flushBits(bitC, kFastFlush);
+    }
+    assert(n == 0);
+
+}
+
+/**
+ * Returns a tight upper bound on the output space needed by Huffman
+ * with 8 bytes buffer to handle over-writes. If the output is at least
+ * this large we don't need to do bounds checks during Huffman encoding.
+ */
+static size_t HUF_tightCompressBound(size_t srcSize, size_t tableLog)
+{
+    return ((srcSize * tableLog) >> 3) + 8;
+}
+
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_compress1X_usingCTable_internal_body(void* dst, size_t dstSize,
+                                   const void* src, size_t srcSize,
+                                   const HUF_CElt* CTable)
+{
+    U32 const tableLog = HUF_readCTableHeader(CTable).tableLog;
+    HUF_CElt const* ct = CTable + 1;
+    const BYTE* ip = (const BYTE*) src;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstSize;
+    HUF_CStream_t bitC;
+
+    /* init */
+    if (dstSize < 8) return 0;   /* not enough space to compress */
+    { BYTE* op = ostart;
+      size_t const initErr = HUF_initCStream(&bitC, op, (size_t)(oend-op));
+      if (HUF_isError(initErr)) return 0; }
+
+    if (dstSize < HUF_tightCompressBound(srcSize, (size_t)tableLog) || tableLog > 11)
+        HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ MEM_32bits() ? 2 : 4, /* kFast */ 0, /* kLastFast */ 0);
+    else {
+        if (MEM_32bits()) {
+            switch (tableLog) {
+            case 11:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 10: ZSTD_FALLTHROUGH;
+            case 9: ZSTD_FALLTHROUGH;
+            case 8:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            case 7: ZSTD_FALLTHROUGH;
+            default:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 3, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            }
+        } else {
+            switch (tableLog) {
+            case 11:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 10:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            case 9:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 6, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 8:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 7, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 7:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 8, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 6: ZSTD_FALLTHROUGH;
+            default:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 9, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            }
+        }
+    }
+    assert(bitC.ptr <= bitC.endPtr);
+
+    return HUF_closeCStream(&bitC);
+}
+
+#if DYNAMIC_BMI2
+
+static BMI2_TARGET_ATTRIBUTE size_t
+HUF_compress1X_usingCTable_internal_bmi2(void* dst, size_t dstSize,
+                                   const void* src, size_t srcSize,
+                                   const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+static size_t
+HUF_compress1X_usingCTable_internal_default(void* dst, size_t dstSize,
+                                      const void* src, size_t srcSize,
+                                      const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+static size_t
+HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, const int flags)
+{
+    if (flags & HUF_flags_bmi2) {
+        return HUF_compress1X_usingCTable_internal_bmi2(dst, dstSize, src, srcSize, CTable);
+    }
+    return HUF_compress1X_usingCTable_internal_default(dst, dstSize, src, srcSize, CTable);
+}
+
+#else
+
+static size_t
+HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, const int flags)
+{
+    (void)flags;
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+#endif
+
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
+{
+    return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
+}
+
+static size_t
+HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, int flags)
+{
+    size_t const segmentSize = (srcSize+3)/4;   /* first 3 segments */
+    const BYTE* ip = (const BYTE*) src;
+    const BYTE* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart;
+
+    if (dstSize < 6 + 1 + 1 + 1 + 8) return 0;   /* minimum space to compress successfully */
+    if (srcSize < 12) return 0;   /* no saving possible : too small input */
+    op += 6;   /* jumpTable */
+
+    assert(op <= oend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
+        MEM_writeLE16(ostart, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    assert(op <= oend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
+        MEM_writeLE16(ostart+2, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    assert(op <= oend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
+        MEM_writeLE16(ostart+4, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    assert(op <= oend);
+    assert(ip <= iend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, (size_t)(iend-ip), CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
+        op += cSize;
+    }
+
+    return (size_t)(op-ostart);
+}
+
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
+{
+    return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
+}
+
+typedef enum { HUF_singleStream, HUF_fourStreams } HUF_nbStreams_e;
+
+static size_t HUF_compressCTable_internal(
+                BYTE* const ostart, BYTE* op, BYTE* const oend,
+                const void* src, size_t srcSize,
+                HUF_nbStreams_e nbStreams, const HUF_CElt* CTable, const int flags)
+{
+    size_t const cSize = (nbStreams==HUF_singleStream) ?
+                         HUF_compress1X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags) :
+                         HUF_compress4X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags);
+    if (HUF_isError(cSize)) { return cSize; }
+    if (cSize==0) { return 0; }   /* uncompressible */
+    op += cSize;
+    /* check compressibility */
+    assert(op >= ostart);
+    if ((size_t)(op-ostart) >= srcSize-1) { return 0; }
+    return (size_t)(op-ostart);
+}
+
+typedef struct {
+    unsigned count[HUF_SYMBOLVALUE_MAX + 1];
+    HUF_CElt CTable[HUF_CTABLE_SIZE_ST(HUF_SYMBOLVALUE_MAX)];
+    union {
+        HUF_buildCTable_wksp_tables buildCTable_wksp;
+        HUF_WriteCTableWksp writeCTable_wksp;
+        U32 hist_wksp[HIST_WKSP_SIZE_U32];
+    } wksps;
+} HUF_compress_tables_t;
+
+#define SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE 4096
+#define SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO 10  /* Must be >= 2 */
+
+unsigned HUF_cardinality(const unsigned* count, unsigned maxSymbolValue)
+{
+    unsigned cardinality = 0;
+    unsigned i;
+
+    for (i = 0; i < maxSymbolValue + 1; i++) {
+        if (count[i] != 0) cardinality += 1;
+    }
+
+    return cardinality;
+}
+
+unsigned HUF_minTableLog(unsigned symbolCardinality)
+{
+    U32 minBitsSymbols = ZSTD_highbit32(symbolCardinality) + 1;
+    return minBitsSymbols;
+}
+
+unsigned HUF_optimalTableLog(
+            unsigned maxTableLog,
+            size_t srcSize,
+            unsigned maxSymbolValue,
+            void* workSpace, size_t wkspSize,
+            HUF_CElt* table,
+      const unsigned* count,
+            int flags)
+{
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
+    assert(wkspSize >= sizeof(HUF_buildCTable_wksp_tables));
+
+    if (!(flags & HUF_flags_optimalDepth)) {
+        /* cheap evaluation, based on FSE */
+        return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
+    }
+
+    {   BYTE* dst = (BYTE*)workSpace + sizeof(HUF_WriteCTableWksp);
+        size_t dstSize = wkspSize - sizeof(HUF_WriteCTableWksp);
+        size_t hSize, newSize;
+        const unsigned symbolCardinality = HUF_cardinality(count, maxSymbolValue);
+        const unsigned minTableLog = HUF_minTableLog(symbolCardinality);
+        size_t optSize = ((size_t) ~0) - 1;
+        unsigned optLog = maxTableLog, optLogGuess;
+
+        DEBUGLOG(6, "HUF_optimalTableLog: probing huf depth (srcSize=%zu)", srcSize);
+
+        /* Search until size increases */
+        for (optLogGuess = minTableLog; optLogGuess <= maxTableLog; optLogGuess++) {
+            DEBUGLOG(7, "checking for huffLog=%u", optLogGuess);
+
+            {   size_t maxBits = HUF_buildCTable_wksp(table, count, maxSymbolValue, optLogGuess, workSpace, wkspSize);
+                if (ERR_isError(maxBits)) continue;
+
+                if (maxBits < optLogGuess && optLogGuess > minTableLog) break;
+
+                hSize = HUF_writeCTable_wksp(dst, dstSize, table, maxSymbolValue, (U32)maxBits, workSpace, wkspSize);
+            }
+
+            if (ERR_isError(hSize)) continue;
+
+            newSize = HUF_estimateCompressedSize(table, count, maxSymbolValue) + hSize;
+
+            if (newSize > optSize + 1) {
+                break;
+            }
+
+            if (newSize < optSize) {
+                optSize = newSize;
+                optLog = optLogGuess;
+            }
+        }
+        assert(optLog <= HUF_TABLELOG_MAX);
+        return optLog;
+    }
+}
+
+/* HUF_compress_internal() :
+ * `workSpace_align4` must be aligned on 4-bytes boundaries,
+ * and occupies the same space as a table of HUF_WORKSPACE_SIZE_U64 unsigned */
+static size_t
+HUF_compress_internal (void* dst, size_t dstSize,
+                 const void* src, size_t srcSize,
+                       unsigned maxSymbolValue, unsigned huffLog,
+                       HUF_nbStreams_e nbStreams,
+                       void* workSpace, size_t wkspSize,
+                       HUF_CElt* oldHufTable, HUF_repeat* repeat, int flags)
+{
+    HUF_compress_tables_t* const table = (HUF_compress_tables_t*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(size_t));
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart;
+
+    DEBUGLOG(5, "HUF_compress_internal (srcSize=%zu)", srcSize);
+    HUF_STATIC_ASSERT(sizeof(*table) + HUF_WORKSPACE_MAX_ALIGNMENT <= HUF_WORKSPACE_SIZE);
+
+    /* checks & inits */
+    if (wkspSize < sizeof(*table)) return ERROR(workSpace_tooSmall);
+    if (!srcSize) return 0;  /* Uncompressed */
+    if (!dstSize) return 0;  /* cannot fit anything within dst budget */
+    if (srcSize > HUF_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);   /* current block size limit */
+    if (huffLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
+    if (!maxSymbolValue) maxSymbolValue = HUF_SYMBOLVALUE_MAX;
+    if (!huffLog) huffLog = HUF_TABLELOG_DEFAULT;
+
+    /* Heuristic : If old table is valid, use it for small inputs */
+    if ((flags & HUF_flags_preferRepeat) && repeat && *repeat == HUF_repeat_valid) {
+        return HUF_compressCTable_internal(ostart, op, oend,
+                                           src, srcSize,
+                                           nbStreams, oldHufTable, flags);
+    }
+
+    /* If uncompressible data is suspected, do a smaller sampling first */
+    DEBUG_STATIC_ASSERT(SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO >= 2);
+    if ((flags & HUF_flags_suspectUncompressible) && srcSize >= (SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE * SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO)) {
+        size_t largestTotal = 0;
+        DEBUGLOG(5, "input suspected incompressible : sampling to check");
+        {   unsigned maxSymbolValueBegin = maxSymbolValue;
+            CHECK_V_F(largestBegin, HIST_count_simple (table->count, &maxSymbolValueBegin, (const BYTE*)src, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
+            largestTotal += largestBegin;
+        }
+        {   unsigned maxSymbolValueEnd = maxSymbolValue;
+            CHECK_V_F(largestEnd, HIST_count_simple (table->count, &maxSymbolValueEnd, (const BYTE*)src + srcSize - SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
+            largestTotal += largestEnd;
+        }
+        if (largestTotal <= ((2 * SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) >> 7)+4) return 0;   /* heuristic : probably not compressible enough */
+    }
+
+    /* Scan input and build symbol stats */
+    {   CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, table->wksps.hist_wksp, sizeof(table->wksps.hist_wksp)) );
+        if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; }   /* single symbol, rle */
+        if (largest <= (srcSize >> 7)+4) return 0;   /* heuristic : probably not compressible enough */
+    }
+    DEBUGLOG(6, "histogram detail completed (%zu symbols)", showU32(table->count, maxSymbolValue+1));
+
+    /* Check validity of previous table */
+    if ( repeat
+      && *repeat == HUF_repeat_check
+      && !HUF_validateCTable(oldHufTable, table->count, maxSymbolValue)) {
+        *repeat = HUF_repeat_none;
+    }
+    /* Heuristic : use existing table for small inputs */
+    if ((flags & HUF_flags_preferRepeat) && repeat && *repeat != HUF_repeat_none) {
+        return HUF_compressCTable_internal(ostart, op, oend,
+                                           src, srcSize,
+                                           nbStreams, oldHufTable, flags);
+    }
+
+    /* Build Huffman Tree */
+    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue, &table->wksps, sizeof(table->wksps), table->CTable, table->count, flags);
+    {   size_t const maxBits = HUF_buildCTable_wksp(table->CTable, table->count,
+                                            maxSymbolValue, huffLog,
+                                            &table->wksps.buildCTable_wksp, sizeof(table->wksps.buildCTable_wksp));
+        CHECK_F(maxBits);
+        huffLog = (U32)maxBits;
+        DEBUGLOG(6, "bit distribution completed (%zu symbols)", showCTableBits(table->CTable + 1, maxSymbolValue+1));
+    }
+
+    /* Write table description header */
+    {   CHECK_V_F(hSize, HUF_writeCTable_wksp(op, dstSize, table->CTable, maxSymbolValue, huffLog,
+                                              &table->wksps.writeCTable_wksp, sizeof(table->wksps.writeCTable_wksp)) );
+        /* Check if using previous huffman table is beneficial */
+        if (repeat && *repeat != HUF_repeat_none) {
+            size_t const oldSize = HUF_estimateCompressedSize(oldHufTable, table->count, maxSymbolValue);
+            size_t const newSize = HUF_estimateCompressedSize(table->CTable, table->count, maxSymbolValue);
+            if (oldSize <= hSize + newSize || hSize + 12 >= srcSize) {
+                return HUF_compressCTable_internal(ostart, op, oend,
+                                                   src, srcSize,
+                                                   nbStreams, oldHufTable, flags);
+        }   }
+
+        /* Use the new huffman table */
+        if (hSize + 12ul >= srcSize) { return 0; }
+        op += hSize;
+        if (repeat) { *repeat = HUF_repeat_none; }
+        if (oldHufTable)
+            ZSTD_memcpy(oldHufTable, table->CTable, sizeof(table->CTable));  /* Save new table */
+    }
+    return HUF_compressCTable_internal(ostart, op, oend,
+                                       src, srcSize,
+                                       nbStreams, table->CTable, flags);
+}
+
+size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize,
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
+{
+    DEBUGLOG(5, "HUF_compress1X_repeat (srcSize = %zu)", srcSize);
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, HUF_singleStream,
+                                 workSpace, wkspSize, hufTable,
+                                 repeat, flags);
+}
+
+/* HUF_compress4X_repeat():
+ * compress input using 4 streams.
+ * consider skipping quickly
+ * reuse an existing huffman compression table */
+size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize,
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
+{
+    DEBUGLOG(5, "HUF_compress4X_repeat (srcSize = %zu)", srcSize);
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, HUF_fourStreams,
+                                 workSpace, wkspSize,
+                                 hufTable, repeat, flags);
+}
diff --git a/lib/compress/zstd_compress.c b/lib/compress/zstd_compress.c
new file mode 100644
index 0000000..c8f6b28
--- /dev/null
+++ b/lib/compress/zstd_compress.c
@@ -0,0 +1,7851 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "../common/allocations.h"  /* ZSTD_customMalloc, ZSTD_customCalloc, ZSTD_customFree */
+#include "../common/zstd_deps.h"  /* INT_MAX, ZSTD_memset, ZSTD_memcpy */
+#include "../common/mem.h"
+#include "../common/error_private.h"
+#include "hist.h"           /* HIST_countFast_wksp */
+#define FSE_STATIC_LINKING_ONLY   /* FSE_encodeSymbol */
+#include "../common/fse.h"
+#include "../common/huf.h"
+#include "zstd_compress_internal.h"
+#include "zstd_compress_sequences.h"
+#include "zstd_compress_literals.h"
+#include "zstd_fast.h"
+#include "zstd_double_fast.h"
+#include "zstd_lazy.h"
+#include "zstd_opt.h"
+#include "zstd_ldm.h"
+#include "zstd_compress_superblock.h"
+#include  "../common/bits.h"      /* ZSTD_highbit32, ZSTD_rotateRight_U64 */
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * COMPRESS_HEAPMODE :
+ * Select how default decompression function ZSTD_compress() allocates its context,
+ * on stack (0, default), or into heap (1).
+ * Note that functions with explicit context such as ZSTD_compressCCtx() are unaffected.
+ */
+#ifndef ZSTD_COMPRESS_HEAPMODE
+#  define ZSTD_COMPRESS_HEAPMODE 0
+#endif
+
+/*!
+ * ZSTD_HASHLOG3_MAX :
+ * Maximum size of the hash table dedicated to find 3-bytes matches,
+ * in log format, aka 17 => 1 << 17 == 128Ki positions.
+ * This structure is only used in zstd_opt.
+ * Since allocation is centralized for all strategies, it has to be known here.
+ * The actual (selected) size of the hash table is then stored in ZSTD_MatchState_t.hashLog3,
+ * so that zstd_opt.c doesn't need to know about this constant.
+ */
+#ifndef ZSTD_HASHLOG3_MAX
+#  define ZSTD_HASHLOG3_MAX 17
+#endif
+
+/*-*************************************
+*  Helper functions
+***************************************/
+/* ZSTD_compressBound()
+ * Note that the result from this function is only valid for
+ * the one-pass compression functions.
+ * When employing the streaming mode,
+ * if flushes are frequently altering the size of blocks,
+ * the overhead from block headers can make the compressed data larger
+ * than the return value of ZSTD_compressBound().
+ */
+size_t ZSTD_compressBound(size_t srcSize) {
+    size_t const r = ZSTD_COMPRESSBOUND(srcSize);
+    if (r==0) return ERROR(srcSize_wrong);
+    return r;
+}
+
+
+/*-*************************************
+*  Context memory management
+***************************************/
+struct ZSTD_CDict_s {
+    const void* dictContent;
+    size_t dictContentSize;
+    ZSTD_dictContentType_e dictContentType; /* The dictContentType the CDict was created with */
+    U32* entropyWorkspace; /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
+    ZSTD_cwksp workspace;
+    ZSTD_MatchState_t matchState;
+    ZSTD_compressedBlockState_t cBlockState;
+    ZSTD_customMem customMem;
+    U32 dictID;
+    int compressionLevel; /* 0 indicates that advanced API was used to select CDict params */
+    ZSTD_ParamSwitch_e useRowMatchFinder; /* Indicates whether the CDict was created with params that would use
+                                           * row-based matchfinder. Unless the cdict is reloaded, we will use
+                                           * the same greedy/lazy matchfinder at compression time.
+                                           */
+};  /* typedef'd to ZSTD_CDict within "zstd.h" */
+
+ZSTD_CCtx* ZSTD_createCCtx(void)
+{
+    return ZSTD_createCCtx_advanced(ZSTD_defaultCMem);
+}
+
+static void ZSTD_initCCtx(ZSTD_CCtx* cctx, ZSTD_customMem memManager)
+{
+    assert(cctx != NULL);
+    ZSTD_memset(cctx, 0, sizeof(*cctx));
+    cctx->customMem = memManager;
+    cctx->bmi2 = ZSTD_cpuSupportsBmi2();
+    {   size_t const err = ZSTD_CCtx_reset(cctx, ZSTD_reset_parameters);
+        assert(!ZSTD_isError(err));
+        (void)err;
+    }
+}
+
+ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem)
+{
+    ZSTD_STATIC_ASSERT(zcss_init==0);
+    ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN==(0ULL - 1));
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
+    {   ZSTD_CCtx* const cctx = (ZSTD_CCtx*)ZSTD_customMalloc(sizeof(ZSTD_CCtx), customMem);
+        if (!cctx) return NULL;
+        ZSTD_initCCtx(cctx, customMem);
+        return cctx;
+    }
+}
+
+ZSTD_CCtx* ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize)
+{
+    ZSTD_cwksp ws;
+    ZSTD_CCtx* cctx;
+    if (workspaceSize <= sizeof(ZSTD_CCtx)) return NULL;  /* minimum size */
+    if ((size_t)workspace & 7) return NULL;  /* must be 8-aligned */
+    ZSTD_cwksp_init(&ws, workspace, workspaceSize, ZSTD_cwksp_static_alloc);
+
+    cctx = (ZSTD_CCtx*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CCtx));
+    if (cctx == NULL) return NULL;
+
+    ZSTD_memset(cctx, 0, sizeof(ZSTD_CCtx));
+    ZSTD_cwksp_move(&cctx->workspace, &ws);
+    cctx->staticSize = workspaceSize;
+
+    /* statically sized space. tmpWorkspace never moves (but prev/next block swap places) */
+    if (!ZSTD_cwksp_check_available(&cctx->workspace, TMP_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t))) return NULL;
+    cctx->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
+    cctx->blockState.nextCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
+    cctx->tmpWorkspace = ZSTD_cwksp_reserve_object(&cctx->workspace, TMP_WORKSPACE_SIZE);
+    cctx->tmpWkspSize = TMP_WORKSPACE_SIZE;
+    cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+    return cctx;
+}
+
+/**
+ * Clears and frees all of the dictionaries in the CCtx.
+ */
+static void ZSTD_clearAllDicts(ZSTD_CCtx* cctx)
+{
+    ZSTD_customFree(cctx->localDict.dictBuffer, cctx->customMem);
+    ZSTD_freeCDict(cctx->localDict.cdict);
+    ZSTD_memset(&cctx->localDict, 0, sizeof(cctx->localDict));
+    ZSTD_memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));
+    cctx->cdict = NULL;
+}
+
+static size_t ZSTD_sizeof_localDict(ZSTD_localDict dict)
+{
+    size_t const bufferSize = dict.dictBuffer != NULL ? dict.dictSize : 0;
+    size_t const cdictSize = ZSTD_sizeof_CDict(dict.cdict);
+    return bufferSize + cdictSize;
+}
+
+static void ZSTD_freeCCtxContent(ZSTD_CCtx* cctx)
+{
+    assert(cctx != NULL);
+    assert(cctx->staticSize == 0);
+    ZSTD_clearAllDicts(cctx);
+#ifdef ZSTD_MULTITHREAD
+    ZSTDMT_freeCCtx(cctx->mtctx); cctx->mtctx = NULL;
+#endif
+    ZSTD_cwksp_free(&cctx->workspace, cctx->customMem);
+}
+
+size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
+{
+    DEBUGLOG(3, "ZSTD_freeCCtx (address: %p)", (void*)cctx);
+    if (cctx==NULL) return 0;   /* support free on NULL */
+    RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
+                    "not compatible with static CCtx");
+    {   int cctxInWorkspace = ZSTD_cwksp_owns_buffer(&cctx->workspace, cctx);
+        ZSTD_freeCCtxContent(cctx);
+        if (!cctxInWorkspace) ZSTD_customFree(cctx, cctx->customMem);
+    }
+    return 0;
+}
+
+
+static size_t ZSTD_sizeof_mtctx(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    return ZSTDMT_sizeof_CCtx(cctx->mtctx);
+#else
+    (void)cctx;
+    return 0;
+#endif
+}
+
+
+size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return 0;   /* support sizeof on NULL */
+    /* cctx may be in the workspace */
+    return (cctx->workspace.workspace == cctx ? 0 : sizeof(*cctx))
+           + ZSTD_cwksp_sizeof(&cctx->workspace)
+           + ZSTD_sizeof_localDict(cctx->localDict)
+           + ZSTD_sizeof_mtctx(cctx);
+}
+
+size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs)
+{
+    return ZSTD_sizeof_CCtx(zcs);  /* same object */
+}
+
+/* private API call, for dictBuilder only */
+const SeqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx) { return &(ctx->seqStore); }
+
+/* Returns true if the strategy supports using a row based matchfinder */
+static int ZSTD_rowMatchFinderSupported(const ZSTD_strategy strategy) {
+    return (strategy >= ZSTD_greedy && strategy <= ZSTD_lazy2);
+}
+
+/* Returns true if the strategy and useRowMatchFinder mode indicate that we will use the row based matchfinder
+ * for this compression.
+ */
+static int ZSTD_rowMatchFinderUsed(const ZSTD_strategy strategy, const ZSTD_ParamSwitch_e mode) {
+    assert(mode != ZSTD_ps_auto);
+    return ZSTD_rowMatchFinderSupported(strategy) && (mode == ZSTD_ps_enable);
+}
+
+/* Returns row matchfinder usage given an initial mode and cParams */
+static ZSTD_ParamSwitch_e ZSTD_resolveRowMatchFinderMode(ZSTD_ParamSwitch_e mode,
+                                                         const ZSTD_compressionParameters* const cParams) {
+#ifdef ZSTD_LINUX_KERNEL
+    /* The Linux Kernel does not use SIMD, and 128KB is a very common size, e.g. in BtrFS.
+     * The row match finder is slower for this size without SIMD, so disable it.
+     */
+    const unsigned kWindowLogLowerBound = 17;
+#else
+    const unsigned kWindowLogLowerBound = 14;
+#endif
+    if (mode != ZSTD_ps_auto) return mode; /* if requested enabled, but no SIMD, we still will use row matchfinder */
+    mode = ZSTD_ps_disable;
+    if (!ZSTD_rowMatchFinderSupported(cParams->strategy)) return mode;
+    if (cParams->windowLog > kWindowLogLowerBound) mode = ZSTD_ps_enable;
+    return mode;
+}
+
+/* Returns block splitter usage (generally speaking, when using slower/stronger compression modes) */
+static ZSTD_ParamSwitch_e ZSTD_resolveBlockSplitterMode(ZSTD_ParamSwitch_e mode,
+                                                        const ZSTD_compressionParameters* const cParams) {
+    if (mode != ZSTD_ps_auto) return mode;
+    return (cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 17) ? ZSTD_ps_enable : ZSTD_ps_disable;
+}
+
+/* Returns 1 if the arguments indicate that we should allocate a chainTable, 0 otherwise */
+static int ZSTD_allocateChainTable(const ZSTD_strategy strategy,
+                                   const ZSTD_ParamSwitch_e useRowMatchFinder,
+                                   const U32 forDDSDict) {
+    assert(useRowMatchFinder != ZSTD_ps_auto);
+    /* We always should allocate a chaintable if we are allocating a matchstate for a DDS dictionary matchstate.
+     * We do not allocate a chaintable if we are using ZSTD_fast, or are using the row-based matchfinder.
+     */
+    return forDDSDict || ((strategy != ZSTD_fast) && !ZSTD_rowMatchFinderUsed(strategy, useRowMatchFinder));
+}
+
+/* Returns ZSTD_ps_enable if compression parameters are such that we should
+ * enable long distance matching (wlog >= 27, strategy >= btopt).
+ * Returns ZSTD_ps_disable otherwise.
+ */
+static ZSTD_ParamSwitch_e ZSTD_resolveEnableLdm(ZSTD_ParamSwitch_e mode,
+                                 const ZSTD_compressionParameters* const cParams) {
+    if (mode != ZSTD_ps_auto) return mode;
+    return (cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 27) ? ZSTD_ps_enable : ZSTD_ps_disable;
+}
+
+static int ZSTD_resolveExternalSequenceValidation(int mode) {
+    return mode;
+}
+
+/* Resolves maxBlockSize to the default if no value is present. */
+static size_t ZSTD_resolveMaxBlockSize(size_t maxBlockSize) {
+    if (maxBlockSize == 0) {
+        return ZSTD_BLOCKSIZE_MAX;
+    } else {
+        return maxBlockSize;
+    }
+}
+
+static ZSTD_ParamSwitch_e ZSTD_resolveExternalRepcodeSearch(ZSTD_ParamSwitch_e value, int cLevel) {
+    if (value != ZSTD_ps_auto) return value;
+    if (cLevel < 10) {
+        return ZSTD_ps_disable;
+    } else {
+        return ZSTD_ps_enable;
+    }
+}
+
+/* Returns 1 if compression parameters are such that CDict hashtable and chaintable indices are tagged.
+ * If so, the tags need to be removed in ZSTD_resetCCtx_byCopyingCDict. */
+static int ZSTD_CDictIndicesAreTagged(const ZSTD_compressionParameters* const cParams) {
+    return cParams->strategy == ZSTD_fast || cParams->strategy == ZSTD_dfast;
+}
+
+static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
+        ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params cctxParams;
+    /* should not matter, as all cParams are presumed properly defined */
+    ZSTD_CCtxParams_init(&cctxParams, ZSTD_CLEVEL_DEFAULT);
+    cctxParams.cParams = cParams;
+
+    /* Adjust advanced params according to cParams */
+    cctxParams.ldmParams.enableLdm = ZSTD_resolveEnableLdm(cctxParams.ldmParams.enableLdm, &cParams);
+    if (cctxParams.ldmParams.enableLdm == ZSTD_ps_enable) {
+        ZSTD_ldm_adjustParameters(&cctxParams.ldmParams, &cParams);
+        assert(cctxParams.ldmParams.hashLog >= cctxParams.ldmParams.bucketSizeLog);
+        assert(cctxParams.ldmParams.hashRateLog < 32);
+    }
+    cctxParams.postBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams.postBlockSplitter, &cParams);
+    cctxParams.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams.useRowMatchFinder, &cParams);
+    cctxParams.validateSequences = ZSTD_resolveExternalSequenceValidation(cctxParams.validateSequences);
+    cctxParams.maxBlockSize = ZSTD_resolveMaxBlockSize(cctxParams.maxBlockSize);
+    cctxParams.searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(cctxParams.searchForExternalRepcodes,
+                                                                             cctxParams.compressionLevel);
+    assert(!ZSTD_checkCParams(cParams));
+    return cctxParams;
+}
+
+static ZSTD_CCtx_params* ZSTD_createCCtxParams_advanced(
+        ZSTD_customMem customMem)
+{
+    ZSTD_CCtx_params* params;
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
+    params = (ZSTD_CCtx_params*)ZSTD_customCalloc(
+            sizeof(ZSTD_CCtx_params), customMem);
+    if (!params) { return NULL; }
+    ZSTD_CCtxParams_init(params, ZSTD_CLEVEL_DEFAULT);
+    params->customMem = customMem;
+    return params;
+}
+
+ZSTD_CCtx_params* ZSTD_createCCtxParams(void)
+{
+    return ZSTD_createCCtxParams_advanced(ZSTD_defaultCMem);
+}
+
+size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params)
+{
+    if (params == NULL) { return 0; }
+    ZSTD_customFree(params, params->customMem);
+    return 0;
+}
+
+size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params)
+{
+    return ZSTD_CCtxParams_init(params, ZSTD_CLEVEL_DEFAULT);
+}
+
+size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel) {
+    RETURN_ERROR_IF(!cctxParams, GENERIC, "NULL pointer!");
+    ZSTD_memset(cctxParams, 0, sizeof(*cctxParams));
+    cctxParams->compressionLevel = compressionLevel;
+    cctxParams->fParams.contentSizeFlag = 1;
+    return 0;
+}
+
+#define ZSTD_NO_CLEVEL 0
+
+/**
+ * Initializes `cctxParams` from `params` and `compressionLevel`.
+ * @param compressionLevel If params are derived from a compression level then that compression level, otherwise ZSTD_NO_CLEVEL.
+ */
+static void
+ZSTD_CCtxParams_init_internal(ZSTD_CCtx_params* cctxParams,
+                        const ZSTD_parameters* params,
+                              int compressionLevel)
+{
+    assert(!ZSTD_checkCParams(params->cParams));
+    ZSTD_memset(cctxParams, 0, sizeof(*cctxParams));
+    cctxParams->cParams = params->cParams;
+    cctxParams->fParams = params->fParams;
+    /* Should not matter, as all cParams are presumed properly defined.
+     * But, set it for tracing anyway.
+     */
+    cctxParams->compressionLevel = compressionLevel;
+    cctxParams->useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams->useRowMatchFinder, &params->cParams);
+    cctxParams->postBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams->postBlockSplitter, &params->cParams);
+    cctxParams->ldmParams.enableLdm = ZSTD_resolveEnableLdm(cctxParams->ldmParams.enableLdm, &params->cParams);
+    cctxParams->validateSequences = ZSTD_resolveExternalSequenceValidation(cctxParams->validateSequences);
+    cctxParams->maxBlockSize = ZSTD_resolveMaxBlockSize(cctxParams->maxBlockSize);
+    cctxParams->searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(cctxParams->searchForExternalRepcodes, compressionLevel);
+    DEBUGLOG(4, "ZSTD_CCtxParams_init_internal: useRowMatchFinder=%d, useBlockSplitter=%d ldm=%d",
+                cctxParams->useRowMatchFinder, cctxParams->postBlockSplitter, cctxParams->ldmParams.enableLdm);
+}
+
+size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
+{
+    RETURN_ERROR_IF(!cctxParams, GENERIC, "NULL pointer!");
+    FORWARD_IF_ERROR( ZSTD_checkCParams(params.cParams) , "");
+    ZSTD_CCtxParams_init_internal(cctxParams, &params, ZSTD_NO_CLEVEL);
+    return 0;
+}
+
+/**
+ * Sets cctxParams' cParams and fParams from params, but otherwise leaves them alone.
+ * @param params Validated zstd parameters.
+ */
+static void ZSTD_CCtxParams_setZstdParams(
+        ZSTD_CCtx_params* cctxParams, const ZSTD_parameters* params)
+{
+    assert(!ZSTD_checkCParams(params->cParams));
+    cctxParams->cParams = params->cParams;
+    cctxParams->fParams = params->fParams;
+    /* Should not matter, as all cParams are presumed properly defined.
+     * But, set it for tracing anyway.
+     */
+    cctxParams->compressionLevel = ZSTD_NO_CLEVEL;
+}
+
+ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
+{
+    ZSTD_bounds bounds = { 0, 0, 0 };
+
+    switch(param)
+    {
+    case ZSTD_c_compressionLevel:
+        bounds.lowerBound = ZSTD_minCLevel();
+        bounds.upperBound = ZSTD_maxCLevel();
+        return bounds;
+
+    case ZSTD_c_windowLog:
+        bounds.lowerBound = ZSTD_WINDOWLOG_MIN;
+        bounds.upperBound = ZSTD_WINDOWLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_hashLog:
+        bounds.lowerBound = ZSTD_HASHLOG_MIN;
+        bounds.upperBound = ZSTD_HASHLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_chainLog:
+        bounds.lowerBound = ZSTD_CHAINLOG_MIN;
+        bounds.upperBound = ZSTD_CHAINLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_searchLog:
+        bounds.lowerBound = ZSTD_SEARCHLOG_MIN;
+        bounds.upperBound = ZSTD_SEARCHLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_minMatch:
+        bounds.lowerBound = ZSTD_MINMATCH_MIN;
+        bounds.upperBound = ZSTD_MINMATCH_MAX;
+        return bounds;
+
+    case ZSTD_c_targetLength:
+        bounds.lowerBound = ZSTD_TARGETLENGTH_MIN;
+        bounds.upperBound = ZSTD_TARGETLENGTH_MAX;
+        return bounds;
+
+    case ZSTD_c_strategy:
+        bounds.lowerBound = ZSTD_STRATEGY_MIN;
+        bounds.upperBound = ZSTD_STRATEGY_MAX;
+        return bounds;
+
+    case ZSTD_c_contentSizeFlag:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_checksumFlag:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_dictIDFlag:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_nbWorkers:
+        bounds.lowerBound = 0;
+#ifdef ZSTD_MULTITHREAD
+        bounds.upperBound = ZSTDMT_NBWORKERS_MAX;
+#else
+        bounds.upperBound = 0;
+#endif
+        return bounds;
+
+    case ZSTD_c_jobSize:
+        bounds.lowerBound = 0;
+#ifdef ZSTD_MULTITHREAD
+        bounds.upperBound = ZSTDMT_JOBSIZE_MAX;
+#else
+        bounds.upperBound = 0;
+#endif
+        return bounds;
+
+    case ZSTD_c_overlapLog:
+#ifdef ZSTD_MULTITHREAD
+        bounds.lowerBound = ZSTD_OVERLAPLOG_MIN;
+        bounds.upperBound = ZSTD_OVERLAPLOG_MAX;
+#else
+        bounds.lowerBound = 0;
+        bounds.upperBound = 0;
+#endif
+        return bounds;
+
+    case ZSTD_c_enableDedicatedDictSearch:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_enableLongDistanceMatching:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_ldmHashLog:
+        bounds.lowerBound = ZSTD_LDM_HASHLOG_MIN;
+        bounds.upperBound = ZSTD_LDM_HASHLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_ldmMinMatch:
+        bounds.lowerBound = ZSTD_LDM_MINMATCH_MIN;
+        bounds.upperBound = ZSTD_LDM_MINMATCH_MAX;
+        return bounds;
+
+    case ZSTD_c_ldmBucketSizeLog:
+        bounds.lowerBound = ZSTD_LDM_BUCKETSIZELOG_MIN;
+        bounds.upperBound = ZSTD_LDM_BUCKETSIZELOG_MAX;
+        return bounds;
+
+    case ZSTD_c_ldmHashRateLog:
+        bounds.lowerBound = ZSTD_LDM_HASHRATELOG_MIN;
+        bounds.upperBound = ZSTD_LDM_HASHRATELOG_MAX;
+        return bounds;
+
+    /* experimental parameters */
+    case ZSTD_c_rsyncable:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_forceMaxWindow :
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_format:
+        ZSTD_STATIC_ASSERT(ZSTD_f_zstd1 < ZSTD_f_zstd1_magicless);
+        bounds.lowerBound = ZSTD_f_zstd1;
+        bounds.upperBound = ZSTD_f_zstd1_magicless;   /* note : how to ensure at compile time that this is the highest value enum ? */
+        return bounds;
+
+    case ZSTD_c_forceAttachDict:
+        ZSTD_STATIC_ASSERT(ZSTD_dictDefaultAttach < ZSTD_dictForceLoad);
+        bounds.lowerBound = ZSTD_dictDefaultAttach;
+        bounds.upperBound = ZSTD_dictForceLoad;       /* note : how to ensure at compile time that this is the highest value enum ? */
+        return bounds;
+
+    case ZSTD_c_literalCompressionMode:
+        ZSTD_STATIC_ASSERT(ZSTD_ps_auto < ZSTD_ps_enable && ZSTD_ps_enable < ZSTD_ps_disable);
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_targetCBlockSize:
+        bounds.lowerBound = ZSTD_TARGETCBLOCKSIZE_MIN;
+        bounds.upperBound = ZSTD_TARGETCBLOCKSIZE_MAX;
+        return bounds;
+
+    case ZSTD_c_srcSizeHint:
+        bounds.lowerBound = ZSTD_SRCSIZEHINT_MIN;
+        bounds.upperBound = ZSTD_SRCSIZEHINT_MAX;
+        return bounds;
+
+    case ZSTD_c_stableInBuffer:
+    case ZSTD_c_stableOutBuffer:
+        bounds.lowerBound = (int)ZSTD_bm_buffered;
+        bounds.upperBound = (int)ZSTD_bm_stable;
+        return bounds;
+
+    case ZSTD_c_blockDelimiters:
+        bounds.lowerBound = (int)ZSTD_sf_noBlockDelimiters;
+        bounds.upperBound = (int)ZSTD_sf_explicitBlockDelimiters;
+        return bounds;
+
+    case ZSTD_c_validateSequences:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_splitAfterSequences:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_blockSplitterLevel:
+        bounds.lowerBound = 0;
+        bounds.upperBound = ZSTD_BLOCKSPLITTER_LEVEL_MAX;
+        return bounds;
+
+    case ZSTD_c_useRowMatchFinder:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_deterministicRefPrefix:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_prefetchCDictTables:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_enableSeqProducerFallback:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_maxBlockSize:
+        bounds.lowerBound = ZSTD_BLOCKSIZE_MAX_MIN;
+        bounds.upperBound = ZSTD_BLOCKSIZE_MAX;
+        return bounds;
+
+    case ZSTD_c_repcodeResolution:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    default:
+        bounds.error = ERROR(parameter_unsupported);
+        return bounds;
+    }
+}
+
+/* ZSTD_cParam_clampBounds:
+ * Clamps the value into the bounded range.
+ */
+static size_t ZSTD_cParam_clampBounds(ZSTD_cParameter cParam, int* value)
+{
+    ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);
+    if (ZSTD_isError(bounds.error)) return bounds.error;
+    if (*value < bounds.lowerBound) *value = bounds.lowerBound;
+    if (*value > bounds.upperBound) *value = bounds.upperBound;
+    return 0;
+}
+
+#define BOUNDCHECK(cParam, val)                                       \
+    do {                                                              \
+        RETURN_ERROR_IF(!ZSTD_cParam_withinBounds(cParam,val),        \
+                        parameter_outOfBound, "Param out of bounds"); \
+    } while (0)
+
+
+static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
+{
+    switch(param)
+    {
+    case ZSTD_c_compressionLevel:
+    case ZSTD_c_hashLog:
+    case ZSTD_c_chainLog:
+    case ZSTD_c_searchLog:
+    case ZSTD_c_minMatch:
+    case ZSTD_c_targetLength:
+    case ZSTD_c_strategy:
+    case ZSTD_c_blockSplitterLevel:
+        return 1;
+
+    case ZSTD_c_format:
+    case ZSTD_c_windowLog:
+    case ZSTD_c_contentSizeFlag:
+    case ZSTD_c_checksumFlag:
+    case ZSTD_c_dictIDFlag:
+    case ZSTD_c_forceMaxWindow :
+    case ZSTD_c_nbWorkers:
+    case ZSTD_c_jobSize:
+    case ZSTD_c_overlapLog:
+    case ZSTD_c_rsyncable:
+    case ZSTD_c_enableDedicatedDictSearch:
+    case ZSTD_c_enableLongDistanceMatching:
+    case ZSTD_c_ldmHashLog:
+    case ZSTD_c_ldmMinMatch:
+    case ZSTD_c_ldmBucketSizeLog:
+    case ZSTD_c_ldmHashRateLog:
+    case ZSTD_c_forceAttachDict:
+    case ZSTD_c_literalCompressionMode:
+    case ZSTD_c_targetCBlockSize:
+    case ZSTD_c_srcSizeHint:
+    case ZSTD_c_stableInBuffer:
+    case ZSTD_c_stableOutBuffer:
+    case ZSTD_c_blockDelimiters:
+    case ZSTD_c_validateSequences:
+    case ZSTD_c_splitAfterSequences:
+    case ZSTD_c_useRowMatchFinder:
+    case ZSTD_c_deterministicRefPrefix:
+    case ZSTD_c_prefetchCDictTables:
+    case ZSTD_c_enableSeqProducerFallback:
+    case ZSTD_c_maxBlockSize:
+    case ZSTD_c_repcodeResolution:
+    default:
+        return 0;
+    }
+}
+
+size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setParameter (%i, %i)", (int)param, value);
+    if (cctx->streamStage != zcss_init) {
+        if (ZSTD_isUpdateAuthorized(param)) {
+            cctx->cParamsChanged = 1;
+        } else {
+            RETURN_ERROR(stage_wrong, "can only set params in cctx init stage");
+    }   }
+
+    switch(param)
+    {
+    case ZSTD_c_nbWorkers:
+        RETURN_ERROR_IF((value!=0) && cctx->staticSize, parameter_unsupported,
+                        "MT not compatible with static alloc");
+        break;
+
+    case ZSTD_c_compressionLevel:
+    case ZSTD_c_windowLog:
+    case ZSTD_c_hashLog:
+    case ZSTD_c_chainLog:
+    case ZSTD_c_searchLog:
+    case ZSTD_c_minMatch:
+    case ZSTD_c_targetLength:
+    case ZSTD_c_strategy:
+    case ZSTD_c_ldmHashRateLog:
+    case ZSTD_c_format:
+    case ZSTD_c_contentSizeFlag:
+    case ZSTD_c_checksumFlag:
+    case ZSTD_c_dictIDFlag:
+    case ZSTD_c_forceMaxWindow:
+    case ZSTD_c_forceAttachDict:
+    case ZSTD_c_literalCompressionMode:
+    case ZSTD_c_jobSize:
+    case ZSTD_c_overlapLog:
+    case ZSTD_c_rsyncable:
+    case ZSTD_c_enableDedicatedDictSearch:
+    case ZSTD_c_enableLongDistanceMatching:
+    case ZSTD_c_ldmHashLog:
+    case ZSTD_c_ldmMinMatch:
+    case ZSTD_c_ldmBucketSizeLog:
+    case ZSTD_c_targetCBlockSize:
+    case ZSTD_c_srcSizeHint:
+    case ZSTD_c_stableInBuffer:
+    case ZSTD_c_stableOutBuffer:
+    case ZSTD_c_blockDelimiters:
+    case ZSTD_c_validateSequences:
+    case ZSTD_c_splitAfterSequences:
+    case ZSTD_c_blockSplitterLevel:
+    case ZSTD_c_useRowMatchFinder:
+    case ZSTD_c_deterministicRefPrefix:
+    case ZSTD_c_prefetchCDictTables:
+    case ZSTD_c_enableSeqProducerFallback:
+    case ZSTD_c_maxBlockSize:
+    case ZSTD_c_repcodeResolution:
+        break;
+
+    default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
+    }
+    return ZSTD_CCtxParams_setParameter(&cctx->requestedParams, param, value);
+}
+
+size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
+                                    ZSTD_cParameter param, int value)
+{
+    DEBUGLOG(4, "ZSTD_CCtxParams_setParameter (%i, %i)", (int)param, value);
+    switch(param)
+    {
+    case ZSTD_c_format :
+        BOUNDCHECK(ZSTD_c_format, value);
+        CCtxParams->format = (ZSTD_format_e)value;
+        return (size_t)CCtxParams->format;
+
+    case ZSTD_c_compressionLevel : {
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
+        if (value == 0)
+            CCtxParams->compressionLevel = ZSTD_CLEVEL_DEFAULT; /* 0 == default */
+        else
+            CCtxParams->compressionLevel = value;
+        if (CCtxParams->compressionLevel >= 0) return (size_t)CCtxParams->compressionLevel;
+        return 0;  /* return type (size_t) cannot represent negative values */
+    }
+
+    case ZSTD_c_windowLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_windowLog, value);
+        CCtxParams->cParams.windowLog = (U32)value;
+        return CCtxParams->cParams.windowLog;
+
+    case ZSTD_c_hashLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_hashLog, value);
+        CCtxParams->cParams.hashLog = (U32)value;
+        return CCtxParams->cParams.hashLog;
+
+    case ZSTD_c_chainLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_chainLog, value);
+        CCtxParams->cParams.chainLog = (U32)value;
+        return CCtxParams->cParams.chainLog;
+
+    case ZSTD_c_searchLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_searchLog, value);
+        CCtxParams->cParams.searchLog = (U32)value;
+        return (size_t)value;
+
+    case ZSTD_c_minMatch :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_minMatch, value);
+        CCtxParams->cParams.minMatch = (U32)value;
+        return CCtxParams->cParams.minMatch;
+
+    case ZSTD_c_targetLength :
+        BOUNDCHECK(ZSTD_c_targetLength, value);
+        CCtxParams->cParams.targetLength = (U32)value;
+        return CCtxParams->cParams.targetLength;
+
+    case ZSTD_c_strategy :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_strategy, value);
+        CCtxParams->cParams.strategy = (ZSTD_strategy)value;
+        return (size_t)CCtxParams->cParams.strategy;
+
+    case ZSTD_c_contentSizeFlag :
+        /* Content size written in frame header _when known_ (default:1) */
+        DEBUGLOG(4, "set content size flag = %u", (value!=0));
+        CCtxParams->fParams.contentSizeFlag = value != 0;
+        return (size_t)CCtxParams->fParams.contentSizeFlag;
+
+    case ZSTD_c_checksumFlag :
+        /* A 32-bits content checksum will be calculated and written at end of frame (default:0) */
+        CCtxParams->fParams.checksumFlag = value != 0;
+        return (size_t)CCtxParams->fParams.checksumFlag;
+
+    case ZSTD_c_dictIDFlag : /* When applicable, dictionary's dictID is provided in frame header (default:1) */
+        DEBUGLOG(4, "set dictIDFlag = %u", (value!=0));
+        CCtxParams->fParams.noDictIDFlag = !value;
+        return !CCtxParams->fParams.noDictIDFlag;
+
+    case ZSTD_c_forceMaxWindow :
+        CCtxParams->forceWindow = (value != 0);
+        return (size_t)CCtxParams->forceWindow;
+
+    case ZSTD_c_forceAttachDict : {
+        const ZSTD_dictAttachPref_e pref = (ZSTD_dictAttachPref_e)value;
+        BOUNDCHECK(ZSTD_c_forceAttachDict, (int)pref);
+        CCtxParams->attachDictPref = pref;
+        return CCtxParams->attachDictPref;
+    }
+
+    case ZSTD_c_literalCompressionMode : {
+        const ZSTD_ParamSwitch_e lcm = (ZSTD_ParamSwitch_e)value;
+        BOUNDCHECK(ZSTD_c_literalCompressionMode, (int)lcm);
+        CCtxParams->literalCompressionMode = lcm;
+        return CCtxParams->literalCompressionMode;
+    }
+
+    case ZSTD_c_nbWorkers :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
+        return 0;
+#else
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
+        CCtxParams->nbWorkers = value;
+        return (size_t)(CCtxParams->nbWorkers);
+#endif
+
+    case ZSTD_c_jobSize :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
+        return 0;
+#else
+        /* Adjust to the minimum non-default value. */
+        if (value != 0 && value < ZSTDMT_JOBSIZE_MIN)
+            value = ZSTDMT_JOBSIZE_MIN;
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
+        assert(value >= 0);
+        CCtxParams->jobSize = (size_t)value;
+        return CCtxParams->jobSize;
+#endif
+
+    case ZSTD_c_overlapLog :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
+        return 0;
+#else
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(ZSTD_c_overlapLog, &value), "");
+        CCtxParams->overlapLog = value;
+        return (size_t)CCtxParams->overlapLog;
+#endif
+
+    case ZSTD_c_rsyncable :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
+        return 0;
+#else
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(ZSTD_c_overlapLog, &value), "");
+        CCtxParams->rsyncable = value;
+        return (size_t)CCtxParams->rsyncable;
+#endif
+
+    case ZSTD_c_enableDedicatedDictSearch :
+        CCtxParams->enableDedicatedDictSearch = (value!=0);
+        return (size_t)CCtxParams->enableDedicatedDictSearch;
+
+    case ZSTD_c_enableLongDistanceMatching :
+        BOUNDCHECK(ZSTD_c_enableLongDistanceMatching, value);
+        CCtxParams->ldmParams.enableLdm = (ZSTD_ParamSwitch_e)value;
+        return CCtxParams->ldmParams.enableLdm;
+
+    case ZSTD_c_ldmHashLog :
+        if (value!=0)   /* 0 ==> auto */
+            BOUNDCHECK(ZSTD_c_ldmHashLog, value);
+        CCtxParams->ldmParams.hashLog = (U32)value;
+        return CCtxParams->ldmParams.hashLog;
+
+    case ZSTD_c_ldmMinMatch :
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_ldmMinMatch, value);
+        CCtxParams->ldmParams.minMatchLength = (U32)value;
+        return CCtxParams->ldmParams.minMatchLength;
+
+    case ZSTD_c_ldmBucketSizeLog :
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_ldmBucketSizeLog, value);
+        CCtxParams->ldmParams.bucketSizeLog = (U32)value;
+        return CCtxParams->ldmParams.bucketSizeLog;
+
+    case ZSTD_c_ldmHashRateLog :
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_ldmHashRateLog, value);
+        CCtxParams->ldmParams.hashRateLog = (U32)value;
+        return CCtxParams->ldmParams.hashRateLog;
+
+    case ZSTD_c_targetCBlockSize :
+        if (value!=0) {  /* 0 ==> default */
+            value = MAX(value, ZSTD_TARGETCBLOCKSIZE_MIN);
+            BOUNDCHECK(ZSTD_c_targetCBlockSize, value);
+        }
+        CCtxParams->targetCBlockSize = (U32)value;
+        return CCtxParams->targetCBlockSize;
+
+    case ZSTD_c_srcSizeHint :
+        if (value!=0)    /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_srcSizeHint, value);
+        CCtxParams->srcSizeHint = value;
+        return (size_t)CCtxParams->srcSizeHint;
+
+    case ZSTD_c_stableInBuffer:
+        BOUNDCHECK(ZSTD_c_stableInBuffer, value);
+        CCtxParams->inBufferMode = (ZSTD_bufferMode_e)value;
+        return CCtxParams->inBufferMode;
+
+    case ZSTD_c_stableOutBuffer:
+        BOUNDCHECK(ZSTD_c_stableOutBuffer, value);
+        CCtxParams->outBufferMode = (ZSTD_bufferMode_e)value;
+        return CCtxParams->outBufferMode;
+
+    case ZSTD_c_blockDelimiters:
+        BOUNDCHECK(ZSTD_c_blockDelimiters, value);
+        CCtxParams->blockDelimiters = (ZSTD_SequenceFormat_e)value;
+        return CCtxParams->blockDelimiters;
+
+    case ZSTD_c_validateSequences:
+        BOUNDCHECK(ZSTD_c_validateSequences, value);
+        CCtxParams->validateSequences = value;
+        return (size_t)CCtxParams->validateSequences;
+
+    case ZSTD_c_splitAfterSequences:
+        BOUNDCHECK(ZSTD_c_splitAfterSequences, value);
+        CCtxParams->postBlockSplitter = (ZSTD_ParamSwitch_e)value;
+        return CCtxParams->postBlockSplitter;
+
+    case ZSTD_c_blockSplitterLevel:
+        BOUNDCHECK(ZSTD_c_blockSplitterLevel, value);
+        CCtxParams->preBlockSplitter_level = value;
+        return (size_t)CCtxParams->preBlockSplitter_level;
+
+    case ZSTD_c_useRowMatchFinder:
+        BOUNDCHECK(ZSTD_c_useRowMatchFinder, value);
+        CCtxParams->useRowMatchFinder = (ZSTD_ParamSwitch_e)value;
+        return CCtxParams->useRowMatchFinder;
+
+    case ZSTD_c_deterministicRefPrefix:
+        BOUNDCHECK(ZSTD_c_deterministicRefPrefix, value);
+        CCtxParams->deterministicRefPrefix = !!value;
+        return (size_t)CCtxParams->deterministicRefPrefix;
+
+    case ZSTD_c_prefetchCDictTables:
+        BOUNDCHECK(ZSTD_c_prefetchCDictTables, value);
+        CCtxParams->prefetchCDictTables = (ZSTD_ParamSwitch_e)value;
+        return CCtxParams->prefetchCDictTables;
+
+    case ZSTD_c_enableSeqProducerFallback:
+        BOUNDCHECK(ZSTD_c_enableSeqProducerFallback, value);
+        CCtxParams->enableMatchFinderFallback = value;
+        return (size_t)CCtxParams->enableMatchFinderFallback;
+
+    case ZSTD_c_maxBlockSize:
+        if (value!=0)    /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_maxBlockSize, value);
+        assert(value>=0);
+        CCtxParams->maxBlockSize = (size_t)value;
+        return CCtxParams->maxBlockSize;
+
+    case ZSTD_c_repcodeResolution:
+        BOUNDCHECK(ZSTD_c_repcodeResolution, value);
+        CCtxParams->searchForExternalRepcodes = (ZSTD_ParamSwitch_e)value;
+        return CCtxParams->searchForExternalRepcodes;
+
+    default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
+    }
+}
+
+size_t ZSTD_CCtx_getParameter(ZSTD_CCtx const* cctx, ZSTD_cParameter param, int* value)
+{
+    return ZSTD_CCtxParams_getParameter(&cctx->requestedParams, param, value);
+}
+
+size_t ZSTD_CCtxParams_getParameter(
+        ZSTD_CCtx_params const* CCtxParams, ZSTD_cParameter param, int* value)
+{
+    switch(param)
+    {
+    case ZSTD_c_format :
+        *value = (int)CCtxParams->format;
+        break;
+    case ZSTD_c_compressionLevel :
+        *value = CCtxParams->compressionLevel;
+        break;
+    case ZSTD_c_windowLog :
+        *value = (int)CCtxParams->cParams.windowLog;
+        break;
+    case ZSTD_c_hashLog :
+        *value = (int)CCtxParams->cParams.hashLog;
+        break;
+    case ZSTD_c_chainLog :
+        *value = (int)CCtxParams->cParams.chainLog;
+        break;
+    case ZSTD_c_searchLog :
+        *value = (int)CCtxParams->cParams.searchLog;
+        break;
+    case ZSTD_c_minMatch :
+        *value = (int)CCtxParams->cParams.minMatch;
+        break;
+    case ZSTD_c_targetLength :
+        *value = (int)CCtxParams->cParams.targetLength;
+        break;
+    case ZSTD_c_strategy :
+        *value = (int)CCtxParams->cParams.strategy;
+        break;
+    case ZSTD_c_contentSizeFlag :
+        *value = CCtxParams->fParams.contentSizeFlag;
+        break;
+    case ZSTD_c_checksumFlag :
+        *value = CCtxParams->fParams.checksumFlag;
+        break;
+    case ZSTD_c_dictIDFlag :
+        *value = !CCtxParams->fParams.noDictIDFlag;
+        break;
+    case ZSTD_c_forceMaxWindow :
+        *value = CCtxParams->forceWindow;
+        break;
+    case ZSTD_c_forceAttachDict :
+        *value = (int)CCtxParams->attachDictPref;
+        break;
+    case ZSTD_c_literalCompressionMode :
+        *value = (int)CCtxParams->literalCompressionMode;
+        break;
+    case ZSTD_c_nbWorkers :
+#ifndef ZSTD_MULTITHREAD
+        assert(CCtxParams->nbWorkers == 0);
+#endif
+        *value = CCtxParams->nbWorkers;
+        break;
+    case ZSTD_c_jobSize :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
+#else
+        assert(CCtxParams->jobSize <= INT_MAX);
+        *value = (int)CCtxParams->jobSize;
+        break;
+#endif
+    case ZSTD_c_overlapLog :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
+#else
+        *value = CCtxParams->overlapLog;
+        break;
+#endif
+    case ZSTD_c_rsyncable :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
+#else
+        *value = CCtxParams->rsyncable;
+        break;
+#endif
+    case ZSTD_c_enableDedicatedDictSearch :
+        *value = CCtxParams->enableDedicatedDictSearch;
+        break;
+    case ZSTD_c_enableLongDistanceMatching :
+        *value = (int)CCtxParams->ldmParams.enableLdm;
+        break;
+    case ZSTD_c_ldmHashLog :
+        *value = (int)CCtxParams->ldmParams.hashLog;
+        break;
+    case ZSTD_c_ldmMinMatch :
+        *value = (int)CCtxParams->ldmParams.minMatchLength;
+        break;
+    case ZSTD_c_ldmBucketSizeLog :
+        *value = (int)CCtxParams->ldmParams.bucketSizeLog;
+        break;
+    case ZSTD_c_ldmHashRateLog :
+        *value = (int)CCtxParams->ldmParams.hashRateLog;
+        break;
+    case ZSTD_c_targetCBlockSize :
+        *value = (int)CCtxParams->targetCBlockSize;
+        break;
+    case ZSTD_c_srcSizeHint :
+        *value = (int)CCtxParams->srcSizeHint;
+        break;
+    case ZSTD_c_stableInBuffer :
+        *value = (int)CCtxParams->inBufferMode;
+        break;
+    case ZSTD_c_stableOutBuffer :
+        *value = (int)CCtxParams->outBufferMode;
+        break;
+    case ZSTD_c_blockDelimiters :
+        *value = (int)CCtxParams->blockDelimiters;
+        break;
+    case ZSTD_c_validateSequences :
+        *value = (int)CCtxParams->validateSequences;
+        break;
+    case ZSTD_c_splitAfterSequences :
+        *value = (int)CCtxParams->postBlockSplitter;
+        break;
+    case ZSTD_c_blockSplitterLevel :
+        *value = CCtxParams->preBlockSplitter_level;
+        break;
+    case ZSTD_c_useRowMatchFinder :
+        *value = (int)CCtxParams->useRowMatchFinder;
+        break;
+    case ZSTD_c_deterministicRefPrefix:
+        *value = (int)CCtxParams->deterministicRefPrefix;
+        break;
+    case ZSTD_c_prefetchCDictTables:
+        *value = (int)CCtxParams->prefetchCDictTables;
+        break;
+    case ZSTD_c_enableSeqProducerFallback:
+        *value = CCtxParams->enableMatchFinderFallback;
+        break;
+    case ZSTD_c_maxBlockSize:
+        *value = (int)CCtxParams->maxBlockSize;
+        break;
+    case ZSTD_c_repcodeResolution:
+        *value = (int)CCtxParams->searchForExternalRepcodes;
+        break;
+    default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
+    }
+    return 0;
+}
+
+/** ZSTD_CCtx_setParametersUsingCCtxParams() :
+ *  just applies `params` into `cctx`
+ *  no action is performed, parameters are merely stored.
+ *  If ZSTDMT is enabled, parameters are pushed to cctx->mtctx.
+ *    This is possible even if a compression is ongoing.
+ *    In which case, new parameters will be applied on the fly, starting with next compression job.
+ */
+size_t ZSTD_CCtx_setParametersUsingCCtxParams(
+        ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setParametersUsingCCtxParams");
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "The context is in the wrong stage!");
+    RETURN_ERROR_IF(cctx->cdict, stage_wrong,
+                    "Can't override parameters with cdict attached (some must "
+                    "be inherited from the cdict).");
+
+    cctx->requestedParams = *params;
+    return 0;
+}
+
+size_t ZSTD_CCtx_setCParams(ZSTD_CCtx* cctx, ZSTD_compressionParameters cparams)
+{
+    ZSTD_STATIC_ASSERT(sizeof(cparams) == 7 * 4 /* all params are listed below */);
+    DEBUGLOG(4, "ZSTD_CCtx_setCParams");
+    /* only update if all parameters are valid */
+    FORWARD_IF_ERROR(ZSTD_checkCParams(cparams), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_windowLog, (int)cparams.windowLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_chainLog, (int)cparams.chainLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_hashLog, (int)cparams.hashLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_searchLog, (int)cparams.searchLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_minMatch, (int)cparams.minMatch), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_targetLength, (int)cparams.targetLength), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_strategy, (int)cparams.strategy), "");
+    return 0;
+}
+
+size_t ZSTD_CCtx_setFParams(ZSTD_CCtx* cctx, ZSTD_frameParameters fparams)
+{
+    ZSTD_STATIC_ASSERT(sizeof(fparams) == 3 * 4 /* all params are listed below */);
+    DEBUGLOG(4, "ZSTD_CCtx_setFParams");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_contentSizeFlag, fparams.contentSizeFlag != 0), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, fparams.checksumFlag != 0), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(cctx, ZSTD_c_dictIDFlag, fparams.noDictIDFlag == 0), "");
+    return 0;
+}
+
+size_t ZSTD_CCtx_setParams(ZSTD_CCtx* cctx, ZSTD_parameters params)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setParams");
+    /* First check cParams, because we want to update all or none. */
+    FORWARD_IF_ERROR(ZSTD_checkCParams(params.cParams), "");
+    /* Next set fParams, because this could fail if the cctx isn't in init stage. */
+    FORWARD_IF_ERROR(ZSTD_CCtx_setFParams(cctx, params.fParams), "");
+    /* Finally set cParams, which should succeed. */
+    FORWARD_IF_ERROR(ZSTD_CCtx_setCParams(cctx, params.cParams), "");
+    return 0;
+}
+
+size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setPledgedSrcSize to %llu bytes", pledgedSrcSize);
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't set pledgedSrcSize when not in init stage.");
+    cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
+    return 0;
+}
+
+static ZSTD_compressionParameters ZSTD_dedicatedDictSearch_getCParams(
+        int const compressionLevel,
+        size_t const dictSize);
+static int ZSTD_dedicatedDictSearch_isSupported(
+        const ZSTD_compressionParameters* cParams);
+static void ZSTD_dedicatedDictSearch_revertCParams(
+        ZSTD_compressionParameters* cParams);
+
+/**
+ * Initializes the local dictionary using requested parameters.
+ * NOTE: Initialization does not employ the pledged src size,
+ * because the dictionary may be used for multiple compressions.
+ */
+static size_t ZSTD_initLocalDict(ZSTD_CCtx* cctx)
+{
+    ZSTD_localDict* const dl = &cctx->localDict;
+    if (dl->dict == NULL) {
+        /* No local dictionary. */
+        assert(dl->dictBuffer == NULL);
+        assert(dl->cdict == NULL);
+        assert(dl->dictSize == 0);
+        return 0;
+    }
+    if (dl->cdict != NULL) {
+        /* Local dictionary already initialized. */
+        assert(cctx->cdict == dl->cdict);
+        return 0;
+    }
+    assert(dl->dictSize > 0);
+    assert(cctx->cdict == NULL);
+    assert(cctx->prefixDict.dict == NULL);
+
+    dl->cdict = ZSTD_createCDict_advanced2(
+            dl->dict,
+            dl->dictSize,
+            ZSTD_dlm_byRef,
+            dl->dictContentType,
+            &cctx->requestedParams,
+            cctx->customMem);
+    RETURN_ERROR_IF(!dl->cdict, memory_allocation, "ZSTD_createCDict_advanced failed");
+    cctx->cdict = dl->cdict;
+    return 0;
+}
+
+size_t ZSTD_CCtx_loadDictionary_advanced(
+        ZSTD_CCtx* cctx,
+        const void* dict, size_t dictSize,
+        ZSTD_dictLoadMethod_e dictLoadMethod,
+        ZSTD_dictContentType_e dictContentType)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_loadDictionary_advanced (size: %u)", (U32)dictSize);
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't load a dictionary when cctx is not in init stage.");
+    ZSTD_clearAllDicts(cctx);  /* erase any previously set dictionary */
+    if (dict == NULL || dictSize == 0)  /* no dictionary */
+        return 0;
+    if (dictLoadMethod == ZSTD_dlm_byRef) {
+        cctx->localDict.dict = dict;
+    } else {
+        /* copy dictionary content inside CCtx to own its lifetime */
+        void* dictBuffer;
+        RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
+                        "static CCtx can't allocate for an internal copy of dictionary");
+        dictBuffer = ZSTD_customMalloc(dictSize, cctx->customMem);
+        RETURN_ERROR_IF(dictBuffer==NULL, memory_allocation,
+                        "allocation failed for dictionary content");
+        ZSTD_memcpy(dictBuffer, dict, dictSize);
+        cctx->localDict.dictBuffer = dictBuffer;  /* owned ptr to free */
+        cctx->localDict.dict = dictBuffer;        /* read-only reference */
+    }
+    cctx->localDict.dictSize = dictSize;
+    cctx->localDict.dictContentType = dictContentType;
+    return 0;
+}
+
+size_t ZSTD_CCtx_loadDictionary_byReference(
+      ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_CCtx_loadDictionary_advanced(
+            cctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto);
+}
+
+size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_CCtx_loadDictionary_advanced(
+            cctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto);
+}
+
+
+size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't ref a dict when ctx not in init stage.");
+    /* Free the existing local cdict (if any) to save memory. */
+    ZSTD_clearAllDicts(cctx);
+    cctx->cdict = cdict;
+    return 0;
+}
+
+size_t ZSTD_CCtx_refThreadPool(ZSTD_CCtx* cctx, ZSTD_threadPool* pool)
+{
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't ref a pool when ctx not in init stage.");
+    cctx->pool = pool;
+    return 0;
+}
+
+size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize)
+{
+    return ZSTD_CCtx_refPrefix_advanced(cctx, prefix, prefixSize, ZSTD_dct_rawContent);
+}
+
+size_t ZSTD_CCtx_refPrefix_advanced(
+        ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
+{
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't ref a prefix when ctx not in init stage.");
+    ZSTD_clearAllDicts(cctx);
+    if (prefix != NULL && prefixSize > 0) {
+        cctx->prefixDict.dict = prefix;
+        cctx->prefixDict.dictSize = prefixSize;
+        cctx->prefixDict.dictContentType = dictContentType;
+    }
+    return 0;
+}
+
+/*! ZSTD_CCtx_reset() :
+ *  Also dumps dictionary */
+size_t ZSTD_CCtx_reset(ZSTD_CCtx* cctx, ZSTD_ResetDirective reset)
+{
+    if ( (reset == ZSTD_reset_session_only)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        cctx->streamStage = zcss_init;
+        cctx->pledgedSrcSizePlusOne = 0;
+    }
+    if ( (reset == ZSTD_reset_parameters)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                        "Reset parameters is only possible during init stage.");
+        ZSTD_clearAllDicts(cctx);
+        return ZSTD_CCtxParams_reset(&cctx->requestedParams);
+    }
+    return 0;
+}
+
+
+/** ZSTD_checkCParams() :
+    control CParam values remain within authorized range.
+    @return : 0, or an error code if one value is beyond authorized range */
+size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
+{
+    BOUNDCHECK(ZSTD_c_windowLog, (int)cParams.windowLog);
+    BOUNDCHECK(ZSTD_c_chainLog,  (int)cParams.chainLog);
+    BOUNDCHECK(ZSTD_c_hashLog,   (int)cParams.hashLog);
+    BOUNDCHECK(ZSTD_c_searchLog, (int)cParams.searchLog);
+    BOUNDCHECK(ZSTD_c_minMatch,  (int)cParams.minMatch);
+    BOUNDCHECK(ZSTD_c_targetLength,(int)cParams.targetLength);
+    BOUNDCHECK(ZSTD_c_strategy,  (int)cParams.strategy);
+    return 0;
+}
+
+/** ZSTD_clampCParams() :
+ *  make CParam values within valid range.
+ *  @return : valid CParams */
+static ZSTD_compressionParameters
+ZSTD_clampCParams(ZSTD_compressionParameters cParams)
+{
+#   define CLAMP_TYPE(cParam, val, type)                                      \
+        do {                                                                  \
+            ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);         \
+            if ((int)val<bounds.lowerBound) val=(type)bounds.lowerBound;      \
+            else if ((int)val>bounds.upperBound) val=(type)bounds.upperBound; \
+        } while (0)
+#   define CLAMP(cParam, val) CLAMP_TYPE(cParam, val, unsigned)
+    CLAMP(ZSTD_c_windowLog, cParams.windowLog);
+    CLAMP(ZSTD_c_chainLog,  cParams.chainLog);
+    CLAMP(ZSTD_c_hashLog,   cParams.hashLog);
+    CLAMP(ZSTD_c_searchLog, cParams.searchLog);
+    CLAMP(ZSTD_c_minMatch,  cParams.minMatch);
+    CLAMP(ZSTD_c_targetLength,cParams.targetLength);
+    CLAMP_TYPE(ZSTD_c_strategy,cParams.strategy, ZSTD_strategy);
+    return cParams;
+}
+
+/** ZSTD_cycleLog() :
+ *  condition for correct operation : hashLog > 1 */
+U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat)
+{
+    U32 const btScale = ((U32)strat >= (U32)ZSTD_btlazy2);
+    return hashLog - btScale;
+}
+
+/** ZSTD_dictAndWindowLog() :
+ * Returns an adjusted window log that is large enough to fit the source and the dictionary.
+ * The zstd format says that the entire dictionary is valid if one byte of the dictionary
+ * is within the window. So the hashLog and chainLog should be large enough to reference both
+ * the dictionary and the window. So we must use this adjusted dictAndWindowLog when downsizing
+ * the hashLog and windowLog.
+ * NOTE: srcSize must not be ZSTD_CONTENTSIZE_UNKNOWN.
+ */
+static U32 ZSTD_dictAndWindowLog(U32 windowLog, U64 srcSize, U64 dictSize)
+{
+    const U64 maxWindowSize = 1ULL << ZSTD_WINDOWLOG_MAX;
+    /* No dictionary ==> No change */
+    if (dictSize == 0) {
+        return windowLog;
+    }
+    assert(windowLog <= ZSTD_WINDOWLOG_MAX);
+    assert(srcSize != ZSTD_CONTENTSIZE_UNKNOWN); /* Handled in ZSTD_adjustCParams_internal() */
+    {
+        U64 const windowSize = 1ULL << windowLog;
+        U64 const dictAndWindowSize = dictSize + windowSize;
+        /* If the window size is already large enough to fit both the source and the dictionary
+         * then just use the window size. Otherwise adjust so that it fits the dictionary and
+         * the window.
+         */
+        if (windowSize >= dictSize + srcSize) {
+            return windowLog; /* Window size large enough already */
+        } else if (dictAndWindowSize >= maxWindowSize) {
+            return ZSTD_WINDOWLOG_MAX; /* Larger than max window log */
+        } else  {
+            return ZSTD_highbit32((U32)dictAndWindowSize - 1) + 1;
+        }
+    }
+}
+
+/** ZSTD_adjustCParams_internal() :
+ *  optimize `cPar` for a specified input (`srcSize` and `dictSize`).
+ *  mostly downsize to reduce memory consumption and initialization latency.
+ * `srcSize` can be ZSTD_CONTENTSIZE_UNKNOWN when not known.
+ * `mode` is the mode for parameter adjustment. See docs for `ZSTD_CParamMode_e`.
+ *  note : `srcSize==0` means 0!
+ *  condition : cPar is presumed validated (can be checked using ZSTD_checkCParams()). */
+static ZSTD_compressionParameters
+ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
+                            unsigned long long srcSize,
+                            size_t dictSize,
+                            ZSTD_CParamMode_e mode,
+                            ZSTD_ParamSwitch_e useRowMatchFinder)
+{
+    const U64 minSrcSize = 513; /* (1<<9) + 1 */
+    const U64 maxWindowResize = 1ULL << (ZSTD_WINDOWLOG_MAX-1);
+    assert(ZSTD_checkCParams(cPar)==0);
+
+    /* Cascade the selected strategy down to the next-highest one built into
+     * this binary. */
+#ifdef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_btultra2) {
+        cPar.strategy = ZSTD_btultra;
+    }
+    if (cPar.strategy == ZSTD_btultra) {
+        cPar.strategy = ZSTD_btopt;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_btopt) {
+        cPar.strategy = ZSTD_btlazy2;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_btlazy2) {
+        cPar.strategy = ZSTD_lazy2;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_lazy2) {
+        cPar.strategy = ZSTD_lazy;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_lazy) {
+        cPar.strategy = ZSTD_greedy;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_greedy) {
+        cPar.strategy = ZSTD_dfast;
+    }
+#endif
+#ifdef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
+    if (cPar.strategy == ZSTD_dfast) {
+        cPar.strategy = ZSTD_fast;
+        cPar.targetLength = 0;
+    }
+#endif
+
+    switch (mode) {
+    case ZSTD_cpm_unknown:
+    case ZSTD_cpm_noAttachDict:
+        /* If we don't know the source size, don't make any
+         * assumptions about it. We will already have selected
+         * smaller parameters if a dictionary is in use.
+         */
+        break;
+    case ZSTD_cpm_createCDict:
+        /* Assume a small source size when creating a dictionary
+         * with an unknown source size.
+         */
+        if (dictSize && srcSize == ZSTD_CONTENTSIZE_UNKNOWN)
+            srcSize = minSrcSize;
+        break;
+    case ZSTD_cpm_attachDict:
+        /* Dictionary has its own dedicated parameters which have
+         * already been selected. We are selecting parameters
+         * for only the source.
+         */
+        dictSize = 0;
+        break;
+    default:
+        assert(0);
+        break;
+    }
+
+    /* resize windowLog if input is small enough, to use less memory */
+    if ( (srcSize <= maxWindowResize)
+      && (dictSize <= maxWindowResize) )  {
+        U32 const tSize = (U32)(srcSize + dictSize);
+        static U32 const hashSizeMin = 1 << ZSTD_HASHLOG_MIN;
+        U32 const srcLog = (tSize < hashSizeMin) ? ZSTD_HASHLOG_MIN :
+                            ZSTD_highbit32(tSize-1) + 1;
+        if (cPar.windowLog > srcLog) cPar.windowLog = srcLog;
+    }
+    if (srcSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+        U32 const dictAndWindowLog = ZSTD_dictAndWindowLog(cPar.windowLog, (U64)srcSize, (U64)dictSize);
+        U32 const cycleLog = ZSTD_cycleLog(cPar.chainLog, cPar.strategy);
+        if (cPar.hashLog > dictAndWindowLog+1) cPar.hashLog = dictAndWindowLog+1;
+        if (cycleLog > dictAndWindowLog)
+            cPar.chainLog -= (cycleLog - dictAndWindowLog);
+    }
+
+    if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN)
+        cPar.windowLog = ZSTD_WINDOWLOG_ABSOLUTEMIN;  /* minimum wlog required for valid frame header */
+
+    /* We can't use more than 32 bits of hash in total, so that means that we require:
+     * (hashLog + 8) <= 32 && (chainLog + 8) <= 32
+     */
+    if (mode == ZSTD_cpm_createCDict && ZSTD_CDictIndicesAreTagged(&cPar)) {
+        U32 const maxShortCacheHashLog = 32 - ZSTD_SHORT_CACHE_TAG_BITS;
+        if (cPar.hashLog > maxShortCacheHashLog) {
+            cPar.hashLog = maxShortCacheHashLog;
+        }
+        if (cPar.chainLog > maxShortCacheHashLog) {
+            cPar.chainLog = maxShortCacheHashLog;
+        }
+    }
+
+
+    /* At this point, we aren't 100% sure if we are using the row match finder.
+     * Unless it is explicitly disabled, conservatively assume that it is enabled.
+     * In this case it will only be disabled for small sources, so shrinking the
+     * hash log a little bit shouldn't result in any ratio loss.
+     */
+    if (useRowMatchFinder == ZSTD_ps_auto)
+        useRowMatchFinder = ZSTD_ps_enable;
+
+    /* We can't hash more than 32-bits in total. So that means that we require:
+     * (hashLog - rowLog + 8) <= 32
+     */
+    if (ZSTD_rowMatchFinderUsed(cPar.strategy, useRowMatchFinder)) {
+        /* Switch to 32-entry rows if searchLog is 5 (or more) */
+        U32 const rowLog = BOUNDED(4, cPar.searchLog, 6);
+        U32 const maxRowHashLog = 32 - ZSTD_ROW_HASH_TAG_BITS;
+        U32 const maxHashLog = maxRowHashLog + rowLog;
+        assert(cPar.hashLog >= rowLog);
+        if (cPar.hashLog > maxHashLog) {
+            cPar.hashLog = maxHashLog;
+        }
+    }
+
+    return cPar;
+}
+
+ZSTD_compressionParameters
+ZSTD_adjustCParams(ZSTD_compressionParameters cPar,
+                   unsigned long long srcSize,
+                   size_t dictSize)
+{
+    cPar = ZSTD_clampCParams(cPar);   /* resulting cPar is necessarily valid (all parameters within range) */
+    if (srcSize == 0) srcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+    return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize, ZSTD_cpm_unknown, ZSTD_ps_auto);
+}
+
+static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_CParamMode_e mode);
+static ZSTD_parameters ZSTD_getParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_CParamMode_e mode);
+
+static void ZSTD_overrideCParams(
+              ZSTD_compressionParameters* cParams,
+        const ZSTD_compressionParameters* overrides)
+{
+    if (overrides->windowLog)    cParams->windowLog    = overrides->windowLog;
+    if (overrides->hashLog)      cParams->hashLog      = overrides->hashLog;
+    if (overrides->chainLog)     cParams->chainLog     = overrides->chainLog;
+    if (overrides->searchLog)    cParams->searchLog    = overrides->searchLog;
+    if (overrides->minMatch)     cParams->minMatch     = overrides->minMatch;
+    if (overrides->targetLength) cParams->targetLength = overrides->targetLength;
+    if (overrides->strategy)     cParams->strategy     = overrides->strategy;
+}
+
+ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize, ZSTD_CParamMode_e mode)
+{
+    ZSTD_compressionParameters cParams;
+    if (srcSizeHint == ZSTD_CONTENTSIZE_UNKNOWN && CCtxParams->srcSizeHint > 0) {
+        assert(CCtxParams->srcSizeHint>=0);
+        srcSizeHint = (U64)CCtxParams->srcSizeHint;
+    }
+    cParams = ZSTD_getCParams_internal(CCtxParams->compressionLevel, srcSizeHint, dictSize, mode);
+    if (CCtxParams->ldmParams.enableLdm == ZSTD_ps_enable) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
+    ZSTD_overrideCParams(&cParams, &CCtxParams->cParams);
+    assert(!ZSTD_checkCParams(cParams));
+    /* srcSizeHint == 0 means 0 */
+    return ZSTD_adjustCParams_internal(cParams, srcSizeHint, dictSize, mode, CCtxParams->useRowMatchFinder);
+}
+
+static size_t
+ZSTD_sizeof_matchState(const ZSTD_compressionParameters* const cParams,
+                       const ZSTD_ParamSwitch_e useRowMatchFinder,
+                       const int enableDedicatedDictSearch,
+                       const U32 forCCtx)
+{
+    /* chain table size should be 0 for fast or row-hash strategies */
+    size_t const chainSize = ZSTD_allocateChainTable(cParams->strategy, useRowMatchFinder, enableDedicatedDictSearch && !forCCtx)
+                                ? ((size_t)1 << cParams->chainLog)
+                                : 0;
+    size_t const hSize = ((size_t)1) << cParams->hashLog;
+    U32    const hashLog3 = (forCCtx && cParams->minMatch==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
+    size_t const h3Size = hashLog3 ? ((size_t)1) << hashLog3 : 0;
+    /* We don't use ZSTD_cwksp_alloc_size() here because the tables aren't
+     * surrounded by redzones in ASAN. */
+    size_t const tableSpace = chainSize * sizeof(U32)
+                            + hSize * sizeof(U32)
+                            + h3Size * sizeof(U32);
+    size_t const optPotentialSpace =
+        ZSTD_cwksp_aligned64_alloc_size((MaxML+1) * sizeof(U32))
+      + ZSTD_cwksp_aligned64_alloc_size((MaxLL+1) * sizeof(U32))
+      + ZSTD_cwksp_aligned64_alloc_size((MaxOff+1) * sizeof(U32))
+      + ZSTD_cwksp_aligned64_alloc_size((1<<Litbits) * sizeof(U32))
+      + ZSTD_cwksp_aligned64_alloc_size(ZSTD_OPT_SIZE * sizeof(ZSTD_match_t))
+      + ZSTD_cwksp_aligned64_alloc_size(ZSTD_OPT_SIZE * sizeof(ZSTD_optimal_t));
+    size_t const lazyAdditionalSpace = ZSTD_rowMatchFinderUsed(cParams->strategy, useRowMatchFinder)
+                                            ? ZSTD_cwksp_aligned64_alloc_size(hSize)
+                                            : 0;
+    size_t const optSpace = (forCCtx && (cParams->strategy >= ZSTD_btopt))
+                                ? optPotentialSpace
+                                : 0;
+    size_t const slackSpace = ZSTD_cwksp_slack_space_required();
+
+    /* tables are guaranteed to be sized in multiples of 64 bytes (or 16 uint32_t) */
+    ZSTD_STATIC_ASSERT(ZSTD_HASHLOG_MIN >= 4 && ZSTD_WINDOWLOG_MIN >= 4 && ZSTD_CHAINLOG_MIN >= 4);
+    assert(useRowMatchFinder != ZSTD_ps_auto);
+
+    DEBUGLOG(4, "chainSize: %u - hSize: %u - h3Size: %u",
+                (U32)chainSize, (U32)hSize, (U32)h3Size);
+    return tableSpace + optSpace + slackSpace + lazyAdditionalSpace;
+}
+
+/* Helper function for calculating memory requirements.
+ * Gives a tighter bound than ZSTD_sequenceBound() by taking minMatch into account. */
+static size_t ZSTD_maxNbSeq(size_t blockSize, unsigned minMatch, int useSequenceProducer) {
+    U32 const divider = (minMatch==3 || useSequenceProducer) ? 3 : 4;
+    return blockSize / divider;
+}
+
+static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+        const ZSTD_compressionParameters* cParams,
+        const ldmParams_t* ldmParams,
+        const int isStatic,
+        const ZSTD_ParamSwitch_e useRowMatchFinder,
+        const size_t buffInSize,
+        const size_t buffOutSize,
+        const U64 pledgedSrcSize,
+        int useSequenceProducer,
+        size_t maxBlockSize)
+{
+    size_t const windowSize = (size_t) BOUNDED(1ULL, 1ULL << cParams->windowLog, pledgedSrcSize);
+    size_t const blockSize = MIN(ZSTD_resolveMaxBlockSize(maxBlockSize), windowSize);
+    size_t const maxNbSeq = ZSTD_maxNbSeq(blockSize, cParams->minMatch, useSequenceProducer);
+    size_t const tokenSpace = ZSTD_cwksp_alloc_size(WILDCOPY_OVERLENGTH + blockSize)
+                            + ZSTD_cwksp_aligned64_alloc_size(maxNbSeq * sizeof(SeqDef))
+                            + 3 * ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(BYTE));
+    size_t const tmpWorkSpace = ZSTD_cwksp_alloc_size(TMP_WORKSPACE_SIZE);
+    size_t const blockStateSpace = 2 * ZSTD_cwksp_alloc_size(sizeof(ZSTD_compressedBlockState_t));
+    size_t const matchStateSize = ZSTD_sizeof_matchState(cParams, useRowMatchFinder, /* enableDedicatedDictSearch */ 0, /* forCCtx */ 1);
+
+    size_t const ldmSpace = ZSTD_ldm_getTableSize(*ldmParams);
+    size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(*ldmParams, blockSize);
+    size_t const ldmSeqSpace = ldmParams->enableLdm == ZSTD_ps_enable ?
+        ZSTD_cwksp_aligned64_alloc_size(maxNbLdmSeq * sizeof(rawSeq)) : 0;
+
+
+    size_t const bufferSpace = ZSTD_cwksp_alloc_size(buffInSize)
+                             + ZSTD_cwksp_alloc_size(buffOutSize);
+
+    size_t const cctxSpace = isStatic ? ZSTD_cwksp_alloc_size(sizeof(ZSTD_CCtx)) : 0;
+
+    size_t const maxNbExternalSeq = ZSTD_sequenceBound(blockSize);
+    size_t const externalSeqSpace = useSequenceProducer
+        ? ZSTD_cwksp_aligned64_alloc_size(maxNbExternalSeq * sizeof(ZSTD_Sequence))
+        : 0;
+
+    size_t const neededSpace =
+        cctxSpace +
+        tmpWorkSpace +
+        blockStateSpace +
+        ldmSpace +
+        ldmSeqSpace +
+        matchStateSize +
+        tokenSpace +
+        bufferSpace +
+        externalSeqSpace;
+
+    DEBUGLOG(5, "estimate workspace : %u", (U32)neededSpace);
+    return neededSpace;
+}
+
+size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
+{
+    ZSTD_compressionParameters const cParams =
+                ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
+    ZSTD_ParamSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder,
+                                                                               &cParams);
+
+    RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
+    /* estimateCCtxSize is for one-shot compression. So no buffers should
+     * be needed. However, we still allocate two 0-sized buffers, which can
+     * take space under ASAN. */
+    return ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+        &cParams, &params->ldmParams, 1, useRowMatchFinder, 0, 0, ZSTD_CONTENTSIZE_UNKNOWN, ZSTD_hasExtSeqProd(params), params->maxBlockSize);
+}
+
+size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params initialParams = ZSTD_makeCCtxParamsFromCParams(cParams);
+    if (ZSTD_rowMatchFinderSupported(cParams.strategy)) {
+        /* Pick bigger of not using and using row-based matchfinder for greedy and lazy strategies */
+        size_t noRowCCtxSize;
+        size_t rowCCtxSize;
+        initialParams.useRowMatchFinder = ZSTD_ps_disable;
+        noRowCCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(&initialParams);
+        initialParams.useRowMatchFinder = ZSTD_ps_enable;
+        rowCCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(&initialParams);
+        return MAX(noRowCCtxSize, rowCCtxSize);
+    } else {
+        return ZSTD_estimateCCtxSize_usingCCtxParams(&initialParams);
+    }
+}
+
+static size_t ZSTD_estimateCCtxSize_internal(int compressionLevel)
+{
+    int tier = 0;
+    size_t largestSize = 0;
+    static const unsigned long long srcSizeTiers[4] = {16 KB, 128 KB, 256 KB, ZSTD_CONTENTSIZE_UNKNOWN};
+    for (; tier < 4; ++tier) {
+        /* Choose the set of cParams for a given level across all srcSizes that give the largest cctxSize */
+        ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, srcSizeTiers[tier], 0, ZSTD_cpm_noAttachDict);
+        largestSize = MAX(ZSTD_estimateCCtxSize_usingCParams(cParams), largestSize);
+    }
+    return largestSize;
+}
+
+size_t ZSTD_estimateCCtxSize(int compressionLevel)
+{
+    int level;
+    size_t memBudget = 0;
+    for (level=MIN(compressionLevel, 1); level<=compressionLevel; level++) {
+        /* Ensure monotonically increasing memory usage as compression level increases */
+        size_t const newMB = ZSTD_estimateCCtxSize_internal(level);
+        if (newMB > memBudget) memBudget = newMB;
+    }
+    return memBudget;
+}
+
+size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params)
+{
+    RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
+    {   ZSTD_compressionParameters const cParams =
+                ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
+        size_t const blockSize = MIN(ZSTD_resolveMaxBlockSize(params->maxBlockSize), (size_t)1 << cParams.windowLog);
+        size_t const inBuffSize = (params->inBufferMode == ZSTD_bm_buffered)
+                ? ((size_t)1 << cParams.windowLog) + blockSize
+                : 0;
+        size_t const outBuffSize = (params->outBufferMode == ZSTD_bm_buffered)
+                ? ZSTD_compressBound(blockSize) + 1
+                : 0;
+        ZSTD_ParamSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder, &params->cParams);
+
+        return ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+            &cParams, &params->ldmParams, 1, useRowMatchFinder, inBuffSize, outBuffSize,
+            ZSTD_CONTENTSIZE_UNKNOWN, ZSTD_hasExtSeqProd(params), params->maxBlockSize);
+    }
+}
+
+size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params initialParams = ZSTD_makeCCtxParamsFromCParams(cParams);
+    if (ZSTD_rowMatchFinderSupported(cParams.strategy)) {
+        /* Pick bigger of not using and using row-based matchfinder for greedy and lazy strategies */
+        size_t noRowCCtxSize;
+        size_t rowCCtxSize;
+        initialParams.useRowMatchFinder = ZSTD_ps_disable;
+        noRowCCtxSize = ZSTD_estimateCStreamSize_usingCCtxParams(&initialParams);
+        initialParams.useRowMatchFinder = ZSTD_ps_enable;
+        rowCCtxSize = ZSTD_estimateCStreamSize_usingCCtxParams(&initialParams);
+        return MAX(noRowCCtxSize, rowCCtxSize);
+    } else {
+        return ZSTD_estimateCStreamSize_usingCCtxParams(&initialParams);
+    }
+}
+
+static size_t ZSTD_estimateCStreamSize_internal(int compressionLevel)
+{
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
+    return ZSTD_estimateCStreamSize_usingCParams(cParams);
+}
+
+size_t ZSTD_estimateCStreamSize(int compressionLevel)
+{
+    int level;
+    size_t memBudget = 0;
+    for (level=MIN(compressionLevel, 1); level<=compressionLevel; level++) {
+        size_t const newMB = ZSTD_estimateCStreamSize_internal(level);
+        if (newMB > memBudget) memBudget = newMB;
+    }
+    return memBudget;
+}
+
+/* ZSTD_getFrameProgression():
+ * tells how much data has been consumed (input) and produced (output) for current frame.
+ * able to count progression inside worker threads (non-blocking mode).
+ */
+ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        return ZSTDMT_getFrameProgression(cctx->mtctx);
+    }
+#endif
+    {   ZSTD_frameProgression fp;
+        size_t const buffered = (cctx->inBuff == NULL) ? 0 :
+                                cctx->inBuffPos - cctx->inToCompress;
+        if (buffered) assert(cctx->inBuffPos >= cctx->inToCompress);
+        assert(buffered <= ZSTD_BLOCKSIZE_MAX);
+        fp.ingested = cctx->consumedSrcSize + buffered;
+        fp.consumed = cctx->consumedSrcSize;
+        fp.produced = cctx->producedCSize;
+        fp.flushed  = cctx->producedCSize;   /* simplified; some data might still be left within streaming output buffer */
+        fp.currentJobID = 0;
+        fp.nbActiveWorkers = 0;
+        return fp;
+}   }
+
+/*! ZSTD_toFlushNow()
+ *  Only useful for multithreading scenarios currently (nbWorkers >= 1).
+ */
+size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        return ZSTDMT_toFlushNow(cctx->mtctx);
+    }
+#endif
+    (void)cctx;
+    return 0;   /* over-simplification; could also check if context is currently running in streaming mode, and in which case, report how many bytes are left to be flushed within output buffer */
+}
+
+static void ZSTD_assertEqualCParams(ZSTD_compressionParameters cParams1,
+                                    ZSTD_compressionParameters cParams2)
+{
+    (void)cParams1;
+    (void)cParams2;
+    assert(cParams1.windowLog    == cParams2.windowLog);
+    assert(cParams1.chainLog     == cParams2.chainLog);
+    assert(cParams1.hashLog      == cParams2.hashLog);
+    assert(cParams1.searchLog    == cParams2.searchLog);
+    assert(cParams1.minMatch     == cParams2.minMatch);
+    assert(cParams1.targetLength == cParams2.targetLength);
+    assert(cParams1.strategy     == cParams2.strategy);
+}
+
+void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
+{
+    int i;
+    for (i = 0; i < ZSTD_REP_NUM; ++i)
+        bs->rep[i] = repStartValue[i];
+    bs->entropy.huf.repeatMode = HUF_repeat_none;
+    bs->entropy.fse.offcode_repeatMode = FSE_repeat_none;
+    bs->entropy.fse.matchlength_repeatMode = FSE_repeat_none;
+    bs->entropy.fse.litlength_repeatMode = FSE_repeat_none;
+}
+
+/*! ZSTD_invalidateMatchState()
+ *  Invalidate all the matches in the match finder tables.
+ *  Requires nextSrc and base to be set (can be NULL).
+ */
+static void ZSTD_invalidateMatchState(ZSTD_MatchState_t* ms)
+{
+    ZSTD_window_clear(&ms->window);
+
+    ms->nextToUpdate = ms->window.dictLimit;
+    ms->loadedDictEnd = 0;
+    ms->opt.litLengthSum = 0;  /* force reset of btopt stats */
+    ms->dictMatchState = NULL;
+}
+
+/**
+ * Controls, for this matchState reset, whether the tables need to be cleared /
+ * prepared for the coming compression (ZSTDcrp_makeClean), or whether the
+ * tables can be left unclean (ZSTDcrp_leaveDirty), because we know that a
+ * subsequent operation will overwrite the table space anyways (e.g., copying
+ * the matchState contents in from a CDict).
+ */
+typedef enum {
+    ZSTDcrp_makeClean,
+    ZSTDcrp_leaveDirty
+} ZSTD_compResetPolicy_e;
+
+/**
+ * Controls, for this matchState reset, whether indexing can continue where it
+ * left off (ZSTDirp_continue), or whether it needs to be restarted from zero
+ * (ZSTDirp_reset).
+ */
+typedef enum {
+    ZSTDirp_continue,
+    ZSTDirp_reset
+} ZSTD_indexResetPolicy_e;
+
+typedef enum {
+    ZSTD_resetTarget_CDict,
+    ZSTD_resetTarget_CCtx
+} ZSTD_resetTarget_e;
+
+/* Mixes bits in a 64 bits in a value, based on XXH3_rrmxmx */
+static U64 ZSTD_bitmix(U64 val, U64 len) {
+    val ^= ZSTD_rotateRight_U64(val, 49) ^ ZSTD_rotateRight_U64(val, 24);
+    val *= 0x9FB21C651E98DF25ULL;
+    val ^= (val >> 35) + len ;
+    val *= 0x9FB21C651E98DF25ULL;
+    return val ^ (val >> 28);
+}
+
+/* Mixes in the hashSalt and hashSaltEntropy to create a new hashSalt */
+static void ZSTD_advanceHashSalt(ZSTD_MatchState_t* ms) {
+    ms->hashSalt = ZSTD_bitmix(ms->hashSalt, 8) ^ ZSTD_bitmix((U64) ms->hashSaltEntropy, 4);
+}
+
+static size_t
+ZSTD_reset_matchState(ZSTD_MatchState_t* ms,
+                      ZSTD_cwksp* ws,
+                const ZSTD_compressionParameters* cParams,
+                const ZSTD_ParamSwitch_e useRowMatchFinder,
+                const ZSTD_compResetPolicy_e crp,
+                const ZSTD_indexResetPolicy_e forceResetIndex,
+                const ZSTD_resetTarget_e forWho)
+{
+    /* disable chain table allocation for fast or row-based strategies */
+    size_t const chainSize = ZSTD_allocateChainTable(cParams->strategy, useRowMatchFinder,
+                                                     ms->dedicatedDictSearch && (forWho == ZSTD_resetTarget_CDict))
+                                ? ((size_t)1 << cParams->chainLog)
+                                : 0;
+    size_t const hSize = ((size_t)1) << cParams->hashLog;
+    U32    const hashLog3 = ((forWho == ZSTD_resetTarget_CCtx) && cParams->minMatch==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
+    size_t const h3Size = hashLog3 ? ((size_t)1) << hashLog3 : 0;
+
+    DEBUGLOG(4, "reset indices : %u", forceResetIndex == ZSTDirp_reset);
+    assert(useRowMatchFinder != ZSTD_ps_auto);
+    if (forceResetIndex == ZSTDirp_reset) {
+        ZSTD_window_init(&ms->window);
+        ZSTD_cwksp_mark_tables_dirty(ws);
+    }
+
+    ms->hashLog3 = hashLog3;
+    ms->lazySkipping = 0;
+
+    ZSTD_invalidateMatchState(ms);
+
+    assert(!ZSTD_cwksp_reserve_failed(ws)); /* check that allocation hasn't already failed */
+
+    ZSTD_cwksp_clear_tables(ws);
+
+    DEBUGLOG(5, "reserving table space");
+    /* table Space */
+    ms->hashTable = (U32*)ZSTD_cwksp_reserve_table(ws, hSize * sizeof(U32));
+    ms->chainTable = (U32*)ZSTD_cwksp_reserve_table(ws, chainSize * sizeof(U32));
+    ms->hashTable3 = (U32*)ZSTD_cwksp_reserve_table(ws, h3Size * sizeof(U32));
+    RETURN_ERROR_IF(ZSTD_cwksp_reserve_failed(ws), memory_allocation,
+                    "failed a workspace allocation in ZSTD_reset_matchState");
+
+    DEBUGLOG(4, "reset table : %u", crp!=ZSTDcrp_leaveDirty);
+    if (crp!=ZSTDcrp_leaveDirty) {
+        /* reset tables only */
+        ZSTD_cwksp_clean_tables(ws);
+    }
+
+    if (ZSTD_rowMatchFinderUsed(cParams->strategy, useRowMatchFinder)) {
+        /* Row match finder needs an additional table of hashes ("tags") */
+        size_t const tagTableSize = hSize;
+        /* We want to generate a new salt in case we reset a Cctx, but we always want to use
+         * 0 when we reset a Cdict */
+        if(forWho == ZSTD_resetTarget_CCtx) {
+            ms->tagTable = (BYTE*) ZSTD_cwksp_reserve_aligned_init_once(ws, tagTableSize);
+            ZSTD_advanceHashSalt(ms);
+        } else {
+            /* When we are not salting we want to always memset the memory */
+            ms->tagTable = (BYTE*) ZSTD_cwksp_reserve_aligned64(ws, tagTableSize);
+            ZSTD_memset(ms->tagTable, 0, tagTableSize);
+            ms->hashSalt = 0;
+        }
+        {   /* Switch to 32-entry rows if searchLog is 5 (or more) */
+            U32 const rowLog = BOUNDED(4, cParams->searchLog, 6);
+            assert(cParams->hashLog >= rowLog);
+            ms->rowHashLog = cParams->hashLog - rowLog;
+        }
+    }
+
+    /* opt parser space */
+    if ((forWho == ZSTD_resetTarget_CCtx) && (cParams->strategy >= ZSTD_btopt)) {
+        DEBUGLOG(4, "reserving optimal parser space");
+        ms->opt.litFreq = (unsigned*)ZSTD_cwksp_reserve_aligned64(ws, (1<<Litbits) * sizeof(unsigned));
+        ms->opt.litLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned64(ws, (MaxLL+1) * sizeof(unsigned));
+        ms->opt.matchLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned64(ws, (MaxML+1) * sizeof(unsigned));
+        ms->opt.offCodeFreq = (unsigned*)ZSTD_cwksp_reserve_aligned64(ws, (MaxOff+1) * sizeof(unsigned));
+        ms->opt.matchTable = (ZSTD_match_t*)ZSTD_cwksp_reserve_aligned64(ws, ZSTD_OPT_SIZE * sizeof(ZSTD_match_t));
+        ms->opt.priceTable = (ZSTD_optimal_t*)ZSTD_cwksp_reserve_aligned64(ws, ZSTD_OPT_SIZE * sizeof(ZSTD_optimal_t));
+    }
+
+    ms->cParams = *cParams;
+
+    RETURN_ERROR_IF(ZSTD_cwksp_reserve_failed(ws), memory_allocation,
+                    "failed a workspace allocation in ZSTD_reset_matchState");
+    return 0;
+}
+
+/* ZSTD_indexTooCloseToMax() :
+ * minor optimization : prefer memset() rather than reduceIndex()
+ * which is measurably slow in some circumstances (reported for Visual Studio).
+ * Works when re-using a context for a lot of smallish inputs :
+ * if all inputs are smaller than ZSTD_INDEXOVERFLOW_MARGIN,
+ * memset() will be triggered before reduceIndex().
+ */
+#define ZSTD_INDEXOVERFLOW_MARGIN (16 MB)
+static int ZSTD_indexTooCloseToMax(ZSTD_window_t w)
+{
+    return (size_t)(w.nextSrc - w.base) > (ZSTD_CURRENT_MAX - ZSTD_INDEXOVERFLOW_MARGIN);
+}
+
+/** ZSTD_dictTooBig():
+ * When dictionaries are larger than ZSTD_CHUNKSIZE_MAX they can't be loaded in
+ * one go generically. So we ensure that in that case we reset the tables to zero,
+ * so that we can load as much of the dictionary as possible.
+ */
+static int ZSTD_dictTooBig(size_t const loadedDictSize)
+{
+    return loadedDictSize > ZSTD_CHUNKSIZE_MAX;
+}
+
+/*! ZSTD_resetCCtx_internal() :
+ * @param loadedDictSize The size of the dictionary to be loaded
+ * into the context, if any. If no dictionary is used, or the
+ * dictionary is being attached / copied, then pass 0.
+ * note : `params` are assumed fully validated at this stage.
+ */
+static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
+                                      ZSTD_CCtx_params const* params,
+                                      U64 const pledgedSrcSize,
+                                      size_t const loadedDictSize,
+                                      ZSTD_compResetPolicy_e const crp,
+                                      ZSTD_buffered_policy_e const zbuff)
+{
+    ZSTD_cwksp* const ws = &zc->workspace;
+    DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u, useRowMatchFinder=%d useBlockSplitter=%d",
+                (U32)pledgedSrcSize, params->cParams.windowLog, (int)params->useRowMatchFinder, (int)params->postBlockSplitter);
+    assert(!ZSTD_isError(ZSTD_checkCParams(params->cParams)));
+
+    zc->isFirstBlock = 1;
+
+    /* Set applied params early so we can modify them for LDM,
+     * and point params at the applied params.
+     */
+    zc->appliedParams = *params;
+    params = &zc->appliedParams;
+
+    assert(params->useRowMatchFinder != ZSTD_ps_auto);
+    assert(params->postBlockSplitter != ZSTD_ps_auto);
+    assert(params->ldmParams.enableLdm != ZSTD_ps_auto);
+    assert(params->maxBlockSize != 0);
+    if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
+        /* Adjust long distance matching parameters */
+        ZSTD_ldm_adjustParameters(&zc->appliedParams.ldmParams, &params->cParams);
+        assert(params->ldmParams.hashLog >= params->ldmParams.bucketSizeLog);
+        assert(params->ldmParams.hashRateLog < 32);
+    }
+
+    {   size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params->cParams.windowLog), pledgedSrcSize));
+        size_t const blockSize = MIN(params->maxBlockSize, windowSize);
+        size_t const maxNbSeq = ZSTD_maxNbSeq(blockSize, params->cParams.minMatch, ZSTD_hasExtSeqProd(params));
+        size_t const buffOutSize = (zbuff == ZSTDb_buffered && params->outBufferMode == ZSTD_bm_buffered)
+                ? ZSTD_compressBound(blockSize) + 1
+                : 0;
+        size_t const buffInSize = (zbuff == ZSTDb_buffered && params->inBufferMode == ZSTD_bm_buffered)
+                ? windowSize + blockSize
+                : 0;
+        size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(params->ldmParams, blockSize);
+
+        int const indexTooClose = ZSTD_indexTooCloseToMax(zc->blockState.matchState.window);
+        int const dictTooBig = ZSTD_dictTooBig(loadedDictSize);
+        ZSTD_indexResetPolicy_e needsIndexReset =
+            (indexTooClose || dictTooBig || !zc->initialized) ? ZSTDirp_reset : ZSTDirp_continue;
+
+        size_t const neededSpace =
+            ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+                &params->cParams, &params->ldmParams, zc->staticSize != 0, params->useRowMatchFinder,
+                buffInSize, buffOutSize, pledgedSrcSize, ZSTD_hasExtSeqProd(params), params->maxBlockSize);
+
+        FORWARD_IF_ERROR(neededSpace, "cctx size estimate failed!");
+
+        if (!zc->staticSize) ZSTD_cwksp_bump_oversized_duration(ws, 0);
+
+        {   /* Check if workspace is large enough, alloc a new one if needed */
+            int const workspaceTooSmall = ZSTD_cwksp_sizeof(ws) < neededSpace;
+            int const workspaceWasteful = ZSTD_cwksp_check_wasteful(ws, neededSpace);
+            int resizeWorkspace = workspaceTooSmall || workspaceWasteful;
+            DEBUGLOG(4, "Need %zu B workspace", neededSpace);
+            DEBUGLOG(4, "windowSize: %zu - blockSize: %zu", windowSize, blockSize);
+
+            if (resizeWorkspace) {
+                DEBUGLOG(4, "Resize workspaceSize from %zuKB to %zuKB",
+                            ZSTD_cwksp_sizeof(ws) >> 10,
+                            neededSpace >> 10);
+
+                RETURN_ERROR_IF(zc->staticSize, memory_allocation, "static cctx : no resize");
+
+                needsIndexReset = ZSTDirp_reset;
+
+                ZSTD_cwksp_free(ws, zc->customMem);
+                FORWARD_IF_ERROR(ZSTD_cwksp_create(ws, neededSpace, zc->customMem), "");
+
+                DEBUGLOG(5, "reserving object space");
+                /* Statically sized space.
+                 * tmpWorkspace never moves,
+                 * though prev/next block swap places */
+                assert(ZSTD_cwksp_check_available(ws, 2 * sizeof(ZSTD_compressedBlockState_t)));
+                zc->blockState.prevCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t));
+                RETURN_ERROR_IF(zc->blockState.prevCBlock == NULL, memory_allocation, "couldn't allocate prevCBlock");
+                zc->blockState.nextCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t));
+                RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate nextCBlock");
+                zc->tmpWorkspace = ZSTD_cwksp_reserve_object(ws, TMP_WORKSPACE_SIZE);
+                RETURN_ERROR_IF(zc->tmpWorkspace == NULL, memory_allocation, "couldn't allocate tmpWorkspace");
+                zc->tmpWkspSize = TMP_WORKSPACE_SIZE;
+        }   }
+
+        ZSTD_cwksp_clear(ws);
+
+        /* init params */
+        zc->blockState.matchState.cParams = params->cParams;
+        zc->blockState.matchState.prefetchCDictTables = params->prefetchCDictTables == ZSTD_ps_enable;
+        zc->pledgedSrcSizePlusOne = pledgedSrcSize+1;
+        zc->consumedSrcSize = 0;
+        zc->producedCSize = 0;
+        if (pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN)
+            zc->appliedParams.fParams.contentSizeFlag = 0;
+        DEBUGLOG(4, "pledged content size : %u ; flag : %u",
+            (unsigned)pledgedSrcSize, zc->appliedParams.fParams.contentSizeFlag);
+        zc->blockSizeMax = blockSize;
+
+        XXH64_reset(&zc->xxhState, 0);
+        zc->stage = ZSTDcs_init;
+        zc->dictID = 0;
+        zc->dictContentSize = 0;
+
+        ZSTD_reset_compressedBlockState(zc->blockState.prevCBlock);
+
+        FORWARD_IF_ERROR(ZSTD_reset_matchState(
+                &zc->blockState.matchState,
+                ws,
+                &params->cParams,
+                params->useRowMatchFinder,
+                crp,
+                needsIndexReset,
+                ZSTD_resetTarget_CCtx), "");
+
+        zc->seqStore.sequencesStart = (SeqDef*)ZSTD_cwksp_reserve_aligned64(ws, maxNbSeq * sizeof(SeqDef));
+
+        /* ldm hash table */
+        if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
+            /* TODO: avoid memset? */
+            size_t const ldmHSize = ((size_t)1) << params->ldmParams.hashLog;
+            zc->ldmState.hashTable = (ldmEntry_t*)ZSTD_cwksp_reserve_aligned64(ws, ldmHSize * sizeof(ldmEntry_t));
+            ZSTD_memset(zc->ldmState.hashTable, 0, ldmHSize * sizeof(ldmEntry_t));
+            zc->ldmSequences = (rawSeq*)ZSTD_cwksp_reserve_aligned64(ws, maxNbLdmSeq * sizeof(rawSeq));
+            zc->maxNbLdmSequences = maxNbLdmSeq;
+
+            ZSTD_window_init(&zc->ldmState.window);
+            zc->ldmState.loadedDictEnd = 0;
+        }
+
+        /* reserve space for block-level external sequences */
+        if (ZSTD_hasExtSeqProd(params)) {
+            size_t const maxNbExternalSeq = ZSTD_sequenceBound(blockSize);
+            zc->extSeqBufCapacity = maxNbExternalSeq;
+            zc->extSeqBuf =
+                (ZSTD_Sequence*)ZSTD_cwksp_reserve_aligned64(ws, maxNbExternalSeq * sizeof(ZSTD_Sequence));
+        }
+
+        /* buffers */
+
+        /* ZSTD_wildcopy() is used to copy into the literals buffer,
+         * so we have to oversize the buffer by WILDCOPY_OVERLENGTH bytes.
+         */
+        zc->seqStore.litStart = ZSTD_cwksp_reserve_buffer(ws, blockSize + WILDCOPY_OVERLENGTH);
+        zc->seqStore.maxNbLit = blockSize;
+
+        zc->bufferedPolicy = zbuff;
+        zc->inBuffSize = buffInSize;
+        zc->inBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffInSize);
+        zc->outBuffSize = buffOutSize;
+        zc->outBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffOutSize);
+
+        /* ldm bucketOffsets table */
+        if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
+            /* TODO: avoid memset? */
+            size_t const numBuckets =
+                  ((size_t)1) << (params->ldmParams.hashLog -
+                                  params->ldmParams.bucketSizeLog);
+            zc->ldmState.bucketOffsets = ZSTD_cwksp_reserve_buffer(ws, numBuckets);
+            ZSTD_memset(zc->ldmState.bucketOffsets, 0, numBuckets);
+        }
+
+        /* sequences storage */
+        ZSTD_referenceExternalSequences(zc, NULL, 0);
+        zc->seqStore.maxNbSeq = maxNbSeq;
+        zc->seqStore.llCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
+        zc->seqStore.mlCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
+        zc->seqStore.ofCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
+
+        DEBUGLOG(3, "wksp: finished allocating, %zd bytes remain available", ZSTD_cwksp_available_space(ws));
+        assert(ZSTD_cwksp_estimated_space_within_bounds(ws, neededSpace));
+
+        zc->initialized = 1;
+
+        return 0;
+    }
+}
+
+/* ZSTD_invalidateRepCodes() :
+ * ensures next compression will not use repcodes from previous block.
+ * Note : only works with regular variant;
+ *        do not use with extDict variant ! */
+void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx) {
+    int i;
+    for (i=0; i<ZSTD_REP_NUM; i++) cctx->blockState.prevCBlock->rep[i] = 0;
+    assert(!ZSTD_window_hasExtDict(cctx->blockState.matchState.window));
+}
+
+/* These are the approximate sizes for each strategy past which copying the
+ * dictionary tables into the working context is faster than using them
+ * in-place.
+ */
+static const size_t attachDictSizeCutoffs[ZSTD_STRATEGY_MAX+1] = {
+    8 KB,  /* unused */
+    8 KB,  /* ZSTD_fast */
+    16 KB, /* ZSTD_dfast */
+    32 KB, /* ZSTD_greedy */
+    32 KB, /* ZSTD_lazy */
+    32 KB, /* ZSTD_lazy2 */
+    32 KB, /* ZSTD_btlazy2 */
+    32 KB, /* ZSTD_btopt */
+    8 KB,  /* ZSTD_btultra */
+    8 KB   /* ZSTD_btultra2 */
+};
+
+static int ZSTD_shouldAttachDict(const ZSTD_CDict* cdict,
+                                 const ZSTD_CCtx_params* params,
+                                 U64 pledgedSrcSize)
+{
+    size_t cutoff = attachDictSizeCutoffs[cdict->matchState.cParams.strategy];
+    int const dedicatedDictSearch = cdict->matchState.dedicatedDictSearch;
+    return dedicatedDictSearch
+        || ( ( pledgedSrcSize <= cutoff
+            || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
+            || params->attachDictPref == ZSTD_dictForceAttach )
+          && params->attachDictPref != ZSTD_dictForceCopy
+          && !params->forceWindow ); /* dictMatchState isn't correctly
+                                      * handled in _enforceMaxDist */
+}
+
+static size_t
+ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx,
+                        const ZSTD_CDict* cdict,
+                        ZSTD_CCtx_params params,
+                        U64 pledgedSrcSize,
+                        ZSTD_buffered_policy_e zbuff)
+{
+    DEBUGLOG(4, "ZSTD_resetCCtx_byAttachingCDict() pledgedSrcSize=%llu",
+                (unsigned long long)pledgedSrcSize);
+    {
+        ZSTD_compressionParameters adjusted_cdict_cParams = cdict->matchState.cParams;
+        unsigned const windowLog = params.cParams.windowLog;
+        assert(windowLog != 0);
+        /* Resize working context table params for input only, since the dict
+         * has its own tables. */
+        /* pledgedSrcSize == 0 means 0! */
+
+        if (cdict->matchState.dedicatedDictSearch) {
+            ZSTD_dedicatedDictSearch_revertCParams(&adjusted_cdict_cParams);
+        }
+
+        params.cParams = ZSTD_adjustCParams_internal(adjusted_cdict_cParams, pledgedSrcSize,
+                                                     cdict->dictContentSize, ZSTD_cpm_attachDict,
+                                                     params.useRowMatchFinder);
+        params.cParams.windowLog = windowLog;
+        params.useRowMatchFinder = cdict->useRowMatchFinder;    /* cdict overrides */
+        FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, &params, pledgedSrcSize,
+                                                 /* loadedDictSize */ 0,
+                                                 ZSTDcrp_makeClean, zbuff), "");
+        assert(cctx->appliedParams.cParams.strategy == adjusted_cdict_cParams.strategy);
+    }
+
+    {   const U32 cdictEnd = (U32)( cdict->matchState.window.nextSrc
+                                  - cdict->matchState.window.base);
+        const U32 cdictLen = cdictEnd - cdict->matchState.window.dictLimit;
+        if (cdictLen == 0) {
+            /* don't even attach dictionaries with no contents */
+            DEBUGLOG(4, "skipping attaching empty dictionary");
+        } else {
+            DEBUGLOG(4, "attaching dictionary into context");
+            cctx->blockState.matchState.dictMatchState = &cdict->matchState;
+
+            /* prep working match state so dict matches never have negative indices
+             * when they are translated to the working context's index space. */
+            if (cctx->blockState.matchState.window.dictLimit < cdictEnd) {
+                cctx->blockState.matchState.window.nextSrc =
+                    cctx->blockState.matchState.window.base + cdictEnd;
+                ZSTD_window_clear(&cctx->blockState.matchState.window);
+            }
+            /* loadedDictEnd is expressed within the referential of the active context */
+            cctx->blockState.matchState.loadedDictEnd = cctx->blockState.matchState.window.dictLimit;
+    }   }
+
+    cctx->dictID = cdict->dictID;
+    cctx->dictContentSize = cdict->dictContentSize;
+
+    /* copy block state */
+    ZSTD_memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
+
+    return 0;
+}
+
+static void ZSTD_copyCDictTableIntoCCtx(U32* dst, U32 const* src, size_t tableSize,
+                                        ZSTD_compressionParameters const* cParams) {
+    if (ZSTD_CDictIndicesAreTagged(cParams)){
+        /* Remove tags from the CDict table if they are present.
+         * See docs on "short cache" in zstd_compress_internal.h for context. */
+        size_t i;
+        for (i = 0; i < tableSize; i++) {
+            U32 const taggedIndex = src[i];
+            U32 const index = taggedIndex >> ZSTD_SHORT_CACHE_TAG_BITS;
+            dst[i] = index;
+        }
+    } else {
+        ZSTD_memcpy(dst, src, tableSize * sizeof(U32));
+    }
+}
+
+static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
+                            const ZSTD_CDict* cdict,
+                            ZSTD_CCtx_params params,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
+{
+    const ZSTD_compressionParameters *cdict_cParams = &cdict->matchState.cParams;
+
+    assert(!cdict->matchState.dedicatedDictSearch);
+    DEBUGLOG(4, "ZSTD_resetCCtx_byCopyingCDict() pledgedSrcSize=%llu",
+                (unsigned long long)pledgedSrcSize);
+
+    {   unsigned const windowLog = params.cParams.windowLog;
+        assert(windowLog != 0);
+        /* Copy only compression parameters related to tables. */
+        params.cParams = *cdict_cParams;
+        params.cParams.windowLog = windowLog;
+        params.useRowMatchFinder = cdict->useRowMatchFinder;
+        FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, &params, pledgedSrcSize,
+                                                 /* loadedDictSize */ 0,
+                                                 ZSTDcrp_leaveDirty, zbuff), "");
+        assert(cctx->appliedParams.cParams.strategy == cdict_cParams->strategy);
+        assert(cctx->appliedParams.cParams.hashLog == cdict_cParams->hashLog);
+        assert(cctx->appliedParams.cParams.chainLog == cdict_cParams->chainLog);
+    }
+
+    ZSTD_cwksp_mark_tables_dirty(&cctx->workspace);
+    assert(params.useRowMatchFinder != ZSTD_ps_auto);
+
+    /* copy tables */
+    {   size_t const chainSize = ZSTD_allocateChainTable(cdict_cParams->strategy, cdict->useRowMatchFinder, 0 /* DDS guaranteed disabled */)
+                                                            ? ((size_t)1 << cdict_cParams->chainLog)
+                                                            : 0;
+        size_t const hSize =  (size_t)1 << cdict_cParams->hashLog;
+
+        ZSTD_copyCDictTableIntoCCtx(cctx->blockState.matchState.hashTable,
+                                cdict->matchState.hashTable,
+                                hSize, cdict_cParams);
+
+        /* Do not copy cdict's chainTable if cctx has parameters such that it would not use chainTable */
+        if (ZSTD_allocateChainTable(cctx->appliedParams.cParams.strategy, cctx->appliedParams.useRowMatchFinder, 0 /* forDDSDict */)) {
+            ZSTD_copyCDictTableIntoCCtx(cctx->blockState.matchState.chainTable,
+                                    cdict->matchState.chainTable,
+                                    chainSize, cdict_cParams);
+        }
+        /* copy tag table */
+        if (ZSTD_rowMatchFinderUsed(cdict_cParams->strategy, cdict->useRowMatchFinder)) {
+            size_t const tagTableSize = hSize;
+            ZSTD_memcpy(cctx->blockState.matchState.tagTable,
+                        cdict->matchState.tagTable,
+                        tagTableSize);
+            cctx->blockState.matchState.hashSalt = cdict->matchState.hashSalt;
+        }
+    }
+
+    /* Zero the hashTable3, since the cdict never fills it */
+    assert(cctx->blockState.matchState.hashLog3 <= 31);
+    {   U32 const h3log = cctx->blockState.matchState.hashLog3;
+        size_t const h3Size = h3log ? ((size_t)1 << h3log) : 0;
+        assert(cdict->matchState.hashLog3 == 0);
+        ZSTD_memset(cctx->blockState.matchState.hashTable3, 0, h3Size * sizeof(U32));
+    }
+
+    ZSTD_cwksp_mark_tables_clean(&cctx->workspace);
+
+    /* copy dictionary offsets */
+    {   ZSTD_MatchState_t const* srcMatchState = &cdict->matchState;
+        ZSTD_MatchState_t* dstMatchState = &cctx->blockState.matchState;
+        dstMatchState->window       = srcMatchState->window;
+        dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
+        dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
+    }
+
+    cctx->dictID = cdict->dictID;
+    cctx->dictContentSize = cdict->dictContentSize;
+
+    /* copy block state */
+    ZSTD_memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
+
+    return 0;
+}
+
+/* We have a choice between copying the dictionary context into the working
+ * context, or referencing the dictionary context from the working context
+ * in-place. We decide here which strategy to use. */
+static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
+                            const ZSTD_CDict* cdict,
+                            const ZSTD_CCtx_params* params,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
+{
+
+    DEBUGLOG(4, "ZSTD_resetCCtx_usingCDict (pledgedSrcSize=%u)",
+                (unsigned)pledgedSrcSize);
+
+    if (ZSTD_shouldAttachDict(cdict, params, pledgedSrcSize)) {
+        return ZSTD_resetCCtx_byAttachingCDict(
+            cctx, cdict, *params, pledgedSrcSize, zbuff);
+    } else {
+        return ZSTD_resetCCtx_byCopyingCDict(
+            cctx, cdict, *params, pledgedSrcSize, zbuff);
+    }
+}
+
+/*! ZSTD_copyCCtx_internal() :
+ *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
+ *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
+ *  The "context", in this case, refers to the hash and chain tables,
+ *  entropy tables, and dictionary references.
+ * `windowLog` value is enforced if != 0, otherwise value is copied from srcCCtx.
+ * @return : 0, or an error code */
+static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
+                            const ZSTD_CCtx* srcCCtx,
+                            ZSTD_frameParameters fParams,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
+{
+    RETURN_ERROR_IF(srcCCtx->stage!=ZSTDcs_init, stage_wrong,
+                    "Can't copy a ctx that's not in init stage.");
+    DEBUGLOG(5, "ZSTD_copyCCtx_internal");
+    ZSTD_memcpy(&dstCCtx->customMem, &srcCCtx->customMem, sizeof(ZSTD_customMem));
+    {   ZSTD_CCtx_params params = dstCCtx->requestedParams;
+        /* Copy only compression parameters related to tables. */
+        params.cParams = srcCCtx->appliedParams.cParams;
+        assert(srcCCtx->appliedParams.useRowMatchFinder != ZSTD_ps_auto);
+        assert(srcCCtx->appliedParams.postBlockSplitter != ZSTD_ps_auto);
+        assert(srcCCtx->appliedParams.ldmParams.enableLdm != ZSTD_ps_auto);
+        params.useRowMatchFinder = srcCCtx->appliedParams.useRowMatchFinder;
+        params.postBlockSplitter = srcCCtx->appliedParams.postBlockSplitter;
+        params.ldmParams = srcCCtx->appliedParams.ldmParams;
+        params.fParams = fParams;
+        params.maxBlockSize = srcCCtx->appliedParams.maxBlockSize;
+        ZSTD_resetCCtx_internal(dstCCtx, &params, pledgedSrcSize,
+                                /* loadedDictSize */ 0,
+                                ZSTDcrp_leaveDirty, zbuff);
+        assert(dstCCtx->appliedParams.cParams.windowLog == srcCCtx->appliedParams.cParams.windowLog);
+        assert(dstCCtx->appliedParams.cParams.strategy == srcCCtx->appliedParams.cParams.strategy);
+        assert(dstCCtx->appliedParams.cParams.hashLog == srcCCtx->appliedParams.cParams.hashLog);
+        assert(dstCCtx->appliedParams.cParams.chainLog == srcCCtx->appliedParams.cParams.chainLog);
+        assert(dstCCtx->blockState.matchState.hashLog3 == srcCCtx->blockState.matchState.hashLog3);
+    }
+
+    ZSTD_cwksp_mark_tables_dirty(&dstCCtx->workspace);
+
+    /* copy tables */
+    {   size_t const chainSize = ZSTD_allocateChainTable(srcCCtx->appliedParams.cParams.strategy,
+                                                         srcCCtx->appliedParams.useRowMatchFinder,
+                                                         0 /* forDDSDict */)
+                                    ? ((size_t)1 << srcCCtx->appliedParams.cParams.chainLog)
+                                    : 0;
+        size_t const hSize =  (size_t)1 << srcCCtx->appliedParams.cParams.hashLog;
+        U32 const h3log = srcCCtx->blockState.matchState.hashLog3;
+        size_t const h3Size = h3log ? ((size_t)1 << h3log) : 0;
+
+        ZSTD_memcpy(dstCCtx->blockState.matchState.hashTable,
+               srcCCtx->blockState.matchState.hashTable,
+               hSize * sizeof(U32));
+        ZSTD_memcpy(dstCCtx->blockState.matchState.chainTable,
+               srcCCtx->blockState.matchState.chainTable,
+               chainSize * sizeof(U32));
+        ZSTD_memcpy(dstCCtx->blockState.matchState.hashTable3,
+               srcCCtx->blockState.matchState.hashTable3,
+               h3Size * sizeof(U32));
+    }
+
+    ZSTD_cwksp_mark_tables_clean(&dstCCtx->workspace);
+
+    /* copy dictionary offsets */
+    {
+        const ZSTD_MatchState_t* srcMatchState = &srcCCtx->blockState.matchState;
+        ZSTD_MatchState_t* dstMatchState = &dstCCtx->blockState.matchState;
+        dstMatchState->window       = srcMatchState->window;
+        dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
+        dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
+    }
+    dstCCtx->dictID = srcCCtx->dictID;
+    dstCCtx->dictContentSize = srcCCtx->dictContentSize;
+
+    /* copy block state */
+    ZSTD_memcpy(dstCCtx->blockState.prevCBlock, srcCCtx->blockState.prevCBlock, sizeof(*srcCCtx->blockState.prevCBlock));
+
+    return 0;
+}
+
+/*! ZSTD_copyCCtx() :
+ *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
+ *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
+ *  pledgedSrcSize==0 means "unknown".
+*   @return : 0, or an error code */
+size_t ZSTD_copyCCtx(ZSTD_CCtx* dstCCtx, const ZSTD_CCtx* srcCCtx, unsigned long long pledgedSrcSize)
+{
+    ZSTD_frameParameters fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    ZSTD_buffered_policy_e const zbuff = srcCCtx->bufferedPolicy;
+    ZSTD_STATIC_ASSERT((U32)ZSTDb_buffered==1);
+    if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+    fParams.contentSizeFlag = (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN);
+
+    return ZSTD_copyCCtx_internal(dstCCtx, srcCCtx,
+                                fParams, pledgedSrcSize,
+                                zbuff);
+}
+
+
+#define ZSTD_ROWSIZE 16
+/*! ZSTD_reduceTable() :
+ *  reduce table indexes by `reducerValue`, or squash to zero.
+ *  PreserveMark preserves "unsorted mark" for btlazy2 strategy.
+ *  It must be set to a clear 0/1 value, to remove branch during inlining.
+ *  Presume table size is a multiple of ZSTD_ROWSIZE
+ *  to help auto-vectorization */
+FORCE_INLINE_TEMPLATE void
+ZSTD_reduceTable_internal (U32* const table, U32 const size, U32 const reducerValue, int const preserveMark)
+{
+    int const nbRows = (int)size / ZSTD_ROWSIZE;
+    int cellNb = 0;
+    int rowNb;
+    /* Protect special index values < ZSTD_WINDOW_START_INDEX. */
+    U32 const reducerThreshold = reducerValue + ZSTD_WINDOW_START_INDEX;
+    assert((size & (ZSTD_ROWSIZE-1)) == 0);  /* multiple of ZSTD_ROWSIZE */
+    assert(size < (1U<<31));   /* can be cast to int */
+
+#if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
+    /* To validate that the table reuse logic is sound, and that we don't
+     * access table space that we haven't cleaned, we re-"poison" the table
+     * space every time we mark it dirty.
+     *
+     * This function however is intended to operate on those dirty tables and
+     * re-clean them. So when this function is used correctly, we can unpoison
+     * the memory it operated on. This introduces a blind spot though, since
+     * if we now try to operate on __actually__ poisoned memory, we will not
+     * detect that. */
+    __msan_unpoison(table, size * sizeof(U32));
+#endif
+
+    for (rowNb=0 ; rowNb < nbRows ; rowNb++) {
+        int column;
+        for (column=0; column<ZSTD_ROWSIZE; column++) {
+            U32 newVal;
+            if (preserveMark && table[cellNb] == ZSTD_DUBT_UNSORTED_MARK) {
+                /* This write is pointless, but is required(?) for the compiler
+                 * to auto-vectorize the loop. */
+                newVal = ZSTD_DUBT_UNSORTED_MARK;
+            } else if (table[cellNb] < reducerThreshold) {
+                newVal = 0;
+            } else {
+                newVal = table[cellNb] - reducerValue;
+            }
+            table[cellNb] = newVal;
+            cellNb++;
+    }   }
+}
+
+static void ZSTD_reduceTable(U32* const table, U32 const size, U32 const reducerValue)
+{
+    ZSTD_reduceTable_internal(table, size, reducerValue, 0);
+}
+
+static void ZSTD_reduceTable_btlazy2(U32* const table, U32 const size, U32 const reducerValue)
+{
+    ZSTD_reduceTable_internal(table, size, reducerValue, 1);
+}
+
+/*! ZSTD_reduceIndex() :
+*   rescale all indexes to avoid future overflow (indexes are U32) */
+static void ZSTD_reduceIndex (ZSTD_MatchState_t* ms, ZSTD_CCtx_params const* params, const U32 reducerValue)
+{
+    {   U32 const hSize = (U32)1 << params->cParams.hashLog;
+        ZSTD_reduceTable(ms->hashTable, hSize, reducerValue);
+    }
+
+    if (ZSTD_allocateChainTable(params->cParams.strategy, params->useRowMatchFinder, (U32)ms->dedicatedDictSearch)) {
+        U32 const chainSize = (U32)1 << params->cParams.chainLog;
+        if (params->cParams.strategy == ZSTD_btlazy2)
+            ZSTD_reduceTable_btlazy2(ms->chainTable, chainSize, reducerValue);
+        else
+            ZSTD_reduceTable(ms->chainTable, chainSize, reducerValue);
+    }
+
+    if (ms->hashLog3) {
+        U32 const h3Size = (U32)1 << ms->hashLog3;
+        ZSTD_reduceTable(ms->hashTable3, h3Size, reducerValue);
+    }
+}
+
+
+/*-*******************************************************
+*  Block entropic compression
+*********************************************************/
+
+/* See doc/zstd_compression_format.md for detailed format description */
+
+int ZSTD_seqToCodes(const SeqStore_t* seqStorePtr)
+{
+    const SeqDef* const sequences = seqStorePtr->sequencesStart;
+    BYTE* const llCodeTable = seqStorePtr->llCode;
+    BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    U32 u;
+    int longOffsets = 0;
+    assert(nbSeq <= seqStorePtr->maxNbSeq);
+    for (u=0; u<nbSeq; u++) {
+        U32 const llv = sequences[u].litLength;
+        U32 const ofCode = ZSTD_highbit32(sequences[u].offBase);
+        U32 const mlv = sequences[u].mlBase;
+        llCodeTable[u] = (BYTE)ZSTD_LLcode(llv);
+        ofCodeTable[u] = (BYTE)ofCode;
+        mlCodeTable[u] = (BYTE)ZSTD_MLcode(mlv);
+        assert(!(MEM_64bits() && ofCode >= STREAM_ACCUMULATOR_MIN));
+        if (MEM_32bits() && ofCode >= STREAM_ACCUMULATOR_MIN)
+            longOffsets = 1;
+    }
+    if (seqStorePtr->longLengthType==ZSTD_llt_literalLength)
+        llCodeTable[seqStorePtr->longLengthPos] = MaxLL;
+    if (seqStorePtr->longLengthType==ZSTD_llt_matchLength)
+        mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
+    return longOffsets;
+}
+
+/* ZSTD_useTargetCBlockSize():
+ * Returns if target compressed block size param is being used.
+ * If used, compression will do best effort to make a compressed block size to be around targetCBlockSize.
+ * Returns 1 if true, 0 otherwise. */
+static int ZSTD_useTargetCBlockSize(const ZSTD_CCtx_params* cctxParams)
+{
+    DEBUGLOG(5, "ZSTD_useTargetCBlockSize (targetCBlockSize=%zu)", cctxParams->targetCBlockSize);
+    return (cctxParams->targetCBlockSize != 0);
+}
+
+/* ZSTD_blockSplitterEnabled():
+ * Returns if block splitting param is being used
+ * If used, compression will do best effort to split a block in order to improve compression ratio.
+ * At the time this function is called, the parameter must be finalized.
+ * Returns 1 if true, 0 otherwise. */
+static int ZSTD_blockSplitterEnabled(ZSTD_CCtx_params* cctxParams)
+{
+    DEBUGLOG(5, "ZSTD_blockSplitterEnabled (postBlockSplitter=%d)", cctxParams->postBlockSplitter);
+    assert(cctxParams->postBlockSplitter != ZSTD_ps_auto);
+    return (cctxParams->postBlockSplitter == ZSTD_ps_enable);
+}
+
+/* Type returned by ZSTD_buildSequencesStatistics containing finalized symbol encoding types
+ * and size of the sequences statistics
+ */
+typedef struct {
+    U32 LLtype;
+    U32 Offtype;
+    U32 MLtype;
+    size_t size;
+    size_t lastCountSize; /* Accounts for bug in 1.3.4. More detail in ZSTD_entropyCompressSeqStore_internal() */
+    int longOffsets;
+} ZSTD_symbolEncodingTypeStats_t;
+
+/* ZSTD_buildSequencesStatistics():
+ * Returns a ZSTD_symbolEncodingTypeStats_t, or a zstd error code in the `size` field.
+ * Modifies `nextEntropy` to have the appropriate values as a side effect.
+ * nbSeq must be greater than 0.
+ *
+ * entropyWkspSize must be of size at least ENTROPY_WORKSPACE_SIZE - (MaxSeq + 1)*sizeof(U32)
+ */
+static ZSTD_symbolEncodingTypeStats_t
+ZSTD_buildSequencesStatistics(
+                const SeqStore_t* seqStorePtr, size_t nbSeq,
+                const ZSTD_fseCTables_t* prevEntropy, ZSTD_fseCTables_t* nextEntropy,
+                      BYTE* dst, const BYTE* const dstEnd,
+                      ZSTD_strategy strategy, unsigned* countWorkspace,
+                      void* entropyWorkspace, size_t entropyWkspSize)
+{
+    BYTE* const ostart = dst;
+    const BYTE* const oend = dstEnd;
+    BYTE* op = ostart;
+    FSE_CTable* CTable_LitLength = nextEntropy->litlengthCTable;
+    FSE_CTable* CTable_OffsetBits = nextEntropy->offcodeCTable;
+    FSE_CTable* CTable_MatchLength = nextEntropy->matchlengthCTable;
+    const BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    const BYTE* const llCodeTable = seqStorePtr->llCode;
+    const BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    ZSTD_symbolEncodingTypeStats_t stats;
+
+    stats.lastCountSize = 0;
+    /* convert length/distances into codes */
+    stats.longOffsets = ZSTD_seqToCodes(seqStorePtr);
+    assert(op <= oend);
+    assert(nbSeq != 0); /* ZSTD_selectEncodingType() divides by nbSeq */
+    /* build CTable for Literal Lengths */
+    {   unsigned max = MaxLL;
+        size_t const mostFrequent = HIST_countFast_wksp(countWorkspace, &max, llCodeTable, nbSeq, entropyWorkspace, entropyWkspSize);   /* can't fail */
+        DEBUGLOG(5, "Building LL table");
+        nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
+        stats.LLtype = ZSTD_selectEncodingType(&nextEntropy->litlength_repeatMode,
+                                        countWorkspace, max, mostFrequent, nbSeq,
+                                        LLFSELog, prevEntropy->litlengthCTable,
+                                        LL_defaultNorm, LL_defaultNormLog,
+                                        ZSTD_defaultAllowed, strategy);
+        assert(set_basic < set_compressed && set_rle < set_compressed);
+        assert(!(stats.LLtype < set_compressed && nextEntropy->litlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(
+                op, (size_t)(oend - op),
+                CTable_LitLength, LLFSELog, (SymbolEncodingType_e)stats.LLtype,
+                countWorkspace, max, llCodeTable, nbSeq,
+                LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                prevEntropy->litlengthCTable,
+                sizeof(prevEntropy->litlengthCTable),
+                entropyWorkspace, entropyWkspSize);
+            if (ZSTD_isError(countSize)) {
+                DEBUGLOG(3, "ZSTD_buildCTable for LitLens failed");
+                stats.size = countSize;
+                return stats;
+            }
+            if (stats.LLtype == set_compressed)
+                stats.lastCountSize = countSize;
+            op += countSize;
+            assert(op <= oend);
+    }   }
+    /* build CTable for Offsets */
+    {   unsigned max = MaxOff;
+        size_t const mostFrequent = HIST_countFast_wksp(
+            countWorkspace, &max, ofCodeTable, nbSeq, entropyWorkspace, entropyWkspSize);  /* can't fail */
+        /* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */
+        ZSTD_DefaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
+        DEBUGLOG(5, "Building OF table");
+        nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
+        stats.Offtype = ZSTD_selectEncodingType(&nextEntropy->offcode_repeatMode,
+                                        countWorkspace, max, mostFrequent, nbSeq,
+                                        OffFSELog, prevEntropy->offcodeCTable,
+                                        OF_defaultNorm, OF_defaultNormLog,
+                                        defaultPolicy, strategy);
+        assert(!(stats.Offtype < set_compressed && nextEntropy->offcode_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(
+                op, (size_t)(oend - op),
+                CTable_OffsetBits, OffFSELog, (SymbolEncodingType_e)stats.Offtype,
+                countWorkspace, max, ofCodeTable, nbSeq,
+                OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                prevEntropy->offcodeCTable,
+                sizeof(prevEntropy->offcodeCTable),
+                entropyWorkspace, entropyWkspSize);
+            if (ZSTD_isError(countSize)) {
+                DEBUGLOG(3, "ZSTD_buildCTable for Offsets failed");
+                stats.size = countSize;
+                return stats;
+            }
+            if (stats.Offtype == set_compressed)
+                stats.lastCountSize = countSize;
+            op += countSize;
+            assert(op <= oend);
+    }   }
+    /* build CTable for MatchLengths */
+    {   unsigned max = MaxML;
+        size_t const mostFrequent = HIST_countFast_wksp(
+            countWorkspace, &max, mlCodeTable, nbSeq, entropyWorkspace, entropyWkspSize);   /* can't fail */
+        DEBUGLOG(5, "Building ML table (remaining space : %i)", (int)(oend-op));
+        nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
+        stats.MLtype = ZSTD_selectEncodingType(&nextEntropy->matchlength_repeatMode,
+                                        countWorkspace, max, mostFrequent, nbSeq,
+                                        MLFSELog, prevEntropy->matchlengthCTable,
+                                        ML_defaultNorm, ML_defaultNormLog,
+                                        ZSTD_defaultAllowed, strategy);
+        assert(!(stats.MLtype < set_compressed && nextEntropy->matchlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(
+                op, (size_t)(oend - op),
+                CTable_MatchLength, MLFSELog, (SymbolEncodingType_e)stats.MLtype,
+                countWorkspace, max, mlCodeTable, nbSeq,
+                ML_defaultNorm, ML_defaultNormLog, MaxML,
+                prevEntropy->matchlengthCTable,
+                sizeof(prevEntropy->matchlengthCTable),
+                entropyWorkspace, entropyWkspSize);
+            if (ZSTD_isError(countSize)) {
+                DEBUGLOG(3, "ZSTD_buildCTable for MatchLengths failed");
+                stats.size = countSize;
+                return stats;
+            }
+            if (stats.MLtype == set_compressed)
+                stats.lastCountSize = countSize;
+            op += countSize;
+            assert(op <= oend);
+    }   }
+    stats.size = (size_t)(op-ostart);
+    return stats;
+}
+
+/* ZSTD_entropyCompressSeqStore_internal():
+ * compresses both literals and sequences
+ * Returns compressed size of block, or a zstd error.
+ */
+#define SUSPECT_UNCOMPRESSIBLE_LITERAL_RATIO 20
+MEM_STATIC size_t
+ZSTD_entropyCompressSeqStore_internal(
+                              void* dst, size_t dstCapacity,
+                        const void* literals, size_t litSize,
+                        const SeqStore_t* seqStorePtr,
+                        const ZSTD_entropyCTables_t* prevEntropy,
+                              ZSTD_entropyCTables_t* nextEntropy,
+                        const ZSTD_CCtx_params* cctxParams,
+                              void* entropyWorkspace, size_t entropyWkspSize,
+                        const int bmi2)
+{
+    ZSTD_strategy const strategy = cctxParams->cParams.strategy;
+    unsigned* count = (unsigned*)entropyWorkspace;
+    FSE_CTable* CTable_LitLength = nextEntropy->fse.litlengthCTable;
+    FSE_CTable* CTable_OffsetBits = nextEntropy->fse.offcodeCTable;
+    FSE_CTable* CTable_MatchLength = nextEntropy->fse.matchlengthCTable;
+    const SeqDef* const sequences = seqStorePtr->sequencesStart;
+    const size_t nbSeq = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    const BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    const BYTE* const llCodeTable = seqStorePtr->llCode;
+    const BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    size_t lastCountSize;
+    int longOffsets = 0;
+
+    entropyWorkspace = count + (MaxSeq + 1);
+    entropyWkspSize -= (MaxSeq + 1) * sizeof(*count);
+
+    DEBUGLOG(5, "ZSTD_entropyCompressSeqStore_internal (nbSeq=%zu, dstCapacity=%zu)", nbSeq, dstCapacity);
+    ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
+    assert(entropyWkspSize >= HUF_WORKSPACE_SIZE);
+
+    /* Compress literals */
+    {   size_t const numSequences = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+        /* Base suspicion of uncompressibility on ratio of literals to sequences */
+        int const suspectUncompressible = (numSequences == 0) || (litSize / numSequences >= SUSPECT_UNCOMPRESSIBLE_LITERAL_RATIO);
+
+        size_t const cSize = ZSTD_compressLiterals(
+                                    op, dstCapacity,
+                                    literals, litSize,
+                                    entropyWorkspace, entropyWkspSize,
+                                    &prevEntropy->huf, &nextEntropy->huf,
+                                    cctxParams->cParams.strategy,
+                                    ZSTD_literalsCompressionIsDisabled(cctxParams),
+                                    suspectUncompressible, bmi2);
+        FORWARD_IF_ERROR(cSize, "ZSTD_compressLiterals failed");
+        assert(cSize <= dstCapacity);
+        op += cSize;
+    }
+
+    /* Sequences Header */
+    RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
+                    dstSize_tooSmall, "Can't fit seq hdr in output buf!");
+    if (nbSeq < 128) {
+        *op++ = (BYTE)nbSeq;
+    } else if (nbSeq < LONGNBSEQ) {
+        op[0] = (BYTE)((nbSeq>>8) + 0x80);
+        op[1] = (BYTE)nbSeq;
+        op+=2;
+    } else {
+        op[0]=0xFF;
+        MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ));
+        op+=3;
+    }
+    assert(op <= oend);
+    if (nbSeq==0) {
+        /* Copy the old tables over as if we repeated them */
+        ZSTD_memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
+        return (size_t)(op - ostart);
+    }
+    {   BYTE* const seqHead = op++;
+        /* build stats for sequences */
+        const ZSTD_symbolEncodingTypeStats_t stats =
+                ZSTD_buildSequencesStatistics(seqStorePtr, nbSeq,
+                                             &prevEntropy->fse, &nextEntropy->fse,
+                                              op, oend,
+                                              strategy, count,
+                                              entropyWorkspace, entropyWkspSize);
+        FORWARD_IF_ERROR(stats.size, "ZSTD_buildSequencesStatistics failed!");
+        *seqHead = (BYTE)((stats.LLtype<<6) + (stats.Offtype<<4) + (stats.MLtype<<2));
+        lastCountSize = stats.lastCountSize;
+        op += stats.size;
+        longOffsets = stats.longOffsets;
+    }
+
+    {   size_t const bitstreamSize = ZSTD_encodeSequences(
+                                        op, (size_t)(oend - op),
+                                        CTable_MatchLength, mlCodeTable,
+                                        CTable_OffsetBits, ofCodeTable,
+                                        CTable_LitLength, llCodeTable,
+                                        sequences, nbSeq,
+                                        longOffsets, bmi2);
+        FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
+        op += bitstreamSize;
+        assert(op <= oend);
+        /* zstd versions <= 1.3.4 mistakenly report corruption when
+         * FSE_readNCount() receives a buffer < 4 bytes.
+         * Fixed by https://github.com/facebook/zstd/pull/1146.
+         * This can happen when the last set_compressed table present is 2
+         * bytes and the bitstream is only one byte.
+         * In this exceedingly rare case, we will simply emit an uncompressed
+         * block, since it isn't worth optimizing.
+         */
+        if (lastCountSize && (lastCountSize + bitstreamSize) < 4) {
+            /* lastCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
+            assert(lastCountSize + bitstreamSize == 3);
+            DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
+                        "emitting an uncompressed block.");
+            return 0;
+        }
+    }
+
+    DEBUGLOG(5, "compressed block size : %u", (unsigned)(op - ostart));
+    return (size_t)(op - ostart);
+}
+
+static size_t
+ZSTD_entropyCompressSeqStore_wExtLitBuffer(
+                          void* dst, size_t dstCapacity,
+                    const void* literals, size_t litSize,
+                          size_t blockSize,
+                    const SeqStore_t* seqStorePtr,
+                    const ZSTD_entropyCTables_t* prevEntropy,
+                          ZSTD_entropyCTables_t* nextEntropy,
+                    const ZSTD_CCtx_params* cctxParams,
+                          void* entropyWorkspace, size_t entropyWkspSize,
+                          int bmi2)
+{
+    size_t const cSize = ZSTD_entropyCompressSeqStore_internal(
+                            dst, dstCapacity,
+                            literals, litSize,
+                            seqStorePtr, prevEntropy, nextEntropy, cctxParams,
+                            entropyWorkspace, entropyWkspSize, bmi2);
+    if (cSize == 0) return 0;
+    /* When srcSize <= dstCapacity, there is enough space to write a raw uncompressed block.
+     * Since we ran out of space, block must be not compressible, so fall back to raw uncompressed block.
+     */
+    if ((cSize == ERROR(dstSize_tooSmall)) & (blockSize <= dstCapacity)) {
+        DEBUGLOG(4, "not enough dstCapacity (%zu) for ZSTD_entropyCompressSeqStore_internal()=> do not compress block", dstCapacity);
+        return 0;  /* block not compressed */
+    }
+    FORWARD_IF_ERROR(cSize, "ZSTD_entropyCompressSeqStore_internal failed");
+
+    /* Check compressibility */
+    {   size_t const maxCSize = blockSize - ZSTD_minGain(blockSize, cctxParams->cParams.strategy);
+        if (cSize >= maxCSize) return 0;  /* block not compressed */
+    }
+    DEBUGLOG(5, "ZSTD_entropyCompressSeqStore() cSize: %zu", cSize);
+    /* libzstd decoder before  > v1.5.4 is not compatible with compressed blocks of size ZSTD_BLOCKSIZE_MAX exactly.
+     * This restriction is indirectly already fulfilled by respecting ZSTD_minGain() condition above.
+     */
+    assert(cSize < ZSTD_BLOCKSIZE_MAX);
+    return cSize;
+}
+
+static size_t
+ZSTD_entropyCompressSeqStore(
+                    const SeqStore_t* seqStorePtr,
+                    const ZSTD_entropyCTables_t* prevEntropy,
+                          ZSTD_entropyCTables_t* nextEntropy,
+                    const ZSTD_CCtx_params* cctxParams,
+                          void* dst, size_t dstCapacity,
+                          size_t srcSize,
+                          void* entropyWorkspace, size_t entropyWkspSize,
+                          int bmi2)
+{
+    return ZSTD_entropyCompressSeqStore_wExtLitBuffer(
+                dst, dstCapacity,
+                seqStorePtr->litStart, (size_t)(seqStorePtr->lit - seqStorePtr->litStart),
+                srcSize,
+                seqStorePtr,
+                prevEntropy, nextEntropy,
+                cctxParams,
+                entropyWorkspace, entropyWkspSize,
+                bmi2);
+}
+
+/* ZSTD_selectBlockCompressor() :
+ * Not static, but internal use only (used by long distance matcher)
+ * assumption : strat is a valid strategy */
+ZSTD_BlockCompressor_f ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_ParamSwitch_e useRowMatchFinder, ZSTD_dictMode_e dictMode)
+{
+    static const ZSTD_BlockCompressor_f blockCompressor[4][ZSTD_STRATEGY_MAX+1] = {
+        { ZSTD_compressBlock_fast  /* default for 0 */,
+          ZSTD_compressBlock_fast,
+          ZSTD_COMPRESSBLOCK_DOUBLEFAST,
+          ZSTD_COMPRESSBLOCK_GREEDY,
+          ZSTD_COMPRESSBLOCK_LAZY,
+          ZSTD_COMPRESSBLOCK_LAZY2,
+          ZSTD_COMPRESSBLOCK_BTLAZY2,
+          ZSTD_COMPRESSBLOCK_BTOPT,
+          ZSTD_COMPRESSBLOCK_BTULTRA,
+          ZSTD_COMPRESSBLOCK_BTULTRA2
+        },
+        { ZSTD_compressBlock_fast_extDict  /* default for 0 */,
+          ZSTD_compressBlock_fast_extDict,
+          ZSTD_COMPRESSBLOCK_DOUBLEFAST_EXTDICT,
+          ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT,
+          ZSTD_COMPRESSBLOCK_LAZY_EXTDICT,
+          ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT,
+          ZSTD_COMPRESSBLOCK_BTLAZY2_EXTDICT,
+          ZSTD_COMPRESSBLOCK_BTOPT_EXTDICT,
+          ZSTD_COMPRESSBLOCK_BTULTRA_EXTDICT,
+          ZSTD_COMPRESSBLOCK_BTULTRA_EXTDICT
+        },
+        { ZSTD_compressBlock_fast_dictMatchState  /* default for 0 */,
+          ZSTD_compressBlock_fast_dictMatchState,
+          ZSTD_COMPRESSBLOCK_DOUBLEFAST_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_BTLAZY2_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_BTOPT_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_BTULTRA_DICTMATCHSTATE,
+          ZSTD_COMPRESSBLOCK_BTULTRA_DICTMATCHSTATE
+        },
+        { NULL  /* default for 0 */,
+          NULL,
+          NULL,
+          ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH,
+          ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH,
+          ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH,
+          NULL,
+          NULL,
+          NULL,
+          NULL }
+    };
+    ZSTD_BlockCompressor_f selectedCompressor;
+    ZSTD_STATIC_ASSERT((unsigned)ZSTD_fast == 1);
+
+    assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, (int)strat));
+    DEBUGLOG(5, "Selected block compressor: dictMode=%d strat=%d rowMatchfinder=%d", (int)dictMode, (int)strat, (int)useRowMatchFinder);
+    if (ZSTD_rowMatchFinderUsed(strat, useRowMatchFinder)) {
+        static const ZSTD_BlockCompressor_f rowBasedBlockCompressors[4][3] = {
+            {
+                ZSTD_COMPRESSBLOCK_GREEDY_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY2_ROW
+            },
+            {
+                ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY_EXTDICT_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT_ROW
+            },
+            {
+                ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE_ROW
+            },
+            {
+                ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH_ROW,
+                ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH_ROW
+            }
+        };
+        DEBUGLOG(5, "Selecting a row-based matchfinder");
+        assert(useRowMatchFinder != ZSTD_ps_auto);
+        selectedCompressor = rowBasedBlockCompressors[(int)dictMode][(int)strat - (int)ZSTD_greedy];
+    } else {
+        selectedCompressor = blockCompressor[(int)dictMode][(int)strat];
+    }
+    assert(selectedCompressor != NULL);
+    return selectedCompressor;
+}
+
+static void ZSTD_storeLastLiterals(SeqStore_t* seqStorePtr,
+                                   const BYTE* anchor, size_t lastLLSize)
+{
+    ZSTD_memcpy(seqStorePtr->lit, anchor, lastLLSize);
+    seqStorePtr->lit += lastLLSize;
+}
+
+void ZSTD_resetSeqStore(SeqStore_t* ssPtr)
+{
+    ssPtr->lit = ssPtr->litStart;
+    ssPtr->sequences = ssPtr->sequencesStart;
+    ssPtr->longLengthType = ZSTD_llt_none;
+}
+
+/* ZSTD_postProcessSequenceProducerResult() :
+ * Validates and post-processes sequences obtained through the external matchfinder API:
+ *   - Checks whether nbExternalSeqs represents an error condition.
+ *   - Appends a block delimiter to outSeqs if one is not already present.
+ *     See zstd.h for context regarding block delimiters.
+ * Returns the number of sequences after post-processing, or an error code. */
+static size_t ZSTD_postProcessSequenceProducerResult(
+    ZSTD_Sequence* outSeqs, size_t nbExternalSeqs, size_t outSeqsCapacity, size_t srcSize
+) {
+    RETURN_ERROR_IF(
+        nbExternalSeqs > outSeqsCapacity,
+        sequenceProducer_failed,
+        "External sequence producer returned error code %lu",
+        (unsigned long)nbExternalSeqs
+    );
+
+    RETURN_ERROR_IF(
+        nbExternalSeqs == 0 && srcSize > 0,
+        sequenceProducer_failed,
+        "Got zero sequences from external sequence producer for a non-empty src buffer!"
+    );
+
+    if (srcSize == 0) {
+        ZSTD_memset(&outSeqs[0], 0, sizeof(ZSTD_Sequence));
+        return 1;
+    }
+
+    {
+        ZSTD_Sequence const lastSeq = outSeqs[nbExternalSeqs - 1];
+
+        /* We can return early if lastSeq is already a block delimiter. */
+        if (lastSeq.offset == 0 && lastSeq.matchLength == 0) {
+            return nbExternalSeqs;
+        }
+
+        /* This error condition is only possible if the external matchfinder
+         * produced an invalid parse, by definition of ZSTD_sequenceBound(). */
+        RETURN_ERROR_IF(
+            nbExternalSeqs == outSeqsCapacity,
+            sequenceProducer_failed,
+            "nbExternalSeqs == outSeqsCapacity but lastSeq is not a block delimiter!"
+        );
+
+        /* lastSeq is not a block delimiter, so we need to append one. */
+        ZSTD_memset(&outSeqs[nbExternalSeqs], 0, sizeof(ZSTD_Sequence));
+        return nbExternalSeqs + 1;
+    }
+}
+
+/* ZSTD_fastSequenceLengthSum() :
+ * Returns sum(litLen) + sum(matchLen) + lastLits for *seqBuf*.
+ * Similar to another function in zstd_compress.c (determine_blockSize),
+ * except it doesn't check for a block delimiter to end summation.
+ * Removing the early exit allows the compiler to auto-vectorize (https://godbolt.org/z/cY1cajz9P).
+ * This function can be deleted and replaced by determine_blockSize after we resolve issue #3456. */
+static size_t ZSTD_fastSequenceLengthSum(ZSTD_Sequence const* seqBuf, size_t seqBufSize) {
+    size_t matchLenSum, litLenSum, i;
+    matchLenSum = 0;
+    litLenSum = 0;
+    for (i = 0; i < seqBufSize; i++) {
+        litLenSum += seqBuf[i].litLength;
+        matchLenSum += seqBuf[i].matchLength;
+    }
+    return litLenSum + matchLenSum;
+}
+
+/**
+ * Function to validate sequences produced by a block compressor.
+ */
+static void ZSTD_validateSeqStore(const SeqStore_t* seqStore, const ZSTD_compressionParameters* cParams)
+{
+#if DEBUGLEVEL >= 1
+    const SeqDef* seq = seqStore->sequencesStart;
+    const SeqDef* const seqEnd = seqStore->sequences;
+    size_t const matchLenLowerBound = cParams->minMatch == 3 ? 3 : 4;
+    for (; seq < seqEnd; ++seq) {
+        const ZSTD_SequenceLength seqLength = ZSTD_getSequenceLength(seqStore, seq);
+        assert(seqLength.matchLength >= matchLenLowerBound);
+        (void)seqLength;
+        (void)matchLenLowerBound;
+    }
+#else
+    (void)seqStore;
+    (void)cParams;
+#endif
+}
+
+static size_t
+ZSTD_transferSequences_wBlockDelim(ZSTD_CCtx* cctx,
+                                   ZSTD_SequencePosition* seqPos,
+                             const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                             const void* src, size_t blockSize,
+                                   ZSTD_ParamSwitch_e externalRepSearch);
+
+typedef enum { ZSTDbss_compress, ZSTDbss_noCompress } ZSTD_BuildSeqStore_e;
+
+static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
+{
+    ZSTD_MatchState_t* const ms = &zc->blockState.matchState;
+    DEBUGLOG(5, "ZSTD_buildSeqStore (srcSize=%zu)", srcSize);
+    assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
+    /* Assert that we have correctly flushed the ctx params into the ms's copy */
+    ZSTD_assertEqualCParams(zc->appliedParams.cParams, ms->cParams);
+    /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+     * additional 1. We need to revisit and change this logic to be more consistent */
+    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1+1) {
+        if (zc->appliedParams.cParams.strategy >= ZSTD_btopt) {
+            ZSTD_ldm_skipRawSeqStoreBytes(&zc->externSeqStore, srcSize);
+        } else {
+            ZSTD_ldm_skipSequences(&zc->externSeqStore, srcSize, zc->appliedParams.cParams.minMatch);
+        }
+        return ZSTDbss_noCompress; /* don't even attempt compression below a certain srcSize */
+    }
+    ZSTD_resetSeqStore(&(zc->seqStore));
+    /* required for optimal parser to read stats from dictionary */
+    ms->opt.symbolCosts = &zc->blockState.prevCBlock->entropy;
+    /* tell the optimal parser how we expect to compress literals */
+    ms->opt.literalCompressionMode = zc->appliedParams.literalCompressionMode;
+    /* a gap between an attached dict and the current window is not safe,
+     * they must remain adjacent,
+     * and when that stops being the case, the dict must be unset */
+    assert(ms->dictMatchState == NULL || ms->loadedDictEnd == ms->window.dictLimit);
+
+    /* limited update after a very long match */
+    {   const BYTE* const base = ms->window.base;
+        const BYTE* const istart = (const BYTE*)src;
+        const U32 curr = (U32)(istart-base);
+        if (sizeof(ptrdiff_t)==8) assert(istart - base < (ptrdiff_t)(U32)(-1));   /* ensure no overflow */
+        if (curr > ms->nextToUpdate + 384)
+            ms->nextToUpdate = curr - MIN(192, (U32)(curr - ms->nextToUpdate - 384));
+    }
+
+    /* select and store sequences */
+    {   ZSTD_dictMode_e const dictMode = ZSTD_matchState_dictMode(ms);
+        size_t lastLLSize;
+        {   int i;
+            for (i = 0; i < ZSTD_REP_NUM; ++i)
+                zc->blockState.nextCBlock->rep[i] = zc->blockState.prevCBlock->rep[i];
+        }
+        if (zc->externSeqStore.pos < zc->externSeqStore.size) {
+            assert(zc->appliedParams.ldmParams.enableLdm == ZSTD_ps_disable);
+
+            /* External matchfinder + LDM is technically possible, just not implemented yet.
+             * We need to revisit soon and implement it. */
+            RETURN_ERROR_IF(
+                ZSTD_hasExtSeqProd(&zc->appliedParams),
+                parameter_combination_unsupported,
+                "Long-distance matching with external sequence producer enabled is not currently supported."
+            );
+
+            /* Updates ldmSeqStore.pos */
+            lastLLSize =
+                ZSTD_ldm_blockCompress(&zc->externSeqStore,
+                                       ms, &zc->seqStore,
+                                       zc->blockState.nextCBlock->rep,
+                                       zc->appliedParams.useRowMatchFinder,
+                                       src, srcSize);
+            assert(zc->externSeqStore.pos <= zc->externSeqStore.size);
+        } else if (zc->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable) {
+            RawSeqStore_t ldmSeqStore = kNullRawSeqStore;
+
+            /* External matchfinder + LDM is technically possible, just not implemented yet.
+             * We need to revisit soon and implement it. */
+            RETURN_ERROR_IF(
+                ZSTD_hasExtSeqProd(&zc->appliedParams),
+                parameter_combination_unsupported,
+                "Long-distance matching with external sequence producer enabled is not currently supported."
+            );
+
+            ldmSeqStore.seq = zc->ldmSequences;
+            ldmSeqStore.capacity = zc->maxNbLdmSequences;
+            /* Updates ldmSeqStore.size */
+            FORWARD_IF_ERROR(ZSTD_ldm_generateSequences(&zc->ldmState, &ldmSeqStore,
+                                               &zc->appliedParams.ldmParams,
+                                               src, srcSize), "");
+            /* Updates ldmSeqStore.pos */
+            lastLLSize =
+                ZSTD_ldm_blockCompress(&ldmSeqStore,
+                                       ms, &zc->seqStore,
+                                       zc->blockState.nextCBlock->rep,
+                                       zc->appliedParams.useRowMatchFinder,
+                                       src, srcSize);
+            assert(ldmSeqStore.pos == ldmSeqStore.size);
+        } else if (ZSTD_hasExtSeqProd(&zc->appliedParams)) {
+            assert(
+                zc->extSeqBufCapacity >= ZSTD_sequenceBound(srcSize)
+            );
+            assert(zc->appliedParams.extSeqProdFunc != NULL);
+
+            {   U32 const windowSize = (U32)1 << zc->appliedParams.cParams.windowLog;
+
+                size_t const nbExternalSeqs = (zc->appliedParams.extSeqProdFunc)(
+                    zc->appliedParams.extSeqProdState,
+                    zc->extSeqBuf,
+                    zc->extSeqBufCapacity,
+                    src, srcSize,
+                    NULL, 0,  /* dict and dictSize, currently not supported */
+                    zc->appliedParams.compressionLevel,
+                    windowSize
+                );
+
+                size_t const nbPostProcessedSeqs = ZSTD_postProcessSequenceProducerResult(
+                    zc->extSeqBuf,
+                    nbExternalSeqs,
+                    zc->extSeqBufCapacity,
+                    srcSize
+                );
+
+                /* Return early if there is no error, since we don't need to worry about last literals */
+                if (!ZSTD_isError(nbPostProcessedSeqs)) {
+                    ZSTD_SequencePosition seqPos = {0,0,0};
+                    size_t const seqLenSum = ZSTD_fastSequenceLengthSum(zc->extSeqBuf, nbPostProcessedSeqs);
+                    RETURN_ERROR_IF(seqLenSum > srcSize, externalSequences_invalid, "External sequences imply too large a block!");
+                    FORWARD_IF_ERROR(
+                        ZSTD_transferSequences_wBlockDelim(
+                            zc, &seqPos,
+                            zc->extSeqBuf, nbPostProcessedSeqs,
+                            src, srcSize,
+                            zc->appliedParams.searchForExternalRepcodes
+                        ),
+                        "Failed to copy external sequences to seqStore!"
+                    );
+                    ms->ldmSeqStore = NULL;
+                    DEBUGLOG(5, "Copied %lu sequences from external sequence producer to internal seqStore.", (unsigned long)nbExternalSeqs);
+                    return ZSTDbss_compress;
+                }
+
+                /* Propagate the error if fallback is disabled */
+                if (!zc->appliedParams.enableMatchFinderFallback) {
+                    return nbPostProcessedSeqs;
+                }
+
+                /* Fallback to software matchfinder */
+                {   ZSTD_BlockCompressor_f const blockCompressor =
+                        ZSTD_selectBlockCompressor(
+                            zc->appliedParams.cParams.strategy,
+                            zc->appliedParams.useRowMatchFinder,
+                            dictMode);
+                    ms->ldmSeqStore = NULL;
+                    DEBUGLOG(
+                        5,
+                        "External sequence producer returned error code %lu. Falling back to internal parser.",
+                        (unsigned long)nbExternalSeqs
+                    );
+                    lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
+            }   }
+        } else {   /* not long range mode and no external matchfinder */
+            ZSTD_BlockCompressor_f const blockCompressor = ZSTD_selectBlockCompressor(
+                    zc->appliedParams.cParams.strategy,
+                    zc->appliedParams.useRowMatchFinder,
+                    dictMode);
+            ms->ldmSeqStore = NULL;
+            lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
+        }
+        {   const BYTE* const lastLiterals = (const BYTE*)src + srcSize - lastLLSize;
+            ZSTD_storeLastLiterals(&zc->seqStore, lastLiterals, lastLLSize);
+    }   }
+    ZSTD_validateSeqStore(&zc->seqStore, &zc->appliedParams.cParams);
+    return ZSTDbss_compress;
+}
+
+static size_t ZSTD_copyBlockSequences(SeqCollector* seqCollector, const SeqStore_t* seqStore, const U32 prevRepcodes[ZSTD_REP_NUM])
+{
+    const SeqDef* inSeqs = seqStore->sequencesStart;
+    const size_t nbInSequences = (size_t)(seqStore->sequences - inSeqs);
+    const size_t nbInLiterals = (size_t)(seqStore->lit - seqStore->litStart);
+
+    ZSTD_Sequence* outSeqs = seqCollector->seqIndex == 0 ? seqCollector->seqStart : seqCollector->seqStart + seqCollector->seqIndex;
+    const size_t nbOutSequences = nbInSequences + 1;
+    size_t nbOutLiterals = 0;
+    Repcodes_t repcodes;
+    size_t i;
+
+    /* Bounds check that we have enough space for every input sequence
+     * and the block delimiter
+     */
+    assert(seqCollector->seqIndex <= seqCollector->maxSequences);
+    RETURN_ERROR_IF(
+        nbOutSequences > (size_t)(seqCollector->maxSequences - seqCollector->seqIndex),
+        dstSize_tooSmall,
+        "Not enough space to copy sequences");
+
+    ZSTD_memcpy(&repcodes, prevRepcodes, sizeof(repcodes));
+    for (i = 0; i < nbInSequences; ++i) {
+        U32 rawOffset;
+        outSeqs[i].litLength = inSeqs[i].litLength;
+        outSeqs[i].matchLength = inSeqs[i].mlBase + MINMATCH;
+        outSeqs[i].rep = 0;
+
+        /* Handle the possible single length >= 64K
+         * There can only be one because we add MINMATCH to every match length,
+         * and blocks are at most 128K.
+         */
+        if (i == seqStore->longLengthPos) {
+            if (seqStore->longLengthType == ZSTD_llt_literalLength) {
+                outSeqs[i].litLength += 0x10000;
+            } else if (seqStore->longLengthType == ZSTD_llt_matchLength) {
+                outSeqs[i].matchLength += 0x10000;
+            }
+        }
+
+        /* Determine the raw offset given the offBase, which may be a repcode. */
+        if (OFFBASE_IS_REPCODE(inSeqs[i].offBase)) {
+            const U32 repcode = OFFBASE_TO_REPCODE(inSeqs[i].offBase);
+            assert(repcode > 0);
+            outSeqs[i].rep = repcode;
+            if (outSeqs[i].litLength != 0) {
+                rawOffset = repcodes.rep[repcode - 1];
+            } else {
+                if (repcode == 3) {
+                    assert(repcodes.rep[0] > 1);
+                    rawOffset = repcodes.rep[0] - 1;
+                } else {
+                    rawOffset = repcodes.rep[repcode];
+                }
+            }
+        } else {
+            rawOffset = OFFBASE_TO_OFFSET(inSeqs[i].offBase);
+        }
+        outSeqs[i].offset = rawOffset;
+
+        /* Update repcode history for the sequence */
+        ZSTD_updateRep(repcodes.rep,
+                       inSeqs[i].offBase,
+                       inSeqs[i].litLength == 0);
+
+        nbOutLiterals += outSeqs[i].litLength;
+    }
+    /* Insert last literals (if any exist) in the block as a sequence with ml == off == 0.
+     * If there are no last literals, then we'll emit (of: 0, ml: 0, ll: 0), which is a marker
+     * for the block boundary, according to the API.
+     */
+    assert(nbInLiterals >= nbOutLiterals);
+    {
+        const size_t lastLLSize = nbInLiterals - nbOutLiterals;
+        outSeqs[nbInSequences].litLength = (U32)lastLLSize;
+        outSeqs[nbInSequences].matchLength = 0;
+        outSeqs[nbInSequences].offset = 0;
+        assert(nbOutSequences == nbInSequences + 1);
+    }
+    seqCollector->seqIndex += nbOutSequences;
+    assert(seqCollector->seqIndex <= seqCollector->maxSequences);
+
+    return 0;
+}
+
+size_t ZSTD_sequenceBound(size_t srcSize) {
+    const size_t maxNbSeq = (srcSize / ZSTD_MINMATCH_MIN) + 1;
+    const size_t maxNbDelims = (srcSize / ZSTD_BLOCKSIZE_MAX_MIN) + 1;
+    return maxNbSeq + maxNbDelims;
+}
+
+size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
+                              size_t outSeqsSize, const void* src, size_t srcSize)
+{
+    const size_t dstCapacity = ZSTD_compressBound(srcSize);
+    void* dst; /* Make C90 happy. */
+    SeqCollector seqCollector;
+    {
+        int targetCBlockSize;
+        FORWARD_IF_ERROR(ZSTD_CCtx_getParameter(zc, ZSTD_c_targetCBlockSize, &targetCBlockSize), "");
+        RETURN_ERROR_IF(targetCBlockSize != 0, parameter_unsupported, "targetCBlockSize != 0");
+    }
+    {
+        int nbWorkers;
+        FORWARD_IF_ERROR(ZSTD_CCtx_getParameter(zc, ZSTD_c_nbWorkers, &nbWorkers), "");
+        RETURN_ERROR_IF(nbWorkers != 0, parameter_unsupported, "nbWorkers != 0");
+    }
+
+    dst = ZSTD_customMalloc(dstCapacity, ZSTD_defaultCMem);
+    RETURN_ERROR_IF(dst == NULL, memory_allocation, "NULL pointer!");
+
+    seqCollector.collectSequences = 1;
+    seqCollector.seqStart = outSeqs;
+    seqCollector.seqIndex = 0;
+    seqCollector.maxSequences = outSeqsSize;
+    zc->seqCollector = seqCollector;
+
+    {
+        const size_t ret = ZSTD_compress2(zc, dst, dstCapacity, src, srcSize);
+        ZSTD_customFree(dst, ZSTD_defaultCMem);
+        FORWARD_IF_ERROR(ret, "ZSTD_compress2 failed");
+    }
+    assert(zc->seqCollector.seqIndex <= ZSTD_sequenceBound(srcSize));
+    return zc->seqCollector.seqIndex;
+}
+
+size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize) {
+    size_t in = 0;
+    size_t out = 0;
+    for (; in < seqsSize; ++in) {
+        if (sequences[in].offset == 0 && sequences[in].matchLength == 0) {
+            if (in != seqsSize - 1) {
+                sequences[in+1].litLength += sequences[in].litLength;
+            }
+        } else {
+            sequences[out] = sequences[in];
+            ++out;
+        }
+    }
+    return out;
+}
+
+/* Unrolled loop to read four size_ts of input at a time. Returns 1 if is RLE, 0 if not. */
+static int ZSTD_isRLE(const BYTE* src, size_t length) {
+    const BYTE* ip = src;
+    const BYTE value = ip[0];
+    const size_t valueST = (size_t)((U64)value * 0x0101010101010101ULL);
+    const size_t unrollSize = sizeof(size_t) * 4;
+    const size_t unrollMask = unrollSize - 1;
+    const size_t prefixLength = length & unrollMask;
+    size_t i;
+    if (length == 1) return 1;
+    /* Check if prefix is RLE first before using unrolled loop */
+    if (prefixLength && ZSTD_count(ip+1, ip, ip+prefixLength) != prefixLength-1) {
+        return 0;
+    }
+    for (i = prefixLength; i != length; i += unrollSize) {
+        size_t u;
+        for (u = 0; u < unrollSize; u += sizeof(size_t)) {
+            if (MEM_readST(ip + i + u) != valueST) {
+                return 0;
+    }   }   }
+    return 1;
+}
+
+/* Returns true if the given block may be RLE.
+ * This is just a heuristic based on the compressibility.
+ * It may return both false positives and false negatives.
+ */
+static int ZSTD_maybeRLE(SeqStore_t const* seqStore)
+{
+    size_t const nbSeqs = (size_t)(seqStore->sequences - seqStore->sequencesStart);
+    size_t const nbLits = (size_t)(seqStore->lit - seqStore->litStart);
+
+    return nbSeqs < 4 && nbLits < 10;
+}
+
+static void
+ZSTD_blockState_confirmRepcodesAndEntropyTables(ZSTD_blockState_t* const bs)
+{
+    ZSTD_compressedBlockState_t* const tmp = bs->prevCBlock;
+    bs->prevCBlock = bs->nextCBlock;
+    bs->nextCBlock = tmp;
+}
+
+/* Writes the block header */
+static void
+writeBlockHeader(void* op, size_t cSize, size_t blockSize, U32 lastBlock)
+{
+    U32 const cBlockHeader = cSize == 1 ?
+                        lastBlock + (((U32)bt_rle)<<1) + (U32)(blockSize << 3) :
+                        lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
+    MEM_writeLE24(op, cBlockHeader);
+    DEBUGLOG(5, "writeBlockHeader: cSize: %zu blockSize: %zu lastBlock: %u", cSize, blockSize, lastBlock);
+}
+
+/** ZSTD_buildBlockEntropyStats_literals() :
+ *  Builds entropy for the literals.
+ *  Stores literals block type (raw, rle, compressed, repeat) and
+ *  huffman description table to hufMetadata.
+ *  Requires ENTROPY_WORKSPACE_SIZE workspace
+ * @return : size of huffman description table, or an error code
+ */
+static size_t
+ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSize,
+                               const ZSTD_hufCTables_t* prevHuf,
+                                     ZSTD_hufCTables_t* nextHuf,
+                                     ZSTD_hufCTablesMetadata_t* hufMetadata,
+                               const int literalsCompressionIsDisabled,
+                                     void* workspace, size_t wkspSize,
+                                     int hufFlags)
+{
+    BYTE* const wkspStart = (BYTE*)workspace;
+    BYTE* const wkspEnd = wkspStart + wkspSize;
+    BYTE* const countWkspStart = wkspStart;
+    unsigned* const countWksp = (unsigned*)workspace;
+    const size_t countWkspSize = (HUF_SYMBOLVALUE_MAX + 1) * sizeof(unsigned);
+    BYTE* const nodeWksp = countWkspStart + countWkspSize;
+    const size_t nodeWkspSize = (size_t)(wkspEnd - nodeWksp);
+    unsigned maxSymbolValue = HUF_SYMBOLVALUE_MAX;
+    unsigned huffLog = LitHufLog;
+    HUF_repeat repeat = prevHuf->repeatMode;
+    DEBUGLOG(5, "ZSTD_buildBlockEntropyStats_literals (srcSize=%zu)", srcSize);
+
+    /* Prepare nextEntropy assuming reusing the existing table */
+    ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+
+    if (literalsCompressionIsDisabled) {
+        DEBUGLOG(5, "set_basic - disabled");
+        hufMetadata->hType = set_basic;
+        return 0;
+    }
+
+    /* small ? don't even attempt compression (speed opt) */
+#ifndef COMPRESS_LITERALS_SIZE_MIN
+# define COMPRESS_LITERALS_SIZE_MIN 63  /* heuristic */
+#endif
+    {   size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
+        if (srcSize <= minLitSize) {
+            DEBUGLOG(5, "set_basic - too small");
+            hufMetadata->hType = set_basic;
+            return 0;
+    }   }
+
+    /* Scan input and build symbol stats */
+    {   size_t const largest =
+            HIST_count_wksp (countWksp, &maxSymbolValue,
+                            (const BYTE*)src, srcSize,
+                            workspace, wkspSize);
+        FORWARD_IF_ERROR(largest, "HIST_count_wksp failed");
+        if (largest == srcSize) {
+            /* only one literal symbol */
+            DEBUGLOG(5, "set_rle");
+            hufMetadata->hType = set_rle;
+            return 0;
+        }
+        if (largest <= (srcSize >> 7)+4) {
+            /* heuristic: likely not compressible */
+            DEBUGLOG(5, "set_basic - no gain");
+            hufMetadata->hType = set_basic;
+            return 0;
+    }   }
+
+    /* Validate the previous Huffman table */
+    if (repeat == HUF_repeat_check
+      && !HUF_validateCTable((HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue)) {
+        repeat = HUF_repeat_none;
+    }
+
+    /* Build Huffman Tree */
+    ZSTD_memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
+    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue, nodeWksp, nodeWkspSize, nextHuf->CTable, countWksp, hufFlags);
+    assert(huffLog <= LitHufLog);
+    {   size_t const maxBits = HUF_buildCTable_wksp((HUF_CElt*)nextHuf->CTable, countWksp,
+                                                    maxSymbolValue, huffLog,
+                                                    nodeWksp, nodeWkspSize);
+        FORWARD_IF_ERROR(maxBits, "HUF_buildCTable_wksp");
+        huffLog = (U32)maxBits;
+    }
+    {   /* Build and write the CTable */
+        size_t const newCSize = HUF_estimateCompressedSize(
+                (HUF_CElt*)nextHuf->CTable, countWksp, maxSymbolValue);
+        size_t const hSize = HUF_writeCTable_wksp(
+                hufMetadata->hufDesBuffer, sizeof(hufMetadata->hufDesBuffer),
+                (HUF_CElt*)nextHuf->CTable, maxSymbolValue, huffLog,
+                nodeWksp, nodeWkspSize);
+        /* Check against repeating the previous CTable */
+        if (repeat != HUF_repeat_none) {
+            size_t const oldCSize = HUF_estimateCompressedSize(
+                    (HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
+            if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
+                DEBUGLOG(5, "set_repeat - smaller");
+                ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+                hufMetadata->hType = set_repeat;
+                return 0;
+        }   }
+        if (newCSize + hSize >= srcSize) {
+            DEBUGLOG(5, "set_basic - no gains");
+            ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+            hufMetadata->hType = set_basic;
+            return 0;
+        }
+        DEBUGLOG(5, "set_compressed (hSize=%u)", (U32)hSize);
+        hufMetadata->hType = set_compressed;
+        nextHuf->repeatMode = HUF_repeat_check;
+        return hSize;
+    }
+}
+
+
+/* ZSTD_buildDummySequencesStatistics():
+ * Returns a ZSTD_symbolEncodingTypeStats_t with all encoding types as set_basic,
+ * and updates nextEntropy to the appropriate repeatMode.
+ */
+static ZSTD_symbolEncodingTypeStats_t
+ZSTD_buildDummySequencesStatistics(ZSTD_fseCTables_t* nextEntropy)
+{
+    ZSTD_symbolEncodingTypeStats_t stats = {set_basic, set_basic, set_basic, 0, 0, 0};
+    nextEntropy->litlength_repeatMode = FSE_repeat_none;
+    nextEntropy->offcode_repeatMode = FSE_repeat_none;
+    nextEntropy->matchlength_repeatMode = FSE_repeat_none;
+    return stats;
+}
+
+/** ZSTD_buildBlockEntropyStats_sequences() :
+ *  Builds entropy for the sequences.
+ *  Stores symbol compression modes and fse table to fseMetadata.
+ *  Requires ENTROPY_WORKSPACE_SIZE wksp.
+ * @return : size of fse tables or error code */
+static size_t
+ZSTD_buildBlockEntropyStats_sequences(
+                const SeqStore_t* seqStorePtr,
+                const ZSTD_fseCTables_t* prevEntropy,
+                      ZSTD_fseCTables_t* nextEntropy,
+                const ZSTD_CCtx_params* cctxParams,
+                      ZSTD_fseCTablesMetadata_t* fseMetadata,
+                      void* workspace, size_t wkspSize)
+{
+    ZSTD_strategy const strategy = cctxParams->cParams.strategy;
+    size_t const nbSeq = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    BYTE* const ostart = fseMetadata->fseTablesBuffer;
+    BYTE* const oend = ostart + sizeof(fseMetadata->fseTablesBuffer);
+    BYTE* op = ostart;
+    unsigned* countWorkspace = (unsigned*)workspace;
+    unsigned* entropyWorkspace = countWorkspace + (MaxSeq + 1);
+    size_t entropyWorkspaceSize = wkspSize - (MaxSeq + 1) * sizeof(*countWorkspace);
+    ZSTD_symbolEncodingTypeStats_t stats;
+
+    DEBUGLOG(5, "ZSTD_buildBlockEntropyStats_sequences (nbSeq=%zu)", nbSeq);
+    stats = nbSeq != 0 ? ZSTD_buildSequencesStatistics(seqStorePtr, nbSeq,
+                                          prevEntropy, nextEntropy, op, oend,
+                                          strategy, countWorkspace,
+                                          entropyWorkspace, entropyWorkspaceSize)
+                       : ZSTD_buildDummySequencesStatistics(nextEntropy);
+    FORWARD_IF_ERROR(stats.size, "ZSTD_buildSequencesStatistics failed!");
+    fseMetadata->llType = (SymbolEncodingType_e) stats.LLtype;
+    fseMetadata->ofType = (SymbolEncodingType_e) stats.Offtype;
+    fseMetadata->mlType = (SymbolEncodingType_e) stats.MLtype;
+    fseMetadata->lastCountSize = stats.lastCountSize;
+    return stats.size;
+}
+
+
+/** ZSTD_buildBlockEntropyStats() :
+ *  Builds entropy for the block.
+ *  Requires workspace size ENTROPY_WORKSPACE_SIZE
+ * @return : 0 on success, or an error code
+ *  Note : also employed in superblock
+ */
+size_t ZSTD_buildBlockEntropyStats(
+            const SeqStore_t* seqStorePtr,
+            const ZSTD_entropyCTables_t* prevEntropy,
+                  ZSTD_entropyCTables_t* nextEntropy,
+            const ZSTD_CCtx_params* cctxParams,
+                  ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                  void* workspace, size_t wkspSize)
+{
+    size_t const litSize = (size_t)(seqStorePtr->lit - seqStorePtr->litStart);
+    int const huf_useOptDepth = (cctxParams->cParams.strategy >= HUF_OPTIMAL_DEPTH_THRESHOLD);
+    int const hufFlags = huf_useOptDepth ? HUF_flags_optimalDepth : 0;
+
+    entropyMetadata->hufMetadata.hufDesSize =
+        ZSTD_buildBlockEntropyStats_literals(seqStorePtr->litStart, litSize,
+                                            &prevEntropy->huf, &nextEntropy->huf,
+                                            &entropyMetadata->hufMetadata,
+                                            ZSTD_literalsCompressionIsDisabled(cctxParams),
+                                            workspace, wkspSize, hufFlags);
+
+    FORWARD_IF_ERROR(entropyMetadata->hufMetadata.hufDesSize, "ZSTD_buildBlockEntropyStats_literals failed");
+    entropyMetadata->fseMetadata.fseTablesSize =
+        ZSTD_buildBlockEntropyStats_sequences(seqStorePtr,
+                                              &prevEntropy->fse, &nextEntropy->fse,
+                                              cctxParams,
+                                              &entropyMetadata->fseMetadata,
+                                              workspace, wkspSize);
+    FORWARD_IF_ERROR(entropyMetadata->fseMetadata.fseTablesSize, "ZSTD_buildBlockEntropyStats_sequences failed");
+    return 0;
+}
+
+/* Returns the size estimate for the literals section (header + content) of a block */
+static size_t
+ZSTD_estimateBlockSize_literal(const BYTE* literals, size_t litSize,
+                               const ZSTD_hufCTables_t* huf,
+                               const ZSTD_hufCTablesMetadata_t* hufMetadata,
+                               void* workspace, size_t wkspSize,
+                               int writeEntropy)
+{
+    unsigned* const countWksp = (unsigned*)workspace;
+    unsigned maxSymbolValue = HUF_SYMBOLVALUE_MAX;
+    size_t literalSectionHeaderSize = 3 + (litSize >= 1 KB) + (litSize >= 16 KB);
+    U32 singleStream = litSize < 256;
+
+    if (hufMetadata->hType == set_basic) return litSize;
+    else if (hufMetadata->hType == set_rle) return 1;
+    else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
+        size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)literals, litSize, workspace, wkspSize);
+        if (ZSTD_isError(largest)) return litSize;
+        {   size_t cLitSizeEstimate = HUF_estimateCompressedSize((const HUF_CElt*)huf->CTable, countWksp, maxSymbolValue);
+            if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
+            if (!singleStream) cLitSizeEstimate += 6; /* multi-stream huffman uses 6-byte jump table */
+            return cLitSizeEstimate + literalSectionHeaderSize;
+    }   }
+    assert(0); /* impossible */
+    return 0;
+}
+
+/* Returns the size estimate for the FSE-compressed symbols (of, ml, ll) of a block */
+static size_t
+ZSTD_estimateBlockSize_symbolType(SymbolEncodingType_e type,
+                    const BYTE* codeTable, size_t nbSeq, unsigned maxCode,
+                    const FSE_CTable* fseCTable,
+                    const U8* additionalBits,
+                    short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+                    void* workspace, size_t wkspSize)
+{
+    unsigned* const countWksp = (unsigned*)workspace;
+    const BYTE* ctp = codeTable;
+    const BYTE* const ctStart = ctp;
+    const BYTE* const ctEnd = ctStart + nbSeq;
+    size_t cSymbolTypeSizeEstimateInBits = 0;
+    unsigned max = maxCode;
+
+    HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize);  /* can't fail */
+    if (type == set_basic) {
+        /* We selected this encoding type, so it must be valid. */
+        assert(max <= defaultMax);
+        (void)defaultMax;
+        cSymbolTypeSizeEstimateInBits = ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max);
+    } else if (type == set_rle) {
+        cSymbolTypeSizeEstimateInBits = 0;
+    } else if (type == set_compressed || type == set_repeat) {
+        cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(fseCTable, countWksp, max);
+    }
+    if (ZSTD_isError(cSymbolTypeSizeEstimateInBits)) {
+        return nbSeq * 10;
+    }
+    while (ctp < ctEnd) {
+        if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
+        else cSymbolTypeSizeEstimateInBits += *ctp; /* for offset, offset code is also the number of additional bits */
+        ctp++;
+    }
+    return cSymbolTypeSizeEstimateInBits >> 3;
+}
+
+/* Returns the size estimate for the sequences section (header + content) of a block */
+static size_t
+ZSTD_estimateBlockSize_sequences(const BYTE* ofCodeTable,
+                                 const BYTE* llCodeTable,
+                                 const BYTE* mlCodeTable,
+                                 size_t nbSeq,
+                                 const ZSTD_fseCTables_t* fseTables,
+                                 const ZSTD_fseCTablesMetadata_t* fseMetadata,
+                                 void* workspace, size_t wkspSize,
+                                 int writeEntropy)
+{
+    size_t sequencesSectionHeaderSize = 1 /* seqHead */ + 1 /* min seqSize size */ + (nbSeq >= 128) + (nbSeq >= LONGNBSEQ);
+    size_t cSeqSizeEstimate = 0;
+    cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, nbSeq, MaxOff,
+                                    fseTables->offcodeCTable, NULL,
+                                    OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                                    workspace, wkspSize);
+    cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->llType, llCodeTable, nbSeq, MaxLL,
+                                    fseTables->litlengthCTable, LL_bits,
+                                    LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                                    workspace, wkspSize);
+    cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, nbSeq, MaxML,
+                                    fseTables->matchlengthCTable, ML_bits,
+                                    ML_defaultNorm, ML_defaultNormLog, MaxML,
+                                    workspace, wkspSize);
+    if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
+    return cSeqSizeEstimate + sequencesSectionHeaderSize;
+}
+
+/* Returns the size estimate for a given stream of literals, of, ll, ml */
+static size_t
+ZSTD_estimateBlockSize(const BYTE* literals, size_t litSize,
+                       const BYTE* ofCodeTable,
+                       const BYTE* llCodeTable,
+                       const BYTE* mlCodeTable,
+                       size_t nbSeq,
+                       const ZSTD_entropyCTables_t* entropy,
+                       const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                       void* workspace, size_t wkspSize,
+                       int writeLitEntropy, int writeSeqEntropy)
+{
+    size_t const literalsSize = ZSTD_estimateBlockSize_literal(literals, litSize,
+                                    &entropy->huf, &entropyMetadata->hufMetadata,
+                                    workspace, wkspSize, writeLitEntropy);
+    size_t const seqSize = ZSTD_estimateBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
+                                    nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
+                                    workspace, wkspSize, writeSeqEntropy);
+    return seqSize + literalsSize + ZSTD_blockHeaderSize;
+}
+
+/* Builds entropy statistics and uses them for blocksize estimation.
+ *
+ * @return: estimated compressed size of the seqStore, or a zstd error.
+ */
+static size_t
+ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(SeqStore_t* seqStore, ZSTD_CCtx* zc)
+{
+    ZSTD_entropyCTablesMetadata_t* const entropyMetadata = &zc->blockSplitCtx.entropyMetadata;
+    DEBUGLOG(6, "ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize()");
+    FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(seqStore,
+                    &zc->blockState.prevCBlock->entropy,
+                    &zc->blockState.nextCBlock->entropy,
+                    &zc->appliedParams,
+                    entropyMetadata,
+                    zc->tmpWorkspace, zc->tmpWkspSize), "");
+    return ZSTD_estimateBlockSize(
+                    seqStore->litStart, (size_t)(seqStore->lit - seqStore->litStart),
+                    seqStore->ofCode, seqStore->llCode, seqStore->mlCode,
+                    (size_t)(seqStore->sequences - seqStore->sequencesStart),
+                    &zc->blockState.nextCBlock->entropy,
+                    entropyMetadata,
+                    zc->tmpWorkspace, zc->tmpWkspSize,
+                    (int)(entropyMetadata->hufMetadata.hType == set_compressed), 1);
+}
+
+/* Returns literals bytes represented in a seqStore */
+static size_t ZSTD_countSeqStoreLiteralsBytes(const SeqStore_t* const seqStore)
+{
+    size_t literalsBytes = 0;
+    size_t const nbSeqs = (size_t)(seqStore->sequences - seqStore->sequencesStart);
+    size_t i;
+    for (i = 0; i < nbSeqs; ++i) {
+        SeqDef const seq = seqStore->sequencesStart[i];
+        literalsBytes += seq.litLength;
+        if (i == seqStore->longLengthPos && seqStore->longLengthType == ZSTD_llt_literalLength) {
+            literalsBytes += 0x10000;
+    }   }
+    return literalsBytes;
+}
+
+/* Returns match bytes represented in a seqStore */
+static size_t ZSTD_countSeqStoreMatchBytes(const SeqStore_t* const seqStore)
+{
+    size_t matchBytes = 0;
+    size_t const nbSeqs = (size_t)(seqStore->sequences - seqStore->sequencesStart);
+    size_t i;
+    for (i = 0; i < nbSeqs; ++i) {
+        SeqDef seq = seqStore->sequencesStart[i];
+        matchBytes += seq.mlBase + MINMATCH;
+        if (i == seqStore->longLengthPos && seqStore->longLengthType == ZSTD_llt_matchLength) {
+            matchBytes += 0x10000;
+    }   }
+    return matchBytes;
+}
+
+/* Derives the seqStore that is a chunk of the originalSeqStore from [startIdx, endIdx).
+ * Stores the result in resultSeqStore.
+ */
+static void ZSTD_deriveSeqStoreChunk(SeqStore_t* resultSeqStore,
+                               const SeqStore_t* originalSeqStore,
+                                     size_t startIdx, size_t endIdx)
+{
+    *resultSeqStore = *originalSeqStore;
+    if (startIdx > 0) {
+        resultSeqStore->sequences = originalSeqStore->sequencesStart + startIdx;
+        resultSeqStore->litStart += ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
+    }
+
+    /* Move longLengthPos into the correct position if necessary */
+    if (originalSeqStore->longLengthType != ZSTD_llt_none) {
+        if (originalSeqStore->longLengthPos < startIdx || originalSeqStore->longLengthPos > endIdx) {
+            resultSeqStore->longLengthType = ZSTD_llt_none;
+        } else {
+            resultSeqStore->longLengthPos -= (U32)startIdx;
+        }
+    }
+    resultSeqStore->sequencesStart = originalSeqStore->sequencesStart + startIdx;
+    resultSeqStore->sequences = originalSeqStore->sequencesStart + endIdx;
+    if (endIdx == (size_t)(originalSeqStore->sequences - originalSeqStore->sequencesStart)) {
+        /* This accounts for possible last literals if the derived chunk reaches the end of the block */
+        assert(resultSeqStore->lit == originalSeqStore->lit);
+    } else {
+        size_t const literalsBytes = ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
+        resultSeqStore->lit = resultSeqStore->litStart + literalsBytes;
+    }
+    resultSeqStore->llCode += startIdx;
+    resultSeqStore->mlCode += startIdx;
+    resultSeqStore->ofCode += startIdx;
+}
+
+/**
+ * Returns the raw offset represented by the combination of offBase, ll0, and repcode history.
+ * offBase must represent a repcode in the numeric representation of ZSTD_storeSeq().
+ */
+static U32
+ZSTD_resolveRepcodeToRawOffset(const U32 rep[ZSTD_REP_NUM], const U32 offBase, const U32 ll0)
+{
+    U32 const adjustedRepCode = OFFBASE_TO_REPCODE(offBase) - 1 + ll0;  /* [ 0 - 3 ] */
+    assert(OFFBASE_IS_REPCODE(offBase));
+    if (adjustedRepCode == ZSTD_REP_NUM) {
+        assert(ll0);
+        /* litlength == 0 and offCode == 2 implies selection of first repcode - 1
+         * This is only valid if it results in a valid offset value, aka > 0.
+         * Note : it may happen that `rep[0]==1` in exceptional circumstances.
+         * In which case this function will return 0, which is an invalid offset.
+         * It's not an issue though, since this value will be
+         * compared and discarded within ZSTD_seqStore_resolveOffCodes().
+         */
+        return rep[0] - 1;
+    }
+    return rep[adjustedRepCode];
+}
+
+/**
+ * ZSTD_seqStore_resolveOffCodes() reconciles any possible divergences in offset history that may arise
+ * due to emission of RLE/raw blocks that disturb the offset history,
+ * and replaces any repcodes within the seqStore that may be invalid.
+ *
+ * dRepcodes are updated as would be on the decompression side.
+ * cRepcodes are updated exactly in accordance with the seqStore.
+ *
+ * Note : this function assumes seq->offBase respects the following numbering scheme :
+ *        0 : invalid
+ *        1-3 : repcode 1-3
+ *        4+ : real_offset+3
+ */
+static void
+ZSTD_seqStore_resolveOffCodes(Repcodes_t* const dRepcodes, Repcodes_t* const cRepcodes,
+                        const SeqStore_t* const seqStore, U32 const nbSeq)
+{
+    U32 idx = 0;
+    U32 const longLitLenIdx = seqStore->longLengthType == ZSTD_llt_literalLength ? seqStore->longLengthPos : nbSeq;
+    for (; idx < nbSeq; ++idx) {
+        SeqDef* const seq = seqStore->sequencesStart + idx;
+        U32 const ll0 = (seq->litLength == 0) && (idx != longLitLenIdx);
+        U32 const offBase = seq->offBase;
+        assert(offBase > 0);
+        if (OFFBASE_IS_REPCODE(offBase)) {
+            U32 const dRawOffset = ZSTD_resolveRepcodeToRawOffset(dRepcodes->rep, offBase, ll0);
+            U32 const cRawOffset = ZSTD_resolveRepcodeToRawOffset(cRepcodes->rep, offBase, ll0);
+            /* Adjust simulated decompression repcode history if we come across a mismatch. Replace
+             * the repcode with the offset it actually references, determined by the compression
+             * repcode history.
+             */
+            if (dRawOffset != cRawOffset) {
+                seq->offBase = OFFSET_TO_OFFBASE(cRawOffset);
+            }
+        }
+        /* Compression repcode history is always updated with values directly from the unmodified seqStore.
+         * Decompression repcode history may use modified seq->offset value taken from compression repcode history.
+         */
+        ZSTD_updateRep(dRepcodes->rep, seq->offBase, ll0);
+        ZSTD_updateRep(cRepcodes->rep, offBase, ll0);
+    }
+}
+
+/* ZSTD_compressSeqStore_singleBlock():
+ * Compresses a seqStore into a block with a block header, into the buffer dst.
+ *
+ * Returns the total size of that block (including header) or a ZSTD error code.
+ */
+static size_t
+ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc,
+                            const SeqStore_t* const seqStore,
+                                  Repcodes_t* const dRep, Repcodes_t* const cRep,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                                  U32 lastBlock, U32 isPartition)
+{
+    const U32 rleMaxLength = 25;
+    BYTE* op = (BYTE*)dst;
+    const BYTE* ip = (const BYTE*)src;
+    size_t cSize;
+    size_t cSeqsSize;
+
+    /* In case of an RLE or raw block, the simulated decompression repcode history must be reset */
+    Repcodes_t const dRepOriginal = *dRep;
+    DEBUGLOG(5, "ZSTD_compressSeqStore_singleBlock");
+    if (isPartition)
+        ZSTD_seqStore_resolveOffCodes(dRep, cRep, seqStore, (U32)(seqStore->sequences - seqStore->sequencesStart));
+
+    RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall, "Block header doesn't fit");
+    cSeqsSize = ZSTD_entropyCompressSeqStore(seqStore,
+                &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
+                &zc->appliedParams,
+                op + ZSTD_blockHeaderSize, dstCapacity - ZSTD_blockHeaderSize,
+                srcSize,
+                zc->tmpWorkspace, zc->tmpWkspSize /* statically allocated in resetCCtx */,
+                zc->bmi2);
+    FORWARD_IF_ERROR(cSeqsSize, "ZSTD_entropyCompressSeqStore failed!");
+
+    if (!zc->isFirstBlock &&
+        cSeqsSize < rleMaxLength &&
+        ZSTD_isRLE((BYTE const*)src, srcSize)) {
+        /* We don't want to emit our first block as a RLE even if it qualifies because
+        * doing so will cause the decoder (cli only) to throw a "should consume all input error."
+        * This is only an issue for zstd <= v1.4.3
+        */
+        cSeqsSize = 1;
+    }
+
+    /* Sequence collection not supported when block splitting */
+    if (zc->seqCollector.collectSequences) {
+        FORWARD_IF_ERROR(ZSTD_copyBlockSequences(&zc->seqCollector, seqStore, dRepOriginal.rep), "copyBlockSequences failed");
+        ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
+        return 0;
+    }
+
+    if (cSeqsSize == 0) {
+        cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, srcSize, lastBlock);
+        FORWARD_IF_ERROR(cSize, "Nocompress block failed");
+        DEBUGLOG(5, "Writing out nocompress block, size: %zu", cSize);
+        *dRep = dRepOriginal; /* reset simulated decompression repcode history */
+    } else if (cSeqsSize == 1) {
+        cSize = ZSTD_rleCompressBlock(op, dstCapacity, *ip, srcSize, lastBlock);
+        FORWARD_IF_ERROR(cSize, "RLE compress block failed");
+        DEBUGLOG(5, "Writing out RLE block, size: %zu", cSize);
+        *dRep = dRepOriginal; /* reset simulated decompression repcode history */
+    } else {
+        ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
+        writeBlockHeader(op, cSeqsSize, srcSize, lastBlock);
+        cSize = ZSTD_blockHeaderSize + cSeqsSize;
+        DEBUGLOG(5, "Writing out compressed block, size: %zu", cSize);
+    }
+
+    if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+        zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+    return cSize;
+}
+
+/* Struct to keep track of where we are in our recursive calls. */
+typedef struct {
+    U32* splitLocations;    /* Array of split indices */
+    size_t idx;             /* The current index within splitLocations being worked on */
+} seqStoreSplits;
+
+#define MIN_SEQUENCES_BLOCK_SPLITTING 300
+
+/* Helper function to perform the recursive search for block splits.
+ * Estimates the cost of seqStore prior to split, and estimates the cost of splitting the sequences in half.
+ * If advantageous to split, then we recurse down the two sub-blocks.
+ * If not, or if an error occurred in estimation, then we do not recurse.
+ *
+ * Note: The recursion depth is capped by a heuristic minimum number of sequences,
+ * defined by MIN_SEQUENCES_BLOCK_SPLITTING.
+ * In theory, this means the absolute largest recursion depth is 10 == log2(maxNbSeqInBlock/MIN_SEQUENCES_BLOCK_SPLITTING).
+ * In practice, recursion depth usually doesn't go beyond 4.
+ *
+ * Furthermore, the number of splits is capped by ZSTD_MAX_NB_BLOCK_SPLITS.
+ * At ZSTD_MAX_NB_BLOCK_SPLITS == 196 with the current existing blockSize
+ * maximum of 128 KB, this value is actually impossible to reach.
+ */
+static void
+ZSTD_deriveBlockSplitsHelper(seqStoreSplits* splits, size_t startIdx, size_t endIdx,
+                             ZSTD_CCtx* zc, const SeqStore_t* origSeqStore)
+{
+    SeqStore_t* const fullSeqStoreChunk = &zc->blockSplitCtx.fullSeqStoreChunk;
+    SeqStore_t* const firstHalfSeqStore = &zc->blockSplitCtx.firstHalfSeqStore;
+    SeqStore_t* const secondHalfSeqStore = &zc->blockSplitCtx.secondHalfSeqStore;
+    size_t estimatedOriginalSize;
+    size_t estimatedFirstHalfSize;
+    size_t estimatedSecondHalfSize;
+    size_t midIdx = (startIdx + endIdx)/2;
+
+    DEBUGLOG(5, "ZSTD_deriveBlockSplitsHelper: startIdx=%zu endIdx=%zu", startIdx, endIdx);
+    assert(endIdx >= startIdx);
+    if (endIdx - startIdx < MIN_SEQUENCES_BLOCK_SPLITTING || splits->idx >= ZSTD_MAX_NB_BLOCK_SPLITS) {
+        DEBUGLOG(6, "ZSTD_deriveBlockSplitsHelper: Too few sequences (%zu)", endIdx - startIdx);
+        return;
+    }
+    ZSTD_deriveSeqStoreChunk(fullSeqStoreChunk, origSeqStore, startIdx, endIdx);
+    ZSTD_deriveSeqStoreChunk(firstHalfSeqStore, origSeqStore, startIdx, midIdx);
+    ZSTD_deriveSeqStoreChunk(secondHalfSeqStore, origSeqStore, midIdx, endIdx);
+    estimatedOriginalSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(fullSeqStoreChunk, zc);
+    estimatedFirstHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(firstHalfSeqStore, zc);
+    estimatedSecondHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(secondHalfSeqStore, zc);
+    DEBUGLOG(5, "Estimated original block size: %zu -- First half split: %zu -- Second half split: %zu",
+             estimatedOriginalSize, estimatedFirstHalfSize, estimatedSecondHalfSize);
+    if (ZSTD_isError(estimatedOriginalSize) || ZSTD_isError(estimatedFirstHalfSize) || ZSTD_isError(estimatedSecondHalfSize)) {
+        return;
+    }
+    if (estimatedFirstHalfSize + estimatedSecondHalfSize < estimatedOriginalSize) {
+        DEBUGLOG(5, "split decided at seqNb:%zu", midIdx);
+        ZSTD_deriveBlockSplitsHelper(splits, startIdx, midIdx, zc, origSeqStore);
+        splits->splitLocations[splits->idx] = (U32)midIdx;
+        splits->idx++;
+        ZSTD_deriveBlockSplitsHelper(splits, midIdx, endIdx, zc, origSeqStore);
+    }
+}
+
+/* Base recursive function.
+ * Populates a table with intra-block partition indices that can improve compression ratio.
+ *
+ * @return: number of splits made (which equals the size of the partition table - 1).
+ */
+static size_t ZSTD_deriveBlockSplits(ZSTD_CCtx* zc, U32 partitions[], U32 nbSeq)
+{
+    seqStoreSplits splits;
+    splits.splitLocations = partitions;
+    splits.idx = 0;
+    if (nbSeq <= 4) {
+        DEBUGLOG(5, "ZSTD_deriveBlockSplits: Too few sequences to split (%u <= 4)", nbSeq);
+        /* Refuse to try and split anything with less than 4 sequences */
+        return 0;
+    }
+    ZSTD_deriveBlockSplitsHelper(&splits, 0, nbSeq, zc, &zc->seqStore);
+    splits.splitLocations[splits.idx] = nbSeq;
+    DEBUGLOG(5, "ZSTD_deriveBlockSplits: final nb partitions: %zu", splits.idx+1);
+    return splits.idx;
+}
+
+/* ZSTD_compressBlock_splitBlock():
+ * Attempts to split a given block into multiple blocks to improve compression ratio.
+ *
+ * Returns combined size of all blocks (which includes headers), or a ZSTD error code.
+ */
+static size_t
+ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc,
+                                    void* dst, size_t dstCapacity,
+                              const void* src, size_t blockSize,
+                                    U32 lastBlock, U32 nbSeq)
+{
+    size_t cSize = 0;
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    size_t i = 0;
+    size_t srcBytesTotal = 0;
+    U32* const partitions = zc->blockSplitCtx.partitions; /* size == ZSTD_MAX_NB_BLOCK_SPLITS */
+    SeqStore_t* const nextSeqStore = &zc->blockSplitCtx.nextSeqStore;
+    SeqStore_t* const currSeqStore = &zc->blockSplitCtx.currSeqStore;
+    size_t const numSplits = ZSTD_deriveBlockSplits(zc, partitions, nbSeq);
+
+    /* If a block is split and some partitions are emitted as RLE/uncompressed, then repcode history
+     * may become invalid. In order to reconcile potentially invalid repcodes, we keep track of two
+     * separate repcode histories that simulate repcode history on compression and decompression side,
+     * and use the histories to determine whether we must replace a particular repcode with its raw offset.
+     *
+     * 1) cRep gets updated for each partition, regardless of whether the block was emitted as uncompressed
+     *    or RLE. This allows us to retrieve the offset value that an invalid repcode references within
+     *    a nocompress/RLE block.
+     * 2) dRep gets updated only for compressed partitions, and when a repcode gets replaced, will use
+     *    the replacement offset value rather than the original repcode to update the repcode history.
+     *    dRep also will be the final repcode history sent to the next block.
+     *
+     * See ZSTD_seqStore_resolveOffCodes() for more details.
+     */
+    Repcodes_t dRep;
+    Repcodes_t cRep;
+    ZSTD_memcpy(dRep.rep, zc->blockState.prevCBlock->rep, sizeof(Repcodes_t));
+    ZSTD_memcpy(cRep.rep, zc->blockState.prevCBlock->rep, sizeof(Repcodes_t));
+    ZSTD_memset(nextSeqStore, 0, sizeof(SeqStore_t));
+
+    DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
+                (unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit,
+                (unsigned)zc->blockState.matchState.nextToUpdate);
+
+    if (numSplits == 0) {
+        size_t cSizeSingleBlock =
+            ZSTD_compressSeqStore_singleBlock(zc, &zc->seqStore,
+                                            &dRep, &cRep,
+                                            op, dstCapacity,
+                                            ip, blockSize,
+                                            lastBlock, 0 /* isPartition */);
+        FORWARD_IF_ERROR(cSizeSingleBlock, "Compressing single block from splitBlock_internal() failed!");
+        DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal: No splits");
+        assert(zc->blockSizeMax <= ZSTD_BLOCKSIZE_MAX);
+        assert(cSizeSingleBlock <= zc->blockSizeMax + ZSTD_blockHeaderSize);
+        return cSizeSingleBlock;
+    }
+
+    ZSTD_deriveSeqStoreChunk(currSeqStore, &zc->seqStore, 0, partitions[0]);
+    for (i = 0; i <= numSplits; ++i) {
+        size_t cSizeChunk;
+        U32 const lastPartition = (i == numSplits);
+        U32 lastBlockEntireSrc = 0;
+
+        size_t srcBytes = ZSTD_countSeqStoreLiteralsBytes(currSeqStore) + ZSTD_countSeqStoreMatchBytes(currSeqStore);
+        srcBytesTotal += srcBytes;
+        if (lastPartition) {
+            /* This is the final partition, need to account for possible last literals */
+            srcBytes += blockSize - srcBytesTotal;
+            lastBlockEntireSrc = lastBlock;
+        } else {
+            ZSTD_deriveSeqStoreChunk(nextSeqStore, &zc->seqStore, partitions[i], partitions[i+1]);
+        }
+
+        cSizeChunk = ZSTD_compressSeqStore_singleBlock(zc, currSeqStore,
+                                                      &dRep, &cRep,
+                                                       op, dstCapacity,
+                                                       ip, srcBytes,
+                                                       lastBlockEntireSrc, 1 /* isPartition */);
+        DEBUGLOG(5, "Estimated size: %zu vs %zu : actual size",
+                    ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(currSeqStore, zc), cSizeChunk);
+        FORWARD_IF_ERROR(cSizeChunk, "Compressing chunk failed!");
+
+        ip += srcBytes;
+        op += cSizeChunk;
+        dstCapacity -= cSizeChunk;
+        cSize += cSizeChunk;
+        *currSeqStore = *nextSeqStore;
+        assert(cSizeChunk <= zc->blockSizeMax + ZSTD_blockHeaderSize);
+    }
+    /* cRep and dRep may have diverged during the compression.
+     * If so, we use the dRep repcodes for the next block.
+     */
+    ZSTD_memcpy(zc->blockState.prevCBlock->rep, dRep.rep, sizeof(Repcodes_t));
+    return cSize;
+}
+
+static size_t
+ZSTD_compressBlock_splitBlock(ZSTD_CCtx* zc,
+                              void* dst, size_t dstCapacity,
+                              const void* src, size_t srcSize, U32 lastBlock)
+{
+    U32 nbSeq;
+    size_t cSize;
+    DEBUGLOG(5, "ZSTD_compressBlock_splitBlock");
+    assert(zc->appliedParams.postBlockSplitter == ZSTD_ps_enable);
+
+    {   const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
+        FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
+        if (bss == ZSTDbss_noCompress) {
+            if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+                zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+            RETURN_ERROR_IF(zc->seqCollector.collectSequences, sequenceProducer_failed, "Uncompressible block");
+            cSize = ZSTD_noCompressBlock(dst, dstCapacity, src, srcSize, lastBlock);
+            FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
+            DEBUGLOG(5, "ZSTD_compressBlock_splitBlock: Nocompress block");
+            return cSize;
+        }
+        nbSeq = (U32)(zc->seqStore.sequences - zc->seqStore.sequencesStart);
+    }
+
+    cSize = ZSTD_compressBlock_splitBlock_internal(zc, dst, dstCapacity, src, srcSize, lastBlock, nbSeq);
+    FORWARD_IF_ERROR(cSize, "Splitting blocks failed!");
+    return cSize;
+}
+
+static size_t
+ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
+                            void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize, U32 frame)
+{
+    /* This is an estimated upper bound for the length of an rle block.
+     * This isn't the actual upper bound.
+     * Finding the real threshold needs further investigation.
+     */
+    const U32 rleMaxLength = 25;
+    size_t cSize;
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
+                (unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit,
+                (unsigned)zc->blockState.matchState.nextToUpdate);
+
+    {   const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
+        FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
+        if (bss == ZSTDbss_noCompress) {
+            RETURN_ERROR_IF(zc->seqCollector.collectSequences, sequenceProducer_failed, "Uncompressible block");
+            cSize = 0;
+            goto out;
+        }
+    }
+
+    if (zc->seqCollector.collectSequences) {
+        FORWARD_IF_ERROR(ZSTD_copyBlockSequences(&zc->seqCollector, ZSTD_getSeqStore(zc), zc->blockState.prevCBlock->rep), "copyBlockSequences failed");
+        ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
+        return 0;
+    }
+
+    /* encode sequences and literals */
+    cSize = ZSTD_entropyCompressSeqStore(&zc->seqStore,
+            &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
+            &zc->appliedParams,
+            dst, dstCapacity,
+            srcSize,
+            zc->tmpWorkspace, zc->tmpWkspSize /* statically allocated in resetCCtx */,
+            zc->bmi2);
+
+    if (frame &&
+        /* We don't want to emit our first block as a RLE even if it qualifies because
+         * doing so will cause the decoder (cli only) to throw a "should consume all input error."
+         * This is only an issue for zstd <= v1.4.3
+         */
+        !zc->isFirstBlock &&
+        cSize < rleMaxLength &&
+        ZSTD_isRLE(ip, srcSize))
+    {
+        cSize = 1;
+        op[0] = ip[0];
+    }
+
+out:
+    if (!ZSTD_isError(cSize) && cSize > 1) {
+        ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
+    }
+    /* We check that dictionaries have offset codes available for the first
+     * block. After the first block, the offcode table might not have large
+     * enough codes to represent the offsets in the data.
+     */
+    if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+        zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+    return cSize;
+}
+
+static size_t ZSTD_compressBlock_targetCBlockSize_body(ZSTD_CCtx* zc,
+                               void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                               const size_t bss, U32 lastBlock)
+{
+    DEBUGLOG(6, "Attempting ZSTD_compressSuperBlock()");
+    if (bss == ZSTDbss_compress) {
+        if (/* We don't want to emit our first block as a RLE even if it qualifies because
+            * doing so will cause the decoder (cli only) to throw a "should consume all input error."
+            * This is only an issue for zstd <= v1.4.3
+            */
+            !zc->isFirstBlock &&
+            ZSTD_maybeRLE(&zc->seqStore) &&
+            ZSTD_isRLE((BYTE const*)src, srcSize))
+        {
+            return ZSTD_rleCompressBlock(dst, dstCapacity, *(BYTE const*)src, srcSize, lastBlock);
+        }
+        /* Attempt superblock compression.
+         *
+         * Note that compressed size of ZSTD_compressSuperBlock() is not bound by the
+         * standard ZSTD_compressBound(). This is a problem, because even if we have
+         * space now, taking an extra byte now could cause us to run out of space later
+         * and violate ZSTD_compressBound().
+         *
+         * Define blockBound(blockSize) = blockSize + ZSTD_blockHeaderSize.
+         *
+         * In order to respect ZSTD_compressBound() we must attempt to emit a raw
+         * uncompressed block in these cases:
+         *   * cSize == 0: Return code for an uncompressed block.
+         *   * cSize == dstSize_tooSmall: We may have expanded beyond blockBound(srcSize).
+         *     ZSTD_noCompressBlock() will return dstSize_tooSmall if we are really out of
+         *     output space.
+         *   * cSize >= blockBound(srcSize): We have expanded the block too much so
+         *     emit an uncompressed block.
+         */
+        {   size_t const cSize =
+                ZSTD_compressSuperBlock(zc, dst, dstCapacity, src, srcSize, lastBlock);
+            if (cSize != ERROR(dstSize_tooSmall)) {
+                size_t const maxCSize =
+                    srcSize - ZSTD_minGain(srcSize, zc->appliedParams.cParams.strategy);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressSuperBlock failed");
+                if (cSize != 0 && cSize < maxCSize + ZSTD_blockHeaderSize) {
+                    ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
+                    return cSize;
+                }
+            }
+        }
+    } /* if (bss == ZSTDbss_compress)*/
+
+    DEBUGLOG(6, "Resorting to ZSTD_noCompressBlock()");
+    /* Superblock compression failed, attempt to emit a single no compress block.
+     * The decoder will be able to stream this block since it is uncompressed.
+     */
+    return ZSTD_noCompressBlock(dst, dstCapacity, src, srcSize, lastBlock);
+}
+
+static size_t ZSTD_compressBlock_targetCBlockSize(ZSTD_CCtx* zc,
+                               void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                               U32 lastBlock)
+{
+    size_t cSize = 0;
+    const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
+    DEBUGLOG(5, "ZSTD_compressBlock_targetCBlockSize (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u, srcSize=%zu)",
+                (unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit, (unsigned)zc->blockState.matchState.nextToUpdate, srcSize);
+    FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
+
+    cSize = ZSTD_compressBlock_targetCBlockSize_body(zc, dst, dstCapacity, src, srcSize, bss, lastBlock);
+    FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_targetCBlockSize_body failed");
+
+    if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+        zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+    return cSize;
+}
+
+static void ZSTD_overflowCorrectIfNeeded(ZSTD_MatchState_t* ms,
+                                         ZSTD_cwksp* ws,
+                                         ZSTD_CCtx_params const* params,
+                                         void const* ip,
+                                         void const* iend)
+{
+    U32 const cycleLog = ZSTD_cycleLog(params->cParams.chainLog, params->cParams.strategy);
+    U32 const maxDist = (U32)1 << params->cParams.windowLog;
+    if (ZSTD_window_needOverflowCorrection(ms->window, cycleLog, maxDist, ms->loadedDictEnd, ip, iend)) {
+        U32 const correction = ZSTD_window_correctOverflow(&ms->window, cycleLog, maxDist, ip);
+        ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
+        ZSTD_cwksp_mark_tables_dirty(ws);
+        ZSTD_reduceIndex(ms, params, correction);
+        ZSTD_cwksp_mark_tables_clean(ws);
+        if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
+        else ms->nextToUpdate -= correction;
+        /* invalidate dictionaries on overflow correction */
+        ms->loadedDictEnd = 0;
+        ms->dictMatchState = NULL;
+    }
+}
+
+#include "zstd_preSplit.h"
+
+static size_t ZSTD_optimalBlockSize(ZSTD_CCtx* cctx, const void* src, size_t srcSize, size_t blockSizeMax, int splitLevel, ZSTD_strategy strat, S64 savings)
+{
+    /* split level based on compression strategy, from `fast` to `btultra2` */
+    static const int splitLevels[] = { 0, 0, 1, 2, 2, 3, 3, 4, 4, 4 };
+    /* note: conservatively only split full blocks (128 KB) currently.
+     * While it's possible to go lower, let's keep it simple for a first implementation.
+     * Besides, benefits of splitting are reduced when blocks are already small.
+     */
+    if (srcSize < 128 KB || blockSizeMax < 128 KB)
+        return MIN(srcSize, blockSizeMax);
+    /* do not split incompressible data though:
+     * require verified savings to allow pre-splitting.
+     * Note: as a consequence, the first full block is not split.
+     */
+    if (savings < 3) {
+        DEBUGLOG(6, "don't attempt splitting: savings (%i) too low", (int)savings);
+        return 128 KB;
+    }
+    /* apply @splitLevel, or use default value (which depends on @strat).
+     * note that splitting heuristic is still conditioned by @savings >= 3,
+     * so the first block will not reach this code path */
+    if (splitLevel == 1) return 128 KB;
+    if (splitLevel == 0) {
+        assert(ZSTD_fast <= strat && strat <= ZSTD_btultra2);
+        splitLevel = splitLevels[strat];
+    } else {
+        assert(2 <= splitLevel && splitLevel <= 6);
+        splitLevel -= 2;
+    }
+    return ZSTD_splitBlock(src, blockSizeMax, splitLevel, cctx->tmpWorkspace, cctx->tmpWkspSize);
+}
+
+/*! ZSTD_compress_frameChunk() :
+*   Compress a chunk of data into one or multiple blocks.
+*   All blocks will be terminated, all input will be consumed.
+*   Function will issue an error if there is not enough `dstCapacity` to hold the compressed content.
+*   Frame is supposed already started (header already produced)
+*  @return : compressed size, or an error code
+*/
+static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
+                                     void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                                     U32 lastFrameChunk)
+{
+    size_t blockSizeMax = cctx->blockSizeMax;
+    size_t remaining = srcSize;
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    U32 const maxDist = (U32)1 << cctx->appliedParams.cParams.windowLog;
+    S64 savings = (S64)cctx->consumedSrcSize - (S64)cctx->producedCSize;
+
+    assert(cctx->appliedParams.cParams.windowLog <= ZSTD_WINDOWLOG_MAX);
+
+    DEBUGLOG(5, "ZSTD_compress_frameChunk (srcSize=%u, blockSizeMax=%u)", (unsigned)srcSize, (unsigned)blockSizeMax);
+    if (cctx->appliedParams.fParams.checksumFlag && srcSize)
+        XXH64_update(&cctx->xxhState, src, srcSize);
+
+    while (remaining) {
+        ZSTD_MatchState_t* const ms = &cctx->blockState.matchState;
+        size_t const blockSize = ZSTD_optimalBlockSize(cctx,
+                                ip, remaining,
+                                blockSizeMax,
+                                cctx->appliedParams.preBlockSplitter_level,
+                                cctx->appliedParams.cParams.strategy,
+                                savings);
+        U32 const lastBlock = lastFrameChunk & (blockSize == remaining);
+        assert(blockSize <= remaining);
+
+        /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+         * additional 1. We need to revisit and change this logic to be more consistent */
+        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE + 1,
+                        dstSize_tooSmall,
+                        "not enough space to store compressed block");
+
+        ZSTD_overflowCorrectIfNeeded(
+            ms, &cctx->workspace, &cctx->appliedParams, ip, ip + blockSize);
+        ZSTD_checkDictValidity(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
+        ZSTD_window_enforceMaxDist(&ms->window, ip, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
+
+        /* Ensure hash/chain table insertion resumes no sooner than lowlimit */
+        if (ms->nextToUpdate < ms->window.lowLimit) ms->nextToUpdate = ms->window.lowLimit;
+
+        {   size_t cSize;
+            if (ZSTD_useTargetCBlockSize(&cctx->appliedParams)) {
+                cSize = ZSTD_compressBlock_targetCBlockSize(cctx, op, dstCapacity, ip, blockSize, lastBlock);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_targetCBlockSize failed");
+                assert(cSize > 0);
+                assert(cSize <= blockSize + ZSTD_blockHeaderSize);
+            } else if (ZSTD_blockSplitterEnabled(&cctx->appliedParams)) {
+                cSize = ZSTD_compressBlock_splitBlock(cctx, op, dstCapacity, ip, blockSize, lastBlock);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_splitBlock failed");
+                assert(cSize > 0 || cctx->seqCollector.collectSequences == 1);
+            } else {
+                cSize = ZSTD_compressBlock_internal(cctx,
+                                        op+ZSTD_blockHeaderSize, dstCapacity-ZSTD_blockHeaderSize,
+                                        ip, blockSize, 1 /* frame */);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_internal failed");
+
+                if (cSize == 0) {  /* block is not compressible */
+                    cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
+                    FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
+                } else {
+                    U32 const cBlockHeader = cSize == 1 ?
+                        lastBlock + (((U32)bt_rle)<<1) + (U32)(blockSize << 3) :
+                        lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
+                    MEM_writeLE24(op, cBlockHeader);
+                    cSize += ZSTD_blockHeaderSize;
+                }
+            }  /* if (ZSTD_useTargetCBlockSize(&cctx->appliedParams))*/
+
+            /* @savings is employed to ensure that splitting doesn't worsen expansion of incompressible data.
+             * Without splitting, the maximum expansion is 3 bytes per full block.
+             * An adversarial input could attempt to fudge the split detector,
+             * and make it split incompressible data, resulting in more block headers.
+             * Note that, since ZSTD_COMPRESSBOUND() assumes a worst case scenario of 1KB per block,
+             * and the splitter never creates blocks that small (current lower limit is 8 KB),
+             * there is already no risk to expand beyond ZSTD_COMPRESSBOUND() limit.
+             * But if the goal is to not expand by more than 3-bytes per 128 KB full block,
+             * then yes, it becomes possible to make the block splitter oversplit incompressible data.
+             * Using @savings, we enforce an even more conservative condition,
+             * requiring the presence of enough savings (at least 3 bytes) to authorize splitting,
+             * otherwise only full blocks are used.
+             * But being conservative is fine,
+             * since splitting barely compressible blocks is not fruitful anyway */
+            savings += (S64)blockSize - (S64)cSize;
+
+            ip += blockSize;
+            assert(remaining >= blockSize);
+            remaining -= blockSize;
+            op += cSize;
+            assert(dstCapacity >= cSize);
+            dstCapacity -= cSize;
+            cctx->isFirstBlock = 0;
+            DEBUGLOG(5, "ZSTD_compress_frameChunk: adding a block of size %u",
+                        (unsigned)cSize);
+    }   }
+
+    if (lastFrameChunk && (op>ostart)) cctx->stage = ZSTDcs_ending;
+    return (size_t)(op-ostart);
+}
+
+
+static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
+                                    const ZSTD_CCtx_params* params,
+                                    U64 pledgedSrcSize, U32 dictID)
+{
+    BYTE* const op = (BYTE*)dst;
+    U32   const dictIDSizeCodeLength = (dictID>0) + (dictID>=256) + (dictID>=65536);   /* 0-3 */
+    U32   const dictIDSizeCode = params->fParams.noDictIDFlag ? 0 : dictIDSizeCodeLength;   /* 0-3 */
+    U32   const checksumFlag = params->fParams.checksumFlag>0;
+    U32   const windowSize = (U32)1 << params->cParams.windowLog;
+    U32   const singleSegment = params->fParams.contentSizeFlag && (windowSize >= pledgedSrcSize);
+    BYTE  const windowLogByte = (BYTE)((params->cParams.windowLog - ZSTD_WINDOWLOG_ABSOLUTEMIN) << 3);
+    U32   const fcsCode = params->fParams.contentSizeFlag ?
+                     (pledgedSrcSize>=256) + (pledgedSrcSize>=65536+256) + (pledgedSrcSize>=0xFFFFFFFFU) : 0;  /* 0-3 */
+    BYTE  const frameHeaderDescriptionByte = (BYTE)(dictIDSizeCode + (checksumFlag<<2) + (singleSegment<<5) + (fcsCode<<6) );
+    size_t pos=0;
+
+    assert(!(params->fParams.contentSizeFlag && pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN));
+    RETURN_ERROR_IF(dstCapacity < ZSTD_FRAMEHEADERSIZE_MAX, dstSize_tooSmall,
+                    "dst buf is too small to fit worst-case frame header size.");
+    DEBUGLOG(4, "ZSTD_writeFrameHeader : dictIDFlag : %u ; dictID : %u ; dictIDSizeCode : %u",
+                !params->fParams.noDictIDFlag, (unsigned)dictID, (unsigned)dictIDSizeCode);
+    if (params->format == ZSTD_f_zstd1) {
+        MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
+        pos = 4;
+    }
+    op[pos++] = frameHeaderDescriptionByte;
+    if (!singleSegment) op[pos++] = windowLogByte;
+    switch(dictIDSizeCode)
+    {
+        default:
+            assert(0); /* impossible */
+            ZSTD_FALLTHROUGH;
+        case 0 : break;
+        case 1 : op[pos] = (BYTE)(dictID); pos++; break;
+        case 2 : MEM_writeLE16(op+pos, (U16)dictID); pos+=2; break;
+        case 3 : MEM_writeLE32(op+pos, dictID); pos+=4; break;
+    }
+    switch(fcsCode)
+    {
+        default:
+            assert(0); /* impossible */
+            ZSTD_FALLTHROUGH;
+        case 0 : if (singleSegment) op[pos++] = (BYTE)(pledgedSrcSize); break;
+        case 1 : MEM_writeLE16(op+pos, (U16)(pledgedSrcSize-256)); pos+=2; break;
+        case 2 : MEM_writeLE32(op+pos, (U32)(pledgedSrcSize)); pos+=4; break;
+        case 3 : MEM_writeLE64(op+pos, (U64)(pledgedSrcSize)); pos+=8; break;
+    }
+    return pos;
+}
+
+/* ZSTD_writeSkippableFrame_advanced() :
+ * Writes out a skippable frame with the specified magic number variant (16 are supported),
+ * from ZSTD_MAGIC_SKIPPABLE_START to ZSTD_MAGIC_SKIPPABLE_START+15, and the desired source data.
+ *
+ * Returns the total number of bytes written, or a ZSTD error code.
+ */
+size_t ZSTD_writeSkippableFrame(void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize, unsigned magicVariant) {
+    BYTE* op = (BYTE*)dst;
+    RETURN_ERROR_IF(dstCapacity < srcSize + ZSTD_SKIPPABLEHEADERSIZE /* Skippable frame overhead */,
+                    dstSize_tooSmall, "Not enough room for skippable frame");
+    RETURN_ERROR_IF(srcSize > (unsigned)0xFFFFFFFF, srcSize_wrong, "Src size too large for skippable frame");
+    RETURN_ERROR_IF(magicVariant > 15, parameter_outOfBound, "Skippable frame magic number variant not supported");
+
+    MEM_writeLE32(op, (U32)(ZSTD_MAGIC_SKIPPABLE_START + magicVariant));
+    MEM_writeLE32(op+4, (U32)srcSize);
+    ZSTD_memcpy(op+8, src, srcSize);
+    return srcSize + ZSTD_SKIPPABLEHEADERSIZE;
+}
+
+/* ZSTD_writeLastEmptyBlock() :
+ * output an empty Block with end-of-frame mark to complete a frame
+ * @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
+ *           or an error code if `dstCapacity` is too small (<ZSTD_blockHeaderSize)
+ */
+size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity)
+{
+    RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall,
+                    "dst buf is too small to write frame trailer empty block.");
+    {   U32 const cBlockHeader24 = 1 /*lastBlock*/ + (((U32)bt_raw)<<1);  /* 0 size */
+        MEM_writeLE24(dst, cBlockHeader24);
+        return ZSTD_blockHeaderSize;
+    }
+}
+
+void ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq)
+{
+    assert(cctx->stage == ZSTDcs_init);
+    assert(nbSeq == 0 || cctx->appliedParams.ldmParams.enableLdm != ZSTD_ps_enable);
+    cctx->externSeqStore.seq = seq;
+    cctx->externSeqStore.size = nbSeq;
+    cctx->externSeqStore.capacity = nbSeq;
+    cctx->externSeqStore.pos = 0;
+    cctx->externSeqStore.posInSequence = 0;
+}
+
+
+static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize,
+                               U32 frame, U32 lastFrameChunk)
+{
+    ZSTD_MatchState_t* const ms = &cctx->blockState.matchState;
+    size_t fhSize = 0;
+
+    DEBUGLOG(5, "ZSTD_compressContinue_internal, stage: %u, srcSize: %u",
+                cctx->stage, (unsigned)srcSize);
+    RETURN_ERROR_IF(cctx->stage==ZSTDcs_created, stage_wrong,
+                    "missing init (ZSTD_compressBegin)");
+
+    if (frame && (cctx->stage==ZSTDcs_init)) {
+        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, &cctx->appliedParams,
+                                       cctx->pledgedSrcSizePlusOne-1, cctx->dictID);
+        FORWARD_IF_ERROR(fhSize, "ZSTD_writeFrameHeader failed");
+        assert(fhSize <= dstCapacity);
+        dstCapacity -= fhSize;
+        dst = (char*)dst + fhSize;
+        cctx->stage = ZSTDcs_ongoing;
+    }
+
+    if (!srcSize) return fhSize;  /* do not generate an empty block if no input */
+
+    if (!ZSTD_window_update(&ms->window, src, srcSize, ms->forceNonContiguous)) {
+        ms->forceNonContiguous = 0;
+        ms->nextToUpdate = ms->window.dictLimit;
+    }
+    if (cctx->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable) {
+        ZSTD_window_update(&cctx->ldmState.window, src, srcSize, /* forceNonContiguous */ 0);
+    }
+
+    if (!frame) {
+        /* overflow check and correction for block mode */
+        ZSTD_overflowCorrectIfNeeded(
+            ms, &cctx->workspace, &cctx->appliedParams,
+            src, (BYTE const*)src + srcSize);
+    }
+
+    DEBUGLOG(5, "ZSTD_compressContinue_internal (blockSize=%u)", (unsigned)cctx->blockSizeMax);
+    {   size_t const cSize = frame ?
+                             ZSTD_compress_frameChunk (cctx, dst, dstCapacity, src, srcSize, lastFrameChunk) :
+                             ZSTD_compressBlock_internal (cctx, dst, dstCapacity, src, srcSize, 0 /* frame */);
+        FORWARD_IF_ERROR(cSize, "%s", frame ? "ZSTD_compress_frameChunk failed" : "ZSTD_compressBlock_internal failed");
+        cctx->consumedSrcSize += srcSize;
+        cctx->producedCSize += (cSize + fhSize);
+        assert(!(cctx->appliedParams.fParams.contentSizeFlag && cctx->pledgedSrcSizePlusOne == 0));
+        if (cctx->pledgedSrcSizePlusOne != 0) {  /* control src size */
+            ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN == (unsigned long long)-1);
+            RETURN_ERROR_IF(
+                cctx->consumedSrcSize+1 > cctx->pledgedSrcSizePlusOne,
+                srcSize_wrong,
+                "error : pledgedSrcSize = %u, while realSrcSize >= %u",
+                (unsigned)cctx->pledgedSrcSizePlusOne-1,
+                (unsigned)cctx->consumedSrcSize);
+        }
+        return cSize + fhSize;
+    }
+}
+
+size_t ZSTD_compressContinue_public(ZSTD_CCtx* cctx,
+                                        void* dst, size_t dstCapacity,
+                                  const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressContinue (srcSize=%u)", (unsigned)srcSize);
+    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1 /* frame mode */, 0 /* last chunk */);
+}
+
+/* NOTE: Must just wrap ZSTD_compressContinue_public() */
+size_t ZSTD_compressContinue(ZSTD_CCtx* cctx,
+                             void* dst, size_t dstCapacity,
+                       const void* src, size_t srcSize)
+{
+    return ZSTD_compressContinue_public(cctx, dst, dstCapacity, src, srcSize);
+}
+
+static size_t ZSTD_getBlockSize_deprecated(const ZSTD_CCtx* cctx)
+{
+    ZSTD_compressionParameters const cParams = cctx->appliedParams.cParams;
+    assert(!ZSTD_checkCParams(cParams));
+    return MIN(cctx->appliedParams.maxBlockSize, (size_t)1 << cParams.windowLog);
+}
+
+/* NOTE: Must just wrap ZSTD_getBlockSize_deprecated() */
+size_t ZSTD_getBlockSize(const ZSTD_CCtx* cctx)
+{
+    return ZSTD_getBlockSize_deprecated(cctx);
+}
+
+/* NOTE: Must just wrap ZSTD_compressBlock_deprecated() */
+size_t ZSTD_compressBlock_deprecated(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressBlock: srcSize = %u", (unsigned)srcSize);
+    { size_t const blockSizeMax = ZSTD_getBlockSize_deprecated(cctx);
+      RETURN_ERROR_IF(srcSize > blockSizeMax, srcSize_wrong, "input is larger than a block"); }
+
+    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 0 /* frame mode */, 0 /* last chunk */);
+}
+
+/* NOTE: Must just wrap ZSTD_compressBlock_deprecated() */
+size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_deprecated(cctx, dst, dstCapacity, src, srcSize);
+}
+
+/*! ZSTD_loadDictionaryContent() :
+ *  @return : 0, or an error code
+ */
+static size_t
+ZSTD_loadDictionaryContent(ZSTD_MatchState_t* ms,
+                        ldmState_t* ls,
+                        ZSTD_cwksp* ws,
+                        ZSTD_CCtx_params const* params,
+                        const void* src, size_t srcSize,
+                        ZSTD_dictTableLoadMethod_e dtlm,
+                        ZSTD_tableFillPurpose_e tfp)
+{
+    const BYTE* ip = (const BYTE*) src;
+    const BYTE* const iend = ip + srcSize;
+    int const loadLdmDict = params->ldmParams.enableLdm == ZSTD_ps_enable && ls != NULL;
+
+    /* Assert that the ms params match the params we're being given */
+    ZSTD_assertEqualCParams(params->cParams, ms->cParams);
+
+    {   /* Ensure large dictionaries can't cause index overflow */
+
+        /* Allow the dictionary to set indices up to exactly ZSTD_CURRENT_MAX.
+         * Dictionaries right at the edge will immediately trigger overflow
+         * correction, but I don't want to insert extra constraints here.
+         */
+        U32 maxDictSize = ZSTD_CURRENT_MAX - ZSTD_WINDOW_START_INDEX;
+
+        int const CDictTaggedIndices = ZSTD_CDictIndicesAreTagged(&params->cParams);
+        if (CDictTaggedIndices && tfp == ZSTD_tfp_forCDict) {
+            /* Some dictionary matchfinders in zstd use "short cache",
+             * which treats the lower ZSTD_SHORT_CACHE_TAG_BITS of each
+             * CDict hashtable entry as a tag rather than as part of an index.
+             * When short cache is used, we need to truncate the dictionary
+             * so that its indices don't overlap with the tag. */
+            U32 const shortCacheMaxDictSize = (1u << (32 - ZSTD_SHORT_CACHE_TAG_BITS)) - ZSTD_WINDOW_START_INDEX;
+            maxDictSize = MIN(maxDictSize, shortCacheMaxDictSize);
+            assert(!loadLdmDict);
+        }
+
+        /* If the dictionary is too large, only load the suffix of the dictionary. */
+        if (srcSize > maxDictSize) {
+            ip = iend - maxDictSize;
+            src = ip;
+            srcSize = maxDictSize;
+        }
+    }
+
+    if (srcSize > ZSTD_CHUNKSIZE_MAX) {
+        /* We must have cleared our windows when our source is this large. */
+        assert(ZSTD_window_isEmpty(ms->window));
+        if (loadLdmDict) assert(ZSTD_window_isEmpty(ls->window));
+    }
+    ZSTD_window_update(&ms->window, src, srcSize, /* forceNonContiguous */ 0);
+
+    DEBUGLOG(4, "ZSTD_loadDictionaryContent: useRowMatchFinder=%d", (int)params->useRowMatchFinder);
+
+    if (loadLdmDict) { /* Load the entire dict into LDM matchfinders. */
+        DEBUGLOG(4, "ZSTD_loadDictionaryContent: Trigger loadLdmDict");
+        ZSTD_window_update(&ls->window, src, srcSize, /* forceNonContiguous */ 0);
+        ls->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ls->window.base);
+        ZSTD_ldm_fillHashTable(ls, ip, iend, &params->ldmParams);
+        DEBUGLOG(4, "ZSTD_loadDictionaryContent: ZSTD_ldm_fillHashTable completes");
+    }
+
+    /* If the dict is larger than we can reasonably index in our tables, only load the suffix. */
+    {   U32 maxDictSize = 1U << MIN(MAX(params->cParams.hashLog + 3, params->cParams.chainLog + 1), 31);
+        if (srcSize > maxDictSize) {
+            ip = iend - maxDictSize;
+            src = ip;
+            srcSize = maxDictSize;
+        }
+    }
+
+    ms->nextToUpdate = (U32)(ip - ms->window.base);
+    ms->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ms->window.base);
+    ms->forceNonContiguous = params->deterministicRefPrefix;
+
+    if (srcSize <= HASH_READ_SIZE) return 0;
+
+    ZSTD_overflowCorrectIfNeeded(ms, ws, params, ip, iend);
+
+    switch(params->cParams.strategy)
+    {
+    case ZSTD_fast:
+        ZSTD_fillHashTable(ms, iend, dtlm, tfp);
+        break;
+    case ZSTD_dfast:
+#ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
+        ZSTD_fillDoubleHashTable(ms, iend, dtlm, tfp);
+#else
+        assert(0); /* shouldn't be called: cparams should've been adjusted. */
+#endif
+        break;
+
+    case ZSTD_greedy:
+    case ZSTD_lazy:
+    case ZSTD_lazy2:
+#if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR)
+        assert(srcSize >= HASH_READ_SIZE);
+        if (ms->dedicatedDictSearch) {
+            assert(ms->chainTable != NULL);
+            ZSTD_dedicatedDictSearch_lazy_loadDictionary(ms, iend-HASH_READ_SIZE);
+        } else {
+            assert(params->useRowMatchFinder != ZSTD_ps_auto);
+            if (params->useRowMatchFinder == ZSTD_ps_enable) {
+                size_t const tagTableSize = ((size_t)1 << params->cParams.hashLog);
+                ZSTD_memset(ms->tagTable, 0, tagTableSize);
+                ZSTD_row_update(ms, iend-HASH_READ_SIZE);
+                DEBUGLOG(4, "Using row-based hash table for lazy dict");
+            } else {
+                ZSTD_insertAndFindFirstIndex(ms, iend-HASH_READ_SIZE);
+                DEBUGLOG(4, "Using chain-based hash table for lazy dict");
+            }
+        }
+#else
+        assert(0); /* shouldn't be called: cparams should've been adjusted. */
+#endif
+        break;
+
+    case ZSTD_btlazy2:   /* we want the dictionary table fully sorted */
+    case ZSTD_btopt:
+    case ZSTD_btultra:
+    case ZSTD_btultra2:
+#if !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR)
+        assert(srcSize >= HASH_READ_SIZE);
+        DEBUGLOG(4, "Fill %u bytes into the Binary Tree", (unsigned)srcSize);
+        ZSTD_updateTree(ms, iend-HASH_READ_SIZE, iend);
+#else
+        assert(0); /* shouldn't be called: cparams should've been adjusted. */
+#endif
+        break;
+
+    default:
+        assert(0);  /* not possible : not a valid strategy id */
+    }
+
+    ms->nextToUpdate = (U32)(iend - ms->window.base);
+    return 0;
+}
+
+
+/* Dictionaries that assign zero probability to symbols that show up causes problems
+ * when FSE encoding. Mark dictionaries with zero probability symbols as FSE_repeat_check
+ * and only dictionaries with 100% valid symbols can be assumed valid.
+ */
+static FSE_repeat ZSTD_dictNCountRepeat(short* normalizedCounter, unsigned dictMaxSymbolValue, unsigned maxSymbolValue)
+{
+    U32 s;
+    if (dictMaxSymbolValue < maxSymbolValue) {
+        return FSE_repeat_check;
+    }
+    for (s = 0; s <= maxSymbolValue; ++s) {
+        if (normalizedCounter[s] == 0) {
+            return FSE_repeat_check;
+        }
+    }
+    return FSE_repeat_valid;
+}
+
+size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
+                         const void* const dict, size_t dictSize)
+{
+    short offcodeNCount[MaxOff+1];
+    unsigned offcodeMaxValue = MaxOff;
+    const BYTE* dictPtr = (const BYTE*)dict;    /* skip magic num and dict ID */
+    const BYTE* const dictEnd = dictPtr + dictSize;
+    dictPtr += 8;
+    bs->entropy.huf.repeatMode = HUF_repeat_check;
+
+    {   unsigned maxSymbolValue = 255;
+        unsigned hasZeroWeights = 1;
+        size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)bs->entropy.huf.CTable, &maxSymbolValue, dictPtr,
+            (size_t)(dictEnd-dictPtr), &hasZeroWeights);
+
+        /* We only set the loaded table as valid if it contains all non-zero
+         * weights. Otherwise, we set it to check */
+        if (!hasZeroWeights && maxSymbolValue == 255)
+            bs->entropy.huf.repeatMode = HUF_repeat_valid;
+
+        RETURN_ERROR_IF(HUF_isError(hufHeaderSize), dictionary_corrupted, "");
+        dictPtr += hufHeaderSize;
+    }
+
+    {   unsigned offcodeLog;
+        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted, "");
+        /* fill all offset symbols to avoid garbage at end of table */
+        RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
+                bs->entropy.fse.offcodeCTable,
+                offcodeNCount, MaxOff, offcodeLog,
+                workspace, HUF_WORKSPACE_SIZE)),
+            dictionary_corrupted, "");
+        /* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
+        dictPtr += offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted, "");
+        RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
+                bs->entropy.fse.matchlengthCTable,
+                matchlengthNCount, matchlengthMaxValue, matchlengthLog,
+                workspace, HUF_WORKSPACE_SIZE)),
+            dictionary_corrupted, "");
+        bs->entropy.fse.matchlength_repeatMode = ZSTD_dictNCountRepeat(matchlengthNCount, matchlengthMaxValue, MaxML);
+        dictPtr += matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted, "");
+        RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
+                bs->entropy.fse.litlengthCTable,
+                litlengthNCount, litlengthMaxValue, litlengthLog,
+                workspace, HUF_WORKSPACE_SIZE)),
+            dictionary_corrupted, "");
+        bs->entropy.fse.litlength_repeatMode = ZSTD_dictNCountRepeat(litlengthNCount, litlengthMaxValue, MaxLL);
+        dictPtr += litlengthHeaderSize;
+    }
+
+    RETURN_ERROR_IF(dictPtr+12 > dictEnd, dictionary_corrupted, "");
+    bs->rep[0] = MEM_readLE32(dictPtr+0);
+    bs->rep[1] = MEM_readLE32(dictPtr+4);
+    bs->rep[2] = MEM_readLE32(dictPtr+8);
+    dictPtr += 12;
+
+    {   size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
+        U32 offcodeMax = MaxOff;
+        if (dictContentSize <= ((U32)-1) - 128 KB) {
+            U32 const maxOffset = (U32)dictContentSize + 128 KB; /* The maximum offset that must be supported */
+            offcodeMax = ZSTD_highbit32(maxOffset); /* Calculate minimum offset code required to represent maxOffset */
+        }
+        /* All offset values <= dictContentSize + 128 KB must be representable for a valid table */
+        bs->entropy.fse.offcode_repeatMode = ZSTD_dictNCountRepeat(offcodeNCount, offcodeMaxValue, MIN(offcodeMax, MaxOff));
+
+        /* All repCodes must be <= dictContentSize and != 0 */
+        {   U32 u;
+            for (u=0; u<3; u++) {
+                RETURN_ERROR_IF(bs->rep[u] == 0, dictionary_corrupted, "");
+                RETURN_ERROR_IF(bs->rep[u] > dictContentSize, dictionary_corrupted, "");
+    }   }   }
+
+    return (size_t)(dictPtr - (const BYTE*)dict);
+}
+
+/* Dictionary format :
+ * See :
+ * https://github.com/facebook/zstd/blob/release/doc/zstd_compression_format.md#dictionary-format
+ */
+/*! ZSTD_loadZstdDictionary() :
+ * @return : dictID, or an error code
+ *  assumptions : magic number supposed already checked
+ *                dictSize supposed >= 8
+ */
+static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
+                                      ZSTD_MatchState_t* ms,
+                                      ZSTD_cwksp* ws,
+                                      ZSTD_CCtx_params const* params,
+                                      const void* dict, size_t dictSize,
+                                      ZSTD_dictTableLoadMethod_e dtlm,
+                                      ZSTD_tableFillPurpose_e tfp,
+                                      void* workspace)
+{
+    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* const dictEnd = dictPtr + dictSize;
+    size_t dictID;
+    size_t eSize;
+    ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
+    assert(dictSize >= 8);
+    assert(MEM_readLE32(dictPtr) == ZSTD_MAGIC_DICTIONARY);
+
+    dictID = params->fParams.noDictIDFlag ? 0 :  MEM_readLE32(dictPtr + 4 /* skip magic number */ );
+    eSize = ZSTD_loadCEntropy(bs, workspace, dict, dictSize);
+    FORWARD_IF_ERROR(eSize, "ZSTD_loadCEntropy failed");
+    dictPtr += eSize;
+
+    {
+        size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
+        FORWARD_IF_ERROR(ZSTD_loadDictionaryContent(
+            ms, NULL, ws, params, dictPtr, dictContentSize, dtlm, tfp), "");
+    }
+    return dictID;
+}
+
+/** ZSTD_compress_insertDictionary() :
+*   @return : dictID, or an error code */
+static size_t
+ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
+                               ZSTD_MatchState_t* ms,
+                               ldmState_t* ls,
+                               ZSTD_cwksp* ws,
+                         const ZSTD_CCtx_params* params,
+                         const void* dict, size_t dictSize,
+                               ZSTD_dictContentType_e dictContentType,
+                               ZSTD_dictTableLoadMethod_e dtlm,
+                               ZSTD_tableFillPurpose_e tfp,
+                               void* workspace)
+{
+    DEBUGLOG(4, "ZSTD_compress_insertDictionary (dictSize=%u)", (U32)dictSize);
+    if ((dict==NULL) || (dictSize<8)) {
+        RETURN_ERROR_IF(dictContentType == ZSTD_dct_fullDict, dictionary_wrong, "");
+        return 0;
+    }
+
+    ZSTD_reset_compressedBlockState(bs);
+
+    /* dict restricted modes */
+    if (dictContentType == ZSTD_dct_rawContent)
+        return ZSTD_loadDictionaryContent(ms, ls, ws, params, dict, dictSize, dtlm, tfp);
+
+    if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) {
+        if (dictContentType == ZSTD_dct_auto) {
+            DEBUGLOG(4, "raw content dictionary detected");
+            return ZSTD_loadDictionaryContent(
+                ms, ls, ws, params, dict, dictSize, dtlm, tfp);
+        }
+        RETURN_ERROR_IF(dictContentType == ZSTD_dct_fullDict, dictionary_wrong, "");
+        assert(0);   /* impossible */
+    }
+
+    /* dict as full zstd dictionary */
+    return ZSTD_loadZstdDictionary(
+        bs, ms, ws, params, dict, dictSize, dtlm, tfp, workspace);
+}
+
+#define ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF (128 KB)
+#define ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER (6ULL)
+
+/*! ZSTD_compressBegin_internal() :
+ * Assumption : either @dict OR @cdict (or none) is non-NULL, never both
+ * @return : 0, or an error code */
+static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    ZSTD_dictTableLoadMethod_e dtlm,
+                                    const ZSTD_CDict* cdict,
+                                    const ZSTD_CCtx_params* params, U64 pledgedSrcSize,
+                                    ZSTD_buffered_policy_e zbuff)
+{
+    size_t const dictContentSize = cdict ? cdict->dictContentSize : dictSize;
+#if ZSTD_TRACE
+    cctx->traceCtx = (ZSTD_trace_compress_begin != NULL) ? ZSTD_trace_compress_begin(cctx) : 0;
+#endif
+    DEBUGLOG(4, "ZSTD_compressBegin_internal: wlog=%u", params->cParams.windowLog);
+    /* params are supposed to be fully validated at this point */
+    assert(!ZSTD_isError(ZSTD_checkCParams(params->cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+    if ( (cdict)
+      && (cdict->dictContentSize > 0)
+      && ( pledgedSrcSize < ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF
+        || pledgedSrcSize < cdict->dictContentSize * ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER
+        || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
+        || cdict->compressionLevel == 0)
+      && (params->attachDictPref != ZSTD_dictForceLoad) ) {
+        return ZSTD_resetCCtx_usingCDict(cctx, cdict, params, pledgedSrcSize, zbuff);
+    }
+
+    FORWARD_IF_ERROR( ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
+                                     dictContentSize,
+                                     ZSTDcrp_makeClean, zbuff) , "");
+    {   size_t const dictID = cdict ?
+                ZSTD_compress_insertDictionary(
+                        cctx->blockState.prevCBlock, &cctx->blockState.matchState,
+                        &cctx->ldmState, &cctx->workspace, &cctx->appliedParams, cdict->dictContent,
+                        cdict->dictContentSize, cdict->dictContentType, dtlm,
+                        ZSTD_tfp_forCCtx, cctx->tmpWorkspace)
+              : ZSTD_compress_insertDictionary(
+                        cctx->blockState.prevCBlock, &cctx->blockState.matchState,
+                        &cctx->ldmState, &cctx->workspace, &cctx->appliedParams, dict, dictSize,
+                        dictContentType, dtlm, ZSTD_tfp_forCCtx, cctx->tmpWorkspace);
+        FORWARD_IF_ERROR(dictID, "ZSTD_compress_insertDictionary failed");
+        assert(dictID <= UINT_MAX);
+        cctx->dictID = (U32)dictID;
+        cctx->dictContentSize = dictContentSize;
+    }
+    return 0;
+}
+
+size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    ZSTD_dictTableLoadMethod_e dtlm,
+                                    const ZSTD_CDict* cdict,
+                                    const ZSTD_CCtx_params* params,
+                                    unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_compressBegin_advanced_internal: wlog=%u", params->cParams.windowLog);
+    /* compression parameters verification and optimization */
+    FORWARD_IF_ERROR( ZSTD_checkCParams(params->cParams) , "");
+    return ZSTD_compressBegin_internal(cctx,
+                                       dict, dictSize, dictContentType, dtlm,
+                                       cdict,
+                                       params, pledgedSrcSize,
+                                       ZSTDb_not_buffered);
+}
+
+/*! ZSTD_compressBegin_advanced() :
+*   @return : 0, or an error code */
+size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx,
+                             const void* dict, size_t dictSize,
+                                   ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    ZSTD_CCtx_params cctxParams;
+    ZSTD_CCtxParams_init_internal(&cctxParams, &params, ZSTD_NO_CLEVEL);
+    return ZSTD_compressBegin_advanced_internal(cctx,
+                                            dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast,
+                                            NULL /*cdict*/,
+                                            &cctxParams, pledgedSrcSize);
+}
+
+static size_t
+ZSTD_compressBegin_usingDict_deprecated(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_CCtx_params cctxParams;
+    {   ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_noAttachDict);
+        ZSTD_CCtxParams_init_internal(&cctxParams, &params, (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : compressionLevel);
+    }
+    DEBUGLOG(4, "ZSTD_compressBegin_usingDict (dictSize=%u)", (unsigned)dictSize);
+    return ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast, NULL,
+                                       &cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, ZSTDb_not_buffered);
+}
+
+size_t
+ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
+{
+    return ZSTD_compressBegin_usingDict_deprecated(cctx, dict, dictSize, compressionLevel);
+}
+
+size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel)
+{
+    return ZSTD_compressBegin_usingDict_deprecated(cctx, NULL, 0, compressionLevel);
+}
+
+
+/*! ZSTD_writeEpilogue() :
+*   Ends a frame.
+*   @return : nb of bytes written into dst (or an error code) */
+static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+
+    DEBUGLOG(4, "ZSTD_writeEpilogue");
+    RETURN_ERROR_IF(cctx->stage == ZSTDcs_created, stage_wrong, "init missing");
+
+    /* special case : empty frame */
+    if (cctx->stage == ZSTDcs_init) {
+        size_t fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, &cctx->appliedParams, 0, 0);
+        FORWARD_IF_ERROR(fhSize, "ZSTD_writeFrameHeader failed");
+        dstCapacity -= fhSize;
+        op += fhSize;
+        cctx->stage = ZSTDcs_ongoing;
+    }
+
+    if (cctx->stage != ZSTDcs_ending) {
+        /* write one last empty block, make it the "last" block */
+        U32 const cBlockHeader24 = 1 /* last block */ + (((U32)bt_raw)<<1) + 0;
+        ZSTD_STATIC_ASSERT(ZSTD_BLOCKHEADERSIZE == 3);
+        RETURN_ERROR_IF(dstCapacity<3, dstSize_tooSmall, "no room for epilogue");
+        MEM_writeLE24(op, cBlockHeader24);
+        op += ZSTD_blockHeaderSize;
+        dstCapacity -= ZSTD_blockHeaderSize;
+    }
+
+    if (cctx->appliedParams.fParams.checksumFlag) {
+        U32 const checksum = (U32) XXH64_digest(&cctx->xxhState);
+        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "no room for checksum");
+        DEBUGLOG(4, "ZSTD_writeEpilogue: write checksum : %08X", (unsigned)checksum);
+        MEM_writeLE32(op, checksum);
+        op += 4;
+    }
+
+    cctx->stage = ZSTDcs_created;  /* return to "created but no init" status */
+    return (size_t)(op-ostart);
+}
+
+void ZSTD_CCtx_trace(ZSTD_CCtx* cctx, size_t extraCSize)
+{
+#if ZSTD_TRACE
+    if (cctx->traceCtx && ZSTD_trace_compress_end != NULL) {
+        int const streaming = cctx->inBuffSize > 0 || cctx->outBuffSize > 0 || cctx->appliedParams.nbWorkers > 0;
+        ZSTD_Trace trace;
+        ZSTD_memset(&trace, 0, sizeof(trace));
+        trace.version = ZSTD_VERSION_NUMBER;
+        trace.streaming = streaming;
+        trace.dictionaryID = cctx->dictID;
+        trace.dictionarySize = cctx->dictContentSize;
+        trace.uncompressedSize = cctx->consumedSrcSize;
+        trace.compressedSize = cctx->producedCSize + extraCSize;
+        trace.params = &cctx->appliedParams;
+        trace.cctx = cctx;
+        ZSTD_trace_compress_end(cctx->traceCtx, &trace);
+    }
+    cctx->traceCtx = 0;
+#else
+    (void)cctx;
+    (void)extraCSize;
+#endif
+}
+
+size_t ZSTD_compressEnd_public(ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize)
+{
+    size_t endResult;
+    size_t const cSize = ZSTD_compressContinue_internal(cctx,
+                                dst, dstCapacity, src, srcSize,
+                                1 /* frame mode */, 1 /* last chunk */);
+    FORWARD_IF_ERROR(cSize, "ZSTD_compressContinue_internal failed");
+    endResult = ZSTD_writeEpilogue(cctx, (char*)dst + cSize, dstCapacity-cSize);
+    FORWARD_IF_ERROR(endResult, "ZSTD_writeEpilogue failed");
+    assert(!(cctx->appliedParams.fParams.contentSizeFlag && cctx->pledgedSrcSizePlusOne == 0));
+    if (cctx->pledgedSrcSizePlusOne != 0) {  /* control src size */
+        ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN == (unsigned long long)-1);
+        DEBUGLOG(4, "end of frame : controlling src size");
+        RETURN_ERROR_IF(
+            cctx->pledgedSrcSizePlusOne != cctx->consumedSrcSize+1,
+            srcSize_wrong,
+             "error : pledgedSrcSize = %u, while realSrcSize = %u",
+            (unsigned)cctx->pledgedSrcSizePlusOne-1,
+            (unsigned)cctx->consumedSrcSize);
+    }
+    ZSTD_CCtx_trace(cctx, endResult);
+    return cSize + endResult;
+}
+
+/* NOTE: Must just wrap ZSTD_compressEnd_public() */
+size_t ZSTD_compressEnd(ZSTD_CCtx* cctx,
+                        void* dst, size_t dstCapacity,
+                  const void* src, size_t srcSize)
+{
+    return ZSTD_compressEnd_public(cctx, dst, dstCapacity, src, srcSize);
+}
+
+size_t ZSTD_compress_advanced (ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const void* dict,size_t dictSize,
+                               ZSTD_parameters params)
+{
+    DEBUGLOG(4, "ZSTD_compress_advanced");
+    FORWARD_IF_ERROR(ZSTD_checkCParams(params.cParams), "");
+    ZSTD_CCtxParams_init_internal(&cctx->simpleApiParams, &params, ZSTD_NO_CLEVEL);
+    return ZSTD_compress_advanced_internal(cctx,
+                                           dst, dstCapacity,
+                                           src, srcSize,
+                                           dict, dictSize,
+                                           &cctx->simpleApiParams);
+}
+
+/* Internal */
+size_t ZSTD_compress_advanced_internal(
+        ZSTD_CCtx* cctx,
+        void* dst, size_t dstCapacity,
+        const void* src, size_t srcSize,
+        const void* dict,size_t dictSize,
+        const ZSTD_CCtx_params* params)
+{
+    DEBUGLOG(4, "ZSTD_compress_advanced_internal (srcSize:%u)", (unsigned)srcSize);
+    FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
+                         dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast, NULL,
+                         params, srcSize, ZSTDb_not_buffered) , "");
+    return ZSTD_compressEnd_public(cctx, dst, dstCapacity, src, srcSize);
+}
+
+size_t ZSTD_compress_usingDict(ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const void* dict, size_t dictSize,
+                               int compressionLevel)
+{
+    {
+        ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, srcSize, dict ? dictSize : 0, ZSTD_cpm_noAttachDict);
+        assert(params.fParams.contentSizeFlag == 1);
+        ZSTD_CCtxParams_init_internal(&cctx->simpleApiParams, &params, (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT: compressionLevel);
+    }
+    DEBUGLOG(4, "ZSTD_compress_usingDict (srcSize=%u)", (unsigned)srcSize);
+    return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, dict, dictSize, &cctx->simpleApiParams);
+}
+
+size_t ZSTD_compressCCtx(ZSTD_CCtx* cctx,
+                         void* dst, size_t dstCapacity,
+                   const void* src, size_t srcSize,
+                         int compressionLevel)
+{
+    DEBUGLOG(4, "ZSTD_compressCCtx (srcSize=%u)", (unsigned)srcSize);
+    assert(cctx != NULL);
+    return ZSTD_compress_usingDict(cctx, dst, dstCapacity, src, srcSize, NULL, 0, compressionLevel);
+}
+
+size_t ZSTD_compress(void* dst, size_t dstCapacity,
+               const void* src, size_t srcSize,
+                     int compressionLevel)
+{
+    size_t result;
+#if ZSTD_COMPRESS_HEAPMODE
+    ZSTD_CCtx* cctx = ZSTD_createCCtx();
+    RETURN_ERROR_IF(!cctx, memory_allocation, "ZSTD_createCCtx failed");
+    result = ZSTD_compressCCtx(cctx, dst, dstCapacity, src, srcSize, compressionLevel);
+    ZSTD_freeCCtx(cctx);
+#else
+    ZSTD_CCtx ctxBody;
+    ZSTD_initCCtx(&ctxBody, ZSTD_defaultCMem);
+    result = ZSTD_compressCCtx(&ctxBody, dst, dstCapacity, src, srcSize, compressionLevel);
+    ZSTD_freeCCtxContent(&ctxBody);   /* can't free ctxBody itself, as it's on stack; free only heap content */
+#endif
+    return result;
+}
+
+
+/* =====  Dictionary API  ===== */
+
+/*! ZSTD_estimateCDictSize_advanced() :
+ *  Estimate amount of memory that will be needed to create a dictionary with following arguments */
+size_t ZSTD_estimateCDictSize_advanced(
+        size_t dictSize, ZSTD_compressionParameters cParams,
+        ZSTD_dictLoadMethod_e dictLoadMethod)
+{
+    DEBUGLOG(5, "sizeof(ZSTD_CDict) : %u", (unsigned)sizeof(ZSTD_CDict));
+    return ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict))
+         + ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE)
+         /* enableDedicatedDictSearch == 1 ensures that CDict estimation will not be too small
+          * in case we are using DDS with row-hash. */
+         + ZSTD_sizeof_matchState(&cParams, ZSTD_resolveRowMatchFinderMode(ZSTD_ps_auto, &cParams),
+                                  /* enableDedicatedDictSearch */ 1, /* forCCtx */ 0)
+         + (dictLoadMethod == ZSTD_dlm_byRef ? 0
+            : ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void *))));
+}
+
+size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel)
+{
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    return ZSTD_estimateCDictSize_advanced(dictSize, cParams, ZSTD_dlm_byCopy);
+}
+
+size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict)
+{
+    if (cdict==NULL) return 0;   /* support sizeof on NULL */
+    DEBUGLOG(5, "sizeof(*cdict) : %u", (unsigned)sizeof(*cdict));
+    /* cdict may be in the workspace */
+    return (cdict->workspace.workspace == cdict ? 0 : sizeof(*cdict))
+        + ZSTD_cwksp_sizeof(&cdict->workspace);
+}
+
+static size_t ZSTD_initCDict_internal(
+                    ZSTD_CDict* cdict,
+              const void* dictBuffer, size_t dictSize,
+                    ZSTD_dictLoadMethod_e dictLoadMethod,
+                    ZSTD_dictContentType_e dictContentType,
+                    ZSTD_CCtx_params params)
+{
+    DEBUGLOG(3, "ZSTD_initCDict_internal (dictContentType:%u)", (unsigned)dictContentType);
+    assert(!ZSTD_checkCParams(params.cParams));
+    cdict->matchState.cParams = params.cParams;
+    cdict->matchState.dedicatedDictSearch = params.enableDedicatedDictSearch;
+    if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dictBuffer) || (!dictSize)) {
+        cdict->dictContent = dictBuffer;
+    } else {
+         void *internalBuffer = ZSTD_cwksp_reserve_object(&cdict->workspace, ZSTD_cwksp_align(dictSize, sizeof(void*)));
+        RETURN_ERROR_IF(!internalBuffer, memory_allocation, "NULL pointer!");
+        cdict->dictContent = internalBuffer;
+        ZSTD_memcpy(internalBuffer, dictBuffer, dictSize);
+    }
+    cdict->dictContentSize = dictSize;
+    cdict->dictContentType = dictContentType;
+
+    cdict->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(&cdict->workspace, HUF_WORKSPACE_SIZE);
+
+
+    /* Reset the state to no dictionary */
+    ZSTD_reset_compressedBlockState(&cdict->cBlockState);
+    FORWARD_IF_ERROR(ZSTD_reset_matchState(
+        &cdict->matchState,
+        &cdict->workspace,
+        &params.cParams,
+        params.useRowMatchFinder,
+        ZSTDcrp_makeClean,
+        ZSTDirp_reset,
+        ZSTD_resetTarget_CDict), "");
+    /* (Maybe) load the dictionary
+     * Skips loading the dictionary if it is < 8 bytes.
+     */
+    {   params.compressionLevel = ZSTD_CLEVEL_DEFAULT;
+        params.fParams.contentSizeFlag = 1;
+        {   size_t const dictID = ZSTD_compress_insertDictionary(
+                    &cdict->cBlockState, &cdict->matchState, NULL, &cdict->workspace,
+                    &params, cdict->dictContent, cdict->dictContentSize,
+                    dictContentType, ZSTD_dtlm_full, ZSTD_tfp_forCDict, cdict->entropyWorkspace);
+            FORWARD_IF_ERROR(dictID, "ZSTD_compress_insertDictionary failed");
+            assert(dictID <= (size_t)(U32)-1);
+            cdict->dictID = (U32)dictID;
+        }
+    }
+
+    return 0;
+}
+
+static ZSTD_CDict*
+ZSTD_createCDict_advanced_internal(size_t dictSize,
+                                ZSTD_dictLoadMethod_e dictLoadMethod,
+                                ZSTD_compressionParameters cParams,
+                                ZSTD_ParamSwitch_e useRowMatchFinder,
+                                int enableDedicatedDictSearch,
+                                ZSTD_customMem customMem)
+{
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
+    DEBUGLOG(3, "ZSTD_createCDict_advanced_internal (dictSize=%u)", (unsigned)dictSize);
+
+    {   size_t const workspaceSize =
+            ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict)) +
+            ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE) +
+            ZSTD_sizeof_matchState(&cParams, useRowMatchFinder, enableDedicatedDictSearch, /* forCCtx */ 0) +
+            (dictLoadMethod == ZSTD_dlm_byRef ? 0
+             : ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void*))));
+        void* const workspace = ZSTD_customMalloc(workspaceSize, customMem);
+        ZSTD_cwksp ws;
+        ZSTD_CDict* cdict;
+
+        if (!workspace) {
+            ZSTD_customFree(workspace, customMem);
+            return NULL;
+        }
+
+        ZSTD_cwksp_init(&ws, workspace, workspaceSize, ZSTD_cwksp_dynamic_alloc);
+
+        cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict));
+        assert(cdict != NULL);
+        ZSTD_cwksp_move(&cdict->workspace, &ws);
+        cdict->customMem = customMem;
+        cdict->compressionLevel = ZSTD_NO_CLEVEL; /* signals advanced API usage */
+        cdict->useRowMatchFinder = useRowMatchFinder;
+        return cdict;
+    }
+}
+
+ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType,
+                                      ZSTD_compressionParameters cParams,
+                                      ZSTD_customMem customMem)
+{
+    ZSTD_CCtx_params cctxParams;
+    ZSTD_memset(&cctxParams, 0, sizeof(cctxParams));
+    DEBUGLOG(3, "ZSTD_createCDict_advanced, dictSize=%u, mode=%u", (unsigned)dictSize, (unsigned)dictContentType);
+    ZSTD_CCtxParams_init(&cctxParams, 0);
+    cctxParams.cParams = cParams;
+    cctxParams.customMem = customMem;
+    return ZSTD_createCDict_advanced2(
+        dictBuffer, dictSize,
+        dictLoadMethod, dictContentType,
+        &cctxParams, customMem);
+}
+
+ZSTD_CDict* ZSTD_createCDict_advanced2(
+        const void* dict, size_t dictSize,
+        ZSTD_dictLoadMethod_e dictLoadMethod,
+        ZSTD_dictContentType_e dictContentType,
+        const ZSTD_CCtx_params* originalCctxParams,
+        ZSTD_customMem customMem)
+{
+    ZSTD_CCtx_params cctxParams = *originalCctxParams;
+    ZSTD_compressionParameters cParams;
+    ZSTD_CDict* cdict;
+
+    DEBUGLOG(3, "ZSTD_createCDict_advanced2, dictSize=%u, mode=%u", (unsigned)dictSize, (unsigned)dictContentType);
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+
+    if (cctxParams.enableDedicatedDictSearch) {
+        cParams = ZSTD_dedicatedDictSearch_getCParams(
+            cctxParams.compressionLevel, dictSize);
+        ZSTD_overrideCParams(&cParams, &cctxParams.cParams);
+    } else {
+        cParams = ZSTD_getCParamsFromCCtxParams(
+            &cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    }
+
+    if (!ZSTD_dedicatedDictSearch_isSupported(&cParams)) {
+        /* Fall back to non-DDSS params */
+        cctxParams.enableDedicatedDictSearch = 0;
+        cParams = ZSTD_getCParamsFromCCtxParams(
+            &cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    }
+
+    DEBUGLOG(3, "ZSTD_createCDict_advanced2: DedicatedDictSearch=%u", cctxParams.enableDedicatedDictSearch);
+    cctxParams.cParams = cParams;
+    cctxParams.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams.useRowMatchFinder, &cParams);
+
+    cdict = ZSTD_createCDict_advanced_internal(dictSize,
+                        dictLoadMethod, cctxParams.cParams,
+                        cctxParams.useRowMatchFinder, cctxParams.enableDedicatedDictSearch,
+                        customMem);
+
+    if (!cdict || ZSTD_isError( ZSTD_initCDict_internal(cdict,
+                                    dict, dictSize,
+                                    dictLoadMethod, dictContentType,
+                                    cctxParams) )) {
+        ZSTD_freeCDict(cdict);
+        return NULL;
+    }
+
+    return cdict;
+}
+
+ZSTD_CDict* ZSTD_createCDict(const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    ZSTD_CDict* const cdict = ZSTD_createCDict_advanced(dict, dictSize,
+                                                  ZSTD_dlm_byCopy, ZSTD_dct_auto,
+                                                  cParams, ZSTD_defaultCMem);
+    if (cdict)
+        cdict->compressionLevel = (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : compressionLevel;
+    return cdict;
+}
+
+ZSTD_CDict* ZSTD_createCDict_byReference(const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    ZSTD_CDict* const cdict = ZSTD_createCDict_advanced(dict, dictSize,
+                                     ZSTD_dlm_byRef, ZSTD_dct_auto,
+                                     cParams, ZSTD_defaultCMem);
+    if (cdict)
+        cdict->compressionLevel = (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : compressionLevel;
+    return cdict;
+}
+
+size_t ZSTD_freeCDict(ZSTD_CDict* cdict)
+{
+    if (cdict==NULL) return 0;   /* support free on NULL */
+    {   ZSTD_customMem const cMem = cdict->customMem;
+        int cdictInWorkspace = ZSTD_cwksp_owns_buffer(&cdict->workspace, cdict);
+        ZSTD_cwksp_free(&cdict->workspace, cMem);
+        if (!cdictInWorkspace) {
+            ZSTD_customFree(cdict, cMem);
+        }
+        return 0;
+    }
+}
+
+/*! ZSTD_initStaticCDict_advanced() :
+ *  Generate a digested dictionary in provided memory area.
+ *  workspace: The memory area to emplace the dictionary into.
+ *             Provided pointer must 8-bytes aligned.
+ *             It must outlive dictionary usage.
+ *  workspaceSize: Use ZSTD_estimateCDictSize()
+ *                 to determine how large workspace must be.
+ *  cParams : use ZSTD_getCParams() to transform a compression level
+ *            into its relevant cParams.
+ * @return : pointer to ZSTD_CDict*, or NULL if error (size too small)
+ *  Note : there is no corresponding "free" function.
+ *         Since workspace was allocated externally, it must be freed externally.
+ */
+const ZSTD_CDict* ZSTD_initStaticCDict(
+                                 void* workspace, size_t workspaceSize,
+                           const void* dict, size_t dictSize,
+                                 ZSTD_dictLoadMethod_e dictLoadMethod,
+                                 ZSTD_dictContentType_e dictContentType,
+                                 ZSTD_compressionParameters cParams)
+{
+    ZSTD_ParamSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(ZSTD_ps_auto, &cParams);
+    /* enableDedicatedDictSearch == 1 ensures matchstate is not too small in case this CDict will be used for DDS + row hash */
+    size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, useRowMatchFinder, /* enableDedicatedDictSearch */ 1, /* forCCtx */ 0);
+    size_t const neededSize = ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict))
+                            + (dictLoadMethod == ZSTD_dlm_byRef ? 0
+                               : ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void*))))
+                            + ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE)
+                            + matchStateSize;
+    ZSTD_CDict* cdict;
+    ZSTD_CCtx_params params;
+
+    DEBUGLOG(4, "ZSTD_initStaticCDict (dictSize==%u)", (unsigned)dictSize);
+    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
+
+    {
+        ZSTD_cwksp ws;
+        ZSTD_cwksp_init(&ws, workspace, workspaceSize, ZSTD_cwksp_static_alloc);
+        cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict));
+        if (cdict == NULL) return NULL;
+        ZSTD_cwksp_move(&cdict->workspace, &ws);
+    }
+
+    if (workspaceSize < neededSize) return NULL;
+
+    ZSTD_CCtxParams_init(&params, 0);
+    params.cParams = cParams;
+    params.useRowMatchFinder = useRowMatchFinder;
+    cdict->useRowMatchFinder = useRowMatchFinder;
+    cdict->compressionLevel = ZSTD_NO_CLEVEL;
+
+    if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
+                                              dict, dictSize,
+                                              dictLoadMethod, dictContentType,
+                                              params) ))
+        return NULL;
+
+    return cdict;
+}
+
+ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict)
+{
+    assert(cdict != NULL);
+    return cdict->matchState.cParams;
+}
+
+/*! ZSTD_getDictID_fromCDict() :
+ *  Provides the dictID of the dictionary loaded into `cdict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+unsigned ZSTD_getDictID_fromCDict(const ZSTD_CDict* cdict)
+{
+    if (cdict==NULL) return 0;
+    return cdict->dictID;
+}
+
+/* ZSTD_compressBegin_usingCDict_internal() :
+ * Implementation of various ZSTD_compressBegin_usingCDict* functions.
+ */
+static size_t ZSTD_compressBegin_usingCDict_internal(
+    ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict,
+    ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize)
+{
+    ZSTD_CCtx_params cctxParams;
+    DEBUGLOG(4, "ZSTD_compressBegin_usingCDict_internal");
+    RETURN_ERROR_IF(cdict==NULL, dictionary_wrong, "NULL pointer!");
+    /* Initialize the cctxParams from the cdict */
+    {
+        ZSTD_parameters params;
+        params.fParams = fParams;
+        params.cParams = ( pledgedSrcSize < ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF
+                        || pledgedSrcSize < cdict->dictContentSize * ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER
+                        || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
+                        || cdict->compressionLevel == 0 ) ?
+                ZSTD_getCParamsFromCDict(cdict)
+              : ZSTD_getCParams(cdict->compressionLevel,
+                                pledgedSrcSize,
+                                cdict->dictContentSize);
+        ZSTD_CCtxParams_init_internal(&cctxParams, &params, cdict->compressionLevel);
+    }
+    /* Increase window log to fit the entire dictionary and source if the
+     * source size is known. Limit the increase to 19, which is the
+     * window log for compression level 1 with the largest source size.
+     */
+    if (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+        U32 const limitedSrcSize = (U32)MIN(pledgedSrcSize, 1U << 19);
+        U32 const limitedSrcLog = limitedSrcSize > 1 ? ZSTD_highbit32(limitedSrcSize - 1) + 1 : 1;
+        cctxParams.cParams.windowLog = MAX(cctxParams.cParams.windowLog, limitedSrcLog);
+    }
+    return ZSTD_compressBegin_internal(cctx,
+                                        NULL, 0, ZSTD_dct_auto, ZSTD_dtlm_fast,
+                                        cdict,
+                                        &cctxParams, pledgedSrcSize,
+                                        ZSTDb_not_buffered);
+}
+
+
+/* ZSTD_compressBegin_usingCDict_advanced() :
+ * This function is DEPRECATED.
+ * cdict must be != NULL */
+size_t ZSTD_compressBegin_usingCDict_advanced(
+    ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict,
+    ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize)
+{
+    return ZSTD_compressBegin_usingCDict_internal(cctx, cdict, fParams, pledgedSrcSize);
+}
+
+/* ZSTD_compressBegin_usingCDict() :
+ * cdict must be != NULL */
+size_t ZSTD_compressBegin_usingCDict_deprecated(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    ZSTD_frameParameters const fParams = { 0 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    return ZSTD_compressBegin_usingCDict_internal(cctx, cdict, fParams, ZSTD_CONTENTSIZE_UNKNOWN);
+}
+
+size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    return ZSTD_compressBegin_usingCDict_deprecated(cctx, cdict);
+}
+
+/*! ZSTD_compress_usingCDict_internal():
+ * Implementation of various ZSTD_compress_usingCDict* functions.
+ */
+static size_t ZSTD_compress_usingCDict_internal(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const ZSTD_CDict* cdict, ZSTD_frameParameters fParams)
+{
+    FORWARD_IF_ERROR(ZSTD_compressBegin_usingCDict_internal(cctx, cdict, fParams, srcSize), ""); /* will check if cdict != NULL */
+    return ZSTD_compressEnd_public(cctx, dst, dstCapacity, src, srcSize);
+}
+
+/*! ZSTD_compress_usingCDict_advanced():
+ * This function is DEPRECATED.
+ */
+size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const ZSTD_CDict* cdict, ZSTD_frameParameters fParams)
+{
+    return ZSTD_compress_usingCDict_internal(cctx, dst, dstCapacity, src, srcSize, cdict, fParams);
+}
+
+/*! ZSTD_compress_usingCDict() :
+ *  Compression using a digested Dictionary.
+ *  Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
+ *  Note that compression parameters are decided at CDict creation time
+ *  while frame parameters are hardcoded */
+size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const ZSTD_CDict* cdict)
+{
+    ZSTD_frameParameters const fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    return ZSTD_compress_usingCDict_internal(cctx, dst, dstCapacity, src, srcSize, cdict, fParams);
+}
+
+
+
+/* ******************************************************************
+*  Streaming
+********************************************************************/
+
+ZSTD_CStream* ZSTD_createCStream(void)
+{
+    DEBUGLOG(3, "ZSTD_createCStream");
+    return ZSTD_createCStream_advanced(ZSTD_defaultCMem);
+}
+
+ZSTD_CStream* ZSTD_initStaticCStream(void *workspace, size_t workspaceSize)
+{
+    return ZSTD_initStaticCCtx(workspace, workspaceSize);
+}
+
+ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem)
+{   /* CStream and CCtx are now same object */
+    return ZSTD_createCCtx_advanced(customMem);
+}
+
+size_t ZSTD_freeCStream(ZSTD_CStream* zcs)
+{
+    return ZSTD_freeCCtx(zcs);   /* same object */
+}
+
+
+
+/*======   Initialization   ======*/
+
+size_t ZSTD_CStreamInSize(void)  { return ZSTD_BLOCKSIZE_MAX; }
+
+size_t ZSTD_CStreamOutSize(void)
+{
+    return ZSTD_compressBound(ZSTD_BLOCKSIZE_MAX) + ZSTD_blockHeaderSize + 4 /* 32-bits hash */ ;
+}
+
+static ZSTD_CParamMode_e ZSTD_getCParamMode(ZSTD_CDict const* cdict, ZSTD_CCtx_params const* params, U64 pledgedSrcSize)
+{
+    if (cdict != NULL && ZSTD_shouldAttachDict(cdict, params, pledgedSrcSize))
+        return ZSTD_cpm_attachDict;
+    else
+        return ZSTD_cpm_noAttachDict;
+}
+
+/* ZSTD_resetCStream():
+ * pledgedSrcSize == 0 means "unknown" */
+size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pss)
+{
+    /* temporary : 0 interpreted as "unknown" during transition period.
+     * Users willing to specify "unknown" **must** use ZSTD_CONTENTSIZE_UNKNOWN.
+     * 0 will be interpreted as "empty" in the future.
+     */
+    U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
+    DEBUGLOG(4, "ZSTD_resetCStream: pledgedSrcSize = %u", (unsigned)pledgedSrcSize);
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    return 0;
+}
+
+/*! ZSTD_initCStream_internal() :
+ *  Note : for lib/compress only. Used by zstdmt_compress.c.
+ *  Assumption 1 : params are valid
+ *  Assumption 2 : either dict, or cdict, is defined, not both */
+size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
+                    const void* dict, size_t dictSize, const ZSTD_CDict* cdict,
+                    const ZSTD_CCtx_params* params,
+                    unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_internal");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    assert(!ZSTD_isError(ZSTD_checkCParams(params->cParams)));
+    zcs->requestedParams = *params;
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+    if (dict) {
+        FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) , "");
+    } else {
+        /* Dictionary is cleared if !cdict */
+        FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) , "");
+    }
+    return 0;
+}
+
+/* ZSTD_initCStream_usingCDict_advanced() :
+ * same as ZSTD_initCStream_usingCDict(), with control over frame parameters */
+size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
+                                            const ZSTD_CDict* cdict,
+                                            ZSTD_frameParameters fParams,
+                                            unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_usingCDict_advanced");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    zcs->requestedParams.fParams = fParams;
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) , "");
+    return 0;
+}
+
+/* note : cdict must outlive compression session */
+size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_usingCDict");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) , "");
+    return 0;
+}
+
+
+/* ZSTD_initCStream_advanced() :
+ * pledgedSrcSize must be exact.
+ * if srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN.
+ * dict is loaded with default parameters ZSTD_dct_auto and ZSTD_dlm_byCopy. */
+size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
+                                 const void* dict, size_t dictSize,
+                                 ZSTD_parameters params, unsigned long long pss)
+{
+    /* for compatibility with older programs relying on this behavior.
+     * Users should now specify ZSTD_CONTENTSIZE_UNKNOWN.
+     * This line will be removed in the future.
+     */
+    U64 const pledgedSrcSize = (pss==0 && params.fParams.contentSizeFlag==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
+    DEBUGLOG(4, "ZSTD_initCStream_advanced");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    FORWARD_IF_ERROR( ZSTD_checkCParams(params.cParams) , "");
+    ZSTD_CCtxParams_setZstdParams(&zcs->requestedParams, &params);
+    FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) , "");
+    return 0;
+}
+
+size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_usingDict");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) , "");
+    return 0;
+}
+
+size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pss)
+{
+    /* temporary : 0 interpreted as "unknown" during transition period.
+     * Users willing to specify "unknown" **must** use ZSTD_CONTENTSIZE_UNKNOWN.
+     * 0 will be interpreted as "empty" in the future.
+     */
+    U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
+    DEBUGLOG(4, "ZSTD_initCStream_srcSize");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, NULL) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    return 0;
+}
+
+size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel)
+{
+    DEBUGLOG(4, "ZSTD_initCStream");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, NULL) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) , "");
+    return 0;
+}
+
+/*======   Compression   ======*/
+
+static size_t ZSTD_nextInputSizeHint(const ZSTD_CCtx* cctx)
+{
+    if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        return cctx->blockSizeMax - cctx->stableIn_notConsumed;
+    }
+    assert(cctx->appliedParams.inBufferMode == ZSTD_bm_buffered);
+    {   size_t hintInSize = cctx->inBuffTarget - cctx->inBuffPos;
+        if (hintInSize==0) hintInSize = cctx->blockSizeMax;
+        return hintInSize;
+    }
+}
+
+/** ZSTD_compressStream_generic():
+ *  internal function for all *compressStream*() variants
+ * @return : hint size for next input to complete ongoing block */
+static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
+                                          ZSTD_outBuffer* output,
+                                          ZSTD_inBuffer* input,
+                                          ZSTD_EndDirective const flushMode)
+{
+    const char* const istart = (assert(input != NULL), (const char*)input->src);
+    const char* const iend = (istart != NULL) ? istart + input->size : istart;
+    const char* ip = (istart != NULL) ? istart + input->pos : istart;
+    char* const ostart = (assert(output != NULL), (char*)output->dst);
+    char* const oend = (ostart != NULL) ? ostart + output->size : ostart;
+    char* op = (ostart != NULL) ? ostart + output->pos : ostart;
+    U32 someMoreWork = 1;
+
+    /* check expectations */
+    DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%i, srcSize = %zu", (int)flushMode, input->size - input->pos);
+    assert(zcs != NULL);
+    if (zcs->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        assert(input->pos >= zcs->stableIn_notConsumed);
+        input->pos -= zcs->stableIn_notConsumed;
+        if (ip) ip -= zcs->stableIn_notConsumed;
+        zcs->stableIn_notConsumed = 0;
+    }
+    if (zcs->appliedParams.inBufferMode == ZSTD_bm_buffered) {
+        assert(zcs->inBuff != NULL);
+        assert(zcs->inBuffSize > 0);
+    }
+    if (zcs->appliedParams.outBufferMode == ZSTD_bm_buffered) {
+        assert(zcs->outBuff !=  NULL);
+        assert(zcs->outBuffSize > 0);
+    }
+    if (input->src == NULL) assert(input->size == 0);
+    assert(input->pos <= input->size);
+    if (output->dst == NULL) assert(output->size == 0);
+    assert(output->pos <= output->size);
+    assert((U32)flushMode <= (U32)ZSTD_e_end);
+
+    while (someMoreWork) {
+        switch(zcs->streamStage)
+        {
+        case zcss_init:
+            RETURN_ERROR(init_missing, "call ZSTD_initCStream() first!");
+
+        case zcss_load:
+            if ( (flushMode == ZSTD_e_end)
+              && ( (size_t)(oend-op) >= ZSTD_compressBound((size_t)(iend-ip))     /* Enough output space */
+                || zcs->appliedParams.outBufferMode == ZSTD_bm_stable)  /* OR we are allowed to return dstSizeTooSmall */
+              && (zcs->inBuffPos == 0) ) {
+                /* shortcut to compression pass directly into output buffer */
+                size_t const cSize = ZSTD_compressEnd_public(zcs,
+                                                op, (size_t)(oend-op),
+                                                ip, (size_t)(iend-ip));
+                DEBUGLOG(4, "ZSTD_compressEnd : cSize=%u", (unsigned)cSize);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressEnd failed");
+                ip = iend;
+                op += cSize;
+                zcs->frameEnded = 1;
+                ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+                someMoreWork = 0; break;
+            }
+            /* complete loading into inBuffer in buffered mode */
+            if (zcs->appliedParams.inBufferMode == ZSTD_bm_buffered) {
+                size_t const toLoad = zcs->inBuffTarget - zcs->inBuffPos;
+                size_t const loaded = ZSTD_limitCopy(
+                                        zcs->inBuff + zcs->inBuffPos, toLoad,
+                                        ip, (size_t)(iend-ip));
+                zcs->inBuffPos += loaded;
+                if (ip) ip += loaded;
+                if ( (flushMode == ZSTD_e_continue)
+                  && (zcs->inBuffPos < zcs->inBuffTarget) ) {
+                    /* not enough input to fill full block : stop here */
+                    someMoreWork = 0; break;
+                }
+                if ( (flushMode == ZSTD_e_flush)
+                  && (zcs->inBuffPos == zcs->inToCompress) ) {
+                    /* empty */
+                    someMoreWork = 0; break;
+                }
+            } else {
+                assert(zcs->appliedParams.inBufferMode == ZSTD_bm_stable);
+                if ( (flushMode == ZSTD_e_continue)
+                  && ( (size_t)(iend - ip) < zcs->blockSizeMax) ) {
+                    /* can't compress a full block : stop here */
+                    zcs->stableIn_notConsumed = (size_t)(iend - ip);
+                    ip = iend;  /* pretend to have consumed input */
+                    someMoreWork = 0; break;
+                }
+                if ( (flushMode == ZSTD_e_flush)
+                  && (ip == iend) ) {
+                    /* empty */
+                    someMoreWork = 0; break;
+                }
+            }
+            /* compress current block (note : this stage cannot be stopped in the middle) */
+            DEBUGLOG(5, "stream compression stage (flushMode==%u)", flushMode);
+            {   int const inputBuffered = (zcs->appliedParams.inBufferMode == ZSTD_bm_buffered);
+                void* cDst;
+                size_t cSize;
+                size_t oSize = (size_t)(oend-op);
+                size_t const iSize = inputBuffered ? zcs->inBuffPos - zcs->inToCompress
+                                                   : MIN((size_t)(iend - ip), zcs->blockSizeMax);
+                if (oSize >= ZSTD_compressBound(iSize) || zcs->appliedParams.outBufferMode == ZSTD_bm_stable)
+                    cDst = op;   /* compress into output buffer, to skip flush stage */
+                else
+                    cDst = zcs->outBuff, oSize = zcs->outBuffSize;
+                if (inputBuffered) {
+                    unsigned const lastBlock = (flushMode == ZSTD_e_end) && (ip==iend);
+                    cSize = lastBlock ?
+                            ZSTD_compressEnd_public(zcs, cDst, oSize,
+                                        zcs->inBuff + zcs->inToCompress, iSize) :
+                            ZSTD_compressContinue_public(zcs, cDst, oSize,
+                                        zcs->inBuff + zcs->inToCompress, iSize);
+                    FORWARD_IF_ERROR(cSize, "%s", lastBlock ? "ZSTD_compressEnd failed" : "ZSTD_compressContinue failed");
+                    zcs->frameEnded = lastBlock;
+                    /* prepare next block */
+                    zcs->inBuffTarget = zcs->inBuffPos + zcs->blockSizeMax;
+                    if (zcs->inBuffTarget > zcs->inBuffSize)
+                        zcs->inBuffPos = 0, zcs->inBuffTarget = zcs->blockSizeMax;
+                    DEBUGLOG(5, "inBuffTarget:%u / inBuffSize:%u",
+                            (unsigned)zcs->inBuffTarget, (unsigned)zcs->inBuffSize);
+                    if (!lastBlock)
+                        assert(zcs->inBuffTarget <= zcs->inBuffSize);
+                    zcs->inToCompress = zcs->inBuffPos;
+                } else { /* !inputBuffered, hence ZSTD_bm_stable */
+                    unsigned const lastBlock = (flushMode == ZSTD_e_end) && (ip + iSize == iend);
+                    cSize = lastBlock ?
+                            ZSTD_compressEnd_public(zcs, cDst, oSize, ip, iSize) :
+                            ZSTD_compressContinue_public(zcs, cDst, oSize, ip, iSize);
+                    /* Consume the input prior to error checking to mirror buffered mode. */
+                    if (ip) ip += iSize;
+                    FORWARD_IF_ERROR(cSize, "%s", lastBlock ? "ZSTD_compressEnd failed" : "ZSTD_compressContinue failed");
+                    zcs->frameEnded = lastBlock;
+                    if (lastBlock) assert(ip == iend);
+                }
+                if (cDst == op) {  /* no need to flush */
+                    op += cSize;
+                    if (zcs->frameEnded) {
+                        DEBUGLOG(5, "Frame completed directly in outBuffer");
+                        someMoreWork = 0;
+                        ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+                    }
+                    break;
+                }
+                zcs->outBuffContentSize = cSize;
+                zcs->outBuffFlushedSize = 0;
+                zcs->streamStage = zcss_flush; /* pass-through to flush stage */
+            }
+	    ZSTD_FALLTHROUGH;
+        case zcss_flush:
+            DEBUGLOG(5, "flush stage");
+            assert(zcs->appliedParams.outBufferMode == ZSTD_bm_buffered);
+            {   size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
+                size_t const flushed = ZSTD_limitCopy(op, (size_t)(oend-op),
+                            zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
+                DEBUGLOG(5, "toFlush: %u into %u ==> flushed: %u",
+                            (unsigned)toFlush, (unsigned)(oend-op), (unsigned)flushed);
+                if (flushed)
+                    op += flushed;
+                zcs->outBuffFlushedSize += flushed;
+                if (toFlush!=flushed) {
+                    /* flush not fully completed, presumably because dst is too small */
+                    assert(op==oend);
+                    someMoreWork = 0;
+                    break;
+                }
+                zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
+                if (zcs->frameEnded) {
+                    DEBUGLOG(5, "Frame completed on flush");
+                    someMoreWork = 0;
+                    ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+                    break;
+                }
+                zcs->streamStage = zcss_load;
+                break;
+            }
+
+        default: /* impossible */
+            assert(0);
+        }
+    }
+
+    input->pos = (size_t)(ip - istart);
+    output->pos = (size_t)(op - ostart);
+    if (zcs->frameEnded) return 0;
+    return ZSTD_nextInputSizeHint(zcs);
+}
+
+static size_t ZSTD_nextInputSizeHint_MTorST(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers >= 1) {
+        assert(cctx->mtctx != NULL);
+        return ZSTDMT_nextInputSizeHint(cctx->mtctx);
+    }
+#endif
+    return ZSTD_nextInputSizeHint(cctx);
+
+}
+
+size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    FORWARD_IF_ERROR( ZSTD_compressStream2(zcs, output, input, ZSTD_e_continue) , "");
+    return ZSTD_nextInputSizeHint_MTorST(zcs);
+}
+
+/* After a compression call set the expected input/output buffer.
+ * This is validated at the start of the next compression call.
+ */
+static void
+ZSTD_setBufferExpectations(ZSTD_CCtx* cctx, const ZSTD_outBuffer* output, const ZSTD_inBuffer* input)
+{
+    DEBUGLOG(5, "ZSTD_setBufferExpectations (for advanced stable in/out modes)");
+    if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        cctx->expectedInBuffer = *input;
+    }
+    if (cctx->appliedParams.outBufferMode == ZSTD_bm_stable) {
+        cctx->expectedOutBufferSize = output->size - output->pos;
+    }
+}
+
+/* Validate that the input/output buffers match the expectations set by
+ * ZSTD_setBufferExpectations.
+ */
+static size_t ZSTD_checkBufferStability(ZSTD_CCtx const* cctx,
+                                        ZSTD_outBuffer const* output,
+                                        ZSTD_inBuffer const* input,
+                                        ZSTD_EndDirective endOp)
+{
+    if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        ZSTD_inBuffer const expect = cctx->expectedInBuffer;
+        if (expect.src != input->src || expect.pos != input->pos)
+            RETURN_ERROR(stabilityCondition_notRespected, "ZSTD_c_stableInBuffer enabled but input differs!");
+    }
+    (void)endOp;
+    if (cctx->appliedParams.outBufferMode == ZSTD_bm_stable) {
+        size_t const outBufferSize = output->size - output->pos;
+        if (cctx->expectedOutBufferSize != outBufferSize)
+            RETURN_ERROR(stabilityCondition_notRespected, "ZSTD_c_stableOutBuffer enabled but output size differs!");
+    }
+    return 0;
+}
+
+/*
+ * If @endOp == ZSTD_e_end, @inSize becomes pledgedSrcSize.
+ * Otherwise, it's ignored.
+ * @return: 0 on success, or a ZSTD_error code otherwise.
+ */
+static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
+                                             ZSTD_EndDirective endOp,
+                                             size_t inSize)
+{
+    ZSTD_CCtx_params params = cctx->requestedParams;
+    ZSTD_prefixDict const prefixDict = cctx->prefixDict;
+    FORWARD_IF_ERROR( ZSTD_initLocalDict(cctx) , ""); /* Init the local dict if present. */
+    ZSTD_memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));   /* single usage */
+    assert(prefixDict.dict==NULL || cctx->cdict==NULL);    /* only one can be set */
+    if (cctx->cdict && !cctx->localDict.cdict) {
+        /* Let the cdict's compression level take priority over the requested params.
+         * But do not take the cdict's compression level if the "cdict" is actually a localDict
+         * generated from ZSTD_initLocalDict().
+         */
+        params.compressionLevel = cctx->cdict->compressionLevel;
+    }
+    DEBUGLOG(4, "ZSTD_CCtx_init_compressStream2 : transparent init stage");
+    if (endOp == ZSTD_e_end) cctx->pledgedSrcSizePlusOne = inSize + 1;  /* auto-determine pledgedSrcSize */
+
+    {   size_t const dictSize = prefixDict.dict
+                ? prefixDict.dictSize
+                : (cctx->cdict ? cctx->cdict->dictContentSize : 0);
+        ZSTD_CParamMode_e const mode = ZSTD_getCParamMode(cctx->cdict, &params, cctx->pledgedSrcSizePlusOne - 1);
+        params.cParams = ZSTD_getCParamsFromCCtxParams(
+                &params, cctx->pledgedSrcSizePlusOne-1,
+                dictSize, mode);
+    }
+
+    params.postBlockSplitter = ZSTD_resolveBlockSplitterMode(params.postBlockSplitter, &params.cParams);
+    params.ldmParams.enableLdm = ZSTD_resolveEnableLdm(params.ldmParams.enableLdm, &params.cParams);
+    params.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params.useRowMatchFinder, &params.cParams);
+    params.validateSequences = ZSTD_resolveExternalSequenceValidation(params.validateSequences);
+    params.maxBlockSize = ZSTD_resolveMaxBlockSize(params.maxBlockSize);
+    params.searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(params.searchForExternalRepcodes, params.compressionLevel);
+
+#ifdef ZSTD_MULTITHREAD
+    /* If external matchfinder is enabled, make sure to fail before checking job size (for consistency) */
+    RETURN_ERROR_IF(
+        ZSTD_hasExtSeqProd(&params) && params.nbWorkers >= 1,
+        parameter_combination_unsupported,
+        "External sequence producer isn't supported with nbWorkers >= 1"
+    );
+
+    if ((cctx->pledgedSrcSizePlusOne-1) <= ZSTDMT_JOBSIZE_MIN) {
+        params.nbWorkers = 0; /* do not invoke multi-threading when src size is too small */
+    }
+    if (params.nbWorkers > 0) {
+# if ZSTD_TRACE
+        cctx->traceCtx = (ZSTD_trace_compress_begin != NULL) ? ZSTD_trace_compress_begin(cctx) : 0;
+# endif
+        /* mt context creation */
+        if (cctx->mtctx == NULL) {
+            DEBUGLOG(4, "ZSTD_compressStream2: creating new mtctx for nbWorkers=%u",
+                        params.nbWorkers);
+            cctx->mtctx = ZSTDMT_createCCtx_advanced((U32)params.nbWorkers, cctx->customMem, cctx->pool);
+            RETURN_ERROR_IF(cctx->mtctx == NULL, memory_allocation, "NULL pointer!");
+        }
+        /* mt compression */
+        DEBUGLOG(4, "call ZSTDMT_initCStream_internal as nbWorkers=%u", params.nbWorkers);
+        FORWARD_IF_ERROR( ZSTDMT_initCStream_internal(
+                    cctx->mtctx,
+                    prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType,
+                    cctx->cdict, params, cctx->pledgedSrcSizePlusOne-1) , "");
+        cctx->dictID = cctx->cdict ? cctx->cdict->dictID : 0;
+        cctx->dictContentSize = cctx->cdict ? cctx->cdict->dictContentSize : prefixDict.dictSize;
+        cctx->consumedSrcSize = 0;
+        cctx->producedCSize = 0;
+        cctx->streamStage = zcss_load;
+        cctx->appliedParams = params;
+    } else
+#endif  /* ZSTD_MULTITHREAD */
+    {   U64 const pledgedSrcSize = cctx->pledgedSrcSizePlusOne - 1;
+        assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+        FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
+                prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType, ZSTD_dtlm_fast,
+                cctx->cdict,
+                &params, pledgedSrcSize,
+                ZSTDb_buffered) , "");
+        assert(cctx->appliedParams.nbWorkers == 0);
+        cctx->inToCompress = 0;
+        cctx->inBuffPos = 0;
+        if (cctx->appliedParams.inBufferMode == ZSTD_bm_buffered) {
+            /* for small input: avoid automatic flush on reaching end of block, since
+            * it would require to add a 3-bytes null block to end frame
+            */
+            cctx->inBuffTarget = cctx->blockSizeMax + (cctx->blockSizeMax == pledgedSrcSize);
+        } else {
+            cctx->inBuffTarget = 0;
+        }
+        cctx->outBuffContentSize = cctx->outBuffFlushedSize = 0;
+        cctx->streamStage = zcss_load;
+        cctx->frameEnded = 0;
+    }
+    return 0;
+}
+
+/* @return provides a minimum amount of data remaining to be flushed from internal buffers
+ */
+size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
+                             ZSTD_outBuffer* output,
+                             ZSTD_inBuffer* input,
+                             ZSTD_EndDirective endOp)
+{
+    DEBUGLOG(5, "ZSTD_compressStream2, endOp=%u ", (unsigned)endOp);
+    /* check conditions */
+    RETURN_ERROR_IF(output->pos > output->size, dstSize_tooSmall, "invalid output buffer");
+    RETURN_ERROR_IF(input->pos  > input->size, srcSize_wrong, "invalid input buffer");
+    RETURN_ERROR_IF((U32)endOp > (U32)ZSTD_e_end, parameter_outOfBound, "invalid endDirective");
+    assert(cctx != NULL);
+
+    /* transparent initialization stage */
+    if (cctx->streamStage == zcss_init) {
+        size_t const inputSize = input->size - input->pos;  /* no obligation to start from pos==0 */
+        size_t const totalInputSize = inputSize + cctx->stableIn_notConsumed;
+        if ( (cctx->requestedParams.inBufferMode == ZSTD_bm_stable) /* input is presumed stable, across invocations */
+          && (endOp == ZSTD_e_continue)                             /* no flush requested, more input to come */
+          && (totalInputSize < ZSTD_BLOCKSIZE_MAX) ) {              /* not even reached one block yet */
+            if (cctx->stableIn_notConsumed) {  /* not the first time */
+                /* check stable source guarantees */
+                RETURN_ERROR_IF(input->src != cctx->expectedInBuffer.src, stabilityCondition_notRespected, "stableInBuffer condition not respected: wrong src pointer");
+                RETURN_ERROR_IF(input->pos != cctx->expectedInBuffer.size, stabilityCondition_notRespected, "stableInBuffer condition not respected: externally modified pos");
+            }
+            /* pretend input was consumed, to give a sense forward progress */
+            input->pos = input->size;
+            /* save stable inBuffer, for later control, and flush/end */
+            cctx->expectedInBuffer = *input;
+            /* but actually input wasn't consumed, so keep track of position from where compression shall resume */
+            cctx->stableIn_notConsumed += inputSize;
+            /* don't initialize yet, wait for the first block of flush() order, for better parameters adaptation */
+            return ZSTD_FRAMEHEADERSIZE_MIN(cctx->requestedParams.format);  /* at least some header to produce */
+        }
+        FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, endOp, totalInputSize), "compressStream2 initialization failed");
+        ZSTD_setBufferExpectations(cctx, output, input);   /* Set initial buffer expectations now that we've initialized */
+    }
+    /* end of transparent initialization stage */
+
+    FORWARD_IF_ERROR(ZSTD_checkBufferStability(cctx, output, input, endOp), "invalid buffers");
+    /* compression stage */
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        size_t flushMin;
+        if (cctx->cParamsChanged) {
+            ZSTDMT_updateCParams_whileCompressing(cctx->mtctx, &cctx->requestedParams);
+            cctx->cParamsChanged = 0;
+        }
+        if (cctx->stableIn_notConsumed) {
+            assert(cctx->appliedParams.inBufferMode == ZSTD_bm_stable);
+            /* some early data was skipped - make it available for consumption */
+            assert(input->pos >= cctx->stableIn_notConsumed);
+            input->pos -= cctx->stableIn_notConsumed;
+            cctx->stableIn_notConsumed = 0;
+        }
+        for (;;) {
+            size_t const ipos = input->pos;
+            size_t const opos = output->pos;
+            flushMin = ZSTDMT_compressStream_generic(cctx->mtctx, output, input, endOp);
+            cctx->consumedSrcSize += (U64)(input->pos - ipos);
+            cctx->producedCSize += (U64)(output->pos - opos);
+            if ( ZSTD_isError(flushMin)
+              || (endOp == ZSTD_e_end && flushMin == 0) ) { /* compression completed */
+                if (flushMin == 0)
+                    ZSTD_CCtx_trace(cctx, 0);
+                ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only);
+            }
+            FORWARD_IF_ERROR(flushMin, "ZSTDMT_compressStream_generic failed");
+
+            if (endOp == ZSTD_e_continue) {
+                /* We only require some progress with ZSTD_e_continue, not maximal progress.
+                 * We're done if we've consumed or produced any bytes, or either buffer is
+                 * full.
+                 */
+                if (input->pos != ipos || output->pos != opos || input->pos == input->size || output->pos == output->size)
+                    break;
+            } else {
+                assert(endOp == ZSTD_e_flush || endOp == ZSTD_e_end);
+                /* We require maximal progress. We're done when the flush is complete or the
+                 * output buffer is full.
+                 */
+                if (flushMin == 0 || output->pos == output->size)
+                    break;
+            }
+        }
+        DEBUGLOG(5, "completed ZSTD_compressStream2 delegating to ZSTDMT_compressStream_generic");
+        /* Either we don't require maximum forward progress, we've finished the
+         * flush, or we are out of output space.
+         */
+        assert(endOp == ZSTD_e_continue || flushMin == 0 || output->pos == output->size);
+        ZSTD_setBufferExpectations(cctx, output, input);
+        return flushMin;
+    }
+#endif /* ZSTD_MULTITHREAD */
+    FORWARD_IF_ERROR( ZSTD_compressStream_generic(cctx, output, input, endOp) , "");
+    DEBUGLOG(5, "completed ZSTD_compressStream2");
+    ZSTD_setBufferExpectations(cctx, output, input);
+    return cctx->outBuffContentSize - cctx->outBuffFlushedSize; /* remaining to flush */
+}
+
+size_t ZSTD_compressStream2_simpleArgs (
+                            ZSTD_CCtx* cctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos,
+                            ZSTD_EndDirective endOp)
+{
+    ZSTD_outBuffer output;
+    ZSTD_inBuffer  input;
+    output.dst = dst;
+    output.size = dstCapacity;
+    output.pos = *dstPos;
+    input.src = src;
+    input.size = srcSize;
+    input.pos = *srcPos;
+    /* ZSTD_compressStream2() will check validity of dstPos and srcPos */
+    {   size_t const cErr = ZSTD_compressStream2(cctx, &output, &input, endOp);
+        *dstPos = output.pos;
+        *srcPos = input.pos;
+        return cErr;
+    }
+}
+
+size_t ZSTD_compress2(ZSTD_CCtx* cctx,
+                      void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    ZSTD_bufferMode_e const originalInBufferMode = cctx->requestedParams.inBufferMode;
+    ZSTD_bufferMode_e const originalOutBufferMode = cctx->requestedParams.outBufferMode;
+    DEBUGLOG(4, "ZSTD_compress2 (srcSize=%u)", (unsigned)srcSize);
+    ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only);
+    /* Enable stable input/output buffers. */
+    cctx->requestedParams.inBufferMode = ZSTD_bm_stable;
+    cctx->requestedParams.outBufferMode = ZSTD_bm_stable;
+    {   size_t oPos = 0;
+        size_t iPos = 0;
+        size_t const result = ZSTD_compressStream2_simpleArgs(cctx,
+                                        dst, dstCapacity, &oPos,
+                                        src, srcSize, &iPos,
+                                        ZSTD_e_end);
+        /* Reset to the original values. */
+        cctx->requestedParams.inBufferMode = originalInBufferMode;
+        cctx->requestedParams.outBufferMode = originalOutBufferMode;
+
+        FORWARD_IF_ERROR(result, "ZSTD_compressStream2_simpleArgs failed");
+        if (result != 0) {  /* compression not completed, due to lack of output space */
+            assert(oPos == dstCapacity);
+            RETURN_ERROR(dstSize_tooSmall, "");
+        }
+        assert(iPos == srcSize);   /* all input is expected consumed */
+        return oPos;
+    }
+}
+
+/* ZSTD_validateSequence() :
+ * @offBase : must use the format required by ZSTD_storeSeq()
+ * @returns a ZSTD error code if sequence is not valid
+ */
+static size_t
+ZSTD_validateSequence(U32 offBase, U32 matchLength, U32 minMatch,
+                      size_t posInSrc, U32 windowLog, size_t dictSize, int useSequenceProducer)
+{
+    U32 const windowSize = 1u << windowLog;
+    /* posInSrc represents the amount of data the decoder would decode up to this point.
+     * As long as the amount of data decoded is less than or equal to window size, offsets may be
+     * larger than the total length of output decoded in order to reference the dict, even larger than
+     * window size. After output surpasses windowSize, we're limited to windowSize offsets again.
+     */
+    size_t const offsetBound = posInSrc > windowSize ? (size_t)windowSize : posInSrc + (size_t)dictSize;
+    size_t const matchLenLowerBound = (minMatch == 3 || useSequenceProducer) ? 3 : 4;
+    RETURN_ERROR_IF(offBase > OFFSET_TO_OFFBASE(offsetBound), externalSequences_invalid, "Offset too large!");
+    /* Validate maxNbSeq is large enough for the given matchLength and minMatch */
+    RETURN_ERROR_IF(matchLength < matchLenLowerBound, externalSequences_invalid, "Matchlength too small for the minMatch");
+    return 0;
+}
+
+/* Returns an offset code, given a sequence's raw offset, the ongoing repcode array, and whether litLength == 0 */
+static U32 ZSTD_finalizeOffBase(U32 rawOffset, const U32 rep[ZSTD_REP_NUM], U32 ll0)
+{
+    U32 offBase = OFFSET_TO_OFFBASE(rawOffset);
+
+    if (!ll0 && rawOffset == rep[0]) {
+        offBase = REPCODE1_TO_OFFBASE;
+    } else if (rawOffset == rep[1]) {
+        offBase = REPCODE_TO_OFFBASE(2 - ll0);
+    } else if (rawOffset == rep[2]) {
+        offBase = REPCODE_TO_OFFBASE(3 - ll0);
+    } else if (ll0 && rawOffset == rep[0] - 1) {
+        offBase = REPCODE3_TO_OFFBASE;
+    }
+    return offBase;
+}
+
+/* This function scans through an array of ZSTD_Sequence,
+ * storing the sequences it reads, until it reaches a block delimiter.
+ * Note that the block delimiter includes the last literals of the block.
+ * @blockSize must be == sum(sequence_lengths).
+ * @returns @blockSize on success, and a ZSTD_error otherwise.
+ */
+static size_t
+ZSTD_transferSequences_wBlockDelim(ZSTD_CCtx* cctx,
+                                   ZSTD_SequencePosition* seqPos,
+                             const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                             const void* src, size_t blockSize,
+                                   ZSTD_ParamSwitch_e externalRepSearch)
+{
+    U32 idx = seqPos->idx;
+    U32 const startIdx = idx;
+    BYTE const* ip = (BYTE const*)(src);
+    const BYTE* const iend = ip + blockSize;
+    Repcodes_t updatedRepcodes;
+    U32 dictSize;
+
+    DEBUGLOG(5, "ZSTD_transferSequences_wBlockDelim (blockSize = %zu)", blockSize);
+
+    if (cctx->cdict) {
+        dictSize = (U32)cctx->cdict->dictContentSize;
+    } else if (cctx->prefixDict.dict) {
+        dictSize = (U32)cctx->prefixDict.dictSize;
+    } else {
+        dictSize = 0;
+    }
+    ZSTD_memcpy(updatedRepcodes.rep, cctx->blockState.prevCBlock->rep, sizeof(Repcodes_t));
+    for (; idx < inSeqsSize && (inSeqs[idx].matchLength != 0 || inSeqs[idx].offset != 0); ++idx) {
+        U32 const litLength = inSeqs[idx].litLength;
+        U32 const matchLength = inSeqs[idx].matchLength;
+        U32 offBase;
+
+        if (externalRepSearch == ZSTD_ps_disable) {
+            offBase = OFFSET_TO_OFFBASE(inSeqs[idx].offset);
+        } else {
+            U32 const ll0 = (litLength == 0);
+            offBase = ZSTD_finalizeOffBase(inSeqs[idx].offset, updatedRepcodes.rep, ll0);
+            ZSTD_updateRep(updatedRepcodes.rep, offBase, ll0);
+        }
+
+        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offBase, matchLength, litLength);
+        if (cctx->appliedParams.validateSequences) {
+            seqPos->posInSrc += litLength + matchLength;
+            FORWARD_IF_ERROR(ZSTD_validateSequence(offBase, matchLength, cctx->appliedParams.cParams.minMatch,
+                                                seqPos->posInSrc,
+                                                cctx->appliedParams.cParams.windowLog, dictSize,
+                                                ZSTD_hasExtSeqProd(&cctx->appliedParams)),
+                                                "Sequence validation failed");
+        }
+        RETURN_ERROR_IF(idx - seqPos->idx >= cctx->seqStore.maxNbSeq, externalSequences_invalid,
+                        "Not enough memory allocated. Try adjusting ZSTD_c_minMatch.");
+        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offBase, matchLength);
+        ip += matchLength + litLength;
+    }
+    RETURN_ERROR_IF(idx == inSeqsSize, externalSequences_invalid, "Block delimiter not found.");
+
+    /* If we skipped repcode search while parsing, we need to update repcodes now */
+    assert(externalRepSearch != ZSTD_ps_auto);
+    assert(idx >= startIdx);
+    if (externalRepSearch == ZSTD_ps_disable && idx != startIdx) {
+        U32* const rep = updatedRepcodes.rep;
+        U32 lastSeqIdx = idx - 1; /* index of last non-block-delimiter sequence */
+
+        if (lastSeqIdx >= startIdx + 2) {
+            rep[2] = inSeqs[lastSeqIdx - 2].offset;
+            rep[1] = inSeqs[lastSeqIdx - 1].offset;
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        } else if (lastSeqIdx == startIdx + 1) {
+            rep[2] = rep[0];
+            rep[1] = inSeqs[lastSeqIdx - 1].offset;
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        } else {
+            assert(lastSeqIdx == startIdx);
+            rep[2] = rep[1];
+            rep[1] = rep[0];
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        }
+    }
+
+    ZSTD_memcpy(cctx->blockState.nextCBlock->rep, updatedRepcodes.rep, sizeof(Repcodes_t));
+
+    if (inSeqs[idx].litLength) {
+        DEBUGLOG(6, "Storing last literals of size: %u", inSeqs[idx].litLength);
+        ZSTD_storeLastLiterals(&cctx->seqStore, ip, inSeqs[idx].litLength);
+        ip += inSeqs[idx].litLength;
+        seqPos->posInSrc += inSeqs[idx].litLength;
+    }
+    RETURN_ERROR_IF(ip != iend, externalSequences_invalid, "Blocksize doesn't agree with block delimiter!");
+    seqPos->idx = idx+1;
+    return blockSize;
+}
+
+/*
+ * This function attempts to scan through @blockSize bytes in @src
+ * represented by the sequences in @inSeqs,
+ * storing any (partial) sequences.
+ *
+ * Occasionally, we may want to reduce the actual number of bytes consumed from @src
+ * to avoid splitting a match, notably if it would produce a match smaller than MINMATCH.
+ *
+ * @returns the number of bytes consumed from @src, necessarily <= @blockSize.
+ * Otherwise, it may return a ZSTD error if something went wrong.
+ */
+static size_t
+ZSTD_transferSequences_noDelim(ZSTD_CCtx* cctx,
+                               ZSTD_SequencePosition* seqPos,
+                         const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                         const void* src, size_t blockSize,
+                               ZSTD_ParamSwitch_e externalRepSearch)
+{
+    U32 idx = seqPos->idx;
+    U32 startPosInSequence = seqPos->posInSequence;
+    U32 endPosInSequence = seqPos->posInSequence + (U32)blockSize;
+    size_t dictSize;
+    const BYTE* const istart = (const BYTE*)(src);
+    const BYTE* ip = istart;
+    const BYTE* iend = istart + blockSize;  /* May be adjusted if we decide to process fewer than blockSize bytes */
+    Repcodes_t updatedRepcodes;
+    U32 bytesAdjustment = 0;
+    U32 finalMatchSplit = 0;
+
+    /* TODO(embg) support fast parsing mode in noBlockDelim mode */
+    (void)externalRepSearch;
+
+    if (cctx->cdict) {
+        dictSize = cctx->cdict->dictContentSize;
+    } else if (cctx->prefixDict.dict) {
+        dictSize = cctx->prefixDict.dictSize;
+    } else {
+        dictSize = 0;
+    }
+    DEBUGLOG(5, "ZSTD_transferSequences_noDelim: idx: %u PIS: %u blockSize: %zu", idx, startPosInSequence, blockSize);
+    DEBUGLOG(5, "Start seq: idx: %u (of: %u ml: %u ll: %u)", idx, inSeqs[idx].offset, inSeqs[idx].matchLength, inSeqs[idx].litLength);
+    ZSTD_memcpy(updatedRepcodes.rep, cctx->blockState.prevCBlock->rep, sizeof(Repcodes_t));
+    while (endPosInSequence && idx < inSeqsSize && !finalMatchSplit) {
+        const ZSTD_Sequence currSeq = inSeqs[idx];
+        U32 litLength = currSeq.litLength;
+        U32 matchLength = currSeq.matchLength;
+        U32 const rawOffset = currSeq.offset;
+        U32 offBase;
+
+        /* Modify the sequence depending on where endPosInSequence lies */
+        if (endPosInSequence >= currSeq.litLength + currSeq.matchLength) {
+            if (startPosInSequence >= litLength) {
+                startPosInSequence -= litLength;
+                litLength = 0;
+                matchLength -= startPosInSequence;
+            } else {
+                litLength -= startPosInSequence;
+            }
+            /* Move to the next sequence */
+            endPosInSequence -= currSeq.litLength + currSeq.matchLength;
+            startPosInSequence = 0;
+        } else {
+            /* This is the final (partial) sequence we're adding from inSeqs, and endPosInSequence
+               does not reach the end of the match. So, we have to split the sequence */
+            DEBUGLOG(6, "Require a split: diff: %u, idx: %u PIS: %u",
+                     currSeq.litLength + currSeq.matchLength - endPosInSequence, idx, endPosInSequence);
+            if (endPosInSequence > litLength) {
+                U32 firstHalfMatchLength;
+                litLength = startPosInSequence >= litLength ? 0 : litLength - startPosInSequence;
+                firstHalfMatchLength = endPosInSequence - startPosInSequence - litLength;
+                if (matchLength > blockSize && firstHalfMatchLength >= cctx->appliedParams.cParams.minMatch) {
+                    /* Only ever split the match if it is larger than the block size */
+                    U32 secondHalfMatchLength = currSeq.matchLength + currSeq.litLength - endPosInSequence;
+                    if (secondHalfMatchLength < cctx->appliedParams.cParams.minMatch) {
+                        /* Move the endPosInSequence backward so that it creates match of minMatch length */
+                        endPosInSequence -= cctx->appliedParams.cParams.minMatch - secondHalfMatchLength;
+                        bytesAdjustment = cctx->appliedParams.cParams.minMatch - secondHalfMatchLength;
+                        firstHalfMatchLength -= bytesAdjustment;
+                    }
+                    matchLength = firstHalfMatchLength;
+                    /* Flag that we split the last match - after storing the sequence, exit the loop,
+                       but keep the value of endPosInSequence */
+                    finalMatchSplit = 1;
+                } else {
+                    /* Move the position in sequence backwards so that we don't split match, and break to store
+                     * the last literals. We use the original currSeq.litLength as a marker for where endPosInSequence
+                     * should go. We prefer to do this whenever it is not necessary to split the match, or if doing so
+                     * would cause the first half of the match to be too small
+                     */
+                    bytesAdjustment = endPosInSequence - currSeq.litLength;
+                    endPosInSequence = currSeq.litLength;
+                    break;
+                }
+            } else {
+                /* This sequence ends inside the literals, break to store the last literals */
+                break;
+            }
+        }
+        /* Check if this offset can be represented with a repcode */
+        {   U32 const ll0 = (litLength == 0);
+            offBase = ZSTD_finalizeOffBase(rawOffset, updatedRepcodes.rep, ll0);
+            ZSTD_updateRep(updatedRepcodes.rep, offBase, ll0);
+        }
+
+        if (cctx->appliedParams.validateSequences) {
+            seqPos->posInSrc += litLength + matchLength;
+            FORWARD_IF_ERROR(ZSTD_validateSequence(offBase, matchLength, cctx->appliedParams.cParams.minMatch, seqPos->posInSrc,
+                                                   cctx->appliedParams.cParams.windowLog, dictSize, ZSTD_hasExtSeqProd(&cctx->appliedParams)),
+                                                   "Sequence validation failed");
+        }
+        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offBase, matchLength, litLength);
+        RETURN_ERROR_IF(idx - seqPos->idx >= cctx->seqStore.maxNbSeq, externalSequences_invalid,
+                        "Not enough memory allocated. Try adjusting ZSTD_c_minMatch.");
+        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offBase, matchLength);
+        ip += matchLength + litLength;
+        if (!finalMatchSplit)
+            idx++; /* Next Sequence */
+    }
+    DEBUGLOG(5, "Ending seq: idx: %u (of: %u ml: %u ll: %u)", idx, inSeqs[idx].offset, inSeqs[idx].matchLength, inSeqs[idx].litLength);
+    assert(idx == inSeqsSize || endPosInSequence <= inSeqs[idx].litLength + inSeqs[idx].matchLength);
+    seqPos->idx = idx;
+    seqPos->posInSequence = endPosInSequence;
+    ZSTD_memcpy(cctx->blockState.nextCBlock->rep, updatedRepcodes.rep, sizeof(Repcodes_t));
+
+    iend -= bytesAdjustment;
+    if (ip != iend) {
+        /* Store any last literals */
+        U32 const lastLLSize = (U32)(iend - ip);
+        assert(ip <= iend);
+        DEBUGLOG(6, "Storing last literals of size: %u", lastLLSize);
+        ZSTD_storeLastLiterals(&cctx->seqStore, ip, lastLLSize);
+        seqPos->posInSrc += lastLLSize;
+    }
+
+    return (size_t)(iend-istart);
+}
+
+/* @seqPos represents a position within @inSeqs,
+ * it is read and updated by this function,
+ * once the goal to produce a block of size @blockSize is reached.
+ * @return: nb of bytes consumed from @src, necessarily <= @blockSize.
+ */
+typedef size_t (*ZSTD_SequenceCopier_f)(ZSTD_CCtx* cctx,
+                                        ZSTD_SequencePosition* seqPos,
+                                  const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                  const void* src, size_t blockSize,
+                                        ZSTD_ParamSwitch_e externalRepSearch);
+
+static ZSTD_SequenceCopier_f ZSTD_selectSequenceCopier(ZSTD_SequenceFormat_e mode)
+{
+    assert(ZSTD_cParam_withinBounds(ZSTD_c_blockDelimiters, (int)mode));
+    if (mode == ZSTD_sf_explicitBlockDelimiters) {
+        return ZSTD_transferSequences_wBlockDelim;
+    }
+    assert(mode == ZSTD_sf_noBlockDelimiters);
+    return ZSTD_transferSequences_noDelim;
+}
+
+/* Discover the size of next block by searching for the delimiter.
+ * Note that a block delimiter **must** exist in this mode,
+ * otherwise it's an input error.
+ * The block size retrieved will be later compared to ensure it remains within bounds */
+static size_t
+blockSize_explicitDelimiter(const ZSTD_Sequence* inSeqs, size_t inSeqsSize, ZSTD_SequencePosition seqPos)
+{
+    int end = 0;
+    size_t blockSize = 0;
+    size_t spos = seqPos.idx;
+    DEBUGLOG(6, "blockSize_explicitDelimiter : seq %zu / %zu", spos, inSeqsSize);
+    assert(spos <= inSeqsSize);
+    while (spos < inSeqsSize) {
+        end = (inSeqs[spos].offset == 0);
+        blockSize += inSeqs[spos].litLength + inSeqs[spos].matchLength;
+        if (end) {
+            if (inSeqs[spos].matchLength != 0)
+                RETURN_ERROR(externalSequences_invalid, "delimiter format error : both matchlength and offset must be == 0");
+            break;
+        }
+        spos++;
+    }
+    if (!end)
+        RETURN_ERROR(externalSequences_invalid, "Reached end of sequences without finding a block delimiter");
+    return blockSize;
+}
+
+static size_t determine_blockSize(ZSTD_SequenceFormat_e mode,
+                           size_t blockSize, size_t remaining,
+                     const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                           ZSTD_SequencePosition seqPos)
+{
+    DEBUGLOG(6, "determine_blockSize : remainingSize = %zu", remaining);
+    if (mode == ZSTD_sf_noBlockDelimiters) {
+        /* Note: more a "target" block size */
+        return MIN(remaining, blockSize);
+    }
+    assert(mode == ZSTD_sf_explicitBlockDelimiters);
+    {   size_t const explicitBlockSize = blockSize_explicitDelimiter(inSeqs, inSeqsSize, seqPos);
+        FORWARD_IF_ERROR(explicitBlockSize, "Error while determining block size with explicit delimiters");
+        if (explicitBlockSize > blockSize)
+            RETURN_ERROR(externalSequences_invalid, "sequences incorrectly define a too large block");
+        if (explicitBlockSize > remaining)
+            RETURN_ERROR(externalSequences_invalid, "sequences define a frame longer than source");
+        return explicitBlockSize;
+    }
+}
+
+/* Compress all provided sequences, block-by-block.
+ *
+ * Returns the cumulative size of all compressed blocks (including their headers),
+ * otherwise a ZSTD error.
+ */
+static size_t
+ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                          const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                          const void* src, size_t srcSize)
+{
+    size_t cSize = 0;
+    size_t remaining = srcSize;
+    ZSTD_SequencePosition seqPos = {0, 0, 0};
+
+    const BYTE* ip = (BYTE const*)src;
+    BYTE* op = (BYTE*)dst;
+    ZSTD_SequenceCopier_f const sequenceCopier = ZSTD_selectSequenceCopier(cctx->appliedParams.blockDelimiters);
+
+    DEBUGLOG(4, "ZSTD_compressSequences_internal srcSize: %zu, inSeqsSize: %zu", srcSize, inSeqsSize);
+    /* Special case: empty frame */
+    if (remaining == 0) {
+        U32 const cBlockHeader24 = 1 /* last block */ + (((U32)bt_raw)<<1);
+        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "No room for empty frame block header");
+        MEM_writeLE32(op, cBlockHeader24);
+        op += ZSTD_blockHeaderSize;
+        dstCapacity -= ZSTD_blockHeaderSize;
+        cSize += ZSTD_blockHeaderSize;
+    }
+
+    while (remaining) {
+        size_t compressedSeqsSize;
+        size_t cBlockSize;
+        size_t blockSize = determine_blockSize(cctx->appliedParams.blockDelimiters,
+                                        cctx->blockSizeMax, remaining,
+                                        inSeqs, inSeqsSize, seqPos);
+        U32 const lastBlock = (blockSize == remaining);
+        FORWARD_IF_ERROR(blockSize, "Error while trying to determine block size");
+        assert(blockSize <= remaining);
+        ZSTD_resetSeqStore(&cctx->seqStore);
+
+        blockSize = sequenceCopier(cctx,
+                                   &seqPos, inSeqs, inSeqsSize,
+                                   ip, blockSize,
+                                   cctx->appliedParams.searchForExternalRepcodes);
+        FORWARD_IF_ERROR(blockSize, "Bad sequence copy");
+
+        /* If blocks are too small, emit as a nocompress block */
+        /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+         * additional 1. We need to revisit and change this logic to be more consistent */
+        if (blockSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1+1) {
+            cBlockSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
+            FORWARD_IF_ERROR(cBlockSize, "Nocompress block failed");
+            DEBUGLOG(5, "Block too small (%zu): data remains uncompressed: cSize=%zu", blockSize, cBlockSize);
+            cSize += cBlockSize;
+            ip += blockSize;
+            op += cBlockSize;
+            remaining -= blockSize;
+            dstCapacity -= cBlockSize;
+            continue;
+        }
+
+        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall, "not enough dstCapacity to write a new compressed block");
+        compressedSeqsSize = ZSTD_entropyCompressSeqStore(&cctx->seqStore,
+                                &cctx->blockState.prevCBlock->entropy, &cctx->blockState.nextCBlock->entropy,
+                                &cctx->appliedParams,
+                                op + ZSTD_blockHeaderSize /* Leave space for block header */, dstCapacity - ZSTD_blockHeaderSize,
+                                blockSize,
+                                cctx->tmpWorkspace, cctx->tmpWkspSize /* statically allocated in resetCCtx */,
+                                cctx->bmi2);
+        FORWARD_IF_ERROR(compressedSeqsSize, "Compressing sequences of block failed");
+        DEBUGLOG(5, "Compressed sequences size: %zu", compressedSeqsSize);
+
+        if (!cctx->isFirstBlock &&
+            ZSTD_maybeRLE(&cctx->seqStore) &&
+            ZSTD_isRLE(ip, blockSize)) {
+            /* Note: don't emit the first block as RLE even if it qualifies because
+             * doing so will cause the decoder (cli <= v1.4.3 only) to throw an (invalid) error
+             * "should consume all input error."
+             */
+            compressedSeqsSize = 1;
+        }
+
+        if (compressedSeqsSize == 0) {
+            /* ZSTD_noCompressBlock writes the block header as well */
+            cBlockSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
+            FORWARD_IF_ERROR(cBlockSize, "ZSTD_noCompressBlock failed");
+            DEBUGLOG(5, "Writing out nocompress block, size: %zu", cBlockSize);
+        } else if (compressedSeqsSize == 1) {
+            cBlockSize = ZSTD_rleCompressBlock(op, dstCapacity, *ip, blockSize, lastBlock);
+            FORWARD_IF_ERROR(cBlockSize, "ZSTD_rleCompressBlock failed");
+            DEBUGLOG(5, "Writing out RLE block, size: %zu", cBlockSize);
+        } else {
+            U32 cBlockHeader;
+            /* Error checking and repcodes update */
+            ZSTD_blockState_confirmRepcodesAndEntropyTables(&cctx->blockState);
+            if (cctx->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+                cctx->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+            /* Write block header into beginning of block*/
+            cBlockHeader = lastBlock + (((U32)bt_compressed)<<1) + (U32)(compressedSeqsSize << 3);
+            MEM_writeLE24(op, cBlockHeader);
+            cBlockSize = ZSTD_blockHeaderSize + compressedSeqsSize;
+            DEBUGLOG(5, "Writing out compressed block, size: %zu", cBlockSize);
+        }
+
+        cSize += cBlockSize;
+
+        if (lastBlock) {
+            break;
+        } else {
+            ip += blockSize;
+            op += cBlockSize;
+            remaining -= blockSize;
+            dstCapacity -= cBlockSize;
+            cctx->isFirstBlock = 0;
+        }
+        DEBUGLOG(5, "cSize running total: %zu (remaining dstCapacity=%zu)", cSize, dstCapacity);
+    }
+
+    DEBUGLOG(4, "cSize final total: %zu", cSize);
+    return cSize;
+}
+
+size_t ZSTD_compressSequences(ZSTD_CCtx* cctx,
+                              void* dst, size_t dstCapacity,
+                              const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                              const void* src, size_t srcSize)
+{
+    BYTE* op = (BYTE*)dst;
+    size_t cSize = 0;
+
+    /* Transparent initialization stage, same as compressStream2() */
+    DEBUGLOG(4, "ZSTD_compressSequences (nbSeqs=%zu,dstCapacity=%zu)", inSeqsSize, dstCapacity);
+    assert(cctx != NULL);
+    FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, ZSTD_e_end, srcSize), "CCtx initialization failed");
+
+    /* Begin writing output, starting with frame header */
+    {   size_t const frameHeaderSize = ZSTD_writeFrameHeader(op, dstCapacity,
+                    &cctx->appliedParams, srcSize, cctx->dictID);
+        op += frameHeaderSize;
+        assert(frameHeaderSize <= dstCapacity);
+        dstCapacity -= frameHeaderSize;
+        cSize += frameHeaderSize;
+    }
+    if (cctx->appliedParams.fParams.checksumFlag && srcSize) {
+        XXH64_update(&cctx->xxhState, src, srcSize);
+    }
+
+    /* Now generate compressed blocks */
+    {   size_t const cBlocksSize = ZSTD_compressSequences_internal(cctx,
+                                                           op, dstCapacity,
+                                                           inSeqs, inSeqsSize,
+                                                           src, srcSize);
+        FORWARD_IF_ERROR(cBlocksSize, "Compressing blocks failed!");
+        cSize += cBlocksSize;
+        assert(cBlocksSize <= dstCapacity);
+        dstCapacity -= cBlocksSize;
+    }
+
+    /* Complete with frame checksum, if needed */
+    if (cctx->appliedParams.fParams.checksumFlag) {
+        U32 const checksum = (U32) XXH64_digest(&cctx->xxhState);
+        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "no room for checksum");
+        DEBUGLOG(4, "Write checksum : %08X", (unsigned)checksum);
+        MEM_writeLE32((char*)dst + cSize, checksum);
+        cSize += 4;
+    }
+
+    DEBUGLOG(4, "Final compressed size: %zu", cSize);
+    return cSize;
+}
+
+
+#if defined(__AVX2__)
+
+#include <immintrin.h>  /* AVX2 intrinsics */
+
+/*
+ * Convert 2 sequences per iteration, using AVX2 intrinsics:
+ *   - offset -> offBase = offset + 2
+ *   - litLength -> (U16) litLength
+ *   - matchLength -> (U16)(matchLength - 3)
+ *   - rep is ignored
+ * Store only 8 bytes per SeqDef (offBase[4], litLength[2], mlBase[2]).
+ *
+ * At the end, instead of extracting two __m128i,
+ * we use _mm256_permute4x64_epi64(..., 0xE8) to move lane2 into lane1,
+ * then store the lower 16 bytes in one go.
+ *
+ * @returns 0 on succes, with no long length detected
+ * @returns > 0 if there is one long length (> 65535),
+ * indicating the position, and type.
+ */
+static size_t convertSequences_noRepcodes(
+    SeqDef* dstSeqs,
+    const ZSTD_Sequence* inSeqs,
+    size_t nbSequences)
+{
+    /*
+     * addition:
+     *   For each 128-bit half: (offset+2, litLength+0, matchLength-3, rep+0)
+     */
+    const __m256i addition = _mm256_setr_epi32(
+        ZSTD_REP_NUM, 0, -MINMATCH, 0,    /* for sequence i */
+        ZSTD_REP_NUM, 0, -MINMATCH, 0     /* for sequence i+1 */
+    );
+
+    /* limit: check if there is a long length */
+    const __m256i limit = _mm256_set1_epi32(65535);
+
+    /*
+     * shuffle mask for byte-level rearrangement in each 128-bit half:
+     *
+     * Input layout (after addition) per 128-bit half:
+     *   [ offset+2 (4 bytes) | litLength (4 bytes) | matchLength (4 bytes) | rep (4 bytes) ]
+     * We only need:
+     *   offBase (4 bytes) = offset+2
+     *   litLength (2 bytes) = low 2 bytes of litLength
+     *   mlBase (2 bytes) = low 2 bytes of (matchLength)
+     * => Bytes [0..3, 4..5, 8..9], zero the rest.
+     */
+    const __m256i mask = _mm256_setr_epi8(
+        /* For the lower 128 bits => sequence i */
+         0, 1, 2, 3,       /* offset+2 */
+         4, 5,             /* litLength (16 bits) */
+         8, 9,             /* matchLength (16 bits) */
+         (BYTE)0x80, (BYTE)0x80, (BYTE)0x80, (BYTE)0x80,
+         (BYTE)0x80, (BYTE)0x80, (BYTE)0x80, (BYTE)0x80,
+
+        /* For the upper 128 bits => sequence i+1 */
+        16,17,18,19,       /* offset+2 */
+        20,21,             /* litLength */
+        24,25,             /* matchLength */
+        (BYTE)0x80, (BYTE)0x80, (BYTE)0x80, (BYTE)0x80,
+        (BYTE)0x80, (BYTE)0x80, (BYTE)0x80, (BYTE)0x80
+    );
+
+    /*
+     * Next, we'll use _mm256_permute4x64_epi64(vshf, 0xE8).
+     * Explanation of 0xE8 = 11101000b => [lane0, lane2, lane2, lane3].
+     * So the lower 128 bits become [lane0, lane2] => combining seq0 and seq1.
+     */
+#define PERM_LANE_0X_E8 0xE8  /* [0,2,2,3] in lane indices */
+
+    size_t longLen = 0, i = 0;
+
+    /* AVX permutation depends on the specific definition of target structures */
+    ZSTD_STATIC_ASSERT(sizeof(ZSTD_Sequence) == 16);
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_Sequence, offset) == 0);
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_Sequence, litLength) == 4);
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_Sequence, matchLength) == 8);
+    ZSTD_STATIC_ASSERT(sizeof(SeqDef) == 8);
+    ZSTD_STATIC_ASSERT(offsetof(SeqDef, offBase) == 0);
+    ZSTD_STATIC_ASSERT(offsetof(SeqDef, litLength) == 4);
+    ZSTD_STATIC_ASSERT(offsetof(SeqDef, mlBase) == 6);
+
+    /* Process 2 sequences per loop iteration */
+    for (; i + 1 < nbSequences; i += 2) {
+        /* Load 2 ZSTD_Sequence (32 bytes) */
+        __m256i vin  = _mm256_loadu_si256((const __m256i*)(const void*)&inSeqs[i]);
+
+        /* Add {2, 0, -3, 0} in each 128-bit half */
+        __m256i vadd = _mm256_add_epi32(vin, addition);
+
+        /* Check for long length */
+        __m256i ll_cmp  = _mm256_cmpgt_epi32(vadd, limit);  /* 0xFFFFFFFF for element > 65535 */
+        int ll_res  = _mm256_movemask_epi8(ll_cmp);
+
+        /* Shuffle bytes so each half gives us the 8 bytes we need */
+        __m256i vshf = _mm256_shuffle_epi8(vadd, mask);
+        /*
+         * Now:
+         *   Lane0 = seq0's 8 bytes
+         *   Lane1 = 0
+         *   Lane2 = seq1's 8 bytes
+         *   Lane3 = 0
+         */
+
+        /* Permute 64-bit lanes => move Lane2 down into Lane1. */
+        __m256i vperm = _mm256_permute4x64_epi64(vshf, PERM_LANE_0X_E8);
+        /*
+         * Now the lower 16 bytes (Lane0+Lane1) = [seq0, seq1].
+         * The upper 16 bytes are [Lane2, Lane3] = [seq1, 0], but we won't use them.
+         */
+
+        /* Store only the lower 16 bytes => 2 SeqDef (8 bytes each) */
+        _mm_storeu_si128((__m128i *)(void*)&dstSeqs[i], _mm256_castsi256_si128(vperm));
+        /*
+         * This writes out 16 bytes total:
+         *   - offset 0..7  => seq0 (offBase, litLength, mlBase)
+         *   - offset 8..15 => seq1 (offBase, litLength, mlBase)
+         */
+
+        /* check (unlikely) long lengths > 65535
+         * indices for lengths correspond to bits [4..7], [8..11], [20..23], [24..27]
+         * => combined mask = 0x0FF00FF0
+         */
+        if (UNLIKELY((ll_res & 0x0FF00FF0) != 0)) {
+            /* long length detected: let's figure out which one*/
+            if (inSeqs[i].matchLength > 65535+MINMATCH) {
+                assert(longLen == 0);
+                longLen = i + 1;
+            }
+            if (inSeqs[i].litLength > 65535) {
+                assert(longLen == 0);
+                longLen = i + nbSequences + 1;
+            }
+            if (inSeqs[i+1].matchLength > 65535+MINMATCH) {
+                assert(longLen == 0);
+                longLen = i + 1 + 1;
+            }
+            if (inSeqs[i+1].litLength > 65535) {
+                assert(longLen == 0);
+                longLen = i + 1 + nbSequences + 1;
+            }
+        }
+    }
+
+    /* Handle leftover if @nbSequences is odd */
+    if (i < nbSequences) {
+        /* process last sequence */
+        assert(i == nbSequences - 1);
+        dstSeqs[i].offBase = OFFSET_TO_OFFBASE(inSeqs[i].offset);
+        dstSeqs[i].litLength = (U16)inSeqs[i].litLength;
+        dstSeqs[i].mlBase = (U16)(inSeqs[i].matchLength - MINMATCH);
+        /* check (unlikely) long lengths > 65535 */
+        if (UNLIKELY(inSeqs[i].matchLength > 65535+MINMATCH)) {
+            assert(longLen == 0);
+            longLen = i + 1;
+        }
+        if (UNLIKELY(inSeqs[i].litLength > 65535)) {
+            assert(longLen == 0);
+            longLen = i + nbSequences + 1;
+        }
+    }
+
+    return longLen;
+}
+
+/* the vector implementation could also be ported to SSSE3,
+ * but since this implementation is targeting modern systems (>= Sapphire Rapid),
+ * it's not useful to develop and maintain code for older pre-AVX2 platforms */
+
+#else /* no AVX2 */
+
+static size_t convertSequences_noRepcodes(
+    SeqDef* dstSeqs,
+    const ZSTD_Sequence* inSeqs,
+    size_t nbSequences)
+{
+    size_t longLen = 0;
+    size_t n;
+    for (n=0; n<nbSequences; n++) {
+        dstSeqs[n].offBase = OFFSET_TO_OFFBASE(inSeqs[n].offset);
+        dstSeqs[n].litLength = (U16)inSeqs[n].litLength;
+        dstSeqs[n].mlBase = (U16)(inSeqs[n].matchLength - MINMATCH);
+        /* check for long length > 65535 */
+        if (UNLIKELY(inSeqs[n].matchLength > 65535+MINMATCH)) {
+            assert(longLen == 0);
+            longLen = n + 1;
+        }
+        if (UNLIKELY(inSeqs[n].litLength > 65535)) {
+            assert(longLen == 0);
+            longLen = n + nbSequences + 1;
+        }
+    }
+    return longLen;
+}
+
+#endif
+
+/*
+ * Precondition: Sequences must end on an explicit Block Delimiter
+ * @return: 0 on success, or an error code.
+ * Note: Sequence validation functionality has been disabled (removed).
+ * This is helpful to generate a lean main pipeline, improving performance.
+ * It may be re-inserted later.
+ */
+size_t ZSTD_convertBlockSequences(ZSTD_CCtx* cctx,
+                const ZSTD_Sequence* const inSeqs, size_t nbSequences,
+                int repcodeResolution)
+{
+    Repcodes_t updatedRepcodes;
+    size_t seqNb = 0;
+
+    DEBUGLOG(5, "ZSTD_convertBlockSequences (nbSequences = %zu)", nbSequences);
+
+    RETURN_ERROR_IF(nbSequences >= cctx->seqStore.maxNbSeq, externalSequences_invalid,
+                    "Not enough memory allocated. Try adjusting ZSTD_c_minMatch.");
+
+    ZSTD_memcpy(updatedRepcodes.rep, cctx->blockState.prevCBlock->rep, sizeof(Repcodes_t));
+
+    /* check end condition */
+    assert(nbSequences >= 1);
+    assert(inSeqs[nbSequences-1].matchLength == 0);
+    assert(inSeqs[nbSequences-1].offset == 0);
+
+    /* Convert Sequences from public format to internal format */
+    if (!repcodeResolution) {
+        size_t const longl = convertSequences_noRepcodes(cctx->seqStore.sequencesStart, inSeqs, nbSequences-1);
+        cctx->seqStore.sequences = cctx->seqStore.sequencesStart + nbSequences-1;
+        if (longl) {
+            DEBUGLOG(5, "long length");
+            assert(cctx->seqStore.longLengthType == ZSTD_llt_none);
+            if (longl <= nbSequences-1) {
+                DEBUGLOG(5, "long match length detected at pos %zu", longl-1);
+                cctx->seqStore.longLengthType = ZSTD_llt_matchLength;
+                cctx->seqStore.longLengthPos = (U32)(longl-1);
+            } else {
+                DEBUGLOG(5, "long literals length detected at pos %zu", longl-nbSequences);
+                assert(longl <= 2* (nbSequences-1));
+                cctx->seqStore.longLengthType = ZSTD_llt_literalLength;
+                cctx->seqStore.longLengthPos = (U32)(longl-(nbSequences-1)-1);
+            }
+        }
+    } else {
+        for (seqNb = 0; seqNb < nbSequences - 1 ; seqNb++) {
+            U32 const litLength = inSeqs[seqNb].litLength;
+            U32 const matchLength = inSeqs[seqNb].matchLength;
+            U32 const ll0 = (litLength == 0);
+            U32 const offBase = ZSTD_finalizeOffBase(inSeqs[seqNb].offset, updatedRepcodes.rep, ll0);
+
+            DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offBase, matchLength, litLength);
+            ZSTD_storeSeqOnly(&cctx->seqStore, litLength, offBase, matchLength);
+            ZSTD_updateRep(updatedRepcodes.rep, offBase, ll0);
+        }
+    }
+
+    /* If we skipped repcode search while parsing, we need to update repcodes now */
+    if (!repcodeResolution && nbSequences > 1) {
+        U32* const rep = updatedRepcodes.rep;
+
+        if (nbSequences >= 4) {
+            U32 lastSeqIdx = (U32)nbSequences - 2; /* index of last full sequence */
+            rep[2] = inSeqs[lastSeqIdx - 2].offset;
+            rep[1] = inSeqs[lastSeqIdx - 1].offset;
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        } else if (nbSequences == 3) {
+            rep[2] = rep[0];
+            rep[1] = inSeqs[0].offset;
+            rep[0] = inSeqs[1].offset;
+        } else {
+            assert(nbSequences == 2);
+            rep[2] = rep[1];
+            rep[1] = rep[0];
+            rep[0] = inSeqs[0].offset;
+        }
+    }
+
+    ZSTD_memcpy(cctx->blockState.nextCBlock->rep, updatedRepcodes.rep, sizeof(Repcodes_t));
+
+    return 0;
+}
+
+#if defined(ZSTD_ARCH_X86_AVX2)
+
+BlockSummary ZSTD_get1BlockSummary(const ZSTD_Sequence* seqs, size_t nbSeqs)
+{
+    size_t i;
+    __m256i const zeroVec = _mm256_setzero_si256();
+    __m256i sumVec = zeroVec;  /* accumulates match+lit in 32-bit lanes */
+    ZSTD_ALIGNED(32) U32 tmp[8];      /* temporary buffer for reduction */
+    size_t mSum = 0, lSum = 0;
+    ZSTD_STATIC_ASSERT(sizeof(ZSTD_Sequence) == 16);
+
+    /* Process 2 structs (32 bytes) at a time */
+    for (i = 0; i + 2 <= nbSeqs; i += 2) {
+        /* Load two consecutive ZSTD_Sequence (8×4 = 32 bytes) */
+        __m256i data     = _mm256_loadu_si256((const __m256i*)(const void*)&seqs[i]);
+        /* check end of block signal */
+        __m256i cmp      = _mm256_cmpeq_epi32(data, zeroVec);
+        int cmp_res      = _mm256_movemask_epi8(cmp);
+        /* indices for match lengths correspond to bits [8..11], [24..27]
+         * => combined mask = 0x0F000F00 */
+        ZSTD_STATIC_ASSERT(offsetof(ZSTD_Sequence, matchLength) == 8);
+        if (cmp_res & 0x0F000F00) break;
+        /* Accumulate in sumVec */
+        sumVec           = _mm256_add_epi32(sumVec, data);
+    }
+
+    /* Horizontal reduction */
+    _mm256_store_si256((__m256i*)tmp, sumVec);
+    lSum = tmp[1] + tmp[5];
+    mSum = tmp[2] + tmp[6];
+
+    /* Handle the leftover */
+    for (; i < nbSeqs; i++) {
+        lSum += seqs[i].litLength;
+        mSum += seqs[i].matchLength;
+        if (seqs[i].matchLength == 0) break; /* end of block */
+    }
+
+    if (i==nbSeqs) {
+        /* reaching end of sequences: end of block signal was not present */
+        BlockSummary bs;
+        bs.nbSequences = ERROR(externalSequences_invalid);
+        return bs;
+    }
+    {   BlockSummary bs;
+        bs.nbSequences = i+1;
+        bs.blockSize = lSum + mSum;
+        bs.litSize = lSum;
+        return bs;
+    }
+}
+
+#else
+
+BlockSummary ZSTD_get1BlockSummary(const ZSTD_Sequence* seqs, size_t nbSeqs)
+{
+    size_t totalMatchSize = 0;
+    size_t litSize = 0;
+    size_t n;
+    assert(seqs);
+    for (n=0; n<nbSeqs; n++) {
+        totalMatchSize += seqs[n].matchLength;
+        litSize += seqs[n].litLength;
+        if (seqs[n].matchLength == 0) {
+            assert(seqs[n].offset == 0);
+            break;
+        }
+    }
+    if (n==nbSeqs) {
+        BlockSummary bs;
+        bs.nbSequences = ERROR(externalSequences_invalid);
+        return bs;
+    }
+    {   BlockSummary bs;
+        bs.nbSequences = n+1;
+        bs.blockSize = litSize + totalMatchSize;
+        bs.litSize = litSize;
+        return bs;
+    }
+}
+#endif
+
+
+static size_t
+ZSTD_compressSequencesAndLiterals_internal(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                          const ZSTD_Sequence* inSeqs, size_t nbSequences,
+                          const void* literals, size_t litSize, size_t srcSize)
+{
+    size_t remaining = srcSize;
+    size_t cSize = 0;
+    BYTE* op = (BYTE*)dst;
+    int const repcodeResolution = (cctx->appliedParams.searchForExternalRepcodes == ZSTD_ps_enable);
+    assert(cctx->appliedParams.searchForExternalRepcodes != ZSTD_ps_auto);
+
+    DEBUGLOG(4, "ZSTD_compressSequencesAndLiterals_internal: nbSeqs=%zu, litSize=%zu", nbSequences, litSize);
+    RETURN_ERROR_IF(nbSequences == 0, externalSequences_invalid, "Requires at least 1 end-of-block");
+
+    /* Special case: empty frame */
+    if ((nbSequences == 1) && (inSeqs[0].litLength == 0)) {
+        U32 const cBlockHeader24 = 1 /* last block */ + (((U32)bt_raw)<<1);
+        RETURN_ERROR_IF(dstCapacity<3, dstSize_tooSmall, "No room for empty frame block header");
+        MEM_writeLE24(op, cBlockHeader24);
+        op += ZSTD_blockHeaderSize;
+        dstCapacity -= ZSTD_blockHeaderSize;
+        cSize += ZSTD_blockHeaderSize;
+    }
+
+    while (nbSequences) {
+        size_t compressedSeqsSize, cBlockSize, conversionStatus;
+        BlockSummary const block = ZSTD_get1BlockSummary(inSeqs, nbSequences);
+        U32 const lastBlock = (block.nbSequences == nbSequences);
+        FORWARD_IF_ERROR(block.nbSequences, "Error while trying to determine nb of sequences for a block");
+        assert(block.nbSequences <= nbSequences);
+        RETURN_ERROR_IF(block.litSize > litSize, externalSequences_invalid, "discrepancy: Sequences require more literals than present in buffer");
+        ZSTD_resetSeqStore(&cctx->seqStore);
+
+        conversionStatus = ZSTD_convertBlockSequences(cctx,
+                            inSeqs, block.nbSequences,
+                            repcodeResolution);
+        FORWARD_IF_ERROR(conversionStatus, "Bad sequence conversion");
+        inSeqs += block.nbSequences;
+        nbSequences -= block.nbSequences;
+        remaining -= block.blockSize;
+
+        /* Note: when blockSize is very small, other variant send it uncompressed.
+         * Here, we still send the sequences, because we don't have the original source to send it uncompressed.
+         * One could imagine in theory reproducing the source from the sequences,
+         * but that's complex and costly memory intensive, and goes against the objectives of this variant. */
+
+        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall, "not enough dstCapacity to write a new compressed block");
+
+        compressedSeqsSize = ZSTD_entropyCompressSeqStore_internal(
+                                op + ZSTD_blockHeaderSize /* Leave space for block header */, dstCapacity - ZSTD_blockHeaderSize,
+                                literals, block.litSize,
+                                &cctx->seqStore,
+                                &cctx->blockState.prevCBlock->entropy, &cctx->blockState.nextCBlock->entropy,
+                                &cctx->appliedParams,
+                                cctx->tmpWorkspace, cctx->tmpWkspSize /* statically allocated in resetCCtx */,
+                                cctx->bmi2);
+        FORWARD_IF_ERROR(compressedSeqsSize, "Compressing sequences of block failed");
+        /* note: the spec forbids for any compressed block to be larger than maximum block size */
+        if (compressedSeqsSize > cctx->blockSizeMax) compressedSeqsSize = 0;
+        DEBUGLOG(5, "Compressed sequences size: %zu", compressedSeqsSize);
+        litSize -= block.litSize;
+        literals = (const char*)literals + block.litSize;
+
+        /* Note: difficult to check source for RLE block when only Literals are provided,
+         * but it could be considered from analyzing the sequence directly */
+
+        if (compressedSeqsSize == 0) {
+            /* Sending uncompressed blocks is out of reach, because the source is not provided.
+             * In theory, one could use the sequences to regenerate the source, like a decompressor,
+             * but it's complex, and memory hungry, killing the purpose of this variant.
+             * Current outcome: generate an error code.
+             */
+            RETURN_ERROR(cannotProduce_uncompressedBlock, "ZSTD_compressSequencesAndLiterals cannot generate an uncompressed block");
+        } else {
+            U32 cBlockHeader;
+            assert(compressedSeqsSize > 1); /* no RLE */
+            /* Error checking and repcodes update */
+            ZSTD_blockState_confirmRepcodesAndEntropyTables(&cctx->blockState);
+            if (cctx->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+                cctx->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+            /* Write block header into beginning of block*/
+            cBlockHeader = lastBlock + (((U32)bt_compressed)<<1) + (U32)(compressedSeqsSize << 3);
+            MEM_writeLE24(op, cBlockHeader);
+            cBlockSize = ZSTD_blockHeaderSize + compressedSeqsSize;
+            DEBUGLOG(5, "Writing out compressed block, size: %zu", cBlockSize);
+        }
+
+        cSize += cBlockSize;
+        op += cBlockSize;
+        dstCapacity -= cBlockSize;
+        cctx->isFirstBlock = 0;
+        DEBUGLOG(5, "cSize running total: %zu (remaining dstCapacity=%zu)", cSize, dstCapacity);
+
+        if (lastBlock) {
+            assert(nbSequences == 0);
+            break;
+        }
+    }
+
+    RETURN_ERROR_IF(litSize != 0, externalSequences_invalid, "literals must be entirely and exactly consumed");
+    RETURN_ERROR_IF(remaining != 0, externalSequences_invalid, "Sequences must represent a total of exactly srcSize=%zu", srcSize);
+    DEBUGLOG(4, "cSize final total: %zu", cSize);
+    return cSize;
+}
+
+size_t
+ZSTD_compressSequencesAndLiterals(ZSTD_CCtx* cctx,
+                    void* dst, size_t dstCapacity,
+                    const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                    const void* literals, size_t litSize, size_t litCapacity,
+                    size_t decompressedSize)
+{
+    BYTE* op = (BYTE*)dst;
+    size_t cSize = 0;
+
+    /* Transparent initialization stage, same as compressStream2() */
+    DEBUGLOG(4, "ZSTD_compressSequencesAndLiterals (dstCapacity=%zu)", dstCapacity);
+    assert(cctx != NULL);
+    if (litCapacity < litSize) {
+        RETURN_ERROR(workSpace_tooSmall, "literals buffer is not large enough: must be at least 8 bytes larger than litSize (risk of read out-of-bound)");
+    }
+    FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, ZSTD_e_end, decompressedSize), "CCtx initialization failed");
+
+    if (cctx->appliedParams.blockDelimiters == ZSTD_sf_noBlockDelimiters) {
+        RETURN_ERROR(frameParameter_unsupported, "This mode is only compatible with explicit delimiters");
+    }
+    if (cctx->appliedParams.validateSequences) {
+        RETURN_ERROR(parameter_unsupported, "This mode is not compatible with Sequence validation");
+    }
+    if (cctx->appliedParams.fParams.checksumFlag) {
+        RETURN_ERROR(frameParameter_unsupported, "this mode is not compatible with frame checksum");
+    }
+
+    /* Begin writing output, starting with frame header */
+    {   size_t const frameHeaderSize = ZSTD_writeFrameHeader(op, dstCapacity,
+                    &cctx->appliedParams, decompressedSize, cctx->dictID);
+        op += frameHeaderSize;
+        assert(frameHeaderSize <= dstCapacity);
+        dstCapacity -= frameHeaderSize;
+        cSize += frameHeaderSize;
+    }
+
+    /* Now generate compressed blocks */
+    {   size_t const cBlocksSize = ZSTD_compressSequencesAndLiterals_internal(cctx,
+                                            op, dstCapacity,
+                                            inSeqs, inSeqsSize,
+                                            literals, litSize, decompressedSize);
+        FORWARD_IF_ERROR(cBlocksSize, "Compressing blocks failed!");
+        cSize += cBlocksSize;
+        assert(cBlocksSize <= dstCapacity);
+        dstCapacity -= cBlocksSize;
+    }
+
+    DEBUGLOG(4, "Final compressed size: %zu", cSize);
+    return cSize;
+}
+
+/*======   Finalize   ======*/
+
+static ZSTD_inBuffer inBuffer_forEndFlush(const ZSTD_CStream* zcs)
+{
+    const ZSTD_inBuffer nullInput = { NULL, 0, 0 };
+    const int stableInput = (zcs->appliedParams.inBufferMode == ZSTD_bm_stable);
+    return stableInput ? zcs->expectedInBuffer : nullInput;
+}
+
+/*! ZSTD_flushStream() :
+ * @return : amount of data remaining to flush */
+size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
+{
+    ZSTD_inBuffer input = inBuffer_forEndFlush(zcs);
+    input.size = input.pos; /* do not ingest more input during flush */
+    return ZSTD_compressStream2(zcs, output, &input, ZSTD_e_flush);
+}
+
+size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
+{
+    ZSTD_inBuffer input = inBuffer_forEndFlush(zcs);
+    size_t const remainingToFlush = ZSTD_compressStream2(zcs, output, &input, ZSTD_e_end);
+    FORWARD_IF_ERROR(remainingToFlush , "ZSTD_compressStream2(,,ZSTD_e_end) failed");
+    if (zcs->appliedParams.nbWorkers > 0) return remainingToFlush;   /* minimal estimation */
+    /* single thread mode : attempt to calculate remaining to flush more precisely */
+    {   size_t const lastBlockSize = zcs->frameEnded ? 0 : ZSTD_BLOCKHEADERSIZE;
+        size_t const checksumSize = (size_t)(zcs->frameEnded ? 0 : zcs->appliedParams.fParams.checksumFlag * 4);
+        size_t const toFlush = remainingToFlush + lastBlockSize + checksumSize;
+        DEBUGLOG(4, "ZSTD_endStream : remaining to flush : %u", (unsigned)toFlush);
+        return toFlush;
+    }
+}
+
+
+/*-=====  Pre-defined compression levels  =====-*/
+#include "clevels.h"
+
+int ZSTD_maxCLevel(void) { return ZSTD_MAX_CLEVEL; }
+int ZSTD_minCLevel(void) { return (int)-ZSTD_TARGETLENGTH_MAX; }
+int ZSTD_defaultCLevel(void) { return ZSTD_CLEVEL_DEFAULT; }
+
+static ZSTD_compressionParameters ZSTD_dedicatedDictSearch_getCParams(int const compressionLevel, size_t const dictSize)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, 0, dictSize, ZSTD_cpm_createCDict);
+    switch (cParams.strategy) {
+        case ZSTD_fast:
+        case ZSTD_dfast:
+            break;
+        case ZSTD_greedy:
+        case ZSTD_lazy:
+        case ZSTD_lazy2:
+            cParams.hashLog += ZSTD_LAZY_DDSS_BUCKET_LOG;
+            break;
+        case ZSTD_btlazy2:
+        case ZSTD_btopt:
+        case ZSTD_btultra:
+        case ZSTD_btultra2:
+            break;
+    }
+    return cParams;
+}
+
+static int ZSTD_dedicatedDictSearch_isSupported(
+        ZSTD_compressionParameters const* cParams)
+{
+    return (cParams->strategy >= ZSTD_greedy)
+        && (cParams->strategy <= ZSTD_lazy2)
+        && (cParams->hashLog > cParams->chainLog)
+        && (cParams->chainLog <= 24);
+}
+
+/**
+ * Reverses the adjustment applied to cparams when enabling dedicated dict
+ * search. This is used to recover the params set to be used in the working
+ * context. (Otherwise, those tables would also grow.)
+ */
+static void ZSTD_dedicatedDictSearch_revertCParams(
+        ZSTD_compressionParameters* cParams) {
+    switch (cParams->strategy) {
+        case ZSTD_fast:
+        case ZSTD_dfast:
+            break;
+        case ZSTD_greedy:
+        case ZSTD_lazy:
+        case ZSTD_lazy2:
+            cParams->hashLog -= ZSTD_LAZY_DDSS_BUCKET_LOG;
+            if (cParams->hashLog < ZSTD_HASHLOG_MIN) {
+                cParams->hashLog = ZSTD_HASHLOG_MIN;
+            }
+            break;
+        case ZSTD_btlazy2:
+        case ZSTD_btopt:
+        case ZSTD_btultra:
+        case ZSTD_btultra2:
+            break;
+    }
+}
+
+static U64 ZSTD_getCParamRowSize(U64 srcSizeHint, size_t dictSize, ZSTD_CParamMode_e mode)
+{
+    switch (mode) {
+    case ZSTD_cpm_unknown:
+    case ZSTD_cpm_noAttachDict:
+    case ZSTD_cpm_createCDict:
+        break;
+    case ZSTD_cpm_attachDict:
+        dictSize = 0;
+        break;
+    default:
+        assert(0);
+        break;
+    }
+    {   int const unknown = srcSizeHint == ZSTD_CONTENTSIZE_UNKNOWN;
+        size_t const addedSize = unknown && dictSize > 0 ? 500 : 0;
+        return unknown && dictSize == 0 ? ZSTD_CONTENTSIZE_UNKNOWN : srcSizeHint+dictSize+addedSize;
+    }
+}
+
+/*! ZSTD_getCParams_internal() :
+ * @return ZSTD_compressionParameters structure for a selected compression level, srcSize and dictSize.
+ *  Note: srcSizeHint 0 means 0, use ZSTD_CONTENTSIZE_UNKNOWN for unknown.
+ *        Use dictSize == 0 for unknown or unused.
+ *  Note: `mode` controls how we treat the `dictSize`. See docs for `ZSTD_CParamMode_e`. */
+static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_CParamMode_e mode)
+{
+    U64 const rSize = ZSTD_getCParamRowSize(srcSizeHint, dictSize, mode);
+    U32 const tableID = (rSize <= 256 KB) + (rSize <= 128 KB) + (rSize <= 16 KB);
+    int row;
+    DEBUGLOG(5, "ZSTD_getCParams_internal (cLevel=%i)", compressionLevel);
+
+    /* row */
+    if (compressionLevel == 0) row = ZSTD_CLEVEL_DEFAULT;   /* 0 == default */
+    else if (compressionLevel < 0) row = 0;   /* entry 0 is baseline for fast mode */
+    else if (compressionLevel > ZSTD_MAX_CLEVEL) row = ZSTD_MAX_CLEVEL;
+    else row = compressionLevel;
+
+    {   ZSTD_compressionParameters cp = ZSTD_defaultCParameters[tableID][row];
+        DEBUGLOG(5, "ZSTD_getCParams_internal selected tableID: %u row: %u strat: %u", tableID, row, (U32)cp.strategy);
+        /* acceleration factor */
+        if (compressionLevel < 0) {
+            int const clampedCompressionLevel = MAX(ZSTD_minCLevel(), compressionLevel);
+            cp.targetLength = (unsigned)(-clampedCompressionLevel);
+        }
+        /* refine parameters based on srcSize & dictSize */
+        return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize, mode, ZSTD_ps_auto);
+    }
+}
+
+/*! ZSTD_getCParams() :
+ * @return ZSTD_compressionParameters structure for a selected compression level, srcSize and dictSize.
+ *  Size values are optional, provide 0 if not known or unused */
+ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize)
+{
+    if (srcSizeHint == 0) srcSizeHint = ZSTD_CONTENTSIZE_UNKNOWN;
+    return ZSTD_getCParams_internal(compressionLevel, srcSizeHint, dictSize, ZSTD_cpm_unknown);
+}
+
+/*! ZSTD_getParams() :
+ *  same idea as ZSTD_getCParams()
+ * @return a `ZSTD_parameters` structure (instead of `ZSTD_compressionParameters`).
+ *  Fields of `ZSTD_frameParameters` are set to default values */
+static ZSTD_parameters
+ZSTD_getParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_CParamMode_e mode)
+{
+    ZSTD_parameters params;
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, srcSizeHint, dictSize, mode);
+    DEBUGLOG(5, "ZSTD_getParams (cLevel=%i)", compressionLevel);
+    ZSTD_memset(&params, 0, sizeof(params));
+    params.cParams = cParams;
+    params.fParams.contentSizeFlag = 1;
+    return params;
+}
+
+/*! ZSTD_getParams() :
+ *  same idea as ZSTD_getCParams()
+ * @return a `ZSTD_parameters` structure (instead of `ZSTD_compressionParameters`).
+ *  Fields of `ZSTD_frameParameters` are set to default values */
+ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize)
+{
+    if (srcSizeHint == 0) srcSizeHint = ZSTD_CONTENTSIZE_UNKNOWN;
+    return ZSTD_getParams_internal(compressionLevel, srcSizeHint, dictSize, ZSTD_cpm_unknown);
+}
+
+void ZSTD_registerSequenceProducer(
+    ZSTD_CCtx* zc,
+    void* extSeqProdState,
+    ZSTD_sequenceProducer_F extSeqProdFunc)
+{
+    assert(zc != NULL);
+    ZSTD_CCtxParams_registerSequenceProducer(
+        &zc->requestedParams, extSeqProdState, extSeqProdFunc
+    );
+}
+
+void ZSTD_CCtxParams_registerSequenceProducer(
+  ZSTD_CCtx_params* params,
+  void* extSeqProdState,
+  ZSTD_sequenceProducer_F extSeqProdFunc)
+{
+    assert(params != NULL);
+    if (extSeqProdFunc != NULL) {
+        params->extSeqProdFunc = extSeqProdFunc;
+        params->extSeqProdState = extSeqProdState;
+    } else {
+        params->extSeqProdFunc = NULL;
+        params->extSeqProdState = NULL;
+    }
+}
diff --git a/lib/compress/zstd_compress_sequences.c b/lib/compress/zstd_compress_sequences.c
new file mode 100644
index 0000000..7beb9da
--- /dev/null
+++ b/lib/compress/zstd_compress_sequences.c
@@ -0,0 +1,442 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+ /*-*************************************
+ *  Dependencies
+ ***************************************/
+#include "zstd_compress_sequences.h"
+
+/**
+ * -log2(x / 256) lookup table for x in [0, 256).
+ * If x == 0: Return 0
+ * Else: Return floor(-log2(x / 256) * 256)
+ */
+static unsigned const kInverseProbabilityLog256[256] = {
+    0,    2048, 1792, 1642, 1536, 1453, 1386, 1329, 1280, 1236, 1197, 1162,
+    1130, 1100, 1073, 1047, 1024, 1001, 980,  960,  941,  923,  906,  889,
+    874,  859,  844,  830,  817,  804,  791,  779,  768,  756,  745,  734,
+    724,  714,  704,  694,  685,  676,  667,  658,  650,  642,  633,  626,
+    618,  610,  603,  595,  588,  581,  574,  567,  561,  554,  548,  542,
+    535,  529,  523,  517,  512,  506,  500,  495,  489,  484,  478,  473,
+    468,  463,  458,  453,  448,  443,  438,  434,  429,  424,  420,  415,
+    411,  407,  402,  398,  394,  390,  386,  382,  377,  373,  370,  366,
+    362,  358,  354,  350,  347,  343,  339,  336,  332,  329,  325,  322,
+    318,  315,  311,  308,  305,  302,  298,  295,  292,  289,  286,  282,
+    279,  276,  273,  270,  267,  264,  261,  258,  256,  253,  250,  247,
+    244,  241,  239,  236,  233,  230,  228,  225,  222,  220,  217,  215,
+    212,  209,  207,  204,  202,  199,  197,  194,  192,  190,  187,  185,
+    182,  180,  178,  175,  173,  171,  168,  166,  164,  162,  159,  157,
+    155,  153,  151,  149,  146,  144,  142,  140,  138,  136,  134,  132,
+    130,  128,  126,  123,  121,  119,  117,  115,  114,  112,  110,  108,
+    106,  104,  102,  100,  98,   96,   94,   93,   91,   89,   87,   85,
+    83,   82,   80,   78,   76,   74,   73,   71,   69,   67,   66,   64,
+    62,   61,   59,   57,   55,   54,   52,   50,   49,   47,   46,   44,
+    42,   41,   39,   37,   36,   34,   33,   31,   30,   28,   26,   25,
+    23,   22,   20,   19,   17,   16,   14,   13,   11,   10,   8,    7,
+    5,    4,    2,    1,
+};
+
+static unsigned ZSTD_getFSEMaxSymbolValue(FSE_CTable const* ctable) {
+  void const* ptr = ctable;
+  U16 const* u16ptr = (U16 const*)ptr;
+  U32 const maxSymbolValue = MEM_read16(u16ptr + 1);
+  return maxSymbolValue;
+}
+
+/**
+ * Returns true if we should use ncount=-1 else we should
+ * use ncount=1 for low probability symbols instead.
+ */
+static unsigned ZSTD_useLowProbCount(size_t const nbSeq)
+{
+    /* Heuristic: This should cover most blocks <= 16K and
+     * start to fade out after 16K to about 32K depending on
+     * compressibility.
+     */
+    return nbSeq >= 2048;
+}
+
+/**
+ * Returns the cost in bytes of encoding the normalized count header.
+ * Returns an error if any of the helper functions return an error.
+ */
+static size_t ZSTD_NCountCost(unsigned const* count, unsigned const max,
+                              size_t const nbSeq, unsigned const FSELog)
+{
+    BYTE wksp[FSE_NCOUNTBOUND];
+    S16 norm[MaxSeq + 1];
+    const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
+    FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max, ZSTD_useLowProbCount(nbSeq)), "");
+    return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
+}
+
+/**
+ * Returns the cost in bits of encoding the distribution described by count
+ * using the entropy bound.
+ */
+static size_t ZSTD_entropyCost(unsigned const* count, unsigned const max, size_t const total)
+{
+    unsigned cost = 0;
+    unsigned s;
+
+    assert(total > 0);
+    for (s = 0; s <= max; ++s) {
+        unsigned norm = (unsigned)((256 * count[s]) / total);
+        if (count[s] != 0 && norm == 0)
+            norm = 1;
+        assert(count[s] < total);
+        cost += count[s] * kInverseProbabilityLog256[norm];
+    }
+    return cost >> 8;
+}
+
+/**
+ * Returns the cost in bits of encoding the distribution in count using ctable.
+ * Returns an error if ctable cannot represent all the symbols in count.
+ */
+size_t ZSTD_fseBitCost(
+    FSE_CTable const* ctable,
+    unsigned const* count,
+    unsigned const max)
+{
+    unsigned const kAccuracyLog = 8;
+    size_t cost = 0;
+    unsigned s;
+    FSE_CState_t cstate;
+    FSE_initCState(&cstate, ctable);
+    if (ZSTD_getFSEMaxSymbolValue(ctable) < max) {
+        DEBUGLOG(5, "Repeat FSE_CTable has maxSymbolValue %u < %u",
+                    ZSTD_getFSEMaxSymbolValue(ctable), max);
+        return ERROR(GENERIC);
+    }
+    for (s = 0; s <= max; ++s) {
+        unsigned const tableLog = cstate.stateLog;
+        unsigned const badCost = (tableLog + 1) << kAccuracyLog;
+        unsigned const bitCost = FSE_bitCost(cstate.symbolTT, tableLog, s, kAccuracyLog);
+        if (count[s] == 0)
+            continue;
+        if (bitCost >= badCost) {
+            DEBUGLOG(5, "Repeat FSE_CTable has Prob[%u] == 0", s);
+            return ERROR(GENERIC);
+        }
+        cost += (size_t)count[s] * bitCost;
+    }
+    return cost >> kAccuracyLog;
+}
+
+/**
+ * Returns the cost in bits of encoding the distribution in count using the
+ * table described by norm. The max symbol support by norm is assumed >= max.
+ * norm must be valid for every symbol with non-zero probability in count.
+ */
+size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
+                             unsigned const* count, unsigned const max)
+{
+    unsigned const shift = 8 - accuracyLog;
+    size_t cost = 0;
+    unsigned s;
+    assert(accuracyLog <= 8);
+    for (s = 0; s <= max; ++s) {
+        unsigned const normAcc = (norm[s] != -1) ? (unsigned)norm[s] : 1;
+        unsigned const norm256 = normAcc << shift;
+        assert(norm256 > 0);
+        assert(norm256 < 256);
+        cost += count[s] * kInverseProbabilityLog256[norm256];
+    }
+    return cost >> 8;
+}
+
+SymbolEncodingType_e
+ZSTD_selectEncodingType(
+        FSE_repeat* repeatMode, unsigned const* count, unsigned const max,
+        size_t const mostFrequent, size_t nbSeq, unsigned const FSELog,
+        FSE_CTable const* prevCTable,
+        short const* defaultNorm, U32 defaultNormLog,
+        ZSTD_DefaultPolicy_e const isDefaultAllowed,
+        ZSTD_strategy const strategy)
+{
+    ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0);
+    if (mostFrequent == nbSeq) {
+        *repeatMode = FSE_repeat_none;
+        if (isDefaultAllowed && nbSeq <= 2) {
+            /* Prefer set_basic over set_rle when there are 2 or fewer symbols,
+             * since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
+             * If basic encoding isn't possible, always choose RLE.
+             */
+            DEBUGLOG(5, "Selected set_basic");
+            return set_basic;
+        }
+        DEBUGLOG(5, "Selected set_rle");
+        return set_rle;
+    }
+    if (strategy < ZSTD_lazy) {
+        if (isDefaultAllowed) {
+            size_t const staticFse_nbSeq_max = 1000;
+            size_t const mult = 10 - strategy;
+            size_t const baseLog = 3;
+            size_t const dynamicFse_nbSeq_min = (((size_t)1 << defaultNormLog) * mult) >> baseLog;  /* 28-36 for offset, 56-72 for lengths */
+            assert(defaultNormLog >= 5 && defaultNormLog <= 6);  /* xx_DEFAULTNORMLOG */
+            assert(mult <= 9 && mult >= 7);
+            if ( (*repeatMode == FSE_repeat_valid)
+              && (nbSeq < staticFse_nbSeq_max) ) {
+                DEBUGLOG(5, "Selected set_repeat");
+                return set_repeat;
+            }
+            if ( (nbSeq < dynamicFse_nbSeq_min)
+              || (mostFrequent < (nbSeq >> (defaultNormLog-1))) ) {
+                DEBUGLOG(5, "Selected set_basic");
+                /* The format allows default tables to be repeated, but it isn't useful.
+                 * When using simple heuristics to select encoding type, we don't want
+                 * to confuse these tables with dictionaries. When running more careful
+                 * analysis, we don't need to waste time checking both repeating tables
+                 * and default tables.
+                 */
+                *repeatMode = FSE_repeat_none;
+                return set_basic;
+            }
+        }
+    } else {
+        size_t const basicCost = isDefaultAllowed ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, count, max) : ERROR(GENERIC);
+        size_t const repeatCost = *repeatMode != FSE_repeat_none ? ZSTD_fseBitCost(prevCTable, count, max) : ERROR(GENERIC);
+        size_t const NCountCost = ZSTD_NCountCost(count, max, nbSeq, FSELog);
+        size_t const compressedCost = (NCountCost << 3) + ZSTD_entropyCost(count, max, nbSeq);
+
+        if (isDefaultAllowed) {
+            assert(!ZSTD_isError(basicCost));
+            assert(!(*repeatMode == FSE_repeat_valid && ZSTD_isError(repeatCost)));
+        }
+        assert(!ZSTD_isError(NCountCost));
+        assert(compressedCost < ERROR(maxCode));
+        DEBUGLOG(5, "Estimated bit costs: basic=%u\trepeat=%u\tcompressed=%u",
+                    (unsigned)basicCost, (unsigned)repeatCost, (unsigned)compressedCost);
+        if (basicCost <= repeatCost && basicCost <= compressedCost) {
+            DEBUGLOG(5, "Selected set_basic");
+            assert(isDefaultAllowed);
+            *repeatMode = FSE_repeat_none;
+            return set_basic;
+        }
+        if (repeatCost <= compressedCost) {
+            DEBUGLOG(5, "Selected set_repeat");
+            assert(!ZSTD_isError(repeatCost));
+            return set_repeat;
+        }
+        assert(compressedCost < basicCost && compressedCost < repeatCost);
+    }
+    DEBUGLOG(5, "Selected set_compressed");
+    *repeatMode = FSE_repeat_check;
+    return set_compressed;
+}
+
+typedef struct {
+    S16 norm[MaxSeq + 1];
+    U32 wksp[FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(MaxSeq, MaxFSELog)];
+} ZSTD_BuildCTableWksp;
+
+size_t
+ZSTD_buildCTable(void* dst, size_t dstCapacity,
+                FSE_CTable* nextCTable, U32 FSELog, SymbolEncodingType_e type,
+                unsigned* count, U32 max,
+                const BYTE* codeTable, size_t nbSeq,
+                const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+                const FSE_CTable* prevCTable, size_t prevCTableSize,
+                void* entropyWorkspace, size_t entropyWorkspaceSize)
+{
+    BYTE* op = (BYTE*)dst;
+    const BYTE* const oend = op + dstCapacity;
+    DEBUGLOG(6, "ZSTD_buildCTable (dstCapacity=%u)", (unsigned)dstCapacity);
+
+    switch (type) {
+    case set_rle:
+        FORWARD_IF_ERROR(FSE_buildCTable_rle(nextCTable, (BYTE)max), "");
+        RETURN_ERROR_IF(dstCapacity==0, dstSize_tooSmall, "not enough space");
+        *op = codeTable[0];
+        return 1;
+    case set_repeat:
+        ZSTD_memcpy(nextCTable, prevCTable, prevCTableSize);
+        return 0;
+    case set_basic:
+        FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, entropyWorkspace, entropyWorkspaceSize), "");  /* note : could be pre-calculated */
+        return 0;
+    case set_compressed: {
+        ZSTD_BuildCTableWksp* wksp = (ZSTD_BuildCTableWksp*)entropyWorkspace;
+        size_t nbSeq_1 = nbSeq;
+        const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
+        if (count[codeTable[nbSeq-1]] > 1) {
+            count[codeTable[nbSeq-1]]--;
+            nbSeq_1--;
+        }
+        assert(nbSeq_1 > 1);
+        assert(entropyWorkspaceSize >= sizeof(ZSTD_BuildCTableWksp));
+        (void)entropyWorkspaceSize;
+        FORWARD_IF_ERROR(FSE_normalizeCount(wksp->norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "FSE_normalizeCount failed");
+        assert(oend >= op);
+        {   size_t const NCountSize = FSE_writeNCount(op, (size_t)(oend - op), wksp->norm, max, tableLog);   /* overflow protected */
+            FORWARD_IF_ERROR(NCountSize, "FSE_writeNCount failed");
+            FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, wksp->norm, max, tableLog, wksp->wksp, sizeof(wksp->wksp)), "FSE_buildCTable_wksp failed");
+            return NCountSize;
+        }
+    }
+    default: assert(0); RETURN_ERROR(GENERIC, "impossible to reach");
+    }
+}
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_encodeSequences_body(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            SeqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    BIT_CStream_t blockStream;
+    FSE_CState_t  stateMatchLength;
+    FSE_CState_t  stateOffsetBits;
+    FSE_CState_t  stateLitLength;
+
+    RETURN_ERROR_IF(
+        ERR_isError(BIT_initCStream(&blockStream, dst, dstCapacity)),
+        dstSize_tooSmall, "not enough space remaining");
+    DEBUGLOG(6, "available space for bitstream : %i  (dstCapacity=%u)",
+                (int)(blockStream.endPtr - blockStream.startPtr),
+                (unsigned)dstCapacity);
+
+    /* first symbols */
+    FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
+    FSE_initCState2(&stateOffsetBits,  CTable_OffsetBits,  ofCodeTable[nbSeq-1]);
+    FSE_initCState2(&stateLitLength,   CTable_LitLength,   llCodeTable[nbSeq-1]);
+    BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
+    if (MEM_32bits()) BIT_flushBits(&blockStream);
+    BIT_addBits(&blockStream, sequences[nbSeq-1].mlBase, ML_bits[mlCodeTable[nbSeq-1]]);
+    if (MEM_32bits()) BIT_flushBits(&blockStream);
+    if (longOffsets) {
+        U32 const ofBits = ofCodeTable[nbSeq-1];
+        unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+        if (extraBits) {
+            BIT_addBits(&blockStream, sequences[nbSeq-1].offBase, extraBits);
+            BIT_flushBits(&blockStream);
+        }
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offBase >> extraBits,
+                    ofBits - extraBits);
+    } else {
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offBase, ofCodeTable[nbSeq-1]);
+    }
+    BIT_flushBits(&blockStream);
+
+    {   size_t n;
+        for (n=nbSeq-2 ; n<nbSeq ; n--) {      /* intentional underflow */
+            BYTE const llCode = llCodeTable[n];
+            BYTE const ofCode = ofCodeTable[n];
+            BYTE const mlCode = mlCodeTable[n];
+            U32  const llBits = LL_bits[llCode];
+            U32  const ofBits = ofCode;
+            U32  const mlBits = ML_bits[mlCode];
+            DEBUGLOG(6, "encoding: litlen:%2u - matchlen:%2u - offCode:%7u",
+                        (unsigned)sequences[n].litLength,
+                        (unsigned)sequences[n].mlBase + MINMATCH,
+                        (unsigned)sequences[n].offBase);
+                                                                            /* 32b*/  /* 64b*/
+                                                                            /* (7)*/  /* (7)*/
+            FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode);       /* 15 */  /* 15 */
+            FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode);      /* 24 */  /* 24 */
+            if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
+            FSE_encodeSymbol(&blockStream, &stateLitLength, llCode);        /* 16 */  /* 33 */
+            if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
+                BIT_flushBits(&blockStream);                                /* (7)*/
+            BIT_addBits(&blockStream, sequences[n].litLength, llBits);
+            if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
+            BIT_addBits(&blockStream, sequences[n].mlBase, mlBits);
+            if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
+            if (longOffsets) {
+                unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+                if (extraBits) {
+                    BIT_addBits(&blockStream, sequences[n].offBase, extraBits);
+                    BIT_flushBits(&blockStream);                            /* (7)*/
+                }
+                BIT_addBits(&blockStream, sequences[n].offBase >> extraBits,
+                            ofBits - extraBits);                            /* 31 */
+            } else {
+                BIT_addBits(&blockStream, sequences[n].offBase, ofBits);     /* 31 */
+            }
+            BIT_flushBits(&blockStream);                                    /* (7)*/
+            DEBUGLOG(7, "remaining space : %i", (int)(blockStream.endPtr - blockStream.ptr));
+    }   }
+
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing ML state with %u bits", stateMatchLength.stateLog);
+    FSE_flushCState(&blockStream, &stateMatchLength);
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing Off state with %u bits", stateOffsetBits.stateLog);
+    FSE_flushCState(&blockStream, &stateOffsetBits);
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing LL state with %u bits", stateLitLength.stateLog);
+    FSE_flushCState(&blockStream, &stateLitLength);
+
+    {   size_t const streamSize = BIT_closeCStream(&blockStream);
+        RETURN_ERROR_IF(streamSize==0, dstSize_tooSmall, "not enough space");
+        return streamSize;
+    }
+}
+
+static size_t
+ZSTD_encodeSequences_default(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            SeqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    return ZSTD_encodeSequences_body(dst, dstCapacity,
+                                    CTable_MatchLength, mlCodeTable,
+                                    CTable_OffsetBits, ofCodeTable,
+                                    CTable_LitLength, llCodeTable,
+                                    sequences, nbSeq, longOffsets);
+}
+
+
+#if DYNAMIC_BMI2
+
+static BMI2_TARGET_ATTRIBUTE size_t
+ZSTD_encodeSequences_bmi2(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            SeqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    return ZSTD_encodeSequences_body(dst, dstCapacity,
+                                    CTable_MatchLength, mlCodeTable,
+                                    CTable_OffsetBits, ofCodeTable,
+                                    CTable_LitLength, llCodeTable,
+                                    sequences, nbSeq, longOffsets);
+}
+
+#endif
+
+size_t ZSTD_encodeSequences(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            SeqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2)
+{
+    DEBUGLOG(5, "ZSTD_encodeSequences: dstCapacity = %u", (unsigned)dstCapacity);
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return ZSTD_encodeSequences_bmi2(dst, dstCapacity,
+                                         CTable_MatchLength, mlCodeTable,
+                                         CTable_OffsetBits, ofCodeTable,
+                                         CTable_LitLength, llCodeTable,
+                                         sequences, nbSeq, longOffsets);
+    }
+#endif
+    (void)bmi2;
+    return ZSTD_encodeSequences_default(dst, dstCapacity,
+                                        CTable_MatchLength, mlCodeTable,
+                                        CTable_OffsetBits, ofCodeTable,
+                                        CTable_LitLength, llCodeTable,
+                                        sequences, nbSeq, longOffsets);
+}
diff --git a/lib/compress/zstd_double_fast.c b/lib/compress/zstd_double_fast.c
new file mode 100644
index 0000000..1a266e7
--- /dev/null
+++ b/lib/compress/zstd_double_fast.c
@@ -0,0 +1,778 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_double_fast.h"
+
+#ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
+
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_fillDoubleHashTableForCDict(ZSTD_MatchState_t* ms,
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashLarge = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    U32  const mls = cParams->minMatch;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
+
+    /* Always insert every fastHashFillStep position into the hash tables.
+     * Insert the other positions into the large hash table if their entry
+     * is empty.
+     */
+    for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
+        U32 const curr = (U32)(ip - base);
+        U32 i;
+        for (i = 0; i < fastHashFillStep; ++i) {
+            size_t const smHashAndTag = ZSTD_hashPtr(ip + i, hBitsS, mls);
+            size_t const lgHashAndTag = ZSTD_hashPtr(ip + i, hBitsL, 8);
+            if (i == 0) {
+                ZSTD_writeTaggedIndex(hashSmall, smHashAndTag, curr + i);
+            }
+            if (i == 0 || hashLarge[lgHashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {
+                ZSTD_writeTaggedIndex(hashLarge, lgHashAndTag, curr + i);
+            }
+            /* Only load extra positions for ZSTD_dtlm_full */
+            if (dtlm == ZSTD_dtlm_fast)
+                break;
+    }   }
+}
+
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_fillDoubleHashTableForCCtx(ZSTD_MatchState_t* ms,
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashLarge = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog;
+    U32  const mls = cParams->minMatch;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
+
+    /* Always insert every fastHashFillStep position into the hash tables.
+     * Insert the other positions into the large hash table if their entry
+     * is empty.
+     */
+    for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
+        U32 const curr = (U32)(ip - base);
+        U32 i;
+        for (i = 0; i < fastHashFillStep; ++i) {
+            size_t const smHash = ZSTD_hashPtr(ip + i, hBitsS, mls);
+            size_t const lgHash = ZSTD_hashPtr(ip + i, hBitsL, 8);
+            if (i == 0)
+                hashSmall[smHash] = curr + i;
+            if (i == 0 || hashLarge[lgHash] == 0)
+                hashLarge[lgHash] = curr + i;
+            /* Only load extra positions for ZSTD_dtlm_full */
+            if (dtlm == ZSTD_dtlm_fast)
+                break;
+        }   }
+}
+
+void ZSTD_fillDoubleHashTable(ZSTD_MatchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm,
+                        ZSTD_tableFillPurpose_e tfp)
+{
+    if (tfp == ZSTD_tfp_forCDict) {
+        ZSTD_fillDoubleHashTableForCDict(ms, end, dtlm);
+    } else {
+        ZSTD_fillDoubleHashTableForCCtx(ms, end, dtlm);
+    }
+}
+
+
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_compressBlock_doubleFast_noDict_generic(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize, U32 const mls /* template */)
+{
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
+    U32* const hashLong = ms->hashTable;
+    const U32 hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    const U32 hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* anchor = istart;
+    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
+    /* presumes that, if there is a dictionary, it must be using Attach mode */
+    const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
+    const BYTE* const prefixLowest = base + prefixLowestIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=rep[0], offset_2=rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
+
+    size_t mLength;
+    U32 offset;
+    U32 curr;
+
+    /* how many positions to search before increasing step size */
+    const size_t kStepIncr = 1 << kSearchStrength;
+    /* the position at which to increment the step size if no match is found */
+    const BYTE* nextStep;
+    size_t step; /* the current step size */
+
+    size_t hl0; /* the long hash at ip */
+    size_t hl1; /* the long hash at ip1 */
+
+    U32 idxl0; /* the long match index for ip */
+    U32 idxl1; /* the long match index for ip1 */
+
+    const BYTE* matchl0; /* the long match for ip */
+    const BYTE* matchs0; /* the short match for ip */
+    const BYTE* matchl1; /* the long match for ip1 */
+    const BYTE* matchs0_safe; /* matchs0 or safe address */
+
+    const BYTE* ip = istart; /* the current position */
+    const BYTE* ip1; /* the next position */
+    /* Array of ~random data, should have low probability of matching data
+     * we load from here instead of from tables, if matchl0/matchl1 are
+     * invalid indices. Used to avoid unpredictable branches. */
+    const BYTE dummy[] = {0x12,0x34,0x56,0x78,0x9a,0xbc,0xde,0xf0,0xe2,0xb4};
+
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_noDict_generic");
+
+    /* init */
+    ip += ((ip - prefixLowest) == 0);
+    {
+        U32 const current = (U32)(ip - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
+        U32 const maxRep = current - windowLow;
+        if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
+    }
+
+    /* Outer Loop: one iteration per match found and stored */
+    while (1) {
+        step = 1;
+        nextStep = ip + kStepIncr;
+        ip1 = ip + step;
+
+        if (ip1 > ilimit) {
+            goto _cleanup;
+        }
+
+        hl0 = ZSTD_hashPtr(ip, hBitsL, 8);
+        idxl0 = hashLong[hl0];
+        matchl0 = base + idxl0;
+
+        /* Inner Loop: one iteration per search / position */
+        do {
+            const size_t hs0 = ZSTD_hashPtr(ip, hBitsS, mls);
+            const U32 idxs0 = hashSmall[hs0];
+            curr = (U32)(ip-base);
+            matchs0 = base + idxs0;
+
+            hashLong[hl0] = hashSmall[hs0] = curr;   /* update hash tables */
+
+            /* check noDict repcode */
+            if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
+                mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+                ip++;
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
+                goto _match_stored;
+            }
+
+            hl1 = ZSTD_hashPtr(ip1, hBitsL, 8);
+
+            /* idxl0 > prefixLowestIndex is a (somewhat) unpredictable branch.
+             * However expression below complies into conditional move. Since
+             * match is unlikely and we only *branch* on idxl0 > prefixLowestIndex
+             * if there is a match, all branches become predictable. */
+            {   const BYTE*  const matchl0_safe = ZSTD_selectAddr(idxl0, prefixLowestIndex, matchl0, &dummy[0]);
+
+                /* check prefix long match */
+                if (MEM_read64(matchl0_safe) == MEM_read64(ip) && matchl0_safe == matchl0) {
+                    mLength = ZSTD_count(ip+8, matchl0+8, iend) + 8;
+                    offset = (U32)(ip-matchl0);
+                    while (((ip>anchor) & (matchl0>prefixLowest)) && (ip[-1] == matchl0[-1])) { ip--; matchl0--; mLength++; } /* catch up */
+                    goto _match_found;
+            }   }
+
+            idxl1 = hashLong[hl1];
+            matchl1 = base + idxl1;
+
+            /* Same optimization as matchl0 above */
+            matchs0_safe = ZSTD_selectAddr(idxs0, prefixLowestIndex, matchs0, &dummy[0]);
+
+            /* check prefix short match */
+            if(MEM_read32(matchs0_safe) == MEM_read32(ip) && matchs0_safe == matchs0) {
+                  goto _search_next_long;
+            }
+
+            if (ip1 >= nextStep) {
+                PREFETCH_L1(ip1 + 64);
+                PREFETCH_L1(ip1 + 128);
+                step++;
+                nextStep += kStepIncr;
+            }
+            ip = ip1;
+            ip1 += step;
+
+            hl0 = hl1;
+            idxl0 = idxl1;
+            matchl0 = matchl1;
+    #if defined(__aarch64__)
+            PREFETCH_L1(ip+256);
+    #endif
+        } while (ip1 <= ilimit);
+
+_cleanup:
+        /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
+         * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
+        offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
+
+        /* save reps for next block */
+        rep[0] = offset_1 ? offset_1 : offsetSaved1;
+        rep[1] = offset_2 ? offset_2 : offsetSaved2;
+
+        /* Return the last literals size */
+        return (size_t)(iend - anchor);
+
+_search_next_long:
+
+        /* short match found: let's check for a longer one */
+        mLength = ZSTD_count(ip+4, matchs0+4, iend) + 4;
+        offset = (U32)(ip - matchs0);
+
+        /* check long match at +1 position */
+        if ((idxl1 > prefixLowestIndex) && (MEM_read64(matchl1) == MEM_read64(ip1))) {
+            size_t const l1len = ZSTD_count(ip1+8, matchl1+8, iend) + 8;
+            if (l1len > mLength) {
+                /* use the long match instead */
+                ip = ip1;
+                mLength = l1len;
+                offset = (U32)(ip-matchl1);
+                matchs0 = matchl1;
+            }
+        }
+
+        while (((ip>anchor) & (matchs0>prefixLowest)) && (ip[-1] == matchs0[-1])) { ip--; matchs0--; mLength++; } /* complete backward */
+
+        /* fall-through */
+
+_match_found: /* requires ip, offset, mLength */
+        offset_2 = offset_1;
+        offset_1 = offset;
+
+        if (step < 4) {
+            /* It is unsafe to write this value back to the hashtable when ip1 is
+             * greater than or equal to the new ip we will have after we're done
+             * processing this match. Rather than perform that test directly
+             * (ip1 >= ip + mLength), which costs speed in practice, we do a simpler
+             * more predictable test. The minmatch even if we take a short match is
+             * 4 bytes, so as long as step, the distance between ip and ip1
+             * (initially) is less than 4, we know ip1 < new ip. */
+            hashLong[hl1] = (U32)(ip1 - base);
+        }
+
+        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+
+_match_stored:
+        /* match found */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Complementary insertion */
+            /* done after iLimit test, as candidates could be > iend-8 */
+            {   U32 const indexToInsert = curr+2;
+                hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
+                hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+                hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
+                hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
+            }
+
+            /* check immediate repcode */
+            while ( (ip <= ilimit)
+                 && ( (offset_2>0)
+                    & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                /* store sequence */
+                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff;  /* swap offset_2 <=> offset_1 */
+                hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
+                hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
+                ip += rLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+            }
+        }
+    }
+}
+
+
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const mls /* template */)
+{
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
+    U32* const hashLong = ms->hashTable;
+    const U32 hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    const U32 hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
+    /* presumes that, if there is a dictionary, it must be using Attach mode */
+    const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
+    const BYTE* const prefixLowest = base + prefixLowestIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=rep[0], offset_2=rep[1];
+
+    const ZSTD_MatchState_t* const dms = ms->dictMatchState;
+    const ZSTD_compressionParameters* const dictCParams = &dms->cParams;
+    const U32* const dictHashLong  = dms->hashTable;
+    const U32* const dictHashSmall = dms->chainTable;
+    const U32 dictStartIndex       = dms->window.dictLimit;
+    const BYTE* const dictBase     = dms->window.base;
+    const BYTE* const dictStart    = dictBase + dictStartIndex;
+    const BYTE* const dictEnd      = dms->window.nextSrc;
+    const U32 dictIndexDelta       = prefixLowestIndex - (U32)(dictEnd - dictBase);
+    const U32 dictHBitsL           = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    const U32 dictHBitsS           = dictCParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    const U32 dictAndPrefixLength  = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
+
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_dictMatchState_generic");
+
+    /* if a dictionary is attached, it must be within window range */
+    assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
+
+    if (ms->prefetchCDictTables) {
+        size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
+        size_t const chainTableBytes = (((size_t)1) << dictCParams->chainLog) * sizeof(U32);
+        PREFETCH_AREA(dictHashLong, hashTableBytes);
+        PREFETCH_AREA(dictHashSmall, chainTableBytes);
+    }
+
+    /* init */
+    ip += (dictAndPrefixLength == 0);
+
+    /* dictMatchState repCode checks don't currently handle repCode == 0
+     * disabling. */
+    assert(offset_1 <= dictAndPrefixLength);
+    assert(offset_2 <= dictAndPrefixLength);
+
+    /* Main Search Loop */
+    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
+        size_t mLength;
+        U32 offset;
+        size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
+        size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
+        size_t const dictHashAndTagL = ZSTD_hashPtr(ip, dictHBitsL, 8);
+        size_t const dictHashAndTagS = ZSTD_hashPtr(ip, dictHBitsS, mls);
+        U32 const dictMatchIndexAndTagL = dictHashLong[dictHashAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS];
+        U32 const dictMatchIndexAndTagS = dictHashSmall[dictHashAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS];
+        int const dictTagsMatchL = ZSTD_comparePackedTags(dictMatchIndexAndTagL, dictHashAndTagL);
+        int const dictTagsMatchS = ZSTD_comparePackedTags(dictMatchIndexAndTagS, dictHashAndTagS);
+        U32 const curr = (U32)(ip-base);
+        U32 const matchIndexL = hashLong[h2];
+        U32 matchIndexS = hashSmall[h];
+        const BYTE* matchLong = base + matchIndexL;
+        const BYTE* match = base + matchIndexS;
+        const U32 repIndex = curr + 1 - offset_1;
+        const BYTE* repMatch = (repIndex < prefixLowestIndex) ?
+                               dictBase + (repIndex - dictIndexDelta) :
+                               base + repIndex;
+        hashLong[h2] = hashSmall[h] = curr;   /* update hash tables */
+
+        /* check repcode */
+        if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex))
+            && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
+            goto _match_stored;
+        }
+
+        if ((matchIndexL >= prefixLowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
+            /* check prefix long match */
+            mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
+            offset = (U32)(ip-matchLong);
+            while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
+            goto _match_found;
+        } else if (dictTagsMatchL) {
+            /* check dictMatchState long match */
+            U32 const dictMatchIndexL = dictMatchIndexAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS;
+            const BYTE* dictMatchL = dictBase + dictMatchIndexL;
+            assert(dictMatchL < dictEnd);
+
+            if (dictMatchL > dictStart && MEM_read64(dictMatchL) == MEM_read64(ip)) {
+                mLength = ZSTD_count_2segments(ip+8, dictMatchL+8, iend, dictEnd, prefixLowest) + 8;
+                offset = (U32)(curr - dictMatchIndexL - dictIndexDelta);
+                while (((ip>anchor) & (dictMatchL>dictStart)) && (ip[-1] == dictMatchL[-1])) { ip--; dictMatchL--; mLength++; } /* catch up */
+                goto _match_found;
+        }   }
+
+        if (matchIndexS > prefixLowestIndex) {
+            /* short match  candidate */
+            if (MEM_read32(match) == MEM_read32(ip)) {
+                goto _search_next_long;
+            }
+        } else if (dictTagsMatchS) {
+            /* check dictMatchState short match */
+            U32 const dictMatchIndexS = dictMatchIndexAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS;
+            match = dictBase + dictMatchIndexS;
+            matchIndexS = dictMatchIndexS + dictIndexDelta;
+
+            if (match > dictStart && MEM_read32(match) == MEM_read32(ip)) {
+                goto _search_next_long;
+        }   }
+
+        ip += ((ip-anchor) >> kSearchStrength) + 1;
+#if defined(__aarch64__)
+        PREFETCH_L1(ip+256);
+#endif
+        continue;
+
+_search_next_long:
+        {   size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+            size_t const dictHashAndTagL3 = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
+            U32 const matchIndexL3 = hashLong[hl3];
+            U32 const dictMatchIndexAndTagL3 = dictHashLong[dictHashAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS];
+            int const dictTagsMatchL3 = ZSTD_comparePackedTags(dictMatchIndexAndTagL3, dictHashAndTagL3);
+            const BYTE* matchL3 = base + matchIndexL3;
+            hashLong[hl3] = curr + 1;
+
+            /* check prefix long +1 match */
+            if ((matchIndexL3 >= prefixLowestIndex) && (MEM_read64(matchL3) == MEM_read64(ip+1))) {
+                mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
+                ip++;
+                offset = (U32)(ip-matchL3);
+                while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
+                goto _match_found;
+            } else if (dictTagsMatchL3) {
+                /* check dict long +1 match */
+                U32 const dictMatchIndexL3 = dictMatchIndexAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS;
+                const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
+                assert(dictMatchL3 < dictEnd);
+                if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
+                    mLength = ZSTD_count_2segments(ip+1+8, dictMatchL3+8, iend, dictEnd, prefixLowest) + 8;
+                    ip++;
+                    offset = (U32)(curr + 1 - dictMatchIndexL3 - dictIndexDelta);
+                    while (((ip>anchor) & (dictMatchL3>dictStart)) && (ip[-1] == dictMatchL3[-1])) { ip--; dictMatchL3--; mLength++; } /* catch up */
+                    goto _match_found;
+        }   }   }
+
+        /* if no long +1 match, explore the short match we found */
+        if (matchIndexS < prefixLowestIndex) {
+            mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
+            offset = (U32)(curr - matchIndexS);
+            while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+        } else {
+            mLength = ZSTD_count(ip+4, match+4, iend) + 4;
+            offset = (U32)(ip - match);
+            while (((ip>anchor) & (match>prefixLowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+        }
+
+_match_found:
+        offset_2 = offset_1;
+        offset_1 = offset;
+
+        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+
+_match_stored:
+        /* match found */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Complementary insertion */
+            /* done after iLimit test, as candidates could be > iend-8 */
+            {   U32 const indexToInsert = curr+2;
+                hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
+                hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+                hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
+                hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
+            }
+
+            /* check immediate repcode */
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < prefixLowestIndex ?
+                        dictBase + repIndex2 - dictIndexDelta :
+                        base + repIndex2;
+                if ( (ZSTD_index_overlap_check(prefixLowestIndex, repIndex2))
+                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
+                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
+                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
+                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
+                    ip += repLength2;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+            }
+        }
+    }   /* while (ip < ilimit) */
+
+    /* save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+
+#define ZSTD_GEN_DFAST_FN(dictMode, mls)                                                                 \
+    static size_t ZSTD_compressBlock_doubleFast_##dictMode##_##mls(                                      \
+            ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                          \
+            void const* src, size_t srcSize)                                                             \
+    {                                                                                                    \
+        return ZSTD_compressBlock_doubleFast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls); \
+    }
+
+ZSTD_GEN_DFAST_FN(noDict, 4)
+ZSTD_GEN_DFAST_FN(noDict, 5)
+ZSTD_GEN_DFAST_FN(noDict, 6)
+ZSTD_GEN_DFAST_FN(noDict, 7)
+
+ZSTD_GEN_DFAST_FN(dictMatchState, 4)
+ZSTD_GEN_DFAST_FN(dictMatchState, 5)
+ZSTD_GEN_DFAST_FN(dictMatchState, 6)
+ZSTD_GEN_DFAST_FN(dictMatchState, 7)
+
+
+size_t ZSTD_compressBlock_doubleFast(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    const U32 mls = ms->cParams.minMatch;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_noDict_4(ms, seqStore, rep, src, srcSize);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_noDict_5(ms, seqStore, rep, src, srcSize);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_noDict_6(ms, seqStore, rep, src, srcSize);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_noDict_7(ms, seqStore, rep, src, srcSize);
+    }
+}
+
+
+size_t ZSTD_compressBlock_doubleFast_dictMatchState(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    const U32 mls = ms->cParams.minMatch;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_dictMatchState_4(ms, seqStore, rep, src, srcSize);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_dictMatchState_5(ms, seqStore, rep, src, srcSize);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_dictMatchState_6(ms, seqStore, rep, src, srcSize);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_dictMatchState_7(ms, seqStore, rep, src, srcSize);
+    }
+}
+
+
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_compressBlock_doubleFast_extDict_generic(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const mls /* template */)
+{
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
+    U32* const hashLong = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ms->window.base;
+    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
+    const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
+    const U32   dictStartIndex = lowLimit;
+    const U32   dictLimit = ms->window.dictLimit;
+    const U32   prefixStartIndex = (dictLimit > lowLimit) ? dictLimit : lowLimit;
+    const BYTE* const prefixStart = base + prefixStartIndex;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const dictStart = dictBase + dictStartIndex;
+    const BYTE* const dictEnd = dictBase + prefixStartIndex;
+    U32 offset_1=rep[0], offset_2=rep[1];
+
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_extDict_generic (srcSize=%zu)", srcSize);
+
+    /* if extDict is invalidated due to maxDistance, switch to "regular" variant */
+    if (prefixStartIndex == dictStartIndex)
+        return ZSTD_compressBlock_doubleFast(ms, seqStore, rep, src, srcSize);
+
+    /* Search Loop */
+    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
+        const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
+        const U32 matchIndex = hashSmall[hSmall];
+        const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
+        const BYTE* match = matchBase + matchIndex;
+
+        const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
+        const U32 matchLongIndex = hashLong[hLong];
+        const BYTE* const matchLongBase = matchLongIndex < prefixStartIndex ? dictBase : base;
+        const BYTE* matchLong = matchLongBase + matchLongIndex;
+
+        const U32 curr = (U32)(ip-base);
+        const U32 repIndex = curr + 1 - offset_1;   /* offset_1 expected <= curr +1 */
+        const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
+        const BYTE* const repMatch = repBase + repIndex;
+        size_t mLength;
+        hashSmall[hSmall] = hashLong[hLong] = curr;   /* update hash table */
+
+        if (((ZSTD_index_overlap_check(prefixStartIndex, repIndex))
+            & (offset_1 <= curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
+          && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
+        } else {
+            if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
+                const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
+                const BYTE* const lowMatchPtr = matchLongIndex < prefixStartIndex ? dictStart : prefixStart;
+                U32 offset;
+                mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, prefixStart) + 8;
+                offset = curr - matchLongIndex;
+                while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+
+            } else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
+                size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+                U32 const matchIndex3 = hashLong[h3];
+                const BYTE* const match3Base = matchIndex3 < prefixStartIndex ? dictBase : base;
+                const BYTE* match3 = match3Base + matchIndex3;
+                U32 offset;
+                hashLong[h3] = curr + 1;
+                if ( (matchIndex3 > dictStartIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
+                    const BYTE* const matchEnd = matchIndex3 < prefixStartIndex ? dictEnd : iend;
+                    const BYTE* const lowMatchPtr = matchIndex3 < prefixStartIndex ? dictStart : prefixStart;
+                    mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, prefixStart) + 8;
+                    ip++;
+                    offset = curr+1 - matchIndex3;
+                    while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
+                } else {
+                    const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
+                    const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
+                    mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
+                    offset = curr - matchIndex;
+                    while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
+                }
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+
+            } else {
+                ip += ((ip-anchor) >> kSearchStrength) + 1;
+                continue;
+        }   }
+
+        /* move to next sequence start */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Complementary insertion */
+            /* done after iLimit test, as candidates could be > iend-8 */
+            {   U32 const indexToInsert = curr+2;
+                hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
+                hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+                hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
+                hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
+            }
+
+            /* check immediate repcode */
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
+                if ( ((ZSTD_index_overlap_check(prefixStartIndex, repIndex2))
+                    & (offset_2 <= current2 - dictStartIndex))
+                  && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                    U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
+                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
+                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
+                    ip += repLength2;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+    }   }   }
+
+    /* save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+
+ZSTD_GEN_DFAST_FN(extDict, 4)
+ZSTD_GEN_DFAST_FN(extDict, 5)
+ZSTD_GEN_DFAST_FN(extDict, 6)
+ZSTD_GEN_DFAST_FN(extDict, 7)
+
+size_t ZSTD_compressBlock_doubleFast_extDict(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    U32 const mls = ms->cParams.minMatch;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_extDict_4(ms, seqStore, rep, src, srcSize);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_extDict_5(ms, seqStore, rep, src, srcSize);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_extDict_6(ms, seqStore, rep, src, srcSize);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_extDict_7(ms, seqStore, rep, src, srcSize);
+    }
+}
+
+#endif /* ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR */
diff --git a/lib/compress/zstd_lazy.c b/lib/compress/zstd_lazy.c
new file mode 100644
index 0000000..272ebe0
--- /dev/null
+++ b/lib/compress/zstd_lazy.c
@@ -0,0 +1,2199 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_lazy.h"
+#include "../common/bits.h" /* ZSTD_countTrailingZeros64 */
+
+#if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR)
+
+#define kLazySkippingStep 8
+
+
+/*-*************************************
+*  Binary Tree search
+***************************************/
+
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_updateDUBT(ZSTD_MatchState_t* ms,
+                const BYTE* ip, const BYTE* iend,
+                U32 mls)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32  const hashLog = cParams->hashLog;
+
+    U32* const bt = ms->chainTable;
+    U32  const btLog  = cParams->chainLog - 1;
+    U32  const btMask = (1 << btLog) - 1;
+
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+
+    if (idx != target)
+        DEBUGLOG(7, "ZSTD_updateDUBT, from %u to %u (dictLimit:%u)",
+                    idx, target, ms->window.dictLimit);
+    assert(ip + 8 <= iend);   /* condition for ZSTD_hashPtr */
+    (void)iend;
+
+    assert(idx >= ms->window.dictLimit);   /* condition for valid base+idx */
+    for ( ; idx < target ; idx++) {
+        size_t const h  = ZSTD_hashPtr(base + idx, hashLog, mls);   /* assumption : ip + 8 <= iend */
+        U32    const matchIndex = hashTable[h];
+
+        U32*   const nextCandidatePtr = bt + 2*(idx&btMask);
+        U32*   const sortMarkPtr  = nextCandidatePtr + 1;
+
+        DEBUGLOG(8, "ZSTD_updateDUBT: insert %u", idx);
+        hashTable[h] = idx;   /* Update Hash Table */
+        *nextCandidatePtr = matchIndex;   /* update BT like a chain */
+        *sortMarkPtr = ZSTD_DUBT_UNSORTED_MARK;
+    }
+    ms->nextToUpdate = target;
+}
+
+
+/** ZSTD_insertDUBT1() :
+ *  sort one already inserted but unsorted position
+ *  assumption : curr >= btlow == (curr - btmask)
+ *  doesn't fail */
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_insertDUBT1(const ZSTD_MatchState_t* ms,
+                 U32 curr, const BYTE* inputEnd,
+                 U32 nbCompares, U32 btLow,
+                 const ZSTD_dictMode_e dictMode)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const bt = ms->chainTable;
+    U32  const btLog  = cParams->chainLog - 1;
+    U32  const btMask = (1 << btLog) - 1;
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const ip = (curr>=dictLimit) ? base + curr : dictBase + curr;
+    const BYTE* const iend = (curr>=dictLimit) ? inputEnd : dictBase + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* match;
+    U32* smallerPtr = bt + 2*(curr&btMask);
+    U32* largerPtr  = smallerPtr + 1;
+    U32 matchIndex = *smallerPtr;   /* this candidate is unsorted : next sorted candidate is reached through *smallerPtr, while *largerPtr contains previous unsorted candidate (which is already saved and can be overwritten) */
+    U32 dummy32;   /* to be nullified at the end */
+    U32 const windowValid = ms->window.lowLimit;
+    U32 const maxDistance = 1U << cParams->windowLog;
+    U32 const windowLow = (curr - windowValid > maxDistance) ? curr - maxDistance : windowValid;
+
+
+    DEBUGLOG(8, "ZSTD_insertDUBT1(%u) (dictLimit=%u, lowLimit=%u)",
+                curr, dictLimit, windowLow);
+    assert(curr >= btLow);
+    assert(ip < iend);   /* condition for ZSTD_count */
+
+    for (; nbCompares && (matchIndex > windowLow); --nbCompares) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(matchIndex < curr);
+        /* note : all candidates are now supposed sorted,
+         * but it's still possible to have nextPtr[1] == ZSTD_DUBT_UNSORTED_MARK
+         * when a real index has the same value as ZSTD_DUBT_UNSORTED_MARK */
+
+        if ( (dictMode != ZSTD_extDict)
+          || (matchIndex+matchLength >= dictLimit)  /* both in current segment*/
+          || (curr < dictLimit) /* both in extDict */) {
+            const BYTE* const mBase = ( (dictMode != ZSTD_extDict)
+                                     || (matchIndex+matchLength >= dictLimit)) ?
+                                        base : dictBase;
+            assert( (matchIndex+matchLength >= dictLimit)   /* might be wrong if extDict is incorrectly set to 0 */
+                 || (curr < dictLimit) );
+            match = mBase + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* preparation for next read of match[matchLength] */
+        }
+
+        DEBUGLOG(8, "ZSTD_insertDUBT1: comparing %u with %u : found %u common bytes ",
+                    curr, matchIndex, (U32)matchLength);
+
+        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
+        }
+
+        if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is smaller : next => %u",
+                        matchIndex, btLow, nextPtr[1]);
+            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is larger => %u",
+                        matchIndex, btLow, nextPtr[0]);
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+}
+
+
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_DUBT_findBetterDictMatch (
+        const ZSTD_MatchState_t* ms,
+        const BYTE* const ip, const BYTE* const iend,
+        size_t* offsetPtr,
+        size_t bestLength,
+        U32 nbCompares,
+        U32 const mls,
+        const ZSTD_dictMode_e dictMode)
+{
+    const ZSTD_MatchState_t * const dms = ms->dictMatchState;
+    const ZSTD_compressionParameters* const dmsCParams = &dms->cParams;
+    const U32 * const dictHashTable = dms->hashTable;
+    U32         const hashLog = dmsCParams->hashLog;
+    size_t      const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32               dictMatchIndex = dictHashTable[h];
+
+    const BYTE* const base = ms->window.base;
+    const BYTE* const prefixStart = base + ms->window.dictLimit;
+    U32         const curr = (U32)(ip-base);
+    const BYTE* const dictBase = dms->window.base;
+    const BYTE* const dictEnd = dms->window.nextSrc;
+    U32         const dictHighLimit = (U32)(dms->window.nextSrc - dms->window.base);
+    U32         const dictLowLimit = dms->window.lowLimit;
+    U32         const dictIndexDelta = ms->window.lowLimit - dictHighLimit;
+
+    U32*        const dictBt = dms->chainTable;
+    U32         const btLog  = dmsCParams->chainLog - 1;
+    U32         const btMask = (1 << btLog) - 1;
+    U32         const btLow = (btMask >= dictHighLimit - dictLowLimit) ? dictLowLimit : dictHighLimit - btMask;
+
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+
+    (void)dictMode;
+    assert(dictMode == ZSTD_dictMatchState);
+
+    for (; nbCompares && (dictMatchIndex > dictLowLimit); --nbCompares) {
+        U32* const nextPtr = dictBt + 2*(dictMatchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        const BYTE* match = dictBase + dictMatchIndex;
+        matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+        if (dictMatchIndex+matchLength >= dictHighLimit)
+            match = base + dictMatchIndex + dictIndexDelta;   /* to prepare for next usage of match[matchLength] */
+
+        if (matchLength > bestLength) {
+            U32 matchIndex = dictMatchIndex + dictIndexDelta;
+            if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) ) {
+                DEBUGLOG(9, "ZSTD_DUBT_findBetterDictMatch(%u) : found better match length %u -> %u and offsetCode %u -> %u (dictMatchIndex %u, matchIndex %u)",
+                    curr, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, OFFSET_TO_OFFBASE(curr - matchIndex), dictMatchIndex, matchIndex);
+                bestLength = matchLength, *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
+            }
+            if (ip+matchLength == iend) {   /* reached end of input : ip[matchLength] is not valid, no way to know if it's larger or smaller than match */
+                break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+            }
+        }
+
+        if (match[matchLength] < ip[matchLength]) {
+            if (dictMatchIndex <= btLow) { break; }   /* beyond tree size, stop the search */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            dictMatchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+        } else {
+            /* match is larger than current */
+            if (dictMatchIndex <= btLow) { break; }   /* beyond tree size, stop the search */
+            commonLengthLarger = matchLength;
+            dictMatchIndex = nextPtr[0];
+        }
+    }
+
+    if (bestLength >= MINMATCH) {
+        U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offsetPtr); (void)mIndex;
+        DEBUGLOG(8, "ZSTD_DUBT_findBetterDictMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
+                    curr, (U32)bestLength, (U32)*offsetPtr, mIndex);
+    }
+    return bestLength;
+
+}
+
+
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_DUBT_findBestMatch(ZSTD_MatchState_t* ms,
+                        const BYTE* const ip, const BYTE* const iend,
+                        size_t* offBasePtr,
+                        U32 const mls,
+                        const ZSTD_dictMode_e dictMode)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32*   const hashTable = ms->hashTable;
+    U32    const hashLog = cParams->hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32          matchIndex  = hashTable[h];
+
+    const BYTE* const base = ms->window.base;
+    U32    const curr = (U32)(ip-base);
+    U32    const windowLow = ZSTD_getLowestMatchIndex(ms, curr, cParams->windowLog);
+
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32    const btLow = (btMask >= curr) ? 0 : curr - btMask;
+    U32    const unsortLimit = MAX(btLow, windowLow);
+
+    U32*         nextCandidate = bt + 2*(matchIndex&btMask);
+    U32*         unsortedMark = bt + 2*(matchIndex&btMask) + 1;
+    U32          nbCompares = 1U << cParams->searchLog;
+    U32          nbCandidates = nbCompares;
+    U32          previousCandidate = 0;
+
+    DEBUGLOG(7, "ZSTD_DUBT_findBestMatch (%u) ", curr);
+    assert(ip <= iend-8);   /* required for h calculation */
+    assert(dictMode != ZSTD_dedicatedDictSearch);
+
+    /* reach end of unsorted candidates list */
+    while ( (matchIndex > unsortLimit)
+         && (*unsortedMark == ZSTD_DUBT_UNSORTED_MARK)
+         && (nbCandidates > 1) ) {
+        DEBUGLOG(8, "ZSTD_DUBT_findBestMatch: candidate %u is unsorted",
+                    matchIndex);
+        *unsortedMark = previousCandidate;  /* the unsortedMark becomes a reversed chain, to move up back to original position */
+        previousCandidate = matchIndex;
+        matchIndex = *nextCandidate;
+        nextCandidate = bt + 2*(matchIndex&btMask);
+        unsortedMark = bt + 2*(matchIndex&btMask) + 1;
+        nbCandidates --;
+    }
+
+    /* nullify last candidate if it's still unsorted
+     * simplification, detrimental to compression ratio, beneficial for speed */
+    if ( (matchIndex > unsortLimit)
+      && (*unsortedMark==ZSTD_DUBT_UNSORTED_MARK) ) {
+        DEBUGLOG(7, "ZSTD_DUBT_findBestMatch: nullify last unsorted candidate %u",
+                    matchIndex);
+        *nextCandidate = *unsortedMark = 0;
+    }
+
+    /* batch sort stacked candidates */
+    matchIndex = previousCandidate;
+    while (matchIndex) {  /* will end on matchIndex == 0 */
+        U32* const nextCandidateIdxPtr = bt + 2*(matchIndex&btMask) + 1;
+        U32 const nextCandidateIdx = *nextCandidateIdxPtr;
+        ZSTD_insertDUBT1(ms, matchIndex, iend,
+                         nbCandidates, unsortLimit, dictMode);
+        matchIndex = nextCandidateIdx;
+        nbCandidates++;
+    }
+
+    /* find longest match */
+    {   size_t commonLengthSmaller = 0, commonLengthLarger = 0;
+        const BYTE* const dictBase = ms->window.dictBase;
+        const U32 dictLimit = ms->window.dictLimit;
+        const BYTE* const dictEnd = dictBase + dictLimit;
+        const BYTE* const prefixStart = base + dictLimit;
+        U32* smallerPtr = bt + 2*(curr&btMask);
+        U32* largerPtr  = bt + 2*(curr&btMask) + 1;
+        U32 matchEndIdx = curr + 8 + 1;
+        U32 dummy32;   /* to be nullified at the end */
+        size_t bestLength = 0;
+
+        matchIndex  = hashTable[h];
+        hashTable[h] = curr;   /* Update Hash Table */
+
+        for (; nbCompares && (matchIndex > windowLow); --nbCompares) {
+            U32* const nextPtr = bt + 2*(matchIndex & btMask);
+            size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+            const BYTE* match;
+
+            if ((dictMode != ZSTD_extDict) || (matchIndex+matchLength >= dictLimit)) {
+                match = base + matchIndex;
+                matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+            } else {
+                match = dictBase + matchIndex;
+                matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+                if (matchIndex+matchLength >= dictLimit)
+                    match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+            }
+
+            if (matchLength > bestLength) {
+                if (matchLength > matchEndIdx - matchIndex)
+                    matchEndIdx = matchIndex + (U32)matchLength;
+                if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr - matchIndex + 1) - ZSTD_highbit32((U32)*offBasePtr)) )
+                    bestLength = matchLength, *offBasePtr = OFFSET_TO_OFFBASE(curr - matchIndex);
+                if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+                    if (dictMode == ZSTD_dictMatchState) {
+                        nbCompares = 0; /* in addition to avoiding checking any
+                                         * further in this loop, make sure we
+                                         * skip checking in the dictionary. */
+                    }
+                    break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+                }
+            }
+
+            if (match[matchLength] < ip[matchLength]) {
+                /* match is smaller than current */
+                *smallerPtr = matchIndex;             /* update smaller idx */
+                commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+                if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+                smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+                matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            } else {
+                /* match is larger than current */
+                *largerPtr = matchIndex;
+                commonLengthLarger = matchLength;
+                if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+                largerPtr = nextPtr;
+                matchIndex = nextPtr[0];
+        }   }
+
+        *smallerPtr = *largerPtr = 0;
+
+        assert(nbCompares <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
+        if (dictMode == ZSTD_dictMatchState && nbCompares) {
+            bestLength = ZSTD_DUBT_findBetterDictMatch(
+                    ms, ip, iend,
+                    offBasePtr, bestLength, nbCompares,
+                    mls, dictMode);
+        }
+
+        assert(matchEndIdx > curr+8); /* ensure nextToUpdate is increased */
+        ms->nextToUpdate = matchEndIdx - 8;   /* skip repetitive patterns */
+        if (bestLength >= MINMATCH) {
+            U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offBasePtr); (void)mIndex;
+            DEBUGLOG(8, "ZSTD_DUBT_findBestMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
+                        curr, (U32)bestLength, (U32)*offBasePtr, mIndex);
+        }
+        return bestLength;
+    }
+}
+
+
+/** ZSTD_BtFindBestMatch() : Tree updater, providing best match */
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_BtFindBestMatch( ZSTD_MatchState_t* ms,
+                const BYTE* const ip, const BYTE* const iLimit,
+                      size_t* offBasePtr,
+                const U32 mls /* template */,
+                const ZSTD_dictMode_e dictMode)
+{
+    DEBUGLOG(7, "ZSTD_BtFindBestMatch");
+    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateDUBT(ms, ip, iLimit, mls);
+    return ZSTD_DUBT_findBestMatch(ms, ip, iLimit, offBasePtr, mls, dictMode);
+}
+
+/***********************************
+* Dedicated dict search
+***********************************/
+
+void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_MatchState_t* ms, const BYTE* const ip)
+{
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32* const hashTable = ms->hashTable;
+    U32* const chainTable = ms->chainTable;
+    U32 const chainSize = 1 << ms->cParams.chainLog;
+    U32 idx = ms->nextToUpdate;
+    U32 const minChain = chainSize < target - idx ? target - chainSize : idx;
+    U32 const bucketSize = 1 << ZSTD_LAZY_DDSS_BUCKET_LOG;
+    U32 const cacheSize = bucketSize - 1;
+    U32 const chainAttempts = (1 << ms->cParams.searchLog) - cacheSize;
+    U32 const chainLimit = chainAttempts > 255 ? 255 : chainAttempts;
+
+    /* We know the hashtable is oversized by a factor of `bucketSize`.
+     * We are going to temporarily pretend `bucketSize == 1`, keeping only a
+     * single entry. We will use the rest of the space to construct a temporary
+     * chaintable.
+     */
+    U32 const hashLog = ms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
+    U32* const tmpHashTable = hashTable;
+    U32* const tmpChainTable = hashTable + ((size_t)1 << hashLog);
+    U32 const tmpChainSize = (U32)((1 << ZSTD_LAZY_DDSS_BUCKET_LOG) - 1) << hashLog;
+    U32 const tmpMinChain = tmpChainSize < target ? target - tmpChainSize : idx;
+    U32 hashIdx;
+
+    assert(ms->cParams.chainLog <= 24);
+    assert(ms->cParams.hashLog > ms->cParams.chainLog);
+    assert(idx != 0);
+    assert(tmpMinChain <= minChain);
+
+    /* fill conventional hash table and conventional chain table */
+    for ( ; idx < target; idx++) {
+        U32 const h = (U32)ZSTD_hashPtr(base + idx, hashLog, ms->cParams.minMatch);
+        if (idx >= tmpMinChain) {
+            tmpChainTable[idx - tmpMinChain] = hashTable[h];
+        }
+        tmpHashTable[h] = idx;
+    }
+
+    /* sort chains into ddss chain table */
+    {
+        U32 chainPos = 0;
+        for (hashIdx = 0; hashIdx < (1U << hashLog); hashIdx++) {
+            U32 count;
+            U32 countBeyondMinChain = 0;
+            U32 i = tmpHashTable[hashIdx];
+            for (count = 0; i >= tmpMinChain && count < cacheSize; count++) {
+                /* skip through the chain to the first position that won't be
+                 * in the hash cache bucket */
+                if (i < minChain) {
+                    countBeyondMinChain++;
+                }
+                i = tmpChainTable[i - tmpMinChain];
+            }
+            if (count == cacheSize) {
+                for (count = 0; count < chainLimit;) {
+                    if (i < minChain) {
+                        if (!i || ++countBeyondMinChain > cacheSize) {
+                            /* only allow pulling `cacheSize` number of entries
+                             * into the cache or chainTable beyond `minChain`,
+                             * to replace the entries pulled out of the
+                             * chainTable into the cache. This lets us reach
+                             * back further without increasing the total number
+                             * of entries in the chainTable, guaranteeing the
+                             * DDSS chain table will fit into the space
+                             * allocated for the regular one. */
+                            break;
+                        }
+                    }
+                    chainTable[chainPos++] = i;
+                    count++;
+                    if (i < tmpMinChain) {
+                        break;
+                    }
+                    i = tmpChainTable[i - tmpMinChain];
+                }
+            } else {
+                count = 0;
+            }
+            if (count) {
+                tmpHashTable[hashIdx] = ((chainPos - count) << 8) + count;
+            } else {
+                tmpHashTable[hashIdx] = 0;
+            }
+        }
+        assert(chainPos <= chainSize); /* I believe this is guaranteed... */
+    }
+
+    /* move chain pointers into the last entry of each hash bucket */
+    for (hashIdx = (1 << hashLog); hashIdx; ) {
+        U32 const bucketIdx = --hashIdx << ZSTD_LAZY_DDSS_BUCKET_LOG;
+        U32 const chainPackedPointer = tmpHashTable[hashIdx];
+        U32 i;
+        for (i = 0; i < cacheSize; i++) {
+            hashTable[bucketIdx + i] = 0;
+        }
+        hashTable[bucketIdx + bucketSize - 1] = chainPackedPointer;
+    }
+
+    /* fill the buckets of the hash table */
+    for (idx = ms->nextToUpdate; idx < target; idx++) {
+        U32 const h = (U32)ZSTD_hashPtr(base + idx, hashLog, ms->cParams.minMatch)
+                   << ZSTD_LAZY_DDSS_BUCKET_LOG;
+        U32 i;
+        /* Shift hash cache down 1. */
+        for (i = cacheSize - 1; i; i--)
+            hashTable[h + i] = hashTable[h + i - 1];
+        hashTable[h] = idx;
+    }
+
+    ms->nextToUpdate = target;
+}
+
+/* Returns the longest match length found in the dedicated dict search structure.
+ * If none are longer than the argument ml, then ml will be returned.
+ */
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_dedicatedDictSearch_lazy_search(size_t* offsetPtr, size_t ml, U32 nbAttempts,
+                                            const ZSTD_MatchState_t* const dms,
+                                            const BYTE* const ip, const BYTE* const iLimit,
+                                            const BYTE* const prefixStart, const U32 curr,
+                                            const U32 dictLimit, const size_t ddsIdx) {
+    const U32 ddsLowestIndex  = dms->window.dictLimit;
+    const BYTE* const ddsBase = dms->window.base;
+    const BYTE* const ddsEnd  = dms->window.nextSrc;
+    const U32 ddsSize         = (U32)(ddsEnd - ddsBase);
+    const U32 ddsIndexDelta   = dictLimit - ddsSize;
+    const U32 bucketSize      = (1 << ZSTD_LAZY_DDSS_BUCKET_LOG);
+    const U32 bucketLimit     = nbAttempts < bucketSize - 1 ? nbAttempts : bucketSize - 1;
+    U32 ddsAttempt;
+    U32 matchIndex;
+
+    for (ddsAttempt = 0; ddsAttempt < bucketSize - 1; ddsAttempt++) {
+        PREFETCH_L1(ddsBase + dms->hashTable[ddsIdx + ddsAttempt]);
+    }
+
+    {
+        U32 const chainPackedPointer = dms->hashTable[ddsIdx + bucketSize - 1];
+        U32 const chainIndex = chainPackedPointer >> 8;
+
+        PREFETCH_L1(&dms->chainTable[chainIndex]);
+    }
+
+    for (ddsAttempt = 0; ddsAttempt < bucketLimit; ddsAttempt++) {
+        size_t currentMl=0;
+        const BYTE* match;
+        matchIndex = dms->hashTable[ddsIdx + ddsAttempt];
+        match = ddsBase + matchIndex;
+
+        if (!matchIndex) {
+            return ml;
+        }
+
+        /* guaranteed by table construction */
+        (void)ddsLowestIndex;
+        assert(matchIndex >= ddsLowestIndex);
+        assert(match+4 <= ddsEnd);
+        if (MEM_read32(match) == MEM_read32(ip)) {
+            /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+            currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, ddsEnd, prefixStart) + 4;
+        }
+
+        /* save best solution */
+        if (currentMl > ml) {
+            ml = currentMl;
+            *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta));
+            if (ip+currentMl == iLimit) {
+                /* best possible, avoids read overflow on next attempt */
+                return ml;
+            }
+        }
+    }
+
+    {
+        U32 const chainPackedPointer = dms->hashTable[ddsIdx + bucketSize - 1];
+        U32 chainIndex = chainPackedPointer >> 8;
+        U32 const chainLength = chainPackedPointer & 0xFF;
+        U32 const chainAttempts = nbAttempts - ddsAttempt;
+        U32 const chainLimit = chainAttempts > chainLength ? chainLength : chainAttempts;
+        U32 chainAttempt;
+
+        for (chainAttempt = 0 ; chainAttempt < chainLimit; chainAttempt++) {
+            PREFETCH_L1(ddsBase + dms->chainTable[chainIndex + chainAttempt]);
+        }
+
+        for (chainAttempt = 0 ; chainAttempt < chainLimit; chainAttempt++, chainIndex++) {
+            size_t currentMl=0;
+            const BYTE* match;
+            matchIndex = dms->chainTable[chainIndex];
+            match = ddsBase + matchIndex;
+
+            /* guaranteed by table construction */
+            assert(matchIndex >= ddsLowestIndex);
+            assert(match+4 <= ddsEnd);
+            if (MEM_read32(match) == MEM_read32(ip)) {
+                /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, ddsEnd, prefixStart) + 4;
+            }
+
+            /* save best solution */
+            if (currentMl > ml) {
+                ml = currentMl;
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta));
+                if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+            }
+        }
+    }
+    return ml;
+}
+
+
+/* *********************************
+*  Hash Chain
+***********************************/
+#define NEXT_IN_CHAIN(d, mask)   chainTable[(d) & (mask)]
+
+/* Update chains up to ip (excluded)
+   Assumption : always within prefix (i.e. not within extDict) */
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_insertAndFindFirstIndex_internal(
+                        ZSTD_MatchState_t* ms,
+                        const ZSTD_compressionParameters* const cParams,
+                        const BYTE* ip, U32 const mls, U32 const lazySkipping)
+{
+    U32* const hashTable  = ms->hashTable;
+    const U32 hashLog = cParams->hashLog;
+    U32* const chainTable = ms->chainTable;
+    const U32 chainMask = (1 << cParams->chainLog) - 1;
+    const BYTE* const base = ms->window.base;
+    const U32 target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+
+    while(idx < target) { /* catch up */
+        size_t const h = ZSTD_hashPtr(base+idx, hashLog, mls);
+        NEXT_IN_CHAIN(idx, chainMask) = hashTable[h];
+        hashTable[h] = idx;
+        idx++;
+        /* Stop inserting every position when in the lazy skipping mode. */
+        if (lazySkipping)
+            break;
+    }
+
+    ms->nextToUpdate = target;
+    return hashTable[ZSTD_hashPtr(ip, hashLog, mls)];
+}
+
+U32 ZSTD_insertAndFindFirstIndex(ZSTD_MatchState_t* ms, const BYTE* ip) {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, ms->cParams.minMatch, /* lazySkipping*/ 0);
+}
+
+/* inlining is important to hardwire a hot branch (template emulation) */
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_HcFindBestMatch(
+                        ZSTD_MatchState_t* ms,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 mls, const ZSTD_dictMode_e dictMode)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const chainTable = ms->chainTable;
+    const U32 chainSize = (1 << cParams->chainLog);
+    const U32 chainMask = chainSize-1;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const U32 curr = (U32)(ip-base);
+    const U32 maxDistance = 1U << cParams->windowLog;
+    const U32 lowestValid = ms->window.lowLimit;
+    const U32 withinMaxDistance = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
+    const U32 isDictionary = (ms->loadedDictEnd != 0);
+    const U32 lowLimit = isDictionary ? lowestValid : withinMaxDistance;
+    const U32 minChain = curr > chainSize ? curr - chainSize : 0;
+    U32 nbAttempts = 1U << cParams->searchLog;
+    size_t ml=4-1;
+
+    const ZSTD_MatchState_t* const dms = ms->dictMatchState;
+    const U32 ddsHashLog = dictMode == ZSTD_dedicatedDictSearch
+                         ? dms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG : 0;
+    const size_t ddsIdx = dictMode == ZSTD_dedicatedDictSearch
+                        ? ZSTD_hashPtr(ip, ddsHashLog, mls) << ZSTD_LAZY_DDSS_BUCKET_LOG : 0;
+
+    U32 matchIndex;
+
+    if (dictMode == ZSTD_dedicatedDictSearch) {
+        const U32* entry = &dms->hashTable[ddsIdx];
+        PREFETCH_L1(entry);
+    }
+
+    /* HC4 match finder */
+    matchIndex = ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, mls, ms->lazySkipping);
+
+    for ( ; (matchIndex>=lowLimit) & (nbAttempts>0) ; nbAttempts--) {
+        size_t currentMl=0;
+        if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
+            const BYTE* const match = base + matchIndex;
+            assert(matchIndex >= dictLimit);   /* ensures this is true if dictMode != ZSTD_extDict */
+            /* read 4B starting from (match + ml + 1 - sizeof(U32)) */
+            if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3))   /* potentially better */
+                currentMl = ZSTD_count(ip, match, iLimit);
+        } else {
+            const BYTE* const match = dictBase + matchIndex;
+            assert(match+4 <= dictEnd);
+            if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4;
+        }
+
+        /* save best solution */
+        if (currentMl > ml) {
+            ml = currentMl;
+            *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
+            if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+        }
+
+        if (matchIndex <= minChain) break;
+        matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
+    }
+
+    assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
+    if (dictMode == ZSTD_dedicatedDictSearch) {
+        ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts, dms,
+                                                  ip, iLimit, prefixStart, curr, dictLimit, ddsIdx);
+    } else if (dictMode == ZSTD_dictMatchState) {
+        const U32* const dmsChainTable = dms->chainTable;
+        const U32 dmsChainSize         = (1 << dms->cParams.chainLog);
+        const U32 dmsChainMask         = dmsChainSize - 1;
+        const U32 dmsLowestIndex       = dms->window.dictLimit;
+        const BYTE* const dmsBase      = dms->window.base;
+        const BYTE* const dmsEnd       = dms->window.nextSrc;
+        const U32 dmsSize              = (U32)(dmsEnd - dmsBase);
+        const U32 dmsIndexDelta        = dictLimit - dmsSize;
+        const U32 dmsMinChain = dmsSize > dmsChainSize ? dmsSize - dmsChainSize : 0;
+
+        matchIndex = dms->hashTable[ZSTD_hashPtr(ip, dms->cParams.hashLog, mls)];
+
+        for ( ; (matchIndex>=dmsLowestIndex) & (nbAttempts>0) ; nbAttempts--) {
+            size_t currentMl=0;
+            const BYTE* const match = dmsBase + matchIndex;
+            assert(match+4 <= dmsEnd);
+            if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dmsEnd, prefixStart) + 4;
+
+            /* save best solution */
+            if (currentMl > ml) {
+                ml = currentMl;
+                assert(curr > matchIndex + dmsIndexDelta);
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta));
+                if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+            }
+
+            if (matchIndex <= dmsMinChain) break;
+
+            matchIndex = dmsChainTable[matchIndex & dmsChainMask];
+        }
+    }
+
+    return ml;
+}
+
+/* *********************************
+* (SIMD) Row-based matchfinder
+***********************************/
+/* Constants for row-based hash */
+#define ZSTD_ROW_HASH_TAG_MASK ((1u << ZSTD_ROW_HASH_TAG_BITS) - 1)
+#define ZSTD_ROW_HASH_MAX_ENTRIES 64    /* absolute maximum number of entries per row, for all configurations */
+
+#define ZSTD_ROW_HASH_CACHE_MASK (ZSTD_ROW_HASH_CACHE_SIZE - 1)
+
+typedef U64 ZSTD_VecMask;   /* Clarifies when we are interacting with a U64 representing a mask of matches */
+
+/* ZSTD_VecMask_next():
+ * Starting from the LSB, returns the idx of the next non-zero bit.
+ * Basically counting the nb of trailing zeroes.
+ */
+MEM_STATIC U32 ZSTD_VecMask_next(ZSTD_VecMask val) {
+    return ZSTD_countTrailingZeros64(val);
+}
+
+/* ZSTD_row_nextIndex():
+ * Returns the next index to insert at within a tagTable row, and updates the "head"
+ * value to reflect the update. Essentially cycles backwards from [1, {entries per row})
+ */
+FORCE_INLINE_TEMPLATE U32 ZSTD_row_nextIndex(BYTE* const tagRow, U32 const rowMask) {
+    U32 next = (*tagRow-1) & rowMask;
+    next += (next == 0) ? rowMask : 0; /* skip first position */
+    *tagRow = (BYTE)next;
+    return next;
+}
+
+/* ZSTD_isAligned():
+ * Checks that a pointer is aligned to "align" bytes which must be a power of 2.
+ */
+MEM_STATIC int ZSTD_isAligned(void const* ptr, size_t align) {
+    assert((align & (align - 1)) == 0);
+    return (((size_t)ptr) & (align - 1)) == 0;
+}
+
+/* ZSTD_row_prefetch():
+ * Performs prefetching for the hashTable and tagTable at a given row.
+ */
+FORCE_INLINE_TEMPLATE void ZSTD_row_prefetch(U32 const* hashTable, BYTE const* tagTable, U32 const relRow, U32 const rowLog) {
+    PREFETCH_L1(hashTable + relRow);
+    if (rowLog >= 5) {
+        PREFETCH_L1(hashTable + relRow + 16);
+        /* Note: prefetching more of the hash table does not appear to be beneficial for 128-entry rows */
+    }
+    PREFETCH_L1(tagTable + relRow);
+    if (rowLog == 6) {
+        PREFETCH_L1(tagTable + relRow + 32);
+    }
+    assert(rowLog == 4 || rowLog == 5 || rowLog == 6);
+    assert(ZSTD_isAligned(hashTable + relRow, 64));                 /* prefetched hash row always 64-byte aligned */
+    assert(ZSTD_isAligned(tagTable + relRow, (size_t)1 << rowLog)); /* prefetched tagRow sits on correct multiple of bytes (32,64,128) */
+}
+
+/* ZSTD_row_fillHashCache():
+ * Fill up the hash cache starting at idx, prefetching up to ZSTD_ROW_HASH_CACHE_SIZE entries,
+ * but not beyond iLimit.
+ */
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_row_fillHashCache(ZSTD_MatchState_t* ms, const BYTE* base,
+                                   U32 const rowLog, U32 const mls,
+                                   U32 idx, const BYTE* const iLimit)
+{
+    U32 const* const hashTable = ms->hashTable;
+    BYTE const* const tagTable = ms->tagTable;
+    U32 const hashLog = ms->rowHashLog;
+    U32 const maxElemsToPrefetch = (base + idx) > iLimit ? 0 : (U32)(iLimit - (base + idx) + 1);
+    U32 const lim = idx + MIN(ZSTD_ROW_HASH_CACHE_SIZE, maxElemsToPrefetch);
+
+    for (; idx < lim; ++idx) {
+        U32 const hash = (U32)ZSTD_hashPtrSalted(base + idx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt);
+        U32 const row = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
+        ZSTD_row_prefetch(hashTable, tagTable, row, rowLog);
+        ms->hashCache[idx & ZSTD_ROW_HASH_CACHE_MASK] = hash;
+    }
+
+    DEBUGLOG(6, "ZSTD_row_fillHashCache(): [%u %u %u %u %u %u %u %u]", ms->hashCache[0], ms->hashCache[1],
+                                                     ms->hashCache[2], ms->hashCache[3], ms->hashCache[4],
+                                                     ms->hashCache[5], ms->hashCache[6], ms->hashCache[7]);
+}
+
+/* ZSTD_row_nextCachedHash():
+ * Returns the hash of base + idx, and replaces the hash in the hash cache with the byte at
+ * base + idx + ZSTD_ROW_HASH_CACHE_SIZE. Also prefetches the appropriate rows from hashTable and tagTable.
+ */
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_row_nextCachedHash(U32* cache, U32 const* hashTable,
+                                                  BYTE const* tagTable, BYTE const* base,
+                                                  U32 idx, U32 const hashLog,
+                                                  U32 const rowLog, U32 const mls,
+                                                  U64 const hashSalt)
+{
+    U32 const newHash = (U32)ZSTD_hashPtrSalted(base+idx+ZSTD_ROW_HASH_CACHE_SIZE, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, hashSalt);
+    U32 const row = (newHash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
+    ZSTD_row_prefetch(hashTable, tagTable, row, rowLog);
+    {   U32 const hash = cache[idx & ZSTD_ROW_HASH_CACHE_MASK];
+        cache[idx & ZSTD_ROW_HASH_CACHE_MASK] = newHash;
+        return hash;
+    }
+}
+
+/* ZSTD_row_update_internalImpl():
+ * Updates the hash table with positions starting from updateStartIdx until updateEndIdx.
+ */
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_row_update_internalImpl(ZSTD_MatchState_t* ms,
+                                  U32 updateStartIdx, U32 const updateEndIdx,
+                                  U32 const mls, U32 const rowLog,
+                                  U32 const rowMask, U32 const useCache)
+{
+    U32* const hashTable = ms->hashTable;
+    BYTE* const tagTable = ms->tagTable;
+    U32 const hashLog = ms->rowHashLog;
+    const BYTE* const base = ms->window.base;
+
+    DEBUGLOG(6, "ZSTD_row_update_internalImpl(): updateStartIdx=%u, updateEndIdx=%u", updateStartIdx, updateEndIdx);
+    for (; updateStartIdx < updateEndIdx; ++updateStartIdx) {
+        U32 const hash = useCache ? ZSTD_row_nextCachedHash(ms->hashCache, hashTable, tagTable, base, updateStartIdx, hashLog, rowLog, mls, ms->hashSalt)
+                                  : (U32)ZSTD_hashPtrSalted(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt);
+        U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
+        U32* const row = hashTable + relRow;
+        BYTE* tagRow = tagTable + relRow;
+        U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask);
+
+        assert(hash == ZSTD_hashPtrSalted(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt));
+        tagRow[pos] = hash & ZSTD_ROW_HASH_TAG_MASK;
+        row[pos] = updateStartIdx;
+    }
+}
+
+/* ZSTD_row_update_internal():
+ * Inserts the byte at ip into the appropriate position in the hash table, and updates ms->nextToUpdate.
+ * Skips sections of long matches as is necessary.
+ */
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_row_update_internal(ZSTD_MatchState_t* ms, const BYTE* ip,
+                              U32 const mls, U32 const rowLog,
+                              U32 const rowMask, U32 const useCache)
+{
+    U32 idx = ms->nextToUpdate;
+    const BYTE* const base = ms->window.base;
+    const U32 target = (U32)(ip - base);
+    const U32 kSkipThreshold = 384;
+    const U32 kMaxMatchStartPositionsToUpdate = 96;
+    const U32 kMaxMatchEndPositionsToUpdate = 32;
+
+    if (useCache) {
+        /* Only skip positions when using hash cache, i.e.
+         * if we are loading a dict, don't skip anything.
+         * If we decide to skip, then we only update a set number
+         * of positions at the beginning and end of the match.
+         */
+        if (UNLIKELY(target - idx > kSkipThreshold)) {
+            U32 const bound = idx + kMaxMatchStartPositionsToUpdate;
+            ZSTD_row_update_internalImpl(ms, idx, bound, mls, rowLog, rowMask, useCache);
+            idx = target - kMaxMatchEndPositionsToUpdate;
+            ZSTD_row_fillHashCache(ms, base, rowLog, mls, idx, ip+1);
+        }
+    }
+    assert(target >= idx);
+    ZSTD_row_update_internalImpl(ms, idx, target, mls, rowLog, rowMask, useCache);
+    ms->nextToUpdate = target;
+}
+
+/* ZSTD_row_update():
+ * External wrapper for ZSTD_row_update_internal(). Used for filling the hashtable during dictionary
+ * processing.
+ */
+void ZSTD_row_update(ZSTD_MatchState_t* const ms, const BYTE* ip) {
+    const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
+    const U32 rowMask = (1u << rowLog) - 1;
+    const U32 mls = MIN(ms->cParams.minMatch, 6 /* mls caps out at 6 */);
+
+    DEBUGLOG(5, "ZSTD_row_update(), rowLog=%u", rowLog);
+    ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 0 /* don't use cache */);
+}
+
+/* Returns the mask width of bits group of which will be set to 1. Given not all
+ * architectures have easy movemask instruction, this helps to iterate over
+ * groups of bits easier and faster.
+ */
+FORCE_INLINE_TEMPLATE U32
+ZSTD_row_matchMaskGroupWidth(const U32 rowEntries)
+{
+    assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
+    assert(rowEntries <= ZSTD_ROW_HASH_MAX_ENTRIES);
+    (void)rowEntries;
+#if defined(ZSTD_ARCH_ARM_NEON)
+    /* NEON path only works for little endian */
+    if (!MEM_isLittleEndian()) {
+        return 1;
+    }
+    if (rowEntries == 16) {
+        return 4;
+    }
+    if (rowEntries == 32) {
+        return 2;
+    }
+    if (rowEntries == 64) {
+        return 1;
+    }
+#endif
+    return 1;
+}
+
+#if defined(ZSTD_ARCH_X86_SSE2)
+FORCE_INLINE_TEMPLATE ZSTD_VecMask
+ZSTD_row_getSSEMask(int nbChunks, const BYTE* const src, const BYTE tag, const U32 head)
+{
+    const __m128i comparisonMask = _mm_set1_epi8((char)tag);
+    int matches[4] = {0};
+    int i;
+    assert(nbChunks == 1 || nbChunks == 2 || nbChunks == 4);
+    for (i=0; i<nbChunks; i++) {
+        const __m128i chunk = _mm_loadu_si128((const __m128i*)(const void*)(src + 16*i));
+        const __m128i equalMask = _mm_cmpeq_epi8(chunk, comparisonMask);
+        matches[i] = _mm_movemask_epi8(equalMask);
+    }
+    if (nbChunks == 1) return ZSTD_rotateRight_U16((U16)matches[0], head);
+    if (nbChunks == 2) return ZSTD_rotateRight_U32((U32)matches[1] << 16 | (U32)matches[0], head);
+    assert(nbChunks == 4);
+    return ZSTD_rotateRight_U64((U64)matches[3] << 48 | (U64)matches[2] << 32 | (U64)matches[1] << 16 | (U64)matches[0], head);
+}
+#endif
+
+#if defined(ZSTD_ARCH_ARM_NEON)
+FORCE_INLINE_TEMPLATE ZSTD_VecMask
+ZSTD_row_getNEONMask(const U32 rowEntries, const BYTE* const src, const BYTE tag, const U32 headGrouped)
+{
+    assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
+    if (rowEntries == 16) {
+        /* vshrn_n_u16 shifts by 4 every u16 and narrows to 8 lower bits.
+         * After that groups of 4 bits represent the equalMask. We lower
+         * all bits except the highest in these groups by doing AND with
+         * 0x88 = 0b10001000.
+         */
+        const uint8x16_t chunk = vld1q_u8(src);
+        const uint16x8_t equalMask = vreinterpretq_u16_u8(vceqq_u8(chunk, vdupq_n_u8(tag)));
+        const uint8x8_t res = vshrn_n_u16(equalMask, 4);
+        const U64 matches = vget_lane_u64(vreinterpret_u64_u8(res), 0);
+        return ZSTD_rotateRight_U64(matches, headGrouped) & 0x8888888888888888ull;
+    } else if (rowEntries == 32) {
+        /* Same idea as with rowEntries == 16 but doing AND with
+         * 0x55 = 0b01010101.
+         */
+        const uint16x8x2_t chunk = vld2q_u16((const uint16_t*)(const void*)src);
+        const uint8x16_t chunk0 = vreinterpretq_u8_u16(chunk.val[0]);
+        const uint8x16_t chunk1 = vreinterpretq_u8_u16(chunk.val[1]);
+        const uint8x16_t dup = vdupq_n_u8(tag);
+        const uint8x8_t t0 = vshrn_n_u16(vreinterpretq_u16_u8(vceqq_u8(chunk0, dup)), 6);
+        const uint8x8_t t1 = vshrn_n_u16(vreinterpretq_u16_u8(vceqq_u8(chunk1, dup)), 6);
+        const uint8x8_t res = vsli_n_u8(t0, t1, 4);
+        const U64 matches = vget_lane_u64(vreinterpret_u64_u8(res), 0) ;
+        return ZSTD_rotateRight_U64(matches, headGrouped) & 0x5555555555555555ull;
+    } else { /* rowEntries == 64 */
+        const uint8x16x4_t chunk = vld4q_u8(src);
+        const uint8x16_t dup = vdupq_n_u8(tag);
+        const uint8x16_t cmp0 = vceqq_u8(chunk.val[0], dup);
+        const uint8x16_t cmp1 = vceqq_u8(chunk.val[1], dup);
+        const uint8x16_t cmp2 = vceqq_u8(chunk.val[2], dup);
+        const uint8x16_t cmp3 = vceqq_u8(chunk.val[3], dup);
+
+        const uint8x16_t t0 = vsriq_n_u8(cmp1, cmp0, 1);
+        const uint8x16_t t1 = vsriq_n_u8(cmp3, cmp2, 1);
+        const uint8x16_t t2 = vsriq_n_u8(t1, t0, 2);
+        const uint8x16_t t3 = vsriq_n_u8(t2, t2, 4);
+        const uint8x8_t t4 = vshrn_n_u16(vreinterpretq_u16_u8(t3), 4);
+        const U64 matches = vget_lane_u64(vreinterpret_u64_u8(t4), 0);
+        return ZSTD_rotateRight_U64(matches, headGrouped);
+    }
+}
+#endif
+
+/* Returns a ZSTD_VecMask (U64) that has the nth group (determined by
+ * ZSTD_row_matchMaskGroupWidth) of bits set to 1 if the newly-computed "tag"
+ * matches the hash at the nth position in a row of the tagTable.
+ * Each row is a circular buffer beginning at the value of "headGrouped". So we
+ * must rotate the "matches" bitfield to match up with the actual layout of the
+ * entries within the hashTable */
+FORCE_INLINE_TEMPLATE ZSTD_VecMask
+ZSTD_row_getMatchMask(const BYTE* const tagRow, const BYTE tag, const U32 headGrouped, const U32 rowEntries)
+{
+    const BYTE* const src = tagRow;
+    assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
+    assert(rowEntries <= ZSTD_ROW_HASH_MAX_ENTRIES);
+    assert(ZSTD_row_matchMaskGroupWidth(rowEntries) * rowEntries <= sizeof(ZSTD_VecMask) * 8);
+
+#if defined(ZSTD_ARCH_X86_SSE2)
+
+    return ZSTD_row_getSSEMask(rowEntries / 16, src, tag, headGrouped);
+
+#else /* SW or NEON-LE */
+
+# if defined(ZSTD_ARCH_ARM_NEON)
+  /* This NEON path only works for little endian - otherwise use SWAR below */
+    if (MEM_isLittleEndian()) {
+        return ZSTD_row_getNEONMask(rowEntries, src, tag, headGrouped);
+    }
+# endif /* ZSTD_ARCH_ARM_NEON */
+    /* SWAR */
+    {   const int chunkSize = sizeof(size_t);
+        const size_t shiftAmount = ((chunkSize * 8) - chunkSize);
+        const size_t xFF = ~((size_t)0);
+        const size_t x01 = xFF / 0xFF;
+        const size_t x80 = x01 << 7;
+        const size_t splatChar = tag * x01;
+        ZSTD_VecMask matches = 0;
+        int i = rowEntries - chunkSize;
+        assert((sizeof(size_t) == 4) || (sizeof(size_t) == 8));
+        if (MEM_isLittleEndian()) { /* runtime check so have two loops */
+            const size_t extractMagic = (xFF / 0x7F) >> chunkSize;
+            do {
+                size_t chunk = MEM_readST(&src[i]);
+                chunk ^= splatChar;
+                chunk = (((chunk | x80) - x01) | chunk) & x80;
+                matches <<= chunkSize;
+                matches |= (chunk * extractMagic) >> shiftAmount;
+                i -= chunkSize;
+            } while (i >= 0);
+        } else { /* big endian: reverse bits during extraction */
+            const size_t msb = xFF ^ (xFF >> 1);
+            const size_t extractMagic = (msb / 0x1FF) | msb;
+            do {
+                size_t chunk = MEM_readST(&src[i]);
+                chunk ^= splatChar;
+                chunk = (((chunk | x80) - x01) | chunk) & x80;
+                matches <<= chunkSize;
+                matches |= ((chunk >> 7) * extractMagic) >> shiftAmount;
+                i -= chunkSize;
+            } while (i >= 0);
+        }
+        matches = ~matches;
+        if (rowEntries == 16) {
+            return ZSTD_rotateRight_U16((U16)matches, headGrouped);
+        } else if (rowEntries == 32) {
+            return ZSTD_rotateRight_U32((U32)matches, headGrouped);
+        } else {
+            return ZSTD_rotateRight_U64((U64)matches, headGrouped);
+        }
+    }
+#endif
+}
+
+/* The high-level approach of the SIMD row based match finder is as follows:
+ * - Figure out where to insert the new entry:
+ *      - Generate a hash for current input position and split it into a one byte of tag and `rowHashLog` bits of index.
+ *           - The hash is salted by a value that changes on every context reset, so when the same table is used
+ *             we will avoid collisions that would otherwise slow us down by introducing phantom matches.
+ *      - The hashTable is effectively split into groups or "rows" of 15 or 31 entries of U32, and the index determines
+ *        which row to insert into.
+ *      - Determine the correct position within the row to insert the entry into. Each row of 15 or 31 can
+ *        be considered as a circular buffer with a "head" index that resides in the tagTable (overall 16 or 32 bytes
+ *        per row).
+ * - Use SIMD to efficiently compare the tags in the tagTable to the 1-byte tag calculated for the position and
+ *   generate a bitfield that we can cycle through to check the collisions in the hash table.
+ * - Pick the longest match.
+ * - Insert the tag into the equivalent row and position in the tagTable.
+ */
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_RowFindBestMatch(
+                        ZSTD_MatchState_t* ms,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 mls, const ZSTD_dictMode_e dictMode,
+                        const U32 rowLog)
+{
+    U32* const hashTable = ms->hashTable;
+    BYTE* const tagTable = ms->tagTable;
+    U32* const hashCache = ms->hashCache;
+    const U32 hashLog = ms->rowHashLog;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const U32 curr = (U32)(ip-base);
+    const U32 maxDistance = 1U << cParams->windowLog;
+    const U32 lowestValid = ms->window.lowLimit;
+    const U32 withinMaxDistance = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
+    const U32 isDictionary = (ms->loadedDictEnd != 0);
+    const U32 lowLimit = isDictionary ? lowestValid : withinMaxDistance;
+    const U32 rowEntries = (1U << rowLog);
+    const U32 rowMask = rowEntries - 1;
+    const U32 cappedSearchLog = MIN(cParams->searchLog, rowLog); /* nb of searches is capped at nb entries per row */
+    const U32 groupWidth = ZSTD_row_matchMaskGroupWidth(rowEntries);
+    const U64 hashSalt = ms->hashSalt;
+    U32 nbAttempts = 1U << cappedSearchLog;
+    size_t ml=4-1;
+    U32 hash;
+
+    /* DMS/DDS variables that may be referenced laster */
+    const ZSTD_MatchState_t* const dms = ms->dictMatchState;
+
+    /* Initialize the following variables to satisfy static analyzer */
+    size_t ddsIdx = 0;
+    U32 ddsExtraAttempts = 0; /* cctx hash tables are limited in searches, but allow extra searches into DDS */
+    U32 dmsTag = 0;
+    U32* dmsRow = NULL;
+    BYTE* dmsTagRow = NULL;
+
+    if (dictMode == ZSTD_dedicatedDictSearch) {
+        const U32 ddsHashLog = dms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
+        {   /* Prefetch DDS hashtable entry */
+            ddsIdx = ZSTD_hashPtr(ip, ddsHashLog, mls) << ZSTD_LAZY_DDSS_BUCKET_LOG;
+            PREFETCH_L1(&dms->hashTable[ddsIdx]);
+        }
+        ddsExtraAttempts = cParams->searchLog > rowLog ? 1U << (cParams->searchLog - rowLog) : 0;
+    }
+
+    if (dictMode == ZSTD_dictMatchState) {
+        /* Prefetch DMS rows */
+        U32* const dmsHashTable = dms->hashTable;
+        BYTE* const dmsTagTable = dms->tagTable;
+        U32 const dmsHash = (U32)ZSTD_hashPtr(ip, dms->rowHashLog + ZSTD_ROW_HASH_TAG_BITS, mls);
+        U32 const dmsRelRow = (dmsHash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
+        dmsTag = dmsHash & ZSTD_ROW_HASH_TAG_MASK;
+        dmsTagRow = (BYTE*)(dmsTagTable + dmsRelRow);
+        dmsRow = dmsHashTable + dmsRelRow;
+        ZSTD_row_prefetch(dmsHashTable, dmsTagTable, dmsRelRow, rowLog);
+    }
+
+    /* Update the hashTable and tagTable up to (but not including) ip */
+    if (!ms->lazySkipping) {
+        ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 1 /* useCache */);
+        hash = ZSTD_row_nextCachedHash(hashCache, hashTable, tagTable, base, curr, hashLog, rowLog, mls, hashSalt);
+    } else {
+        /* Stop inserting every position when in the lazy skipping mode.
+         * The hash cache is also not kept up to date in this mode.
+         */
+        hash = (U32)ZSTD_hashPtrSalted(ip, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, hashSalt);
+        ms->nextToUpdate = curr;
+    }
+    ms->hashSaltEntropy += hash; /* collect salt entropy */
+
+    {   /* Get the hash for ip, compute the appropriate row */
+        U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
+        U32 const tag = hash & ZSTD_ROW_HASH_TAG_MASK;
+        U32* const row = hashTable + relRow;
+        BYTE* tagRow = (BYTE*)(tagTable + relRow);
+        U32 const headGrouped = (*tagRow & rowMask) * groupWidth;
+        U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
+        size_t numMatches = 0;
+        size_t currMatch = 0;
+        ZSTD_VecMask matches = ZSTD_row_getMatchMask(tagRow, (BYTE)tag, headGrouped, rowEntries);
+
+        /* Cycle through the matches and prefetch */
+        for (; (matches > 0) && (nbAttempts > 0); matches &= (matches - 1)) {
+            U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask;
+            U32 const matchIndex = row[matchPos];
+            if(matchPos == 0) continue;
+            assert(numMatches < rowEntries);
+            if (matchIndex < lowLimit)
+                break;
+            if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
+                PREFETCH_L1(base + matchIndex);
+            } else {
+                PREFETCH_L1(dictBase + matchIndex);
+            }
+            matchBuffer[numMatches++] = matchIndex;
+            --nbAttempts;
+        }
+
+        /* Speed opt: insert current byte into hashtable too. This allows us to avoid one iteration of the loop
+           in ZSTD_row_update_internal() at the next search. */
+        {
+            U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask);
+            tagRow[pos] = (BYTE)tag;
+            row[pos] = ms->nextToUpdate++;
+        }
+
+        /* Return the longest match */
+        for (; currMatch < numMatches; ++currMatch) {
+            U32 const matchIndex = matchBuffer[currMatch];
+            size_t currentMl=0;
+            assert(matchIndex < curr);
+            assert(matchIndex >= lowLimit);
+
+            if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
+                const BYTE* const match = base + matchIndex;
+                assert(matchIndex >= dictLimit);   /* ensures this is true if dictMode != ZSTD_extDict */
+                /* read 4B starting from (match + ml + 1 - sizeof(U32)) */
+                if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3))   /* potentially better */
+                    currentMl = ZSTD_count(ip, match, iLimit);
+            } else {
+                const BYTE* const match = dictBase + matchIndex;
+                assert(match+4 <= dictEnd);
+                if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                    currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4;
+            }
+
+            /* Save best solution */
+            if (currentMl > ml) {
+                ml = currentMl;
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
+                if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+            }
+        }
+    }
+
+    assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
+    if (dictMode == ZSTD_dedicatedDictSearch) {
+        ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts + ddsExtraAttempts, dms,
+                                                  ip, iLimit, prefixStart, curr, dictLimit, ddsIdx);
+    } else if (dictMode == ZSTD_dictMatchState) {
+        /* TODO: Measure and potentially add prefetching to DMS */
+        const U32 dmsLowestIndex       = dms->window.dictLimit;
+        const BYTE* const dmsBase      = dms->window.base;
+        const BYTE* const dmsEnd       = dms->window.nextSrc;
+        const U32 dmsSize              = (U32)(dmsEnd - dmsBase);
+        const U32 dmsIndexDelta        = dictLimit - dmsSize;
+
+        {   U32 const headGrouped = (*dmsTagRow & rowMask) * groupWidth;
+            U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
+            size_t numMatches = 0;
+            size_t currMatch = 0;
+            ZSTD_VecMask matches = ZSTD_row_getMatchMask(dmsTagRow, (BYTE)dmsTag, headGrouped, rowEntries);
+
+            for (; (matches > 0) && (nbAttempts > 0); matches &= (matches - 1)) {
+                U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask;
+                U32 const matchIndex = dmsRow[matchPos];
+                if(matchPos == 0) continue;
+                if (matchIndex < dmsLowestIndex)
+                    break;
+                PREFETCH_L1(dmsBase + matchIndex);
+                matchBuffer[numMatches++] = matchIndex;
+                --nbAttempts;
+            }
+
+            /* Return the longest match */
+            for (; currMatch < numMatches; ++currMatch) {
+                U32 const matchIndex = matchBuffer[currMatch];
+                size_t currentMl=0;
+                assert(matchIndex >= dmsLowestIndex);
+                assert(matchIndex < curr);
+
+                {   const BYTE* const match = dmsBase + matchIndex;
+                    assert(match+4 <= dmsEnd);
+                    if (MEM_read32(match) == MEM_read32(ip))
+                        currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dmsEnd, prefixStart) + 4;
+                }
+
+                if (currentMl > ml) {
+                    ml = currentMl;
+                    assert(curr > matchIndex + dmsIndexDelta);
+                    *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta));
+                    if (ip+currentMl == iLimit) break;
+                }
+            }
+        }
+    }
+    return ml;
+}
+
+
+/**
+ * Generate search functions templated on (dictMode, mls, rowLog).
+ * These functions are outlined for code size & compilation time.
+ * ZSTD_searchMax() dispatches to the correct implementation function.
+ *
+ * TODO: The start of the search function involves loading and calculating a
+ * bunch of constants from the ZSTD_MatchState_t. These computations could be
+ * done in an initialization function, and saved somewhere in the match state.
+ * Then we could pass a pointer to the saved state instead of the match state,
+ * and avoid duplicate computations.
+ *
+ * TODO: Move the match re-winding into searchMax. This improves compression
+ * ratio, and unlocks further simplifications with the next TODO.
+ *
+ * TODO: Try moving the repcode search into searchMax. After the re-winding
+ * and repcode search are in searchMax, there is no more logic in the match
+ * finder loop that requires knowledge about the dictMode. So we should be
+ * able to avoid force inlining it, and we can join the extDict loop with
+ * the single segment loop. It should go in searchMax instead of its own
+ * function to avoid having multiple virtual function calls per search.
+ */
+
+#define ZSTD_BT_SEARCH_FN(dictMode, mls) ZSTD_BtFindBestMatch_##dictMode##_##mls
+#define ZSTD_HC_SEARCH_FN(dictMode, mls) ZSTD_HcFindBestMatch_##dictMode##_##mls
+#define ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog) ZSTD_RowFindBestMatch_##dictMode##_##mls##_##rowLog
+
+#define ZSTD_SEARCH_FN_ATTRS FORCE_NOINLINE
+
+#define GEN_ZSTD_BT_SEARCH_FN(dictMode, mls)                                           \
+    ZSTD_SEARCH_FN_ATTRS size_t ZSTD_BT_SEARCH_FN(dictMode, mls)(                      \
+            ZSTD_MatchState_t* ms,                                                     \
+            const BYTE* ip, const BYTE* const iLimit,                                  \
+            size_t* offBasePtr)                                                        \
+    {                                                                                  \
+        assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                           \
+        return ZSTD_BtFindBestMatch(ms, ip, iLimit, offBasePtr, mls, ZSTD_##dictMode); \
+    }                                                                                  \
+
+#define GEN_ZSTD_HC_SEARCH_FN(dictMode, mls)                                          \
+    ZSTD_SEARCH_FN_ATTRS size_t ZSTD_HC_SEARCH_FN(dictMode, mls)(                     \
+            ZSTD_MatchState_t* ms,                                                    \
+            const BYTE* ip, const BYTE* const iLimit,                                 \
+            size_t* offsetPtr)                                                        \
+    {                                                                                 \
+        assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                          \
+        return ZSTD_HcFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode); \
+    }                                                                                 \
+
+#define GEN_ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)                                          \
+    ZSTD_SEARCH_FN_ATTRS size_t ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)(                     \
+            ZSTD_MatchState_t* ms,                                                             \
+            const BYTE* ip, const BYTE* const iLimit,                                          \
+            size_t* offsetPtr)                                                                 \
+    {                                                                                          \
+        assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                                   \
+        assert(MAX(4, MIN(6, ms->cParams.searchLog)) == rowLog);                               \
+        return ZSTD_RowFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode, rowLog); \
+    }                                                                                          \
+
+#define ZSTD_FOR_EACH_ROWLOG(X, dictMode, mls) \
+    X(dictMode, mls, 4)                        \
+    X(dictMode, mls, 5)                        \
+    X(dictMode, mls, 6)
+
+#define ZSTD_FOR_EACH_MLS_ROWLOG(X, dictMode) \
+    ZSTD_FOR_EACH_ROWLOG(X, dictMode, 4)      \
+    ZSTD_FOR_EACH_ROWLOG(X, dictMode, 5)      \
+    ZSTD_FOR_EACH_ROWLOG(X, dictMode, 6)
+
+#define ZSTD_FOR_EACH_MLS(X, dictMode) \
+    X(dictMode, 4)                     \
+    X(dictMode, 5)                     \
+    X(dictMode, 6)
+
+#define ZSTD_FOR_EACH_DICT_MODE(X, ...) \
+    X(__VA_ARGS__, noDict)              \
+    X(__VA_ARGS__, extDict)             \
+    X(__VA_ARGS__, dictMatchState)      \
+    X(__VA_ARGS__, dedicatedDictSearch)
+
+/* Generate row search fns for each combination of (dictMode, mls, rowLog) */
+ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS_ROWLOG, GEN_ZSTD_ROW_SEARCH_FN)
+/* Generate binary Tree search fns for each combination of (dictMode, mls) */
+ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_BT_SEARCH_FN)
+/* Generate hash chain search fns for each combination of (dictMode, mls) */
+ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_HC_SEARCH_FN)
+
+typedef enum { search_hashChain=0, search_binaryTree=1, search_rowHash=2 } searchMethod_e;
+
+#define GEN_ZSTD_CALL_BT_SEARCH_FN(dictMode, mls)                         \
+    case mls:                                                             \
+        return ZSTD_BT_SEARCH_FN(dictMode, mls)(ms, ip, iend, offsetPtr);
+#define GEN_ZSTD_CALL_HC_SEARCH_FN(dictMode, mls)                         \
+    case mls:                                                             \
+        return ZSTD_HC_SEARCH_FN(dictMode, mls)(ms, ip, iend, offsetPtr);
+#define GEN_ZSTD_CALL_ROW_SEARCH_FN(dictMode, mls, rowLog)                         \
+    case rowLog:                                                                   \
+        return ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)(ms, ip, iend, offsetPtr);
+
+#define ZSTD_SWITCH_MLS(X, dictMode)   \
+    switch (mls) {                     \
+        ZSTD_FOR_EACH_MLS(X, dictMode) \
+    }
+
+#define ZSTD_SWITCH_ROWLOG(dictMode, mls)                                    \
+    case mls:                                                                \
+        switch (rowLog) {                                                    \
+            ZSTD_FOR_EACH_ROWLOG(GEN_ZSTD_CALL_ROW_SEARCH_FN, dictMode, mls) \
+        }                                                                    \
+        ZSTD_UNREACHABLE;                                                    \
+        break;
+
+#define ZSTD_SWITCH_SEARCH_METHOD(dictMode)                       \
+    switch (searchMethod) {                                       \
+        case search_hashChain:                                    \
+            ZSTD_SWITCH_MLS(GEN_ZSTD_CALL_HC_SEARCH_FN, dictMode) \
+            break;                                                \
+        case search_binaryTree:                                   \
+            ZSTD_SWITCH_MLS(GEN_ZSTD_CALL_BT_SEARCH_FN, dictMode) \
+            break;                                                \
+        case search_rowHash:                                      \
+            ZSTD_SWITCH_MLS(ZSTD_SWITCH_ROWLOG, dictMode)         \
+            break;                                                \
+    }                                                             \
+    ZSTD_UNREACHABLE;
+
+/**
+ * Searches for the longest match at @p ip.
+ * Dispatches to the correct implementation function based on the
+ * (searchMethod, dictMode, mls, rowLog). We use switch statements
+ * here instead of using an indirect function call through a function
+ * pointer because after Spectre and Meltdown mitigations, indirect
+ * function calls can be very costly, especially in the kernel.
+ *
+ * NOTE: dictMode and searchMethod should be templated, so those switch
+ * statements should be optimized out. Only the mls & rowLog switches
+ * should be left.
+ *
+ * @param ms The match state.
+ * @param ip The position to search at.
+ * @param iend The end of the input data.
+ * @param[out] offsetPtr Stores the match offset into this pointer.
+ * @param mls The minimum search length, in the range [4, 6].
+ * @param rowLog The row log (if applicable), in the range [4, 6].
+ * @param searchMethod The search method to use (templated).
+ * @param dictMode The dictMode (templated).
+ *
+ * @returns The length of the longest match found, or < mls if no match is found.
+ * If a match is found its offset is stored in @p offsetPtr.
+ */
+FORCE_INLINE_TEMPLATE size_t ZSTD_searchMax(
+    ZSTD_MatchState_t* ms,
+    const BYTE* ip,
+    const BYTE* iend,
+    size_t* offsetPtr,
+    U32 const mls,
+    U32 const rowLog,
+    searchMethod_e const searchMethod,
+    ZSTD_dictMode_e const dictMode)
+{
+    if (dictMode == ZSTD_noDict) {
+        ZSTD_SWITCH_SEARCH_METHOD(noDict)
+    } else if (dictMode == ZSTD_extDict) {
+        ZSTD_SWITCH_SEARCH_METHOD(extDict)
+    } else if (dictMode == ZSTD_dictMatchState) {
+        ZSTD_SWITCH_SEARCH_METHOD(dictMatchState)
+    } else if (dictMode == ZSTD_dedicatedDictSearch) {
+        ZSTD_SWITCH_SEARCH_METHOD(dedicatedDictSearch)
+    }
+    ZSTD_UNREACHABLE;
+    return 0;
+}
+
+/* *******************************
+*  Common parser - lazy strategy
+*********************************/
+
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_compressBlock_lazy_generic(
+                        ZSTD_MatchState_t* ms, SeqStore_t* seqStore,
+                        U32 rep[ZSTD_REP_NUM],
+                        const void* src, size_t srcSize,
+                        const searchMethod_e searchMethod, const U32 depth,
+                        ZSTD_dictMode_e const dictMode)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = (searchMethod == search_rowHash) ? iend - 8 - ZSTD_ROW_HASH_CACHE_SIZE : iend - 8;
+    const BYTE* const base = ms->window.base;
+    const U32 prefixLowestIndex = ms->window.dictLimit;
+    const BYTE* const prefixLowest = base + prefixLowestIndex;
+    const U32 mls = BOUNDED(4, ms->cParams.minMatch, 6);
+    const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
+
+    U32 offset_1 = rep[0], offset_2 = rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
+
+    const int isDMS = dictMode == ZSTD_dictMatchState;
+    const int isDDS = dictMode == ZSTD_dedicatedDictSearch;
+    const int isDxS = isDMS || isDDS;
+    const ZSTD_MatchState_t* const dms = ms->dictMatchState;
+    const U32 dictLowestIndex      = isDxS ? dms->window.dictLimit : 0;
+    const BYTE* const dictBase     = isDxS ? dms->window.base : NULL;
+    const BYTE* const dictLowest   = isDxS ? dictBase + dictLowestIndex : NULL;
+    const BYTE* const dictEnd      = isDxS ? dms->window.nextSrc : NULL;
+    const U32 dictIndexDelta       = isDxS ?
+                                     prefixLowestIndex - (U32)(dictEnd - dictBase) :
+                                     0;
+    const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictLowest));
+
+    DEBUGLOG(5, "ZSTD_compressBlock_lazy_generic (dictMode=%u) (searchFunc=%u)", (U32)dictMode, (U32)searchMethod);
+    ip += (dictAndPrefixLength == 0);
+    if (dictMode == ZSTD_noDict) {
+        U32 const curr = (U32)(ip - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, ms->cParams.windowLog);
+        U32 const maxRep = curr - windowLow;
+        if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
+    }
+    if (isDxS) {
+        /* dictMatchState repCode checks don't currently handle repCode == 0
+         * disabling. */
+        assert(offset_1 <= dictAndPrefixLength);
+        assert(offset_2 <= dictAndPrefixLength);
+    }
+
+    /* Reset the lazy skipping state */
+    ms->lazySkipping = 0;
+
+    if (searchMethod == search_rowHash) {
+        ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
+    }
+
+    /* Match Loop */
+#if defined(__GNUC__) && defined(__x86_64__)
+    /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the
+     * code alignment is perturbed. To fix the instability align the loop on 32-bytes.
+     */
+    __asm__(".p2align 5");
+#endif
+    while (ip < ilimit) {
+        size_t matchLength=0;
+        size_t offBase = REPCODE1_TO_OFFBASE;
+        const BYTE* start=ip+1;
+        DEBUGLOG(7, "search baseline (depth 0)");
+
+        /* check repCode */
+        if (isDxS) {
+            const U32 repIndex = (U32)(ip - base) + 1 - offset_1;
+            const BYTE* repMatch = ((dictMode == ZSTD_dictMatchState || dictMode == ZSTD_dedicatedDictSearch)
+                                && repIndex < prefixLowestIndex) ?
+                                   dictBase + (repIndex - dictIndexDelta) :
+                                   base + repIndex;
+            if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex))
+                && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+                const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                if (depth==0) goto _storeSequence;
+            }
+        }
+        if ( dictMode == ZSTD_noDict
+          && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
+            matchLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+            if (depth==0) goto _storeSequence;
+        }
+
+        /* first search (depth 0) */
+        {   size_t offbaseFound = 999999999;
+            size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offbaseFound, mls, rowLog, searchMethod, dictMode);
+            if (ml2 > matchLength)
+                matchLength = ml2, start = ip, offBase = offbaseFound;
+        }
+
+        if (matchLength < 4) {
+            size_t const step = ((size_t)(ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */;
+            ip += step;
+            /* Enter the lazy skipping mode once we are skipping more than 8 bytes at a time.
+             * In this mode we stop inserting every position into our tables, and only insert
+             * positions that we search, which is one in step positions.
+             * The exact cutoff is flexible, I've just chosen a number that is reasonably high,
+             * so we minimize the compression ratio loss in "normal" scenarios. This mode gets
+             * triggered once we've gone 2KB without finding any matches.
+             */
+            ms->lazySkipping = step > kLazySkippingStep;
+            continue;
+        }
+
+        /* let's try to find a better solution */
+        if (depth>=1)
+        while (ip<ilimit) {
+            DEBUGLOG(7, "search depth 1");
+            ip ++;
+            if ( (dictMode == ZSTD_noDict)
+              && (offBase) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                int const gain2 = (int)(mlRep * 3);
+                int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
+                if ((mlRep >= 4) && (gain2 > gain1))
+                    matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
+            }
+            if (isDxS) {
+                const U32 repIndex = (U32)(ip - base) - offset_1;
+                const BYTE* repMatch = repIndex < prefixLowestIndex ?
+                               dictBase + (repIndex - dictIndexDelta) :
+                               base + repIndex;
+                if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex))
+                    && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                    const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                    size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                    int const gain2 = (int)(mlRep * 3);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
+                    if ((mlRep >= 4) && (gain2 > gain1))
+                        matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
+                }
+            }
+            {   size_t ofbCandidate=999999999;
+                size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4);
+                if ((ml2 >= 4) && (gain2 > gain1)) {
+                    matchLength = ml2, offBase = ofbCandidate, start = ip;
+                    continue;   /* search a better one */
+            }   }
+
+            /* let's find an even better one */
+            if ((depth==2) && (ip<ilimit)) {
+                DEBUGLOG(7, "search depth 2");
+                ip ++;
+                if ( (dictMode == ZSTD_noDict)
+                  && (offBase) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                    size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                    int const gain2 = (int)(mlRep * 4);
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
+                    if ((mlRep >= 4) && (gain2 > gain1))
+                        matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
+                }
+                if (isDxS) {
+                    const U32 repIndex = (U32)(ip - base) - offset_1;
+                    const BYTE* repMatch = repIndex < prefixLowestIndex ?
+                                   dictBase + (repIndex - dictIndexDelta) :
+                                   base + repIndex;
+                    if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex))
+                        && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                        const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                        size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                        int const gain2 = (int)(mlRep * 4);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
+                        if ((mlRep >= 4) && (gain2 > gain1))
+                            matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
+                    }
+                }
+                {   size_t ofbCandidate=999999999;
+                    size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7);
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
+                        matchLength = ml2, offBase = ofbCandidate, start = ip;
+                        continue;
+            }   }   }
+            break;  /* nothing found : store previous solution */
+        }
+
+        /* NOTE:
+         * Pay attention that `start[-value]` can lead to strange undefined behavior
+         * notably if `value` is unsigned, resulting in a large positive `-value`.
+         */
+        /* catch up */
+        if (OFFBASE_IS_OFFSET(offBase)) {
+            if (dictMode == ZSTD_noDict) {
+                while ( ((start > anchor) & (start - OFFBASE_TO_OFFSET(offBase) > prefixLowest))
+                     && (start[-1] == (start-OFFBASE_TO_OFFSET(offBase))[-1]) )  /* only search for offset within prefix */
+                    { start--; matchLength++; }
+            }
+            if (isDxS) {
+                U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase));
+                const BYTE* match = (matchIndex < prefixLowestIndex) ? dictBase + matchIndex - dictIndexDelta : base + matchIndex;
+                const BYTE* const mStart = (matchIndex < prefixLowestIndex) ? dictLowest : prefixLowest;
+                while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
+            }
+            offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase);
+        }
+        /* store sequence */
+_storeSequence:
+        {   size_t const litLength = (size_t)(start - anchor);
+            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength);
+            anchor = ip = start + matchLength;
+        }
+        if (ms->lazySkipping) {
+            /* We've found a match, disable lazy skipping mode, and refill the hash cache. */
+            if (searchMethod == search_rowHash) {
+                ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
+            }
+            ms->lazySkipping = 0;
+        }
+
+        /* check immediate repcode */
+        if (isDxS) {
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex = current2 - offset_2;
+                const BYTE* repMatch = repIndex < prefixLowestIndex ?
+                        dictBase - dictIndexDelta + repIndex :
+                        base + repIndex;
+                if ( (ZSTD_index_overlap_check(prefixLowestIndex, repIndex))
+                   && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex < prefixLowestIndex ? dictEnd : iend;
+                    matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd2, prefixLowest) + 4;
+                    offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
+                    ip += matchLength;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+            }
+        }
+
+        if (dictMode == ZSTD_noDict) {
+            while ( ((ip <= ilimit) & (offset_2>0))
+                 && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) {
+                /* store sequence */
+                matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase; /* swap repcodes */
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
+                ip += matchLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+    }   }   }
+
+    /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
+     * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
+    offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
+
+    /* save reps for next block */
+    rep[0] = offset_1 ? offset_1 : offsetSaved1;
+    rep[1] = offset_2 ? offset_2 : offsetSaved2;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+#endif /* build exclusions */
+
+
+#ifndef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_greedy(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_greedy_dictMatchState(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dedicatedDictSearch);
+}
+
+size_t ZSTD_compressBlock_greedy_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_greedy_dictMatchState_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dedicatedDictSearch);
+}
+#endif
+
+#ifndef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_lazy(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_lazy_dictMatchState(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dedicatedDictSearch);
+}
+
+size_t ZSTD_compressBlock_lazy_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_lazy_dictMatchState_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dedicatedDictSearch);
+}
+#endif
+
+#ifndef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_lazy2(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_lazy2_dictMatchState(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dedicatedDictSearch);
+}
+
+size_t ZSTD_compressBlock_lazy2_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_lazy2_dictMatchState_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_dedicatedDictSearch);
+}
+#endif
+
+#ifndef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_btlazy2(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_btlazy2_dictMatchState(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_dictMatchState);
+}
+#endif
+
+#if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR)
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_compressBlock_lazy_extDict_generic(
+                        ZSTD_MatchState_t* ms, SeqStore_t* seqStore,
+                        U32 rep[ZSTD_REP_NUM],
+                        const void* src, size_t srcSize,
+                        const searchMethod_e searchMethod, const U32 depth)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = searchMethod == search_rowHash ? iend - 8 - ZSTD_ROW_HASH_CACHE_SIZE : iend - 8;
+    const BYTE* const base = ms->window.base;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const dictEnd  = dictBase + dictLimit;
+    const BYTE* const dictStart  = dictBase + ms->window.lowLimit;
+    const U32 windowLog = ms->cParams.windowLog;
+    const U32 mls = BOUNDED(4, ms->cParams.minMatch, 6);
+    const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
+
+    U32 offset_1 = rep[0], offset_2 = rep[1];
+
+    DEBUGLOG(5, "ZSTD_compressBlock_lazy_extDict_generic (searchFunc=%u)", (U32)searchMethod);
+
+    /* Reset the lazy skipping state */
+    ms->lazySkipping = 0;
+
+    /* init */
+    ip += (ip == prefixStart);
+    if (searchMethod == search_rowHash) {
+        ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
+    }
+
+    /* Match Loop */
+#if defined(__GNUC__) && defined(__x86_64__)
+    /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the
+     * code alignment is perturbed. To fix the instability align the loop on 32-bytes.
+     */
+    __asm__(".p2align 5");
+#endif
+    while (ip < ilimit) {
+        size_t matchLength=0;
+        size_t offBase = REPCODE1_TO_OFFBASE;
+        const BYTE* start=ip+1;
+        U32 curr = (U32)(ip-base);
+
+        /* check repCode */
+        {   const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr+1, windowLog);
+            const U32 repIndex = (U32)(curr+1 - offset_1);
+            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+            const BYTE* const repMatch = repBase + repIndex;
+            if ( (ZSTD_index_overlap_check(dictLimit, repIndex))
+               & (offset_1 <= curr+1 - windowLow) ) /* note: we are searching at curr+1 */
+            if (MEM_read32(ip+1) == MEM_read32(repMatch)) {
+                /* repcode detected we should take it */
+                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                if (depth==0) goto _storeSequence;
+        }   }
+
+        /* first search (depth 0) */
+        {   size_t ofbCandidate = 999999999;
+            size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
+            if (ml2 > matchLength)
+                matchLength = ml2, start = ip, offBase = ofbCandidate;
+        }
+
+        if (matchLength < 4) {
+            size_t const step = ((size_t)(ip-anchor) >> kSearchStrength);
+            ip += step + 1;   /* jump faster over incompressible sections */
+            /* Enter the lazy skipping mode once we are skipping more than 8 bytes at a time.
+             * In this mode we stop inserting every position into our tables, and only insert
+             * positions that we search, which is one in step positions.
+             * The exact cutoff is flexible, I've just chosen a number that is reasonably high,
+             * so we minimize the compression ratio loss in "normal" scenarios. This mode gets
+             * triggered once we've gone 2KB without finding any matches.
+             */
+            ms->lazySkipping = step > kLazySkippingStep;
+            continue;
+        }
+
+        /* let's try to find a better solution */
+        if (depth>=1)
+        while (ip<ilimit) {
+            ip ++;
+            curr++;
+            /* check repCode */
+            if (offBase) {
+                const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
+                const U32 repIndex = (U32)(curr - offset_1);
+                const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                const BYTE* const repMatch = repBase + repIndex;
+                if ( (ZSTD_index_overlap_check(dictLimit, repIndex))
+                   & (offset_1 <= curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
+                if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                    /* repcode detected */
+                    const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                    size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                    int const gain2 = (int)(repLength * 3);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
+                    if ((repLength >= 4) && (gain2 > gain1))
+                        matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip;
+            }   }
+
+            /* search match, depth 1 */
+            {   size_t ofbCandidate = 999999999;
+                size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4);
+                if ((ml2 >= 4) && (gain2 > gain1)) {
+                    matchLength = ml2, offBase = ofbCandidate, start = ip;
+                    continue;   /* search a better one */
+            }   }
+
+            /* let's find an even better one */
+            if ((depth==2) && (ip<ilimit)) {
+                ip ++;
+                curr++;
+                /* check repCode */
+                if (offBase) {
+                    const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
+                    const U32 repIndex = (U32)(curr - offset_1);
+                    const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                    const BYTE* const repMatch = repBase + repIndex;
+                    if ( (ZSTD_index_overlap_check(dictLimit, repIndex))
+                       & (offset_1 <= curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
+                    if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                        /* repcode detected */
+                        const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                        size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                        int const gain2 = (int)(repLength * 4);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
+                        if ((repLength >= 4) && (gain2 > gain1))
+                            matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip;
+                }   }
+
+                /* search match, depth 2 */
+                {   size_t ofbCandidate = 999999999;
+                    size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7);
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
+                        matchLength = ml2, offBase = ofbCandidate, start = ip;
+                        continue;
+            }   }   }
+            break;  /* nothing found : store previous solution */
+        }
+
+        /* catch up */
+        if (OFFBASE_IS_OFFSET(offBase)) {
+            U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase));
+            const BYTE* match = (matchIndex < dictLimit) ? dictBase + matchIndex : base + matchIndex;
+            const BYTE* const mStart = (matchIndex < dictLimit) ? dictStart : prefixStart;
+            while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
+            offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase);
+        }
+
+        /* store sequence */
+_storeSequence:
+        {   size_t const litLength = (size_t)(start - anchor);
+            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength);
+            anchor = ip = start + matchLength;
+        }
+        if (ms->lazySkipping) {
+            /* We've found a match, disable lazy skipping mode, and refill the hash cache. */
+            if (searchMethod == search_rowHash) {
+                ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
+            }
+            ms->lazySkipping = 0;
+        }
+
+        /* check immediate repcode */
+        while (ip <= ilimit) {
+            const U32 repCurrent = (U32)(ip-base);
+            const U32 windowLow = ZSTD_getLowestMatchIndex(ms, repCurrent, windowLog);
+            const U32 repIndex = repCurrent - offset_2;
+            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+            const BYTE* const repMatch = repBase + repIndex;
+            if ( (ZSTD_index_overlap_check(dictLimit, repIndex))
+               & (offset_2 <= repCurrent - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
+            if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                /* repcode detected we should take it */
+                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase;   /* swap offset history */
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
+                ip += matchLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+            }
+            break;
+    }   }
+
+    /* Save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+#endif /* build exclusions */
+
+#ifndef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_greedy_extDict(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0);
+}
+
+size_t ZSTD_compressBlock_greedy_extDict_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0);
+}
+#endif
+
+#ifndef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_lazy_extDict(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1);
+}
+
+size_t ZSTD_compressBlock_lazy_extDict_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1);
+}
+#endif
+
+#ifndef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_lazy2_extDict(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2);
+}
+
+size_t ZSTD_compressBlock_lazy2_extDict_row(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2);
+}
+#endif
+
+#ifndef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_btlazy2_extDict(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2);
+}
+#endif
diff --git a/lib/compress/zstd_ldm.c b/lib/compress/zstd_ldm.c
new file mode 100644
index 0000000..070551c
--- /dev/null
+++ b/lib/compress/zstd_ldm.c
@@ -0,0 +1,745 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_ldm.h"
+
+#include "../common/debug.h"
+#include "../common/xxhash.h"
+#include "zstd_fast.h"          /* ZSTD_fillHashTable() */
+#include "zstd_double_fast.h"   /* ZSTD_fillDoubleHashTable() */
+#include "zstd_ldm_geartab.h"
+
+#define LDM_BUCKET_SIZE_LOG 4
+#define LDM_MIN_MATCH_LENGTH 64
+#define LDM_HASH_RLOG 7
+
+typedef struct {
+    U64 rolling;
+    U64 stopMask;
+} ldmRollingHashState_t;
+
+/** ZSTD_ldm_gear_init():
+ *
+ * Initializes the rolling hash state such that it will honor the
+ * settings in params. */
+static void ZSTD_ldm_gear_init(ldmRollingHashState_t* state, ldmParams_t const* params)
+{
+    unsigned maxBitsInMask = MIN(params->minMatchLength, 64);
+    unsigned hashRateLog = params->hashRateLog;
+
+    state->rolling = ~(U32)0;
+
+    /* The choice of the splitting criterion is subject to two conditions:
+     *   1. it has to trigger on average every 2^(hashRateLog) bytes;
+     *   2. ideally, it has to depend on a window of minMatchLength bytes.
+     *
+     * In the gear hash algorithm, bit n depends on the last n bytes;
+     * so in order to obtain a good quality splitting criterion it is
+     * preferable to use bits with high weight.
+     *
+     * To match condition 1 we use a mask with hashRateLog bits set
+     * and, because of the previous remark, we make sure these bits
+     * have the highest possible weight while still respecting
+     * condition 2.
+     */
+    if (hashRateLog > 0 && hashRateLog <= maxBitsInMask) {
+        state->stopMask = (((U64)1 << hashRateLog) - 1) << (maxBitsInMask - hashRateLog);
+    } else {
+        /* In this degenerate case we simply honor the hash rate. */
+        state->stopMask = ((U64)1 << hashRateLog) - 1;
+    }
+}
+
+/** ZSTD_ldm_gear_reset()
+ * Feeds [data, data + minMatchLength) into the hash without registering any
+ * splits. This effectively resets the hash state. This is used when skipping
+ * over data, either at the beginning of a block, or skipping sections.
+ */
+static void ZSTD_ldm_gear_reset(ldmRollingHashState_t* state,
+                                BYTE const* data, size_t minMatchLength)
+{
+    U64 hash = state->rolling;
+    size_t n = 0;
+
+#define GEAR_ITER_ONCE() do {                                  \
+        hash = (hash << 1) + ZSTD_ldm_gearTab[data[n] & 0xff]; \
+        n += 1;                                                \
+    } while (0)
+    while (n + 3 < minMatchLength) {
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+    }
+    while (n < minMatchLength) {
+        GEAR_ITER_ONCE();
+    }
+#undef GEAR_ITER_ONCE
+}
+
+/** ZSTD_ldm_gear_feed():
+ *
+ * Registers in the splits array all the split points found in the first
+ * size bytes following the data pointer. This function terminates when
+ * either all the data has been processed or LDM_BATCH_SIZE splits are
+ * present in the splits array.
+ *
+ * Precondition: The splits array must not be full.
+ * Returns: The number of bytes processed. */
+static size_t ZSTD_ldm_gear_feed(ldmRollingHashState_t* state,
+                                 BYTE const* data, size_t size,
+                                 size_t* splits, unsigned* numSplits)
+{
+    size_t n;
+    U64 hash, mask;
+
+    hash = state->rolling;
+    mask = state->stopMask;
+    n = 0;
+
+#define GEAR_ITER_ONCE() do { \
+        hash = (hash << 1) + ZSTD_ldm_gearTab[data[n] & 0xff]; \
+        n += 1; \
+        if (UNLIKELY((hash & mask) == 0)) { \
+            splits[*numSplits] = n; \
+            *numSplits += 1; \
+            if (*numSplits == LDM_BATCH_SIZE) \
+                goto done; \
+        } \
+    } while (0)
+
+    while (n + 3 < size) {
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+        GEAR_ITER_ONCE();
+    }
+    while (n < size) {
+        GEAR_ITER_ONCE();
+    }
+
+#undef GEAR_ITER_ONCE
+
+done:
+    state->rolling = hash;
+    return n;
+}
+
+void ZSTD_ldm_adjustParameters(ldmParams_t* params,
+                        const ZSTD_compressionParameters* cParams)
+{
+    params->windowLog = cParams->windowLog;
+    ZSTD_STATIC_ASSERT(LDM_BUCKET_SIZE_LOG <= ZSTD_LDM_BUCKETSIZELOG_MAX);
+    DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
+    if (params->hashRateLog == 0) {
+        if (params->hashLog > 0) {
+            /* if params->hashLog is set, derive hashRateLog from it */
+            assert(params->hashLog <= ZSTD_HASHLOG_MAX);
+            if (params->windowLog > params->hashLog) {
+                params->hashRateLog = params->windowLog - params->hashLog;
+            }
+        } else {
+            assert(1 <= (int)cParams->strategy && (int)cParams->strategy <= 9);
+            /* mapping from [fast, rate7] to [btultra2, rate4] */
+            params->hashRateLog = 7 - (cParams->strategy/3);
+        }
+    }
+    if (params->hashLog == 0) {
+        params->hashLog = BOUNDED(ZSTD_HASHLOG_MIN, params->windowLog - params->hashRateLog, ZSTD_HASHLOG_MAX);
+    }
+    if (params->minMatchLength == 0) {
+        params->minMatchLength = LDM_MIN_MATCH_LENGTH;
+        if (cParams->strategy >= ZSTD_btultra)
+            params->minMatchLength /= 2;
+    }
+    if (params->bucketSizeLog==0) {
+        assert(1 <= (int)cParams->strategy && (int)cParams->strategy <= 9);
+        params->bucketSizeLog = BOUNDED(LDM_BUCKET_SIZE_LOG, (U32)cParams->strategy, ZSTD_LDM_BUCKETSIZELOG_MAX);
+    }
+    params->bucketSizeLog = MIN(params->bucketSizeLog, params->hashLog);
+}
+
+size_t ZSTD_ldm_getTableSize(ldmParams_t params)
+{
+    size_t const ldmHSize = ((size_t)1) << params.hashLog;
+    size_t const ldmBucketSizeLog = MIN(params.bucketSizeLog, params.hashLog);
+    size_t const ldmBucketSize = ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
+    size_t const totalSize = ZSTD_cwksp_alloc_size(ldmBucketSize)
+                           + ZSTD_cwksp_alloc_size(ldmHSize * sizeof(ldmEntry_t));
+    return params.enableLdm == ZSTD_ps_enable ? totalSize : 0;
+}
+
+size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
+{
+    return params.enableLdm == ZSTD_ps_enable ? (maxChunkSize / params.minMatchLength) : 0;
+}
+
+/** ZSTD_ldm_getBucket() :
+ *  Returns a pointer to the start of the bucket associated with hash. */
+static ldmEntry_t* ZSTD_ldm_getBucket(
+        const ldmState_t* ldmState, size_t hash, U32 const bucketSizeLog)
+{
+    return ldmState->hashTable + (hash << bucketSizeLog);
+}
+
+/** ZSTD_ldm_insertEntry() :
+ *  Insert the entry with corresponding hash into the hash table */
+static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
+                                 size_t const hash, const ldmEntry_t entry,
+                                 U32 const bucketSizeLog)
+{
+    BYTE* const pOffset = ldmState->bucketOffsets + hash;
+    unsigned const offset = *pOffset;
+
+    *(ZSTD_ldm_getBucket(ldmState, hash, bucketSizeLog) + offset) = entry;
+    *pOffset = (BYTE)((offset + 1) & ((1u << bucketSizeLog) - 1));
+
+}
+
+/** ZSTD_ldm_countBackwardsMatch() :
+ *  Returns the number of bytes that match backwards before pIn and pMatch.
+ *
+ *  We count only bytes where pMatch >= pBase and pIn >= pAnchor. */
+static size_t ZSTD_ldm_countBackwardsMatch(
+            const BYTE* pIn, const BYTE* pAnchor,
+            const BYTE* pMatch, const BYTE* pMatchBase)
+{
+    size_t matchLength = 0;
+    while (pIn > pAnchor && pMatch > pMatchBase && pIn[-1] == pMatch[-1]) {
+        pIn--;
+        pMatch--;
+        matchLength++;
+    }
+    return matchLength;
+}
+
+/** ZSTD_ldm_countBackwardsMatch_2segments() :
+ *  Returns the number of bytes that match backwards from pMatch,
+ *  even with the backwards match spanning 2 different segments.
+ *
+ *  On reaching `pMatchBase`, start counting from mEnd */
+static size_t ZSTD_ldm_countBackwardsMatch_2segments(
+                    const BYTE* pIn, const BYTE* pAnchor,
+                    const BYTE* pMatch, const BYTE* pMatchBase,
+                    const BYTE* pExtDictStart, const BYTE* pExtDictEnd)
+{
+    size_t matchLength = ZSTD_ldm_countBackwardsMatch(pIn, pAnchor, pMatch, pMatchBase);
+    if (pMatch - matchLength != pMatchBase || pMatchBase == pExtDictStart) {
+        /* If backwards match is entirely in the extDict or prefix, immediately return */
+        return matchLength;
+    }
+    DEBUGLOG(7, "ZSTD_ldm_countBackwardsMatch_2segments: found 2-parts backwards match (length in prefix==%zu)", matchLength);
+    matchLength += ZSTD_ldm_countBackwardsMatch(pIn - matchLength, pAnchor, pExtDictEnd, pExtDictStart);
+    DEBUGLOG(7, "final backwards match length = %zu", matchLength);
+    return matchLength;
+}
+
+/** ZSTD_ldm_fillFastTables() :
+ *
+ *  Fills the relevant tables for the ZSTD_fast and ZSTD_dfast strategies.
+ *  This is similar to ZSTD_loadDictionaryContent.
+ *
+ *  The tables for the other strategies are filled within their
+ *  block compressors. */
+static size_t ZSTD_ldm_fillFastTables(ZSTD_MatchState_t* ms,
+                                      void const* end)
+{
+    const BYTE* const iend = (const BYTE*)end;
+
+    switch(ms->cParams.strategy)
+    {
+    case ZSTD_fast:
+        ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast, ZSTD_tfp_forCCtx);
+        break;
+
+    case ZSTD_dfast:
+#ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
+        ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast, ZSTD_tfp_forCCtx);
+#else
+        assert(0); /* shouldn't be called: cparams should've been adjusted. */
+#endif
+        break;
+
+    case ZSTD_greedy:
+    case ZSTD_lazy:
+    case ZSTD_lazy2:
+    case ZSTD_btlazy2:
+    case ZSTD_btopt:
+    case ZSTD_btultra:
+    case ZSTD_btultra2:
+        break;
+    default:
+        assert(0);  /* not possible : not a valid strategy id */
+    }
+
+    return 0;
+}
+
+void ZSTD_ldm_fillHashTable(
+            ldmState_t* ldmState, const BYTE* ip,
+            const BYTE* iend, ldmParams_t const* params)
+{
+    U32 const minMatchLength = params->minMatchLength;
+    U32 const bucketSizeLog = params->bucketSizeLog;
+    U32 const hBits = params->hashLog - bucketSizeLog;
+    BYTE const* const base = ldmState->window.base;
+    BYTE const* const istart = ip;
+    ldmRollingHashState_t hashState;
+    size_t* const splits = ldmState->splitIndices;
+    unsigned numSplits;
+
+    DEBUGLOG(5, "ZSTD_ldm_fillHashTable");
+
+    ZSTD_ldm_gear_init(&hashState, params);
+    while (ip < iend) {
+        size_t hashed;
+        unsigned n;
+
+        numSplits = 0;
+        hashed = ZSTD_ldm_gear_feed(&hashState, ip, (size_t)(iend - ip), splits, &numSplits);
+
+        for (n = 0; n < numSplits; n++) {
+            if (ip + splits[n] >= istart + minMatchLength) {
+                BYTE const* const split = ip + splits[n] - minMatchLength;
+                U64 const xxhash = XXH64(split, minMatchLength, 0);
+                U32 const hash = (U32)(xxhash & (((U32)1 << hBits) - 1));
+                ldmEntry_t entry;
+
+                entry.offset = (U32)(split - base);
+                entry.checksum = (U32)(xxhash >> 32);
+                ZSTD_ldm_insertEntry(ldmState, hash, entry, params->bucketSizeLog);
+            }
+        }
+
+        ip += hashed;
+    }
+}
+
+
+/** ZSTD_ldm_limitTableUpdate() :
+ *
+ *  Sets cctx->nextToUpdate to a position corresponding closer to anchor
+ *  if it is far way
+ *  (after a long match, only update tables a limited amount). */
+static void ZSTD_ldm_limitTableUpdate(ZSTD_MatchState_t* ms, const BYTE* anchor)
+{
+    U32 const curr = (U32)(anchor - ms->window.base);
+    if (curr > ms->nextToUpdate + 1024) {
+        ms->nextToUpdate =
+            curr - MIN(512, curr - ms->nextToUpdate - 1024);
+    }
+}
+
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_ldm_generateSequences_internal(
+        ldmState_t* ldmState, RawSeqStore_t* rawSeqStore,
+        ldmParams_t const* params, void const* src, size_t srcSize)
+{
+    /* LDM parameters */
+    int const extDict = ZSTD_window_hasExtDict(ldmState->window);
+    U32 const minMatchLength = params->minMatchLength;
+    U32 const entsPerBucket = 1U << params->bucketSizeLog;
+    U32 const hBits = params->hashLog - params->bucketSizeLog;
+    /* Prefix and extDict parameters */
+    U32 const dictLimit = ldmState->window.dictLimit;
+    U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
+    BYTE const* const base = ldmState->window.base;
+    BYTE const* const dictBase = extDict ? ldmState->window.dictBase : NULL;
+    BYTE const* const dictStart = extDict ? dictBase + lowestIndex : NULL;
+    BYTE const* const dictEnd = extDict ? dictBase + dictLimit : NULL;
+    BYTE const* const lowPrefixPtr = base + dictLimit;
+    /* Input bounds */
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    BYTE const* const ilimit = iend - HASH_READ_SIZE;
+    /* Input positions */
+    BYTE const* anchor = istart;
+    BYTE const* ip = istart;
+    /* Rolling hash state */
+    ldmRollingHashState_t hashState;
+    /* Arrays for staged-processing */
+    size_t* const splits = ldmState->splitIndices;
+    ldmMatchCandidate_t* const candidates = ldmState->matchCandidates;
+    unsigned numSplits;
+
+    if (srcSize < minMatchLength)
+        return iend - anchor;
+
+    /* Initialize the rolling hash state with the first minMatchLength bytes */
+    ZSTD_ldm_gear_init(&hashState, params);
+    ZSTD_ldm_gear_reset(&hashState, ip, minMatchLength);
+    ip += minMatchLength;
+
+    while (ip < ilimit) {
+        size_t hashed;
+        unsigned n;
+
+        numSplits = 0;
+        hashed = ZSTD_ldm_gear_feed(&hashState, ip, ilimit - ip,
+                                    splits, &numSplits);
+
+        for (n = 0; n < numSplits; n++) {
+            BYTE const* const split = ip + splits[n] - minMatchLength;
+            U64 const xxhash = XXH64(split, minMatchLength, 0);
+            U32 const hash = (U32)(xxhash & (((U32)1 << hBits) - 1));
+
+            candidates[n].split = split;
+            candidates[n].hash = hash;
+            candidates[n].checksum = (U32)(xxhash >> 32);
+            candidates[n].bucket = ZSTD_ldm_getBucket(ldmState, hash, params->bucketSizeLog);
+            PREFETCH_L1(candidates[n].bucket);
+        }
+
+        for (n = 0; n < numSplits; n++) {
+            size_t forwardMatchLength = 0, backwardMatchLength = 0,
+                   bestMatchLength = 0, mLength;
+            U32 offset;
+            BYTE const* const split = candidates[n].split;
+            U32 const checksum = candidates[n].checksum;
+            U32 const hash = candidates[n].hash;
+            ldmEntry_t* const bucket = candidates[n].bucket;
+            ldmEntry_t const* cur;
+            ldmEntry_t const* bestEntry = NULL;
+            ldmEntry_t newEntry;
+
+            newEntry.offset = (U32)(split - base);
+            newEntry.checksum = checksum;
+
+            /* If a split point would generate a sequence overlapping with
+             * the previous one, we merely register it in the hash table and
+             * move on */
+            if (split < anchor) {
+                ZSTD_ldm_insertEntry(ldmState, hash, newEntry, params->bucketSizeLog);
+                continue;
+            }
+
+            for (cur = bucket; cur < bucket + entsPerBucket; cur++) {
+                size_t curForwardMatchLength, curBackwardMatchLength,
+                       curTotalMatchLength;
+                if (cur->checksum != checksum || cur->offset <= lowestIndex) {
+                    continue;
+                }
+                if (extDict) {
+                    BYTE const* const curMatchBase =
+                        cur->offset < dictLimit ? dictBase : base;
+                    BYTE const* const pMatch = curMatchBase + cur->offset;
+                    BYTE const* const matchEnd =
+                        cur->offset < dictLimit ? dictEnd : iend;
+                    BYTE const* const lowMatchPtr =
+                        cur->offset < dictLimit ? dictStart : lowPrefixPtr;
+                    curForwardMatchLength =
+                        ZSTD_count_2segments(split, pMatch, iend, matchEnd, lowPrefixPtr);
+                    if (curForwardMatchLength < minMatchLength) {
+                        continue;
+                    }
+                    curBackwardMatchLength = ZSTD_ldm_countBackwardsMatch_2segments(
+                            split, anchor, pMatch, lowMatchPtr, dictStart, dictEnd);
+                } else { /* !extDict */
+                    BYTE const* const pMatch = base + cur->offset;
+                    curForwardMatchLength = ZSTD_count(split, pMatch, iend);
+                    if (curForwardMatchLength < minMatchLength) {
+                        continue;
+                    }
+                    curBackwardMatchLength =
+                        ZSTD_ldm_countBackwardsMatch(split, anchor, pMatch, lowPrefixPtr);
+                }
+                curTotalMatchLength = curForwardMatchLength + curBackwardMatchLength;
+
+                if (curTotalMatchLength > bestMatchLength) {
+                    bestMatchLength = curTotalMatchLength;
+                    forwardMatchLength = curForwardMatchLength;
+                    backwardMatchLength = curBackwardMatchLength;
+                    bestEntry = cur;
+                }
+            }
+
+            /* No match found -- insert an entry into the hash table
+             * and process the next candidate match */
+            if (bestEntry == NULL) {
+                ZSTD_ldm_insertEntry(ldmState, hash, newEntry, params->bucketSizeLog);
+                continue;
+            }
+
+            /* Match found */
+            offset = (U32)(split - base) - bestEntry->offset;
+            mLength = forwardMatchLength + backwardMatchLength;
+            {
+                rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
+
+                /* Out of sequence storage */
+                if (rawSeqStore->size == rawSeqStore->capacity)
+                    return ERROR(dstSize_tooSmall);
+                seq->litLength = (U32)(split - backwardMatchLength - anchor);
+                seq->matchLength = (U32)mLength;
+                seq->offset = offset;
+                rawSeqStore->size++;
+            }
+
+            /* Insert the current entry into the hash table --- it must be
+             * done after the previous block to avoid clobbering bestEntry */
+            ZSTD_ldm_insertEntry(ldmState, hash, newEntry, params->bucketSizeLog);
+
+            anchor = split + forwardMatchLength;
+
+            /* If we find a match that ends after the data that we've hashed
+             * then we have a repeating, overlapping, pattern. E.g. all zeros.
+             * If one repetition of the pattern matches our `stopMask` then all
+             * repetitions will. We don't need to insert them all into out table,
+             * only the first one. So skip over overlapping matches.
+             * This is a major speed boost (20x) for compressing a single byte
+             * repeated, when that byte ends up in the table.
+             */
+            if (anchor > ip + hashed) {
+                ZSTD_ldm_gear_reset(&hashState, anchor - minMatchLength, minMatchLength);
+                /* Continue the outer loop at anchor (ip + hashed == anchor). */
+                ip = anchor - hashed;
+                break;
+            }
+        }
+
+        ip += hashed;
+    }
+
+    return iend - anchor;
+}
+
+/*! ZSTD_ldm_reduceTable() :
+ *  reduce table indexes by `reducerValue` */
+static void ZSTD_ldm_reduceTable(ldmEntry_t* const table, U32 const size,
+                                 U32 const reducerValue)
+{
+    U32 u;
+    for (u = 0; u < size; u++) {
+        if (table[u].offset < reducerValue) table[u].offset = 0;
+        else table[u].offset -= reducerValue;
+    }
+}
+
+size_t ZSTD_ldm_generateSequences(
+        ldmState_t* ldmState, RawSeqStore_t* sequences,
+        ldmParams_t const* params, void const* src, size_t srcSize)
+{
+    U32 const maxDist = 1U << params->windowLog;
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    size_t const kMaxChunkSize = 1 << 20;
+    size_t const nbChunks = (srcSize / kMaxChunkSize) + ((srcSize % kMaxChunkSize) != 0);
+    size_t chunk;
+    size_t leftoverSize = 0;
+
+    assert(ZSTD_CHUNKSIZE_MAX >= kMaxChunkSize);
+    /* Check that ZSTD_window_update() has been called for this chunk prior
+     * to passing it to this function.
+     */
+    assert(ldmState->window.nextSrc >= (BYTE const*)src + srcSize);
+    /* The input could be very large (in zstdmt), so it must be broken up into
+     * chunks to enforce the maximum distance and handle overflow correction.
+     */
+    assert(sequences->pos <= sequences->size);
+    assert(sequences->size <= sequences->capacity);
+    for (chunk = 0; chunk < nbChunks && sequences->size < sequences->capacity; ++chunk) {
+        BYTE const* const chunkStart = istart + chunk * kMaxChunkSize;
+        size_t const remaining = (size_t)(iend - chunkStart);
+        BYTE const *const chunkEnd =
+            (remaining < kMaxChunkSize) ? iend : chunkStart + kMaxChunkSize;
+        size_t const chunkSize = chunkEnd - chunkStart;
+        size_t newLeftoverSize;
+        size_t const prevSize = sequences->size;
+
+        assert(chunkStart < iend);
+        /* 1. Perform overflow correction if necessary. */
+        if (ZSTD_window_needOverflowCorrection(ldmState->window, 0, maxDist, ldmState->loadedDictEnd, chunkStart, chunkEnd)) {
+            U32 const ldmHSize = 1U << params->hashLog;
+            U32 const correction = ZSTD_window_correctOverflow(
+                &ldmState->window, /* cycleLog */ 0, maxDist, chunkStart);
+            ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
+            /* invalidate dictionaries on overflow correction */
+            ldmState->loadedDictEnd = 0;
+        }
+        /* 2. We enforce the maximum offset allowed.
+         *
+         * kMaxChunkSize should be small enough that we don't lose too much of
+         * the window through early invalidation.
+         * TODO: * Test the chunk size.
+         *       * Try invalidation after the sequence generation and test the
+         *         offset against maxDist directly.
+         *
+         * NOTE: Because of dictionaries + sequence splitting we MUST make sure
+         * that any offset used is valid at the END of the sequence, since it may
+         * be split into two sequences. This condition holds when using
+         * ZSTD_window_enforceMaxDist(), but if we move to checking offsets
+         * against maxDist directly, we'll have to carefully handle that case.
+         */
+        ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, &ldmState->loadedDictEnd, NULL);
+        /* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
+        newLeftoverSize = ZSTD_ldm_generateSequences_internal(
+            ldmState, sequences, params, chunkStart, chunkSize);
+        if (ZSTD_isError(newLeftoverSize))
+            return newLeftoverSize;
+        /* 4. We add the leftover literals from previous iterations to the first
+         *    newly generated sequence, or add the `newLeftoverSize` if none are
+         *    generated.
+         */
+        /* Prepend the leftover literals from the last call */
+        if (prevSize < sequences->size) {
+            sequences->seq[prevSize].litLength += (U32)leftoverSize;
+            leftoverSize = newLeftoverSize;
+        } else {
+            assert(newLeftoverSize == chunkSize);
+            leftoverSize += chunkSize;
+        }
+    }
+    return 0;
+}
+
+void
+ZSTD_ldm_skipSequences(RawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch)
+{
+    while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
+        if (srcSize <= seq->litLength) {
+            /* Skip past srcSize literals */
+            seq->litLength -= (U32)srcSize;
+            return;
+        }
+        srcSize -= seq->litLength;
+        seq->litLength = 0;
+        if (srcSize < seq->matchLength) {
+            /* Skip past the first srcSize of the match */
+            seq->matchLength -= (U32)srcSize;
+            if (seq->matchLength < minMatch) {
+                /* The match is too short, omit it */
+                if (rawSeqStore->pos + 1 < rawSeqStore->size) {
+                    seq[1].litLength += seq[0].matchLength;
+                }
+                rawSeqStore->pos++;
+            }
+            return;
+        }
+        srcSize -= seq->matchLength;
+        seq->matchLength = 0;
+        rawSeqStore->pos++;
+    }
+}
+
+/**
+ * If the sequence length is longer than remaining then the sequence is split
+ * between this block and the next.
+ *
+ * Returns the current sequence to handle, or if the rest of the block should
+ * be literals, it returns a sequence with offset == 0.
+ */
+static rawSeq maybeSplitSequence(RawSeqStore_t* rawSeqStore,
+                                 U32 const remaining, U32 const minMatch)
+{
+    rawSeq sequence = rawSeqStore->seq[rawSeqStore->pos];
+    assert(sequence.offset > 0);
+    /* Likely: No partial sequence */
+    if (remaining >= sequence.litLength + sequence.matchLength) {
+        rawSeqStore->pos++;
+        return sequence;
+    }
+    /* Cut the sequence short (offset == 0 ==> rest is literals). */
+    if (remaining <= sequence.litLength) {
+        sequence.offset = 0;
+    } else if (remaining < sequence.litLength + sequence.matchLength) {
+        sequence.matchLength = remaining - sequence.litLength;
+        if (sequence.matchLength < minMatch) {
+            sequence.offset = 0;
+        }
+    }
+    /* Skip past `remaining` bytes for the future sequences. */
+    ZSTD_ldm_skipSequences(rawSeqStore, remaining, minMatch);
+    return sequence;
+}
+
+void ZSTD_ldm_skipRawSeqStoreBytes(RawSeqStore_t* rawSeqStore, size_t nbBytes) {
+    U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
+    while (currPos && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
+        if (currPos >= currSeq.litLength + currSeq.matchLength) {
+            currPos -= currSeq.litLength + currSeq.matchLength;
+            rawSeqStore->pos++;
+        } else {
+            rawSeqStore->posInSequence = currPos;
+            break;
+        }
+    }
+    if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
+        rawSeqStore->posInSequence = 0;
+    }
+}
+
+size_t ZSTD_ldm_blockCompress(RawSeqStore_t* rawSeqStore,
+    ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+    ZSTD_ParamSwitch_e useRowMatchFinder,
+    void const* src, size_t srcSize)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    unsigned const minMatch = cParams->minMatch;
+    ZSTD_BlockCompressor_f const blockCompressor =
+        ZSTD_selectBlockCompressor(cParams->strategy, useRowMatchFinder, ZSTD_matchState_dictMode(ms));
+    /* Input bounds */
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    /* Input positions */
+    BYTE const* ip = istart;
+
+    DEBUGLOG(5, "ZSTD_ldm_blockCompress: srcSize=%zu", srcSize);
+    /* If using opt parser, use LDMs only as candidates rather than always accepting them */
+    if (cParams->strategy >= ZSTD_btopt) {
+        size_t lastLLSize;
+        ms->ldmSeqStore = rawSeqStore;
+        lastLLSize = blockCompressor(ms, seqStore, rep, src, srcSize);
+        ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore, srcSize);
+        return lastLLSize;
+    }
+
+    assert(rawSeqStore->pos <= rawSeqStore->size);
+    assert(rawSeqStore->size <= rawSeqStore->capacity);
+    /* Loop through each sequence and apply the block compressor to the literals */
+    while (rawSeqStore->pos < rawSeqStore->size && ip < iend) {
+        /* maybeSplitSequence updates rawSeqStore->pos */
+        rawSeq const sequence = maybeSplitSequence(rawSeqStore,
+                                                   (U32)(iend - ip), minMatch);
+        /* End signal */
+        if (sequence.offset == 0)
+            break;
+
+        assert(ip + sequence.litLength + sequence.matchLength <= iend);
+
+        /* Fill tables for block compressor */
+        ZSTD_ldm_limitTableUpdate(ms, ip);
+        ZSTD_ldm_fillFastTables(ms, ip);
+        /* Run the block compressor */
+        DEBUGLOG(5, "pos %u : calling block compressor on segment of size %u", (unsigned)(ip-istart), sequence.litLength);
+        {
+            int i;
+            size_t const newLitLength =
+                blockCompressor(ms, seqStore, rep, ip, sequence.litLength);
+            ip += sequence.litLength;
+            /* Update the repcodes */
+            for (i = ZSTD_REP_NUM - 1; i > 0; i--)
+                rep[i] = rep[i-1];
+            rep[0] = sequence.offset;
+            /* Store the sequence */
+            ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength, iend,
+                          OFFSET_TO_OFFBASE(sequence.offset),
+                          sequence.matchLength);
+            ip += sequence.matchLength;
+        }
+    }
+    /* Fill the tables for the block compressor */
+    ZSTD_ldm_limitTableUpdate(ms, ip);
+    ZSTD_ldm_fillFastTables(ms, ip);
+    /* Compress the last literals */
+    return blockCompressor(ms, seqStore, rep, ip, iend - ip);
+}
diff --git a/lib/compress/zstd_opt.c b/lib/compress/zstd_opt.c
new file mode 100644
index 0000000..3d7171b
--- /dev/null
+++ b/lib/compress/zstd_opt.c
@@ -0,0 +1,1580 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "hist.h"
+#include "zstd_opt.h"
+
+#if !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR) \
+ || !defined(ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR)
+
+#define ZSTD_LITFREQ_ADD    2   /* scaling factor for litFreq, so that frequencies adapt faster to new stats */
+#define ZSTD_MAX_PRICE     (1<<30)
+
+#define ZSTD_PREDEF_THRESHOLD 8   /* if srcSize < ZSTD_PREDEF_THRESHOLD, symbols' cost is assumed static, directly determined by pre-defined distributions */
+
+
+/*-*************************************
+*  Price functions for optimal parser
+***************************************/
+
+#if 0    /* approximation at bit level (for tests) */
+#  define BITCOST_ACCURACY 0
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat, opt) ((void)(opt), ZSTD_bitWeight(stat))
+#elif 0  /* fractional bit accuracy (for tests) */
+#  define BITCOST_ACCURACY 8
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat,opt) ((void)(opt), ZSTD_fracWeight(stat))
+#else    /* opt==approx, ultra==accurate */
+#  define BITCOST_ACCURACY 8
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat,opt) ((opt) ? ZSTD_fracWeight(stat) : ZSTD_bitWeight(stat))
+#endif
+
+/* ZSTD_bitWeight() :
+ * provide estimated "cost" of a stat in full bits only */
+MEM_STATIC U32 ZSTD_bitWeight(U32 stat)
+{
+    return (ZSTD_highbit32(stat+1) * BITCOST_MULTIPLIER);
+}
+
+/* ZSTD_fracWeight() :
+ * provide fractional-bit "cost" of a stat,
+ * using linear interpolation approximation */
+MEM_STATIC U32 ZSTD_fracWeight(U32 rawStat)
+{
+    U32 const stat = rawStat + 1;
+    U32 const hb = ZSTD_highbit32(stat);
+    U32 const BWeight = hb * BITCOST_MULTIPLIER;
+    /* Fweight was meant for "Fractional weight"
+     * but it's effectively a value between 1 and 2
+     * using fixed point arithmetic */
+    U32 const FWeight = (stat << BITCOST_ACCURACY) >> hb;
+    U32 const weight = BWeight + FWeight;
+    assert(hb + BITCOST_ACCURACY < 31);
+    return weight;
+}
+
+#if (DEBUGLEVEL>=2)
+/* debugging function,
+ * @return price in bytes as fractional value
+ * for debug messages only */
+MEM_STATIC double ZSTD_fCost(int price)
+{
+    return (double)price / (BITCOST_MULTIPLIER*8);
+}
+#endif
+
+static int ZSTD_compressedLiterals(optState_t const* const optPtr)
+{
+    return optPtr->literalCompressionMode != ZSTD_ps_disable;
+}
+
+static void ZSTD_setBasePrices(optState_t* optPtr, int optLevel)
+{
+    if (ZSTD_compressedLiterals(optPtr))
+        optPtr->litSumBasePrice = WEIGHT(optPtr->litSum, optLevel);
+    optPtr->litLengthSumBasePrice = WEIGHT(optPtr->litLengthSum, optLevel);
+    optPtr->matchLengthSumBasePrice = WEIGHT(optPtr->matchLengthSum, optLevel);
+    optPtr->offCodeSumBasePrice = WEIGHT(optPtr->offCodeSum, optLevel);
+}
+
+
+static U32 sum_u32(const unsigned table[], size_t nbElts)
+{
+    size_t n;
+    U32 total = 0;
+    for (n=0; n<nbElts; n++) {
+        total += table[n];
+    }
+    return total;
+}
+
+typedef enum { base_0possible=0, base_1guaranteed=1 } base_directive_e;
+
+static U32
+ZSTD_downscaleStats(unsigned* table, U32 lastEltIndex, U32 shift, base_directive_e base1)
+{
+    U32 s, sum=0;
+    DEBUGLOG(5, "ZSTD_downscaleStats (nbElts=%u, shift=%u)",
+            (unsigned)lastEltIndex+1, (unsigned)shift );
+    assert(shift < 30);
+    for (s=0; s<lastEltIndex+1; s++) {
+        unsigned const base = base1 ? 1 : (table[s]>0);
+        unsigned const newStat = base + (table[s] >> shift);
+        sum += newStat;
+        table[s] = newStat;
+    }
+    return sum;
+}
+
+/* ZSTD_scaleStats() :
+ * reduce all elt frequencies in table if sum too large
+ * return the resulting sum of elements */
+static U32 ZSTD_scaleStats(unsigned* table, U32 lastEltIndex, U32 logTarget)
+{
+    U32 const prevsum = sum_u32(table, lastEltIndex+1);
+    U32 const factor = prevsum >> logTarget;
+    DEBUGLOG(5, "ZSTD_scaleStats (nbElts=%u, target=%u)", (unsigned)lastEltIndex+1, (unsigned)logTarget);
+    assert(logTarget < 30);
+    if (factor <= 1) return prevsum;
+    return ZSTD_downscaleStats(table, lastEltIndex, ZSTD_highbit32(factor), base_1guaranteed);
+}
+
+/* ZSTD_rescaleFreqs() :
+ * if first block (detected by optPtr->litLengthSum == 0) : init statistics
+ *    take hints from dictionary if there is one
+ *    and init from zero if there is none,
+ *    using src for literals stats, and baseline stats for sequence symbols
+ * otherwise downscale existing stats, to be used as seed for next block.
+ */
+static void
+ZSTD_rescaleFreqs(optState_t* const optPtr,
+            const BYTE* const src, size_t const srcSize,
+                  int const optLevel)
+{
+    int const compressedLiterals = ZSTD_compressedLiterals(optPtr);
+    DEBUGLOG(5, "ZSTD_rescaleFreqs (srcSize=%u)", (unsigned)srcSize);
+    optPtr->priceType = zop_dynamic;
+
+    if (optPtr->litLengthSum == 0) {  /* no literals stats collected -> first block assumed -> init */
+
+        /* heuristic: use pre-defined stats for too small inputs */
+        if (srcSize <= ZSTD_PREDEF_THRESHOLD) {
+            DEBUGLOG(5, "srcSize <= %i : use predefined stats", ZSTD_PREDEF_THRESHOLD);
+            optPtr->priceType = zop_predef;
+        }
+
+        assert(optPtr->symbolCosts != NULL);
+        if (optPtr->symbolCosts->huf.repeatMode == HUF_repeat_valid) {
+
+            /* huffman stats covering the full value set : table presumed generated by dictionary */
+            optPtr->priceType = zop_dynamic;
+
+            if (compressedLiterals) {
+                /* generate literals statistics from huffman table */
+                unsigned lit;
+                assert(optPtr->litFreq != NULL);
+                optPtr->litSum = 0;
+                for (lit=0; lit<=MaxLit; lit++) {
+                    U32 const scaleLog = 11;   /* scale to 2K */
+                    U32 const bitCost = HUF_getNbBitsFromCTable(optPtr->symbolCosts->huf.CTable, lit);
+                    assert(bitCost <= scaleLog);
+                    optPtr->litFreq[lit] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->litSum += optPtr->litFreq[lit];
+            }   }
+
+            {   unsigned ll;
+                FSE_CState_t llstate;
+                FSE_initCState(&llstate, optPtr->symbolCosts->fse.litlengthCTable);
+                optPtr->litLengthSum = 0;
+                for (ll=0; ll<=MaxLL; ll++) {
+                    U32 const scaleLog = 10;   /* scale to 1K */
+                    U32 const bitCost = FSE_getMaxNbBits(llstate.symbolTT, ll);
+                    assert(bitCost < scaleLog);
+                    optPtr->litLengthFreq[ll] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->litLengthSum += optPtr->litLengthFreq[ll];
+            }   }
+
+            {   unsigned ml;
+                FSE_CState_t mlstate;
+                FSE_initCState(&mlstate, optPtr->symbolCosts->fse.matchlengthCTable);
+                optPtr->matchLengthSum = 0;
+                for (ml=0; ml<=MaxML; ml++) {
+                    U32 const scaleLog = 10;
+                    U32 const bitCost = FSE_getMaxNbBits(mlstate.symbolTT, ml);
+                    assert(bitCost < scaleLog);
+                    optPtr->matchLengthFreq[ml] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->matchLengthSum += optPtr->matchLengthFreq[ml];
+            }   }
+
+            {   unsigned of;
+                FSE_CState_t ofstate;
+                FSE_initCState(&ofstate, optPtr->symbolCosts->fse.offcodeCTable);
+                optPtr->offCodeSum = 0;
+                for (of=0; of<=MaxOff; of++) {
+                    U32 const scaleLog = 10;
+                    U32 const bitCost = FSE_getMaxNbBits(ofstate.symbolTT, of);
+                    assert(bitCost < scaleLog);
+                    optPtr->offCodeFreq[of] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->offCodeSum += optPtr->offCodeFreq[of];
+            }   }
+
+        } else {  /* first block, no dictionary */
+
+            assert(optPtr->litFreq != NULL);
+            if (compressedLiterals) {
+                /* base initial cost of literals on direct frequency within src */
+                unsigned lit = MaxLit;
+                HIST_count_simple(optPtr->litFreq, &lit, src, srcSize);   /* use raw first block to init statistics */
+                optPtr->litSum = ZSTD_downscaleStats(optPtr->litFreq, MaxLit, 8, base_0possible);
+            }
+
+            {   unsigned const baseLLfreqs[MaxLL+1] = {
+                    4, 2, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1
+                };
+                ZSTD_memcpy(optPtr->litLengthFreq, baseLLfreqs, sizeof(baseLLfreqs));
+                optPtr->litLengthSum = sum_u32(baseLLfreqs, MaxLL+1);
+            }
+
+            {   unsigned ml;
+                for (ml=0; ml<=MaxML; ml++)
+                    optPtr->matchLengthFreq[ml] = 1;
+            }
+            optPtr->matchLengthSum = MaxML+1;
+
+            {   unsigned const baseOFCfreqs[MaxOff+1] = {
+                    6, 2, 1, 1, 2, 3, 4, 4,
+                    4, 3, 2, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1
+                };
+                ZSTD_memcpy(optPtr->offCodeFreq, baseOFCfreqs, sizeof(baseOFCfreqs));
+                optPtr->offCodeSum = sum_u32(baseOFCfreqs, MaxOff+1);
+            }
+
+        }
+
+    } else {   /* new block : scale down accumulated statistics */
+
+        if (compressedLiterals)
+            optPtr->litSum = ZSTD_scaleStats(optPtr->litFreq, MaxLit, 12);
+        optPtr->litLengthSum = ZSTD_scaleStats(optPtr->litLengthFreq, MaxLL, 11);
+        optPtr->matchLengthSum = ZSTD_scaleStats(optPtr->matchLengthFreq, MaxML, 11);
+        optPtr->offCodeSum = ZSTD_scaleStats(optPtr->offCodeFreq, MaxOff, 11);
+    }
+
+    ZSTD_setBasePrices(optPtr, optLevel);
+}
+
+/* ZSTD_rawLiteralsCost() :
+ * price of literals (only) in specified segment (which length can be 0).
+ * does not include price of literalLength symbol */
+static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
+                                const optState_t* const optPtr,
+                                int optLevel)
+{
+    DEBUGLOG(8, "ZSTD_rawLiteralsCost (%u literals)", litLength);
+    if (litLength == 0) return 0;
+
+    if (!ZSTD_compressedLiterals(optPtr))
+        return (litLength << 3) * BITCOST_MULTIPLIER;  /* Uncompressed - 8 bytes per literal. */
+
+    if (optPtr->priceType == zop_predef)
+        return (litLength*6) * BITCOST_MULTIPLIER;  /* 6 bit per literal - no statistic used */
+
+    /* dynamic statistics */
+    {   U32 price = optPtr->litSumBasePrice * litLength;
+        U32 const litPriceMax = optPtr->litSumBasePrice - BITCOST_MULTIPLIER;
+        U32 u;
+        assert(optPtr->litSumBasePrice >= BITCOST_MULTIPLIER);
+        for (u=0; u < litLength; u++) {
+            U32 litPrice = WEIGHT(optPtr->litFreq[literals[u]], optLevel);
+            if (UNLIKELY(litPrice > litPriceMax)) litPrice = litPriceMax;
+            price -= litPrice;
+        }
+        return price;
+    }
+}
+
+/* ZSTD_litLengthPrice() :
+ * cost of literalLength symbol */
+static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optPtr, int optLevel)
+{
+    assert(litLength <= ZSTD_BLOCKSIZE_MAX);
+    if (optPtr->priceType == zop_predef)
+        return WEIGHT(litLength, optLevel);
+
+    /* ZSTD_LLcode() can't compute litLength price for sizes >= ZSTD_BLOCKSIZE_MAX
+     * because it isn't representable in the zstd format.
+     * So instead just pretend it would cost 1 bit more than ZSTD_BLOCKSIZE_MAX - 1.
+     * In such a case, the block would be all literals.
+     */
+    if (litLength == ZSTD_BLOCKSIZE_MAX)
+        return BITCOST_MULTIPLIER + ZSTD_litLengthPrice(ZSTD_BLOCKSIZE_MAX - 1, optPtr, optLevel);
+
+    /* dynamic statistics */
+    {   U32 const llCode = ZSTD_LLcode(litLength);
+        return (LL_bits[llCode] * BITCOST_MULTIPLIER)
+             + optPtr->litLengthSumBasePrice
+             - WEIGHT(optPtr->litLengthFreq[llCode], optLevel);
+    }
+}
+
+/* ZSTD_getMatchPrice() :
+ * Provides the cost of the match part (offset + matchLength) of a sequence.
+ * Must be combined with ZSTD_fullLiteralsCost() to get the full cost of a sequence.
+ * @offBase : sumtype, representing an offset or a repcode, and using numeric representation of ZSTD_storeSeq()
+ * @optLevel: when <2, favors small offset for decompression speed (improved cache efficiency)
+ */
+FORCE_INLINE_TEMPLATE U32
+ZSTD_getMatchPrice(U32 const offBase,
+                   U32 const matchLength,
+             const optState_t* const optPtr,
+                   int const optLevel)
+{
+    U32 price;
+    U32 const offCode = ZSTD_highbit32(offBase);
+    U32 const mlBase = matchLength - MINMATCH;
+    assert(matchLength >= MINMATCH);
+
+    if (optPtr->priceType == zop_predef)  /* fixed scheme, does not use statistics */
+        return WEIGHT(mlBase, optLevel)
+             + ((16 + offCode) * BITCOST_MULTIPLIER); /* emulated offset cost */
+
+    /* dynamic statistics */
+    price = (offCode * BITCOST_MULTIPLIER) + (optPtr->offCodeSumBasePrice - WEIGHT(optPtr->offCodeFreq[offCode], optLevel));
+    if ((optLevel<2) /*static*/ && offCode >= 20)
+        price += (offCode-19)*2 * BITCOST_MULTIPLIER; /* handicap for long distance offsets, favor decompression speed */
+
+    /* match Length */
+    {   U32 const mlCode = ZSTD_MLcode(mlBase);
+        price += (ML_bits[mlCode] * BITCOST_MULTIPLIER) + (optPtr->matchLengthSumBasePrice - WEIGHT(optPtr->matchLengthFreq[mlCode], optLevel));
+    }
+
+    price += BITCOST_MULTIPLIER / 5;   /* heuristic : make matches a bit more costly to favor less sequences -> faster decompression speed */
+
+    DEBUGLOG(8, "ZSTD_getMatchPrice(ml:%u) = %u", matchLength, price);
+    return price;
+}
+
+/* ZSTD_updateStats() :
+ * assumption : literals + litLength <= iend */
+static void ZSTD_updateStats(optState_t* const optPtr,
+                             U32 litLength, const BYTE* literals,
+                             U32 offBase, U32 matchLength)
+{
+    /* literals */
+    if (ZSTD_compressedLiterals(optPtr)) {
+        U32 u;
+        for (u=0; u < litLength; u++)
+            optPtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
+        optPtr->litSum += litLength*ZSTD_LITFREQ_ADD;
+    }
+
+    /* literal Length */
+    {   U32 const llCode = ZSTD_LLcode(litLength);
+        optPtr->litLengthFreq[llCode]++;
+        optPtr->litLengthSum++;
+    }
+
+    /* offset code : follows storeSeq() numeric representation */
+    {   U32 const offCode = ZSTD_highbit32(offBase);
+        assert(offCode <= MaxOff);
+        optPtr->offCodeFreq[offCode]++;
+        optPtr->offCodeSum++;
+    }
+
+    /* match Length */
+    {   U32 const mlBase = matchLength - MINMATCH;
+        U32 const mlCode = ZSTD_MLcode(mlBase);
+        optPtr->matchLengthFreq[mlCode]++;
+        optPtr->matchLengthSum++;
+    }
+}
+
+
+/* ZSTD_readMINMATCH() :
+ * function safe only for comparisons
+ * assumption : memPtr must be at least 4 bytes before end of buffer */
+MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
+{
+    switch (length)
+    {
+    default :
+    case 4 : return MEM_read32(memPtr);
+    case 3 : if (MEM_isLittleEndian())
+                return MEM_read32(memPtr)<<8;
+             else
+                return MEM_read32(memPtr)>>8;
+    }
+}
+
+
+/* Update hashTable3 up to ip (excluded)
+   Assumption : always within prefix (i.e. not within extDict) */
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_insertAndFindFirstIndexHash3 (const ZSTD_MatchState_t* ms,
+                                       U32* nextToUpdate3,
+                                       const BYTE* const ip)
+{
+    U32* const hashTable3 = ms->hashTable3;
+    U32 const hashLog3 = ms->hashLog3;
+    const BYTE* const base = ms->window.base;
+    U32 idx = *nextToUpdate3;
+    U32 const target = (U32)(ip - base);
+    size_t const hash3 = ZSTD_hash3Ptr(ip, hashLog3);
+    assert(hashLog3 > 0);
+
+    while(idx < target) {
+        hashTable3[ZSTD_hash3Ptr(base+idx, hashLog3)] = idx;
+        idx++;
+    }
+
+    *nextToUpdate3 = target;
+    return hashTable3[hash3];
+}
+
+
+/*-*************************************
+*  Binary Tree search
+***************************************/
+/** ZSTD_insertBt1() : add one or multiple positions to tree.
+ * @param ip assumed <= iend-8 .
+ * @param target The target of ZSTD_updateTree_internal() - we are filling to this position
+ * @return : nb of positions added */
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_insertBt1(
+                const ZSTD_MatchState_t* ms,
+                const BYTE* const ip, const BYTE* const iend,
+                U32 const target,
+                U32 const mls, const int extDict)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32*   const hashTable = ms->hashTable;
+    U32    const hashLog = cParams->hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32 matchIndex = hashTable[h];
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* match;
+    const U32 curr = (U32)(ip-base);
+    const U32 btLow = btMask >= curr ? 0 : curr - btMask;
+    U32* smallerPtr = bt + 2*(curr&btMask);
+    U32* largerPtr  = smallerPtr + 1;
+    U32 dummy32;   /* to be nullified at the end */
+    /* windowLow is based on target because
+     * we only need positions that will be in the window at the end of the tree update.
+     */
+    U32 const windowLow = ZSTD_getLowestMatchIndex(ms, target, cParams->windowLog);
+    U32 matchEndIdx = curr+8+1;
+    size_t bestLength = 8;
+    U32 nbCompares = 1U << cParams->searchLog;
+#ifdef ZSTD_C_PREDICT
+    U32 predictedSmall = *(bt + 2*((curr-1)&btMask) + 0);
+    U32 predictedLarge = *(bt + 2*((curr-1)&btMask) + 1);
+    predictedSmall += (predictedSmall>0);
+    predictedLarge += (predictedLarge>0);
+#endif /* ZSTD_C_PREDICT */
+
+    DEBUGLOG(8, "ZSTD_insertBt1 (%u)", curr);
+
+    assert(curr <= target);
+    assert(ip <= iend-8);   /* required for h calculation */
+    hashTable[h] = curr;   /* Update Hash Table */
+
+    assert(windowLow > 0);
+    for (; nbCompares && (matchIndex >= windowLow); --nbCompares) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(matchIndex < curr);
+
+#ifdef ZSTD_C_PREDICT   /* note : can create issues when hlog small <= 11 */
+        const U32* predictPtr = bt + 2*((matchIndex-1) & btMask);   /* written this way, as bt is a roll buffer */
+        if (matchIndex == predictedSmall) {
+            /* no need to check length, result known */
+            *smallerPtr = matchIndex;
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            predictedSmall = predictPtr[1] + (predictPtr[1]>0);
+            continue;
+        }
+        if (matchIndex == predictedLarge) {
+            *largerPtr = matchIndex;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+            predictedLarge = predictPtr[0] + (predictPtr[0]>0);
+            continue;
+        }
+#endif
+
+        if (!extDict || (matchIndex+matchLength >= dictLimit)) {
+            assert(matchIndex+matchLength >= dictLimit);   /* might be wrong if actually extDict */
+            match = base + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+        }
+
+        if (matchLength > bestLength) {
+            bestLength = matchLength;
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
+        }
+
+        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
+        }
+
+        if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+    {   U32 positions = 0;
+        if (bestLength > 384) positions = MIN(192, (U32)(bestLength - 384));   /* speed optimization */
+        assert(matchEndIdx > curr + 8);
+        return MAX(positions, matchEndIdx - (curr + 8));
+    }
+}
+
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_updateTree_internal(
+                ZSTD_MatchState_t* ms,
+                const BYTE* const ip, const BYTE* const iend,
+                const U32 mls, const ZSTD_dictMode_e dictMode)
+{
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+    DEBUGLOG(7, "ZSTD_updateTree_internal, from %u to %u  (dictMode:%u)",
+                idx, target, dictMode);
+
+    while(idx < target) {
+        U32 const forward = ZSTD_insertBt1(ms, base+idx, iend, target, mls, dictMode == ZSTD_extDict);
+        assert(idx < (U32)(idx + forward));
+        idx += forward;
+    }
+    assert((size_t)(ip - base) <= (size_t)(U32)(-1));
+    assert((size_t)(iend - base) <= (size_t)(U32)(-1));
+    ms->nextToUpdate = target;
+}
+
+void ZSTD_updateTree(ZSTD_MatchState_t* ms, const BYTE* ip, const BYTE* iend) {
+    ZSTD_updateTree_internal(ms, ip, iend, ms->cParams.minMatch, ZSTD_noDict);
+}
+
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32
+ZSTD_insertBtAndGetAllMatches (
+                ZSTD_match_t* matches,  /* store result (found matches) in this table (presumed large enough) */
+                ZSTD_MatchState_t* ms,
+                U32* nextToUpdate3,
+                const BYTE* const ip, const BYTE* const iLimit,
+                const ZSTD_dictMode_e dictMode,
+                const U32 rep[ZSTD_REP_NUM],
+                const U32 ll0,  /* tells if associated literal length is 0 or not. This value must be 0 or 1 */
+                const U32 lengthToBeat,
+                const U32 mls /* template */)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
+    const BYTE* const base = ms->window.base;
+    U32 const curr = (U32)(ip-base);
+    U32 const hashLog = cParams->hashLog;
+    U32 const minMatch = (mls==3) ? 3 : 4;
+    U32* const hashTable = ms->hashTable;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32 matchIndex  = hashTable[h];
+    U32* const bt   = ms->chainTable;
+    U32 const btLog = cParams->chainLog - 1;
+    U32 const btMask= (1U << btLog) - 1;
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const dictBase = ms->window.dictBase;
+    U32 const dictLimit = ms->window.dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    U32 const btLow = (btMask >= curr) ? 0 : curr - btMask;
+    U32 const windowLow = ZSTD_getLowestMatchIndex(ms, curr, cParams->windowLog);
+    U32 const matchLow = windowLow ? windowLow : 1;
+    U32* smallerPtr = bt + 2*(curr&btMask);
+    U32* largerPtr  = bt + 2*(curr&btMask) + 1;
+    U32 matchEndIdx = curr+8+1;   /* farthest referenced position of any match => detects repetitive patterns */
+    U32 dummy32;   /* to be nullified at the end */
+    U32 mnum = 0;
+    U32 nbCompares = 1U << cParams->searchLog;
+
+    const ZSTD_MatchState_t* dms    = dictMode == ZSTD_dictMatchState ? ms->dictMatchState : NULL;
+    const ZSTD_compressionParameters* const dmsCParams =
+                                      dictMode == ZSTD_dictMatchState ? &dms->cParams : NULL;
+    const BYTE* const dmsBase       = dictMode == ZSTD_dictMatchState ? dms->window.base : NULL;
+    const BYTE* const dmsEnd        = dictMode == ZSTD_dictMatchState ? dms->window.nextSrc : NULL;
+    U32         const dmsHighLimit  = dictMode == ZSTD_dictMatchState ? (U32)(dmsEnd - dmsBase) : 0;
+    U32         const dmsLowLimit   = dictMode == ZSTD_dictMatchState ? dms->window.lowLimit : 0;
+    U32         const dmsIndexDelta = dictMode == ZSTD_dictMatchState ? windowLow - dmsHighLimit : 0;
+    U32         const dmsHashLog    = dictMode == ZSTD_dictMatchState ? dmsCParams->hashLog : hashLog;
+    U32         const dmsBtLog      = dictMode == ZSTD_dictMatchState ? dmsCParams->chainLog - 1 : btLog;
+    U32         const dmsBtMask     = dictMode == ZSTD_dictMatchState ? (1U << dmsBtLog) - 1 : 0;
+    U32         const dmsBtLow      = dictMode == ZSTD_dictMatchState && dmsBtMask < dmsHighLimit - dmsLowLimit ? dmsHighLimit - dmsBtMask : dmsLowLimit;
+
+    size_t bestLength = lengthToBeat-1;
+    DEBUGLOG(8, "ZSTD_insertBtAndGetAllMatches: current=%u", curr);
+
+    /* check repCode */
+    assert(ll0 <= 1);   /* necessarily 1 or 0 */
+    {   U32 const lastR = ZSTD_REP_NUM + ll0;
+        U32 repCode;
+        for (repCode = ll0; repCode < lastR; repCode++) {
+            U32 const repOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+            U32 const repIndex = curr - repOffset;
+            U32 repLen = 0;
+            assert(curr >= dictLimit);
+            if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < curr-dictLimit) {  /* equivalent to `curr > repIndex >= dictLimit` */
+                /* We must validate the repcode offset because when we're using a dictionary the
+                 * valid offset range shrinks when the dictionary goes out of bounds.
+                 */
+                if ((repIndex >= windowLow) & (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(ip - repOffset, minMatch))) {
+                    repLen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repOffset, iLimit) + minMatch;
+                }
+            } else {  /* repIndex < dictLimit || repIndex >= curr */
+                const BYTE* const repMatch = dictMode == ZSTD_dictMatchState ?
+                                             dmsBase + repIndex - dmsIndexDelta :
+                                             dictBase + repIndex;
+                assert(curr >= windowLow);
+                if ( dictMode == ZSTD_extDict
+                  && ( ((repOffset-1) /*intentional overflow*/ < curr - windowLow)  /* equivalent to `curr > repIndex >= windowLow` */
+                     & (ZSTD_index_overlap_check(dictLimit, repIndex)) )
+                  && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
+                    repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dictEnd, prefixStart) + minMatch;
+                }
+                if (dictMode == ZSTD_dictMatchState
+                  && ( ((repOffset-1) /*intentional overflow*/ < curr - (dmsLowLimit + dmsIndexDelta))  /* equivalent to `curr > repIndex >= dmsLowLimit` */
+                     & (ZSTD_index_overlap_check(dictLimit, repIndex)) )
+                  && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
+                    repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dmsEnd, prefixStart) + minMatch;
+            }   }
+            /* save longer solution */
+            if (repLen > bestLength) {
+                DEBUGLOG(8, "found repCode %u (ll0:%u, offset:%u) of length %u",
+                            repCode, ll0, repOffset, repLen);
+                bestLength = repLen;
+                matches[mnum].off = REPCODE_TO_OFFBASE(repCode - ll0 + 1);  /* expect value between 1 and 3 */
+                matches[mnum].len = (U32)repLen;
+                mnum++;
+                if ( (repLen > sufficient_len)
+                   | (ip+repLen == iLimit) ) {  /* best possible */
+                    return mnum;
+    }   }   }   }
+
+    /* HC3 match finder */
+    if ((mls == 3) /*static*/ && (bestLength < mls)) {
+        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, nextToUpdate3, ip);
+        if ((matchIndex3 >= matchLow)
+          & (curr - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
+            size_t mlen;
+            if ((dictMode == ZSTD_noDict) /*static*/ || (dictMode == ZSTD_dictMatchState) /*static*/ || (matchIndex3 >= dictLimit)) {
+                const BYTE* const match = base + matchIndex3;
+                mlen = ZSTD_count(ip, match, iLimit);
+            } else {
+                const BYTE* const match = dictBase + matchIndex3;
+                mlen = ZSTD_count_2segments(ip, match, iLimit, dictEnd, prefixStart);
+            }
+
+            /* save best solution */
+            if (mlen >= mls /* == 3 > bestLength */) {
+                DEBUGLOG(8, "found small match with hlog3, of length %u",
+                            (U32)mlen);
+                bestLength = mlen;
+                assert(curr > matchIndex3);
+                assert(mnum==0);  /* no prior solution */
+                matches[0].off = OFFSET_TO_OFFBASE(curr - matchIndex3);
+                matches[0].len = (U32)mlen;
+                mnum = 1;
+                if ( (mlen > sufficient_len) |
+                     (ip+mlen == iLimit) ) {  /* best possible length */
+                    ms->nextToUpdate = curr+1;  /* skip insertion */
+                    return 1;
+        }   }   }
+        /* no dictMatchState lookup: dicts don't have a populated HC3 table */
+    }  /* if (mls == 3) */
+
+    hashTable[h] = curr;   /* Update Hash Table */
+
+    for (; nbCompares && (matchIndex >= matchLow); --nbCompares) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        const BYTE* match;
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(curr > matchIndex);
+
+        if ((dictMode == ZSTD_noDict) || (dictMode == ZSTD_dictMatchState) || (matchIndex+matchLength >= dictLimit)) {
+            assert(matchIndex+matchLength >= dictLimit);  /* ensure the condition is correct when !extDict */
+            match = base + matchIndex;
+            if (matchIndex >= dictLimit) assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iLimit);
+        } else {
+            match = dictBase + matchIndex;
+            assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* prepare for match[matchLength] read */
+        }
+
+        if (matchLength > bestLength) {
+            DEBUGLOG(8, "found match of length %u at distance %u (offBase=%u)",
+                    (U32)matchLength, curr - matchIndex, OFFSET_TO_OFFBASE(curr - matchIndex));
+            assert(matchEndIdx > matchIndex);
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
+            bestLength = matchLength;
+            matches[mnum].off = OFFSET_TO_OFFBASE(curr - matchIndex);
+            matches[mnum].len = (U32)matchLength;
+            mnum++;
+            if ( (matchLength > ZSTD_OPT_NUM)
+               | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
+                if (dictMode == ZSTD_dictMatchState) nbCompares = 0; /* break should also skip searching dms */
+                break; /* drop, to preserve bt consistency (miss a little bit of compression) */
+        }   }
+
+        if (match[matchLength] < ip[matchLength]) {
+            /* match smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new candidate => larger than match, which was smaller than current */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous, closer to current */
+        } else {
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+
+    assert(nbCompares <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
+    if (dictMode == ZSTD_dictMatchState && nbCompares) {
+        size_t const dmsH = ZSTD_hashPtr(ip, dmsHashLog, mls);
+        U32 dictMatchIndex = dms->hashTable[dmsH];
+        const U32* const dmsBt = dms->chainTable;
+        commonLengthSmaller = commonLengthLarger = 0;
+        for (; nbCompares && (dictMatchIndex > dmsLowLimit); --nbCompares) {
+            const U32* const nextPtr = dmsBt + 2*(dictMatchIndex & dmsBtMask);
+            size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+            const BYTE* match = dmsBase + dictMatchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dmsEnd, prefixStart);
+            if (dictMatchIndex+matchLength >= dmsHighLimit)
+                match = base + dictMatchIndex + dmsIndexDelta;   /* to prepare for next usage of match[matchLength] */
+
+            if (matchLength > bestLength) {
+                matchIndex = dictMatchIndex + dmsIndexDelta;
+                DEBUGLOG(8, "found dms match of length %u at distance %u (offBase=%u)",
+                        (U32)matchLength, curr - matchIndex, OFFSET_TO_OFFBASE(curr - matchIndex));
+                if (matchLength > matchEndIdx - matchIndex)
+                    matchEndIdx = matchIndex + (U32)matchLength;
+                bestLength = matchLength;
+                matches[mnum].off = OFFSET_TO_OFFBASE(curr - matchIndex);
+                matches[mnum].len = (U32)matchLength;
+                mnum++;
+                if ( (matchLength > ZSTD_OPT_NUM)
+                   | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
+                    break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+            }   }
+
+            if (dictMatchIndex <= dmsBtLow) { break; }   /* beyond tree size, stop the search */
+            if (match[matchLength] < ip[matchLength]) {
+                commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+                dictMatchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            } else {
+                /* match is larger than current */
+                commonLengthLarger = matchLength;
+                dictMatchIndex = nextPtr[0];
+    }   }   }  /* if (dictMode == ZSTD_dictMatchState) */
+
+    assert(matchEndIdx > curr+8);
+    ms->nextToUpdate = matchEndIdx - 8;  /* skip repetitive patterns */
+    return mnum;
+}
+
+typedef U32 (*ZSTD_getAllMatchesFn)(
+    ZSTD_match_t*,
+    ZSTD_MatchState_t*,
+    U32*,
+    const BYTE*,
+    const BYTE*,
+    const U32 rep[ZSTD_REP_NUM],
+    U32 const ll0,
+    U32 const lengthToBeat);
+
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+U32 ZSTD_btGetAllMatches_internal(
+        ZSTD_match_t* matches,
+        ZSTD_MatchState_t* ms,
+        U32* nextToUpdate3,
+        const BYTE* ip,
+        const BYTE* const iHighLimit,
+        const U32 rep[ZSTD_REP_NUM],
+        U32 const ll0,
+        U32 const lengthToBeat,
+        const ZSTD_dictMode_e dictMode,
+        const U32 mls)
+{
+    assert(BOUNDED(3, ms->cParams.minMatch, 6) == mls);
+    DEBUGLOG(8, "ZSTD_BtGetAllMatches(dictMode=%d, mls=%u)", (int)dictMode, mls);
+    if (ip < ms->window.base + ms->nextToUpdate)
+        return 0;   /* skipped area */
+    ZSTD_updateTree_internal(ms, ip, iHighLimit, mls, dictMode);
+    return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, mls);
+}
+
+#define ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, mls) ZSTD_btGetAllMatches_##dictMode##_##mls
+
+#define GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, mls)            \
+    static U32 ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, mls)(      \
+            ZSTD_match_t* matches,                             \
+            ZSTD_MatchState_t* ms,                             \
+            U32* nextToUpdate3,                                \
+            const BYTE* ip,                                    \
+            const BYTE* const iHighLimit,                      \
+            const U32 rep[ZSTD_REP_NUM],                       \
+            U32 const ll0,                                     \
+            U32 const lengthToBeat)                            \
+    {                                                          \
+        return ZSTD_btGetAllMatches_internal(                  \
+                matches, ms, nextToUpdate3, ip, iHighLimit,    \
+                rep, ll0, lengthToBeat, ZSTD_##dictMode, mls); \
+    }
+
+#define GEN_ZSTD_BT_GET_ALL_MATCHES(dictMode)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 3)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 4)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 5)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 6)
+
+GEN_ZSTD_BT_GET_ALL_MATCHES(noDict)
+GEN_ZSTD_BT_GET_ALL_MATCHES(extDict)
+GEN_ZSTD_BT_GET_ALL_MATCHES(dictMatchState)
+
+#define ZSTD_BT_GET_ALL_MATCHES_ARRAY(dictMode)  \
+    {                                            \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 3), \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 4), \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 5), \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 6)  \
+    }
+
+static ZSTD_getAllMatchesFn
+ZSTD_selectBtGetAllMatches(ZSTD_MatchState_t const* ms, ZSTD_dictMode_e const dictMode)
+{
+    ZSTD_getAllMatchesFn const getAllMatchesFns[3][4] = {
+        ZSTD_BT_GET_ALL_MATCHES_ARRAY(noDict),
+        ZSTD_BT_GET_ALL_MATCHES_ARRAY(extDict),
+        ZSTD_BT_GET_ALL_MATCHES_ARRAY(dictMatchState)
+    };
+    U32 const mls = BOUNDED(3, ms->cParams.minMatch, 6);
+    assert((U32)dictMode < 3);
+    assert(mls - 3 < 4);
+    return getAllMatchesFns[(int)dictMode][mls - 3];
+}
+
+/*************************
+*  LDM helper functions  *
+*************************/
+
+/* Struct containing info needed to make decision about ldm inclusion */
+typedef struct {
+    RawSeqStore_t seqStore;   /* External match candidates store for this block */
+    U32 startPosInBlock;      /* Start position of the current match candidate */
+    U32 endPosInBlock;        /* End position of the current match candidate */
+    U32 offset;               /* Offset of the match candidate */
+} ZSTD_optLdm_t;
+
+/* ZSTD_optLdm_skipRawSeqStoreBytes():
+ * Moves forward in @rawSeqStore by @nbBytes,
+ * which will update the fields 'pos' and 'posInSequence'.
+ */
+static void ZSTD_optLdm_skipRawSeqStoreBytes(RawSeqStore_t* rawSeqStore, size_t nbBytes)
+{
+    U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
+    while (currPos && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
+        if (currPos >= currSeq.litLength + currSeq.matchLength) {
+            currPos -= currSeq.litLength + currSeq.matchLength;
+            rawSeqStore->pos++;
+        } else {
+            rawSeqStore->posInSequence = currPos;
+            break;
+        }
+    }
+    if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
+        rawSeqStore->posInSequence = 0;
+    }
+}
+
+/* ZSTD_opt_getNextMatchAndUpdateSeqStore():
+ * Calculates the beginning and end of the next match in the current block.
+ * Updates 'pos' and 'posInSequence' of the ldmSeqStore.
+ */
+static void
+ZSTD_opt_getNextMatchAndUpdateSeqStore(ZSTD_optLdm_t* optLdm, U32 currPosInBlock,
+                                       U32 blockBytesRemaining)
+{
+    rawSeq currSeq;
+    U32 currBlockEndPos;
+    U32 literalsBytesRemaining;
+    U32 matchBytesRemaining;
+
+    /* Setting match end position to MAX to ensure we never use an LDM during this block */
+    if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
+        optLdm->startPosInBlock = UINT_MAX;
+        optLdm->endPosInBlock = UINT_MAX;
+        return;
+    }
+    /* Calculate appropriate bytes left in matchLength and litLength
+     * after adjusting based on ldmSeqStore->posInSequence */
+    currSeq = optLdm->seqStore.seq[optLdm->seqStore.pos];
+    assert(optLdm->seqStore.posInSequence <= currSeq.litLength + currSeq.matchLength);
+    currBlockEndPos = currPosInBlock + blockBytesRemaining;
+    literalsBytesRemaining = (optLdm->seqStore.posInSequence < currSeq.litLength) ?
+            currSeq.litLength - (U32)optLdm->seqStore.posInSequence :
+            0;
+    matchBytesRemaining = (literalsBytesRemaining == 0) ?
+            currSeq.matchLength - ((U32)optLdm->seqStore.posInSequence - currSeq.litLength) :
+            currSeq.matchLength;
+
+    /* If there are more literal bytes than bytes remaining in block, no ldm is possible */
+    if (literalsBytesRemaining >= blockBytesRemaining) {
+        optLdm->startPosInBlock = UINT_MAX;
+        optLdm->endPosInBlock = UINT_MAX;
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, blockBytesRemaining);
+        return;
+    }
+
+    /* Matches may be < minMatch by this process. In that case, we will reject them
+       when we are deciding whether or not to add the ldm */
+    optLdm->startPosInBlock = currPosInBlock + literalsBytesRemaining;
+    optLdm->endPosInBlock = optLdm->startPosInBlock + matchBytesRemaining;
+    optLdm->offset = currSeq.offset;
+
+    if (optLdm->endPosInBlock > currBlockEndPos) {
+        /* Match ends after the block ends, we can't use the whole match */
+        optLdm->endPosInBlock = currBlockEndPos;
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, currBlockEndPos - currPosInBlock);
+    } else {
+        /* Consume nb of bytes equal to size of sequence left */
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, literalsBytesRemaining + matchBytesRemaining);
+    }
+}
+
+/* ZSTD_optLdm_maybeAddMatch():
+ * Adds a match if it's long enough,
+ * based on it's 'matchStartPosInBlock' and 'matchEndPosInBlock',
+ * into 'matches'. Maintains the correct ordering of 'matches'.
+ */
+static void ZSTD_optLdm_maybeAddMatch(ZSTD_match_t* matches, U32* nbMatches,
+                                      const ZSTD_optLdm_t* optLdm, U32 currPosInBlock,
+                                      U32 minMatch)
+{
+    U32 const posDiff = currPosInBlock - optLdm->startPosInBlock;
+    /* Note: ZSTD_match_t actually contains offBase and matchLength (before subtracting MINMATCH) */
+    U32 const candidateMatchLength = optLdm->endPosInBlock - optLdm->startPosInBlock - posDiff;
+
+    /* Ensure that current block position is not outside of the match */
+    if (currPosInBlock < optLdm->startPosInBlock
+      || currPosInBlock >= optLdm->endPosInBlock
+      || candidateMatchLength < minMatch) {
+        return;
+    }
+
+    if (*nbMatches == 0 || ((candidateMatchLength > matches[*nbMatches-1].len) && *nbMatches < ZSTD_OPT_NUM)) {
+        U32 const candidateOffBase = OFFSET_TO_OFFBASE(optLdm->offset);
+        DEBUGLOG(6, "ZSTD_optLdm_maybeAddMatch(): Adding ldm candidate match (offBase: %u matchLength %u) at block position=%u",
+                 candidateOffBase, candidateMatchLength, currPosInBlock);
+        matches[*nbMatches].len = candidateMatchLength;
+        matches[*nbMatches].off = candidateOffBase;
+        (*nbMatches)++;
+    }
+}
+
+/* ZSTD_optLdm_processMatchCandidate():
+ * Wrapper function to update ldm seq store and call ldm functions as necessary.
+ */
+static void
+ZSTD_optLdm_processMatchCandidate(ZSTD_optLdm_t* optLdm,
+                                  ZSTD_match_t* matches, U32* nbMatches,
+                                  U32 currPosInBlock, U32 remainingBytes,
+                                  U32 minMatch)
+{
+    if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
+        return;
+    }
+
+    if (currPosInBlock >= optLdm->endPosInBlock) {
+        if (currPosInBlock > optLdm->endPosInBlock) {
+            /* The position at which ZSTD_optLdm_processMatchCandidate() is called is not necessarily
+             * at the end of a match from the ldm seq store, and will often be some bytes
+             * over beyond matchEndPosInBlock. As such, we need to correct for these "overshoots"
+             */
+            U32 const posOvershoot = currPosInBlock - optLdm->endPosInBlock;
+            ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, posOvershoot);
+        }
+        ZSTD_opt_getNextMatchAndUpdateSeqStore(optLdm, currPosInBlock, remainingBytes);
+    }
+    ZSTD_optLdm_maybeAddMatch(matches, nbMatches, optLdm, currPosInBlock, minMatch);
+}
+
+
+/*-*******************************
+*  Optimal parser
+*********************************/
+
+#if 0 /* debug */
+
+static void
+listStats(const U32* table, int lastEltID)
+{
+    int const nbElts = lastEltID + 1;
+    int enb;
+    for (enb=0; enb < nbElts; enb++) {
+        (void)table;
+        /* RAWLOG(2, "%3i:%3i,  ", enb, table[enb]); */
+        RAWLOG(2, "%4i,", table[enb]);
+    }
+    RAWLOG(2, " \n");
+}
+
+#endif
+
+#define LIT_PRICE(_p) (int)ZSTD_rawLiteralsCost(_p, 1, optStatePtr, optLevel)
+#define LL_PRICE(_l) (int)ZSTD_litLengthPrice(_l, optStatePtr, optLevel)
+#define LL_INCPRICE(_l) (LL_PRICE(_l) - LL_PRICE(_l-1))
+
+FORCE_INLINE_TEMPLATE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t
+ZSTD_compressBlock_opt_generic(ZSTD_MatchState_t* ms,
+                               SeqStore_t* seqStore,
+                               U32 rep[ZSTD_REP_NUM],
+                         const void* src, size_t srcSize,
+                         const int optLevel,
+                         const ZSTD_dictMode_e dictMode)
+{
+    optState_t* const optStatePtr = &ms->opt;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const prefixStart = base + ms->window.dictLimit;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+
+    ZSTD_getAllMatchesFn getAllMatches = ZSTD_selectBtGetAllMatches(ms, dictMode);
+
+    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
+    U32 const minMatch = (cParams->minMatch == 3) ? 3 : 4;
+    U32 nextToUpdate3 = ms->nextToUpdate;
+
+    ZSTD_optimal_t* const opt = optStatePtr->priceTable;
+    ZSTD_match_t* const matches = optStatePtr->matchTable;
+    ZSTD_optimal_t lastStretch;
+    ZSTD_optLdm_t optLdm;
+
+    ZSTD_memset(&lastStretch, 0, sizeof(ZSTD_optimal_t));
+
+    optLdm.seqStore = ms->ldmSeqStore ? *ms->ldmSeqStore : kNullRawSeqStore;
+    optLdm.endPosInBlock = optLdm.startPosInBlock = optLdm.offset = 0;
+    ZSTD_opt_getNextMatchAndUpdateSeqStore(&optLdm, (U32)(ip-istart), (U32)(iend-ip));
+
+    /* init */
+    DEBUGLOG(5, "ZSTD_compressBlock_opt_generic: current=%u, prefix=%u, nextToUpdate=%u",
+                (U32)(ip - base), ms->window.dictLimit, ms->nextToUpdate);
+    assert(optLevel <= 2);
+    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize, optLevel);
+    ip += (ip==prefixStart);
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        U32 cur, last_pos = 0;
+
+        /* find first match */
+        {   U32 const litlen = (U32)(ip - anchor);
+            U32 const ll0 = !litlen;
+            U32 nbMatches = getAllMatches(matches, ms, &nextToUpdate3, ip, iend, rep, ll0, minMatch);
+            ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
+                                              (U32)(ip-istart), (U32)(iend-ip),
+                                              minMatch);
+            if (!nbMatches) {
+                DEBUGLOG(8, "no match found at cPos %u", (unsigned)(ip-istart));
+                ip++;
+                continue;
+            }
+
+            /* Match found: let's store this solution, and eventually find more candidates.
+             * During this forward pass, @opt is used to store stretches,
+             * defined as "a match followed by N literals".
+             * Note how this is different from a Sequence, which is "N literals followed by a match".
+             * Storing stretches allows us to store different match predecessors
+             * for each literal position part of a literals run. */
+
+            /* initialize opt[0] */
+            opt[0].mlen = 0;  /* there are only literals so far */
+            opt[0].litlen = litlen;
+            /* No need to include the actual price of the literals before the first match
+             * because it is static for the duration of the forward pass, and is included
+             * in every subsequent price. But, we include the literal length because
+             * the cost variation of litlen depends on the value of litlen.
+             */
+            opt[0].price = LL_PRICE(litlen);
+            ZSTD_STATIC_ASSERT(sizeof(opt[0].rep[0]) == sizeof(rep[0]));
+            ZSTD_memcpy(&opt[0].rep, rep, sizeof(opt[0].rep));
+
+            /* large match -> immediate encoding */
+            {   U32 const maxML = matches[nbMatches-1].len;
+                U32 const maxOffBase = matches[nbMatches-1].off;
+                DEBUGLOG(6, "found %u matches of maxLength=%u and maxOffBase=%u at cPos=%u => start new series",
+                            nbMatches, maxML, maxOffBase, (U32)(ip-prefixStart));
+
+                if (maxML > sufficient_len) {
+                    lastStretch.litlen = 0;
+                    lastStretch.mlen = maxML;
+                    lastStretch.off = maxOffBase;
+                    DEBUGLOG(6, "large match (%u>%u) => immediate encoding",
+                                maxML, sufficient_len);
+                    cur = 0;
+                    last_pos = maxML;
+                    goto _shortestPath;
+            }   }
+
+            /* set prices for first matches starting position == 0 */
+            assert(opt[0].price >= 0);
+            {   U32 pos;
+                U32 matchNb;
+                for (pos = 1; pos < minMatch; pos++) {
+                    opt[pos].price = ZSTD_MAX_PRICE;
+                    opt[pos].mlen = 0;
+                    opt[pos].litlen = litlen + pos;
+                }
+                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
+                    U32 const offBase = matches[matchNb].off;
+                    U32 const end = matches[matchNb].len;
+                    for ( ; pos <= end ; pos++ ) {
+                        int const matchPrice = (int)ZSTD_getMatchPrice(offBase, pos, optStatePtr, optLevel);
+                        int const sequencePrice = opt[0].price + matchPrice;
+                        DEBUGLOG(7, "rPos:%u => set initial price : %.2f",
+                                    pos, ZSTD_fCost(sequencePrice));
+                        opt[pos].mlen = pos;
+                        opt[pos].off = offBase;
+                        opt[pos].litlen = 0; /* end of match */
+                        opt[pos].price = sequencePrice + LL_PRICE(0);
+                    }
+                }
+                last_pos = pos-1;
+                opt[pos].price = ZSTD_MAX_PRICE;
+            }
+        }
+
+        /* check further positions */
+        for (cur = 1; cur <= last_pos; cur++) {
+            const BYTE* const inr = ip + cur;
+            assert(cur <= ZSTD_OPT_NUM);
+            DEBUGLOG(7, "cPos:%i==rPos:%u", (int)(inr-istart), cur);
+
+            /* Fix current position with one literal if cheaper */
+            {   U32 const litlen = opt[cur-1].litlen + 1;
+                int const price = opt[cur-1].price
+                                + LIT_PRICE(ip+cur-1)
+                                + LL_INCPRICE(litlen);
+                assert(price < 1000000000); /* overflow check */
+                if (price <= opt[cur].price) {
+                    ZSTD_optimal_t const prevMatch = opt[cur];
+                    DEBUGLOG(7, "cPos:%i==rPos:%u : better price (%.2f<=%.2f) using literal (ll==%u) (hist:%u,%u,%u)",
+                                (int)(inr-istart), cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price), litlen,
+                                opt[cur-1].rep[0], opt[cur-1].rep[1], opt[cur-1].rep[2]);
+                    opt[cur] = opt[cur-1];
+                    opt[cur].litlen = litlen;
+                    opt[cur].price = price;
+                    if ( (optLevel >= 1) /* additional check only for higher modes */
+                      && (prevMatch.litlen == 0) /* replace a match */
+                      && (LL_INCPRICE(1) < 0) /* ll1 is cheaper than ll0 */
+                      && LIKELY(ip + cur < iend)
+                    ) {
+                        /* check next position, in case it would be cheaper */
+                        int with1literal = prevMatch.price + LIT_PRICE(ip+cur) + LL_INCPRICE(1);
+                        int withMoreLiterals = price + LIT_PRICE(ip+cur) + LL_INCPRICE(litlen+1);
+                        DEBUGLOG(7, "then at next rPos %u : match+1lit %.2f vs %ulits %.2f",
+                                cur+1, ZSTD_fCost(with1literal), litlen+1, ZSTD_fCost(withMoreLiterals));
+                        if ( (with1literal < withMoreLiterals)
+                          && (with1literal < opt[cur+1].price) ) {
+                            /* update offset history - before it disappears */
+                            U32 const prev = cur - prevMatch.mlen;
+                            Repcodes_t const newReps = ZSTD_newRep(opt[prev].rep, prevMatch.off, opt[prev].litlen==0);
+                            assert(cur >= prevMatch.mlen);
+                            DEBUGLOG(7, "==> match+1lit is cheaper (%.2f < %.2f) (hist:%u,%u,%u) !",
+                                        ZSTD_fCost(with1literal), ZSTD_fCost(withMoreLiterals),
+                                        newReps.rep[0], newReps.rep[1], newReps.rep[2] );
+                            opt[cur+1] = prevMatch;  /* mlen & offbase */
+                            ZSTD_memcpy(opt[cur+1].rep, &newReps, sizeof(Repcodes_t));
+                            opt[cur+1].litlen = 1;
+                            opt[cur+1].price = with1literal;
+                            if (last_pos < cur+1) last_pos = cur+1;
+                        }
+                    }
+                } else {
+                    DEBUGLOG(7, "cPos:%i==rPos:%u : literal would cost more (%.2f>%.2f)",
+                                (int)(inr-istart), cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price));
+                }
+            }
+
+            /* Offset history is not updated during match comparison.
+             * Do it here, now that the match is selected and confirmed.
+             */
+            ZSTD_STATIC_ASSERT(sizeof(opt[cur].rep) == sizeof(Repcodes_t));
+            assert(cur >= opt[cur].mlen);
+            if (opt[cur].litlen == 0) {
+                /* just finished a match => alter offset history */
+                U32 const prev = cur - opt[cur].mlen;
+                Repcodes_t const newReps = ZSTD_newRep(opt[prev].rep, opt[cur].off, opt[prev].litlen==0);
+                ZSTD_memcpy(opt[cur].rep, &newReps, sizeof(Repcodes_t));
+            }
+
+            /* last match must start at a minimum distance of 8 from oend */
+            if (inr > ilimit) continue;
+
+            if (cur == last_pos) break;
+
+            if ( (optLevel==0) /*static_test*/
+              && (opt[cur+1].price <= opt[cur].price + (BITCOST_MULTIPLIER/2)) ) {
+                DEBUGLOG(7, "skip current position : next rPos(%u) price is cheaper", cur+1);
+                continue;  /* skip unpromising positions; about ~+6% speed, -0.01 ratio */
+            }
+
+            assert(opt[cur].price >= 0);
+            {   U32 const ll0 = (opt[cur].litlen == 0);
+                int const previousPrice = opt[cur].price;
+                int const basePrice = previousPrice + LL_PRICE(0);
+                U32 nbMatches = getAllMatches(matches, ms, &nextToUpdate3, inr, iend, opt[cur].rep, ll0, minMatch);
+                U32 matchNb;
+
+                ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
+                                                  (U32)(inr-istart), (U32)(iend-inr),
+                                                  minMatch);
+
+                if (!nbMatches) {
+                    DEBUGLOG(7, "rPos:%u : no match found", cur);
+                    continue;
+                }
+
+                {   U32 const longestML = matches[nbMatches-1].len;
+                    DEBUGLOG(7, "cPos:%i==rPos:%u, found %u matches, of longest ML=%u",
+                                (int)(inr-istart), cur, nbMatches, longestML);
+
+                    if ( (longestML > sufficient_len)
+                      || (cur + longestML >= ZSTD_OPT_NUM)
+                      || (ip + cur + longestML >= iend) ) {
+                        lastStretch.mlen = longestML;
+                        lastStretch.off = matches[nbMatches-1].off;
+                        lastStretch.litlen = 0;
+                        last_pos = cur + longestML;
+                        goto _shortestPath;
+                }   }
+
+                /* set prices using matches found at position == cur */
+                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
+                    U32 const offset = matches[matchNb].off;
+                    U32 const lastML = matches[matchNb].len;
+                    U32 const startML = (matchNb>0) ? matches[matchNb-1].len+1 : minMatch;
+                    U32 mlen;
+
+                    DEBUGLOG(7, "testing match %u => offBase=%4u, mlen=%2u, llen=%2u",
+                                matchNb, matches[matchNb].off, lastML, opt[cur].litlen);
+
+                    for (mlen = lastML; mlen >= startML; mlen--) {  /* scan downward */
+                        U32 const pos = cur + mlen;
+                        int const price = basePrice + (int)ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
+
+                        if ((pos > last_pos) || (price < opt[pos].price)) {
+                            DEBUGLOG(7, "rPos:%u (ml=%2u) => new better price (%.2f<%.2f)",
+                                        pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
+                            while (last_pos < pos) {
+                                /* fill empty positions, for future comparisons */
+                                last_pos++;
+                                opt[last_pos].price = ZSTD_MAX_PRICE;
+                                opt[last_pos].litlen = !0;  /* just needs to be != 0, to mean "not an end of match" */
+                            }
+                            opt[pos].mlen = mlen;
+                            opt[pos].off = offset;
+                            opt[pos].litlen = 0;
+                            opt[pos].price = price;
+                        } else {
+                            DEBUGLOG(7, "rPos:%u (ml=%2u) => new price is worse (%.2f>=%.2f)",
+                                        pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
+                            if (optLevel==0) break;  /* early update abort; gets ~+10% speed for about -0.01 ratio loss */
+                        }
+            }   }   }
+            opt[last_pos+1].price = ZSTD_MAX_PRICE;
+        }  /* for (cur = 1; cur <= last_pos; cur++) */
+
+        lastStretch = opt[last_pos];
+        assert(cur >= lastStretch.mlen);
+        cur = last_pos - lastStretch.mlen;
+
+_shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
+        assert(opt[0].mlen == 0);
+        assert(last_pos >= lastStretch.mlen);
+        assert(cur == last_pos - lastStretch.mlen);
+
+        if (lastStretch.mlen==0) {
+            /* no solution : all matches have been converted into literals */
+            assert(lastStretch.litlen == (ip - anchor) + last_pos);
+            ip += last_pos;
+            continue;
+        }
+        assert(lastStretch.off > 0);
+
+        /* Update offset history */
+        if (lastStretch.litlen == 0) {
+            /* finishing on a match : update offset history */
+            Repcodes_t const reps = ZSTD_newRep(opt[cur].rep, lastStretch.off, opt[cur].litlen==0);
+            ZSTD_memcpy(rep, &reps, sizeof(Repcodes_t));
+        } else {
+            ZSTD_memcpy(rep, lastStretch.rep, sizeof(Repcodes_t));
+            assert(cur >= lastStretch.litlen);
+            cur -= lastStretch.litlen;
+        }
+
+        /* Let's write the shortest path solution.
+         * It is stored in @opt in reverse order,
+         * starting from @storeEnd (==cur+2),
+         * effectively partially @opt overwriting.
+         * Content is changed too:
+         * - So far, @opt stored stretches, aka a match followed by literals
+         * - Now, it will store sequences, aka literals followed by a match
+         */
+        {   U32 const storeEnd = cur + 2;
+            U32 storeStart = storeEnd;
+            U32 stretchPos = cur;
+
+            DEBUGLOG(6, "start reverse traversal (last_pos:%u, cur:%u)",
+                        last_pos, cur); (void)last_pos;
+            assert(storeEnd < ZSTD_OPT_SIZE);
+            DEBUGLOG(6, "last stretch copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
+                        storeEnd, lastStretch.litlen, lastStretch.mlen, lastStretch.off);
+            if (lastStretch.litlen > 0) {
+                /* last "sequence" is unfinished: just a bunch of literals */
+                opt[storeEnd].litlen = lastStretch.litlen;
+                opt[storeEnd].mlen = 0;
+                storeStart = storeEnd-1;
+                opt[storeStart] = lastStretch;
+            } {
+                opt[storeEnd] = lastStretch;  /* note: litlen will be fixed */
+                storeStart = storeEnd;
+            }
+            while (1) {
+                ZSTD_optimal_t nextStretch = opt[stretchPos];
+                opt[storeStart].litlen = nextStretch.litlen;
+                DEBUGLOG(6, "selected sequence (llen=%u,mlen=%u,ofc=%u)",
+                            opt[storeStart].litlen, opt[storeStart].mlen, opt[storeStart].off);
+                if (nextStretch.mlen == 0) {
+                    /* reaching beginning of segment */
+                    break;
+                }
+                storeStart--;
+                opt[storeStart] = nextStretch; /* note: litlen will be fixed */
+                assert(nextStretch.litlen + nextStretch.mlen <= stretchPos);
+                stretchPos -= nextStretch.litlen + nextStretch.mlen;
+            }
+
+            /* save sequences */
+            DEBUGLOG(6, "sending selected sequences into seqStore");
+            {   U32 storePos;
+                for (storePos=storeStart; storePos <= storeEnd; storePos++) {
+                    U32 const llen = opt[storePos].litlen;
+                    U32 const mlen = opt[storePos].mlen;
+                    U32 const offBase = opt[storePos].off;
+                    U32 const advance = llen + mlen;
+                    DEBUGLOG(6, "considering seq starting at %i, llen=%u, mlen=%u",
+                                (int)(anchor - istart), (unsigned)llen, (unsigned)mlen);
+
+                    if (mlen==0) {  /* only literals => must be last "sequence", actually starting a new stream of sequences */
+                        assert(storePos == storeEnd);   /* must be last sequence */
+                        ip = anchor + llen;     /* last "sequence" is a bunch of literals => don't progress anchor */
+                        continue;   /* will finish */
+                    }
+
+                    assert(anchor + llen <= iend);
+                    ZSTD_updateStats(optStatePtr, llen, anchor, offBase, mlen);
+                    ZSTD_storeSeq(seqStore, llen, anchor, iend, offBase, mlen);
+                    anchor += advance;
+                    ip = anchor;
+            }   }
+            DEBUGLOG(7, "new offset history : %u, %u, %u", rep[0], rep[1], rep[2]);
+
+            /* update all costs */
+            ZSTD_setBasePrices(optStatePtr, optLevel);
+        }
+    }   /* while (ip < ilimit) */
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+#endif /* build exclusions */
+
+#ifndef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
+static size_t ZSTD_compressBlock_opt0(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize, const ZSTD_dictMode_e dictMode)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /* optLevel */, dictMode);
+}
+#endif
+
+#ifndef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
+static size_t ZSTD_compressBlock_opt2(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize, const ZSTD_dictMode_e dictMode)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /* optLevel */, dictMode);
+}
+#endif
+
+#ifndef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_btopt(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressBlock_btopt");
+    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
+}
+#endif
+
+
+
+
+#ifndef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
+/* ZSTD_initStats_ultra():
+ * make a first compression pass, just to seed stats with more accurate starting values.
+ * only works on first block, with no dictionary and no ldm.
+ * this function cannot error out, its narrow contract must be respected.
+ */
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_initStats_ultra(ZSTD_MatchState_t* ms,
+                          SeqStore_t* seqStore,
+                          U32 rep[ZSTD_REP_NUM],
+                    const void* src, size_t srcSize)
+{
+    U32 tmpRep[ZSTD_REP_NUM];  /* updated rep codes will sink here */
+    ZSTD_memcpy(tmpRep, rep, sizeof(tmpRep));
+
+    DEBUGLOG(4, "ZSTD_initStats_ultra (srcSize=%zu)", srcSize);
+    assert(ms->opt.litLengthSum == 0);    /* first block */
+    assert(seqStore->sequences == seqStore->sequencesStart);   /* no ldm */
+    assert(ms->window.dictLimit == ms->window.lowLimit);   /* no dictionary */
+    assert(ms->window.dictLimit - ms->nextToUpdate <= 1);  /* no prefix (note: intentional overflow, defined as 2-complement) */
+
+    ZSTD_compressBlock_opt2(ms, seqStore, tmpRep, src, srcSize, ZSTD_noDict);   /* generate stats into ms->opt*/
+
+    /* invalidate first scan from history, only keep entropy stats */
+    ZSTD_resetSeqStore(seqStore);
+    ms->window.base -= srcSize;
+    ms->window.dictLimit += (U32)srcSize;
+    ms->window.lowLimit = ms->window.dictLimit;
+    ms->nextToUpdate = ms->window.dictLimit;
+
+}
+
+size_t ZSTD_compressBlock_btultra(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressBlock_btultra (srcSize=%zu)", srcSize);
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_btultra2(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    U32 const curr = (U32)((const BYTE*)src - ms->window.base);
+    DEBUGLOG(5, "ZSTD_compressBlock_btultra2 (srcSize=%zu)", srcSize);
+
+    /* 2-passes strategy:
+     * this strategy makes a first pass over first block to collect statistics
+     * in order to seed next round's statistics with it.
+     * After 1st pass, function forgets history, and starts a new block.
+     * Consequently, this can only work if no data has been previously loaded in tables,
+     * aka, no dictionary, no prefix, no ldm preprocessing.
+     * The compression ratio gain is generally small (~0.5% on first block),
+     * the cost is 2x cpu time on first block. */
+    assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
+    if ( (ms->opt.litLengthSum==0)   /* first block */
+      && (seqStore->sequences == seqStore->sequencesStart)  /* no ldm */
+      && (ms->window.dictLimit == ms->window.lowLimit)   /* no dictionary */
+      && (curr == ms->window.dictLimit)    /* start of frame, nothing already loaded nor skipped */
+      && (srcSize > ZSTD_PREDEF_THRESHOLD) /* input large enough to not employ default stats */
+      ) {
+        ZSTD_initStats_ultra(ms, seqStore, rep, src, srcSize);
+    }
+
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
+}
+#endif
+
+#ifndef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_btopt_dictMatchState(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_btopt_extDict(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_extDict);
+}
+#endif
+
+#ifndef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
+size_t ZSTD_compressBlock_btultra_dictMatchState(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_dictMatchState);
+}
+
+size_t ZSTD_compressBlock_btultra_extDict(
+        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_extDict);
+}
+#endif
+
+/* note : no btultra2 variant for extDict nor dictMatchState,
+ * because btultra2 is not meant to work with dictionaries
+ * and is only specific for the first block (no prefix) */
diff --git a/lib/compress/zstdmt_compress.c b/lib/compress/zstdmt_compress.c
new file mode 100644
index 0000000..0f1fe6d
--- /dev/null
+++ b/lib/compress/zstdmt_compress.c
@@ -0,0 +1,1923 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/* ======   Compiler specifics   ====== */
+#if defined(_MSC_VER)
+#  pragma warning(disable : 4204)   /* disable: C4204: non-constant aggregate initializer */
+#endif
+
+
+/* ======   Dependencies   ====== */
+#include "../common/allocations.h" /* ZSTD_customMalloc, ZSTD_customCalloc, ZSTD_customFree */
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memset, INT_MAX, UINT_MAX */
+#include "../common/mem.h"         /* MEM_STATIC */
+#include "../common/pool.h"        /* threadpool */
+#include "../common/threading.h"   /* mutex */
+#include "zstd_compress_internal.h" /* MIN, ERROR, ZSTD_*, ZSTD_highbit32 */
+#include "zstd_ldm.h"
+#include "zstdmt_compress.h"
+
+/* Guards code to support resizing the SeqPool.
+ * We will want to resize the SeqPool to save memory in the future.
+ * Until then, comment the code out since it is unused.
+ */
+#define ZSTD_RESIZE_SEQPOOL 0
+
+/* ======   Debug   ====== */
+#if defined(DEBUGLEVEL) && (DEBUGLEVEL>=2) \
+    && !defined(_MSC_VER) \
+    && !defined(__MINGW32__)
+
+#  include <stdio.h>
+#  include <unistd.h>
+#  include <sys/times.h>
+
+#  define DEBUG_PRINTHEX(l,p,n)                                       \
+    do {                                                              \
+        unsigned debug_u;                                             \
+        for (debug_u=0; debug_u<(n); debug_u++)                       \
+            RAWLOG(l, "%02X ", ((const unsigned char*)(p))[debug_u]); \
+        RAWLOG(l, " \n");                                             \
+    } while (0)
+
+static unsigned long long GetCurrentClockTimeMicroseconds(void)
+{
+   static clock_t _ticksPerSecond = 0;
+   if (_ticksPerSecond <= 0) _ticksPerSecond = sysconf(_SC_CLK_TCK);
+
+   {   struct tms junk; clock_t newTicks = (clock_t) times(&junk);
+       return ((((unsigned long long)newTicks)*(1000000))/_ticksPerSecond);
+}  }
+
+#define MUTEX_WAIT_TIME_DLEVEL 6
+#define ZSTD_PTHREAD_MUTEX_LOCK(mutex)                                                  \
+    do {                                                                                \
+        if (DEBUGLEVEL >= MUTEX_WAIT_TIME_DLEVEL) {                                     \
+            unsigned long long const beforeTime = GetCurrentClockTimeMicroseconds();    \
+            ZSTD_pthread_mutex_lock(mutex);                                             \
+            {   unsigned long long const afterTime = GetCurrentClockTimeMicroseconds(); \
+                unsigned long long const elapsedTime = (afterTime-beforeTime);          \
+                if (elapsedTime > 1000) {                                               \
+                    /* or whatever threshold you like; I'm using 1 millisecond here */  \
+                    DEBUGLOG(MUTEX_WAIT_TIME_DLEVEL,                                    \
+                        "Thread took %llu microseconds to acquire mutex %s \n",         \
+                        elapsedTime, #mutex);                                           \
+            }   }                                                                       \
+        } else {                                                                        \
+            ZSTD_pthread_mutex_lock(mutex);                                             \
+        }                                                                               \
+    } while (0)
+
+#else
+
+#  define ZSTD_PTHREAD_MUTEX_LOCK(m) ZSTD_pthread_mutex_lock(m)
+#  define DEBUG_PRINTHEX(l,p,n) do { } while (0)
+
+#endif
+
+
+/* =====   Buffer Pool   ===== */
+/* a single Buffer Pool can be invoked from multiple threads in parallel */
+
+typedef struct buffer_s {
+    void* start;
+    size_t capacity;
+} Buffer;
+
+static const Buffer g_nullBuffer = { NULL, 0 };
+
+typedef struct ZSTDMT_bufferPool_s {
+    ZSTD_pthread_mutex_t poolMutex;
+    size_t bufferSize;
+    unsigned totalBuffers;
+    unsigned nbBuffers;
+    ZSTD_customMem cMem;
+    Buffer* buffers;
+} ZSTDMT_bufferPool;
+
+static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool)
+{
+    DEBUGLOG(3, "ZSTDMT_freeBufferPool (address:%08X)", (U32)(size_t)bufPool);
+    if (!bufPool) return;   /* compatibility with free on NULL */
+    if (bufPool->buffers) {
+        unsigned u;
+        for (u=0; u<bufPool->totalBuffers; u++) {
+            DEBUGLOG(4, "free buffer %2u (address:%08X)", u, (U32)(size_t)bufPool->buffers[u].start);
+            ZSTD_customFree(bufPool->buffers[u].start, bufPool->cMem);
+        }
+        ZSTD_customFree(bufPool->buffers, bufPool->cMem);
+    }
+    ZSTD_pthread_mutex_destroy(&bufPool->poolMutex);
+    ZSTD_customFree(bufPool, bufPool->cMem);
+}
+
+static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned maxNbBuffers, ZSTD_customMem cMem)
+{
+    ZSTDMT_bufferPool* const bufPool =
+        (ZSTDMT_bufferPool*)ZSTD_customCalloc(sizeof(ZSTDMT_bufferPool), cMem);
+    if (bufPool==NULL) return NULL;
+    if (ZSTD_pthread_mutex_init(&bufPool->poolMutex, NULL)) {
+        ZSTD_customFree(bufPool, cMem);
+        return NULL;
+    }
+    bufPool->buffers = (Buffer*)ZSTD_customCalloc(maxNbBuffers * sizeof(Buffer), cMem);
+    if (bufPool->buffers==NULL) {
+        ZSTDMT_freeBufferPool(bufPool);
+        return NULL;
+    }
+    bufPool->bufferSize = 64 KB;
+    bufPool->totalBuffers = maxNbBuffers;
+    bufPool->nbBuffers = 0;
+    bufPool->cMem = cMem;
+    return bufPool;
+}
+
+/* only works at initialization, not during compression */
+static size_t ZSTDMT_sizeof_bufferPool(ZSTDMT_bufferPool* bufPool)
+{
+    size_t const poolSize = sizeof(*bufPool);
+    size_t const arraySize = bufPool->totalBuffers * sizeof(Buffer);
+    unsigned u;
+    size_t totalBufferSize = 0;
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    for (u=0; u<bufPool->totalBuffers; u++)
+        totalBufferSize += bufPool->buffers[u].capacity;
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+
+    return poolSize + arraySize + totalBufferSize;
+}
+
+/* ZSTDMT_setBufferSize() :
+ * all future buffers provided by this buffer pool will have _at least_ this size
+ * note : it's better for all buffers to have same size,
+ * as they become freely interchangeable, reducing malloc/free usages and memory fragmentation */
+static void ZSTDMT_setBufferSize(ZSTDMT_bufferPool* const bufPool, size_t const bSize)
+{
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    DEBUGLOG(4, "ZSTDMT_setBufferSize: bSize = %u", (U32)bSize);
+    bufPool->bufferSize = bSize;
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+}
+
+
+static ZSTDMT_bufferPool* ZSTDMT_expandBufferPool(ZSTDMT_bufferPool* srcBufPool, unsigned maxNbBuffers)
+{
+    if (srcBufPool==NULL) return NULL;
+    if (srcBufPool->totalBuffers >= maxNbBuffers) /* good enough */
+        return srcBufPool;
+    /* need a larger buffer pool */
+    {   ZSTD_customMem const cMem = srcBufPool->cMem;
+        size_t const bSize = srcBufPool->bufferSize;   /* forward parameters */
+        ZSTDMT_bufferPool* newBufPool;
+        ZSTDMT_freeBufferPool(srcBufPool);
+        newBufPool = ZSTDMT_createBufferPool(maxNbBuffers, cMem);
+        if (newBufPool==NULL) return newBufPool;
+        ZSTDMT_setBufferSize(newBufPool, bSize);
+        return newBufPool;
+    }
+}
+
+/** ZSTDMT_getBuffer() :
+ *  assumption : bufPool must be valid
+ * @return : a buffer, with start pointer and size
+ *  note: allocation may fail, in this case, start==NULL and size==0 */
+static Buffer ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool)
+{
+    size_t const bSize = bufPool->bufferSize;
+    DEBUGLOG(5, "ZSTDMT_getBuffer: bSize = %u", (U32)bufPool->bufferSize);
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    if (bufPool->nbBuffers) {   /* try to use an existing buffer */
+        Buffer const buf = bufPool->buffers[--(bufPool->nbBuffers)];
+        size_t const availBufferSize = buf.capacity;
+        bufPool->buffers[bufPool->nbBuffers] = g_nullBuffer;
+        if ((availBufferSize >= bSize) & ((availBufferSize>>3) <= bSize)) {
+            /* large enough, but not too much */
+            DEBUGLOG(5, "ZSTDMT_getBuffer: provide buffer %u of size %u",
+                        bufPool->nbBuffers, (U32)buf.capacity);
+            ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+            return buf;
+        }
+        /* size conditions not respected : scratch this buffer, create new one */
+        DEBUGLOG(5, "ZSTDMT_getBuffer: existing buffer does not meet size conditions => freeing");
+        ZSTD_customFree(buf.start, bufPool->cMem);
+    }
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+    /* create new buffer */
+    DEBUGLOG(5, "ZSTDMT_getBuffer: create a new buffer");
+    {   Buffer buffer;
+        void* const start = ZSTD_customMalloc(bSize, bufPool->cMem);
+        buffer.start = start;   /* note : start can be NULL if malloc fails ! */
+        buffer.capacity = (start==NULL) ? 0 : bSize;
+        if (start==NULL) {
+            DEBUGLOG(5, "ZSTDMT_getBuffer: buffer allocation failure !!");
+        } else {
+            DEBUGLOG(5, "ZSTDMT_getBuffer: created buffer of size %u", (U32)bSize);
+        }
+        return buffer;
+    }
+}
+
+#if ZSTD_RESIZE_SEQPOOL
+/** ZSTDMT_resizeBuffer() :
+ * assumption : bufPool must be valid
+ * @return : a buffer that is at least the buffer pool buffer size.
+ *           If a reallocation happens, the data in the input buffer is copied.
+ */
+static Buffer ZSTDMT_resizeBuffer(ZSTDMT_bufferPool* bufPool, Buffer buffer)
+{
+    size_t const bSize = bufPool->bufferSize;
+    if (buffer.capacity < bSize) {
+        void* const start = ZSTD_customMalloc(bSize, bufPool->cMem);
+        Buffer newBuffer;
+        newBuffer.start = start;
+        newBuffer.capacity = start == NULL ? 0 : bSize;
+        if (start != NULL) {
+            assert(newBuffer.capacity >= buffer.capacity);
+            ZSTD_memcpy(newBuffer.start, buffer.start, buffer.capacity);
+            DEBUGLOG(5, "ZSTDMT_resizeBuffer: created buffer of size %u", (U32)bSize);
+            return newBuffer;
+        }
+        DEBUGLOG(5, "ZSTDMT_resizeBuffer: buffer allocation failure !!");
+    }
+    return buffer;
+}
+#endif
+
+/* store buffer for later re-use, up to pool capacity */
+static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, Buffer buf)
+{
+    DEBUGLOG(5, "ZSTDMT_releaseBuffer");
+    if (buf.start == NULL) return;   /* compatible with release on NULL */
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    if (bufPool->nbBuffers < bufPool->totalBuffers) {
+        bufPool->buffers[bufPool->nbBuffers++] = buf;  /* stored for later use */
+        DEBUGLOG(5, "ZSTDMT_releaseBuffer: stored buffer of size %u in slot %u",
+                    (U32)buf.capacity, (U32)(bufPool->nbBuffers-1));
+        ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+        return;
+    }
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+    /* Reached bufferPool capacity (note: should not happen) */
+    DEBUGLOG(5, "ZSTDMT_releaseBuffer: pool capacity reached => freeing ");
+    ZSTD_customFree(buf.start, bufPool->cMem);
+}
+
+/* We need 2 output buffers per worker since each dstBuff must be flushed after it is released.
+ * The 3 additional buffers are as follows:
+ *   1 buffer for input loading
+ *   1 buffer for "next input" when submitting current one
+ *   1 buffer stuck in queue */
+#define BUF_POOL_MAX_NB_BUFFERS(nbWorkers) (2*(nbWorkers) + 3)
+
+/* After a worker releases its rawSeqStore, it is immediately ready for reuse.
+ * So we only need one seq buffer per worker. */
+#define SEQ_POOL_MAX_NB_BUFFERS(nbWorkers) (nbWorkers)
+
+/* =====   Seq Pool Wrapper   ====== */
+
+typedef ZSTDMT_bufferPool ZSTDMT_seqPool;
+
+static size_t ZSTDMT_sizeof_seqPool(ZSTDMT_seqPool* seqPool)
+{
+    return ZSTDMT_sizeof_bufferPool(seqPool);
+}
+
+static RawSeqStore_t bufferToSeq(Buffer buffer)
+{
+    RawSeqStore_t seq = kNullRawSeqStore;
+    seq.seq = (rawSeq*)buffer.start;
+    seq.capacity = buffer.capacity / sizeof(rawSeq);
+    return seq;
+}
+
+static Buffer seqToBuffer(RawSeqStore_t seq)
+{
+    Buffer buffer;
+    buffer.start = seq.seq;
+    buffer.capacity = seq.capacity * sizeof(rawSeq);
+    return buffer;
+}
+
+static RawSeqStore_t ZSTDMT_getSeq(ZSTDMT_seqPool* seqPool)
+{
+    if (seqPool->bufferSize == 0) {
+        return kNullRawSeqStore;
+    }
+    return bufferToSeq(ZSTDMT_getBuffer(seqPool));
+}
+
+#if ZSTD_RESIZE_SEQPOOL
+static RawSeqStore_t ZSTDMT_resizeSeq(ZSTDMT_seqPool* seqPool, RawSeqStore_t seq)
+{
+  return bufferToSeq(ZSTDMT_resizeBuffer(seqPool, seqToBuffer(seq)));
+}
+#endif
+
+static void ZSTDMT_releaseSeq(ZSTDMT_seqPool* seqPool, RawSeqStore_t seq)
+{
+  ZSTDMT_releaseBuffer(seqPool, seqToBuffer(seq));
+}
+
+static void ZSTDMT_setNbSeq(ZSTDMT_seqPool* const seqPool, size_t const nbSeq)
+{
+  ZSTDMT_setBufferSize(seqPool, nbSeq * sizeof(rawSeq));
+}
+
+static ZSTDMT_seqPool* ZSTDMT_createSeqPool(unsigned nbWorkers, ZSTD_customMem cMem)
+{
+    ZSTDMT_seqPool* const seqPool = ZSTDMT_createBufferPool(SEQ_POOL_MAX_NB_BUFFERS(nbWorkers), cMem);
+    if (seqPool == NULL) return NULL;
+    ZSTDMT_setNbSeq(seqPool, 0);
+    return seqPool;
+}
+
+static void ZSTDMT_freeSeqPool(ZSTDMT_seqPool* seqPool)
+{
+    ZSTDMT_freeBufferPool(seqPool);
+}
+
+static ZSTDMT_seqPool* ZSTDMT_expandSeqPool(ZSTDMT_seqPool* pool, U32 nbWorkers)
+{
+    return ZSTDMT_expandBufferPool(pool, SEQ_POOL_MAX_NB_BUFFERS(nbWorkers));
+}
+
+
+/* =====   CCtx Pool   ===== */
+/* a single CCtx Pool can be invoked from multiple threads in parallel */
+
+typedef struct {
+    ZSTD_pthread_mutex_t poolMutex;
+    int totalCCtx;
+    int availCCtx;
+    ZSTD_customMem cMem;
+    ZSTD_CCtx** cctxs;
+} ZSTDMT_CCtxPool;
+
+/* note : all CCtx borrowed from the pool must be reverted back to the pool _before_ freeing the pool */
+static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
+{
+    if (!pool) return;
+    ZSTD_pthread_mutex_destroy(&pool->poolMutex);
+    if (pool->cctxs) {
+        int cid;
+        for (cid=0; cid<pool->totalCCtx; cid++)
+            ZSTD_freeCCtx(pool->cctxs[cid]);  /* free compatible with NULL */
+        ZSTD_customFree(pool->cctxs, pool->cMem);
+    }
+    ZSTD_customFree(pool, pool->cMem);
+}
+
+/* ZSTDMT_createCCtxPool() :
+ * implies nbWorkers >= 1 , checked by caller ZSTDMT_createCCtx() */
+static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(int nbWorkers,
+                                              ZSTD_customMem cMem)
+{
+    ZSTDMT_CCtxPool* const cctxPool =
+        (ZSTDMT_CCtxPool*) ZSTD_customCalloc(sizeof(ZSTDMT_CCtxPool), cMem);
+    assert(nbWorkers > 0);
+    if (!cctxPool) return NULL;
+    if (ZSTD_pthread_mutex_init(&cctxPool->poolMutex, NULL)) {
+        ZSTD_customFree(cctxPool, cMem);
+        return NULL;
+    }
+    cctxPool->totalCCtx = nbWorkers;
+    cctxPool->cctxs = (ZSTD_CCtx**)ZSTD_customCalloc(nbWorkers * sizeof(ZSTD_CCtx*), cMem);
+    if (!cctxPool->cctxs) {
+        ZSTDMT_freeCCtxPool(cctxPool);
+        return NULL;
+    }
+    cctxPool->cMem = cMem;
+    cctxPool->cctxs[0] = ZSTD_createCCtx_advanced(cMem);
+    if (!cctxPool->cctxs[0]) { ZSTDMT_freeCCtxPool(cctxPool); return NULL; }
+    cctxPool->availCCtx = 1;   /* at least one cctx for single-thread mode */
+    DEBUGLOG(3, "cctxPool created, with %u workers", nbWorkers);
+    return cctxPool;
+}
+
+static ZSTDMT_CCtxPool* ZSTDMT_expandCCtxPool(ZSTDMT_CCtxPool* srcPool,
+                                              int nbWorkers)
+{
+    if (srcPool==NULL) return NULL;
+    if (nbWorkers <= srcPool->totalCCtx) return srcPool;   /* good enough */
+    /* need a larger cctx pool */
+    {   ZSTD_customMem const cMem = srcPool->cMem;
+        ZSTDMT_freeCCtxPool(srcPool);
+        return ZSTDMT_createCCtxPool(nbWorkers, cMem);
+    }
+}
+
+/* only works during initialization phase, not during compression */
+static size_t ZSTDMT_sizeof_CCtxPool(ZSTDMT_CCtxPool* cctxPool)
+{
+    ZSTD_pthread_mutex_lock(&cctxPool->poolMutex);
+    {   unsigned const nbWorkers = cctxPool->totalCCtx;
+        size_t const poolSize = sizeof(*cctxPool);
+        size_t const arraySize = cctxPool->totalCCtx * sizeof(ZSTD_CCtx*);
+        size_t totalCCtxSize = 0;
+        unsigned u;
+        for (u=0; u<nbWorkers; u++) {
+            totalCCtxSize += ZSTD_sizeof_CCtx(cctxPool->cctxs[u]);
+        }
+        ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+        assert(nbWorkers > 0);
+        return poolSize + arraySize + totalCCtxSize;
+    }
+}
+
+static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* cctxPool)
+{
+    DEBUGLOG(5, "ZSTDMT_getCCtx");
+    ZSTD_pthread_mutex_lock(&cctxPool->poolMutex);
+    if (cctxPool->availCCtx) {
+        cctxPool->availCCtx--;
+        {   ZSTD_CCtx* const cctx = cctxPool->cctxs[cctxPool->availCCtx];
+            ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+            return cctx;
+    }   }
+    ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+    DEBUGLOG(5, "create one more CCtx");
+    return ZSTD_createCCtx_advanced(cctxPool->cMem);   /* note : can be NULL, when creation fails ! */
+}
+
+static void ZSTDMT_releaseCCtx(ZSTDMT_CCtxPool* pool, ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return;   /* compatibility with release on NULL */
+    ZSTD_pthread_mutex_lock(&pool->poolMutex);
+    if (pool->availCCtx < pool->totalCCtx)
+        pool->cctxs[pool->availCCtx++] = cctx;
+    else {
+        /* pool overflow : should not happen, since totalCCtx==nbWorkers */
+        DEBUGLOG(4, "CCtx pool overflow : free cctx");
+        ZSTD_freeCCtx(cctx);
+    }
+    ZSTD_pthread_mutex_unlock(&pool->poolMutex);
+}
+
+/* ====   Serial State   ==== */
+
+typedef struct {
+    void const* start;
+    size_t size;
+} Range;
+
+typedef struct {
+    /* All variables in the struct are protected by mutex. */
+    ZSTD_pthread_mutex_t mutex;
+    ZSTD_pthread_cond_t cond;
+    ZSTD_CCtx_params params;
+    ldmState_t ldmState;
+    XXH64_state_t xxhState;
+    unsigned nextJobID;
+    /* Protects ldmWindow.
+     * Must be acquired after the main mutex when acquiring both.
+     */
+    ZSTD_pthread_mutex_t ldmWindowMutex;
+    ZSTD_pthread_cond_t ldmWindowCond;  /* Signaled when ldmWindow is updated */
+    ZSTD_window_t ldmWindow;  /* A thread-safe copy of ldmState.window */
+} SerialState;
+
+static int
+ZSTDMT_serialState_reset(SerialState* serialState,
+                         ZSTDMT_seqPool* seqPool,
+                         ZSTD_CCtx_params params,
+                         size_t jobSize,
+                         const void* dict, size_t const dictSize,
+                         ZSTD_dictContentType_e dictContentType)
+{
+    /* Adjust parameters */
+    if (params.ldmParams.enableLdm == ZSTD_ps_enable) {
+        DEBUGLOG(4, "LDM window size = %u KB", (1U << params.cParams.windowLog) >> 10);
+        ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
+        assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
+        assert(params.ldmParams.hashRateLog < 32);
+    } else {
+        ZSTD_memset(&params.ldmParams, 0, sizeof(params.ldmParams));
+    }
+    serialState->nextJobID = 0;
+    if (params.fParams.checksumFlag)
+        XXH64_reset(&serialState->xxhState, 0);
+    if (params.ldmParams.enableLdm == ZSTD_ps_enable) {
+        ZSTD_customMem cMem = params.customMem;
+        unsigned const hashLog = params.ldmParams.hashLog;
+        size_t const hashSize = ((size_t)1 << hashLog) * sizeof(ldmEntry_t);
+        unsigned const bucketLog =
+            params.ldmParams.hashLog - params.ldmParams.bucketSizeLog;
+        unsigned const prevBucketLog =
+            serialState->params.ldmParams.hashLog -
+            serialState->params.ldmParams.bucketSizeLog;
+        size_t const numBuckets = (size_t)1 << bucketLog;
+        /* Size the seq pool tables */
+        ZSTDMT_setNbSeq(seqPool, ZSTD_ldm_getMaxNbSeq(params.ldmParams, jobSize));
+        /* Reset the window */
+        ZSTD_window_init(&serialState->ldmState.window);
+        /* Resize tables and output space if necessary. */
+        if (serialState->ldmState.hashTable == NULL || serialState->params.ldmParams.hashLog < hashLog) {
+            ZSTD_customFree(serialState->ldmState.hashTable, cMem);
+            serialState->ldmState.hashTable = (ldmEntry_t*)ZSTD_customMalloc(hashSize, cMem);
+        }
+        if (serialState->ldmState.bucketOffsets == NULL || prevBucketLog < bucketLog) {
+            ZSTD_customFree(serialState->ldmState.bucketOffsets, cMem);
+            serialState->ldmState.bucketOffsets = (BYTE*)ZSTD_customMalloc(numBuckets, cMem);
+        }
+        if (!serialState->ldmState.hashTable || !serialState->ldmState.bucketOffsets)
+            return 1;
+        /* Zero the tables */
+        ZSTD_memset(serialState->ldmState.hashTable, 0, hashSize);
+        ZSTD_memset(serialState->ldmState.bucketOffsets, 0, numBuckets);
+
+        /* Update window state and fill hash table with dict */
+        serialState->ldmState.loadedDictEnd = 0;
+        if (dictSize > 0) {
+            if (dictContentType == ZSTD_dct_rawContent) {
+                BYTE const* const dictEnd = (const BYTE*)dict + dictSize;
+                ZSTD_window_update(&serialState->ldmState.window, dict, dictSize, /* forceNonContiguous */ 0);
+                ZSTD_ldm_fillHashTable(&serialState->ldmState, (const BYTE*)dict, dictEnd, &params.ldmParams);
+                serialState->ldmState.loadedDictEnd = params.forceWindow ? 0 : (U32)(dictEnd - serialState->ldmState.window.base);
+            } else {
+                /* don't even load anything */
+            }
+        }
+
+        /* Initialize serialState's copy of ldmWindow. */
+        serialState->ldmWindow = serialState->ldmState.window;
+    }
+
+    serialState->params = params;
+    serialState->params.jobSize = (U32)jobSize;
+    return 0;
+}
+
+static int ZSTDMT_serialState_init(SerialState* serialState)
+{
+    int initError = 0;
+    ZSTD_memset(serialState, 0, sizeof(*serialState));
+    initError |= ZSTD_pthread_mutex_init(&serialState->mutex, NULL);
+    initError |= ZSTD_pthread_cond_init(&serialState->cond, NULL);
+    initError |= ZSTD_pthread_mutex_init(&serialState->ldmWindowMutex, NULL);
+    initError |= ZSTD_pthread_cond_init(&serialState->ldmWindowCond, NULL);
+    return initError;
+}
+
+static void ZSTDMT_serialState_free(SerialState* serialState)
+{
+    ZSTD_customMem cMem = serialState->params.customMem;
+    ZSTD_pthread_mutex_destroy(&serialState->mutex);
+    ZSTD_pthread_cond_destroy(&serialState->cond);
+    ZSTD_pthread_mutex_destroy(&serialState->ldmWindowMutex);
+    ZSTD_pthread_cond_destroy(&serialState->ldmWindowCond);
+    ZSTD_customFree(serialState->ldmState.hashTable, cMem);
+    ZSTD_customFree(serialState->ldmState.bucketOffsets, cMem);
+}
+
+static void
+ZSTDMT_serialState_genSequences(SerialState* serialState,
+                                RawSeqStore_t* seqStore,
+                                Range src, unsigned jobID)
+{
+    /* Wait for our turn */
+    ZSTD_PTHREAD_MUTEX_LOCK(&serialState->mutex);
+    while (serialState->nextJobID < jobID) {
+        DEBUGLOG(5, "wait for serialState->cond");
+        ZSTD_pthread_cond_wait(&serialState->cond, &serialState->mutex);
+    }
+    /* A future job may error and skip our job */
+    if (serialState->nextJobID == jobID) {
+        /* It is now our turn, do any processing necessary */
+        if (serialState->params.ldmParams.enableLdm == ZSTD_ps_enable) {
+            size_t error;
+            DEBUGLOG(6, "ZSTDMT_serialState_genSequences: LDM update");
+            assert(seqStore->seq != NULL && seqStore->pos == 0 &&
+                   seqStore->size == 0 && seqStore->capacity > 0);
+            assert(src.size <= serialState->params.jobSize);
+            ZSTD_window_update(&serialState->ldmState.window, src.start, src.size, /* forceNonContiguous */ 0);
+            error = ZSTD_ldm_generateSequences(
+                &serialState->ldmState, seqStore,
+                &serialState->params.ldmParams, src.start, src.size);
+            /* We provide a large enough buffer to never fail. */
+            assert(!ZSTD_isError(error)); (void)error;
+            /* Update ldmWindow to match the ldmState.window and signal the main
+             * thread if it is waiting for a buffer.
+             */
+            ZSTD_PTHREAD_MUTEX_LOCK(&serialState->ldmWindowMutex);
+            serialState->ldmWindow = serialState->ldmState.window;
+            ZSTD_pthread_cond_signal(&serialState->ldmWindowCond);
+            ZSTD_pthread_mutex_unlock(&serialState->ldmWindowMutex);
+        }
+        if (serialState->params.fParams.checksumFlag && src.size > 0)
+            XXH64_update(&serialState->xxhState, src.start, src.size);
+    }
+    /* Now it is the next jobs turn */
+    serialState->nextJobID++;
+    ZSTD_pthread_cond_broadcast(&serialState->cond);
+    ZSTD_pthread_mutex_unlock(&serialState->mutex);
+}
+
+static void
+ZSTDMT_serialState_applySequences(const SerialState* serialState, /* just for an assert() check */
+                                  ZSTD_CCtx* jobCCtx,
+                                  const RawSeqStore_t* seqStore)
+{
+    if (seqStore->size > 0) {
+        DEBUGLOG(5, "ZSTDMT_serialState_applySequences: uploading %u external sequences", (unsigned)seqStore->size);
+        assert(serialState->params.ldmParams.enableLdm == ZSTD_ps_enable); (void)serialState;
+        assert(jobCCtx);
+        ZSTD_referenceExternalSequences(jobCCtx, seqStore->seq, seqStore->size);
+    }
+}
+
+static void ZSTDMT_serialState_ensureFinished(SerialState* serialState,
+                                              unsigned jobID, size_t cSize)
+{
+    ZSTD_PTHREAD_MUTEX_LOCK(&serialState->mutex);
+    if (serialState->nextJobID <= jobID) {
+        assert(ZSTD_isError(cSize)); (void)cSize;
+        DEBUGLOG(5, "Skipping past job %u because of error", jobID);
+        serialState->nextJobID = jobID + 1;
+        ZSTD_pthread_cond_broadcast(&serialState->cond);
+
+        ZSTD_PTHREAD_MUTEX_LOCK(&serialState->ldmWindowMutex);
+        ZSTD_window_clear(&serialState->ldmWindow);
+        ZSTD_pthread_cond_signal(&serialState->ldmWindowCond);
+        ZSTD_pthread_mutex_unlock(&serialState->ldmWindowMutex);
+    }
+    ZSTD_pthread_mutex_unlock(&serialState->mutex);
+
+}
+
+
+/* ------------------------------------------ */
+/* =====          Worker thread         ===== */
+/* ------------------------------------------ */
+
+static const Range kNullRange = { NULL, 0 };
+
+typedef struct {
+    size_t   consumed;                 /* SHARED - set0 by mtctx, then modified by worker AND read by mtctx */
+    size_t   cSize;                    /* SHARED - set0 by mtctx, then modified by worker AND read by mtctx, then set0 by mtctx */
+    ZSTD_pthread_mutex_t job_mutex;    /* Thread-safe - used by mtctx and worker */
+    ZSTD_pthread_cond_t job_cond;      /* Thread-safe - used by mtctx and worker */
+    ZSTDMT_CCtxPool* cctxPool;         /* Thread-safe - used by mtctx and (all) workers */
+    ZSTDMT_bufferPool* bufPool;        /* Thread-safe - used by mtctx and (all) workers */
+    ZSTDMT_seqPool* seqPool;           /* Thread-safe - used by mtctx and (all) workers */
+    SerialState* serial;               /* Thread-safe - used by mtctx and (all) workers */
+    Buffer dstBuff;                    /* set by worker (or mtctx), then read by worker & mtctx, then modified by mtctx => no barrier */
+    Range prefix;                      /* set by mtctx, then read by worker & mtctx => no barrier */
+    Range src;                         /* set by mtctx, then read by worker & mtctx => no barrier */
+    unsigned jobID;                    /* set by mtctx, then read by worker => no barrier */
+    unsigned firstJob;                 /* set by mtctx, then read by worker => no barrier */
+    unsigned lastJob;                  /* set by mtctx, then read by worker => no barrier */
+    ZSTD_CCtx_params params;           /* set by mtctx, then read by worker => no barrier */
+    const ZSTD_CDict* cdict;           /* set by mtctx, then read by worker => no barrier */
+    unsigned long long fullFrameSize;  /* set by mtctx, then read by worker => no barrier */
+    size_t   dstFlushed;               /* used only by mtctx */
+    unsigned frameChecksumNeeded;      /* used only by mtctx */
+} ZSTDMT_jobDescription;
+
+#define JOB_ERROR(e)                                \
+    do {                                            \
+        ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);   \
+        job->cSize = e;                             \
+        ZSTD_pthread_mutex_unlock(&job->job_mutex); \
+        goto _endJob;                               \
+    } while (0)
+
+/* ZSTDMT_compressionJob() is a POOL_function type */
+static void ZSTDMT_compressionJob(void* jobDescription)
+{
+    ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription;
+    ZSTD_CCtx_params jobParams = job->params;   /* do not modify job->params ! copy it, modify the copy */
+    ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(job->cctxPool);
+    RawSeqStore_t rawSeqStore = ZSTDMT_getSeq(job->seqPool);
+    Buffer dstBuff = job->dstBuff;
+    size_t lastCBlockSize = 0;
+
+    DEBUGLOG(5, "ZSTDMT_compressionJob: job %u", job->jobID);
+    /* resources */
+    if (cctx==NULL) JOB_ERROR(ERROR(memory_allocation));
+    if (dstBuff.start == NULL) {   /* streaming job : doesn't provide a dstBuffer */
+        dstBuff = ZSTDMT_getBuffer(job->bufPool);
+        if (dstBuff.start==NULL) JOB_ERROR(ERROR(memory_allocation));
+        job->dstBuff = dstBuff;   /* this value can be read in ZSTDMT_flush, when it copies the whole job */
+    }
+    if (jobParams.ldmParams.enableLdm == ZSTD_ps_enable && rawSeqStore.seq == NULL)
+        JOB_ERROR(ERROR(memory_allocation));
+
+    /* Don't compute the checksum for chunks, since we compute it externally,
+     * but write it in the header.
+     */
+    if (job->jobID != 0) jobParams.fParams.checksumFlag = 0;
+    /* Don't run LDM for the chunks, since we handle it externally */
+    jobParams.ldmParams.enableLdm = ZSTD_ps_disable;
+    /* Correct nbWorkers to 0. */
+    jobParams.nbWorkers = 0;
+
+
+    /* init */
+
+    /* Perform serial step as early as possible */
+    ZSTDMT_serialState_genSequences(job->serial, &rawSeqStore, job->src, job->jobID);
+
+    if (job->cdict) {
+        size_t const initError = ZSTD_compressBegin_advanced_internal(cctx, NULL, 0, ZSTD_dct_auto, ZSTD_dtlm_fast, job->cdict, &jobParams, job->fullFrameSize);
+        assert(job->firstJob);  /* only allowed for first job */
+        if (ZSTD_isError(initError)) JOB_ERROR(initError);
+    } else {
+        U64 const pledgedSrcSize = job->firstJob ? job->fullFrameSize : job->src.size;
+        {   size_t const forceWindowError = ZSTD_CCtxParams_setParameter(&jobParams, ZSTD_c_forceMaxWindow, !job->firstJob);
+            if (ZSTD_isError(forceWindowError)) JOB_ERROR(forceWindowError);
+        }
+        if (!job->firstJob) {
+            size_t const err = ZSTD_CCtxParams_setParameter(&jobParams, ZSTD_c_deterministicRefPrefix, 0);
+            if (ZSTD_isError(err)) JOB_ERROR(err);
+        }
+        DEBUGLOG(6, "ZSTDMT_compressionJob: job %u: loading prefix of size %zu", job->jobID, job->prefix.size);
+        {   size_t const initError = ZSTD_compressBegin_advanced_internal(cctx,
+                                        job->prefix.start, job->prefix.size, ZSTD_dct_rawContent,
+                                        ZSTD_dtlm_fast,
+                                        NULL, /*cdict*/
+                                        &jobParams, pledgedSrcSize);
+            if (ZSTD_isError(initError)) JOB_ERROR(initError);
+    }   }
+
+    /* External Sequences can only be applied after CCtx initialization */
+    ZSTDMT_serialState_applySequences(job->serial, cctx, &rawSeqStore);
+
+    if (!job->firstJob) {  /* flush and overwrite frame header when it's not first job */
+        size_t const hSize = ZSTD_compressContinue_public(cctx, dstBuff.start, dstBuff.capacity, job->src.start, 0);
+        if (ZSTD_isError(hSize)) JOB_ERROR(hSize);
+        DEBUGLOG(5, "ZSTDMT_compressionJob: flush and overwrite %u bytes of frame header (not first job)", (U32)hSize);
+        ZSTD_invalidateRepCodes(cctx);
+    }
+
+    /* compress the entire job by smaller chunks, for better granularity */
+    {   size_t const chunkSize = 4*ZSTD_BLOCKSIZE_MAX;
+        int const nbChunks = (int)((job->src.size + (chunkSize-1)) / chunkSize);
+        const BYTE* ip = (const BYTE*) job->src.start;
+        BYTE* const ostart = (BYTE*)dstBuff.start;
+        BYTE* op = ostart;
+        BYTE* oend = op + dstBuff.capacity;
+        int chunkNb;
+        if (sizeof(size_t) > sizeof(int)) assert(job->src.size < ((size_t)INT_MAX) * chunkSize);   /* check overflow */
+        DEBUGLOG(5, "ZSTDMT_compressionJob: compress %u bytes in %i blocks", (U32)job->src.size, nbChunks);
+        assert(job->cSize == 0);
+        for (chunkNb = 1; chunkNb < nbChunks; chunkNb++) {
+            size_t const cSize = ZSTD_compressContinue_public(cctx, op, oend-op, ip, chunkSize);
+            if (ZSTD_isError(cSize)) JOB_ERROR(cSize);
+            ip += chunkSize;
+            op += cSize; assert(op < oend);
+            /* stats */
+            ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
+            job->cSize += cSize;
+            job->consumed = chunkSize * chunkNb;
+            DEBUGLOG(5, "ZSTDMT_compressionJob: compress new block : cSize==%u bytes (total: %u)",
+                        (U32)cSize, (U32)job->cSize);
+            ZSTD_pthread_cond_signal(&job->job_cond);   /* warns some more data is ready to be flushed */
+            ZSTD_pthread_mutex_unlock(&job->job_mutex);
+        }
+        /* last block */
+        assert(chunkSize > 0);
+        assert((chunkSize & (chunkSize - 1)) == 0);  /* chunkSize must be power of 2 for mask==(chunkSize-1) to work */
+        if ((nbChunks > 0) | job->lastJob /*must output a "last block" flag*/ ) {
+            size_t const lastBlockSize1 = job->src.size & (chunkSize-1);
+            size_t const lastBlockSize = ((lastBlockSize1==0) & (job->src.size>=chunkSize)) ? chunkSize : lastBlockSize1;
+            size_t const cSize = (job->lastJob) ?
+                 ZSTD_compressEnd_public(cctx, op, oend-op, ip, lastBlockSize) :
+                 ZSTD_compressContinue_public(cctx, op, oend-op, ip, lastBlockSize);
+            if (ZSTD_isError(cSize)) JOB_ERROR(cSize);
+            lastCBlockSize = cSize;
+    }   }
+    if (!job->firstJob) {
+        /* Double check that we don't have an ext-dict, because then our
+         * repcode invalidation doesn't work.
+         */
+        assert(!ZSTD_window_hasExtDict(cctx->blockState.matchState.window));
+    }
+    ZSTD_CCtx_trace(cctx, 0);
+
+_endJob:
+    ZSTDMT_serialState_ensureFinished(job->serial, job->jobID, job->cSize);
+    if (job->prefix.size > 0)
+        DEBUGLOG(5, "Finished with prefix: %zx", (size_t)job->prefix.start);
+    DEBUGLOG(5, "Finished with source: %zx", (size_t)job->src.start);
+    /* release resources */
+    ZSTDMT_releaseSeq(job->seqPool, rawSeqStore);
+    ZSTDMT_releaseCCtx(job->cctxPool, cctx);
+    /* report */
+    ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
+    if (ZSTD_isError(job->cSize)) assert(lastCBlockSize == 0);
+    job->cSize += lastCBlockSize;
+    job->consumed = job->src.size;  /* when job->consumed == job->src.size , compression job is presumed completed */
+    ZSTD_pthread_cond_signal(&job->job_cond);
+    ZSTD_pthread_mutex_unlock(&job->job_mutex);
+}
+
+
+/* ------------------------------------------ */
+/* =====   Multi-threaded compression   ===== */
+/* ------------------------------------------ */
+
+typedef struct {
+    Range prefix;         /* read-only non-owned prefix buffer */
+    Buffer buffer;
+    size_t filled;
+} InBuff_t;
+
+typedef struct {
+  BYTE* buffer;     /* The round input buffer. All jobs get references
+                     * to pieces of the buffer. ZSTDMT_tryGetInputRange()
+                     * handles handing out job input buffers, and makes
+                     * sure it doesn't overlap with any pieces still in use.
+                     */
+  size_t capacity;  /* The capacity of buffer. */
+  size_t pos;       /* The position of the current inBuff in the round
+                     * buffer. Updated past the end if the inBuff once
+                     * the inBuff is sent to the worker thread.
+                     * pos <= capacity.
+                     */
+} RoundBuff_t;
+
+static const RoundBuff_t kNullRoundBuff = {NULL, 0, 0};
+
+#define RSYNC_LENGTH 32
+/* Don't create chunks smaller than the zstd block size.
+ * This stops us from regressing compression ratio too much,
+ * and ensures our output fits in ZSTD_compressBound().
+ *
+ * If this is shrunk < ZSTD_BLOCKSIZELOG_MIN then
+ * ZSTD_COMPRESSBOUND() will need to be updated.
+ */
+#define RSYNC_MIN_BLOCK_LOG ZSTD_BLOCKSIZELOG_MAX
+#define RSYNC_MIN_BLOCK_SIZE (1<<RSYNC_MIN_BLOCK_LOG)
+
+typedef struct {
+  U64 hash;
+  U64 hitMask;
+  U64 primePower;
+} RSyncState_t;
+
+struct ZSTDMT_CCtx_s {
+    POOL_ctx* factory;
+    ZSTDMT_jobDescription* jobs;
+    ZSTDMT_bufferPool* bufPool;
+    ZSTDMT_CCtxPool* cctxPool;
+    ZSTDMT_seqPool* seqPool;
+    ZSTD_CCtx_params params;
+    size_t targetSectionSize;
+    size_t targetPrefixSize;
+    int jobReady;        /* 1 => one job is already prepared, but pool has shortage of workers. Don't create a new job. */
+    InBuff_t inBuff;
+    RoundBuff_t roundBuff;
+    SerialState serial;
+    RSyncState_t rsync;
+    unsigned jobIDMask;
+    unsigned doneJobID;
+    unsigned nextJobID;
+    unsigned frameEnded;
+    unsigned allJobsCompleted;
+    unsigned long long frameContentSize;
+    unsigned long long consumed;
+    unsigned long long produced;
+    ZSTD_customMem cMem;
+    ZSTD_CDict* cdictLocal;
+    const ZSTD_CDict* cdict;
+    unsigned providedFactory: 1;
+};
+
+static void ZSTDMT_freeJobsTable(ZSTDMT_jobDescription* jobTable, U32 nbJobs, ZSTD_customMem cMem)
+{
+    U32 jobNb;
+    if (jobTable == NULL) return;
+    for (jobNb=0; jobNb<nbJobs; jobNb++) {
+        ZSTD_pthread_mutex_destroy(&jobTable[jobNb].job_mutex);
+        ZSTD_pthread_cond_destroy(&jobTable[jobNb].job_cond);
+    }
+    ZSTD_customFree(jobTable, cMem);
+}
+
+/* ZSTDMT_allocJobsTable()
+ * allocate and init a job table.
+ * update *nbJobsPtr to next power of 2 value, as size of table */
+static ZSTDMT_jobDescription* ZSTDMT_createJobsTable(U32* nbJobsPtr, ZSTD_customMem cMem)
+{
+    U32 const nbJobsLog2 = ZSTD_highbit32(*nbJobsPtr) + 1;
+    U32 const nbJobs = 1 << nbJobsLog2;
+    U32 jobNb;
+    ZSTDMT_jobDescription* const jobTable = (ZSTDMT_jobDescription*)
+                ZSTD_customCalloc(nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
+    int initError = 0;
+    if (jobTable==NULL) return NULL;
+    *nbJobsPtr = nbJobs;
+    for (jobNb=0; jobNb<nbJobs; jobNb++) {
+        initError |= ZSTD_pthread_mutex_init(&jobTable[jobNb].job_mutex, NULL);
+        initError |= ZSTD_pthread_cond_init(&jobTable[jobNb].job_cond, NULL);
+    }
+    if (initError != 0) {
+        ZSTDMT_freeJobsTable(jobTable, nbJobs, cMem);
+        return NULL;
+    }
+    return jobTable;
+}
+
+static size_t ZSTDMT_expandJobsTable (ZSTDMT_CCtx* mtctx, U32 nbWorkers) {
+    U32 nbJobs = nbWorkers + 2;
+    if (nbJobs > mtctx->jobIDMask+1) {  /* need more job capacity */
+        ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
+        mtctx->jobIDMask = 0;
+        mtctx->jobs = ZSTDMT_createJobsTable(&nbJobs, mtctx->cMem);
+        if (mtctx->jobs==NULL) return ERROR(memory_allocation);
+        assert((nbJobs != 0) && ((nbJobs & (nbJobs - 1)) == 0));  /* ensure nbJobs is a power of 2 */
+        mtctx->jobIDMask = nbJobs - 1;
+    }
+    return 0;
+}
+
+
+/* ZSTDMT_CCtxParam_setNbWorkers():
+ * Internal use only */
+static size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers)
+{
+    return ZSTD_CCtxParams_setParameter(params, ZSTD_c_nbWorkers, (int)nbWorkers);
+}
+
+MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers, ZSTD_customMem cMem, ZSTD_threadPool* pool)
+{
+    ZSTDMT_CCtx* mtctx;
+    U32 nbJobs = nbWorkers + 2;
+    int initError;
+    DEBUGLOG(3, "ZSTDMT_createCCtx_advanced (nbWorkers = %u)", nbWorkers);
+
+    if (nbWorkers < 1) return NULL;
+    nbWorkers = MIN(nbWorkers , ZSTDMT_NBWORKERS_MAX);
+    if ((cMem.customAlloc!=NULL) ^ (cMem.customFree!=NULL))
+        /* invalid custom allocator */
+        return NULL;
+
+    mtctx = (ZSTDMT_CCtx*) ZSTD_customCalloc(sizeof(ZSTDMT_CCtx), cMem);
+    if (!mtctx) return NULL;
+    ZSTDMT_CCtxParam_setNbWorkers(&mtctx->params, nbWorkers);
+    mtctx->cMem = cMem;
+    mtctx->allJobsCompleted = 1;
+    if (pool != NULL) {
+      mtctx->factory = pool;
+      mtctx->providedFactory = 1;
+    }
+    else {
+      mtctx->factory = POOL_create_advanced(nbWorkers, 0, cMem);
+      mtctx->providedFactory = 0;
+    }
+    mtctx->jobs = ZSTDMT_createJobsTable(&nbJobs, cMem);
+    assert(nbJobs > 0); assert((nbJobs & (nbJobs - 1)) == 0);  /* ensure nbJobs is a power of 2 */
+    mtctx->jobIDMask = nbJobs - 1;
+    mtctx->bufPool = ZSTDMT_createBufferPool(BUF_POOL_MAX_NB_BUFFERS(nbWorkers), cMem);
+    mtctx->cctxPool = ZSTDMT_createCCtxPool(nbWorkers, cMem);
+    mtctx->seqPool = ZSTDMT_createSeqPool(nbWorkers, cMem);
+    initError = ZSTDMT_serialState_init(&mtctx->serial);
+    mtctx->roundBuff = kNullRoundBuff;
+    if (!mtctx->factory | !mtctx->jobs | !mtctx->bufPool | !mtctx->cctxPool | !mtctx->seqPool | initError) {
+        ZSTDMT_freeCCtx(mtctx);
+        return NULL;
+    }
+    DEBUGLOG(3, "mt_cctx created, for %u threads", nbWorkers);
+    return mtctx;
+}
+
+ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers, ZSTD_customMem cMem, ZSTD_threadPool* pool)
+{
+#ifdef ZSTD_MULTITHREAD
+    return ZSTDMT_createCCtx_advanced_internal(nbWorkers, cMem, pool);
+#else
+    (void)nbWorkers;
+    (void)cMem;
+    (void)pool;
+    return NULL;
+#endif
+}
+
+
+/* ZSTDMT_releaseAllJobResources() :
+ * note : ensure all workers are killed first ! */
+static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
+{
+    unsigned jobID;
+    DEBUGLOG(3, "ZSTDMT_releaseAllJobResources");
+    for (jobID=0; jobID <= mtctx->jobIDMask; jobID++) {
+        /* Copy the mutex/cond out */
+        ZSTD_pthread_mutex_t const mutex = mtctx->jobs[jobID].job_mutex;
+        ZSTD_pthread_cond_t const cond = mtctx->jobs[jobID].job_cond;
+
+        DEBUGLOG(4, "job%02u: release dst address %08X", jobID, (U32)(size_t)mtctx->jobs[jobID].dstBuff.start);
+        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
+
+        /* Clear the job description, but keep the mutex/cond */
+        ZSTD_memset(&mtctx->jobs[jobID], 0, sizeof(mtctx->jobs[jobID]));
+        mtctx->jobs[jobID].job_mutex = mutex;
+        mtctx->jobs[jobID].job_cond = cond;
+    }
+    mtctx->inBuff.buffer = g_nullBuffer;
+    mtctx->inBuff.filled = 0;
+    mtctx->allJobsCompleted = 1;
+}
+
+static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* mtctx)
+{
+    DEBUGLOG(4, "ZSTDMT_waitForAllJobsCompleted");
+    while (mtctx->doneJobID < mtctx->nextJobID) {
+        unsigned const jobID = mtctx->doneJobID & mtctx->jobIDMask;
+        ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[jobID].job_mutex);
+        while (mtctx->jobs[jobID].consumed < mtctx->jobs[jobID].src.size) {
+            DEBUGLOG(4, "waiting for jobCompleted signal from job %u", mtctx->doneJobID);   /* we want to block when waiting for data to flush */
+            ZSTD_pthread_cond_wait(&mtctx->jobs[jobID].job_cond, &mtctx->jobs[jobID].job_mutex);
+        }
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[jobID].job_mutex);
+        mtctx->doneJobID++;
+    }
+}
+
+size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
+{
+    if (mtctx==NULL) return 0;   /* compatible with free on NULL */
+    if (!mtctx->providedFactory)
+        POOL_free(mtctx->factory);   /* stop and free worker threads */
+    ZSTDMT_releaseAllJobResources(mtctx);  /* release job resources into pools first */
+    ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
+    ZSTDMT_freeBufferPool(mtctx->bufPool);
+    ZSTDMT_freeCCtxPool(mtctx->cctxPool);
+    ZSTDMT_freeSeqPool(mtctx->seqPool);
+    ZSTDMT_serialState_free(&mtctx->serial);
+    ZSTD_freeCDict(mtctx->cdictLocal);
+    if (mtctx->roundBuff.buffer)
+        ZSTD_customFree(mtctx->roundBuff.buffer, mtctx->cMem);
+    ZSTD_customFree(mtctx, mtctx->cMem);
+    return 0;
+}
+
+size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx)
+{
+    if (mtctx == NULL) return 0;   /* supports sizeof NULL */
+    return sizeof(*mtctx)
+            + POOL_sizeof(mtctx->factory)
+            + ZSTDMT_sizeof_bufferPool(mtctx->bufPool)
+            + (mtctx->jobIDMask+1) * sizeof(ZSTDMT_jobDescription)
+            + ZSTDMT_sizeof_CCtxPool(mtctx->cctxPool)
+            + ZSTDMT_sizeof_seqPool(mtctx->seqPool)
+            + ZSTD_sizeof_CDict(mtctx->cdictLocal)
+            + mtctx->roundBuff.capacity;
+}
+
+
+/* ZSTDMT_resize() :
+ * @return : error code if fails, 0 on success */
+static size_t ZSTDMT_resize(ZSTDMT_CCtx* mtctx, unsigned nbWorkers)
+{
+    if (POOL_resize(mtctx->factory, nbWorkers)) return ERROR(memory_allocation);
+    FORWARD_IF_ERROR( ZSTDMT_expandJobsTable(mtctx, nbWorkers) , "");
+    mtctx->bufPool = ZSTDMT_expandBufferPool(mtctx->bufPool, BUF_POOL_MAX_NB_BUFFERS(nbWorkers));
+    if (mtctx->bufPool == NULL) return ERROR(memory_allocation);
+    mtctx->cctxPool = ZSTDMT_expandCCtxPool(mtctx->cctxPool, nbWorkers);
+    if (mtctx->cctxPool == NULL) return ERROR(memory_allocation);
+    mtctx->seqPool = ZSTDMT_expandSeqPool(mtctx->seqPool, nbWorkers);
+    if (mtctx->seqPool == NULL) return ERROR(memory_allocation);
+    ZSTDMT_CCtxParam_setNbWorkers(&mtctx->params, nbWorkers);
+    return 0;
+}
+
+
+/*! ZSTDMT_updateCParams_whileCompressing() :
+ *  Updates a selected set of compression parameters, remaining compatible with currently active frame.
+ *  New parameters will be applied to next compression job. */
+void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_params* cctxParams)
+{
+    U32 const saved_wlog = mtctx->params.cParams.windowLog;   /* Do not modify windowLog while compressing */
+    int const compressionLevel = cctxParams->compressionLevel;
+    DEBUGLOG(5, "ZSTDMT_updateCParams_whileCompressing (level:%i)",
+                compressionLevel);
+    mtctx->params.compressionLevel = compressionLevel;
+    {   ZSTD_compressionParameters cParams = ZSTD_getCParamsFromCCtxParams(cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
+        cParams.windowLog = saved_wlog;
+        mtctx->params.cParams = cParams;
+    }
+}
+
+/* ZSTDMT_getFrameProgression():
+ * tells how much data has been consumed (input) and produced (output) for current frame.
+ * able to count progression inside worker threads.
+ * Note : mutex will be acquired during statistics collection inside workers. */
+ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx)
+{
+    ZSTD_frameProgression fps;
+    DEBUGLOG(5, "ZSTDMT_getFrameProgression");
+    fps.ingested = mtctx->consumed + mtctx->inBuff.filled;
+    fps.consumed = mtctx->consumed;
+    fps.produced = fps.flushed = mtctx->produced;
+    fps.currentJobID = mtctx->nextJobID;
+    fps.nbActiveWorkers = 0;
+    {   unsigned jobNb;
+        unsigned lastJobNb = mtctx->nextJobID + mtctx->jobReady; assert(mtctx->jobReady <= 1);
+        DEBUGLOG(6, "ZSTDMT_getFrameProgression: jobs: from %u to <%u (jobReady:%u)",
+                    mtctx->doneJobID, lastJobNb, mtctx->jobReady);
+        for (jobNb = mtctx->doneJobID ; jobNb < lastJobNb ; jobNb++) {
+            unsigned const wJobID = jobNb & mtctx->jobIDMask;
+            ZSTDMT_jobDescription* jobPtr = &mtctx->jobs[wJobID];
+            ZSTD_pthread_mutex_lock(&jobPtr->job_mutex);
+            {   size_t const cResult = jobPtr->cSize;
+                size_t const produced = ZSTD_isError(cResult) ? 0 : cResult;
+                size_t const flushed = ZSTD_isError(cResult) ? 0 : jobPtr->dstFlushed;
+                assert(flushed <= produced);
+                fps.ingested += jobPtr->src.size;
+                fps.consumed += jobPtr->consumed;
+                fps.produced += produced;
+                fps.flushed  += flushed;
+                fps.nbActiveWorkers += (jobPtr->consumed < jobPtr->src.size);
+            }
+            ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+        }
+    }
+    return fps;
+}
+
+
+size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx)
+{
+    size_t toFlush;
+    unsigned const jobID = mtctx->doneJobID;
+    assert(jobID <= mtctx->nextJobID);
+    if (jobID == mtctx->nextJobID) return 0;   /* no active job => nothing to flush */
+
+    /* look into oldest non-fully-flushed job */
+    {   unsigned const wJobID = jobID & mtctx->jobIDMask;
+        ZSTDMT_jobDescription* const jobPtr = &mtctx->jobs[wJobID];
+        ZSTD_pthread_mutex_lock(&jobPtr->job_mutex);
+        {   size_t const cResult = jobPtr->cSize;
+            size_t const produced = ZSTD_isError(cResult) ? 0 : cResult;
+            size_t const flushed = ZSTD_isError(cResult) ? 0 : jobPtr->dstFlushed;
+            assert(flushed <= produced);
+            assert(jobPtr->consumed <= jobPtr->src.size);
+            toFlush = produced - flushed;
+            /* if toFlush==0, nothing is available to flush.
+             * However, jobID is expected to still be active:
+             * if jobID was already completed and fully flushed,
+             * ZSTDMT_flushProduced() should have already moved onto next job.
+             * Therefore, some input has not yet been consumed. */
+            if (toFlush==0) {
+                assert(jobPtr->consumed < jobPtr->src.size);
+            }
+        }
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+    }
+
+    return toFlush;
+}
+
+
+/* ------------------------------------------ */
+/* =====   Multi-threaded compression   ===== */
+/* ------------------------------------------ */
+
+static unsigned ZSTDMT_computeTargetJobLog(const ZSTD_CCtx_params* params)
+{
+    unsigned jobLog;
+    if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
+        /* In Long Range Mode, the windowLog is typically oversized.
+         * In which case, it's preferable to determine the jobSize
+         * based on cycleLog instead. */
+        jobLog = MAX(21, ZSTD_cycleLog(params->cParams.chainLog, params->cParams.strategy) + 3);
+    } else {
+        jobLog = MAX(20, params->cParams.windowLog + 2);
+    }
+    return MIN(jobLog, (unsigned)ZSTDMT_JOBLOG_MAX);
+}
+
+static int ZSTDMT_overlapLog_default(ZSTD_strategy strat)
+{
+    switch(strat)
+    {
+        case ZSTD_btultra2:
+            return 9;
+        case ZSTD_btultra:
+        case ZSTD_btopt:
+            return 8;
+        case ZSTD_btlazy2:
+        case ZSTD_lazy2:
+            return 7;
+        case ZSTD_lazy:
+        case ZSTD_greedy:
+        case ZSTD_dfast:
+        case ZSTD_fast:
+        default:;
+    }
+    return 6;
+}
+
+static int ZSTDMT_overlapLog(int ovlog, ZSTD_strategy strat)
+{
+    assert(0 <= ovlog && ovlog <= 9);
+    if (ovlog == 0) return ZSTDMT_overlapLog_default(strat);
+    return ovlog;
+}
+
+static size_t ZSTDMT_computeOverlapSize(const ZSTD_CCtx_params* params)
+{
+    int const overlapRLog = 9 - ZSTDMT_overlapLog(params->overlapLog, params->cParams.strategy);
+    int ovLog = (overlapRLog >= 8) ? 0 : (params->cParams.windowLog - overlapRLog);
+    assert(0 <= overlapRLog && overlapRLog <= 8);
+    if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
+        /* In Long Range Mode, the windowLog is typically oversized.
+         * In which case, it's preferable to determine the jobSize
+         * based on chainLog instead.
+         * Then, ovLog becomes a fraction of the jobSize, rather than windowSize */
+        ovLog = MIN(params->cParams.windowLog, ZSTDMT_computeTargetJobLog(params) - 2)
+                - overlapRLog;
+    }
+    assert(0 <= ovLog && ovLog <= ZSTD_WINDOWLOG_MAX);
+    DEBUGLOG(4, "overlapLog : %i", params->overlapLog);
+    DEBUGLOG(4, "overlap size : %i", 1 << ovLog);
+    return (ovLog==0) ? 0 : (size_t)1 << ovLog;
+}
+
+/* ====================================== */
+/* =======      Streaming API     ======= */
+/* ====================================== */
+
+size_t ZSTDMT_initCStream_internal(
+        ZSTDMT_CCtx* mtctx,
+        const void* dict, size_t dictSize, ZSTD_dictContentType_e dictContentType,
+        const ZSTD_CDict* cdict, ZSTD_CCtx_params params,
+        unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTDMT_initCStream_internal (pledgedSrcSize=%u, nbWorkers=%u, cctxPool=%u)",
+                (U32)pledgedSrcSize, params.nbWorkers, mtctx->cctxPool->totalCCtx);
+
+    /* params supposed partially fully validated at this point */
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+
+    /* init */
+    if (params.nbWorkers != mtctx->params.nbWorkers)
+        FORWARD_IF_ERROR( ZSTDMT_resize(mtctx, (unsigned)params.nbWorkers) , "");
+
+    if (params.jobSize != 0 && params.jobSize < ZSTDMT_JOBSIZE_MIN) params.jobSize = ZSTDMT_JOBSIZE_MIN;
+    if (params.jobSize > (size_t)ZSTDMT_JOBSIZE_MAX) params.jobSize = (size_t)ZSTDMT_JOBSIZE_MAX;
+
+    if (mtctx->allJobsCompleted == 0) {   /* previous compression not correctly finished */
+        ZSTDMT_waitForAllJobsCompleted(mtctx);
+        ZSTDMT_releaseAllJobResources(mtctx);
+        mtctx->allJobsCompleted = 1;
+    }
+
+    mtctx->params = params;
+    mtctx->frameContentSize = pledgedSrcSize;
+    ZSTD_freeCDict(mtctx->cdictLocal);
+    if (dict) {
+        mtctx->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
+                                                    ZSTD_dlm_byCopy, dictContentType, /* note : a loadPrefix becomes an internal CDict */
+                                                    params.cParams, mtctx->cMem);
+        mtctx->cdict = mtctx->cdictLocal;
+        if (mtctx->cdictLocal == NULL) return ERROR(memory_allocation);
+    } else {
+        mtctx->cdictLocal = NULL;
+        mtctx->cdict = cdict;
+    }
+
+    mtctx->targetPrefixSize = ZSTDMT_computeOverlapSize(&params);
+    DEBUGLOG(4, "overlapLog=%i => %u KB", params.overlapLog, (U32)(mtctx->targetPrefixSize>>10));
+    mtctx->targetSectionSize = params.jobSize;
+    if (mtctx->targetSectionSize == 0) {
+        mtctx->targetSectionSize = 1ULL << ZSTDMT_computeTargetJobLog(&params);
+    }
+    assert(mtctx->targetSectionSize <= (size_t)ZSTDMT_JOBSIZE_MAX);
+
+    if (params.rsyncable) {
+        /* Aim for the targetsectionSize as the average job size. */
+        U32 const jobSizeKB = (U32)(mtctx->targetSectionSize >> 10);
+        U32 const rsyncBits = (assert(jobSizeKB >= 1), ZSTD_highbit32(jobSizeKB) + 10);
+        /* We refuse to create jobs < RSYNC_MIN_BLOCK_SIZE bytes, so make sure our
+         * expected job size is at least 4x larger. */
+        assert(rsyncBits >= RSYNC_MIN_BLOCK_LOG + 2);
+        DEBUGLOG(4, "rsyncLog = %u", rsyncBits);
+        mtctx->rsync.hash = 0;
+        mtctx->rsync.hitMask = (1ULL << rsyncBits) - 1;
+        mtctx->rsync.primePower = ZSTD_rollingHash_primePower(RSYNC_LENGTH);
+    }
+    if (mtctx->targetSectionSize < mtctx->targetPrefixSize) mtctx->targetSectionSize = mtctx->targetPrefixSize;  /* job size must be >= overlap size */
+    DEBUGLOG(4, "Job Size : %u KB (note : set to %u)", (U32)(mtctx->targetSectionSize>>10), (U32)params.jobSize);
+    DEBUGLOG(4, "inBuff Size : %u KB", (U32)(mtctx->targetSectionSize>>10));
+    ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(mtctx->targetSectionSize));
+    {
+        /* If ldm is enabled we need windowSize space. */
+        size_t const windowSize = mtctx->params.ldmParams.enableLdm == ZSTD_ps_enable ? (1U << mtctx->params.cParams.windowLog) : 0;
+        /* Two buffers of slack, plus extra space for the overlap
+         * This is the minimum slack that LDM works with. One extra because
+         * flush might waste up to targetSectionSize-1 bytes. Another extra
+         * for the overlap (if > 0), then one to fill which doesn't overlap
+         * with the LDM window.
+         */
+        size_t const nbSlackBuffers = 2 + (mtctx->targetPrefixSize > 0);
+        size_t const slackSize = mtctx->targetSectionSize * nbSlackBuffers;
+        /* Compute the total size, and always have enough slack */
+        size_t const nbWorkers = MAX(mtctx->params.nbWorkers, 1);
+        size_t const sectionsSize = mtctx->targetSectionSize * nbWorkers;
+        size_t const capacity = MAX(windowSize, sectionsSize) + slackSize;
+        if (mtctx->roundBuff.capacity < capacity) {
+            if (mtctx->roundBuff.buffer)
+                ZSTD_customFree(mtctx->roundBuff.buffer, mtctx->cMem);
+            mtctx->roundBuff.buffer = (BYTE*)ZSTD_customMalloc(capacity, mtctx->cMem);
+            if (mtctx->roundBuff.buffer == NULL) {
+                mtctx->roundBuff.capacity = 0;
+                return ERROR(memory_allocation);
+            }
+            mtctx->roundBuff.capacity = capacity;
+        }
+    }
+    DEBUGLOG(4, "roundBuff capacity : %u KB", (U32)(mtctx->roundBuff.capacity>>10));
+    mtctx->roundBuff.pos = 0;
+    mtctx->inBuff.buffer = g_nullBuffer;
+    mtctx->inBuff.filled = 0;
+    mtctx->inBuff.prefix = kNullRange;
+    mtctx->doneJobID = 0;
+    mtctx->nextJobID = 0;
+    mtctx->frameEnded = 0;
+    mtctx->allJobsCompleted = 0;
+    mtctx->consumed = 0;
+    mtctx->produced = 0;
+
+    /* update dictionary */
+    ZSTD_freeCDict(mtctx->cdictLocal);
+    mtctx->cdictLocal = NULL;
+    mtctx->cdict = NULL;
+    if (dict) {
+        if (dictContentType == ZSTD_dct_rawContent) {
+            mtctx->inBuff.prefix.start = (const BYTE*)dict;
+            mtctx->inBuff.prefix.size = dictSize;
+        } else {
+            /* note : a loadPrefix becomes an internal CDict */
+            mtctx->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
+                                                        ZSTD_dlm_byRef, dictContentType,
+                                                        params.cParams, mtctx->cMem);
+            mtctx->cdict = mtctx->cdictLocal;
+            if (mtctx->cdictLocal == NULL) return ERROR(memory_allocation);
+        }
+    } else {
+        mtctx->cdict = cdict;
+    }
+
+    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params, mtctx->targetSectionSize,
+                                 dict, dictSize, dictContentType))
+        return ERROR(memory_allocation);
+
+
+    return 0;
+}
+
+
+/* ZSTDMT_writeLastEmptyBlock()
+ * Write a single empty block with an end-of-frame to finish a frame.
+ * Job must be created from streaming variant.
+ * This function is always successful if expected conditions are fulfilled.
+ */
+static void ZSTDMT_writeLastEmptyBlock(ZSTDMT_jobDescription* job)
+{
+    assert(job->lastJob == 1);
+    assert(job->src.size == 0);   /* last job is empty -> will be simplified into a last empty block */
+    assert(job->firstJob == 0);   /* cannot be first job, as it also needs to create frame header */
+    assert(job->dstBuff.start == NULL);   /* invoked from streaming variant only (otherwise, dstBuff might be user's output) */
+    job->dstBuff = ZSTDMT_getBuffer(job->bufPool);
+    if (job->dstBuff.start == NULL) {
+      job->cSize = ERROR(memory_allocation);
+      return;
+    }
+    assert(job->dstBuff.capacity >= ZSTD_blockHeaderSize);   /* no buffer should ever be that small */
+    job->src = kNullRange;
+    job->cSize = ZSTD_writeLastEmptyBlock(job->dstBuff.start, job->dstBuff.capacity);
+    assert(!ZSTD_isError(job->cSize));
+    assert(job->consumed == 0);
+}
+
+static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* mtctx, size_t srcSize, ZSTD_EndDirective endOp)
+{
+    unsigned const jobID = mtctx->nextJobID & mtctx->jobIDMask;
+    int const endFrame = (endOp == ZSTD_e_end);
+
+    if (mtctx->nextJobID > mtctx->doneJobID + mtctx->jobIDMask) {
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: will not create new job : table is full");
+        assert((mtctx->nextJobID & mtctx->jobIDMask) == (mtctx->doneJobID & mtctx->jobIDMask));
+        return 0;
+    }
+
+    if (!mtctx->jobReady) {
+        BYTE const* src = (BYTE const*)mtctx->inBuff.buffer.start;
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: preparing job %u to compress %u bytes with %u preload ",
+                    mtctx->nextJobID, (U32)srcSize, (U32)mtctx->inBuff.prefix.size);
+        mtctx->jobs[jobID].src.start = src;
+        mtctx->jobs[jobID].src.size = srcSize;
+        assert(mtctx->inBuff.filled >= srcSize);
+        mtctx->jobs[jobID].prefix = mtctx->inBuff.prefix;
+        mtctx->jobs[jobID].consumed = 0;
+        mtctx->jobs[jobID].cSize = 0;
+        mtctx->jobs[jobID].params = mtctx->params;
+        mtctx->jobs[jobID].cdict = mtctx->nextJobID==0 ? mtctx->cdict : NULL;
+        mtctx->jobs[jobID].fullFrameSize = mtctx->frameContentSize;
+        mtctx->jobs[jobID].dstBuff = g_nullBuffer;
+        mtctx->jobs[jobID].cctxPool = mtctx->cctxPool;
+        mtctx->jobs[jobID].bufPool = mtctx->bufPool;
+        mtctx->jobs[jobID].seqPool = mtctx->seqPool;
+        mtctx->jobs[jobID].serial = &mtctx->serial;
+        mtctx->jobs[jobID].jobID = mtctx->nextJobID;
+        mtctx->jobs[jobID].firstJob = (mtctx->nextJobID==0);
+        mtctx->jobs[jobID].lastJob = endFrame;
+        mtctx->jobs[jobID].frameChecksumNeeded = mtctx->params.fParams.checksumFlag && endFrame && (mtctx->nextJobID>0);
+        mtctx->jobs[jobID].dstFlushed = 0;
+
+        /* Update the round buffer pos and clear the input buffer to be reset */
+        mtctx->roundBuff.pos += srcSize;
+        mtctx->inBuff.buffer = g_nullBuffer;
+        mtctx->inBuff.filled = 0;
+        /* Set the prefix for next job */
+        if (!endFrame) {
+            size_t const newPrefixSize = MIN(srcSize, mtctx->targetPrefixSize);
+            mtctx->inBuff.prefix.start = src + srcSize - newPrefixSize;
+            mtctx->inBuff.prefix.size = newPrefixSize;
+        } else {   /* endFrame==1 => no need for another input buffer */
+            mtctx->inBuff.prefix = kNullRange;
+            mtctx->frameEnded = endFrame;
+            if (mtctx->nextJobID == 0) {
+                /* single job exception : checksum is already calculated directly within worker thread */
+                mtctx->params.fParams.checksumFlag = 0;
+        }   }
+
+        if ( (srcSize == 0)
+          && (mtctx->nextJobID>0)/*single job must also write frame header*/ ) {
+            DEBUGLOG(5, "ZSTDMT_createCompressionJob: creating a last empty block to end frame");
+            assert(endOp == ZSTD_e_end);  /* only possible case : need to end the frame with an empty last block */
+            ZSTDMT_writeLastEmptyBlock(mtctx->jobs + jobID);
+            mtctx->nextJobID++;
+            return 0;
+        }
+    }
+
+    DEBUGLOG(5, "ZSTDMT_createCompressionJob: posting job %u : %u bytes  (end:%u, jobNb == %u (mod:%u))",
+                mtctx->nextJobID,
+                (U32)mtctx->jobs[jobID].src.size,
+                mtctx->jobs[jobID].lastJob,
+                mtctx->nextJobID,
+                jobID);
+    if (POOL_tryAdd(mtctx->factory, ZSTDMT_compressionJob, &mtctx->jobs[jobID])) {
+        mtctx->nextJobID++;
+        mtctx->jobReady = 0;
+    } else {
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: no worker available for job %u", mtctx->nextJobID);
+        mtctx->jobReady = 1;
+    }
+    return 0;
+}
+
+
+/*! ZSTDMT_flushProduced() :
+ *  flush whatever data has been produced but not yet flushed in current job.
+ *  move to next job if current one is fully flushed.
+ * `output` : `pos` will be updated with amount of data flushed .
+ * `blockToFlush` : if >0, the function will block and wait if there is no data available to flush .
+ * @return : amount of data remaining within internal buffer, 0 if no more, 1 if unknown but > 0, or an error code */
+static size_t ZSTDMT_flushProduced(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, unsigned blockToFlush, ZSTD_EndDirective end)
+{
+    unsigned const wJobID = mtctx->doneJobID & mtctx->jobIDMask;
+    DEBUGLOG(5, "ZSTDMT_flushProduced (blocking:%u , job %u <= %u)",
+                blockToFlush, mtctx->doneJobID, mtctx->nextJobID);
+    assert(output->size >= output->pos);
+
+    ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[wJobID].job_mutex);
+    if (  blockToFlush
+      && (mtctx->doneJobID < mtctx->nextJobID) ) {
+        assert(mtctx->jobs[wJobID].dstFlushed <= mtctx->jobs[wJobID].cSize);
+        while (mtctx->jobs[wJobID].dstFlushed == mtctx->jobs[wJobID].cSize) {  /* nothing to flush */
+            if (mtctx->jobs[wJobID].consumed == mtctx->jobs[wJobID].src.size) {
+                DEBUGLOG(5, "job %u is completely consumed (%u == %u) => don't wait for cond, there will be none",
+                            mtctx->doneJobID, (U32)mtctx->jobs[wJobID].consumed, (U32)mtctx->jobs[wJobID].src.size);
+                break;
+            }
+            DEBUGLOG(5, "waiting for something to flush from job %u (currently flushed: %u bytes)",
+                        mtctx->doneJobID, (U32)mtctx->jobs[wJobID].dstFlushed);
+            ZSTD_pthread_cond_wait(&mtctx->jobs[wJobID].job_cond, &mtctx->jobs[wJobID].job_mutex);  /* block when nothing to flush but some to come */
+    }   }
+
+    /* try to flush something */
+    {   size_t cSize = mtctx->jobs[wJobID].cSize;                  /* shared */
+        size_t const srcConsumed = mtctx->jobs[wJobID].consumed;   /* shared */
+        size_t const srcSize = mtctx->jobs[wJobID].src.size;       /* read-only, could be done after mutex lock, but no-declaration-after-statement */
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+        if (ZSTD_isError(cSize)) {
+            DEBUGLOG(5, "ZSTDMT_flushProduced: job %u : compression error detected : %s",
+                        mtctx->doneJobID, ZSTD_getErrorName(cSize));
+            ZSTDMT_waitForAllJobsCompleted(mtctx);
+            ZSTDMT_releaseAllJobResources(mtctx);
+            return cSize;
+        }
+        /* add frame checksum if necessary (can only happen once) */
+        assert(srcConsumed <= srcSize);
+        if ( (srcConsumed == srcSize)   /* job completed -> worker no longer active */
+          && mtctx->jobs[wJobID].frameChecksumNeeded ) {
+            U32 const checksum = (U32)XXH64_digest(&mtctx->serial.xxhState);
+            DEBUGLOG(4, "ZSTDMT_flushProduced: writing checksum : %08X \n", checksum);
+            MEM_writeLE32((char*)mtctx->jobs[wJobID].dstBuff.start + mtctx->jobs[wJobID].cSize, checksum);
+            cSize += 4;
+            mtctx->jobs[wJobID].cSize += 4;  /* can write this shared value, as worker is no longer active */
+            mtctx->jobs[wJobID].frameChecksumNeeded = 0;
+        }
+
+        if (cSize > 0) {   /* compression is ongoing or completed */
+            size_t const toFlush = MIN(cSize - mtctx->jobs[wJobID].dstFlushed, output->size - output->pos);
+            DEBUGLOG(5, "ZSTDMT_flushProduced: Flushing %u bytes from job %u (completion:%u/%u, generated:%u)",
+                        (U32)toFlush, mtctx->doneJobID, (U32)srcConsumed, (U32)srcSize, (U32)cSize);
+            assert(mtctx->doneJobID < mtctx->nextJobID);
+            assert(cSize >= mtctx->jobs[wJobID].dstFlushed);
+            assert(mtctx->jobs[wJobID].dstBuff.start != NULL);
+            if (toFlush > 0) {
+                ZSTD_memcpy((char*)output->dst + output->pos,
+                    (const char*)mtctx->jobs[wJobID].dstBuff.start + mtctx->jobs[wJobID].dstFlushed,
+                    toFlush);
+            }
+            output->pos += toFlush;
+            mtctx->jobs[wJobID].dstFlushed += toFlush;  /* can write : this value is only used by mtctx */
+
+            if ( (srcConsumed == srcSize)    /* job is completed */
+              && (mtctx->jobs[wJobID].dstFlushed == cSize) ) {   /* output buffer fully flushed => free this job position */
+                DEBUGLOG(5, "Job %u completed (%u bytes), moving to next one",
+                        mtctx->doneJobID, (U32)mtctx->jobs[wJobID].dstFlushed);
+                ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[wJobID].dstBuff);
+                DEBUGLOG(5, "dstBuffer released");
+                mtctx->jobs[wJobID].dstBuff = g_nullBuffer;
+                mtctx->jobs[wJobID].cSize = 0;   /* ensure this job slot is considered "not started" in future check */
+                mtctx->consumed += srcSize;
+                mtctx->produced += cSize;
+                mtctx->doneJobID++;
+        }   }
+
+        /* return value : how many bytes left in buffer ; fake it to 1 when unknown but >0 */
+        if (cSize > mtctx->jobs[wJobID].dstFlushed) return (cSize - mtctx->jobs[wJobID].dstFlushed);
+        if (srcSize > srcConsumed) return 1;   /* current job not completely compressed */
+    }
+    if (mtctx->doneJobID < mtctx->nextJobID) return 1;   /* some more jobs ongoing */
+    if (mtctx->jobReady) return 1;      /* one job is ready to push, just not yet in the list */
+    if (mtctx->inBuff.filled > 0) return 1;   /* input is not empty, and still needs to be converted into a job */
+    mtctx->allJobsCompleted = mtctx->frameEnded;   /* all jobs are entirely flushed => if this one is last one, frame is completed */
+    if (end == ZSTD_e_end) return !mtctx->frameEnded;  /* for ZSTD_e_end, question becomes : is frame completed ? instead of : are internal buffers fully flushed ? */
+    return 0;   /* internal buffers fully flushed */
+}
+
+/**
+ * Returns the range of data used by the earliest job that is not yet complete.
+ * If the data of the first job is broken up into two segments, we cover both
+ * sections.
+ */
+static Range ZSTDMT_getInputDataInUse(ZSTDMT_CCtx* mtctx)
+{
+    unsigned const firstJobID = mtctx->doneJobID;
+    unsigned const lastJobID = mtctx->nextJobID;
+    unsigned jobID;
+
+    /* no need to check during first round */
+    size_t roundBuffCapacity = mtctx->roundBuff.capacity;
+    size_t nbJobs1stRoundMin = roundBuffCapacity / mtctx->targetSectionSize;
+    if (lastJobID < nbJobs1stRoundMin) return kNullRange;
+
+    for (jobID = firstJobID; jobID < lastJobID; ++jobID) {
+        unsigned const wJobID = jobID & mtctx->jobIDMask;
+        size_t consumed;
+
+        ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[wJobID].job_mutex);
+        consumed = mtctx->jobs[wJobID].consumed;
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+
+        if (consumed < mtctx->jobs[wJobID].src.size) {
+            Range range = mtctx->jobs[wJobID].prefix;
+            if (range.size == 0) {
+                /* Empty prefix */
+                range = mtctx->jobs[wJobID].src;
+            }
+            /* Job source in multiple segments not supported yet */
+            assert(range.start <= mtctx->jobs[wJobID].src.start);
+            return range;
+        }
+    }
+    return kNullRange;
+}
+
+/**
+ * Returns non-zero iff buffer and range overlap.
+ */
+static int ZSTDMT_isOverlapped(Buffer buffer, Range range)
+{
+    BYTE const* const bufferStart = (BYTE const*)buffer.start;
+    BYTE const* const rangeStart = (BYTE const*)range.start;
+
+    if (rangeStart == NULL || bufferStart == NULL)
+        return 0;
+
+    {
+        BYTE const* const bufferEnd = bufferStart + buffer.capacity;
+        BYTE const* const rangeEnd = rangeStart + range.size;
+
+        /* Empty ranges cannot overlap */
+        if (bufferStart == bufferEnd || rangeStart == rangeEnd)
+            return 0;
+
+        return bufferStart < rangeEnd && rangeStart < bufferEnd;
+    }
+}
+
+static int ZSTDMT_doesOverlapWindow(Buffer buffer, ZSTD_window_t window)
+{
+    Range extDict;
+    Range prefix;
+
+    DEBUGLOG(5, "ZSTDMT_doesOverlapWindow");
+    extDict.start = window.dictBase + window.lowLimit;
+    extDict.size = window.dictLimit - window.lowLimit;
+
+    prefix.start = window.base + window.dictLimit;
+    prefix.size = window.nextSrc - (window.base + window.dictLimit);
+    DEBUGLOG(5, "extDict [0x%zx, 0x%zx)",
+                (size_t)extDict.start,
+                (size_t)extDict.start + extDict.size);
+    DEBUGLOG(5, "prefix  [0x%zx, 0x%zx)",
+                (size_t)prefix.start,
+                (size_t)prefix.start + prefix.size);
+
+    return ZSTDMT_isOverlapped(buffer, extDict)
+        || ZSTDMT_isOverlapped(buffer, prefix);
+}
+
+static void ZSTDMT_waitForLdmComplete(ZSTDMT_CCtx* mtctx, Buffer buffer)
+{
+    if (mtctx->params.ldmParams.enableLdm == ZSTD_ps_enable) {
+        ZSTD_pthread_mutex_t* mutex = &mtctx->serial.ldmWindowMutex;
+        DEBUGLOG(5, "ZSTDMT_waitForLdmComplete");
+        DEBUGLOG(5, "source  [0x%zx, 0x%zx)",
+                    (size_t)buffer.start,
+                    (size_t)buffer.start + buffer.capacity);
+        ZSTD_PTHREAD_MUTEX_LOCK(mutex);
+        while (ZSTDMT_doesOverlapWindow(buffer, mtctx->serial.ldmWindow)) {
+            DEBUGLOG(5, "Waiting for LDM to finish...");
+            ZSTD_pthread_cond_wait(&mtctx->serial.ldmWindowCond, mutex);
+        }
+        DEBUGLOG(6, "Done waiting for LDM to finish");
+        ZSTD_pthread_mutex_unlock(mutex);
+    }
+}
+
+/**
+ * Attempts to set the inBuff to the next section to fill.
+ * If any part of the new section is still in use we give up.
+ * Returns non-zero if the buffer is filled.
+ */
+static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
+{
+    Range const inUse = ZSTDMT_getInputDataInUse(mtctx);
+    size_t const spaceLeft = mtctx->roundBuff.capacity - mtctx->roundBuff.pos;
+    size_t const spaceNeeded = mtctx->targetSectionSize;
+    Buffer buffer;
+
+    DEBUGLOG(5, "ZSTDMT_tryGetInputRange");
+    assert(mtctx->inBuff.buffer.start == NULL);
+    assert(mtctx->roundBuff.capacity >= spaceNeeded);
+
+    if (spaceLeft < spaceNeeded) {
+        /* ZSTD_invalidateRepCodes() doesn't work for extDict variants.
+         * Simply copy the prefix to the beginning in that case.
+         */
+        BYTE* const start = (BYTE*)mtctx->roundBuff.buffer;
+        size_t const prefixSize = mtctx->inBuff.prefix.size;
+
+        buffer.start = start;
+        buffer.capacity = prefixSize;
+        if (ZSTDMT_isOverlapped(buffer, inUse)) {
+            DEBUGLOG(5, "Waiting for buffer...");
+            return 0;
+        }
+        ZSTDMT_waitForLdmComplete(mtctx, buffer);
+        ZSTD_memmove(start, mtctx->inBuff.prefix.start, prefixSize);
+        mtctx->inBuff.prefix.start = start;
+        mtctx->roundBuff.pos = prefixSize;
+    }
+    buffer.start = mtctx->roundBuff.buffer + mtctx->roundBuff.pos;
+    buffer.capacity = spaceNeeded;
+
+    if (ZSTDMT_isOverlapped(buffer, inUse)) {
+        DEBUGLOG(5, "Waiting for buffer...");
+        return 0;
+    }
+    assert(!ZSTDMT_isOverlapped(buffer, mtctx->inBuff.prefix));
+
+    ZSTDMT_waitForLdmComplete(mtctx, buffer);
+
+    DEBUGLOG(5, "Using prefix range [%zx, %zx)",
+                (size_t)mtctx->inBuff.prefix.start,
+                (size_t)mtctx->inBuff.prefix.start + mtctx->inBuff.prefix.size);
+    DEBUGLOG(5, "Using source range [%zx, %zx)",
+                (size_t)buffer.start,
+                (size_t)buffer.start + buffer.capacity);
+
+
+    mtctx->inBuff.buffer = buffer;
+    mtctx->inBuff.filled = 0;
+    assert(mtctx->roundBuff.pos + buffer.capacity <= mtctx->roundBuff.capacity);
+    return 1;
+}
+
+typedef struct {
+  size_t toLoad;  /* The number of bytes to load from the input. */
+  int flush;      /* Boolean declaring if we must flush because we found a synchronization point. */
+} SyncPoint;
+
+/**
+ * Searches through the input for a synchronization point. If one is found, we
+ * will instruct the caller to flush, and return the number of bytes to load.
+ * Otherwise, we will load as many bytes as possible and instruct the caller
+ * to continue as normal.
+ */
+static SyncPoint
+findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
+{
+    BYTE const* const istart = (BYTE const*)input.src + input.pos;
+    U64 const primePower = mtctx->rsync.primePower;
+    U64 const hitMask = mtctx->rsync.hitMask;
+
+    SyncPoint syncPoint;
+    U64 hash;
+    BYTE const* prev;
+    size_t pos;
+
+    syncPoint.toLoad = MIN(input.size - input.pos, mtctx->targetSectionSize - mtctx->inBuff.filled);
+    syncPoint.flush = 0;
+    if (!mtctx->params.rsyncable)
+        /* Rsync is disabled. */
+        return syncPoint;
+    if (mtctx->inBuff.filled + input.size - input.pos < RSYNC_MIN_BLOCK_SIZE)
+        /* We don't emit synchronization points if it would produce too small blocks.
+         * We don't have enough input to find a synchronization point, so don't look.
+         */
+        return syncPoint;
+    if (mtctx->inBuff.filled + syncPoint.toLoad < RSYNC_LENGTH)
+        /* Not enough to compute the hash.
+         * We will miss any synchronization points in this RSYNC_LENGTH byte
+         * window. However, since it depends only in the internal buffers, if the
+         * state is already synchronized, we will remain synchronized.
+         * Additionally, the probability that we miss a synchronization point is
+         * low: RSYNC_LENGTH / targetSectionSize.
+         */
+        return syncPoint;
+    /* Initialize the loop variables. */
+    if (mtctx->inBuff.filled < RSYNC_MIN_BLOCK_SIZE) {
+        /* We don't need to scan the first RSYNC_MIN_BLOCK_SIZE positions
+         * because they can't possibly be a sync point. So we can start
+         * part way through the input buffer.
+         */
+        pos = RSYNC_MIN_BLOCK_SIZE - mtctx->inBuff.filled;
+        if (pos >= RSYNC_LENGTH) {
+            prev = istart + pos - RSYNC_LENGTH;
+            hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
+        } else {
+            assert(mtctx->inBuff.filled >= RSYNC_LENGTH);
+            prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
+            hash = ZSTD_rollingHash_compute(prev + pos, (RSYNC_LENGTH - pos));
+            hash = ZSTD_rollingHash_append(hash, istart, pos);
+        }
+    } else {
+        /* We have enough bytes buffered to initialize the hash,
+         * and have processed enough bytes to find a sync point.
+         * Start scanning at the beginning of the input.
+         */
+        assert(mtctx->inBuff.filled >= RSYNC_MIN_BLOCK_SIZE);
+        assert(RSYNC_MIN_BLOCK_SIZE >= RSYNC_LENGTH);
+        pos = 0;
+        prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
+        hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
+        if ((hash & hitMask) == hitMask) {
+            /* We're already at a sync point so don't load any more until
+             * we're able to flush this sync point.
+             * This likely happened because the job table was full so we
+             * couldn't add our job.
+             */
+            syncPoint.toLoad = 0;
+            syncPoint.flush = 1;
+            return syncPoint;
+        }
+    }
+    /* Starting with the hash of the previous RSYNC_LENGTH bytes, roll
+     * through the input. If we hit a synchronization point, then cut the
+     * job off, and tell the compressor to flush the job. Otherwise, load
+     * all the bytes and continue as normal.
+     * If we go too long without a synchronization point (targetSectionSize)
+     * then a block will be emitted anyways, but this is okay, since if we
+     * are already synchronized we will remain synchronized.
+     */
+    assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
+    for (; pos < syncPoint.toLoad; ++pos) {
+        BYTE const toRemove = pos < RSYNC_LENGTH ? prev[pos] : istart[pos - RSYNC_LENGTH];
+        /* This assert is very expensive, and Debian compiles with asserts enabled.
+         * So disable it for now. We can get similar coverage by checking it at the
+         * beginning & end of the loop.
+         * assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
+         */
+        hash = ZSTD_rollingHash_rotate(hash, toRemove, istart[pos], primePower);
+        assert(mtctx->inBuff.filled + pos >= RSYNC_MIN_BLOCK_SIZE);
+        if ((hash & hitMask) == hitMask) {
+            syncPoint.toLoad = pos + 1;
+            syncPoint.flush = 1;
+            ++pos; /* for assert */
+            break;
+        }
+    }
+    assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
+    return syncPoint;
+}
+
+size_t ZSTDMT_nextInputSizeHint(const ZSTDMT_CCtx* mtctx)
+{
+    size_t hintInSize = mtctx->targetSectionSize - mtctx->inBuff.filled;
+    if (hintInSize==0) hintInSize = mtctx->targetSectionSize;
+    return hintInSize;
+}
+
+/** ZSTDMT_compressStream_generic() :
+ *  internal use only - exposed to be invoked from zstd_compress.c
+ *  assumption : output and input are valid (pos <= size)
+ * @return : minimum amount of data remaining to flush, 0 if none */
+size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
+                                     ZSTD_outBuffer* output,
+                                     ZSTD_inBuffer* input,
+                                     ZSTD_EndDirective endOp)
+{
+    unsigned forwardInputProgress = 0;
+    DEBUGLOG(5, "ZSTDMT_compressStream_generic (endOp=%u, srcSize=%u)",
+                (U32)endOp, (U32)(input->size - input->pos));
+    assert(output->pos <= output->size);
+    assert(input->pos  <= input->size);
+
+    if ((mtctx->frameEnded) && (endOp==ZSTD_e_continue)) {
+        /* current frame being ended. Only flush/end are allowed */
+        return ERROR(stage_wrong);
+    }
+
+    /* fill input buffer */
+    if ( (!mtctx->jobReady)
+      && (input->size > input->pos) ) {   /* support NULL input */
+        if (mtctx->inBuff.buffer.start == NULL) {
+            assert(mtctx->inBuff.filled == 0); /* Can't fill an empty buffer */
+            if (!ZSTDMT_tryGetInputRange(mtctx)) {
+                /* It is only possible for this operation to fail if there are
+                 * still compression jobs ongoing.
+                 */
+                DEBUGLOG(5, "ZSTDMT_tryGetInputRange failed");
+                assert(mtctx->doneJobID != mtctx->nextJobID);
+            } else
+                DEBUGLOG(5, "ZSTDMT_tryGetInputRange completed successfully : mtctx->inBuff.buffer.start = %p", mtctx->inBuff.buffer.start);
+        }
+        if (mtctx->inBuff.buffer.start != NULL) {
+            SyncPoint const syncPoint = findSynchronizationPoint(mtctx, *input);
+            if (syncPoint.flush && endOp == ZSTD_e_continue) {
+                endOp = ZSTD_e_flush;
+            }
+            assert(mtctx->inBuff.buffer.capacity >= mtctx->targetSectionSize);
+            DEBUGLOG(5, "ZSTDMT_compressStream_generic: adding %u bytes on top of %u to buffer of size %u",
+                        (U32)syncPoint.toLoad, (U32)mtctx->inBuff.filled, (U32)mtctx->targetSectionSize);
+            ZSTD_memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, syncPoint.toLoad);
+            input->pos += syncPoint.toLoad;
+            mtctx->inBuff.filled += syncPoint.toLoad;
+            forwardInputProgress = syncPoint.toLoad>0;
+        }
+    }
+    if ((input->pos < input->size) && (endOp == ZSTD_e_end)) {
+        /* Can't end yet because the input is not fully consumed.
+            * We are in one of these cases:
+            * - mtctx->inBuff is NULL & empty: we couldn't get an input buffer so don't create a new job.
+            * - We filled the input buffer: flush this job but don't end the frame.
+            * - We hit a synchronization point: flush this job but don't end the frame.
+            */
+        assert(mtctx->inBuff.filled == 0 || mtctx->inBuff.filled == mtctx->targetSectionSize || mtctx->params.rsyncable);
+        endOp = ZSTD_e_flush;
+    }
+
+    if ( (mtctx->jobReady)
+      || (mtctx->inBuff.filled >= mtctx->targetSectionSize)  /* filled enough : let's compress */
+      || ((endOp != ZSTD_e_continue) && (mtctx->inBuff.filled > 0))  /* something to flush : let's go */
+      || ((endOp == ZSTD_e_end) && (!mtctx->frameEnded)) ) {   /* must finish the frame with a zero-size block */
+        size_t const jobSize = mtctx->inBuff.filled;
+        assert(mtctx->inBuff.filled <= mtctx->targetSectionSize);
+        FORWARD_IF_ERROR( ZSTDMT_createCompressionJob(mtctx, jobSize, endOp) , "");
+    }
+
+    /* check for potential compressed data ready to be flushed */
+    {   size_t const remainingToFlush = ZSTDMT_flushProduced(mtctx, output, !forwardInputProgress, endOp); /* block if there was no forward input progress */
+        if (input->pos < input->size) return MAX(remainingToFlush, 1);  /* input not consumed : do not end flush yet */
+        DEBUGLOG(5, "end of ZSTDMT_compressStream_generic: remainingToFlush = %u", (U32)remainingToFlush);
+        return remainingToFlush;
+    }
+}
diff --git a/lib/decompress/huf_decompress.c b/lib/decompress/huf_decompress.c
new file mode 100644
index 0000000..c7e3426
--- /dev/null
+++ b/lib/decompress/huf_decompress.c
@@ -0,0 +1,1945 @@
+/* ******************************************************************
+ * huff0 huffman decoder,
+ * part of Finite State Entropy library
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* **************************************************************
+*  Dependencies
+****************************************************************/
+#include <stddef.h>               /* size_t */
+#include "../common/zstd_deps.h"  /* ZSTD_memcpy, ZSTD_memset */
+#include "../common/compiler.h"
+#include "../common/bitstream.h"  /* BIT_* */
+#include "../common/fse.h"        /* to compress headers */
+#include "../common/huf.h"
+#include "../common/error_private.h"
+#include "../common/zstd_internal.h"
+#include "../common/bits.h"       /* ZSTD_highbit32, ZSTD_countTrailingZeros64 */
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+
+#define HUF_DECODER_FAST_TABLELOG 11
+
+/* **************************************************************
+*  Macros
+****************************************************************/
+
+#ifdef HUF_DISABLE_FAST_DECODE
+# define HUF_ENABLE_FAST_DECODE 0
+#else
+# define HUF_ENABLE_FAST_DECODE 1
+#endif
+
+/* These two optional macros force the use one way or another of the two
+ * Huffman decompression implementations. You can't force in both directions
+ * at the same time.
+ */
+#if defined(HUF_FORCE_DECOMPRESS_X1) && \
+    defined(HUF_FORCE_DECOMPRESS_X2)
+#error "Cannot force the use of the X1 and X2 decoders at the same time!"
+#endif
+
+/* When DYNAMIC_BMI2 is enabled, fast decoders are only called when bmi2 is
+ * supported at runtime, so we can add the BMI2 target attribute.
+ * When it is disabled, we will still get BMI2 if it is enabled statically.
+ */
+#if DYNAMIC_BMI2
+# define HUF_FAST_BMI2_ATTRS BMI2_TARGET_ATTRIBUTE
+#else
+# define HUF_FAST_BMI2_ATTRS
+#endif
+
+#ifdef __cplusplus
+# define HUF_EXTERN_C extern "C"
+#else
+# define HUF_EXTERN_C
+#endif
+#define HUF_ASM_DECL HUF_EXTERN_C
+
+#if DYNAMIC_BMI2
+# define HUF_NEED_BMI2_FUNCTION 1
+#else
+# define HUF_NEED_BMI2_FUNCTION 0
+#endif
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_isError ERR_isError
+
+
+/* **************************************************************
+*  Byte alignment for workSpace management
+****************************************************************/
+#define HUF_ALIGN(x, a)         HUF_ALIGN_MASK((x), (a) - 1)
+#define HUF_ALIGN_MASK(x, mask) (((x) + (mask)) & ~(mask))
+
+
+/* **************************************************************
+*  BMI2 Variant Wrappers
+****************************************************************/
+typedef size_t (*HUF_DecompressUsingDTableFn)(void *dst, size_t dstSize,
+                                              const void *cSrc,
+                                              size_t cSrcSize,
+                                              const HUF_DTable *DTable);
+
+#if DYNAMIC_BMI2
+
+#define HUF_DGEN(fn)                                                        \
+                                                                            \
+    static size_t fn##_default(                                             \
+                  void* dst,  size_t dstSize,                               \
+            const void* cSrc, size_t cSrcSize,                              \
+            const HUF_DTable* DTable)                                       \
+    {                                                                       \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }                                                                       \
+                                                                            \
+    static BMI2_TARGET_ATTRIBUTE size_t fn##_bmi2(                          \
+                  void* dst,  size_t dstSize,                               \
+            const void* cSrc, size_t cSrcSize,                              \
+            const HUF_DTable* DTable)                                       \
+    {                                                                       \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }                                                                       \
+                                                                            \
+    static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
+                     size_t cSrcSize, HUF_DTable const* DTable, int flags)  \
+    {                                                                       \
+        if (flags & HUF_flags_bmi2) {                                       \
+            return fn##_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);         \
+        }                                                                   \
+        return fn##_default(dst, dstSize, cSrc, cSrcSize, DTable);          \
+    }
+
+#else
+
+#define HUF_DGEN(fn)                                                        \
+    static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
+                     size_t cSrcSize, HUF_DTable const* DTable, int flags)  \
+    {                                                                       \
+        (void)flags;                                                        \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }
+
+#endif
+
+
+/*-***************************/
+/*  generic DTableDesc       */
+/*-***************************/
+typedef struct { BYTE maxTableLog; BYTE tableType; BYTE tableLog; BYTE reserved; } DTableDesc;
+
+static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
+{
+    DTableDesc dtd;
+    ZSTD_memcpy(&dtd, table, sizeof(dtd));
+    return dtd;
+}
+
+static size_t HUF_initFastDStream(BYTE const* ip) {
+    BYTE const lastByte = ip[7];
+    size_t const bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0;
+    size_t const value = MEM_readLEST(ip) | 1;
+    assert(bitsConsumed <= 8);
+    assert(sizeof(size_t) == 8);
+    return value << bitsConsumed;
+}
+
+
+/**
+ * The input/output arguments to the Huffman fast decoding loop:
+ *
+ * ip [in/out] - The input pointers, must be updated to reflect what is consumed.
+ * op [in/out] - The output pointers, must be updated to reflect what is written.
+ * bits [in/out] - The bitstream containers, must be updated to reflect the current state.
+ * dt [in] - The decoding table.
+ * ilowest [in] - The beginning of the valid range of the input. Decoders may read
+ *                down to this pointer. It may be below iend[0].
+ * oend [in] - The end of the output stream. op[3] must not cross oend.
+ * iend [in] - The end of each input stream. ip[i] may cross iend[i],
+ *             as long as it is above ilowest, but that indicates corruption.
+ */
+typedef struct {
+    BYTE const* ip[4];
+    BYTE* op[4];
+    U64 bits[4];
+    void const* dt;
+    BYTE const* ilowest;
+    BYTE* oend;
+    BYTE const* iend[4];
+} HUF_DecompressFastArgs;
+
+typedef void (*HUF_DecompressFastLoopFn)(HUF_DecompressFastArgs*);
+
+/**
+ * Initializes args for the fast decoding loop.
+ * @returns 1 on success
+ *          0 if the fallback implementation should be used.
+ *          Or an error code on failure.
+ */
+static size_t HUF_DecompressFastArgs_init(HUF_DecompressFastArgs* args, void* dst, size_t dstSize, void const* src, size_t srcSize, const HUF_DTable* DTable)
+{
+    void const* dt = DTable + 1;
+    U32 const dtLog = HUF_getDTableDesc(DTable).tableLog;
+
+    const BYTE* const istart = (const BYTE*)src;
+
+    BYTE* const oend = (BYTE*)ZSTD_maybeNullPtrAdd(dst, (ptrdiff_t)dstSize);
+
+    /* The fast decoding loop assumes 64-bit little-endian.
+     * This condition is false on x32.
+     */
+    if (!MEM_isLittleEndian() || MEM_32bits())
+        return 0;
+
+    /* Avoid nullptr addition */
+    if (dstSize == 0)
+        return 0;
+    assert(dst != NULL);
+
+    /* strict minimum : jump table + 1 byte per stream */
+    if (srcSize < 10)
+        return ERROR(corruption_detected);
+
+    /* Must have at least 8 bytes per stream because we don't handle initializing smaller bit containers.
+     * If table log is not correct at this point, fallback to the old decoder.
+     * On small inputs we don't have enough data to trigger the fast loop, so use the old decoder.
+     */
+    if (dtLog != HUF_DECODER_FAST_TABLELOG)
+        return 0;
+
+    /* Read the jump table. */
+    {
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = srcSize - (length1 + length2 + length3 + 6);
+        args->iend[0] = istart + 6;  /* jumpTable */
+        args->iend[1] = args->iend[0] + length1;
+        args->iend[2] = args->iend[1] + length2;
+        args->iend[3] = args->iend[2] + length3;
+
+        /* HUF_initFastDStream() requires this, and this small of an input
+         * won't benefit from the ASM loop anyways.
+         */
+        if (length1 < 8 || length2 < 8 || length3 < 8 || length4 < 8)
+            return 0;
+        if (length4 > srcSize) return ERROR(corruption_detected);   /* overflow */
+    }
+    /* ip[] contains the position that is currently loaded into bits[]. */
+    args->ip[0] = args->iend[1] - sizeof(U64);
+    args->ip[1] = args->iend[2] - sizeof(U64);
+    args->ip[2] = args->iend[3] - sizeof(U64);
+    args->ip[3] = (BYTE const*)src + srcSize - sizeof(U64);
+
+    /* op[] contains the output pointers. */
+    args->op[0] = (BYTE*)dst;
+    args->op[1] = args->op[0] + (dstSize+3)/4;
+    args->op[2] = args->op[1] + (dstSize+3)/4;
+    args->op[3] = args->op[2] + (dstSize+3)/4;
+
+    /* No point to call the ASM loop for tiny outputs. */
+    if (args->op[3] >= oend)
+        return 0;
+
+    /* bits[] is the bit container.
+        * It is read from the MSB down to the LSB.
+        * It is shifted left as it is read, and zeros are
+        * shifted in. After the lowest valid bit a 1 is
+        * set, so that CountTrailingZeros(bits[]) can be used
+        * to count how many bits we've consumed.
+        */
+    args->bits[0] = HUF_initFastDStream(args->ip[0]);
+    args->bits[1] = HUF_initFastDStream(args->ip[1]);
+    args->bits[2] = HUF_initFastDStream(args->ip[2]);
+    args->bits[3] = HUF_initFastDStream(args->ip[3]);
+
+    /* The decoders must be sure to never read beyond ilowest.
+     * This is lower than iend[0], but allowing decoders to read
+     * down to ilowest can allow an extra iteration or two in the
+     * fast loop.
+     */
+    args->ilowest = istart;
+
+    args->oend = oend;
+    args->dt = dt;
+
+    return 1;
+}
+
+static size_t HUF_initRemainingDStream(BIT_DStream_t* bit, HUF_DecompressFastArgs const* args, int stream, BYTE* segmentEnd)
+{
+    /* Validate that we haven't overwritten. */
+    if (args->op[stream] > segmentEnd)
+        return ERROR(corruption_detected);
+    /* Validate that we haven't read beyond iend[].
+        * Note that ip[] may be < iend[] because the MSB is
+        * the next bit to read, and we may have consumed 100%
+        * of the stream, so down to iend[i] - 8 is valid.
+        */
+    if (args->ip[stream] < args->iend[stream] - 8)
+        return ERROR(corruption_detected);
+
+    /* Construct the BIT_DStream_t. */
+    assert(sizeof(size_t) == 8);
+    bit->bitContainer = MEM_readLEST(args->ip[stream]);
+    bit->bitsConsumed = ZSTD_countTrailingZeros64(args->bits[stream]);
+    bit->start = (const char*)args->ilowest;
+    bit->limitPtr = bit->start + sizeof(size_t);
+    bit->ptr = (const char*)args->ip[stream];
+
+    return 0;
+}
+
+/* Calls X(N) for each stream 0, 1, 2, 3. */
+#define HUF_4X_FOR_EACH_STREAM(X) \
+    do {                          \
+        X(0);                     \
+        X(1);                     \
+        X(2);                     \
+        X(3);                     \
+    } while (0)
+
+/* Calls X(N, var) for each stream 0, 1, 2, 3. */
+#define HUF_4X_FOR_EACH_STREAM_WITH_VAR(X, var) \
+    do {                                        \
+        X(0, (var));                            \
+        X(1, (var));                            \
+        X(2, (var));                            \
+        X(3, (var));                            \
+    } while (0)
+
+
+#ifndef HUF_FORCE_DECOMPRESS_X2
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+typedef struct { BYTE nbBits; BYTE byte; } HUF_DEltX1;   /* single-symbol decoding */
+
+/**
+ * Packs 4 HUF_DEltX1 structs into a U64. This is used to lay down 4 entries at
+ * a time.
+ */
+static U64 HUF_DEltX1_set4(BYTE symbol, BYTE nbBits) {
+    U64 D4;
+    if (MEM_isLittleEndian()) {
+        D4 = (U64)((symbol << 8) + nbBits);
+    } else {
+        D4 = (U64)(symbol + (nbBits << 8));
+    }
+    assert(D4 < (1U << 16));
+    D4 *= 0x0001000100010001ULL;
+    return D4;
+}
+
+/**
+ * Increase the tableLog to targetTableLog and rescales the stats.
+ * If tableLog > targetTableLog this is a no-op.
+ * @returns New tableLog
+ */
+static U32 HUF_rescaleStats(BYTE* huffWeight, U32* rankVal, U32 nbSymbols, U32 tableLog, U32 targetTableLog)
+{
+    if (tableLog > targetTableLog)
+        return tableLog;
+    if (tableLog < targetTableLog) {
+        U32 const scale = targetTableLog - tableLog;
+        U32 s;
+        /* Increase the weight for all non-zero probability symbols by scale. */
+        for (s = 0; s < nbSymbols; ++s) {
+            huffWeight[s] += (BYTE)((huffWeight[s] == 0) ? 0 : scale);
+        }
+        /* Update rankVal to reflect the new weights.
+         * All weights except 0 get moved to weight + scale.
+         * Weights [1, scale] are empty.
+         */
+        for (s = targetTableLog; s > scale; --s) {
+            rankVal[s] = rankVal[s - scale];
+        }
+        for (s = scale; s > 0; --s) {
+            rankVal[s] = 0;
+        }
+    }
+    return targetTableLog;
+}
+
+typedef struct {
+        U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];
+        U32 rankStart[HUF_TABLELOG_ABSOLUTEMAX + 1];
+        U32 statsWksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
+        BYTE symbols[HUF_SYMBOLVALUE_MAX + 1];
+        BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];
+} HUF_ReadDTableX1_Workspace;
+
+size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags)
+{
+    U32 tableLog = 0;
+    U32 nbSymbols = 0;
+    size_t iSize;
+    void* const dtPtr = DTable + 1;
+    HUF_DEltX1* const dt = (HUF_DEltX1*)dtPtr;
+    HUF_ReadDTableX1_Workspace* wksp = (HUF_ReadDTableX1_Workspace*)workSpace;
+
+    DEBUG_STATIC_ASSERT(HUF_DECOMPRESS_WORKSPACE_SIZE >= sizeof(*wksp));
+    if (sizeof(*wksp) > wkspSize) return ERROR(tableLog_tooLarge);
+
+    DEBUG_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
+    /* ZSTD_memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats_wksp(wksp->huffWeight, HUF_SYMBOLVALUE_MAX + 1, wksp->rankVal, &nbSymbols, &tableLog, src, srcSize, wksp->statsWksp, sizeof(wksp->statsWksp), flags);
+    if (HUF_isError(iSize)) return iSize;
+
+
+    /* Table header */
+    {   DTableDesc dtd = HUF_getDTableDesc(DTable);
+        U32 const maxTableLog = dtd.maxTableLog + 1;
+        U32 const targetTableLog = MIN(maxTableLog, HUF_DECODER_FAST_TABLELOG);
+        tableLog = HUF_rescaleStats(wksp->huffWeight, wksp->rankVal, nbSymbols, tableLog, targetTableLog);
+        if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge);   /* DTable too small, Huffman tree cannot fit in */
+        dtd.tableType = 0;
+        dtd.tableLog = (BYTE)tableLog;
+        ZSTD_memcpy(DTable, &dtd, sizeof(dtd));
+    }
+
+    /* Compute symbols and rankStart given rankVal:
+     *
+     * rankVal already contains the number of values of each weight.
+     *
+     * symbols contains the symbols ordered by weight. First are the rankVal[0]
+     * weight 0 symbols, followed by the rankVal[1] weight 1 symbols, and so on.
+     * symbols[0] is filled (but unused) to avoid a branch.
+     *
+     * rankStart contains the offset where each rank belongs in the DTable.
+     * rankStart[0] is not filled because there are no entries in the table for
+     * weight 0.
+     */
+    {   int n;
+        U32 nextRankStart = 0;
+        int const unroll = 4;
+        int const nLimit = (int)nbSymbols - unroll + 1;
+        for (n=0; n<(int)tableLog+1; n++) {
+            U32 const curr = nextRankStart;
+            nextRankStart += wksp->rankVal[n];
+            wksp->rankStart[n] = curr;
+        }
+        for (n=0; n < nLimit; n += unroll) {
+            int u;
+            for (u=0; u < unroll; ++u) {
+                size_t const w = wksp->huffWeight[n+u];
+                wksp->symbols[wksp->rankStart[w]++] = (BYTE)(n+u);
+            }
+        }
+        for (; n < (int)nbSymbols; ++n) {
+            size_t const w = wksp->huffWeight[n];
+            wksp->symbols[wksp->rankStart[w]++] = (BYTE)n;
+        }
+    }
+
+    /* fill DTable
+     * We fill all entries of each weight in order.
+     * That way length is a constant for each iteration of the outer loop.
+     * We can switch based on the length to a different inner loop which is
+     * optimized for that particular case.
+     */
+    {   U32 w;
+        int symbol = wksp->rankVal[0];
+        int rankStart = 0;
+        for (w=1; w<tableLog+1; ++w) {
+            int const symbolCount = wksp->rankVal[w];
+            int const length = (1 << w) >> 1;
+            int uStart = rankStart;
+            BYTE const nbBits = (BYTE)(tableLog + 1 - w);
+            int s;
+            int u;
+            switch (length) {
+            case 1:
+                for (s=0; s<symbolCount; ++s) {
+                    HUF_DEltX1 D;
+                    D.byte = wksp->symbols[symbol + s];
+                    D.nbBits = nbBits;
+                    dt[uStart] = D;
+                    uStart += 1;
+                }
+                break;
+            case 2:
+                for (s=0; s<symbolCount; ++s) {
+                    HUF_DEltX1 D;
+                    D.byte = wksp->symbols[symbol + s];
+                    D.nbBits = nbBits;
+                    dt[uStart+0] = D;
+                    dt[uStart+1] = D;
+                    uStart += 2;
+                }
+                break;
+            case 4:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    MEM_write64(dt + uStart, D4);
+                    uStart += 4;
+                }
+                break;
+            case 8:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    MEM_write64(dt + uStart, D4);
+                    MEM_write64(dt + uStart + 4, D4);
+                    uStart += 8;
+                }
+                break;
+            default:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    for (u=0; u < length; u += 16) {
+                        MEM_write64(dt + uStart + u + 0, D4);
+                        MEM_write64(dt + uStart + u + 4, D4);
+                        MEM_write64(dt + uStart + u + 8, D4);
+                        MEM_write64(dt + uStart + u + 12, D4);
+                    }
+                    assert(u == length);
+                    uStart += length;
+                }
+                break;
+            }
+            symbol += symbolCount;
+            rankStart += symbolCount * length;
+        }
+    }
+    return iSize;
+}
+
+FORCE_INLINE_TEMPLATE BYTE
+HUF_decodeSymbolX1(BIT_DStream_t* Dstream, const HUF_DEltX1* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+    BYTE const c = dt[val].byte;
+    BIT_skipBits(Dstream, dt[val].nbBits);
+    return c;
+}
+
+#define HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr) \
+    do { *ptr++ = HUF_decodeSymbolX1(DStreamPtr, dt, dtLog); } while (0)
+
+#define HUF_DECODE_SYMBOLX1_1(ptr, DStreamPtr)      \
+    do {                                            \
+        if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
+            HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr); \
+    } while (0)
+
+#define HUF_DECODE_SYMBOLX1_2(ptr, DStreamPtr)      \
+    do {                                            \
+        if (MEM_64bits())                           \
+            HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr); \
+    } while (0)
+
+HINT_INLINE size_t
+HUF_decodeStreamX1(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX1* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    if ((pEnd - p) > 3) {
+        while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-3)) {
+            HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
+            HUF_DECODE_SYMBOLX1_1(p, bitDPtr);
+            HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
+            HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
+        }
+    } else {
+        BIT_reloadDStream(bitDPtr);
+    }
+
+    /* [0-3] symbols remaining */
+    if (MEM_32bits())
+        while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd))
+            HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
+
+    return (size_t)(pEnd-pStart);
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress1X1_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = (BYTE*)ZSTD_maybeNullPtrAdd(op, (ptrdiff_t)dstSize);
+    const void* dtPtr = DTable + 1;
+    const HUF_DEltX1* const dt = (const HUF_DEltX1*)dtPtr;
+    BIT_DStream_t bitD;
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+    U32 const dtLog = dtd.tableLog;
+
+    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
+
+    HUF_decodeStreamX1(op, &bitD, oend, dt, dtLog);
+
+    if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+/* HUF_decompress4X1_usingDTable_internal_body():
+ * Conditions :
+ * @dstSize >= 6
+ */
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress4X1_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);  /* strict minimum : jump table + 1 byte per stream */
+    if (dstSize < 6) return ERROR(corruption_detected);         /* stream 4-split doesn't work */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        BYTE* const olimit = oend - 3;
+        const void* const dtPtr = DTable + 1;
+        const HUF_DEltX1* const dt = (const HUF_DEltX1*)dtPtr;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+        U32 endSignal = 1;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        if (opStart4 > oend) return ERROR(corruption_detected);      /* overflow */
+        assert(dstSize >= 6); /* validated above */
+        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
+        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
+        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
+        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
+
+        /* up to 16 symbols per loop (4 symbols per stream) in 64-bit mode */
+        if ((size_t)(oend - op4) >= sizeof(size_t)) {
+            for ( ; (endSignal) & (op4 < olimit) ; ) {
+                HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX1_1(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_1(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_1(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_1(op4, &bitD4);
+                HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX1_0(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_0(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_0(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_0(op4, &bitD4);
+                endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
+            }
+        }
+
+        /* check corruption */
+        /* note : should not be necessary : op# advance in lock step, and we control op4.
+         *        but curiously, binary generated by gcc 7.2 & 7.3 with -mbmi2 runs faster when >=1 test is present */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX1(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX1(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX1(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX1(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+#if HUF_NEED_BMI2_FUNCTION
+static BMI2_TARGET_ATTRIBUTE
+size_t HUF_decompress4X1_usingDTable_internal_bmi2(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X1_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+#endif
+
+static
+size_t HUF_decompress4X1_usingDTable_internal_default(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X1_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2
+
+HUF_ASM_DECL void HUF_decompress4X1_usingDTable_internal_fast_asm_loop(HUF_DecompressFastArgs* args) ZSTDLIB_HIDDEN;
+
+#endif
+
+static HUF_FAST_BMI2_ATTRS
+void HUF_decompress4X1_usingDTable_internal_fast_c_loop(HUF_DecompressFastArgs* args)
+{
+    U64 bits[4];
+    BYTE const* ip[4];
+    BYTE* op[4];
+    U16 const* const dtable = (U16 const*)args->dt;
+    BYTE* const oend = args->oend;
+    BYTE const* const ilowest = args->ilowest;
+
+    /* Copy the arguments to local variables */
+    ZSTD_memcpy(&bits, &args->bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&ip), &args->ip, sizeof(ip));
+    ZSTD_memcpy(&op, &args->op, sizeof(op));
+
+    assert(MEM_isLittleEndian());
+    assert(!MEM_32bits());
+
+    for (;;) {
+        BYTE* olimit;
+        int stream;
+
+        /* Assert loop preconditions */
+#ifndef NDEBUG
+        for (stream = 0; stream < 4; ++stream) {
+            assert(op[stream] <= (stream == 3 ? oend : op[stream + 1]));
+            assert(ip[stream] >= ilowest);
+        }
+#endif
+        /* Compute olimit */
+        {
+            /* Each iteration produces 5 output symbols per stream */
+            size_t const oiters = (size_t)(oend - op[3]) / 5;
+            /* Each iteration consumes up to 11 bits * 5 = 55 bits < 7 bytes
+             * per stream.
+             */
+            size_t const iiters = (size_t)(ip[0] - ilowest) / 7;
+            /* We can safely run iters iterations before running bounds checks */
+            size_t const iters = MIN(oiters, iiters);
+            size_t const symbols = iters * 5;
+
+            /* We can simply check that op[3] < olimit, instead of checking all
+             * of our bounds, since we can't hit the other bounds until we've run
+             * iters iterations, which only happens when op[3] == olimit.
+             */
+            olimit = op[3] + symbols;
+
+            /* Exit fast decoding loop once we reach the end. */
+            if (op[3] == olimit)
+                break;
+
+            /* Exit the decoding loop if any input pointer has crossed the
+             * previous one. This indicates corruption, and a precondition
+             * to our loop is that ip[i] >= ip[0].
+             */
+            for (stream = 1; stream < 4; ++stream) {
+                if (ip[stream] < ip[stream - 1])
+                    goto _out;
+            }
+        }
+
+#ifndef NDEBUG
+        for (stream = 1; stream < 4; ++stream) {
+            assert(ip[stream] >= ip[stream - 1]);
+        }
+#endif
+
+#define HUF_4X1_DECODE_SYMBOL(_stream, _symbol)                 \
+    do {                                                        \
+        int const index = (int)(bits[(_stream)] >> 53);         \
+        int const entry = (int)dtable[index];                   \
+        bits[(_stream)] <<= (entry & 0x3F);                     \
+        op[(_stream)][(_symbol)] = (BYTE)((entry >> 8) & 0xFF); \
+    } while (0)
+
+#define HUF_4X1_RELOAD_STREAM(_stream)                              \
+    do {                                                            \
+        int const ctz = ZSTD_countTrailingZeros64(bits[(_stream)]); \
+        int const nbBits = ctz & 7;                                 \
+        int const nbBytes = ctz >> 3;                               \
+        op[(_stream)] += 5;                                         \
+        ip[(_stream)] -= nbBytes;                                   \
+        bits[(_stream)] = MEM_read64(ip[(_stream)]) | 1;            \
+        bits[(_stream)] <<= nbBits;                                 \
+    } while (0)
+
+        /* Manually unroll the loop because compilers don't consistently
+         * unroll the inner loops, which destroys performance.
+         */
+        do {
+            /* Decode 5 symbols in each of the 4 streams */
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X1_DECODE_SYMBOL, 0);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X1_DECODE_SYMBOL, 1);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X1_DECODE_SYMBOL, 2);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X1_DECODE_SYMBOL, 3);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X1_DECODE_SYMBOL, 4);
+
+            /* Reload each of the 4 the bitstreams */
+            HUF_4X_FOR_EACH_STREAM(HUF_4X1_RELOAD_STREAM);
+        } while (op[3] < olimit);
+
+#undef HUF_4X1_DECODE_SYMBOL
+#undef HUF_4X1_RELOAD_STREAM
+    }
+
+_out:
+
+    /* Save the final values of each of the state variables back to args. */
+    ZSTD_memcpy(&args->bits, &bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&args->ip), &ip, sizeof(ip));
+    ZSTD_memcpy(&args->op, &op, sizeof(op));
+}
+
+/**
+ * @returns @p dstSize on success (>= 6)
+ *          0 if the fallback implementation should be used
+ *          An error if an error occurred
+ */
+static HUF_FAST_BMI2_ATTRS
+size_t
+HUF_decompress4X1_usingDTable_internal_fast(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable,
+    HUF_DecompressFastLoopFn loopFn)
+{
+    void const* dt = DTable + 1;
+    BYTE const* const ilowest = (BYTE const*)cSrc;
+    BYTE* const oend = (BYTE*)ZSTD_maybeNullPtrAdd(dst, (ptrdiff_t)dstSize);
+    HUF_DecompressFastArgs args;
+    {   size_t const ret = HUF_DecompressFastArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
+        FORWARD_IF_ERROR(ret, "Failed to init fast loop args");
+        if (ret == 0)
+            return 0;
+    }
+
+    assert(args.ip[0] >= args.ilowest);
+    loopFn(&args);
+
+    /* Our loop guarantees that ip[] >= ilowest and that we haven't
+    * overwritten any op[].
+    */
+    assert(args.ip[0] >= ilowest);
+    assert(args.ip[0] >= ilowest);
+    assert(args.ip[1] >= ilowest);
+    assert(args.ip[2] >= ilowest);
+    assert(args.ip[3] >= ilowest);
+    assert(args.op[3] <= oend);
+
+    assert(ilowest == args.ilowest);
+    assert(ilowest + 6 == args.iend[0]);
+    (void)ilowest;
+
+    /* finish bit streams one by one. */
+    {   size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* segmentEnd = (BYTE*)dst;
+        int i;
+        for (i = 0; i < 4; ++i) {
+            BIT_DStream_t bit;
+            if (segmentSize <= (size_t)(oend - segmentEnd))
+                segmentEnd += segmentSize;
+            else
+                segmentEnd = oend;
+            FORWARD_IF_ERROR(HUF_initRemainingDStream(&bit, &args, i, segmentEnd), "corruption");
+            /* Decompress and validate that we've produced exactly the expected length. */
+            args.op[i] += HUF_decodeStreamX1(args.op[i], &bit, segmentEnd, (HUF_DEltX1 const*)dt, HUF_DECODER_FAST_TABLELOG);
+            if (args.op[i] != segmentEnd) return ERROR(corruption_detected);
+        }
+    }
+
+    /* decoded size */
+    assert(dstSize != 0);
+    return dstSize;
+}
+
+HUF_DGEN(HUF_decompress1X1_usingDTable_internal)
+
+static size_t HUF_decompress4X1_usingDTable_internal(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable, int flags)
+{
+    HUF_DecompressUsingDTableFn fallbackFn = HUF_decompress4X1_usingDTable_internal_default;
+    HUF_DecompressFastLoopFn loopFn = HUF_decompress4X1_usingDTable_internal_fast_c_loop;
+
+#if DYNAMIC_BMI2
+    if (flags & HUF_flags_bmi2) {
+        fallbackFn = HUF_decompress4X1_usingDTable_internal_bmi2;
+# if ZSTD_ENABLE_ASM_X86_64_BMI2
+        if (!(flags & HUF_flags_disableAsm)) {
+            loopFn = HUF_decompress4X1_usingDTable_internal_fast_asm_loop;
+        }
+# endif
+    } else {
+        return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
+    }
+#endif
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__)
+    if (!(flags & HUF_flags_disableAsm)) {
+        loopFn = HUF_decompress4X1_usingDTable_internal_fast_asm_loop;
+    }
+#endif
+
+    if (HUF_ENABLE_FAST_DECODE && !(flags & HUF_flags_disableFast)) {
+        size_t const ret = HUF_decompress4X1_usingDTable_internal_fast(dst, dstSize, cSrc, cSrcSize, DTable, loopFn);
+        if (ret != 0)
+            return ret;
+    }
+    return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+static size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int flags)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize, flags);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
+}
+
+#endif /* HUF_FORCE_DECOMPRESS_X2 */
+
+
+#ifndef HUF_FORCE_DECOMPRESS_X1
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX2;  /* double-symbols decoding */
+typedef struct { BYTE symbol; } sortedSymbol_t;
+typedef U32 rankValCol_t[HUF_TABLELOG_MAX + 1];
+typedef rankValCol_t rankVal_t[HUF_TABLELOG_MAX];
+
+/**
+ * Constructs a HUF_DEltX2 in a U32.
+ */
+static U32 HUF_buildDEltX2U32(U32 symbol, U32 nbBits, U32 baseSeq, int level)
+{
+    U32 seq;
+    DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, sequence) == 0);
+    DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, nbBits) == 2);
+    DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, length) == 3);
+    DEBUG_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U32));
+    if (MEM_isLittleEndian()) {
+        seq = level == 1 ? symbol : (baseSeq + (symbol << 8));
+        return seq + (nbBits << 16) + ((U32)level << 24);
+    } else {
+        seq = level == 1 ? (symbol << 8) : ((baseSeq << 8) + symbol);
+        return (seq << 16) + (nbBits << 8) + (U32)level;
+    }
+}
+
+/**
+ * Constructs a HUF_DEltX2.
+ */
+static HUF_DEltX2 HUF_buildDEltX2(U32 symbol, U32 nbBits, U32 baseSeq, int level)
+{
+    HUF_DEltX2 DElt;
+    U32 const val = HUF_buildDEltX2U32(symbol, nbBits, baseSeq, level);
+    DEBUG_STATIC_ASSERT(sizeof(DElt) == sizeof(val));
+    ZSTD_memcpy(&DElt, &val, sizeof(val));
+    return DElt;
+}
+
+/**
+ * Constructs 2 HUF_DEltX2s and packs them into a U64.
+ */
+static U64 HUF_buildDEltX2U64(U32 symbol, U32 nbBits, U16 baseSeq, int level)
+{
+    U32 DElt = HUF_buildDEltX2U32(symbol, nbBits, baseSeq, level);
+    return (U64)DElt + ((U64)DElt << 32);
+}
+
+/**
+ * Fills the DTable rank with all the symbols from [begin, end) that are each
+ * nbBits long.
+ *
+ * @param DTableRank The start of the rank in the DTable.
+ * @param begin The first symbol to fill (inclusive).
+ * @param end The last symbol to fill (exclusive).
+ * @param nbBits Each symbol is nbBits long.
+ * @param tableLog The table log.
+ * @param baseSeq If level == 1 { 0 } else { the first level symbol }
+ * @param level The level in the table. Must be 1 or 2.
+ */
+static void HUF_fillDTableX2ForWeight(
+    HUF_DEltX2* DTableRank,
+    sortedSymbol_t const* begin, sortedSymbol_t const* end,
+    U32 nbBits, U32 tableLog,
+    U16 baseSeq, int const level)
+{
+    U32 const length = 1U << ((tableLog - nbBits) & 0x1F /* quiet static-analyzer */);
+    const sortedSymbol_t* ptr;
+    assert(level >= 1 && level <= 2);
+    switch (length) {
+    case 1:
+        for (ptr = begin; ptr != end; ++ptr) {
+            HUF_DEltX2 const DElt = HUF_buildDEltX2(ptr->symbol, nbBits, baseSeq, level);
+            *DTableRank++ = DElt;
+        }
+        break;
+    case 2:
+        for (ptr = begin; ptr != end; ++ptr) {
+            HUF_DEltX2 const DElt = HUF_buildDEltX2(ptr->symbol, nbBits, baseSeq, level);
+            DTableRank[0] = DElt;
+            DTableRank[1] = DElt;
+            DTableRank += 2;
+        }
+        break;
+    case 4:
+        for (ptr = begin; ptr != end; ++ptr) {
+            U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
+            ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
+            DTableRank += 4;
+        }
+        break;
+    case 8:
+        for (ptr = begin; ptr != end; ++ptr) {
+            U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
+            ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 4, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 6, &DEltX2, sizeof(DEltX2));
+            DTableRank += 8;
+        }
+        break;
+    default:
+        for (ptr = begin; ptr != end; ++ptr) {
+            U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
+            HUF_DEltX2* const DTableRankEnd = DTableRank + length;
+            for (; DTableRank != DTableRankEnd; DTableRank += 8) {
+                ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
+                ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
+                ZSTD_memcpy(DTableRank + 4, &DEltX2, sizeof(DEltX2));
+                ZSTD_memcpy(DTableRank + 6, &DEltX2, sizeof(DEltX2));
+            }
+        }
+        break;
+    }
+}
+
+/* HUF_fillDTableX2Level2() :
+ * `rankValOrigin` must be a table of at least (HUF_TABLELOG_MAX + 1) U32 */
+static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 targetLog, const U32 consumedBits,
+                           const U32* rankVal, const int minWeight, const int maxWeight1,
+                           const sortedSymbol_t* sortedSymbols, U32 const* rankStart,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    /* Fill skipped values (all positions up to rankVal[minWeight]).
+     * These are positions only get a single symbol because the combined weight
+     * is too large.
+     */
+    if (minWeight>1) {
+        U32 const length = 1U << ((targetLog - consumedBits) & 0x1F /* quiet static-analyzer */);
+        U64 const DEltX2 = HUF_buildDEltX2U64(baseSeq, consumedBits, /* baseSeq */ 0, /* level */ 1);
+        int const skipSize = rankVal[minWeight];
+        assert(length > 1);
+        assert((U32)skipSize < length);
+        switch (length) {
+        case 2:
+            assert(skipSize == 1);
+            ZSTD_memcpy(DTable, &DEltX2, sizeof(DEltX2));
+            break;
+        case 4:
+            assert(skipSize <= 4);
+            ZSTD_memcpy(DTable + 0, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTable + 2, &DEltX2, sizeof(DEltX2));
+            break;
+        default:
+            {
+                int i;
+                for (i = 0; i < skipSize; i += 8) {
+                    ZSTD_memcpy(DTable + i + 0, &DEltX2, sizeof(DEltX2));
+                    ZSTD_memcpy(DTable + i + 2, &DEltX2, sizeof(DEltX2));
+                    ZSTD_memcpy(DTable + i + 4, &DEltX2, sizeof(DEltX2));
+                    ZSTD_memcpy(DTable + i + 6, &DEltX2, sizeof(DEltX2));
+                }
+            }
+        }
+    }
+
+    /* Fill each of the second level symbols by weight. */
+    {
+        int w;
+        for (w = minWeight; w < maxWeight1; ++w) {
+            int const begin = rankStart[w];
+            int const end = rankStart[w+1];
+            U32 const nbBits = nbBitsBaseline - w;
+            U32 const totalBits = nbBits + consumedBits;
+            HUF_fillDTableX2ForWeight(
+                DTable + rankVal[w],
+                sortedSymbols + begin, sortedSymbols + end,
+                totalBits, targetLog,
+                baseSeq, /* level */ 2);
+        }
+    }
+}
+
+static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList,
+                           const U32* rankStart, rankValCol_t* rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32* const rankVal = rankValOrigin[0];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    int w;
+    int const wEnd = (int)maxWeight + 1;
+
+    /* Fill DTable in order of weight. */
+    for (w = 1; w < wEnd; ++w) {
+        int const begin = (int)rankStart[w];
+        int const end = (int)rankStart[w+1];
+        U32 const nbBits = nbBitsBaseline - w;
+
+        if (targetLog-nbBits >= minBits) {
+            /* Enough room for a second symbol. */
+            int start = rankVal[w];
+            U32 const length = 1U << ((targetLog - nbBits) & 0x1F /* quiet static-analyzer */);
+            int minWeight = nbBits + scaleLog;
+            int s;
+            if (minWeight < 1) minWeight = 1;
+            /* Fill the DTable for every symbol of weight w.
+             * These symbols get at least 1 second symbol.
+             */
+            for (s = begin; s != end; ++s) {
+                HUF_fillDTableX2Level2(
+                    DTable + start, targetLog, nbBits,
+                    rankValOrigin[nbBits], minWeight, wEnd,
+                    sortedList, rankStart,
+                    nbBitsBaseline, sortedList[s].symbol);
+                start += length;
+            }
+        } else {
+            /* Only a single symbol. */
+            HUF_fillDTableX2ForWeight(
+                DTable + rankVal[w],
+                sortedList + begin, sortedList + end,
+                nbBits, targetLog,
+                /* baseSeq */ 0, /* level */ 1);
+        }
+    }
+}
+
+typedef struct {
+    rankValCol_t rankVal[HUF_TABLELOG_MAX];
+    U32 rankStats[HUF_TABLELOG_MAX + 1];
+    U32 rankStart0[HUF_TABLELOG_MAX + 3];
+    sortedSymbol_t sortedSymbol[HUF_SYMBOLVALUE_MAX + 1];
+    BYTE weightList[HUF_SYMBOLVALUE_MAX + 1];
+    U32 calleeWksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
+} HUF_ReadDTableX2_Workspace;
+
+size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
+                       const void* src, size_t srcSize,
+                             void* workSpace, size_t wkspSize, int flags)
+{
+    U32 tableLog, maxW, nbSymbols;
+    DTableDesc dtd = HUF_getDTableDesc(DTable);
+    U32 maxTableLog = dtd.maxTableLog;
+    size_t iSize;
+    void* dtPtr = DTable+1;   /* force compiler to avoid strict-aliasing */
+    HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
+    U32 *rankStart;
+
+    HUF_ReadDTableX2_Workspace* const wksp = (HUF_ReadDTableX2_Workspace*)workSpace;
+
+    if (sizeof(*wksp) > wkspSize) return ERROR(GENERIC);
+
+    rankStart = wksp->rankStart0 + 1;
+    ZSTD_memset(wksp->rankStats, 0, sizeof(wksp->rankStats));
+    ZSTD_memset(wksp->rankStart0, 0, sizeof(wksp->rankStart0));
+
+    DEBUG_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(HUF_DTable));   /* if compiler fails here, assertion is wrong */
+    if (maxTableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    /* ZSTD_memset(weightList, 0, sizeof(weightList)); */  /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats_wksp(wksp->weightList, HUF_SYMBOLVALUE_MAX + 1, wksp->rankStats, &nbSymbols, &tableLog, src, srcSize, wksp->calleeWksp, sizeof(wksp->calleeWksp), flags);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > maxTableLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+    if (tableLog <= HUF_DECODER_FAST_TABLELOG && maxTableLog > HUF_DECODER_FAST_TABLELOG) maxTableLog = HUF_DECODER_FAST_TABLELOG;
+
+    /* find maxWeight */
+    for (maxW = tableLog; wksp->rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {   U32 w, nextRankStart = 0;
+        for (w=1; w<maxW+1; w++) {
+            U32 curr = nextRankStart;
+            nextRankStart += wksp->rankStats[w];
+            rankStart[w] = curr;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        rankStart[maxW+1] = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {   U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 const w = wksp->weightList[s];
+            U32 const r = rankStart[w]++;
+            wksp->sortedSymbol[r].symbol = (BYTE)s;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {   U32* const rankVal0 = wksp->rankVal[0];
+        {   int const rescale = (maxTableLog-tableLog) - 1;   /* tableLog <= maxTableLog */
+            U32 nextRankVal = 0;
+            U32 w;
+            for (w=1; w<maxW+1; w++) {
+                U32 curr = nextRankVal;
+                nextRankVal += wksp->rankStats[w] << (w+rescale);
+                rankVal0[w] = curr;
+        }   }
+        {   U32 const minBits = tableLog+1 - maxW;
+            U32 consumed;
+            for (consumed = minBits; consumed < maxTableLog - minBits + 1; consumed++) {
+                U32* const rankValPtr = wksp->rankVal[consumed];
+                U32 w;
+                for (w = 1; w < maxW+1; w++) {
+                    rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }   }
+
+    HUF_fillDTableX2(dt, maxTableLog,
+                   wksp->sortedSymbol,
+                   wksp->rankStart0, wksp->rankVal, maxW,
+                   tableLog+1);
+
+    dtd.tableLog = (BYTE)maxTableLog;
+    dtd.tableType = 1;
+    ZSTD_memcpy(DTable, &dtd, sizeof(dtd));
+    return iSize;
+}
+
+
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    ZSTD_memcpy(op, &dt[val].sequence, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeLastSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    ZSTD_memcpy(op, &dt[val].sequence, 1);
+    if (dt[val].length==1) {
+        BIT_skipBits(DStream, dt[val].nbBits);
+    } else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);
+        }
+    }
+    return 1;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    do { ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog); } while (0)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr)                     \
+    do {                                                           \
+        if (MEM_64bits() || (HUF_TABLELOG_MAX<=12))                \
+            ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog); \
+    } while (0)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr)                     \
+    do {                                                           \
+        if (MEM_64bits())                                          \
+            ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog); \
+    } while (0)
+
+HINT_INLINE size_t
+HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd,
+                const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    if ((size_t)(pEnd - p) >= sizeof(bitDPtr->bitContainer)) {
+        if (dtLog <= 11 && MEM_64bits()) {
+            /* up to 10 symbols at a time */
+            while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-9)) {
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+            }
+        } else {
+            /* up to 8 symbols at a time */
+            while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
+                HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+            }
+        }
+    } else {
+        BIT_reloadDStream(bitDPtr);
+    }
+
+    /* closer to end : up to 2 symbols at a time */
+    if ((size_t)(pEnd - p) >= 2) {
+        while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
+            HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+        while (p <= pEnd-2)
+            HUF_DECODE_SYMBOLX2_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+    }
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX2(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress1X2_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    BIT_DStream_t bitD;
+
+    /* Init */
+    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
+
+    /* decode */
+    {   BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = (BYTE*)ZSTD_maybeNullPtrAdd(ostart, (ptrdiff_t)dstSize);
+        const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
+        const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        HUF_decodeStreamX2(ostart, &bitD, oend, dt, dtd.tableLog);
+    }
+
+    /* check */
+    if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+/* HUF_decompress4X2_usingDTable_internal_body():
+ * Conditions:
+ * @dstSize >= 6
+ */
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress4X2_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+    if (dstSize < 6) return ERROR(corruption_detected);         /* stream 4-split doesn't work */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        BYTE* const olimit = oend - (sizeof(size_t)-1);
+        const void* const dtPtr = DTable+1;
+        const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal = 1;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);  /* overflow */
+        if (opStart4 > oend) return ERROR(corruption_detected);     /* overflow */
+        assert(dstSize >= 6 /* validated above */);
+        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
+        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
+        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
+        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        if ((size_t)(oend - op4) >= sizeof(size_t)) {
+            for ( ; (endSignal) & (op4 < olimit); ) {
+#if defined(__clang__) && (defined(__x86_64__) || defined(__i386__))
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+                endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+                endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
+#else
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+                endSignal = (U32)LIKELY((U32)
+                            (BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished)
+                        & (BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished)
+                        & (BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished)
+                        & (BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished));
+#endif
+            }
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+#if HUF_NEED_BMI2_FUNCTION
+static BMI2_TARGET_ATTRIBUTE
+size_t HUF_decompress4X2_usingDTable_internal_bmi2(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X2_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+#endif
+
+static
+size_t HUF_decompress4X2_usingDTable_internal_default(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X2_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2
+
+HUF_ASM_DECL void HUF_decompress4X2_usingDTable_internal_fast_asm_loop(HUF_DecompressFastArgs* args) ZSTDLIB_HIDDEN;
+
+#endif
+
+static HUF_FAST_BMI2_ATTRS
+void HUF_decompress4X2_usingDTable_internal_fast_c_loop(HUF_DecompressFastArgs* args)
+{
+    U64 bits[4];
+    BYTE const* ip[4];
+    BYTE* op[4];
+    BYTE* oend[4];
+    HUF_DEltX2 const* const dtable = (HUF_DEltX2 const*)args->dt;
+    BYTE const* const ilowest = args->ilowest;
+
+    /* Copy the arguments to local registers. */
+    ZSTD_memcpy(&bits, &args->bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&ip), &args->ip, sizeof(ip));
+    ZSTD_memcpy(&op, &args->op, sizeof(op));
+
+    oend[0] = op[1];
+    oend[1] = op[2];
+    oend[2] = op[3];
+    oend[3] = args->oend;
+
+    assert(MEM_isLittleEndian());
+    assert(!MEM_32bits());
+
+    for (;;) {
+        BYTE* olimit;
+        int stream;
+
+        /* Assert loop preconditions */
+#ifndef NDEBUG
+        for (stream = 0; stream < 4; ++stream) {
+            assert(op[stream] <= oend[stream]);
+            assert(ip[stream] >= ilowest);
+        }
+#endif
+        /* Compute olimit */
+        {
+            /* Each loop does 5 table lookups for each of the 4 streams.
+             * Each table lookup consumes up to 11 bits of input, and produces
+             * up to 2 bytes of output.
+             */
+            /* We can consume up to 7 bytes of input per iteration per stream.
+             * We also know that each input pointer is >= ip[0]. So we can run
+             * iters loops before running out of input.
+             */
+            size_t iters = (size_t)(ip[0] - ilowest) / 7;
+            /* Each iteration can produce up to 10 bytes of output per stream.
+             * Each output stream my advance at different rates. So take the
+             * minimum number of safe iterations among all the output streams.
+             */
+            for (stream = 0; stream < 4; ++stream) {
+                size_t const oiters = (size_t)(oend[stream] - op[stream]) / 10;
+                iters = MIN(iters, oiters);
+            }
+
+            /* Each iteration produces at least 5 output symbols. So until
+             * op[3] crosses olimit, we know we haven't executed iters
+             * iterations yet. This saves us maintaining an iters counter,
+             * at the expense of computing the remaining # of iterations
+             * more frequently.
+             */
+            olimit = op[3] + (iters * 5);
+
+            /* Exit the fast decoding loop once we reach the end. */
+            if (op[3] == olimit)
+                break;
+
+            /* Exit the decoding loop if any input pointer has crossed the
+             * previous one. This indicates corruption, and a precondition
+             * to our loop is that ip[i] >= ip[0].
+             */
+            for (stream = 1; stream < 4; ++stream) {
+                if (ip[stream] < ip[stream - 1])
+                    goto _out;
+            }
+        }
+
+#ifndef NDEBUG
+        for (stream = 1; stream < 4; ++stream) {
+            assert(ip[stream] >= ip[stream - 1]);
+        }
+#endif
+
+#define HUF_4X2_DECODE_SYMBOL(_stream, _decode3)                      \
+    do {                                                              \
+        if ((_decode3) || (_stream) != 3) {                           \
+            int const index = (int)(bits[(_stream)] >> 53);           \
+            HUF_DEltX2 const entry = dtable[index];                   \
+            MEM_write16(op[(_stream)], entry.sequence); \
+            bits[(_stream)] <<= (entry.nbBits) & 0x3F;                \
+            op[(_stream)] += (entry.length);                          \
+        }                                                             \
+    } while (0)
+
+#define HUF_4X2_RELOAD_STREAM(_stream)                                  \
+    do {                                                                \
+        HUF_4X2_DECODE_SYMBOL(3, 1);                                    \
+        {                                                               \
+            int const ctz = ZSTD_countTrailingZeros64(bits[(_stream)]); \
+            int const nbBits = ctz & 7;                                 \
+            int const nbBytes = ctz >> 3;                               \
+            ip[(_stream)] -= nbBytes;                                   \
+            bits[(_stream)] = MEM_read64(ip[(_stream)]) | 1;            \
+            bits[(_stream)] <<= nbBits;                                 \
+        }                                                               \
+    } while (0)
+
+        /* Manually unroll the loop because compilers don't consistently
+         * unroll the inner loops, which destroys performance.
+         */
+        do {
+            /* Decode 5 symbols from each of the first 3 streams.
+             * The final stream will be decoded during the reload phase
+             * to reduce register pressure.
+             */
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X2_DECODE_SYMBOL, 0);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X2_DECODE_SYMBOL, 0);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X2_DECODE_SYMBOL, 0);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X2_DECODE_SYMBOL, 0);
+            HUF_4X_FOR_EACH_STREAM_WITH_VAR(HUF_4X2_DECODE_SYMBOL, 0);
+
+            /* Decode one symbol from the final stream */
+            HUF_4X2_DECODE_SYMBOL(3, 1);
+
+            /* Decode 4 symbols from the final stream & reload bitstreams.
+             * The final stream is reloaded last, meaning that all 5 symbols
+             * are decoded from the final stream before it is reloaded.
+             */
+            HUF_4X_FOR_EACH_STREAM(HUF_4X2_RELOAD_STREAM);
+        } while (op[3] < olimit);
+    }
+
+#undef HUF_4X2_DECODE_SYMBOL
+#undef HUF_4X2_RELOAD_STREAM
+
+_out:
+
+    /* Save the final values of each of the state variables back to args. */
+    ZSTD_memcpy(&args->bits, &bits, sizeof(bits));
+    ZSTD_memcpy((void*)(&args->ip), &ip, sizeof(ip));
+    ZSTD_memcpy(&args->op, &op, sizeof(op));
+}
+
+
+static HUF_FAST_BMI2_ATTRS size_t
+HUF_decompress4X2_usingDTable_internal_fast(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable,
+    HUF_DecompressFastLoopFn loopFn) {
+    void const* dt = DTable + 1;
+    const BYTE* const ilowest = (const BYTE*)cSrc;
+    BYTE* const oend = (BYTE*)ZSTD_maybeNullPtrAdd(dst, (ptrdiff_t)dstSize);
+    HUF_DecompressFastArgs args;
+    {
+        size_t const ret = HUF_DecompressFastArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
+        FORWARD_IF_ERROR(ret, "Failed to init asm args");
+        if (ret == 0)
+            return 0;
+    }
+
+    assert(args.ip[0] >= args.ilowest);
+    loopFn(&args);
+
+    /* note : op4 already verified within main loop */
+    assert(args.ip[0] >= ilowest);
+    assert(args.ip[1] >= ilowest);
+    assert(args.ip[2] >= ilowest);
+    assert(args.ip[3] >= ilowest);
+    assert(args.op[3] <= oend);
+
+    assert(ilowest == args.ilowest);
+    assert(ilowest + 6 == args.iend[0]);
+    (void)ilowest;
+
+    /* finish bitStreams one by one */
+    {
+        size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* segmentEnd = (BYTE*)dst;
+        int i;
+        for (i = 0; i < 4; ++i) {
+            BIT_DStream_t bit;
+            if (segmentSize <= (size_t)(oend - segmentEnd))
+                segmentEnd += segmentSize;
+            else
+                segmentEnd = oend;
+            FORWARD_IF_ERROR(HUF_initRemainingDStream(&bit, &args, i, segmentEnd), "corruption");
+            args.op[i] += HUF_decodeStreamX2(args.op[i], &bit, segmentEnd, (HUF_DEltX2 const*)dt, HUF_DECODER_FAST_TABLELOG);
+            if (args.op[i] != segmentEnd)
+                return ERROR(corruption_detected);
+        }
+    }
+
+    /* decoded size */
+    return dstSize;
+}
+
+static size_t HUF_decompress4X2_usingDTable_internal(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable, int flags)
+{
+    HUF_DecompressUsingDTableFn fallbackFn = HUF_decompress4X2_usingDTable_internal_default;
+    HUF_DecompressFastLoopFn loopFn = HUF_decompress4X2_usingDTable_internal_fast_c_loop;
+
+#if DYNAMIC_BMI2
+    if (flags & HUF_flags_bmi2) {
+        fallbackFn = HUF_decompress4X2_usingDTable_internal_bmi2;
+# if ZSTD_ENABLE_ASM_X86_64_BMI2
+        if (!(flags & HUF_flags_disableAsm)) {
+            loopFn = HUF_decompress4X2_usingDTable_internal_fast_asm_loop;
+        }
+# endif
+    } else {
+        return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
+    }
+#endif
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__)
+    if (!(flags & HUF_flags_disableAsm)) {
+        loopFn = HUF_decompress4X2_usingDTable_internal_fast_asm_loop;
+    }
+#endif
+
+    if (HUF_ENABLE_FAST_DECODE && !(flags & HUF_flags_disableFast)) {
+        size_t const ret = HUF_decompress4X2_usingDTable_internal_fast(dst, dstSize, cSrc, cSrcSize, DTable, loopFn);
+        if (ret != 0)
+            return ret;
+    }
+    return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+HUF_DGEN(HUF_decompress1X2_usingDTable_internal)
+
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int flags)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX2_wksp(DCtx, cSrc, cSrcSize,
+                                               workSpace, wkspSize, flags);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, flags);
+}
+
+static size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int flags)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX2_wksp(dctx, cSrc, cSrcSize,
+                                         workSpace, wkspSize, flags);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
+}
+
+#endif /* HUF_FORCE_DECOMPRESS_X1 */
+
+
+/* ***********************************/
+/* Universal decompression selectors */
+/* ***********************************/
+
+
+#if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][2 /* single, double */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}},  /* Q==1 : impossible */
+    {{ 150,216}, { 381,119}},   /* Q == 2 : 12-18% */
+    {{ 170,205}, { 514,112}},   /* Q == 3 : 18-25% */
+    {{ 177,199}, { 539,110}},   /* Q == 4 : 25-32% */
+    {{ 197,194}, { 644,107}},   /* Q == 5 : 32-38% */
+    {{ 221,192}, { 735,107}},   /* Q == 6 : 38-44% */
+    {{ 256,189}, { 881,106}},   /* Q == 7 : 44-50% */
+    {{ 359,188}, {1167,109}},   /* Q == 8 : 50-56% */
+    {{ 582,187}, {1570,114}},   /* Q == 9 : 56-62% */
+    {{ 688,187}, {1712,122}},   /* Q ==10 : 62-69% */
+    {{ 825,186}, {1965,136}},   /* Q ==11 : 69-75% */
+    {{ 976,185}, {2131,150}},   /* Q ==12 : 75-81% */
+    {{1180,186}, {2070,175}},   /* Q ==13 : 81-87% */
+    {{1377,185}, {1731,202}},   /* Q ==14 : 87-93% */
+    {{1412,185}, {1695,202}},   /* Q ==15 : 93-99% */
+};
+#endif
+
+/** HUF_selectDecoder() :
+ *  Tells which decoder is likely to decode faster,
+ *  based on a set of pre-computed metrics.
+ * @return : 0==HUF_decompress4X1, 1==HUF_decompress4X2 .
+ *  Assumption : 0 < dstSize <= 128 KB */
+U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize)
+{
+    assert(dstSize > 0);
+    assert(dstSize <= 128*1024);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+    (void)dstSize;
+    (void)cSrcSize;
+    return 0;
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+    (void)dstSize;
+    (void)cSrcSize;
+    return 1;
+#else
+    /* decoder timing evaluation */
+    {   U32 const Q = (cSrcSize >= dstSize) ? 15 : (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 */
+        U32 const D256 = (U32)(dstSize >> 8);
+        U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
+        U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
+        DTime1 += DTime1 >> 5;  /* small advantage to algorithm using less memory, to reduce cache eviction */
+        return DTime1 < DTime0;
+    }
+#endif
+}
+
+size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                  const void* cSrc, size_t cSrcSize,
+                                  void* workSpace, size_t wkspSize, int flags)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { ZSTD_memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { ZSTD_memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+        (void)algoNb;
+        assert(algoNb == 0);
+        return HUF_decompress1X1_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize, flags);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+        (void)algoNb;
+        assert(algoNb == 1);
+        return HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize, flags);
+#else
+        return algoNb ? HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize, flags):
+                        HUF_decompress1X1_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize, flags);
+#endif
+    }
+}
+
+
+size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+    (void)dtd;
+    assert(dtd.tableType == 0);
+    return HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+    (void)dtd;
+    assert(dtd.tableType == 1);
+    return HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#else
+    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags) :
+                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#endif
+}
+
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize, flags);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
+}
+#endif
+
+size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+    (void)dtd;
+    assert(dtd.tableType == 0);
+    return HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+    (void)dtd;
+    assert(dtd.tableType == 1);
+    return HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#else
+    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags) :
+                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
+#endif
+}
+
+size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize == 0) return ERROR(corruption_detected);
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+        (void)algoNb;
+        assert(algoNb == 0);
+        return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+        (void)algoNb;
+        assert(algoNb == 1);
+        return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
+#else
+        return algoNb ? HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags) :
+                        HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
+#endif
+    }
+}
diff --git a/lib/decompress/zstd_decompress.c b/lib/decompress/zstd_decompress.c
new file mode 100644
index 0000000..9eb9832
--- /dev/null
+++ b/lib/decompress/zstd_decompress.c
@@ -0,0 +1,2410 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTD_decompress() allocates its context,
+ * on stack (0), or into heap (1, default; requires malloc()).
+ * Note that functions with explicit context such as ZSTD_decompressDCtx() are unaffected.
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif
+
+/*!
+*  LEGACY_SUPPORT :
+*  if set to 1+, ZSTD_decompress() can decode older formats (v0.1+)
+*/
+#ifndef ZSTD_LEGACY_SUPPORT
+#  define ZSTD_LEGACY_SUPPORT 0
+#endif
+
+/*!
+ *  MAXWINDOWSIZE_DEFAULT :
+ *  maximum window size accepted by DStream __by default__.
+ *  Frames requiring more memory will be rejected.
+ *  It's possible to set a different limit using ZSTD_DCtx_setMaxWindowSize().
+ */
+#ifndef ZSTD_MAXWINDOWSIZE_DEFAULT
+#  define ZSTD_MAXWINDOWSIZE_DEFAULT (((U32)1 << ZSTD_WINDOWLOG_LIMIT_DEFAULT) + 1)
+#endif
+
+/*!
+ *  NO_FORWARD_PROGRESS_MAX :
+ *  maximum allowed nb of calls to ZSTD_decompressStream()
+ *  without any forward progress
+ *  (defined as: no byte read from input, and no byte flushed to output)
+ *  before triggering an error.
+ */
+#ifndef ZSTD_NO_FORWARD_PROGRESS_MAX
+#  define ZSTD_NO_FORWARD_PROGRESS_MAX 16
+#endif
+
+
+/*-*******************************************************
+*  Dependencies
+*********************************************************/
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
+#include "../common/allocations.h"  /* ZSTD_customMalloc, ZSTD_customCalloc, ZSTD_customFree */
+#include "../common/error_private.h"
+#include "../common/zstd_internal.h"  /* blockProperties_t */
+#include "../common/mem.h"         /* low level memory routines */
+#include "../common/bits.h"  /* ZSTD_highbit32 */
+#define FSE_STATIC_LINKING_ONLY
+#include "../common/fse.h"
+#include "../common/huf.h"
+#include "../common/xxhash.h" /* XXH64_reset, XXH64_update, XXH64_digest, XXH64 */
+#include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
+#include "zstd_ddict.h"  /* ZSTD_DDictDictContent */
+#include "zstd_decompress_block.h"   /* ZSTD_decompressBlock_internal */
+
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+#  include "../legacy/zstd_legacy.h"
+#endif
+
+
+
+/*************************************
+ * Multiple DDicts Hashset internals *
+ *************************************/
+
+#define DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT 4
+#define DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT 3  /* These two constants represent SIZE_MULT/COUNT_MULT load factor without using a float.
+                                                    * Currently, that means a 0.75 load factor.
+                                                    * So, if count * COUNT_MULT / size * SIZE_MULT != 0, then we've exceeded
+                                                    * the load factor of the ddict hash set.
+                                                    */
+
+#define DDICT_HASHSET_TABLE_BASE_SIZE 64
+#define DDICT_HASHSET_RESIZE_FACTOR 2
+
+/* Hash function to determine starting position of dict insertion within the table
+ * Returns an index between [0, hashSet->ddictPtrTableSize]
+ */
+static size_t ZSTD_DDictHashSet_getIndex(const ZSTD_DDictHashSet* hashSet, U32 dictID) {
+    const U64 hash = XXH64(&dictID, sizeof(U32), 0);
+    /* DDict ptr table size is a multiple of 2, use size - 1 as mask to get index within [0, hashSet->ddictPtrTableSize) */
+    return hash & (hashSet->ddictPtrTableSize - 1);
+}
+
+/* Adds DDict to a hashset without resizing it.
+ * If inserting a DDict with a dictID that already exists in the set, replaces the one in the set.
+ * Returns 0 if successful, or a zstd error code if something went wrong.
+ */
+static size_t ZSTD_DDictHashSet_emplaceDDict(ZSTD_DDictHashSet* hashSet, const ZSTD_DDict* ddict) {
+    const U32 dictID = ZSTD_getDictID_fromDDict(ddict);
+    size_t idx = ZSTD_DDictHashSet_getIndex(hashSet, dictID);
+    const size_t idxRangeMask = hashSet->ddictPtrTableSize - 1;
+    RETURN_ERROR_IF(hashSet->ddictPtrCount == hashSet->ddictPtrTableSize, GENERIC, "Hash set is full!");
+    DEBUGLOG(4, "Hashed index: for dictID: %u is %zu", dictID, idx);
+    while (hashSet->ddictPtrTable[idx] != NULL) {
+        /* Replace existing ddict if inserting ddict with same dictID */
+        if (ZSTD_getDictID_fromDDict(hashSet->ddictPtrTable[idx]) == dictID) {
+            DEBUGLOG(4, "DictID already exists, replacing rather than adding");
+            hashSet->ddictPtrTable[idx] = ddict;
+            return 0;
+        }
+        idx &= idxRangeMask;
+        idx++;
+    }
+    DEBUGLOG(4, "Final idx after probing for dictID %u is: %zu", dictID, idx);
+    hashSet->ddictPtrTable[idx] = ddict;
+    hashSet->ddictPtrCount++;
+    return 0;
+}
+
+/* Expands hash table by factor of DDICT_HASHSET_RESIZE_FACTOR and
+ * rehashes all values, allocates new table, frees old table.
+ * Returns 0 on success, otherwise a zstd error code.
+ */
+static size_t ZSTD_DDictHashSet_expand(ZSTD_DDictHashSet* hashSet, ZSTD_customMem customMem) {
+    size_t newTableSize = hashSet->ddictPtrTableSize * DDICT_HASHSET_RESIZE_FACTOR;
+    const ZSTD_DDict** newTable = (const ZSTD_DDict**)ZSTD_customCalloc(sizeof(ZSTD_DDict*) * newTableSize, customMem);
+    const ZSTD_DDict** oldTable = hashSet->ddictPtrTable;
+    size_t oldTableSize = hashSet->ddictPtrTableSize;
+    size_t i;
+
+    DEBUGLOG(4, "Expanding DDict hash table! Old size: %zu new size: %zu", oldTableSize, newTableSize);
+    RETURN_ERROR_IF(!newTable, memory_allocation, "Expanded hashset allocation failed!");
+    hashSet->ddictPtrTable = newTable;
+    hashSet->ddictPtrTableSize = newTableSize;
+    hashSet->ddictPtrCount = 0;
+    for (i = 0; i < oldTableSize; ++i) {
+        if (oldTable[i] != NULL) {
+            FORWARD_IF_ERROR(ZSTD_DDictHashSet_emplaceDDict(hashSet, oldTable[i]), "");
+        }
+    }
+    ZSTD_customFree((void*)oldTable, customMem);
+    DEBUGLOG(4, "Finished re-hash");
+    return 0;
+}
+
+/* Fetches a DDict with the given dictID
+ * Returns the ZSTD_DDict* with the requested dictID. If it doesn't exist, then returns NULL.
+ */
+static const ZSTD_DDict* ZSTD_DDictHashSet_getDDict(ZSTD_DDictHashSet* hashSet, U32 dictID) {
+    size_t idx = ZSTD_DDictHashSet_getIndex(hashSet, dictID);
+    const size_t idxRangeMask = hashSet->ddictPtrTableSize - 1;
+    DEBUGLOG(4, "Hashed index: for dictID: %u is %zu", dictID, idx);
+    for (;;) {
+        size_t currDictID = ZSTD_getDictID_fromDDict(hashSet->ddictPtrTable[idx]);
+        if (currDictID == dictID || currDictID == 0) {
+            /* currDictID == 0 implies a NULL ddict entry */
+            break;
+        } else {
+            idx &= idxRangeMask;    /* Goes to start of table when we reach the end */
+            idx++;
+        }
+    }
+    DEBUGLOG(4, "Final idx after probing for dictID %u is: %zu", dictID, idx);
+    return hashSet->ddictPtrTable[idx];
+}
+
+/* Allocates space for and returns a ddict hash set
+ * The hash set's ZSTD_DDict* table has all values automatically set to NULL to begin with.
+ * Returns NULL if allocation failed.
+ */
+static ZSTD_DDictHashSet* ZSTD_createDDictHashSet(ZSTD_customMem customMem) {
+    ZSTD_DDictHashSet* ret = (ZSTD_DDictHashSet*)ZSTD_customMalloc(sizeof(ZSTD_DDictHashSet), customMem);
+    DEBUGLOG(4, "Allocating new hash set");
+    if (!ret)
+        return NULL;
+    ret->ddictPtrTable = (const ZSTD_DDict**)ZSTD_customCalloc(DDICT_HASHSET_TABLE_BASE_SIZE * sizeof(ZSTD_DDict*), customMem);
+    if (!ret->ddictPtrTable) {
+        ZSTD_customFree(ret, customMem);
+        return NULL;
+    }
+    ret->ddictPtrTableSize = DDICT_HASHSET_TABLE_BASE_SIZE;
+    ret->ddictPtrCount = 0;
+    return ret;
+}
+
+/* Frees the table of ZSTD_DDict* within a hashset, then frees the hashset itself.
+ * Note: The ZSTD_DDict* within the table are NOT freed.
+ */
+static void ZSTD_freeDDictHashSet(ZSTD_DDictHashSet* hashSet, ZSTD_customMem customMem) {
+    DEBUGLOG(4, "Freeing ddict hash set");
+    if (hashSet && hashSet->ddictPtrTable) {
+        ZSTD_customFree((void*)hashSet->ddictPtrTable, customMem);
+    }
+    if (hashSet) {
+        ZSTD_customFree(hashSet, customMem);
+    }
+}
+
+/* Public function: Adds a DDict into the ZSTD_DDictHashSet, possibly triggering a resize of the hash set.
+ * Returns 0 on success, or a ZSTD error.
+ */
+static size_t ZSTD_DDictHashSet_addDDict(ZSTD_DDictHashSet* hashSet, const ZSTD_DDict* ddict, ZSTD_customMem customMem) {
+    DEBUGLOG(4, "Adding dict ID: %u to hashset with - Count: %zu Tablesize: %zu", ZSTD_getDictID_fromDDict(ddict), hashSet->ddictPtrCount, hashSet->ddictPtrTableSize);
+    if (hashSet->ddictPtrCount * DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT / hashSet->ddictPtrTableSize * DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT != 0) {
+        FORWARD_IF_ERROR(ZSTD_DDictHashSet_expand(hashSet, customMem), "");
+    }
+    FORWARD_IF_ERROR(ZSTD_DDictHashSet_emplaceDDict(hashSet, ddict), "");
+    return 0;
+}
+
+/*-*************************************************************
+*   Context management
+***************************************************************/
+size_t ZSTD_sizeof_DCtx (const ZSTD_DCtx* dctx)
+{
+    if (dctx==NULL) return 0;   /* support sizeof NULL */
+    return sizeof(*dctx)
+           + ZSTD_sizeof_DDict(dctx->ddictLocal)
+           + dctx->inBuffSize + dctx->outBuffSize;
+}
+
+size_t ZSTD_estimateDCtxSize(void) { return sizeof(ZSTD_DCtx); }
+
+
+static size_t ZSTD_startingInputLength(ZSTD_format_e format)
+{
+    size_t const startingInputLength = ZSTD_FRAMEHEADERSIZE_PREFIX(format);
+    /* only supports formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless */
+    assert( (format == ZSTD_f_zstd1) || (format == ZSTD_f_zstd1_magicless) );
+    return startingInputLength;
+}
+
+static void ZSTD_DCtx_resetParameters(ZSTD_DCtx* dctx)
+{
+    assert(dctx->streamStage == zdss_init);
+    dctx->format = ZSTD_f_zstd1;
+    dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
+    dctx->outBufferMode = ZSTD_bm_buffered;
+    dctx->forceIgnoreChecksum = ZSTD_d_validateChecksum;
+    dctx->refMultipleDDicts = ZSTD_rmd_refSingleDDict;
+    dctx->disableHufAsm = 0;
+    dctx->maxBlockSizeParam = 0;
+}
+
+static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
+{
+    dctx->staticSize  = 0;
+    dctx->ddict       = NULL;
+    dctx->ddictLocal  = NULL;
+    dctx->dictEnd     = NULL;
+    dctx->ddictIsCold = 0;
+    dctx->dictUses = ZSTD_dont_use;
+    dctx->inBuff      = NULL;
+    dctx->inBuffSize  = 0;
+    dctx->outBuffSize = 0;
+    dctx->streamStage = zdss_init;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+    dctx->legacyContext = NULL;
+    dctx->previousLegacyVersion = 0;
+#endif
+    dctx->noForwardProgress = 0;
+    dctx->oversizedDuration = 0;
+    dctx->isFrameDecompression = 1;
+#if DYNAMIC_BMI2
+    dctx->bmi2 = ZSTD_cpuSupportsBmi2();
+#endif
+    dctx->ddictSet = NULL;
+    ZSTD_DCtx_resetParameters(dctx);
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    dctx->dictContentEndForFuzzing = NULL;
+#endif
+}
+
+ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
+{
+    ZSTD_DCtx* const dctx = (ZSTD_DCtx*) workspace;
+
+    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
+    if (workspaceSize < sizeof(ZSTD_DCtx)) return NULL;  /* minimum size */
+
+    ZSTD_initDCtx_internal(dctx);
+    dctx->staticSize = workspaceSize;
+    dctx->inBuff = (char*)(dctx+1);
+    return dctx;
+}
+
+static ZSTD_DCtx* ZSTD_createDCtx_internal(ZSTD_customMem customMem) {
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
+
+    {   ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_customMalloc(sizeof(*dctx), customMem);
+        if (!dctx) return NULL;
+        dctx->customMem = customMem;
+        ZSTD_initDCtx_internal(dctx);
+        return dctx;
+    }
+}
+
+ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
+{
+    return ZSTD_createDCtx_internal(customMem);
+}
+
+ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    DEBUGLOG(3, "ZSTD_createDCtx");
+    return ZSTD_createDCtx_internal(ZSTD_defaultCMem);
+}
+
+static void ZSTD_clearDict(ZSTD_DCtx* dctx)
+{
+    ZSTD_freeDDict(dctx->ddictLocal);
+    dctx->ddictLocal = NULL;
+    dctx->ddict = NULL;
+    dctx->dictUses = ZSTD_dont_use;
+}
+
+size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    if (dctx==NULL) return 0;   /* support free on NULL */
+    RETURN_ERROR_IF(dctx->staticSize, memory_allocation, "not compatible with static DCtx");
+    {   ZSTD_customMem const cMem = dctx->customMem;
+        ZSTD_clearDict(dctx);
+        ZSTD_customFree(dctx->inBuff, cMem);
+        dctx->inBuff = NULL;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+        if (dctx->legacyContext)
+            ZSTD_freeLegacyStreamContext(dctx->legacyContext, dctx->previousLegacyVersion);
+#endif
+        if (dctx->ddictSet) {
+            ZSTD_freeDDictHashSet(dctx->ddictSet, cMem);
+            dctx->ddictSet = NULL;
+        }
+        ZSTD_customFree(dctx, cMem);
+        return 0;
+    }
+}
+
+/* no longer useful */
+void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
+{
+    size_t const toCopy = (size_t)((char*)(&dstDCtx->inBuff) - (char*)dstDCtx);
+    ZSTD_memcpy(dstDCtx, srcDCtx, toCopy);  /* no need to copy workspace */
+}
+
+/* Given a dctx with a digested frame params, re-selects the correct ZSTD_DDict based on
+ * the requested dict ID from the frame. If there exists a reference to the correct ZSTD_DDict, then
+ * accordingly sets the ddict to be used to decompress the frame.
+ *
+ * If no DDict is found, then no action is taken, and the ZSTD_DCtx::ddict remains as-is.
+ *
+ * ZSTD_d_refMultipleDDicts must be enabled for this function to be called.
+ */
+static void ZSTD_DCtx_selectFrameDDict(ZSTD_DCtx* dctx) {
+    assert(dctx->refMultipleDDicts && dctx->ddictSet);
+    DEBUGLOG(4, "Adjusting DDict based on requested dict ID from frame");
+    if (dctx->ddict) {
+        const ZSTD_DDict* frameDDict = ZSTD_DDictHashSet_getDDict(dctx->ddictSet, dctx->fParams.dictID);
+        if (frameDDict) {
+            DEBUGLOG(4, "DDict found!");
+            ZSTD_clearDict(dctx);
+            dctx->dictID = dctx->fParams.dictID;
+            dctx->ddict = frameDDict;
+            dctx->dictUses = ZSTD_use_indefinitely;
+        }
+    }
+}
+
+
+/*-*************************************************************
+ *   Frame header decoding
+ ***************************************************************/
+
+/*! ZSTD_isFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier.
+ *  Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
+ *  Note 2 : Legacy Frame Identifiers are considered valid only if Legacy Support is enabled.
+ *  Note 3 : Skippable Frame Identifiers are considered valid. */
+unsigned ZSTD_isFrame(const void* buffer, size_t size)
+{
+    if (size < ZSTD_FRAMEIDSIZE) return 0;
+    {   U32 const magic = MEM_readLE32(buffer);
+        if (magic == ZSTD_MAGICNUMBER) return 1;
+        if ((magic & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
+    }
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(buffer, size)) return 1;
+#endif
+    return 0;
+}
+
+/*! ZSTD_isSkippableFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier for a skippable frame.
+ *  Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
+ */
+unsigned ZSTD_isSkippableFrame(const void* buffer, size_t size)
+{
+    if (size < ZSTD_FRAMEIDSIZE) return 0;
+    {   U32 const magic = MEM_readLE32(buffer);
+        if ((magic & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
+    }
+    return 0;
+}
+
+/** ZSTD_frameHeaderSize_internal() :
+ *  srcSize must be large enough to reach header size fields.
+ *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless.
+ * @return : size of the Frame Header
+ *           or an error code, which can be tested with ZSTD_isError() */
+static size_t ZSTD_frameHeaderSize_internal(const void* src, size_t srcSize, ZSTD_format_e format)
+{
+    size_t const minInputSize = ZSTD_startingInputLength(format);
+    RETURN_ERROR_IF(srcSize < minInputSize, srcSize_wrong, "");
+
+    {   BYTE const fhd = ((const BYTE*)src)[minInputSize-1];
+        U32 const dictID= fhd & 3;
+        U32 const singleSegment = (fhd >> 5) & 1;
+        U32 const fcsId = fhd >> 6;
+        return minInputSize + !singleSegment
+             + ZSTD_did_fieldSize[dictID] + ZSTD_fcs_fieldSize[fcsId]
+             + (singleSegment && !fcsId);
+    }
+}
+
+/** ZSTD_frameHeaderSize() :
+ *  srcSize must be >= ZSTD_frameHeaderSize_prefix.
+ * @return : size of the Frame Header,
+ *           or an error code (if srcSize is too small) */
+size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
+{
+    return ZSTD_frameHeaderSize_internal(src, srcSize, ZSTD_f_zstd1);
+}
+
+
+/** ZSTD_getFrameHeader_advanced() :
+ *  decode Frame Header, or require larger `srcSize`.
+ *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+**           or an error code, which can be tested using ZSTD_isError() */
+size_t ZSTD_getFrameHeader_advanced(ZSTD_FrameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t const minInputSize = ZSTD_startingInputLength(format);
+
+    DEBUGLOG(5, "ZSTD_getFrameHeader_advanced: minInputSize = %zu, srcSize = %zu", minInputSize, srcSize);
+
+    if (srcSize > 0) {
+        /* note : technically could be considered an assert(), since it's an invalid entry */
+        RETURN_ERROR_IF(src==NULL, GENERIC, "invalid parameter : src==NULL, but srcSize>0");
+    }
+    if (srcSize < minInputSize) {
+        if (srcSize > 0 && format != ZSTD_f_zstd1_magicless) {
+            /* when receiving less than @minInputSize bytes,
+             * control these bytes at least correspond to a supported magic number
+             * in order to error out early if they don't.
+            **/
+            size_t const toCopy = MIN(4, srcSize);
+            unsigned char hbuf[4]; MEM_writeLE32(hbuf, ZSTD_MAGICNUMBER);
+            assert(src != NULL);
+            ZSTD_memcpy(hbuf, src, toCopy);
+            if ( MEM_readLE32(hbuf) != ZSTD_MAGICNUMBER ) {
+                /* not a zstd frame : let's check if it's a skippable frame */
+                MEM_writeLE32(hbuf, ZSTD_MAGIC_SKIPPABLE_START);
+                ZSTD_memcpy(hbuf, src, toCopy);
+                if ((MEM_readLE32(hbuf) & ZSTD_MAGIC_SKIPPABLE_MASK) != ZSTD_MAGIC_SKIPPABLE_START) {
+                    RETURN_ERROR(prefix_unknown,
+                                "first bytes don't correspond to any supported magic number");
+        }   }   }
+        return minInputSize;
+    }
+
+    ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzers may not understand that zfhPtr will be read only if return value is zero, since they are 2 different signals */
+    if ( (format != ZSTD_f_zstd1_magicless)
+      && (MEM_readLE32(src) != ZSTD_MAGICNUMBER) ) {
+        if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+            /* skippable frame */
+            if (srcSize < ZSTD_SKIPPABLEHEADERSIZE)
+                return ZSTD_SKIPPABLEHEADERSIZE; /* magic number + frame length */
+            ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));
+            zfhPtr->frameType = ZSTD_skippableFrame;
+            zfhPtr->dictID = MEM_readLE32(src) - ZSTD_MAGIC_SKIPPABLE_START;
+            zfhPtr->headerSize = ZSTD_SKIPPABLEHEADERSIZE;
+            zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_FRAMEIDSIZE);
+            return 0;
+        }
+        RETURN_ERROR(prefix_unknown, "");
+    }
+
+    /* ensure there is enough `srcSize` to fully read/decode frame header */
+    {   size_t const fhsize = ZSTD_frameHeaderSize_internal(src, srcSize, format);
+        if (srcSize < fhsize) return fhsize;
+        zfhPtr->headerSize = (U32)fhsize;
+    }
+
+    {   BYTE const fhdByte = ip[minInputSize-1];
+        size_t pos = minInputSize;
+        U32 const dictIDSizeCode = fhdByte&3;
+        U32 const checksumFlag = (fhdByte>>2)&1;
+        U32 const singleSegment = (fhdByte>>5)&1;
+        U32 const fcsID = fhdByte>>6;
+        U64 windowSize = 0;
+        U32 dictID = 0;
+        U64 frameContentSize = ZSTD_CONTENTSIZE_UNKNOWN;
+        RETURN_ERROR_IF((fhdByte & 0x08) != 0, frameParameter_unsupported,
+                        "reserved bits, must be zero");
+
+        if (!singleSegment) {
+            BYTE const wlByte = ip[pos++];
+            U32 const windowLog = (wlByte >> 3) + ZSTD_WINDOWLOG_ABSOLUTEMIN;
+            RETURN_ERROR_IF(windowLog > ZSTD_WINDOWLOG_MAX, frameParameter_windowTooLarge, "");
+            windowSize = (1ULL << windowLog);
+            windowSize += (windowSize >> 3) * (wlByte&7);
+        }
+        switch(dictIDSizeCode)
+        {
+            default:
+                assert(0);  /* impossible */
+                ZSTD_FALLTHROUGH;
+            case 0 : break;
+            case 1 : dictID = ip[pos]; pos++; break;
+            case 2 : dictID = MEM_readLE16(ip+pos); pos+=2; break;
+            case 3 : dictID = MEM_readLE32(ip+pos); pos+=4; break;
+        }
+        switch(fcsID)
+        {
+            default:
+                assert(0);  /* impossible */
+                ZSTD_FALLTHROUGH;
+            case 0 : if (singleSegment) frameContentSize = ip[pos]; break;
+            case 1 : frameContentSize = MEM_readLE16(ip+pos)+256; break;
+            case 2 : frameContentSize = MEM_readLE32(ip+pos); break;
+            case 3 : frameContentSize = MEM_readLE64(ip+pos); break;
+        }
+        if (singleSegment) windowSize = frameContentSize;
+
+        zfhPtr->frameType = ZSTD_frame;
+        zfhPtr->frameContentSize = frameContentSize;
+        zfhPtr->windowSize = windowSize;
+        zfhPtr->blockSizeMax = (unsigned) MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
+        zfhPtr->dictID = dictID;
+        zfhPtr->checksumFlag = checksumFlag;
+    }
+    return 0;
+}
+
+/** ZSTD_getFrameHeader() :
+ *  decode Frame Header, or require larger `srcSize`.
+ *  note : this function does not consume input, it only reads it.
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+ *           or an error code, which can be tested using ZSTD_isError() */
+size_t ZSTD_getFrameHeader(ZSTD_FrameHeader* zfhPtr, const void* src, size_t srcSize)
+{
+    return ZSTD_getFrameHeader_advanced(zfhPtr, src, srcSize, ZSTD_f_zstd1);
+}
+
+/** ZSTD_getFrameContentSize() :
+ *  compatible with legacy mode
+ * @return : decompressed size of the single frame pointed to be `src` if known, otherwise
+ *         - ZSTD_CONTENTSIZE_UNKNOWN if the size cannot be determined
+ *         - ZSTD_CONTENTSIZE_ERROR if an error occurred (e.g. invalid magic number, srcSize too small) */
+unsigned long long ZSTD_getFrameContentSize(const void *src, size_t srcSize)
+{
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(src, srcSize)) {
+        unsigned long long const ret = ZSTD_getDecompressedSize_legacy(src, srcSize);
+        return ret == 0 ? ZSTD_CONTENTSIZE_UNKNOWN : ret;
+    }
+#endif
+    {   ZSTD_FrameHeader zfh;
+        if (ZSTD_getFrameHeader(&zfh, src, srcSize) != 0)
+            return ZSTD_CONTENTSIZE_ERROR;
+        if (zfh.frameType == ZSTD_skippableFrame) {
+            return 0;
+        } else {
+            return zfh.frameContentSize;
+    }   }
+}
+
+static size_t readSkippableFrameSize(void const* src, size_t srcSize)
+{
+    size_t const skippableHeaderSize = ZSTD_SKIPPABLEHEADERSIZE;
+    U32 sizeU32;
+
+    RETURN_ERROR_IF(srcSize < ZSTD_SKIPPABLEHEADERSIZE, srcSize_wrong, "");
+
+    sizeU32 = MEM_readLE32((BYTE const*)src + ZSTD_FRAMEIDSIZE);
+    RETURN_ERROR_IF((U32)(sizeU32 + ZSTD_SKIPPABLEHEADERSIZE) < sizeU32,
+                    frameParameter_unsupported, "");
+    {   size_t const skippableSize = skippableHeaderSize + sizeU32;
+        RETURN_ERROR_IF(skippableSize > srcSize, srcSize_wrong, "");
+        return skippableSize;
+    }
+}
+
+/*! ZSTD_readSkippableFrame() :
+ * Retrieves content of a skippable frame, and writes it to dst buffer.
+ *
+ * The parameter magicVariant will receive the magicVariant that was supplied when the frame was written,
+ * i.e. magicNumber - ZSTD_MAGIC_SKIPPABLE_START.  This can be NULL if the caller is not interested
+ * in the magicVariant.
+ *
+ * Returns an error if destination buffer is not large enough, or if this is not a valid skippable frame.
+ *
+ * @return : number of bytes written or a ZSTD error.
+ */
+size_t ZSTD_readSkippableFrame(void* dst, size_t dstCapacity,
+                               unsigned* magicVariant,  /* optional, can be NULL */
+                         const void* src, size_t srcSize)
+{
+    RETURN_ERROR_IF(srcSize < ZSTD_SKIPPABLEHEADERSIZE, srcSize_wrong, "");
+
+    {   U32 const magicNumber = MEM_readLE32(src);
+        size_t skippableFrameSize = readSkippableFrameSize(src, srcSize);
+        size_t skippableContentSize = skippableFrameSize - ZSTD_SKIPPABLEHEADERSIZE;
+
+        /* check input validity */
+        RETURN_ERROR_IF(!ZSTD_isSkippableFrame(src, srcSize), frameParameter_unsupported, "");
+        RETURN_ERROR_IF(skippableFrameSize < ZSTD_SKIPPABLEHEADERSIZE || skippableFrameSize > srcSize, srcSize_wrong, "");
+        RETURN_ERROR_IF(skippableContentSize > dstCapacity, dstSize_tooSmall, "");
+
+        /* deliver payload */
+        if (skippableContentSize > 0  && dst != NULL)
+            ZSTD_memcpy(dst, (const BYTE *)src + ZSTD_SKIPPABLEHEADERSIZE, skippableContentSize);
+        if (magicVariant != NULL)
+            *magicVariant = magicNumber - ZSTD_MAGIC_SKIPPABLE_START;
+        return skippableContentSize;
+    }
+}
+
+/** ZSTD_findDecompressedSize() :
+ *  `srcSize` must be the exact length of some number of ZSTD compressed and/or
+ *      skippable frames
+ *  note: compatible with legacy mode
+ * @return : decompressed size of the frames contained */
+unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
+{
+    unsigned long long totalDstSize = 0;
+
+    while (srcSize >= ZSTD_startingInputLength(ZSTD_f_zstd1)) {
+        U32 const magicNumber = MEM_readLE32(src);
+
+        if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+            size_t const skippableSize = readSkippableFrameSize(src, srcSize);
+            if (ZSTD_isError(skippableSize)) return ZSTD_CONTENTSIZE_ERROR;
+            assert(skippableSize <= srcSize);
+
+            src = (const BYTE *)src + skippableSize;
+            srcSize -= skippableSize;
+            continue;
+        }
+
+        {   unsigned long long const fcs = ZSTD_getFrameContentSize(src, srcSize);
+            if (fcs >= ZSTD_CONTENTSIZE_ERROR) return fcs;
+
+            if (totalDstSize + fcs < totalDstSize)
+                return ZSTD_CONTENTSIZE_ERROR; /* check for overflow */
+            totalDstSize += fcs;
+        }
+        /* skip to next frame */
+        {   size_t const frameSrcSize = ZSTD_findFrameCompressedSize(src, srcSize);
+            if (ZSTD_isError(frameSrcSize)) return ZSTD_CONTENTSIZE_ERROR;
+            assert(frameSrcSize <= srcSize);
+
+            src = (const BYTE *)src + frameSrcSize;
+            srcSize -= frameSrcSize;
+        }
+    }  /* while (srcSize >= ZSTD_frameHeaderSize_prefix) */
+
+    if (srcSize) return ZSTD_CONTENTSIZE_ERROR;
+
+    return totalDstSize;
+}
+
+/** ZSTD_getDecompressedSize() :
+ *  compatible with legacy mode
+ * @return : decompressed size if known, 0 otherwise
+             note : 0 can mean any of the following :
+                   - frame content is empty
+                   - decompressed size field is not present in frame header
+                   - frame header unknown / not supported
+                   - frame header not complete (`srcSize` too small) */
+unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize)
+{
+    unsigned long long const ret = ZSTD_getFrameContentSize(src, srcSize);
+    ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_ERROR < ZSTD_CONTENTSIZE_UNKNOWN);
+    return (ret >= ZSTD_CONTENTSIZE_ERROR) ? 0 : ret;
+}
+
+
+/** ZSTD_decodeFrameHeader() :
+ * `headerSize` must be the size provided by ZSTD_frameHeaderSize().
+ * If multiple DDict references are enabled, also will choose the correct DDict to use.
+ * @return : 0 if success, or an error code, which can be tested using ZSTD_isError() */
+static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t headerSize)
+{
+    size_t const result = ZSTD_getFrameHeader_advanced(&(dctx->fParams), src, headerSize, dctx->format);
+    if (ZSTD_isError(result)) return result;    /* invalid header */
+    RETURN_ERROR_IF(result>0, srcSize_wrong, "headerSize too small");
+
+    /* Reference DDict requested by frame if dctx references multiple ddicts */
+    if (dctx->refMultipleDDicts == ZSTD_rmd_refMultipleDDicts && dctx->ddictSet) {
+        ZSTD_DCtx_selectFrameDDict(dctx);
+    }
+
+#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    /* Skip the dictID check in fuzzing mode, because it makes the search
+     * harder.
+     */
+    RETURN_ERROR_IF(dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID),
+                    dictionary_wrong, "");
+#endif
+    dctx->validateChecksum = (dctx->fParams.checksumFlag && !dctx->forceIgnoreChecksum) ? 1 : 0;
+    if (dctx->validateChecksum) XXH64_reset(&dctx->xxhState, 0);
+    dctx->processedCSize += headerSize;
+    return 0;
+}
+
+static ZSTD_frameSizeInfo ZSTD_errorFrameSizeInfo(size_t ret)
+{
+    ZSTD_frameSizeInfo frameSizeInfo;
+    frameSizeInfo.compressedSize = ret;
+    frameSizeInfo.decompressedBound = ZSTD_CONTENTSIZE_ERROR;
+    return frameSizeInfo;
+}
+
+static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize, ZSTD_format_e format)
+{
+    ZSTD_frameSizeInfo frameSizeInfo;
+    ZSTD_memset(&frameSizeInfo, 0, sizeof(ZSTD_frameSizeInfo));
+
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (format == ZSTD_f_zstd1 && ZSTD_isLegacy(src, srcSize))
+        return ZSTD_findFrameSizeInfoLegacy(src, srcSize);
+#endif
+
+    if (format == ZSTD_f_zstd1 && (srcSize >= ZSTD_SKIPPABLEHEADERSIZE)
+        && (MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+        frameSizeInfo.compressedSize = readSkippableFrameSize(src, srcSize);
+        assert(ZSTD_isError(frameSizeInfo.compressedSize) ||
+               frameSizeInfo.compressedSize <= srcSize);
+        return frameSizeInfo;
+    } else {
+        const BYTE* ip = (const BYTE*)src;
+        const BYTE* const ipstart = ip;
+        size_t remainingSize = srcSize;
+        size_t nbBlocks = 0;
+        ZSTD_FrameHeader zfh;
+
+        /* Extract Frame Header */
+        {   size_t const ret = ZSTD_getFrameHeader_advanced(&zfh, src, srcSize, format);
+            if (ZSTD_isError(ret))
+                return ZSTD_errorFrameSizeInfo(ret);
+            if (ret > 0)
+                return ZSTD_errorFrameSizeInfo(ERROR(srcSize_wrong));
+        }
+
+        ip += zfh.headerSize;
+        remainingSize -= zfh.headerSize;
+
+        /* Iterate over each block */
+        while (1) {
+            blockProperties_t blockProperties;
+            size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+            if (ZSTD_isError(cBlockSize))
+                return ZSTD_errorFrameSizeInfo(cBlockSize);
+
+            if (ZSTD_blockHeaderSize + cBlockSize > remainingSize)
+                return ZSTD_errorFrameSizeInfo(ERROR(srcSize_wrong));
+
+            ip += ZSTD_blockHeaderSize + cBlockSize;
+            remainingSize -= ZSTD_blockHeaderSize + cBlockSize;
+            nbBlocks++;
+
+            if (blockProperties.lastBlock) break;
+        }
+
+        /* Final frame content checksum */
+        if (zfh.checksumFlag) {
+            if (remainingSize < 4)
+                return ZSTD_errorFrameSizeInfo(ERROR(srcSize_wrong));
+            ip += 4;
+        }
+
+        frameSizeInfo.nbBlocks = nbBlocks;
+        frameSizeInfo.compressedSize = (size_t)(ip - ipstart);
+        frameSizeInfo.decompressedBound = (zfh.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN)
+                                        ? zfh.frameContentSize
+                                        : (unsigned long long)nbBlocks * zfh.blockSizeMax;
+        return frameSizeInfo;
+    }
+}
+
+static size_t ZSTD_findFrameCompressedSize_advanced(const void *src, size_t srcSize, ZSTD_format_e format) {
+    ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize, format);
+    return frameSizeInfo.compressedSize;
+}
+
+/** ZSTD_findFrameCompressedSize() :
+ * See docs in zstd.h
+ * Note: compatible with legacy mode */
+size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
+{
+    return ZSTD_findFrameCompressedSize_advanced(src, srcSize, ZSTD_f_zstd1);
+}
+
+/** ZSTD_decompressBound() :
+ *  compatible with legacy mode
+ *  `src` must point to the start of a ZSTD frame or a skippable frame
+ *  `srcSize` must be at least as large as the frame contained
+ *  @return : the maximum decompressed size of the compressed source
+ */
+unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize)
+{
+    unsigned long long bound = 0;
+    /* Iterate over each frame */
+    while (srcSize > 0) {
+        ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize, ZSTD_f_zstd1);
+        size_t const compressedSize = frameSizeInfo.compressedSize;
+        unsigned long long const decompressedBound = frameSizeInfo.decompressedBound;
+        if (ZSTD_isError(compressedSize) || decompressedBound == ZSTD_CONTENTSIZE_ERROR)
+            return ZSTD_CONTENTSIZE_ERROR;
+        assert(srcSize >= compressedSize);
+        src = (const BYTE*)src + compressedSize;
+        srcSize -= compressedSize;
+        bound += decompressedBound;
+    }
+    return bound;
+}
+
+size_t ZSTD_decompressionMargin(void const* src, size_t srcSize)
+{
+    size_t margin = 0;
+    unsigned maxBlockSize = 0;
+
+    /* Iterate over each frame */
+    while (srcSize > 0) {
+        ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize, ZSTD_f_zstd1);
+        size_t const compressedSize = frameSizeInfo.compressedSize;
+        unsigned long long const decompressedBound = frameSizeInfo.decompressedBound;
+        ZSTD_FrameHeader zfh;
+
+        FORWARD_IF_ERROR(ZSTD_getFrameHeader(&zfh, src, srcSize), "");
+        if (ZSTD_isError(compressedSize) || decompressedBound == ZSTD_CONTENTSIZE_ERROR)
+            return ERROR(corruption_detected);
+
+        if (zfh.frameType == ZSTD_frame) {
+            /* Add the frame header to our margin */
+            margin += zfh.headerSize;
+            /* Add the checksum to our margin */
+            margin += zfh.checksumFlag ? 4 : 0;
+            /* Add 3 bytes per block */
+            margin += 3 * frameSizeInfo.nbBlocks;
+
+            /* Compute the max block size */
+            maxBlockSize = MAX(maxBlockSize, zfh.blockSizeMax);
+        } else {
+            assert(zfh.frameType == ZSTD_skippableFrame);
+            /* Add the entire skippable frame size to our margin. */
+            margin += compressedSize;
+        }
+
+        assert(srcSize >= compressedSize);
+        src = (const BYTE*)src + compressedSize;
+        srcSize -= compressedSize;
+    }
+
+    /* Add the max block size back to the margin. */
+    margin += maxBlockSize;
+
+    return margin;
+}
+
+/*-*************************************************************
+ *   Frame decoding
+ ***************************************************************/
+
+/** ZSTD_insertBlock() :
+ *  insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
+size_t ZSTD_insertBlock(ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize)
+{
+    DEBUGLOG(5, "ZSTD_insertBlock: %u bytes", (unsigned)blockSize);
+    ZSTD_checkContinuity(dctx, blockStart, blockSize);
+    dctx->previousDstEnd = (const char*)blockStart + blockSize;
+    return blockSize;
+}
+
+
+static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
+                          const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_copyRawBlock");
+    RETURN_ERROR_IF(srcSize > dstCapacity, dstSize_tooSmall, "");
+    if (dst == NULL) {
+        if (srcSize == 0) return 0;
+        RETURN_ERROR(dstBuffer_null, "");
+    }
+    ZSTD_memmove(dst, src, srcSize);
+    return srcSize;
+}
+
+static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity,
+                               BYTE b,
+                               size_t regenSize)
+{
+    RETURN_ERROR_IF(regenSize > dstCapacity, dstSize_tooSmall, "");
+    if (dst == NULL) {
+        if (regenSize == 0) return 0;
+        RETURN_ERROR(dstBuffer_null, "");
+    }
+    ZSTD_memset(dst, b, regenSize);
+    return regenSize;
+}
+
+static void ZSTD_DCtx_trace_end(ZSTD_DCtx const* dctx, U64 uncompressedSize, U64 compressedSize, int streaming)
+{
+#if ZSTD_TRACE
+    if (dctx->traceCtx && ZSTD_trace_decompress_end != NULL) {
+        ZSTD_Trace trace;
+        ZSTD_memset(&trace, 0, sizeof(trace));
+        trace.version = ZSTD_VERSION_NUMBER;
+        trace.streaming = streaming;
+        if (dctx->ddict) {
+            trace.dictionaryID = ZSTD_getDictID_fromDDict(dctx->ddict);
+            trace.dictionarySize = ZSTD_DDict_dictSize(dctx->ddict);
+            trace.dictionaryIsCold = dctx->ddictIsCold;
+        }
+        trace.uncompressedSize = (size_t)uncompressedSize;
+        trace.compressedSize = (size_t)compressedSize;
+        trace.dctx = dctx;
+        ZSTD_trace_decompress_end(dctx->traceCtx, &trace);
+    }
+#else
+    (void)dctx;
+    (void)uncompressedSize;
+    (void)compressedSize;
+    (void)streaming;
+#endif
+}
+
+
+/*! ZSTD_decompressFrame() :
+ * @dctx must be properly initialized
+ *  will update *srcPtr and *srcSizePtr,
+ *  to make *srcPtr progress by one frame. */
+static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
+                                   void* dst, size_t dstCapacity,
+                             const void** srcPtr, size_t *srcSizePtr)
+{
+    const BYTE* const istart = (const BYTE*)(*srcPtr);
+    const BYTE* ip = istart;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = dstCapacity != 0 ? ostart + dstCapacity : ostart;
+    BYTE* op = ostart;
+    size_t remainingSrcSize = *srcSizePtr;
+
+    DEBUGLOG(4, "ZSTD_decompressFrame (srcSize:%i)", (int)*srcSizePtr);
+
+    /* check */
+    RETURN_ERROR_IF(
+        remainingSrcSize < ZSTD_FRAMEHEADERSIZE_MIN(dctx->format)+ZSTD_blockHeaderSize,
+        srcSize_wrong, "");
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize_internal(
+                ip, ZSTD_FRAMEHEADERSIZE_PREFIX(dctx->format), dctx->format);
+        if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
+        RETURN_ERROR_IF(remainingSrcSize < frameHeaderSize+ZSTD_blockHeaderSize,
+                        srcSize_wrong, "");
+        FORWARD_IF_ERROR( ZSTD_decodeFrameHeader(dctx, ip, frameHeaderSize) , "");
+        ip += frameHeaderSize; remainingSrcSize -= frameHeaderSize;
+    }
+
+    /* Shrink the blockSizeMax if enabled */
+    if (dctx->maxBlockSizeParam != 0)
+        dctx->fParams.blockSizeMax = MIN(dctx->fParams.blockSizeMax, (unsigned)dctx->maxBlockSizeParam);
+
+    /* Loop on each block */
+    while (1) {
+        BYTE* oBlockEnd = oend;
+        size_t decodedSize;
+        blockProperties_t blockProperties;
+        size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSrcSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSrcSize -= ZSTD_blockHeaderSize;
+        RETURN_ERROR_IF(cBlockSize > remainingSrcSize, srcSize_wrong, "");
+
+        if (ip >= op && ip < oBlockEnd) {
+            /* We are decompressing in-place. Limit the output pointer so that we
+             * don't overwrite the block that we are currently reading. This will
+             * fail decompression if the input & output pointers aren't spaced
+             * far enough apart.
+             *
+             * This is important to set, even when the pointers are far enough
+             * apart, because ZSTD_decompressBlock_internal() can decide to store
+             * literals in the output buffer, after the block it is decompressing.
+             * Since we don't want anything to overwrite our input, we have to tell
+             * ZSTD_decompressBlock_internal to never write past ip.
+             *
+             * See ZSTD_allocateLiteralsBuffer() for reference.
+             */
+            oBlockEnd = op + (ip - op);
+        }
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            assert(dctx->isFrameDecompression == 1);
+            decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oBlockEnd-op), ip, cBlockSize, not_streaming);
+            break;
+        case bt_raw :
+            /* Use oend instead of oBlockEnd because this function is safe to overlap. It uses memmove. */
+            decodedSize = ZSTD_copyRawBlock(op, (size_t)(oend-op), ip, cBlockSize);
+            break;
+        case bt_rle :
+            decodedSize = ZSTD_setRleBlock(op, (size_t)(oBlockEnd-op), *ip, blockProperties.origSize);
+            break;
+        case bt_reserved :
+        default:
+            RETURN_ERROR(corruption_detected, "invalid block type");
+        }
+        FORWARD_IF_ERROR(decodedSize, "Block decompression failure");
+        DEBUGLOG(5, "Decompressed block of dSize = %u", (unsigned)decodedSize);
+        if (dctx->validateChecksum) {
+            XXH64_update(&dctx->xxhState, op, decodedSize);
+        }
+        if (decodedSize) /* support dst = NULL,0 */ {
+            op += decodedSize;
+        }
+        assert(ip != NULL);
+        ip += cBlockSize;
+        remainingSrcSize -= cBlockSize;
+        if (blockProperties.lastBlock) break;
+    }
+
+    if (dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+        RETURN_ERROR_IF((U64)(op-ostart) != dctx->fParams.frameContentSize,
+                        corruption_detected, "");
+    }
+    if (dctx->fParams.checksumFlag) { /* Frame content checksum verification */
+        RETURN_ERROR_IF(remainingSrcSize<4, checksum_wrong, "");
+        if (!dctx->forceIgnoreChecksum) {
+            U32 const checkCalc = (U32)XXH64_digest(&dctx->xxhState);
+            U32 checkRead;
+            checkRead = MEM_readLE32(ip);
+            RETURN_ERROR_IF(checkRead != checkCalc, checksum_wrong, "");
+        }
+        ip += 4;
+        remainingSrcSize -= 4;
+    }
+    ZSTD_DCtx_trace_end(dctx, (U64)(op-ostart), (U64)(ip-istart), /* streaming */ 0);
+    /* Allow caller to get size read */
+    DEBUGLOG(4, "ZSTD_decompressFrame: decompressed frame of size %i, consuming %i bytes of input", (int)(op-ostart), (int)(ip - (const BYTE*)*srcPtr));
+    *srcPtr = ip;
+    *srcSizePtr = remainingSrcSize;
+    return (size_t)(op-ostart);
+}
+
+static
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
+                                        void* dst, size_t dstCapacity,
+                                  const void* src, size_t srcSize,
+                                  const void* dict, size_t dictSize,
+                                  const ZSTD_DDict* ddict)
+{
+    void* const dststart = dst;
+    int moreThan1Frame = 0;
+
+    DEBUGLOG(5, "ZSTD_decompressMultiFrame");
+    assert(dict==NULL || ddict==NULL);  /* either dict or ddict set, not both */
+
+    if (ddict) {
+        dict = ZSTD_DDict_dictContent(ddict);
+        dictSize = ZSTD_DDict_dictSize(ddict);
+    }
+
+    while (srcSize >= ZSTD_startingInputLength(dctx->format)) {
+
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+        if (dctx->format == ZSTD_f_zstd1 && ZSTD_isLegacy(src, srcSize)) {
+            size_t decodedSize;
+            size_t const frameSize = ZSTD_findFrameCompressedSizeLegacy(src, srcSize);
+            if (ZSTD_isError(frameSize)) return frameSize;
+            RETURN_ERROR_IF(dctx->staticSize, memory_allocation,
+                "legacy support is not compatible with static dctx");
+
+            decodedSize = ZSTD_decompressLegacy(dst, dstCapacity, src, frameSize, dict, dictSize);
+            if (ZSTD_isError(decodedSize)) return decodedSize;
+
+            {
+                unsigned long long const expectedSize = ZSTD_getFrameContentSize(src, srcSize);
+                RETURN_ERROR_IF(expectedSize == ZSTD_CONTENTSIZE_ERROR, corruption_detected, "Corrupted frame header!");
+                if (expectedSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+                    RETURN_ERROR_IF(expectedSize != decodedSize, corruption_detected,
+                        "Frame header size does not match decoded size!");
+                }
+            }
+
+            assert(decodedSize <= dstCapacity);
+            dst = (BYTE*)dst + decodedSize;
+            dstCapacity -= decodedSize;
+
+            src = (const BYTE*)src + frameSize;
+            srcSize -= frameSize;
+
+            continue;
+        }
+#endif
+
+        if (dctx->format == ZSTD_f_zstd1 && srcSize >= 4) {
+            U32 const magicNumber = MEM_readLE32(src);
+            DEBUGLOG(5, "reading magic number %08X", (unsigned)magicNumber);
+            if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+                /* skippable frame detected : skip it */
+                size_t const skippableSize = readSkippableFrameSize(src, srcSize);
+                FORWARD_IF_ERROR(skippableSize, "invalid skippable frame");
+                assert(skippableSize <= srcSize);
+
+                src = (const BYTE *)src + skippableSize;
+                srcSize -= skippableSize;
+                continue; /* check next frame */
+        }   }
+
+        if (ddict) {
+            /* we were called from ZSTD_decompress_usingDDict */
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDDict(dctx, ddict), "");
+        } else {
+            /* this will initialize correctly with no dict if dict == NULL, so
+             * use this in all cases but ddict */
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDict(dctx, dict, dictSize), "");
+        }
+        ZSTD_checkContinuity(dctx, dst, dstCapacity);
+
+        {   const size_t res = ZSTD_decompressFrame(dctx, dst, dstCapacity,
+                                                    &src, &srcSize);
+            RETURN_ERROR_IF(
+                (ZSTD_getErrorCode(res) == ZSTD_error_prefix_unknown)
+             && (moreThan1Frame==1),
+                srcSize_wrong,
+                "At least one frame successfully completed, "
+                "but following bytes are garbage: "
+                "it's more likely to be a srcSize error, "
+                "specifying more input bytes than size of frame(s). "
+                "Note: one could be unlucky, it might be a corruption error instead, "
+                "happening right at the place where we expect zstd magic bytes. "
+                "But this is _much_ less likely than a srcSize field error.");
+            if (ZSTD_isError(res)) return res;
+            assert(res <= dstCapacity);
+            if (res != 0)
+                dst = (BYTE*)dst + res;
+            dstCapacity -= res;
+        }
+        moreThan1Frame = 1;
+    }  /* while (srcSize >= ZSTD_frameHeaderSize_prefix) */
+
+    RETURN_ERROR_IF(srcSize, srcSize_wrong, "input not entirely consumed");
+
+    return (size_t)((BYTE*)dst - (BYTE*)dststart);
+}
+
+size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                           const void* src, size_t srcSize,
+                           const void* dict, size_t dictSize)
+{
+    return ZSTD_decompressMultiFrame(dctx, dst, dstCapacity, src, srcSize, dict, dictSize, NULL);
+}
+
+
+static ZSTD_DDict const* ZSTD_getDDict(ZSTD_DCtx* dctx)
+{
+    switch (dctx->dictUses) {
+    default:
+        assert(0 /* Impossible */);
+        ZSTD_FALLTHROUGH;
+    case ZSTD_dont_use:
+        ZSTD_clearDict(dctx);
+        return NULL;
+    case ZSTD_use_indefinitely:
+        return dctx->ddict;
+    case ZSTD_use_once:
+        dctx->dictUses = ZSTD_dont_use;
+        return dctx->ddict;
+    }
+}
+
+size_t ZSTD_decompressDCtx(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTD_decompress_usingDDict(dctx, dst, dstCapacity, src, srcSize, ZSTD_getDDict(dctx));
+}
+
+
+size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+#if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE>=1)
+    size_t regenSize;
+    ZSTD_DCtx* const dctx =  ZSTD_createDCtx_internal(ZSTD_defaultCMem);
+    RETURN_ERROR_IF(dctx==NULL, memory_allocation, "NULL pointer!");
+    regenSize = ZSTD_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
+    ZSTD_freeDCtx(dctx);
+    return regenSize;
+#else   /* stack mode */
+    ZSTD_DCtx dctx;
+    ZSTD_initDCtx_internal(&dctx);
+    return ZSTD_decompressDCtx(&dctx, dst, dstCapacity, src, srcSize);
+#endif
+}
+
+
+/*-**************************************
+*   Advanced Streaming Decompression API
+*   Bufferless and synchronous
+****************************************/
+size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx) { return dctx->expected; }
+
+/**
+ * Similar to ZSTD_nextSrcSizeToDecompress(), but when a block input can be streamed, we
+ * allow taking a partial block as the input. Currently only raw uncompressed blocks can
+ * be streamed.
+ *
+ * For blocks that can be streamed, this allows us to reduce the latency until we produce
+ * output, and avoid copying the input.
+ *
+ * @param inputSize - The total amount of input that the caller currently has.
+ */
+static size_t ZSTD_nextSrcSizeToDecompressWithInputSize(ZSTD_DCtx* dctx, size_t inputSize) {
+    if (!(dctx->stage == ZSTDds_decompressBlock || dctx->stage == ZSTDds_decompressLastBlock))
+        return dctx->expected;
+    if (dctx->bType != bt_raw)
+        return dctx->expected;
+    return BOUNDED(1, inputSize, dctx->expected);
+}
+
+ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx) {
+    switch(dctx->stage)
+    {
+    default:   /* should not happen */
+        assert(0);
+        ZSTD_FALLTHROUGH;
+    case ZSTDds_getFrameHeaderSize:
+        ZSTD_FALLTHROUGH;
+    case ZSTDds_decodeFrameHeader:
+        return ZSTDnit_frameHeader;
+    case ZSTDds_decodeBlockHeader:
+        return ZSTDnit_blockHeader;
+    case ZSTDds_decompressBlock:
+        return ZSTDnit_block;
+    case ZSTDds_decompressLastBlock:
+        return ZSTDnit_lastBlock;
+    case ZSTDds_checkChecksum:
+        return ZSTDnit_checksum;
+    case ZSTDds_decodeSkippableHeader:
+        ZSTD_FALLTHROUGH;
+    case ZSTDds_skipFrame:
+        return ZSTDnit_skippableFrame;
+    }
+}
+
+static int ZSTD_isSkipFrame(ZSTD_DCtx* dctx) { return dctx->stage == ZSTDds_skipFrame; }
+
+/** ZSTD_decompressContinue() :
+ *  srcSize : must be the exact nb of bytes expected (see ZSTD_nextSrcSizeToDecompress())
+ *  @return : nb of bytes generated into `dst` (necessarily <= `dstCapacity)
+ *            or an error code, which can be tested using ZSTD_isError() */
+size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_decompressContinue (srcSize:%u)", (unsigned)srcSize);
+    /* Sanity check */
+    RETURN_ERROR_IF(srcSize != ZSTD_nextSrcSizeToDecompressWithInputSize(dctx, srcSize), srcSize_wrong, "not allowed");
+    ZSTD_checkContinuity(dctx, dst, dstCapacity);
+
+    dctx->processedCSize += srcSize;
+
+    switch (dctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        assert(src != NULL);
+        if (dctx->format == ZSTD_f_zstd1) {  /* allows header */
+            assert(srcSize >= ZSTD_FRAMEIDSIZE);  /* to read skippable magic number */
+            if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
+                ZSTD_memcpy(dctx->headerBuffer, src, srcSize);
+                dctx->expected = ZSTD_SKIPPABLEHEADERSIZE - srcSize;  /* remaining to load to get full skippable frame header */
+                dctx->stage = ZSTDds_decodeSkippableHeader;
+                return 0;
+        }   }
+        dctx->headerSize = ZSTD_frameHeaderSize_internal(src, srcSize, dctx->format);
+        if (ZSTD_isError(dctx->headerSize)) return dctx->headerSize;
+        ZSTD_memcpy(dctx->headerBuffer, src, srcSize);
+        dctx->expected = dctx->headerSize - srcSize;
+        dctx->stage = ZSTDds_decodeFrameHeader;
+        return 0;
+
+    case ZSTDds_decodeFrameHeader:
+        assert(src != NULL);
+        ZSTD_memcpy(dctx->headerBuffer + (dctx->headerSize - srcSize), src, srcSize);
+        FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize), "");
+        dctx->expected = ZSTD_blockHeaderSize;
+        dctx->stage = ZSTDds_decodeBlockHeader;
+        return 0;
+
+    case ZSTDds_decodeBlockHeader:
+        {   blockProperties_t bp;
+            size_t const cBlockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+            if (ZSTD_isError(cBlockSize)) return cBlockSize;
+            RETURN_ERROR_IF(cBlockSize > dctx->fParams.blockSizeMax, corruption_detected, "Block Size Exceeds Maximum");
+            dctx->expected = cBlockSize;
+            dctx->bType = bp.blockType;
+            dctx->rleSize = bp.origSize;
+            if (cBlockSize) {
+                dctx->stage = bp.lastBlock ? ZSTDds_decompressLastBlock : ZSTDds_decompressBlock;
+                return 0;
+            }
+            /* empty block */
+            if (bp.lastBlock) {
+                if (dctx->fParams.checksumFlag) {
+                    dctx->expected = 4;
+                    dctx->stage = ZSTDds_checkChecksum;
+                } else {
+                    dctx->expected = 0; /* end of frame */
+                    dctx->stage = ZSTDds_getFrameHeaderSize;
+                }
+            } else {
+                dctx->expected = ZSTD_blockHeaderSize;  /* jump to next header */
+                dctx->stage = ZSTDds_decodeBlockHeader;
+            }
+            return 0;
+        }
+
+    case ZSTDds_decompressLastBlock:
+    case ZSTDds_decompressBlock:
+        DEBUGLOG(5, "ZSTD_decompressContinue: case ZSTDds_decompressBlock");
+        {   size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                DEBUGLOG(5, "ZSTD_decompressContinue: case bt_compressed");
+                assert(dctx->isFrameDecompression == 1);
+                rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, is_streaming);
+                dctx->expected = 0;  /* Streaming not supported */
+                break;
+            case bt_raw :
+                assert(srcSize <= dctx->expected);
+                rSize = ZSTD_copyRawBlock(dst, dstCapacity, src, srcSize);
+                FORWARD_IF_ERROR(rSize, "ZSTD_copyRawBlock failed");
+                assert(rSize == srcSize);
+                dctx->expected -= rSize;
+                break;
+            case bt_rle :
+                rSize = ZSTD_setRleBlock(dst, dstCapacity, *(const BYTE*)src, dctx->rleSize);
+                dctx->expected = 0;  /* Streaming not supported */
+                break;
+            case bt_reserved :   /* should never happen */
+            default:
+                RETURN_ERROR(corruption_detected, "invalid block type");
+            }
+            FORWARD_IF_ERROR(rSize, "");
+            RETURN_ERROR_IF(rSize > dctx->fParams.blockSizeMax, corruption_detected, "Decompressed Block Size Exceeds Maximum");
+            DEBUGLOG(5, "ZSTD_decompressContinue: decoded size from block : %u", (unsigned)rSize);
+            dctx->decodedSize += rSize;
+            if (dctx->validateChecksum) XXH64_update(&dctx->xxhState, dst, rSize);
+            dctx->previousDstEnd = (char*)dst + rSize;
+
+            /* Stay on the same stage until we are finished streaming the block. */
+            if (dctx->expected > 0) {
+                return rSize;
+            }
+
+            if (dctx->stage == ZSTDds_decompressLastBlock) {   /* end of frame */
+                DEBUGLOG(4, "ZSTD_decompressContinue: decoded size from frame : %u", (unsigned)dctx->decodedSize);
+                RETURN_ERROR_IF(
+                    dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
+                 && dctx->decodedSize != dctx->fParams.frameContentSize,
+                    corruption_detected, "");
+                if (dctx->fParams.checksumFlag) {  /* another round for frame checksum */
+                    dctx->expected = 4;
+                    dctx->stage = ZSTDds_checkChecksum;
+                } else {
+                    ZSTD_DCtx_trace_end(dctx, dctx->decodedSize, dctx->processedCSize, /* streaming */ 1);
+                    dctx->expected = 0;   /* ends here */
+                    dctx->stage = ZSTDds_getFrameHeaderSize;
+                }
+            } else {
+                dctx->stage = ZSTDds_decodeBlockHeader;
+                dctx->expected = ZSTD_blockHeaderSize;
+            }
+            return rSize;
+        }
+
+    case ZSTDds_checkChecksum:
+        assert(srcSize == 4);  /* guaranteed by dctx->expected */
+        {
+            if (dctx->validateChecksum) {
+                U32 const h32 = (U32)XXH64_digest(&dctx->xxhState);
+                U32 const check32 = MEM_readLE32(src);
+                DEBUGLOG(4, "ZSTD_decompressContinue: checksum : calculated %08X :: %08X read", (unsigned)h32, (unsigned)check32);
+                RETURN_ERROR_IF(check32 != h32, checksum_wrong, "");
+            }
+            ZSTD_DCtx_trace_end(dctx, dctx->decodedSize, dctx->processedCSize, /* streaming */ 1);
+            dctx->expected = 0;
+            dctx->stage = ZSTDds_getFrameHeaderSize;
+            return 0;
+        }
+
+    case ZSTDds_decodeSkippableHeader:
+        assert(src != NULL);
+        assert(srcSize <= ZSTD_SKIPPABLEHEADERSIZE);
+        assert(dctx->format != ZSTD_f_zstd1_magicless);
+        ZSTD_memcpy(dctx->headerBuffer + (ZSTD_SKIPPABLEHEADERSIZE - srcSize), src, srcSize);   /* complete skippable header */
+        dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_FRAMEIDSIZE);   /* note : dctx->expected can grow seriously large, beyond local buffer size */
+        dctx->stage = ZSTDds_skipFrame;
+        return 0;
+
+    case ZSTDds_skipFrame:
+        dctx->expected = 0;
+        dctx->stage = ZSTDds_getFrameHeaderSize;
+        return 0;
+
+    default:
+        assert(0);   /* impossible */
+        RETURN_ERROR(GENERIC, "impossible to reach");   /* some compilers require default to do something */
+    }
+}
+
+
+static size_t ZSTD_refDictContent(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->virtualStart = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
+    dctx->prefixStart = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    dctx->dictContentBeginForFuzzing = dctx->prefixStart;
+    dctx->dictContentEndForFuzzing = dctx->previousDstEnd;
+#endif
+    return 0;
+}
+
+/*! ZSTD_loadDEntropy() :
+ *  dict : must point at beginning of a valid zstd dictionary.
+ * @return : size of entropy tables read */
+size_t
+ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
+                  const void* const dict, size_t const dictSize)
+{
+    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* const dictEnd = dictPtr + dictSize;
+
+    RETURN_ERROR_IF(dictSize <= 8, dictionary_corrupted, "dict is too small");
+    assert(MEM_readLE32(dict) == ZSTD_MAGIC_DICTIONARY);   /* dict must be valid */
+    dictPtr += 8;   /* skip header = magic + dictID */
+
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_entropyDTables_t, OFTable) == offsetof(ZSTD_entropyDTables_t, LLTable) + sizeof(entropy->LLTable));
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_entropyDTables_t, MLTable) == offsetof(ZSTD_entropyDTables_t, OFTable) + sizeof(entropy->OFTable));
+    ZSTD_STATIC_ASSERT(sizeof(entropy->LLTable) + sizeof(entropy->OFTable) + sizeof(entropy->MLTable) >= HUF_DECOMPRESS_WORKSPACE_SIZE);
+    {   void* const workspace = &entropy->LLTable;   /* use fse tables as temporary workspace; implies fse tables are grouped together */
+        size_t const workspaceSize = sizeof(entropy->LLTable) + sizeof(entropy->OFTable) + sizeof(entropy->MLTable);
+#ifdef HUF_FORCE_DECOMPRESS_X1
+        /* in minimal huffman, we always use X1 variants */
+        size_t const hSize = HUF_readDTableX1_wksp(entropy->hufTable,
+                                                dictPtr, dictEnd - dictPtr,
+                                                workspace, workspaceSize, /* flags */ 0);
+#else
+        size_t const hSize = HUF_readDTableX2_wksp(entropy->hufTable,
+                                                dictPtr, (size_t)(dictEnd - dictPtr),
+                                                workspace, workspaceSize, /* flags */ 0);
+#endif
+        RETURN_ERROR_IF(HUF_isError(hSize), dictionary_corrupted, "");
+        dictPtr += hSize;
+    }
+
+    {   short offcodeNCount[MaxOff+1];
+        unsigned offcodeMaxValue = MaxOff, offcodeLog;
+        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(offcodeMaxValue > MaxOff, dictionary_corrupted, "");
+        RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted, "");
+        ZSTD_buildFSETable( entropy->OFTable,
+                            offcodeNCount, offcodeMaxValue,
+                            OF_base, OF_bits,
+                            offcodeLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */0);
+        dictPtr += offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(matchlengthMaxValue > MaxML, dictionary_corrupted, "");
+        RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted, "");
+        ZSTD_buildFSETable( entropy->MLTable,
+                            matchlengthNCount, matchlengthMaxValue,
+                            ML_base, ML_bits,
+                            matchlengthLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */ 0);
+        dictPtr += matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
+        RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(litlengthMaxValue > MaxLL, dictionary_corrupted, "");
+        RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted, "");
+        ZSTD_buildFSETable( entropy->LLTable,
+                            litlengthNCount, litlengthMaxValue,
+                            LL_base, LL_bits,
+                            litlengthLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */ 0);
+        dictPtr += litlengthHeaderSize;
+    }
+
+    RETURN_ERROR_IF(dictPtr+12 > dictEnd, dictionary_corrupted, "");
+    {   int i;
+        size_t const dictContentSize = (size_t)(dictEnd - (dictPtr+12));
+        for (i=0; i<3; i++) {
+            U32 const rep = MEM_readLE32(dictPtr); dictPtr += 4;
+            RETURN_ERROR_IF(rep==0 || rep > dictContentSize,
+                            dictionary_corrupted, "");
+            entropy->rep[i] = rep;
+    }   }
+
+    return (size_t)(dictPtr - (const BYTE*)dict);
+}
+
+static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    if (dictSize < 8) return ZSTD_refDictContent(dctx, dict, dictSize);
+    {   U32 const magic = MEM_readLE32(dict);
+        if (magic != ZSTD_MAGIC_DICTIONARY) {
+            return ZSTD_refDictContent(dctx, dict, dictSize);   /* pure content mode */
+    }   }
+    dctx->dictID = MEM_readLE32((const char*)dict + ZSTD_FRAMEIDSIZE);
+
+    /* load entropy tables */
+    {   size_t const eSize = ZSTD_loadDEntropy(&dctx->entropy, dict, dictSize);
+        RETURN_ERROR_IF(ZSTD_isError(eSize), dictionary_corrupted, "");
+        dict = (const char*)dict + eSize;
+        dictSize -= eSize;
+    }
+    dctx->litEntropy = dctx->fseEntropy = 1;
+
+    /* reference dictionary content */
+    return ZSTD_refDictContent(dctx, dict, dictSize);
+}
+
+size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
+{
+    assert(dctx != NULL);
+#if ZSTD_TRACE
+    dctx->traceCtx = (ZSTD_trace_decompress_begin != NULL) ? ZSTD_trace_decompress_begin(dctx) : 0;
+#endif
+    dctx->expected = ZSTD_startingInputLength(dctx->format);  /* dctx->format must be properly set */
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->processedCSize = 0;
+    dctx->decodedSize = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->prefixStart = NULL;
+    dctx->virtualStart = NULL;
+    dctx->dictEnd = NULL;
+    dctx->entropy.hufTable[0] = (HUF_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001);  /* cover both little and big endian */
+    dctx->litEntropy = dctx->fseEntropy = 0;
+    dctx->dictID = 0;
+    dctx->bType = bt_reserved;
+    dctx->isFrameDecompression = 1;
+    ZSTD_STATIC_ASSERT(sizeof(dctx->entropy.rep) == sizeof(repStartValue));
+    ZSTD_memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
+    dctx->LLTptr = dctx->entropy.LLTable;
+    dctx->MLTptr = dctx->entropy.MLTable;
+    dctx->OFTptr = dctx->entropy.OFTable;
+    dctx->HUFptr = dctx->entropy.hufTable;
+    return 0;
+}
+
+size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    FORWARD_IF_ERROR( ZSTD_decompressBegin(dctx) , "");
+    if (dict && dictSize)
+        RETURN_ERROR_IF(
+            ZSTD_isError(ZSTD_decompress_insertDictionary(dctx, dict, dictSize)),
+            dictionary_corrupted, "");
+    return 0;
+}
+
+
+/* ======   ZSTD_DDict   ====== */
+
+size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
+{
+    DEBUGLOG(4, "ZSTD_decompressBegin_usingDDict");
+    assert(dctx != NULL);
+    if (ddict) {
+        const char* const dictStart = (const char*)ZSTD_DDict_dictContent(ddict);
+        size_t const dictSize = ZSTD_DDict_dictSize(ddict);
+        const void* const dictEnd = dictStart + dictSize;
+        dctx->ddictIsCold = (dctx->dictEnd != dictEnd);
+        DEBUGLOG(4, "DDict is %s",
+                    dctx->ddictIsCold ? "~cold~" : "hot!");
+    }
+    FORWARD_IF_ERROR( ZSTD_decompressBegin(dctx) , "");
+    if (ddict) {   /* NULL ddict is equivalent to no dictionary */
+        ZSTD_copyDDictParameters(dctx, ddict);
+    }
+    return 0;
+}
+
+/*! ZSTD_getDictID_fromDict() :
+ *  Provides the dictID stored within dictionary.
+ *  if @return == 0, the dictionary is not conformant with Zstandard specification.
+ *  It can still be loaded, but as a content-only dictionary. */
+unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
+{
+    if (dictSize < 8) return 0;
+    if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) return 0;
+    return MEM_readLE32((const char*)dict + ZSTD_FRAMEIDSIZE);
+}
+
+/*! ZSTD_getDictID_fromFrame() :
+ *  Provides the dictID required to decompress frame stored within `src`.
+ *  If @return == 0, the dictID could not be decoded.
+ *  This could for one of the following reasons :
+ *  - The frame does not require a dictionary (most common case).
+ *  - The frame was built with dictID intentionally removed.
+ *    Needed dictionary is a hidden piece of information.
+ *    Note : this use case also happens when using a non-conformant dictionary.
+ *  - `srcSize` is too small, and as a result, frame header could not be decoded.
+ *    Note : possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`.
+ *  - This is not a Zstandard frame.
+ *  When identifying the exact failure cause, it's possible to use
+ *  ZSTD_getFrameHeader(), which will provide a more precise error code. */
+unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize)
+{
+    ZSTD_FrameHeader zfp = { 0, 0, 0, ZSTD_frame, 0, 0, 0, 0, 0 };
+    size_t const hError = ZSTD_getFrameHeader(&zfp, src, srcSize);
+    if (ZSTD_isError(hError)) return 0;
+    return zfp.dictID;
+}
+
+
+/*! ZSTD_decompress_usingDDict() :
+*   Decompression using a pre-digested Dictionary
+*   Use dictionary without significant overhead. */
+size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                            const ZSTD_DDict* ddict)
+{
+    /* pass content and size in case legacy frames are encountered */
+    return ZSTD_decompressMultiFrame(dctx, dst, dstCapacity, src, srcSize,
+                                     NULL, 0,
+                                     ddict);
+}
+
+
+/*=====================================
+*   Streaming decompression
+*====================================*/
+
+ZSTD_DStream* ZSTD_createDStream(void)
+{
+    DEBUGLOG(3, "ZSTD_createDStream");
+    return ZSTD_createDCtx_internal(ZSTD_defaultCMem);
+}
+
+ZSTD_DStream* ZSTD_initStaticDStream(void *workspace, size_t workspaceSize)
+{
+    return ZSTD_initStaticDCtx(workspace, workspaceSize);
+}
+
+ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem)
+{
+    return ZSTD_createDCtx_internal(customMem);
+}
+
+size_t ZSTD_freeDStream(ZSTD_DStream* zds)
+{
+    return ZSTD_freeDCtx(zds);
+}
+
+
+/* ***  Initialization  *** */
+
+size_t ZSTD_DStreamInSize(void)  { return ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize; }
+size_t ZSTD_DStreamOutSize(void) { return ZSTD_BLOCKSIZE_MAX; }
+
+size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx,
+                                   const void* dict, size_t dictSize,
+                                         ZSTD_dictLoadMethod_e dictLoadMethod,
+                                         ZSTD_dictContentType_e dictContentType)
+{
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    ZSTD_clearDict(dctx);
+    if (dict && dictSize != 0) {
+        dctx->ddictLocal = ZSTD_createDDict_advanced(dict, dictSize, dictLoadMethod, dictContentType, dctx->customMem);
+        RETURN_ERROR_IF(dctx->ddictLocal == NULL, memory_allocation, "NULL pointer!");
+        dctx->ddict = dctx->ddictLocal;
+        dctx->dictUses = ZSTD_use_indefinitely;
+    }
+    return 0;
+}
+
+size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto);
+}
+
+size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto);
+}
+
+size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
+{
+    FORWARD_IF_ERROR(ZSTD_DCtx_loadDictionary_advanced(dctx, prefix, prefixSize, ZSTD_dlm_byRef, dictContentType), "");
+    dctx->dictUses = ZSTD_use_once;
+    return 0;
+}
+
+size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize)
+{
+    return ZSTD_DCtx_refPrefix_advanced(dctx, prefix, prefixSize, ZSTD_dct_rawContent);
+}
+
+
+/* ZSTD_initDStream_usingDict() :
+ * return : expected size, aka ZSTD_startingInputLength().
+ * this function cannot fail */
+size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize)
+{
+    DEBUGLOG(4, "ZSTD_initDStream_usingDict");
+    FORWARD_IF_ERROR( ZSTD_DCtx_reset(zds, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_DCtx_loadDictionary(zds, dict, dictSize) , "");
+    return ZSTD_startingInputLength(zds->format);
+}
+
+/* note : this variant can't fail */
+size_t ZSTD_initDStream(ZSTD_DStream* zds)
+{
+    DEBUGLOG(4, "ZSTD_initDStream");
+    FORWARD_IF_ERROR(ZSTD_DCtx_reset(zds, ZSTD_reset_session_only), "");
+    FORWARD_IF_ERROR(ZSTD_DCtx_refDDict(zds, NULL), "");
+    return ZSTD_startingInputLength(zds->format);
+}
+
+/* ZSTD_initDStream_usingDDict() :
+ * ddict will just be referenced, and must outlive decompression session
+ * this function cannot fail */
+size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
+{
+    DEBUGLOG(4, "ZSTD_initDStream_usingDDict");
+    FORWARD_IF_ERROR( ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_DCtx_refDDict(dctx, ddict) , "");
+    return ZSTD_startingInputLength(dctx->format);
+}
+
+/* ZSTD_resetDStream() :
+ * return : expected size, aka ZSTD_startingInputLength().
+ * this function cannot fail */
+size_t ZSTD_resetDStream(ZSTD_DStream* dctx)
+{
+    DEBUGLOG(4, "ZSTD_resetDStream");
+    FORWARD_IF_ERROR(ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only), "");
+    return ZSTD_startingInputLength(dctx->format);
+}
+
+
+size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
+{
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    ZSTD_clearDict(dctx);
+    if (ddict) {
+        dctx->ddict = ddict;
+        dctx->dictUses = ZSTD_use_indefinitely;
+        if (dctx->refMultipleDDicts == ZSTD_rmd_refMultipleDDicts) {
+            if (dctx->ddictSet == NULL) {
+                dctx->ddictSet = ZSTD_createDDictHashSet(dctx->customMem);
+                if (!dctx->ddictSet) {
+                    RETURN_ERROR(memory_allocation, "Failed to allocate memory for hash set!");
+                }
+            }
+            assert(!dctx->staticSize);  /* Impossible: ddictSet cannot have been allocated if static dctx */
+            FORWARD_IF_ERROR(ZSTD_DDictHashSet_addDDict(dctx->ddictSet, ddict, dctx->customMem), "");
+        }
+    }
+    return 0;
+}
+
+/* ZSTD_DCtx_setMaxWindowSize() :
+ * note : no direct equivalence in ZSTD_DCtx_setParameter,
+ * since this version sets windowSize, and the other sets windowLog */
+size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize)
+{
+    ZSTD_bounds const bounds = ZSTD_dParam_getBounds(ZSTD_d_windowLogMax);
+    size_t const min = (size_t)1 << bounds.lowerBound;
+    size_t const max = (size_t)1 << bounds.upperBound;
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    RETURN_ERROR_IF(maxWindowSize < min, parameter_outOfBound, "");
+    RETURN_ERROR_IF(maxWindowSize > max, parameter_outOfBound, "");
+    dctx->maxWindowSize = maxWindowSize;
+    return 0;
+}
+
+size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format)
+{
+    return ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, (int)format);
+}
+
+ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
+{
+    ZSTD_bounds bounds = { 0, 0, 0 };
+    switch(dParam) {
+        case ZSTD_d_windowLogMax:
+            bounds.lowerBound = ZSTD_WINDOWLOG_ABSOLUTEMIN;
+            bounds.upperBound = ZSTD_WINDOWLOG_MAX;
+            return bounds;
+        case ZSTD_d_format:
+            bounds.lowerBound = (int)ZSTD_f_zstd1;
+            bounds.upperBound = (int)ZSTD_f_zstd1_magicless;
+            ZSTD_STATIC_ASSERT(ZSTD_f_zstd1 < ZSTD_f_zstd1_magicless);
+            return bounds;
+        case ZSTD_d_stableOutBuffer:
+            bounds.lowerBound = (int)ZSTD_bm_buffered;
+            bounds.upperBound = (int)ZSTD_bm_stable;
+            return bounds;
+        case ZSTD_d_forceIgnoreChecksum:
+            bounds.lowerBound = (int)ZSTD_d_validateChecksum;
+            bounds.upperBound = (int)ZSTD_d_ignoreChecksum;
+            return bounds;
+        case ZSTD_d_refMultipleDDicts:
+            bounds.lowerBound = (int)ZSTD_rmd_refSingleDDict;
+            bounds.upperBound = (int)ZSTD_rmd_refMultipleDDicts;
+            return bounds;
+        case ZSTD_d_disableHuffmanAssembly:
+            bounds.lowerBound = 0;
+            bounds.upperBound = 1;
+            return bounds;
+        case ZSTD_d_maxBlockSize:
+            bounds.lowerBound = ZSTD_BLOCKSIZE_MAX_MIN;
+            bounds.upperBound = ZSTD_BLOCKSIZE_MAX;
+            return bounds;
+
+        default:;
+    }
+    bounds.error = ERROR(parameter_unsupported);
+    return bounds;
+}
+
+/* ZSTD_dParam_withinBounds:
+ * @return 1 if value is within dParam bounds,
+ * 0 otherwise */
+static int ZSTD_dParam_withinBounds(ZSTD_dParameter dParam, int value)
+{
+    ZSTD_bounds const bounds = ZSTD_dParam_getBounds(dParam);
+    if (ZSTD_isError(bounds.error)) return 0;
+    if (value < bounds.lowerBound) return 0;
+    if (value > bounds.upperBound) return 0;
+    return 1;
+}
+
+#define CHECK_DBOUNDS(p,v) {                \
+    RETURN_ERROR_IF(!ZSTD_dParam_withinBounds(p, v), parameter_outOfBound, ""); \
+}
+
+size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value)
+{
+    switch (param) {
+        case ZSTD_d_windowLogMax:
+            *value = (int)ZSTD_highbit32((U32)dctx->maxWindowSize);
+            return 0;
+        case ZSTD_d_format:
+            *value = (int)dctx->format;
+            return 0;
+        case ZSTD_d_stableOutBuffer:
+            *value = (int)dctx->outBufferMode;
+            return 0;
+        case ZSTD_d_forceIgnoreChecksum:
+            *value = (int)dctx->forceIgnoreChecksum;
+            return 0;
+        case ZSTD_d_refMultipleDDicts:
+            *value = (int)dctx->refMultipleDDicts;
+            return 0;
+        case ZSTD_d_disableHuffmanAssembly:
+            *value = (int)dctx->disableHufAsm;
+            return 0;
+        case ZSTD_d_maxBlockSize:
+            *value = dctx->maxBlockSizeParam;
+            return 0;
+        default:;
+    }
+    RETURN_ERROR(parameter_unsupported, "");
+}
+
+size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value)
+{
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    switch(dParam) {
+        case ZSTD_d_windowLogMax:
+            if (value == 0) value = ZSTD_WINDOWLOG_LIMIT_DEFAULT;
+            CHECK_DBOUNDS(ZSTD_d_windowLogMax, value);
+            dctx->maxWindowSize = ((size_t)1) << value;
+            return 0;
+        case ZSTD_d_format:
+            CHECK_DBOUNDS(ZSTD_d_format, value);
+            dctx->format = (ZSTD_format_e)value;
+            return 0;
+        case ZSTD_d_stableOutBuffer:
+            CHECK_DBOUNDS(ZSTD_d_stableOutBuffer, value);
+            dctx->outBufferMode = (ZSTD_bufferMode_e)value;
+            return 0;
+        case ZSTD_d_forceIgnoreChecksum:
+            CHECK_DBOUNDS(ZSTD_d_forceIgnoreChecksum, value);
+            dctx->forceIgnoreChecksum = (ZSTD_forceIgnoreChecksum_e)value;
+            return 0;
+        case ZSTD_d_refMultipleDDicts:
+            CHECK_DBOUNDS(ZSTD_d_refMultipleDDicts, value);
+            if (dctx->staticSize != 0) {
+                RETURN_ERROR(parameter_unsupported, "Static dctx does not support multiple DDicts!");
+            }
+            dctx->refMultipleDDicts = (ZSTD_refMultipleDDicts_e)value;
+            return 0;
+        case ZSTD_d_disableHuffmanAssembly:
+            CHECK_DBOUNDS(ZSTD_d_disableHuffmanAssembly, value);
+            dctx->disableHufAsm = value != 0;
+            return 0;
+        case ZSTD_d_maxBlockSize:
+            if (value != 0) CHECK_DBOUNDS(ZSTD_d_maxBlockSize, value);
+            dctx->maxBlockSizeParam = value;
+            return 0;
+        default:;
+    }
+    RETURN_ERROR(parameter_unsupported, "");
+}
+
+size_t ZSTD_DCtx_reset(ZSTD_DCtx* dctx, ZSTD_ResetDirective reset)
+{
+    if ( (reset == ZSTD_reset_session_only)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        dctx->streamStage = zdss_init;
+        dctx->noForwardProgress = 0;
+        dctx->isFrameDecompression = 1;
+    }
+    if ( (reset == ZSTD_reset_parameters)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+        ZSTD_clearDict(dctx);
+        ZSTD_DCtx_resetParameters(dctx);
+    }
+    return 0;
+}
+
+
+size_t ZSTD_sizeof_DStream(const ZSTD_DStream* dctx)
+{
+    return ZSTD_sizeof_DCtx(dctx);
+}
+
+static size_t ZSTD_decodingBufferSize_internal(unsigned long long windowSize, unsigned long long frameContentSize, size_t blockSizeMax)
+{
+    size_t const blockSize = MIN((size_t)MIN(windowSize, ZSTD_BLOCKSIZE_MAX), blockSizeMax);
+    /* We need blockSize + WILDCOPY_OVERLENGTH worth of buffer so that if a block
+     * ends at windowSize + WILDCOPY_OVERLENGTH + 1 bytes, we can start writing
+     * the block at the beginning of the output buffer, and maintain a full window.
+     *
+     * We need another blockSize worth of buffer so that we can store split
+     * literals at the end of the block without overwriting the extDict window.
+     */
+    unsigned long long const neededRBSize = windowSize + (blockSize * 2) + (WILDCOPY_OVERLENGTH * 2);
+    unsigned long long const neededSize = MIN(frameContentSize, neededRBSize);
+    size_t const minRBSize = (size_t) neededSize;
+    RETURN_ERROR_IF((unsigned long long)minRBSize != neededSize,
+                    frameParameter_windowTooLarge, "");
+    return minRBSize;
+}
+
+size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize)
+{
+    return ZSTD_decodingBufferSize_internal(windowSize, frameContentSize, ZSTD_BLOCKSIZE_MAX);
+}
+
+size_t ZSTD_estimateDStreamSize(size_t windowSize)
+{
+    size_t const blockSize = MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
+    size_t const inBuffSize = blockSize;  /* no block can be larger */
+    size_t const outBuffSize = ZSTD_decodingBufferSize_min(windowSize, ZSTD_CONTENTSIZE_UNKNOWN);
+    return ZSTD_estimateDCtxSize() + inBuffSize + outBuffSize;
+}
+
+size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize)
+{
+    U32 const windowSizeMax = 1U << ZSTD_WINDOWLOG_MAX;   /* note : should be user-selectable, but requires an additional parameter (or a dctx) */
+    ZSTD_FrameHeader zfh;
+    size_t const err = ZSTD_getFrameHeader(&zfh, src, srcSize);
+    if (ZSTD_isError(err)) return err;
+    RETURN_ERROR_IF(err>0, srcSize_wrong, "");
+    RETURN_ERROR_IF(zfh.windowSize > windowSizeMax,
+                    frameParameter_windowTooLarge, "");
+    return ZSTD_estimateDStreamSize((size_t)zfh.windowSize);
+}
+
+
+/* *****   Decompression   ***** */
+
+static int ZSTD_DCtx_isOverflow(ZSTD_DStream* zds, size_t const neededInBuffSize, size_t const neededOutBuffSize)
+{
+    return (zds->inBuffSize + zds->outBuffSize) >= (neededInBuffSize + neededOutBuffSize) * ZSTD_WORKSPACETOOLARGE_FACTOR;
+}
+
+static void ZSTD_DCtx_updateOversizedDuration(ZSTD_DStream* zds, size_t const neededInBuffSize, size_t const neededOutBuffSize)
+{
+    if (ZSTD_DCtx_isOverflow(zds, neededInBuffSize, neededOutBuffSize))
+        zds->oversizedDuration++;
+    else
+        zds->oversizedDuration = 0;
+}
+
+static int ZSTD_DCtx_isOversizedTooLong(ZSTD_DStream* zds)
+{
+    return zds->oversizedDuration >= ZSTD_WORKSPACETOOLARGE_MAXDURATION;
+}
+
+/* Checks that the output buffer hasn't changed if ZSTD_obm_stable is used. */
+static size_t ZSTD_checkOutBuffer(ZSTD_DStream const* zds, ZSTD_outBuffer const* output)
+{
+    ZSTD_outBuffer const expect = zds->expectedOutBuffer;
+    /* No requirement when ZSTD_obm_stable is not enabled. */
+    if (zds->outBufferMode != ZSTD_bm_stable)
+        return 0;
+    /* Any buffer is allowed in zdss_init, this must be the same for every other call until
+     * the context is reset.
+     */
+    if (zds->streamStage == zdss_init)
+        return 0;
+    /* The buffer must match our expectation exactly. */
+    if (expect.dst == output->dst && expect.pos == output->pos && expect.size == output->size)
+        return 0;
+    RETURN_ERROR(dstBuffer_wrong, "ZSTD_d_stableOutBuffer enabled but output differs!");
+}
+
+/* Calls ZSTD_decompressContinue() with the right parameters for ZSTD_decompressStream()
+ * and updates the stage and the output buffer state. This call is extracted so it can be
+ * used both when reading directly from the ZSTD_inBuffer, and in buffered input mode.
+ * NOTE: You must break after calling this function since the streamStage is modified.
+ */
+static size_t ZSTD_decompressContinueStream(
+            ZSTD_DStream* zds, char** op, char* oend,
+            void const* src, size_t srcSize) {
+    int const isSkipFrame = ZSTD_isSkipFrame(zds);
+    if (zds->outBufferMode == ZSTD_bm_buffered) {
+        size_t const dstSize = isSkipFrame ? 0 : zds->outBuffSize - zds->outStart;
+        size_t const decodedSize = ZSTD_decompressContinue(zds,
+                zds->outBuff + zds->outStart, dstSize, src, srcSize);
+        FORWARD_IF_ERROR(decodedSize, "");
+        if (!decodedSize && !isSkipFrame) {
+            zds->streamStage = zdss_read;
+        } else {
+            zds->outEnd = zds->outStart + decodedSize;
+            zds->streamStage = zdss_flush;
+        }
+    } else {
+        /* Write directly into the output buffer */
+        size_t const dstSize = isSkipFrame ? 0 : (size_t)(oend - *op);
+        size_t const decodedSize = ZSTD_decompressContinue(zds, *op, dstSize, src, srcSize);
+        FORWARD_IF_ERROR(decodedSize, "");
+        *op += decodedSize;
+        /* Flushing is not needed. */
+        zds->streamStage = zdss_read;
+        assert(*op <= oend);
+        assert(zds->outBufferMode == ZSTD_bm_stable);
+    }
+    return 0;
+}
+
+size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    const char* const src = (const char*)input->src;
+    const char* const istart = input->pos != 0 ? src + input->pos : src;
+    const char* const iend = input->size != 0 ? src + input->size : src;
+    const char* ip = istart;
+    char* const dst = (char*)output->dst;
+    char* const ostart = output->pos != 0 ? dst + output->pos : dst;
+    char* const oend = output->size != 0 ? dst + output->size : dst;
+    char* op = ostart;
+    U32 someMoreWork = 1;
+
+    DEBUGLOG(5, "ZSTD_decompressStream");
+    assert(zds != NULL);
+    RETURN_ERROR_IF(
+        input->pos > input->size,
+        srcSize_wrong,
+        "forbidden. in: pos: %u   vs size: %u",
+        (U32)input->pos, (U32)input->size);
+    RETURN_ERROR_IF(
+        output->pos > output->size,
+        dstSize_tooSmall,
+        "forbidden. out: pos: %u   vs size: %u",
+        (U32)output->pos, (U32)output->size);
+    DEBUGLOG(5, "input size : %u", (U32)(input->size - input->pos));
+    FORWARD_IF_ERROR(ZSTD_checkOutBuffer(zds, output), "");
+
+    while (someMoreWork) {
+        switch(zds->streamStage)
+        {
+        case zdss_init :
+            DEBUGLOG(5, "stage zdss_init => transparent reset ");
+            zds->streamStage = zdss_loadHeader;
+            zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+            zds->legacyVersion = 0;
+#endif
+            zds->hostageByte = 0;
+            zds->expectedOutBuffer = *output;
+            ZSTD_FALLTHROUGH;
+
+        case zdss_loadHeader :
+            DEBUGLOG(5, "stage zdss_loadHeader (srcSize : %u)", (U32)(iend - ip));
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+            if (zds->legacyVersion) {
+                RETURN_ERROR_IF(zds->staticSize, memory_allocation,
+                    "legacy support is incompatible with static dctx");
+                {   size_t const hint = ZSTD_decompressLegacyStream(zds->legacyContext, zds->legacyVersion, output, input);
+                    if (hint==0) zds->streamStage = zdss_init;
+                    return hint;
+            }   }
+#endif
+            {   size_t const hSize = ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format);
+                if (zds->refMultipleDDicts && zds->ddictSet) {
+                    ZSTD_DCtx_selectFrameDDict(zds);
+                }
+                if (ZSTD_isError(hSize)) {
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+                    U32 const legacyVersion = ZSTD_isLegacy(istart, iend-istart);
+                    if (legacyVersion) {
+                        ZSTD_DDict const* const ddict = ZSTD_getDDict(zds);
+                        const void* const dict = ddict ? ZSTD_DDict_dictContent(ddict) : NULL;
+                        size_t const dictSize = ddict ? ZSTD_DDict_dictSize(ddict) : 0;
+                        DEBUGLOG(5, "ZSTD_decompressStream: detected legacy version v0.%u", legacyVersion);
+                        RETURN_ERROR_IF(zds->staticSize, memory_allocation,
+                            "legacy support is incompatible with static dctx");
+                        FORWARD_IF_ERROR(ZSTD_initLegacyStream(&zds->legacyContext,
+                                    zds->previousLegacyVersion, legacyVersion,
+                                    dict, dictSize), "");
+                        zds->legacyVersion = zds->previousLegacyVersion = legacyVersion;
+                        {   size_t const hint = ZSTD_decompressLegacyStream(zds->legacyContext, legacyVersion, output, input);
+                            if (hint==0) zds->streamStage = zdss_init;   /* or stay in stage zdss_loadHeader */
+                            return hint;
+                    }   }
+#endif
+                    return hSize;   /* error */
+                }
+                if (hSize != 0) {   /* need more input */
+                    size_t const toLoad = hSize - zds->lhSize;   /* if hSize!=0, hSize > zds->lhSize */
+                    size_t const remainingInput = (size_t)(iend-ip);
+                    assert(iend >= ip);
+                    if (toLoad > remainingInput) {   /* not enough input to load full header */
+                        if (remainingInput > 0) {
+                            ZSTD_memcpy(zds->headerBuffer + zds->lhSize, ip, remainingInput);
+                            zds->lhSize += remainingInput;
+                        }
+                        input->pos = input->size;
+                        /* check first few bytes */
+                        FORWARD_IF_ERROR(
+                            ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format),
+                            "First few bytes detected incorrect" );
+                        /* return hint input size */
+                        return (MAX((size_t)ZSTD_FRAMEHEADERSIZE_MIN(zds->format), hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
+                    }
+                    assert(ip != NULL);
+                    ZSTD_memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;
+                    break;
+            }   }
+
+            /* check for single-pass mode opportunity */
+            if (zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
+                && zds->fParams.frameType != ZSTD_skippableFrame
+                && (U64)(size_t)(oend-op) >= zds->fParams.frameContentSize) {
+                size_t const cSize = ZSTD_findFrameCompressedSize_advanced(istart, (size_t)(iend-istart), zds->format);
+                if (cSize <= (size_t)(iend-istart)) {
+                    /* shortcut : using single-pass mode */
+                    size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, (size_t)(oend-op), istart, cSize, ZSTD_getDDict(zds));
+                    if (ZSTD_isError(decompressedSize)) return decompressedSize;
+                    DEBUGLOG(4, "shortcut to single-pass ZSTD_decompress_usingDDict()");
+                    assert(istart != NULL);
+                    ip = istart + cSize;
+                    op = op ? op + decompressedSize : op; /* can occur if frameContentSize = 0 (empty frame) */
+                    zds->expected = 0;
+                    zds->streamStage = zdss_init;
+                    someMoreWork = 0;
+                    break;
+            }   }
+
+            /* Check output buffer is large enough for ZSTD_odm_stable. */
+            if (zds->outBufferMode == ZSTD_bm_stable
+                && zds->fParams.frameType != ZSTD_skippableFrame
+                && zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
+                && (U64)(size_t)(oend-op) < zds->fParams.frameContentSize) {
+                RETURN_ERROR(dstSize_tooSmall, "ZSTD_obm_stable passed but ZSTD_outBuffer is too small");
+            }
+
+            /* Consume header (see ZSTDds_decodeFrameHeader) */
+            DEBUGLOG(4, "Consume header");
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDDict(zds, ZSTD_getDDict(zds)), "");
+
+            if (zds->format == ZSTD_f_zstd1
+                && (MEM_readLE32(zds->headerBuffer) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {  /* skippable frame */
+                zds->expected = MEM_readLE32(zds->headerBuffer + ZSTD_FRAMEIDSIZE);
+                zds->stage = ZSTDds_skipFrame;
+            } else {
+                FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(zds, zds->headerBuffer, zds->lhSize), "");
+                zds->expected = ZSTD_blockHeaderSize;
+                zds->stage = ZSTDds_decodeBlockHeader;
+            }
+
+            /* control buffer memory usage */
+            DEBUGLOG(4, "Control max memory usage (%u KB <= max %u KB)",
+                        (U32)(zds->fParams.windowSize >>10),
+                        (U32)(zds->maxWindowSize >> 10) );
+            zds->fParams.windowSize = MAX(zds->fParams.windowSize, 1U << ZSTD_WINDOWLOG_ABSOLUTEMIN);
+            RETURN_ERROR_IF(zds->fParams.windowSize > zds->maxWindowSize,
+                            frameParameter_windowTooLarge, "");
+            if (zds->maxBlockSizeParam != 0)
+                zds->fParams.blockSizeMax = MIN(zds->fParams.blockSizeMax, (unsigned)zds->maxBlockSizeParam);
+
+            /* Adapt buffer sizes to frame header instructions */
+            {   size_t const neededInBuffSize = MAX(zds->fParams.blockSizeMax, 4 /* frame checksum */);
+                size_t const neededOutBuffSize = zds->outBufferMode == ZSTD_bm_buffered
+                        ? ZSTD_decodingBufferSize_internal(zds->fParams.windowSize, zds->fParams.frameContentSize, zds->fParams.blockSizeMax)
+                        : 0;
+
+                ZSTD_DCtx_updateOversizedDuration(zds, neededInBuffSize, neededOutBuffSize);
+
+                {   int const tooSmall = (zds->inBuffSize < neededInBuffSize) || (zds->outBuffSize < neededOutBuffSize);
+                    int const tooLarge = ZSTD_DCtx_isOversizedTooLong(zds);
+
+                    if (tooSmall || tooLarge) {
+                        size_t const bufferSize = neededInBuffSize + neededOutBuffSize;
+                        DEBUGLOG(4, "inBuff  : from %u to %u",
+                                    (U32)zds->inBuffSize, (U32)neededInBuffSize);
+                        DEBUGLOG(4, "outBuff : from %u to %u",
+                                    (U32)zds->outBuffSize, (U32)neededOutBuffSize);
+                        if (zds->staticSize) {  /* static DCtx */
+                            DEBUGLOG(4, "staticSize : %u", (U32)zds->staticSize);
+                            assert(zds->staticSize >= sizeof(ZSTD_DCtx));  /* controlled at init */
+                            RETURN_ERROR_IF(
+                                bufferSize > zds->staticSize - sizeof(ZSTD_DCtx),
+                                memory_allocation, "");
+                        } else {
+                            ZSTD_customFree(zds->inBuff, zds->customMem);
+                            zds->inBuffSize = 0;
+                            zds->outBuffSize = 0;
+                            zds->inBuff = (char*)ZSTD_customMalloc(bufferSize, zds->customMem);
+                            RETURN_ERROR_IF(zds->inBuff == NULL, memory_allocation, "");
+                        }
+                        zds->inBuffSize = neededInBuffSize;
+                        zds->outBuff = zds->inBuff + zds->inBuffSize;
+                        zds->outBuffSize = neededOutBuffSize;
+            }   }   }
+            zds->streamStage = zdss_read;
+            ZSTD_FALLTHROUGH;
+
+        case zdss_read:
+            DEBUGLOG(5, "stage zdss_read");
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompressWithInputSize(zds, (size_t)(iend - ip));
+                DEBUGLOG(5, "neededInSize = %u", (U32)neededInSize);
+                if (neededInSize==0) {  /* end of frame */
+                    zds->streamStage = zdss_init;
+                    someMoreWork = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
+                    FORWARD_IF_ERROR(ZSTD_decompressContinueStream(zds, &op, oend, ip, neededInSize), "");
+                    assert(ip != NULL);
+                    ip += neededInSize;
+                    /* Function modifies the stage so we must break */
+                    break;
+            }   }
+            if (ip==iend) { someMoreWork = 0; break; }   /* no more input */
+            zds->streamStage = zdss_load;
+            ZSTD_FALLTHROUGH;
+
+        case zdss_load:
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds);
+                size_t const toLoad = neededInSize - zds->inPos;
+                int const isSkipFrame = ZSTD_isSkipFrame(zds);
+                size_t loadedSize;
+                /* At this point we shouldn't be decompressing a block that we can stream. */
+                assert(neededInSize == ZSTD_nextSrcSizeToDecompressWithInputSize(zds, (size_t)(iend - ip)));
+                if (isSkipFrame) {
+                    loadedSize = MIN(toLoad, (size_t)(iend-ip));
+                } else {
+                    RETURN_ERROR_IF(toLoad > zds->inBuffSize - zds->inPos,
+                                    corruption_detected,
+                                    "should never happen");
+                    loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, (size_t)(iend-ip));
+                }
+                if (loadedSize != 0) {
+                    /* ip may be NULL */
+                    ip += loadedSize;
+                    zds->inPos += loadedSize;
+                }
+                if (loadedSize < toLoad) { someMoreWork = 0; break; }   /* not enough input, wait for more */
+
+                /* decode loaded input */
+                zds->inPos = 0;   /* input is consumed */
+                FORWARD_IF_ERROR(ZSTD_decompressContinueStream(zds, &op, oend, zds->inBuff, neededInSize), "");
+                /* Function modifies the stage so we must break */
+                break;
+            }
+        case zdss_flush:
+            {
+                size_t const toFlushSize = zds->outEnd - zds->outStart;
+                size_t const flushedSize = ZSTD_limitCopy(op, (size_t)(oend-op), zds->outBuff + zds->outStart, toFlushSize);
+
+                op = op ? op + flushedSize : op;
+
+                zds->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {  /* flush completed */
+                    zds->streamStage = zdss_read;
+                    if ( (zds->outBuffSize < zds->fParams.frameContentSize)
+                        && (zds->outStart + zds->fParams.blockSizeMax > zds->outBuffSize) ) {
+                        DEBUGLOG(5, "restart filling outBuff from beginning (left:%i, needed:%u)",
+                                (int)(zds->outBuffSize - zds->outStart),
+                                (U32)zds->fParams.blockSizeMax);
+                        zds->outStart = zds->outEnd = 0;
+                    }
+                    break;
+            }   }
+            /* cannot complete flush */
+            someMoreWork = 0;
+            break;
+
+        default:
+            assert(0);    /* impossible */
+            RETURN_ERROR(GENERIC, "impossible to reach");   /* some compilers require default to do something */
+    }   }
+
+    /* result */
+    input->pos = (size_t)(ip - (const char*)(input->src));
+    output->pos = (size_t)(op - (char*)(output->dst));
+
+    /* Update the expected output buffer for ZSTD_obm_stable. */
+    zds->expectedOutBuffer = *output;
+
+    if ((ip==istart) && (op==ostart)) {  /* no forward progress */
+        zds->noForwardProgress ++;
+        if (zds->noForwardProgress >= ZSTD_NO_FORWARD_PROGRESS_MAX) {
+            RETURN_ERROR_IF(op==oend, noForwardProgress_destFull, "");
+            RETURN_ERROR_IF(ip==iend, noForwardProgress_inputEmpty, "");
+            assert(0);
+        }
+    } else {
+        zds->noForwardProgress = 0;
+    }
+    {   size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zds);
+        if (!nextSrcSizeHint) {   /* frame fully decoded */
+            if (zds->outEnd == zds->outStart) {  /* output fully flushed */
+                if (zds->hostageByte) {
+                    if (input->pos >= input->size) {
+                        /* can't release hostage (not present) */
+                        zds->streamStage = zdss_read;
+                        return 1;
+                    }
+                    input->pos++;  /* release hostage */
+                }   /* zds->hostageByte */
+                return 0;
+            }  /* zds->outEnd == zds->outStart */
+            if (!zds->hostageByte) { /* output not fully flushed; keep last byte as hostage; will be released when all output is flushed */
+                input->pos--;   /* note : pos > 0, otherwise, impossible to finish reading last block */
+                zds->hostageByte=1;
+            }
+            return 1;
+        }  /* nextSrcSizeHint==0 */
+        nextSrcSizeHint += ZSTD_blockHeaderSize * (ZSTD_nextInputType(zds) == ZSTDnit_block);   /* preload header of next block */
+        assert(zds->inPos <= nextSrcSizeHint);
+        nextSrcSizeHint -= zds->inPos;   /* part already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+size_t ZSTD_decompressStream_simpleArgs (
+                            ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos)
+{
+    ZSTD_outBuffer output;
+    ZSTD_inBuffer  input;
+    output.dst = dst;
+    output.size = dstCapacity;
+    output.pos = *dstPos;
+    input.src = src;
+    input.size = srcSize;
+    input.pos = *srcPos;
+    {   size_t const cErr = ZSTD_decompressStream(dctx, &output, &input);
+        *dstPos = output.pos;
+        *srcPos = input.pos;
+        return cErr;
+    }
+}
diff --git a/lib/decompress/zstd_decompress_block.c b/lib/decompress/zstd_decompress_block.c
new file mode 100644
index 0000000..6174a25
--- /dev/null
+++ b/lib/decompress/zstd_decompress_block.c
@@ -0,0 +1,2217 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* zstd_decompress_block :
+ * this module takes care of decompressing _compressed_ block */
+
+/*-*******************************************************
+*  Dependencies
+*********************************************************/
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
+#include "../common/compiler.h"    /* prefetch */
+#include "../common/mem.h"         /* low level memory routines */
+#include <stddef.h>
+#define FSE_STATIC_LINKING_ONLY
+#include "../common/fse.h"
+#include "../common/huf.h"
+#include "../common/zstd_internal.h"
+#include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
+#include "zstd_decompress_block.h"
+#include "../common/bits.h"  /* ZSTD_highbit32 */
+
+/*_*******************************************************
+*  Macros
+**********************************************************/
+
+/* These two optional macros force the use one way or another of the two
+ * ZSTD_decompressSequences implementations. You can't force in both directions
+ * at the same time.
+ */
+#if defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
+    defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
+#error "Cannot force the use of the short and the long ZSTD_decompressSequences variants!"
+#endif
+
+
+/*_*******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTD_copy4(void* dst, const void* src) { ZSTD_memcpy(dst, src, 4); }
+
+
+/*-*************************************************************
+ *   Block decoding
+ ***************************************************************/
+
+static size_t ZSTD_blockSizeMax(ZSTD_DCtx const* dctx)
+{
+    size_t const blockSizeMax = dctx->isFrameDecompression ? dctx->fParams.blockSizeMax : ZSTD_BLOCKSIZE_MAX;
+    assert(blockSizeMax <= ZSTD_BLOCKSIZE_MAX);
+    return blockSizeMax;
+}
+
+/*! ZSTD_getcBlockSize() :
+ *  Provides the size of compressed block from block header `src` */
+size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
+                          blockProperties_t* bpPtr)
+{
+    RETURN_ERROR_IF(srcSize < ZSTD_blockHeaderSize, srcSize_wrong, "");
+
+    {   U32 const cBlockHeader = MEM_readLE24(src);
+        U32 const cSize = cBlockHeader >> 3;
+        bpPtr->lastBlock = cBlockHeader & 1;
+        bpPtr->blockType = (blockType_e)((cBlockHeader >> 1) & 3);
+        bpPtr->origSize = cSize;   /* only useful for RLE */
+        if (bpPtr->blockType == bt_rle) return 1;
+        RETURN_ERROR_IF(bpPtr->blockType == bt_reserved, corruption_detected, "");
+        return cSize;
+    }
+}
+
+/* Allocate buffer for literals, either overlapping current dst, or split between dst and litExtraBuffer, or stored entirely within litExtraBuffer */
+static void ZSTD_allocateLiteralsBuffer(ZSTD_DCtx* dctx, void* const dst, const size_t dstCapacity, const size_t litSize,
+    const streaming_operation streaming, const size_t expectedWriteSize, const unsigned splitImmediately)
+{
+    size_t const blockSizeMax = ZSTD_blockSizeMax(dctx);
+    assert(litSize <= blockSizeMax);
+    assert(dctx->isFrameDecompression || streaming == not_streaming);
+    assert(expectedWriteSize <= blockSizeMax);
+    if (streaming == not_streaming && dstCapacity > blockSizeMax + WILDCOPY_OVERLENGTH + litSize + WILDCOPY_OVERLENGTH) {
+        /* If we aren't streaming, we can just put the literals after the output
+         * of the current block. We don't need to worry about overwriting the
+         * extDict of our window, because it doesn't exist.
+         * So if we have space after the end of the block, just put it there.
+         */
+        dctx->litBuffer = (BYTE*)dst + blockSizeMax + WILDCOPY_OVERLENGTH;
+        dctx->litBufferEnd = dctx->litBuffer + litSize;
+        dctx->litBufferLocation = ZSTD_in_dst;
+    } else if (litSize <= ZSTD_LITBUFFEREXTRASIZE) {
+        /* Literals fit entirely within the extra buffer, put them there to avoid
+         * having to split the literals.
+         */
+        dctx->litBuffer = dctx->litExtraBuffer;
+        dctx->litBufferEnd = dctx->litBuffer + litSize;
+        dctx->litBufferLocation = ZSTD_not_in_dst;
+    } else {
+        assert(blockSizeMax > ZSTD_LITBUFFEREXTRASIZE);
+        /* Literals must be split between the output block and the extra lit
+         * buffer. We fill the extra lit buffer with the tail of the literals,
+         * and put the rest of the literals at the end of the block, with
+         * WILDCOPY_OVERLENGTH of buffer room to allow for overreads.
+         * This MUST not write more than our maxBlockSize beyond dst, because in
+         * streaming mode, that could overwrite part of our extDict window.
+         */
+        if (splitImmediately) {
+            /* won't fit in litExtraBuffer, so it will be split between end of dst and extra buffer */
+            dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
+            dctx->litBufferEnd = dctx->litBuffer + litSize - ZSTD_LITBUFFEREXTRASIZE;
+        } else {
+            /* initially this will be stored entirely in dst during huffman decoding, it will partially be shifted to litExtraBuffer after */
+            dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize;
+            dctx->litBufferEnd = (BYTE*)dst + expectedWriteSize;
+        }
+        dctx->litBufferLocation = ZSTD_split;
+        assert(dctx->litBufferEnd <= (BYTE*)dst + expectedWriteSize);
+    }
+}
+
+/*! ZSTD_decodeLiteralsBlock() :
+ * Where it is possible to do so without being stomped by the output during decompression, the literals block will be stored
+ * in the dstBuffer.  If there is room to do so, it will be stored in full in the excess dst space after where the current
+ * block will be output.  Otherwise it will be stored at the end of the current dst blockspace, with a small portion being
+ * stored in dctx->litExtraBuffer to help keep it "ahead" of the current output write.
+ *
+ * @return : nb of bytes read from src (< srcSize )
+ *  note : symbol not declared but exposed for fullbench */
+static size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize,   /* note : srcSize < BLOCKSIZE */
+                          void* dst, size_t dstCapacity, const streaming_operation streaming)
+{
+    DEBUGLOG(5, "ZSTD_decodeLiteralsBlock");
+    RETURN_ERROR_IF(srcSize < MIN_CBLOCK_SIZE, corruption_detected, "");
+
+    {   const BYTE* const istart = (const BYTE*) src;
+        SymbolEncodingType_e const litEncType = (SymbolEncodingType_e)(istart[0] & 3);
+        size_t const blockSizeMax = ZSTD_blockSizeMax(dctx);
+
+        switch(litEncType)
+        {
+        case set_repeat:
+            DEBUGLOG(5, "set_repeat flag : re-using stats from previous compressed literals block");
+            RETURN_ERROR_IF(dctx->litEntropy==0, dictionary_corrupted, "");
+            ZSTD_FALLTHROUGH;
+
+        case set_compressed:
+            RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need up to 5 for case 3");
+            {   size_t lhSize, litSize, litCSize;
+                U32 singleStream=0;
+                U32 const lhlCode = (istart[0] >> 2) & 3;
+                U32 const lhc = MEM_readLE32(istart);
+                size_t hufSuccess;
+                size_t expectedWriteSize = MIN(blockSizeMax, dstCapacity);
+                int const flags = 0
+                    | (ZSTD_DCtx_get_bmi2(dctx) ? HUF_flags_bmi2 : 0)
+                    | (dctx->disableHufAsm ? HUF_flags_disableAsm : 0);
+                switch(lhlCode)
+                {
+                case 0: case 1: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    /* 2 - 2 - 10 - 10 */
+                    singleStream = !lhlCode;
+                    lhSize = 3;
+                    litSize  = (lhc >> 4) & 0x3FF;
+                    litCSize = (lhc >> 14) & 0x3FF;
+                    break;
+                case 2:
+                    /* 2 - 2 - 14 - 14 */
+                    lhSize = 4;
+                    litSize  = (lhc >> 4) & 0x3FFF;
+                    litCSize = lhc >> 18;
+                    break;
+                case 3:
+                    /* 2 - 2 - 18 - 18 */
+                    lhSize = 5;
+                    litSize  = (lhc >> 4) & 0x3FFFF;
+                    litCSize = (lhc >> 22) + ((size_t)istart[4] << 10);
+                    break;
+                }
+                RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
+                RETURN_ERROR_IF(litSize > blockSizeMax, corruption_detected, "");
+                if (!singleStream)
+                    RETURN_ERROR_IF(litSize < MIN_LITERALS_FOR_4_STREAMS, literals_headerWrong,
+                        "Not enough literals (%zu) for the 4-streams mode (min %u)",
+                        litSize, MIN_LITERALS_FOR_4_STREAMS);
+                RETURN_ERROR_IF(litCSize + lhSize > srcSize, corruption_detected, "");
+                RETURN_ERROR_IF(expectedWriteSize < litSize , dstSize_tooSmall, "");
+                ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 0);
+
+                /* prefetch huffman table if cold */
+                if (dctx->ddictIsCold && (litSize > 768 /* heuristic */)) {
+                    PREFETCH_AREA(dctx->HUFptr, sizeof(dctx->entropy.hufTable));
+                }
+
+                if (litEncType==set_repeat) {
+                    if (singleStream) {
+                        hufSuccess = HUF_decompress1X_usingDTable(
+                            dctx->litBuffer, litSize, istart+lhSize, litCSize,
+                            dctx->HUFptr, flags);
+                    } else {
+                        assert(litSize >= MIN_LITERALS_FOR_4_STREAMS);
+                        hufSuccess = HUF_decompress4X_usingDTable(
+                            dctx->litBuffer, litSize, istart+lhSize, litCSize,
+                            dctx->HUFptr, flags);
+                    }
+                } else {
+                    if (singleStream) {
+#if defined(HUF_FORCE_DECOMPRESS_X2)
+                        hufSuccess = HUF_decompress1X_DCtx_wksp(
+                            dctx->entropy.hufTable, dctx->litBuffer, litSize,
+                            istart+lhSize, litCSize, dctx->workspace,
+                            sizeof(dctx->workspace), flags);
+#else
+                        hufSuccess = HUF_decompress1X1_DCtx_wksp(
+                            dctx->entropy.hufTable, dctx->litBuffer, litSize,
+                            istart+lhSize, litCSize, dctx->workspace,
+                            sizeof(dctx->workspace), flags);
+#endif
+                    } else {
+                        hufSuccess = HUF_decompress4X_hufOnly_wksp(
+                            dctx->entropy.hufTable, dctx->litBuffer, litSize,
+                            istart+lhSize, litCSize, dctx->workspace,
+                            sizeof(dctx->workspace), flags);
+                    }
+                }
+                if (dctx->litBufferLocation == ZSTD_split)
+                {
+                    assert(litSize > ZSTD_LITBUFFEREXTRASIZE);
+                    ZSTD_memcpy(dctx->litExtraBuffer, dctx->litBufferEnd - ZSTD_LITBUFFEREXTRASIZE, ZSTD_LITBUFFEREXTRASIZE);
+                    ZSTD_memmove(dctx->litBuffer + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH, dctx->litBuffer, litSize - ZSTD_LITBUFFEREXTRASIZE);
+                    dctx->litBuffer += ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
+                    dctx->litBufferEnd -= WILDCOPY_OVERLENGTH;
+                    assert(dctx->litBufferEnd <= (BYTE*)dst + blockSizeMax);
+                }
+
+                RETURN_ERROR_IF(HUF_isError(hufSuccess), corruption_detected, "");
+
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                dctx->litEntropy = 1;
+                if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
+                return litCSize + lhSize;
+            }
+
+        case set_basic:
+            {   size_t litSize, lhSize;
+                U32 const lhlCode = ((istart[0]) >> 2) & 3;
+                size_t expectedWriteSize = MIN(blockSizeMax, dstCapacity);
+                switch(lhlCode)
+                {
+                case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    lhSize = 1;
+                    litSize = istart[0] >> 3;
+                    break;
+                case 1:
+                    lhSize = 2;
+                    litSize = MEM_readLE16(istart) >> 4;
+                    break;
+                case 3:
+                    lhSize = 3;
+                    RETURN_ERROR_IF(srcSize<3, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize = 3");
+                    litSize = MEM_readLE24(istart) >> 4;
+                    break;
+                }
+
+                RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
+                RETURN_ERROR_IF(litSize > blockSizeMax, corruption_detected, "");
+                RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
+                ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
+                if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                    RETURN_ERROR_IF(litSize+lhSize > srcSize, corruption_detected, "");
+                    if (dctx->litBufferLocation == ZSTD_split)
+                    {
+                        ZSTD_memcpy(dctx->litBuffer, istart + lhSize, litSize - ZSTD_LITBUFFEREXTRASIZE);
+                        ZSTD_memcpy(dctx->litExtraBuffer, istart + lhSize + litSize - ZSTD_LITBUFFEREXTRASIZE, ZSTD_LITBUFFEREXTRASIZE);
+                    }
+                    else
+                    {
+                        ZSTD_memcpy(dctx->litBuffer, istart + lhSize, litSize);
+                    }
+                    dctx->litPtr = dctx->litBuffer;
+                    dctx->litSize = litSize;
+                    return lhSize+litSize;
+                }
+                /* direct reference into compressed stream */
+                dctx->litPtr = istart+lhSize;
+                dctx->litSize = litSize;
+                dctx->litBufferEnd = dctx->litPtr + litSize;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
+                return lhSize+litSize;
+            }
+
+        case set_rle:
+            {   U32 const lhlCode = ((istart[0]) >> 2) & 3;
+                size_t litSize, lhSize;
+                size_t expectedWriteSize = MIN(blockSizeMax, dstCapacity);
+                switch(lhlCode)
+                {
+                case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    lhSize = 1;
+                    litSize = istart[0] >> 3;
+                    break;
+                case 1:
+                    lhSize = 2;
+                    RETURN_ERROR_IF(srcSize<3, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize+1 = 3");
+                    litSize = MEM_readLE16(istart) >> 4;
+                    break;
+                case 3:
+                    lhSize = 3;
+                    RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize+1 = 4");
+                    litSize = MEM_readLE24(istart) >> 4;
+                    break;
+                }
+                RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
+                RETURN_ERROR_IF(litSize > blockSizeMax, corruption_detected, "");
+                RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
+                ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
+                if (dctx->litBufferLocation == ZSTD_split)
+                {
+                    ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize - ZSTD_LITBUFFEREXTRASIZE);
+                    ZSTD_memset(dctx->litExtraBuffer, istart[lhSize], ZSTD_LITBUFFEREXTRASIZE);
+                }
+                else
+                {
+                    ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize);
+                }
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                return lhSize+1;
+            }
+        default:
+            RETURN_ERROR(corruption_detected, "impossible");
+        }
+    }
+}
+
+/* Hidden declaration for fullbench */
+size_t ZSTD_decodeLiteralsBlock_wrapper(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize,
+                          void* dst, size_t dstCapacity);
+size_t ZSTD_decodeLiteralsBlock_wrapper(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize,
+                          void* dst, size_t dstCapacity)
+{
+    dctx->isFrameDecompression = 0;
+    return ZSTD_decodeLiteralsBlock(dctx, src, srcSize, dst, dstCapacity, not_streaming);
+}
+
+/* Default FSE distribution tables.
+ * These are pre-calculated FSE decoding tables using default distributions as defined in specification :
+ * https://github.com/facebook/zstd/blob/release/doc/zstd_compression_format.md#default-distributions
+ * They were generated programmatically with following method :
+ * - start from default distributions, present in /lib/common/zstd_internal.h
+ * - generate tables normally, using ZSTD_buildFSETable()
+ * - printout the content of tables
+ * - prettify output, report below, test with fuzzer to ensure it's correct */
+
+/* Default FSE distribution table for Literal Lengths */
+static const ZSTD_seqSymbol LL_defaultDTable[(1<<LL_DEFAULTNORMLOG)+1] = {
+     {  1,  1,  1, LL_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+     /* nextState, nbAddBits, nbBits, baseVal */
+     {  0,  0,  4,    0},  { 16,  0,  4,    0},
+     { 32,  0,  5,    1},  {  0,  0,  5,    3},
+     {  0,  0,  5,    4},  {  0,  0,  5,    6},
+     {  0,  0,  5,    7},  {  0,  0,  5,    9},
+     {  0,  0,  5,   10},  {  0,  0,  5,   12},
+     {  0,  0,  6,   14},  {  0,  1,  5,   16},
+     {  0,  1,  5,   20},  {  0,  1,  5,   22},
+     {  0,  2,  5,   28},  {  0,  3,  5,   32},
+     {  0,  4,  5,   48},  { 32,  6,  5,   64},
+     {  0,  7,  5,  128},  {  0,  8,  6,  256},
+     {  0, 10,  6, 1024},  {  0, 12,  6, 4096},
+     { 32,  0,  4,    0},  {  0,  0,  4,    1},
+     {  0,  0,  5,    2},  { 32,  0,  5,    4},
+     {  0,  0,  5,    5},  { 32,  0,  5,    7},
+     {  0,  0,  5,    8},  { 32,  0,  5,   10},
+     {  0,  0,  5,   11},  {  0,  0,  6,   13},
+     { 32,  1,  5,   16},  {  0,  1,  5,   18},
+     { 32,  1,  5,   22},  {  0,  2,  5,   24},
+     { 32,  3,  5,   32},  {  0,  3,  5,   40},
+     {  0,  6,  4,   64},  { 16,  6,  4,   64},
+     { 32,  7,  5,  128},  {  0,  9,  6,  512},
+     {  0, 11,  6, 2048},  { 48,  0,  4,    0},
+     { 16,  0,  4,    1},  { 32,  0,  5,    2},
+     { 32,  0,  5,    3},  { 32,  0,  5,    5},
+     { 32,  0,  5,    6},  { 32,  0,  5,    8},
+     { 32,  0,  5,    9},  { 32,  0,  5,   11},
+     { 32,  0,  5,   12},  {  0,  0,  6,   15},
+     { 32,  1,  5,   18},  { 32,  1,  5,   20},
+     { 32,  2,  5,   24},  { 32,  2,  5,   28},
+     { 32,  3,  5,   40},  { 32,  4,  5,   48},
+     {  0, 16,  6,65536},  {  0, 15,  6,32768},
+     {  0, 14,  6,16384},  {  0, 13,  6, 8192},
+};   /* LL_defaultDTable */
+
+/* Default FSE distribution table for Offset Codes */
+static const ZSTD_seqSymbol OF_defaultDTable[(1<<OF_DEFAULTNORMLOG)+1] = {
+    {  1,  1,  1, OF_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+    /* nextState, nbAddBits, nbBits, baseVal */
+    {  0,  0,  5,    0},     {  0,  6,  4,   61},
+    {  0,  9,  5,  509},     {  0, 15,  5,32765},
+    {  0, 21,  5,2097149},   {  0,  3,  5,    5},
+    {  0,  7,  4,  125},     {  0, 12,  5, 4093},
+    {  0, 18,  5,262141},    {  0, 23,  5,8388605},
+    {  0,  5,  5,   29},     {  0,  8,  4,  253},
+    {  0, 14,  5,16381},     {  0, 20,  5,1048573},
+    {  0,  2,  5,    1},     { 16,  7,  4,  125},
+    {  0, 11,  5, 2045},     {  0, 17,  5,131069},
+    {  0, 22,  5,4194301},   {  0,  4,  5,   13},
+    { 16,  8,  4,  253},     {  0, 13,  5, 8189},
+    {  0, 19,  5,524285},    {  0,  1,  5,    1},
+    { 16,  6,  4,   61},     {  0, 10,  5, 1021},
+    {  0, 16,  5,65533},     {  0, 28,  5,268435453},
+    {  0, 27,  5,134217725}, {  0, 26,  5,67108861},
+    {  0, 25,  5,33554429},  {  0, 24,  5,16777213},
+};   /* OF_defaultDTable */
+
+
+/* Default FSE distribution table for Match Lengths */
+static const ZSTD_seqSymbol ML_defaultDTable[(1<<ML_DEFAULTNORMLOG)+1] = {
+    {  1,  1,  1, ML_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+    /* nextState, nbAddBits, nbBits, baseVal */
+    {  0,  0,  6,    3},  {  0,  0,  4,    4},
+    { 32,  0,  5,    5},  {  0,  0,  5,    6},
+    {  0,  0,  5,    8},  {  0,  0,  5,    9},
+    {  0,  0,  5,   11},  {  0,  0,  6,   13},
+    {  0,  0,  6,   16},  {  0,  0,  6,   19},
+    {  0,  0,  6,   22},  {  0,  0,  6,   25},
+    {  0,  0,  6,   28},  {  0,  0,  6,   31},
+    {  0,  0,  6,   34},  {  0,  1,  6,   37},
+    {  0,  1,  6,   41},  {  0,  2,  6,   47},
+    {  0,  3,  6,   59},  {  0,  4,  6,   83},
+    {  0,  7,  6,  131},  {  0,  9,  6,  515},
+    { 16,  0,  4,    4},  {  0,  0,  4,    5},
+    { 32,  0,  5,    6},  {  0,  0,  5,    7},
+    { 32,  0,  5,    9},  {  0,  0,  5,   10},
+    {  0,  0,  6,   12},  {  0,  0,  6,   15},
+    {  0,  0,  6,   18},  {  0,  0,  6,   21},
+    {  0,  0,  6,   24},  {  0,  0,  6,   27},
+    {  0,  0,  6,   30},  {  0,  0,  6,   33},
+    {  0,  1,  6,   35},  {  0,  1,  6,   39},
+    {  0,  2,  6,   43},  {  0,  3,  6,   51},
+    {  0,  4,  6,   67},  {  0,  5,  6,   99},
+    {  0,  8,  6,  259},  { 32,  0,  4,    4},
+    { 48,  0,  4,    4},  { 16,  0,  4,    5},
+    { 32,  0,  5,    7},  { 32,  0,  5,    8},
+    { 32,  0,  5,   10},  { 32,  0,  5,   11},
+    {  0,  0,  6,   14},  {  0,  0,  6,   17},
+    {  0,  0,  6,   20},  {  0,  0,  6,   23},
+    {  0,  0,  6,   26},  {  0,  0,  6,   29},
+    {  0,  0,  6,   32},  {  0, 16,  6,65539},
+    {  0, 15,  6,32771},  {  0, 14,  6,16387},
+    {  0, 13,  6, 8195},  {  0, 12,  6, 4099},
+    {  0, 11,  6, 2051},  {  0, 10,  6, 1027},
+};   /* ML_defaultDTable */
+
+
+static void ZSTD_buildSeqTable_rle(ZSTD_seqSymbol* dt, U32 baseValue, U8 nbAddBits)
+{
+    void* ptr = dt;
+    ZSTD_seqSymbol_header* const DTableH = (ZSTD_seqSymbol_header*)ptr;
+    ZSTD_seqSymbol* const cell = dt + 1;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->nbBits = 0;
+    cell->nextState = 0;
+    assert(nbAddBits < 255);
+    cell->nbAdditionalBits = nbAddBits;
+    cell->baseValue = baseValue;
+}
+
+
+/* ZSTD_buildFSETable() :
+ * generate FSE decoding table for one symbol (ll, ml or off)
+ * cannot fail if input is valid =>
+ * all inputs are presumed validated at this stage */
+FORCE_INLINE_TEMPLATE
+void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U8* nbAdditionalBits,
+            unsigned tableLog, void* wksp, size_t wkspSize)
+{
+    ZSTD_seqSymbol* const tableDecode = dt+1;
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+
+    U16* symbolNext = (U16*)wksp;
+    BYTE* spread = (BYTE*)(symbolNext + MaxSeq + 1);
+    U32 highThreshold = tableSize - 1;
+
+
+    /* Sanity Checks */
+    assert(maxSymbolValue <= MaxSeq);
+    assert(tableLog <= MaxFSELog);
+    assert(wkspSize >= ZSTD_BUILD_FSE_TABLE_WKSP_SIZE);
+    (void)wkspSize;
+    /* Init, lay down lowprob symbols */
+    {   ZSTD_seqSymbol_header DTableH;
+        DTableH.tableLog = tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].baseValue = s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    assert(normalizedCounter[s]>=0);
+                    symbolNext[s] = (U16)normalizedCounter[s];
+        }   }   }
+        ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    assert(tableSize <= 512);
+    /* Specialized symbol spreading for the case when there are
+     * no low probability (-1 count) symbols. When compressing
+     * small blocks we avoid low probability symbols to hit this
+     * case, since header decoding speed matters more.
+     */
+    if (highThreshold == tableSize - 1) {
+        size_t const tableMask = tableSize-1;
+        size_t const step = FSE_TABLESTEP(tableSize);
+        /* First lay down the symbols in order.
+         * We use a uint64_t to lay down 8 bytes at a time. This reduces branch
+         * misses since small blocks generally have small table logs, so nearly
+         * all symbols have counts <= 8. We ensure we have 8 bytes at the end of
+         * our buffer to handle the over-write.
+         */
+        {
+            U64 const add = 0x0101010101010101ull;
+            size_t pos = 0;
+            U64 sv = 0;
+            U32 s;
+            for (s=0; s<maxSV1; ++s, sv += add) {
+                int i;
+                int const n = normalizedCounter[s];
+                MEM_write64(spread + pos, sv);
+                for (i = 8; i < n; i += 8) {
+                    MEM_write64(spread + pos + i, sv);
+                }
+                assert(n>=0);
+                pos += (size_t)n;
+            }
+        }
+        /* Now we spread those positions across the table.
+         * The benefit of doing it in two stages is that we avoid the
+         * variable size inner loop, which caused lots of branch misses.
+         * Now we can run through all the positions without any branch misses.
+         * We unroll the loop twice, since that is what empirically worked best.
+         */
+        {
+            size_t position = 0;
+            size_t s;
+            size_t const unroll = 2;
+            assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
+            for (s = 0; s < (size_t)tableSize; s += unroll) {
+                size_t u;
+                for (u = 0; u < unroll; ++u) {
+                    size_t const uPosition = (position + (u * step)) & tableMask;
+                    tableDecode[uPosition].baseValue = spread[s + u];
+                }
+                position = (position + (unroll * step)) & tableMask;
+            }
+            assert(position == 0);
+        }
+    } else {
+        U32 const tableMask = tableSize-1;
+        U32 const step = FSE_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            int const n = normalizedCounter[s];
+            for (i=0; i<n; i++) {
+                tableDecode[position].baseValue = s;
+                position = (position + step) & tableMask;
+                while (UNLIKELY(position > highThreshold)) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+        assert(position == 0); /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {
+        U32 u;
+        for (u=0; u<tableSize; u++) {
+            U32 const symbol = tableDecode[u].baseValue;
+            U32 const nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - ZSTD_highbit32(nextState) );
+            tableDecode[u].nextState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+            assert(nbAdditionalBits[symbol] < 255);
+            tableDecode[u].nbAdditionalBits = nbAdditionalBits[symbol];
+            tableDecode[u].baseValue = baseValue[symbol];
+        }
+    }
+}
+
+/* Avoids the FORCE_INLINE of the _body() function. */
+static void ZSTD_buildFSETable_body_default(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U8* nbAdditionalBits,
+            unsigned tableLog, void* wksp, size_t wkspSize)
+{
+    ZSTD_buildFSETable_body(dt, normalizedCounter, maxSymbolValue,
+            baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
+}
+
+#if DYNAMIC_BMI2
+BMI2_TARGET_ATTRIBUTE static void ZSTD_buildFSETable_body_bmi2(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U8* nbAdditionalBits,
+            unsigned tableLog, void* wksp, size_t wkspSize)
+{
+    ZSTD_buildFSETable_body(dt, normalizedCounter, maxSymbolValue,
+            baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
+}
+#endif
+
+void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U8* nbAdditionalBits,
+            unsigned tableLog, void* wksp, size_t wkspSize, int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        ZSTD_buildFSETable_body_bmi2(dt, normalizedCounter, maxSymbolValue,
+                baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
+        return;
+    }
+#endif
+    (void)bmi2;
+    ZSTD_buildFSETable_body_default(dt, normalizedCounter, maxSymbolValue,
+            baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
+}
+
+
+/*! ZSTD_buildSeqTable() :
+ * @return : nb bytes read from src,
+ *           or an error code if it fails */
+static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymbol** DTablePtr,
+                                 SymbolEncodingType_e type, unsigned max, U32 maxLog,
+                                 const void* src, size_t srcSize,
+                                 const U32* baseValue, const U8* nbAdditionalBits,
+                                 const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable,
+                                 int ddictIsCold, int nbSeq, U32* wksp, size_t wkspSize,
+                                 int bmi2)
+{
+    switch(type)
+    {
+    case set_rle :
+        RETURN_ERROR_IF(!srcSize, srcSize_wrong, "");
+        RETURN_ERROR_IF((*(const BYTE*)src) > max, corruption_detected, "");
+        {   U32 const symbol = *(const BYTE*)src;
+            U32 const baseline = baseValue[symbol];
+            U8 const nbBits = nbAdditionalBits[symbol];
+            ZSTD_buildSeqTable_rle(DTableSpace, baseline, nbBits);
+        }
+        *DTablePtr = DTableSpace;
+        return 1;
+    case set_basic :
+        *DTablePtr = defaultTable;
+        return 0;
+    case set_repeat:
+        RETURN_ERROR_IF(!flagRepeatTable, corruption_detected, "");
+        /* prefetch FSE table if used */
+        if (ddictIsCold && (nbSeq > 24 /* heuristic */)) {
+            const void* const pStart = *DTablePtr;
+            size_t const pSize = sizeof(ZSTD_seqSymbol) * (SEQSYMBOL_TABLE_SIZE(maxLog));
+            PREFETCH_AREA(pStart, pSize);
+        }
+        return 0;
+    case set_compressed :
+        {   unsigned tableLog;
+            S16 norm[MaxSeq+1];
+            size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize);
+            RETURN_ERROR_IF(FSE_isError(headerSize), corruption_detected, "");
+            RETURN_ERROR_IF(tableLog > maxLog, corruption_detected, "");
+            ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog, wksp, wkspSize, bmi2);
+            *DTablePtr = DTableSpace;
+            return headerSize;
+        }
+    default :
+        assert(0);
+        RETURN_ERROR(GENERIC, "impossible");
+    }
+}
+
+size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
+                             const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip = istart;
+    int nbSeq;
+    DEBUGLOG(5, "ZSTD_decodeSeqHeaders");
+
+    /* check */
+    RETURN_ERROR_IF(srcSize < MIN_SEQUENCES_SIZE, srcSize_wrong, "");
+
+    /* SeqHead */
+    nbSeq = *ip++;
+    if (nbSeq > 0x7F) {
+        if (nbSeq == 0xFF) {
+            RETURN_ERROR_IF(ip+2 > iend, srcSize_wrong, "");
+            nbSeq = MEM_readLE16(ip) + LONGNBSEQ;
+            ip+=2;
+        } else {
+            RETURN_ERROR_IF(ip >= iend, srcSize_wrong, "");
+            nbSeq = ((nbSeq-0x80)<<8) + *ip++;
+        }
+    }
+    *nbSeqPtr = nbSeq;
+
+    if (nbSeq == 0) {
+        /* No sequence : section ends immediately */
+        RETURN_ERROR_IF(ip != iend, corruption_detected,
+            "extraneous data present in the Sequences section");
+        return (size_t)(ip - istart);
+    }
+
+    /* FSE table descriptors */
+    RETURN_ERROR_IF(ip+1 > iend, srcSize_wrong, ""); /* minimum possible size: 1 byte for symbol encoding types */
+    RETURN_ERROR_IF(*ip & 3, corruption_detected, ""); /* The last field, Reserved, must be all-zeroes. */
+    {   SymbolEncodingType_e const LLtype = (SymbolEncodingType_e)(*ip >> 6);
+        SymbolEncodingType_e const OFtype = (SymbolEncodingType_e)((*ip >> 4) & 3);
+        SymbolEncodingType_e const MLtype = (SymbolEncodingType_e)((*ip >> 2) & 3);
+        ip++;
+
+        /* Build DTables */
+        assert(ip <= iend);
+        {   size_t const llhSize = ZSTD_buildSeqTable(dctx->entropy.LLTable, &dctx->LLTptr,
+                                                      LLtype, MaxLL, LLFSELog,
+                                                      ip, (size_t)(iend-ip),
+                                                      LL_base, LL_bits,
+                                                      LL_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq,
+                                                      dctx->workspace, sizeof(dctx->workspace),
+                                                      ZSTD_DCtx_get_bmi2(dctx));
+            RETURN_ERROR_IF(ZSTD_isError(llhSize), corruption_detected, "ZSTD_buildSeqTable failed");
+            ip += llhSize;
+        }
+
+        assert(ip <= iend);
+        {   size_t const ofhSize = ZSTD_buildSeqTable(dctx->entropy.OFTable, &dctx->OFTptr,
+                                                      OFtype, MaxOff, OffFSELog,
+                                                      ip, (size_t)(iend-ip),
+                                                      OF_base, OF_bits,
+                                                      OF_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq,
+                                                      dctx->workspace, sizeof(dctx->workspace),
+                                                      ZSTD_DCtx_get_bmi2(dctx));
+            RETURN_ERROR_IF(ZSTD_isError(ofhSize), corruption_detected, "ZSTD_buildSeqTable failed");
+            ip += ofhSize;
+        }
+
+        assert(ip <= iend);
+        {   size_t const mlhSize = ZSTD_buildSeqTable(dctx->entropy.MLTable, &dctx->MLTptr,
+                                                      MLtype, MaxML, MLFSELog,
+                                                      ip, (size_t)(iend-ip),
+                                                      ML_base, ML_bits,
+                                                      ML_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq,
+                                                      dctx->workspace, sizeof(dctx->workspace),
+                                                      ZSTD_DCtx_get_bmi2(dctx));
+            RETURN_ERROR_IF(ZSTD_isError(mlhSize), corruption_detected, "ZSTD_buildSeqTable failed");
+            ip += mlhSize;
+        }
+    }
+
+    return (size_t)(ip-istart);
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+} seq_t;
+
+typedef struct {
+    size_t state;
+    const ZSTD_seqSymbol* table;
+} ZSTD_fseState;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    ZSTD_fseState stateLL;
+    ZSTD_fseState stateOffb;
+    ZSTD_fseState stateML;
+    size_t prevOffset[ZSTD_REP_NUM];
+} seqState_t;
+
+/*! ZSTD_overlapCopy8() :
+ *  Copies 8 bytes from ip to op and updates op and ip where ip <= op.
+ *  If the offset is < 8 then the offset is spread to at least 8 bytes.
+ *
+ *  Precondition: *ip <= *op
+ *  Postcondition: *op - *op >= 8
+ */
+HINT_INLINE void ZSTD_overlapCopy8(BYTE** op, BYTE const** ip, size_t offset)
+{
+    assert(*ip <= *op);
+    if (offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[offset];
+        (*op)[0] = (*ip)[0];
+        (*op)[1] = (*ip)[1];
+        (*op)[2] = (*ip)[2];
+        (*op)[3] = (*ip)[3];
+        *ip += dec32table[offset];
+        ZSTD_copy4(*op+4, *ip);
+        *ip -= sub2;
+    } else {
+        ZSTD_copy8(*op, *ip);
+    }
+    *ip += 8;
+    *op += 8;
+    assert(*op - *ip >= 8);
+}
+
+/*! ZSTD_safecopy() :
+ *  Specialized version of memcpy() that is allowed to READ up to WILDCOPY_OVERLENGTH past the input buffer
+ *  and write up to 16 bytes past oend_w (op >= oend_w is allowed).
+ *  This function is only called in the uncommon case where the sequence is near the end of the block. It
+ *  should be fast for a single long sequence, but can be slow for several short sequences.
+ *
+ *  @param ovtype controls the overlap detection
+ *         - ZSTD_no_overlap: The source and destination are guaranteed to be at least WILDCOPY_VECLEN bytes apart.
+ *         - ZSTD_overlap_src_before_dst: The src and dst may overlap and may be any distance apart.
+ *           The src buffer must be before the dst buffer.
+ */
+static void
+ZSTD_safecopy(BYTE* op, const BYTE* const oend_w, BYTE const* ip, size_t length, ZSTD_overlap_e ovtype)
+{
+    ptrdiff_t const diff = op - ip;
+    BYTE* const oend = op + length;
+
+    assert((ovtype == ZSTD_no_overlap && (diff <= -8 || diff >= 8 || op >= oend_w)) ||
+           (ovtype == ZSTD_overlap_src_before_dst && diff >= 0));
+
+    if (length < 8) {
+        /* Handle short lengths. */
+        while (op < oend) *op++ = *ip++;
+        return;
+    }
+    if (ovtype == ZSTD_overlap_src_before_dst) {
+        /* Copy 8 bytes and ensure the offset >= 8 when there can be overlap. */
+        assert(length >= 8);
+        assert(diff > 0);
+        ZSTD_overlapCopy8(&op, &ip, (size_t)diff);
+        length -= 8;
+        assert(op - ip >= 8);
+        assert(op <= oend);
+    }
+
+    if (oend <= oend_w) {
+        /* No risk of overwrite. */
+        ZSTD_wildcopy(op, ip, length, ovtype);
+        return;
+    }
+    if (op <= oend_w) {
+        /* Wildcopy until we get close to the end. */
+        assert(oend > oend_w);
+        ZSTD_wildcopy(op, ip, (size_t)(oend_w - op), ovtype);
+        ip += oend_w - op;
+        op += oend_w - op;
+    }
+    /* Handle the leftovers. */
+    while (op < oend) *op++ = *ip++;
+}
+
+/* ZSTD_safecopyDstBeforeSrc():
+ * This version allows overlap with dst before src, or handles the non-overlap case with dst after src
+ * Kept separate from more common ZSTD_safecopy case to avoid performance impact to the safecopy common case */
+static void ZSTD_safecopyDstBeforeSrc(BYTE* op, const BYTE* ip, size_t length)
+{
+    ptrdiff_t const diff = op - ip;
+    BYTE* const oend = op + length;
+
+    if (length < 8 || diff > -8) {
+        /* Handle short lengths, close overlaps, and dst not before src. */
+        while (op < oend) *op++ = *ip++;
+        return;
+    }
+
+    if (op <= oend - WILDCOPY_OVERLENGTH && diff < -WILDCOPY_VECLEN) {
+        ZSTD_wildcopy(op, ip, (size_t)(oend - WILDCOPY_OVERLENGTH - op), ZSTD_no_overlap);
+        ip += oend - WILDCOPY_OVERLENGTH - op;
+        op += oend - WILDCOPY_OVERLENGTH - op;
+    }
+
+    /* Handle the leftovers. */
+    while (op < oend) *op++ = *ip++;
+}
+
+/* ZSTD_execSequenceEnd():
+ * This version handles cases that are near the end of the output buffer. It requires
+ * more careful checks to make sure there is no overflow. By separating out these hard
+ * and unlikely cases, we can speed up the common cases.
+ *
+ * NOTE: This function needs to be fast for a single long sequence, but doesn't need
+ * to be optimized for many small sequences, since those fall into ZSTD_execSequence().
+ */
+FORCE_NOINLINE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_execSequenceEnd(BYTE* op,
+    BYTE* const oend, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
+
+    /* bounds checks : careful of address space overflow in 32-bit mode */
+    RETURN_ERROR_IF(sequenceLength > (size_t)(oend - op), dstSize_tooSmall, "last match must fit within dstBuffer");
+    RETURN_ERROR_IF(sequence.litLength > (size_t)(litLimit - *litPtr), corruption_detected, "try to read beyond literal buffer");
+    assert(op < op + sequenceLength);
+    assert(oLitEnd < op + sequenceLength);
+
+    /* copy literals */
+    ZSTD_safecopy(op, oend_w, *litPtr, sequence.litLength, ZSTD_no_overlap);
+    op = oLitEnd;
+    *litPtr = iLitEnd;
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix */
+        RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
+        match = dictEnd - (prefixStart - match);
+        if (match + sequence.matchLength <= dictEnd) {
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = (size_t)(dictEnd - match);
+            ZSTD_memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = prefixStart;
+        }
+    }
+    ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
+    return sequenceLength;
+}
+
+/* ZSTD_execSequenceEndSplitLitBuffer():
+ * This version is intended to be used during instances where the litBuffer is still split.  It is kept separate to avoid performance impact for the good case.
+ */
+FORCE_NOINLINE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_execSequenceEndSplitLitBuffer(BYTE* op,
+    BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+
+    /* bounds checks : careful of address space overflow in 32-bit mode */
+    RETURN_ERROR_IF(sequenceLength > (size_t)(oend - op), dstSize_tooSmall, "last match must fit within dstBuffer");
+    RETURN_ERROR_IF(sequence.litLength > (size_t)(litLimit - *litPtr), corruption_detected, "try to read beyond literal buffer");
+    assert(op < op + sequenceLength);
+    assert(oLitEnd < op + sequenceLength);
+
+    /* copy literals */
+    RETURN_ERROR_IF(op > *litPtr && op < *litPtr + sequence.litLength, dstSize_tooSmall, "output should not catch up to and overwrite literal buffer");
+    ZSTD_safecopyDstBeforeSrc(op, *litPtr, sequence.litLength);
+    op = oLitEnd;
+    *litPtr = iLitEnd;
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix */
+        RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
+        match = dictEnd - (prefixStart - match);
+        if (match + sequence.matchLength <= dictEnd) {
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = (size_t)(dictEnd - match);
+            ZSTD_memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = prefixStart;
+        }
+    }
+    ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
+    return sequenceLength;
+}
+
+HINT_INLINE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_execSequence(BYTE* op,
+    BYTE* const oend, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;   /* risk : address space underflow on oend=NULL */
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    assert(op != NULL /* Precondition */);
+    assert(oend_w < oend /* No underflow */);
+
+#if defined(__aarch64__)
+    /* prefetch sequence starting from match that will be used for copy later */
+    PREFETCH_L1(match);
+#endif
+    /* Handle edge cases in a slow path:
+     *   - Read beyond end of literals
+     *   - Match end is within WILDCOPY_OVERLIMIT of oend
+     *   - 32-bit mode and the match length overflows
+     */
+    if (UNLIKELY(
+        iLitEnd > litLimit ||
+        oMatchEnd > oend_w ||
+        (MEM_32bits() && (size_t)(oend - op) < sequenceLength + WILDCOPY_OVERLENGTH)))
+        return ZSTD_execSequenceEnd(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
+
+    /* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
+    assert(op <= oLitEnd /* No overflow */);
+    assert(oLitEnd < oMatchEnd /* Non-zero match & no overflow */);
+    assert(oMatchEnd <= oend /* No underflow */);
+    assert(iLitEnd <= litLimit /* Literal length is in bounds */);
+    assert(oLitEnd <= oend_w /* Can wildcopy literals */);
+    assert(oMatchEnd <= oend_w /* Can wildcopy matches */);
+
+    /* Copy Literals:
+     * Split out litLength <= 16 since it is nearly always true. +1.6% on gcc-9.
+     * We likely don't need the full 32-byte wildcopy.
+     */
+    assert(WILDCOPY_OVERLENGTH >= 16);
+    ZSTD_copy16(op, (*litPtr));
+    if (UNLIKELY(sequence.litLength > 16)) {
+        ZSTD_wildcopy(op + 16, (*litPtr) + 16, sequence.litLength - 16, ZSTD_no_overlap);
+    }
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* Copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix -> go into extDict */
+        RETURN_ERROR_IF(UNLIKELY(sequence.offset > (size_t)(oLitEnd - virtualStart)), corruption_detected, "");
+        match = dictEnd + (match - prefixStart);
+        if (match + sequence.matchLength <= dictEnd) {
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = (size_t)(dictEnd - match);
+            ZSTD_memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = prefixStart;
+        }
+    }
+    /* Match within prefix of 1 or more bytes */
+    assert(op <= oMatchEnd);
+    assert(oMatchEnd <= oend_w);
+    assert(match >= prefixStart);
+    assert(sequence.matchLength >= 1);
+
+    /* Nearly all offsets are >= WILDCOPY_VECLEN bytes, which means we can use wildcopy
+     * without overlap checking.
+     */
+    if (LIKELY(sequence.offset >= WILDCOPY_VECLEN)) {
+        /* We bet on a full wildcopy for matches, since we expect matches to be
+         * longer than literals (in general). In silesia, ~10% of matches are longer
+         * than 16 bytes.
+         */
+        ZSTD_wildcopy(op, match, sequence.matchLength, ZSTD_no_overlap);
+        return sequenceLength;
+    }
+    assert(sequence.offset < WILDCOPY_VECLEN);
+
+    /* Copy 8 bytes and spread the offset to be >= 8. */
+    ZSTD_overlapCopy8(&op, &match, sequence.offset);
+
+    /* If the match length is > 8 bytes, then continue with the wildcopy. */
+    if (sequence.matchLength > 8) {
+        assert(op < oMatchEnd);
+        ZSTD_wildcopy(op, match, sequence.matchLength - 8, ZSTD_overlap_src_before_dst);
+    }
+    return sequenceLength;
+}
+
+HINT_INLINE
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+size_t ZSTD_execSequenceSplitLitBuffer(BYTE* op,
+    BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    assert(op != NULL /* Precondition */);
+    assert(oend_w < oend /* No underflow */);
+    /* Handle edge cases in a slow path:
+     *   - Read beyond end of literals
+     *   - Match end is within WILDCOPY_OVERLIMIT of oend
+     *   - 32-bit mode and the match length overflows
+     */
+    if (UNLIKELY(
+            iLitEnd > litLimit ||
+            oMatchEnd > oend_w ||
+            (MEM_32bits() && (size_t)(oend - op) < sequenceLength + WILDCOPY_OVERLENGTH)))
+        return ZSTD_execSequenceEndSplitLitBuffer(op, oend, oend_w, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
+
+    /* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
+    assert(op <= oLitEnd /* No overflow */);
+    assert(oLitEnd < oMatchEnd /* Non-zero match & no overflow */);
+    assert(oMatchEnd <= oend /* No underflow */);
+    assert(iLitEnd <= litLimit /* Literal length is in bounds */);
+    assert(oLitEnd <= oend_w /* Can wildcopy literals */);
+    assert(oMatchEnd <= oend_w /* Can wildcopy matches */);
+
+    /* Copy Literals:
+     * Split out litLength <= 16 since it is nearly always true. +1.6% on gcc-9.
+     * We likely don't need the full 32-byte wildcopy.
+     */
+    assert(WILDCOPY_OVERLENGTH >= 16);
+    ZSTD_copy16(op, (*litPtr));
+    if (UNLIKELY(sequence.litLength > 16)) {
+        ZSTD_wildcopy(op+16, (*litPtr)+16, sequence.litLength-16, ZSTD_no_overlap);
+    }
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* Copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix -> go into extDict */
+        RETURN_ERROR_IF(UNLIKELY(sequence.offset > (size_t)(oLitEnd - virtualStart)), corruption_detected, "");
+        match = dictEnd + (match - prefixStart);
+        if (match + sequence.matchLength <= dictEnd) {
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = (size_t)(dictEnd - match);
+            ZSTD_memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = prefixStart;
+    }   }
+    /* Match within prefix of 1 or more bytes */
+    assert(op <= oMatchEnd);
+    assert(oMatchEnd <= oend_w);
+    assert(match >= prefixStart);
+    assert(sequence.matchLength >= 1);
+
+    /* Nearly all offsets are >= WILDCOPY_VECLEN bytes, which means we can use wildcopy
+     * without overlap checking.
+     */
+    if (LIKELY(sequence.offset >= WILDCOPY_VECLEN)) {
+        /* We bet on a full wildcopy for matches, since we expect matches to be
+         * longer than literals (in general). In silesia, ~10% of matches are longer
+         * than 16 bytes.
+         */
+        ZSTD_wildcopy(op, match, sequence.matchLength, ZSTD_no_overlap);
+        return sequenceLength;
+    }
+    assert(sequence.offset < WILDCOPY_VECLEN);
+
+    /* Copy 8 bytes and spread the offset to be >= 8. */
+    ZSTD_overlapCopy8(&op, &match, sequence.offset);
+
+    /* If the match length is > 8 bytes, then continue with the wildcopy. */
+    if (sequence.matchLength > 8) {
+        assert(op < oMatchEnd);
+        ZSTD_wildcopy(op, match, sequence.matchLength-8, ZSTD_overlap_src_before_dst);
+    }
+    return sequenceLength;
+}
+
+
+static void
+ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqSymbol* dt)
+{
+    const void* ptr = dt;
+    const ZSTD_seqSymbol_header* const DTableH = (const ZSTD_seqSymbol_header*)ptr;
+    DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog);
+    DEBUGLOG(6, "ZSTD_initFseState : val=%u using %u bits",
+                (U32)DStatePtr->state, DTableH->tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+FORCE_INLINE_TEMPLATE void
+ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, U16 nextState, U32 nbBits)
+{
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+    DStatePtr->state = nextState + lowBits;
+}
+
+/* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum
+ * offset bits. But we can only read at most STREAM_ACCUMULATOR_MIN_32
+ * bits before reloading. This value is the maximum number of bytes we read
+ * after reloading when we are decoding long offsets.
+ */
+#define LONG_OFFSETS_MAX_EXTRA_BITS_32                       \
+    (ZSTD_WINDOWLOG_MAX_32 > STREAM_ACCUMULATOR_MIN_32       \
+        ? ZSTD_WINDOWLOG_MAX_32 - STREAM_ACCUMULATOR_MIN_32  \
+        : 0)
+
+typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset_e;
+
+/**
+ * ZSTD_decodeSequence():
+ * @p longOffsets : tells the decoder to reload more bit while decoding large offsets
+ *                  only used in 32-bit mode
+ * @return : Sequence (litL + matchL + offset)
+ */
+FORCE_INLINE_TEMPLATE seq_t
+ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets, const int isLastSeq)
+{
+    seq_t seq;
+    /*
+     * ZSTD_seqSymbol is a 64 bits wide structure.
+     * It can be loaded in one operation
+     * and its fields extracted by simply shifting or bit-extracting on aarch64.
+     * GCC doesn't recognize this and generates more unnecessary ldr/ldrb/ldrh
+     * operations that cause performance drop. This can be avoided by using this
+     * ZSTD_memcpy hack.
+     */
+#if defined(__aarch64__) && (defined(__GNUC__) && !defined(__clang__))
+    ZSTD_seqSymbol llDInfoS, mlDInfoS, ofDInfoS;
+    ZSTD_seqSymbol* const llDInfo = &llDInfoS;
+    ZSTD_seqSymbol* const mlDInfo = &mlDInfoS;
+    ZSTD_seqSymbol* const ofDInfo = &ofDInfoS;
+    ZSTD_memcpy(llDInfo, seqState->stateLL.table + seqState->stateLL.state, sizeof(ZSTD_seqSymbol));
+    ZSTD_memcpy(mlDInfo, seqState->stateML.table + seqState->stateML.state, sizeof(ZSTD_seqSymbol));
+    ZSTD_memcpy(ofDInfo, seqState->stateOffb.table + seqState->stateOffb.state, sizeof(ZSTD_seqSymbol));
+#else
+    const ZSTD_seqSymbol* const llDInfo = seqState->stateLL.table + seqState->stateLL.state;
+    const ZSTD_seqSymbol* const mlDInfo = seqState->stateML.table + seqState->stateML.state;
+    const ZSTD_seqSymbol* const ofDInfo = seqState->stateOffb.table + seqState->stateOffb.state;
+#endif
+    seq.matchLength = mlDInfo->baseValue;
+    seq.litLength = llDInfo->baseValue;
+    {   U32 const ofBase = ofDInfo->baseValue;
+        BYTE const llBits = llDInfo->nbAdditionalBits;
+        BYTE const mlBits = mlDInfo->nbAdditionalBits;
+        BYTE const ofBits = ofDInfo->nbAdditionalBits;
+        BYTE const totalBits = llBits+mlBits+ofBits;
+
+        U16 const llNext = llDInfo->nextState;
+        U16 const mlNext = mlDInfo->nextState;
+        U16 const ofNext = ofDInfo->nextState;
+        U32 const llnbBits = llDInfo->nbBits;
+        U32 const mlnbBits = mlDInfo->nbBits;
+        U32 const ofnbBits = ofDInfo->nbBits;
+
+        assert(llBits <= MaxLLBits);
+        assert(mlBits <= MaxMLBits);
+        assert(ofBits <= MaxOff);
+        /*
+         * As gcc has better branch and block analyzers, sometimes it is only
+         * valuable to mark likeliness for clang, it gives around 3-4% of
+         * performance.
+         */
+
+        /* sequence */
+        {   size_t offset;
+            if (ofBits > 1) {
+                ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
+                ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
+                ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 > LONG_OFFSETS_MAX_EXTRA_BITS_32);
+                ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 - LONG_OFFSETS_MAX_EXTRA_BITS_32 >= MaxMLBits);
+                if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
+                    /* Always read extra bits, this keeps the logic simple,
+                     * avoids branches, and avoids accidentally reading 0 bits.
+                     */
+                    U32 const extraBits = LONG_OFFSETS_MAX_EXTRA_BITS_32;
+                    offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
+                    BIT_reloadDStream(&seqState->DStream);
+                    offset += BIT_readBitsFast(&seqState->DStream, extraBits);
+                } else {
+                    offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+                    if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
+                }
+                seqState->prevOffset[2] = seqState->prevOffset[1];
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset;
+            } else {
+                U32 const ll0 = (llDInfo->baseValue == 0);
+                if (LIKELY((ofBits == 0))) {
+                    offset = seqState->prevOffset[ll0];
+                    seqState->prevOffset[1] = seqState->prevOffset[!ll0];
+                    seqState->prevOffset[0] = offset;
+                } else {
+                    offset = ofBase + ll0 + BIT_readBitsFast(&seqState->DStream, 1);
+                    {   size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
+                        temp -= !temp; /* 0 is not valid: input corrupted => force offset to -1 => corruption detected at execSequence */
+                        if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
+                        seqState->prevOffset[1] = seqState->prevOffset[0];
+                        seqState->prevOffset[0] = offset = temp;
+            }   }   }
+            seq.offset = offset;
+        }
+
+        if (mlBits > 0)
+            seq.matchLength += BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/);
+
+        if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
+            BIT_reloadDStream(&seqState->DStream);
+        if (MEM_64bits() && UNLIKELY(totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
+            BIT_reloadDStream(&seqState->DStream);
+        /* Ensure there are enough bits to read the rest of data in 64-bit mode. */
+        ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
+
+        if (llBits > 0)
+            seq.litLength += BIT_readBitsFast(&seqState->DStream, llBits/*>0*/);
+
+        if (MEM_32bits())
+            BIT_reloadDStream(&seqState->DStream);
+
+        DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
+                    (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
+
+        if (!isLastSeq) {
+            /* don't update FSE state for last Sequence */
+            ZSTD_updateFseStateWithDInfo(&seqState->stateLL, &seqState->DStream, llNext, llnbBits);    /* <=  9 bits */
+            ZSTD_updateFseStateWithDInfo(&seqState->stateML, &seqState->DStream, mlNext, mlnbBits);    /* <=  9 bits */
+            if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
+            ZSTD_updateFseStateWithDInfo(&seqState->stateOffb, &seqState->DStream, ofNext, ofnbBits);  /* <=  8 bits */
+            BIT_reloadDStream(&seqState->DStream);
+        }
+    }
+
+    return seq;
+}
+
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+#if DEBUGLEVEL >= 1
+static int ZSTD_dictionaryIsActive(ZSTD_DCtx const* dctx, BYTE const* prefixStart, BYTE const* oLitEnd)
+{
+    size_t const windowSize = dctx->fParams.windowSize;
+    /* No dictionary used. */
+    if (dctx->dictContentEndForFuzzing == NULL) return 0;
+    /* Dictionary is our prefix. */
+    if (prefixStart == dctx->dictContentBeginForFuzzing) return 1;
+    /* Dictionary is not our ext-dict. */
+    if (dctx->dictEnd != dctx->dictContentEndForFuzzing) return 0;
+    /* Dictionary is not within our window size. */
+    if ((size_t)(oLitEnd - prefixStart) >= windowSize) return 0;
+    /* Dictionary is active. */
+    return 1;
+}
+#endif
+
+static void ZSTD_assertValidSequence(
+        ZSTD_DCtx const* dctx,
+        BYTE const* op, BYTE const* oend,
+        seq_t const seq,
+        BYTE const* prefixStart, BYTE const* virtualStart)
+{
+#if DEBUGLEVEL >= 1
+    if (dctx->isFrameDecompression) {
+        size_t const windowSize = dctx->fParams.windowSize;
+        size_t const sequenceSize = seq.litLength + seq.matchLength;
+        BYTE const* const oLitEnd = op + seq.litLength;
+        DEBUGLOG(6, "Checking sequence: litL=%u matchL=%u offset=%u",
+                (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
+        assert(op <= oend);
+        assert((size_t)(oend - op) >= sequenceSize);
+        assert(sequenceSize <= ZSTD_blockSizeMax(dctx));
+        if (ZSTD_dictionaryIsActive(dctx, prefixStart, oLitEnd)) {
+            size_t const dictSize = (size_t)((char const*)dctx->dictContentEndForFuzzing - (char const*)dctx->dictContentBeginForFuzzing);
+            /* Offset must be within the dictionary. */
+            assert(seq.offset <= (size_t)(oLitEnd - virtualStart));
+            assert(seq.offset <= windowSize + dictSize);
+        } else {
+            /* Offset must be within our window. */
+            assert(seq.offset <= windowSize);
+        }
+    }
+#else
+    (void)dctx, (void)op, (void)oend, (void)seq, (void)prefixStart, (void)virtualStart;
+#endif
+}
+#endif
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+
+
+FORCE_INLINE_TEMPLATE size_t
+DONT_VECTORIZE
+ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize, int nbSeq,
+                         const ZSTD_longOffset_e isLongOffset)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = (BYTE*)ZSTD_maybeNullPtrAdd(ostart, (ptrdiff_t)maxDstSize);
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* litBufferEnd = dctx->litBufferEnd;
+    const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
+    const BYTE* const vBase = (const BYTE*) (dctx->virtualStart);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer (%i seqs)", nbSeq);
+
+    /* Literals are split between internal buffer & output buffer */
+    if (nbSeq) {
+        seqState_t seqState;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        RETURN_ERROR_IF(
+            ERR_isError(BIT_initDStream(&seqState.DStream, seqStart, seqSize)),
+            corruption_detected, "");
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+        assert(dst != NULL);
+
+        ZSTD_STATIC_ASSERT(
+                BIT_DStream_unfinished < BIT_DStream_completed &&
+                BIT_DStream_endOfBuffer < BIT_DStream_completed &&
+                BIT_DStream_completed < BIT_DStream_overflow);
+
+        /* decompress without overrunning litPtr begins */
+        {   seq_t sequence = {0,0,0};  /* some static analyzer believe that @sequence is not initialized (it necessarily is, since for(;;) loop as at least one iteration) */
+            /* Align the decompression loop to 32 + 16 bytes.
+                *
+                * zstd compiled with gcc-9 on an Intel i9-9900k shows 10% decompression
+                * speed swings based on the alignment of the decompression loop. This
+                * performance swing is caused by parts of the decompression loop falling
+                * out of the DSB. The entire decompression loop should fit in the DSB,
+                * when it can't we get much worse performance. You can measure if you've
+                * hit the good case or the bad case with this perf command for some
+                * compressed file test.zst:
+                *
+                *   perf stat -e cycles -e instructions -e idq.all_dsb_cycles_any_uops \
+                *             -e idq.all_mite_cycles_any_uops -- ./zstd -tq test.zst
+                *
+                * If you see most cycles served out of the MITE you've hit the bad case.
+                * If you see most cycles served out of the DSB you've hit the good case.
+                * If it is pretty even then you may be in an okay case.
+                *
+                * This issue has been reproduced on the following CPUs:
+                *   - Kabylake: Macbook Pro (15-inch, 2019) 2.4 GHz Intel Core i9
+                *               Use Instruments->Counters to get DSB/MITE cycles.
+                *               I never got performance swings, but I was able to
+                *               go from the good case of mostly DSB to half of the
+                *               cycles served from MITE.
+                *   - Coffeelake: Intel i9-9900k
+                *   - Coffeelake: Intel i7-9700k
+                *
+                * I haven't been able to reproduce the instability or DSB misses on any
+                * of the following CPUS:
+                *   - Haswell
+                *   - Broadwell: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GH
+                *   - Skylake
+                *
+                * Alignment is done for each of the three major decompression loops:
+                *   - ZSTD_decompressSequences_bodySplitLitBuffer - presplit section of the literal buffer
+                *   - ZSTD_decompressSequences_bodySplitLitBuffer - postsplit section of the literal buffer
+                *   - ZSTD_decompressSequences_body
+                * Alignment choices are made to minimize large swings on bad cases and influence on performance
+                * from changes external to this code, rather than to overoptimize on the current commit.
+                *
+                * If you are seeing performance stability this script can help test.
+                * It tests on 4 commits in zstd where I saw performance change.
+                *
+                *   https://gist.github.com/terrelln/9889fc06a423fd5ca6e99351564473f4
+                */
+#if defined(__GNUC__) && defined(__x86_64__)
+            __asm__(".p2align 6");
+#  if __GNUC__ >= 7
+	    /* good for gcc-7, gcc-9, and gcc-11 */
+            __asm__("nop");
+            __asm__(".p2align 5");
+            __asm__("nop");
+            __asm__(".p2align 4");
+#    if __GNUC__ == 8 || __GNUC__ == 10
+	    /* good for gcc-8 and gcc-10 */
+            __asm__("nop");
+            __asm__(".p2align 3");
+#    endif
+#  endif
+#endif
+
+            /* Handle the initial state where litBuffer is currently split between dst and litExtraBuffer */
+            for ( ; nbSeq; nbSeq--) {
+                sequence = ZSTD_decodeSequence(&seqState, isLongOffset, nbSeq==1);
+                if (litPtr + sequence.litLength > dctx->litBufferEnd) break;
+                {   size_t const oneSeqSize = ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence.litLength - WILDCOPY_OVERLENGTH, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                    assert(!ZSTD_isError(oneSeqSize));
+                    ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+                    if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                        return oneSeqSize;
+                    DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+                    op += oneSeqSize;
+            }   }
+            DEBUGLOG(6, "reached: (litPtr + sequence.litLength > dctx->litBufferEnd)");
+
+            /* If there are more sequences, they will need to read literals from litExtraBuffer; copy over the remainder from dst and update litPtr and litEnd */
+            if (nbSeq > 0) {
+                const size_t leftoverLit = (size_t)(dctx->litBufferEnd - litPtr);
+                assert(dctx->litBufferEnd >= litPtr);
+                DEBUGLOG(6, "There are %i sequences left, and %zu/%zu literals left in buffer", nbSeq, leftoverLit, sequence.litLength);
+                if (leftoverLit) {
+                    RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
+                    ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
+                    sequence.litLength -= leftoverLit;
+                    op += leftoverLit;
+                }
+                litPtr = dctx->litExtraBuffer;
+                litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
+                {   size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                    assert(!ZSTD_isError(oneSeqSize));
+                    ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+                    if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                        return oneSeqSize;
+                    DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+                    op += oneSeqSize;
+                }
+                nbSeq--;
+            }
+        }
+
+        if (nbSeq > 0) {
+            /* there is remaining lit from extra buffer */
+
+#if defined(__GNUC__) && defined(__x86_64__)
+            __asm__(".p2align 6");
+            __asm__("nop");
+#  if __GNUC__ != 7
+            /* worse for gcc-7 better for gcc-8, gcc-9, and gcc-10 and clang */
+            __asm__(".p2align 4");
+            __asm__("nop");
+            __asm__(".p2align 3");
+#  elif __GNUC__ >= 11
+            __asm__(".p2align 3");
+#  else
+            __asm__(".p2align 5");
+            __asm__("nop");
+            __asm__(".p2align 3");
+#  endif
+#endif
+
+            for ( ; nbSeq ; nbSeq--) {
+                seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, nbSeq==1);
+                size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+                if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                    return oneSeqSize;
+                DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+                op += oneSeqSize;
+            }
+        }
+
+        /* check if reached exact end */
+        DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer: after decode loop, remaining nbSeq : %i", nbSeq);
+        RETURN_ERROR_IF(nbSeq, corruption_detected, "");
+        DEBUGLOG(5, "bitStream : start=%p, ptr=%p, bitsConsumed=%u", seqState.DStream.start, seqState.DStream.ptr, seqState.DStream.bitsConsumed);
+        RETURN_ERROR_IF(!BIT_endOfDStream(&seqState.DStream), corruption_detected, "");
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    if (dctx->litBufferLocation == ZSTD_split) {
+        /* split hasn't been reached yet, first get dst then copy litExtraBuffer */
+        size_t const lastLLSize = (size_t)(litBufferEnd - litPtr);
+        DEBUGLOG(6, "copy last literals from segment : %u", (U32)lastLLSize);
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memmove(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+        litPtr = dctx->litExtraBuffer;
+        litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+        dctx->litBufferLocation = ZSTD_not_in_dst;
+    }
+    /* copy last literals from internal buffer */
+    {   size_t const lastLLSize = (size_t)(litBufferEnd - litPtr);
+        DEBUGLOG(6, "copy last literals from internal buffer : %u", (U32)lastLLSize);
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+    }   }
+
+    DEBUGLOG(6, "decoded block of size %u bytes", (U32)(op - ostart));
+    return (size_t)(op - ostart);
+}
+
+FORCE_INLINE_TEMPLATE size_t
+DONT_VECTORIZE
+ZSTD_decompressSequences_body(ZSTD_DCtx* dctx,
+    void* dst, size_t maxDstSize,
+    const void* seqStart, size_t seqSize, int nbSeq,
+    const ZSTD_longOffset_e isLongOffset)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = (dctx->litBufferLocation == ZSTD_not_in_dst) ?
+                        (BYTE*)ZSTD_maybeNullPtrAdd(ostart, (ptrdiff_t)maxDstSize) :
+                        dctx->litBuffer;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    const BYTE* const prefixStart = (const BYTE*)(dctx->prefixStart);
+    const BYTE* const vBase = (const BYTE*)(dctx->virtualStart);
+    const BYTE* const dictEnd = (const BYTE*)(dctx->dictEnd);
+    DEBUGLOG(5, "ZSTD_decompressSequences_body: nbSeq = %d", nbSeq);
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seqState_t seqState;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i = 0; i < ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        RETURN_ERROR_IF(
+            ERR_isError(BIT_initDStream(&seqState.DStream, seqStart, seqSize)),
+            corruption_detected, "");
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+        assert(dst != NULL);
+
+#if defined(__GNUC__) && defined(__x86_64__)
+            __asm__(".p2align 6");
+            __asm__("nop");
+#  if __GNUC__ >= 7
+            __asm__(".p2align 5");
+            __asm__("nop");
+            __asm__(".p2align 3");
+#  else
+            __asm__(".p2align 4");
+            __asm__("nop");
+            __asm__(".p2align 3");
+#  endif
+#endif
+
+        for ( ; nbSeq ; nbSeq--) {
+            seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, nbSeq==1);
+            size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+            assert(!ZSTD_isError(oneSeqSize));
+            ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+            if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                return oneSeqSize;
+            DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        assert(nbSeq == 0);
+        RETURN_ERROR_IF(!BIT_endOfDStream(&seqState.DStream), corruption_detected, "");
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = (size_t)(litEnd - litPtr);
+        DEBUGLOG(6, "copy last literals : %u", (U32)lastLLSize);
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+    }   }
+
+    DEBUGLOG(6, "decoded block of size %u bytes", (U32)(op - ostart));
+    return (size_t)(op - ostart);
+}
+
+static size_t
+ZSTD_decompressSequences_default(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+static size_t
+ZSTD_decompressSequencesSplitLitBuffer_default(ZSTD_DCtx* dctx,
+                                               void* dst, size_t maxDstSize,
+                                         const void* seqStart, size_t seqSize, int nbSeq,
+                                         const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+
+FORCE_INLINE_TEMPLATE
+
+size_t ZSTD_prefetchMatch(size_t prefetchPos, seq_t const sequence,
+                   const BYTE* const prefixStart, const BYTE* const dictEnd)
+{
+    prefetchPos += sequence.litLength;
+    {   const BYTE* const matchBase = (sequence.offset > prefetchPos) ? dictEnd : prefixStart;
+        /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted.
+         * No consequence though : memory address is only used for prefetching, not for dereferencing */
+        const BYTE* const match = (const BYTE*)ZSTD_wrappedPtrSub(ZSTD_wrappedPtrAdd(matchBase, (ptrdiff_t)prefetchPos), (ptrdiff_t)sequence.offset);
+        PREFETCH_L1(match); PREFETCH_L1(ZSTD_wrappedPtrAdd(match, CACHELINE_SIZE));   /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
+    }
+    return prefetchPos + sequence.matchLength;
+}
+
+/* This decoding function employs prefetching
+ * to reduce latency impact of cache misses.
+ * It's generally employed when block contains a significant portion of long-distance matches
+ * or when coupled with a "cold" dictionary */
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_decompressSequencesLong_body(
+                               ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize, int nbSeq,
+                         const ZSTD_longOffset_e isLongOffset)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = (dctx->litBufferLocation == ZSTD_in_dst) ?
+                        dctx->litBuffer :
+                        (BYTE*)ZSTD_maybeNullPtrAdd(ostart, (ptrdiff_t)maxDstSize);
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* litBufferEnd = dctx->litBufferEnd;
+    const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
+    const BYTE* const dictStart = (const BYTE*) (dctx->virtualStart);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+
+    /* Regen sequences */
+    if (nbSeq) {
+#define STORED_SEQS 8
+#define STORED_SEQS_MASK (STORED_SEQS-1)
+#define ADVANCED_SEQS STORED_SEQS
+        seq_t sequences[STORED_SEQS];
+        int const seqAdvance = MIN(nbSeq, ADVANCED_SEQS);
+        seqState_t seqState;
+        int seqNb;
+        size_t prefetchPos = (size_t)(op-prefixStart); /* track position relative to prefixStart */
+
+        dctx->fseEntropy = 1;
+        { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        assert(dst != NULL);
+        RETURN_ERROR_IF(
+            ERR_isError(BIT_initDStream(&seqState.DStream, seqStart, seqSize)),
+            corruption_detected, "");
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+
+        /* prepare in advance */
+        for (seqNb=0; seqNb<seqAdvance; seqNb++) {
+            seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, seqNb == nbSeq-1);
+            prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
+            sequences[seqNb] = sequence;
+        }
+
+        /* decompress without stomping litBuffer */
+        for (; seqNb < nbSeq; seqNb++) {
+            seq_t sequence = ZSTD_decodeSequence(&seqState, isLongOffset, seqNb == nbSeq-1);
+
+            if (dctx->litBufferLocation == ZSTD_split && litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength > dctx->litBufferEnd) {
+                /* lit buffer is reaching split point, empty out the first buffer and transition to litExtraBuffer */
+                const size_t leftoverLit = (size_t)(dctx->litBufferEnd - litPtr);
+                assert(dctx->litBufferEnd >= litPtr);
+                if (leftoverLit) {
+                    RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
+                    ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
+                    sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength -= leftoverLit;
+                    op += leftoverLit;
+                }
+                litPtr = dctx->litExtraBuffer;
+                litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
+                {   size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                    assert(!ZSTD_isError(oneSeqSize));
+                    ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+                    if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+
+                    prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
+                    sequences[seqNb & STORED_SEQS_MASK] = sequence;
+                    op += oneSeqSize;
+            }   }
+            else
+            {
+                /* lit buffer is either wholly contained in first or second split, or not split at all*/
+                size_t const oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
+                    ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength - WILDCOPY_OVERLENGTH, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd) :
+                    ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+                if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+
+                prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
+                sequences[seqNb & STORED_SEQS_MASK] = sequence;
+                op += oneSeqSize;
+            }
+        }
+        RETURN_ERROR_IF(!BIT_endOfDStream(&seqState.DStream), corruption_detected, "");
+
+        /* finish queue */
+        seqNb -= seqAdvance;
+        for ( ; seqNb<nbSeq ; seqNb++) {
+            seq_t *sequence = &(sequences[seqNb&STORED_SEQS_MASK]);
+            if (dctx->litBufferLocation == ZSTD_split && litPtr + sequence->litLength > dctx->litBufferEnd) {
+                const size_t leftoverLit = (size_t)(dctx->litBufferEnd - litPtr);
+                assert(dctx->litBufferEnd >= litPtr);
+                if (leftoverLit) {
+                    RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
+                    ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
+                    sequence->litLength -= leftoverLit;
+                    op += leftoverLit;
+                }
+                litPtr = dctx->litExtraBuffer;
+                litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
+                {   size_t const oneSeqSize = ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                    assert(!ZSTD_isError(oneSeqSize));
+                    ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+                    if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+                    op += oneSeqSize;
+                }
+            }
+            else
+            {
+                size_t const oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
+                    ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence->litLength - WILDCOPY_OVERLENGTH, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd) :
+                    ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+                if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+                op += oneSeqSize;
+            }
+        }
+
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    if (dctx->litBufferLocation == ZSTD_split) { /* first deplete literal buffer in dst, then copy litExtraBuffer */
+        size_t const lastLLSize = (size_t)(litBufferEnd - litPtr);
+        assert(litBufferEnd >= litPtr);
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memmove(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+        litPtr = dctx->litExtraBuffer;
+        litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+    }
+    {   size_t const lastLLSize = (size_t)(litBufferEnd - litPtr);
+        assert(litBufferEnd >= litPtr);
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memmove(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return (size_t)(op - ostart);
+}
+
+static size_t
+ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
+
+
+
+#if DYNAMIC_BMI2
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+static BMI2_TARGET_ATTRIBUTE size_t
+DONT_VECTORIZE
+ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+static BMI2_TARGET_ATTRIBUTE size_t
+DONT_VECTORIZE
+ZSTD_decompressSequencesSplitLitBuffer_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+static BMI2_TARGET_ATTRIBUTE size_t
+ZSTD_decompressSequencesLong_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
+
+#endif /* DYNAMIC_BMI2 */
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+static size_t
+ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
+                   const void* seqStart, size_t seqSize, int nbSeq,
+                   const ZSTD_longOffset_e isLongOffset)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequences");
+#if DYNAMIC_BMI2
+    if (ZSTD_DCtx_get_bmi2(dctx)) {
+        return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+    }
+#endif
+    return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+static size_t
+ZSTD_decompressSequencesSplitLitBuffer(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
+                                 const void* seqStart, size_t seqSize, int nbSeq,
+                                 const ZSTD_longOffset_e isLongOffset)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequencesSplitLitBuffer");
+#if DYNAMIC_BMI2
+    if (ZSTD_DCtx_get_bmi2(dctx)) {
+        return ZSTD_decompressSequencesSplitLitBuffer_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+    }
+#endif
+    return ZSTD_decompressSequencesSplitLitBuffer_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
+
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+/* ZSTD_decompressSequencesLong() :
+ * decompression function triggered when a minimum share of offsets is considered "long",
+ * aka out of cache.
+ * note : "long" definition seems overloaded here, sometimes meaning "wider than bitstream register", and sometimes meaning "farther than memory cache distance".
+ * This function will try to mitigate main memory latency through the use of prefetching */
+static size_t
+ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
+                             void* dst, size_t maxDstSize,
+                             const void* seqStart, size_t seqSize, int nbSeq,
+                             const ZSTD_longOffset_e isLongOffset)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequencesLong");
+#if DYNAMIC_BMI2
+    if (ZSTD_DCtx_get_bmi2(dctx)) {
+        return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+    }
+#endif
+  return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
+
+
+/**
+ * @returns The total size of the history referenceable by zstd, including
+ * both the prefix and the extDict. At @p op any offset larger than this
+ * is invalid.
+ */
+static size_t ZSTD_totalHistorySize(void* curPtr, const void* virtualStart)
+{
+    return (size_t)((char*)curPtr - (const char*)virtualStart);
+}
+
+typedef struct {
+    unsigned longOffsetShare;
+    unsigned maxNbAdditionalBits;
+} ZSTD_OffsetInfo;
+
+/* ZSTD_getOffsetInfo() :
+ * condition : offTable must be valid
+ * @return : "share" of long offsets (arbitrarily defined as > (1<<23))
+ *           compared to maximum possible of (1<<OffFSELog),
+ *           as well as the maximum number additional bits required.
+ */
+static ZSTD_OffsetInfo
+ZSTD_getOffsetInfo(const ZSTD_seqSymbol* offTable, int nbSeq)
+{
+    ZSTD_OffsetInfo info = {0, 0};
+    /* If nbSeq == 0, then the offTable is uninitialized, but we have
+     * no sequences, so both values should be 0.
+     */
+    if (nbSeq != 0) {
+        const void* ptr = offTable;
+        U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog;
+        const ZSTD_seqSymbol* table = offTable + 1;
+        U32 const max = 1 << tableLog;
+        U32 u;
+        DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog);
+
+        assert(max <= (1 << OffFSELog));  /* max not too large */
+        for (u=0; u<max; u++) {
+            info.maxNbAdditionalBits = MAX(info.maxNbAdditionalBits, table[u].nbAdditionalBits);
+            if (table[u].nbAdditionalBits > 22) info.longOffsetShare += 1;
+        }
+
+        assert(tableLog <= OffFSELog);
+        info.longOffsetShare <<= (OffFSELog - tableLog);  /* scale to OffFSELog */
+    }
+
+    return info;
+}
+
+/**
+ * @returns The maximum offset we can decode in one read of our bitstream, without
+ * reloading more bits in the middle of the offset bits read. Any offsets larger
+ * than this must use the long offset decoder.
+ */
+static size_t ZSTD_maxShortOffset(void)
+{
+    if (MEM_64bits()) {
+        /* We can decode any offset without reloading bits.
+         * This might change if the max window size grows.
+         */
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
+        return (size_t)-1;
+    } else {
+        /* The maximum offBase is (1 << (STREAM_ACCUMULATOR_MIN + 1)) - 1.
+         * This offBase would require STREAM_ACCUMULATOR_MIN extra bits.
+         * Then we have to subtract ZSTD_REP_NUM to get the maximum possible offset.
+         */
+        size_t const maxOffbase = ((size_t)1 << (STREAM_ACCUMULATOR_MIN + 1)) - 1;
+        size_t const maxOffset = maxOffbase - ZSTD_REP_NUM;
+        assert(ZSTD_highbit32((U32)maxOffbase) == STREAM_ACCUMULATOR_MIN);
+        return maxOffset;
+    }
+}
+
+size_t
+ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize, const streaming_operation streaming)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+    DEBUGLOG(5, "ZSTD_decompressBlock_internal (cSize : %u)", (unsigned)srcSize);
+
+    /* Note : the wording of the specification
+     * allows compressed block to be sized exactly ZSTD_blockSizeMax(dctx).
+     * This generally does not happen, as it makes little sense,
+     * since an uncompressed block would feature same size and have no decompression cost.
+     * Also, note that decoder from reference libzstd before < v1.5.4
+     * would consider this edge case as an error.
+     * As a consequence, avoid generating compressed blocks of size ZSTD_blockSizeMax(dctx)
+     * for broader compatibility with the deployed ecosystem of zstd decoders */
+    RETURN_ERROR_IF(srcSize > ZSTD_blockSizeMax(dctx), srcSize_wrong, "");
+
+    /* Decode literals section */
+    {   size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize, dst, dstCapacity, streaming);
+        DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : cSize=%u, nbLiterals=%zu", (U32)litCSize, dctx->litSize);
+        if (ZSTD_isError(litCSize)) return litCSize;
+        ip += litCSize;
+        srcSize -= litCSize;
+    }
+
+    /* Build Decoding Tables */
+    {
+        /* Compute the maximum block size, which must also work when !frame and fParams are unset.
+         * Additionally, take the min with dstCapacity to ensure that the totalHistorySize fits in a size_t.
+         */
+        size_t const blockSizeMax = MIN(dstCapacity, ZSTD_blockSizeMax(dctx));
+        size_t const totalHistorySize = ZSTD_totalHistorySize(ZSTD_maybeNullPtrAdd(dst, (ptrdiff_t)blockSizeMax), (BYTE const*)dctx->virtualStart);
+        /* isLongOffset must be true if there are long offsets.
+         * Offsets are long if they are larger than ZSTD_maxShortOffset().
+         * We don't expect that to be the case in 64-bit mode.
+         *
+         * We check here to see if our history is large enough to allow long offsets.
+         * If it isn't, then we can't possible have (valid) long offsets. If the offset
+         * is invalid, then it is okay to read it incorrectly.
+         *
+         * If isLongOffsets is true, then we will later check our decoding table to see
+         * if it is even possible to generate long offsets.
+         */
+        ZSTD_longOffset_e isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (totalHistorySize > ZSTD_maxShortOffset()));
+        /* These macros control at build-time which decompressor implementation
+         * we use. If neither is defined, we do some inspection and dispatch at
+         * runtime.
+         */
+#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
+    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
+        int usePrefetchDecoder = dctx->ddictIsCold;
+#else
+        /* Set to 1 to avoid computing offset info if we don't need to.
+         * Otherwise this value is ignored.
+         */
+        int usePrefetchDecoder = 1;
+#endif
+        int nbSeq;
+        size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize);
+        if (ZSTD_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+        srcSize -= seqHSize;
+
+        RETURN_ERROR_IF((dst == NULL || dstCapacity == 0) && nbSeq > 0, dstSize_tooSmall, "NULL not handled");
+        RETURN_ERROR_IF(MEM_64bits() && sizeof(size_t) == sizeof(void*) && (size_t)(-1) - (size_t)dst < (size_t)(1 << 20), dstSize_tooSmall,
+                "invalid dst");
+
+        /* If we could potentially have long offsets, or we might want to use the prefetch decoder,
+         * compute information about the share of long offsets, and the maximum nbAdditionalBits.
+         * NOTE: could probably use a larger nbSeq limit
+         */
+        if (isLongOffset || (!usePrefetchDecoder && (totalHistorySize > (1u << 24)) && (nbSeq > 8))) {
+            ZSTD_OffsetInfo const info = ZSTD_getOffsetInfo(dctx->OFTptr, nbSeq);
+            if (isLongOffset && info.maxNbAdditionalBits <= STREAM_ACCUMULATOR_MIN) {
+                /* If isLongOffset, but the maximum number of additional bits that we see in our table is small
+                 * enough, then we know it is impossible to have too long an offset in this block, so we can
+                 * use the regular offset decoder.
+                 */
+                isLongOffset = ZSTD_lo_isRegularOffset;
+            }
+            if (!usePrefetchDecoder) {
+                U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
+                usePrefetchDecoder = (info.longOffsetShare >= minShare);
+            }
+        }
+
+        dctx->ddictIsCold = 0;
+
+#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
+    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
+        if (usePrefetchDecoder) {
+#else
+        (void)usePrefetchDecoder;
+        {
+#endif
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+            return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
+#endif
+        }
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+        /* else */
+        if (dctx->litBufferLocation == ZSTD_split)
+            return ZSTD_decompressSequencesSplitLitBuffer(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
+        else
+            return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
+#endif
+    }
+}
+
+
+ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
+void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst, size_t dstSize)
+{
+    if (dst != dctx->previousDstEnd && dstSize > 0) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->virtualStart = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
+        dctx->prefixStart = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+size_t ZSTD_decompressBlock_deprecated(ZSTD_DCtx* dctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize)
+{
+    size_t dSize;
+    dctx->isFrameDecompression = 0;
+    ZSTD_checkContinuity(dctx, dst, dstCapacity);
+    dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, not_streaming);
+    FORWARD_IF_ERROR(dSize, "");
+    dctx->previousDstEnd = (char*)dst + dSize;
+    return dSize;
+}
+
+
+/* NOTE: Must just wrap ZSTD_decompressBlock_deprecated() */
+size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    return ZSTD_decompressBlock_deprecated(dctx, dst, dstCapacity, src, srcSize);
+}
diff --git a/lib/deprecated/zbuff_compress.c b/lib/deprecated/zbuff_compress.c
new file mode 100644
index 0000000..1d86821
--- /dev/null
+++ b/lib/deprecated/zbuff_compress.c
@@ -0,0 +1,167 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+
+/* *************************************
+*  Dependencies
+***************************************/
+#define ZBUFF_STATIC_LINKING_ONLY
+#include "zbuff.h"
+#include "../common/error_private.h"
+
+
+/*-***********************************************************
+*  Streaming compression
+*
+*  A ZBUFF_CCtx object is required to track streaming operation.
+*  Use ZBUFF_createCCtx() and ZBUFF_freeCCtx() to create/release resources.
+*  Use ZBUFF_compressInit() to start a new compression operation.
+*  ZBUFF_CCtx objects can be reused multiple times.
+*
+*  Use ZBUFF_compressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to call again the function with remaining input.
+*  The content of dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters or change dst .
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  ZBUFF_compressFlush() can be used to instruct ZBUFF to compress and output whatever remains within its buffer.
+*  Note that it will not output more than *dstCapacityPtr.
+*  Therefore, some content might still be left into its internal buffer if dst buffer is too small.
+*  @return : nb of bytes still present into internal buffer (0 if it's empty)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  ZBUFF_compressEnd() instructs to finish a frame.
+*  It will perform a flush and write frame epilogue.
+*  Similar to ZBUFF_compressFlush(), it may not be able to output the entire internal buffer content if *dstCapacityPtr is too small.
+*  @return : nb of bytes still present into internal buffer (0 if it's empty)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory)
+*  input : ZSTD_BLOCKSIZE_MAX (128 KB), internal unit size, it improves latency to use this value.
+*  output : ZSTD_compressBound(ZSTD_BLOCKSIZE_MAX) + ZSTD_blockHeaderSize + ZBUFF_endFrameSize : ensures it's always possible to write/flush/end a full block at best speed.
+* ***********************************************************/
+
+ZBUFF_CCtx* ZBUFF_createCCtx(void)
+{
+    return ZSTD_createCStream();
+}
+
+ZBUFF_CCtx* ZBUFF_createCCtx_advanced(ZSTD_customMem customMem)
+{
+    return ZSTD_createCStream_advanced(customMem);
+}
+
+size_t ZBUFF_freeCCtx(ZBUFF_CCtx* zbc)
+{
+    return ZSTD_freeCStream(zbc);
+}
+
+
+/* ======   Initialization   ====== */
+
+size_t ZBUFF_compressInit_advanced(ZBUFF_CCtx* zbc,
+                                   const void* dict, size_t dictSize,
+                                   ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;  /* preserve "0 == unknown" behavior */
+    FORWARD_IF_ERROR(ZSTD_CCtx_reset(zbc, ZSTD_reset_session_only), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setPledgedSrcSize(zbc, pledgedSrcSize), "");
+
+    FORWARD_IF_ERROR(ZSTD_checkCParams(params.cParams), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_windowLog, params.cParams.windowLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_hashLog, params.cParams.hashLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_chainLog, params.cParams.chainLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_searchLog, params.cParams.searchLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_minMatch, params.cParams.minMatch), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_targetLength, params.cParams.targetLength), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_strategy, params.cParams.strategy), "");
+
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_contentSizeFlag, params.fParams.contentSizeFlag), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_checksumFlag, params.fParams.checksumFlag), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_dictIDFlag, params.fParams.noDictIDFlag), "");
+
+    FORWARD_IF_ERROR(ZSTD_CCtx_loadDictionary(zbc, dict, dictSize), "");
+    return 0;
+}
+
+size_t ZBUFF_compressInitDictionary(ZBUFF_CCtx* zbc, const void* dict, size_t dictSize, int compressionLevel)
+{
+    FORWARD_IF_ERROR(ZSTD_CCtx_reset(zbc, ZSTD_reset_session_only), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_compressionLevel, compressionLevel), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_loadDictionary(zbc, dict, dictSize), "");
+    return 0;
+}
+
+size_t ZBUFF_compressInit(ZBUFF_CCtx* zbc, int compressionLevel)
+{
+    return ZSTD_initCStream(zbc, compressionLevel);
+}
+
+/* ======   Compression   ====== */
+
+
+size_t ZBUFF_compressContinue(ZBUFF_CCtx* zbc,
+                              void* dst, size_t* dstCapacityPtr,
+                        const void* src, size_t* srcSizePtr)
+{
+    size_t result;
+    ZSTD_outBuffer outBuff;
+    ZSTD_inBuffer inBuff;
+    outBuff.dst = dst;
+    outBuff.pos = 0;
+    outBuff.size = *dstCapacityPtr;
+    inBuff.src = src;
+    inBuff.pos = 0;
+    inBuff.size = *srcSizePtr;
+    result = ZSTD_compressStream(zbc, &outBuff, &inBuff);
+    *dstCapacityPtr = outBuff.pos;
+    *srcSizePtr = inBuff.pos;
+    return result;
+}
+
+
+
+/* ======   Finalize   ====== */
+
+size_t ZBUFF_compressFlush(ZBUFF_CCtx* zbc, void* dst, size_t* dstCapacityPtr)
+{
+    size_t result;
+    ZSTD_outBuffer outBuff;
+    outBuff.dst = dst;
+    outBuff.pos = 0;
+    outBuff.size = *dstCapacityPtr;
+    result = ZSTD_flushStream(zbc, &outBuff);
+    *dstCapacityPtr = outBuff.pos;
+    return result;
+}
+
+
+size_t ZBUFF_compressEnd(ZBUFF_CCtx* zbc, void* dst, size_t* dstCapacityPtr)
+{
+    size_t result;
+    ZSTD_outBuffer outBuff;
+    outBuff.dst = dst;
+    outBuff.pos = 0;
+    outBuff.size = *dstCapacityPtr;
+    result = ZSTD_endStream(zbc, &outBuff);
+    *dstCapacityPtr = outBuff.pos;
+    return result;
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+size_t ZBUFF_recommendedCInSize(void)  { return ZSTD_CStreamInSize(); }
+size_t ZBUFF_recommendedCOutSize(void) { return ZSTD_CStreamOutSize(); }
diff --git a/lib/dictBuilder/divsufsort.c b/lib/dictBuilder/divsufsort.c
new file mode 100644
index 0000000..e5b117b
--- /dev/null
+++ b/lib/dictBuilder/divsufsort.c
@@ -0,0 +1,1913 @@
+/*
+ * divsufsort.c for libdivsufsort-lite
+ * Copyright (c) 2003-2008 Yuta Mori All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use,
+ * copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following
+ * conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/*- Compiler specifics -*/
+#ifdef __clang__
+#pragma clang diagnostic ignored "-Wshorten-64-to-32"
+#endif
+
+#if defined(_MSC_VER)
+#  pragma warning(disable : 4244)
+#  pragma warning(disable : 4127)    /* C4127 : Condition expression is constant */
+#endif
+
+
+/*- Dependencies -*/
+#include <assert.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "divsufsort.h"
+
+/*- Constants -*/
+#if defined(INLINE)
+# undef INLINE
+#endif
+#if !defined(INLINE)
+# define INLINE __inline
+#endif
+#if defined(ALPHABET_SIZE) && (ALPHABET_SIZE < 1)
+# undef ALPHABET_SIZE
+#endif
+#if !defined(ALPHABET_SIZE)
+# define ALPHABET_SIZE (256)
+#endif
+#define BUCKET_A_SIZE (ALPHABET_SIZE)
+#define BUCKET_B_SIZE (ALPHABET_SIZE * ALPHABET_SIZE)
+#if defined(SS_INSERTIONSORT_THRESHOLD)
+# if SS_INSERTIONSORT_THRESHOLD < 1
+#  undef SS_INSERTIONSORT_THRESHOLD
+#  define SS_INSERTIONSORT_THRESHOLD (1)
+# endif
+#else
+# define SS_INSERTIONSORT_THRESHOLD (8)
+#endif
+#if defined(SS_BLOCKSIZE)
+# if SS_BLOCKSIZE < 0
+#  undef SS_BLOCKSIZE
+#  define SS_BLOCKSIZE (0)
+# elif 32768 <= SS_BLOCKSIZE
+#  undef SS_BLOCKSIZE
+#  define SS_BLOCKSIZE (32767)
+# endif
+#else
+# define SS_BLOCKSIZE (1024)
+#endif
+/* minstacksize = log(SS_BLOCKSIZE) / log(3) * 2 */
+#if SS_BLOCKSIZE == 0
+# define SS_MISORT_STACKSIZE (96)
+#elif SS_BLOCKSIZE <= 4096
+# define SS_MISORT_STACKSIZE (16)
+#else
+# define SS_MISORT_STACKSIZE (24)
+#endif
+#define SS_SMERGE_STACKSIZE (32)
+#define TR_INSERTIONSORT_THRESHOLD (8)
+#define TR_STACKSIZE (64)
+
+
+/*- Macros -*/
+#ifndef SWAP
+# define SWAP(_a, _b) do { t = (_a); (_a) = (_b); (_b) = t; } while(0)
+#endif /* SWAP */
+#ifndef MIN
+# define MIN(_a, _b) (((_a) < (_b)) ? (_a) : (_b))
+#endif /* MIN */
+#ifndef MAX
+# define MAX(_a, _b) (((_a) > (_b)) ? (_a) : (_b))
+#endif /* MAX */
+#define STACK_PUSH(_a, _b, _c, _d)\
+  do {\
+    assert(ssize < STACK_SIZE);\
+    stack[ssize].a = (_a), stack[ssize].b = (_b),\
+    stack[ssize].c = (_c), stack[ssize++].d = (_d);\
+  } while(0)
+#define STACK_PUSH5(_a, _b, _c, _d, _e)\
+  do {\
+    assert(ssize < STACK_SIZE);\
+    stack[ssize].a = (_a), stack[ssize].b = (_b),\
+    stack[ssize].c = (_c), stack[ssize].d = (_d), stack[ssize++].e = (_e);\
+  } while(0)
+#define STACK_POP(_a, _b, _c, _d)\
+  do {\
+    assert(0 <= ssize);\
+    if(ssize == 0) { return; }\
+    (_a) = stack[--ssize].a, (_b) = stack[ssize].b,\
+    (_c) = stack[ssize].c, (_d) = stack[ssize].d;\
+  } while(0)
+#define STACK_POP5(_a, _b, _c, _d, _e)\
+  do {\
+    assert(0 <= ssize);\
+    if(ssize == 0) { return; }\
+    (_a) = stack[--ssize].a, (_b) = stack[ssize].b,\
+    (_c) = stack[ssize].c, (_d) = stack[ssize].d, (_e) = stack[ssize].e;\
+  } while(0)
+#define BUCKET_A(_c0) bucket_A[(_c0)]
+#if ALPHABET_SIZE == 256
+#define BUCKET_B(_c0, _c1) (bucket_B[((_c1) << 8) | (_c0)])
+#define BUCKET_BSTAR(_c0, _c1) (bucket_B[((_c0) << 8) | (_c1)])
+#else
+#define BUCKET_B(_c0, _c1) (bucket_B[(_c1) * ALPHABET_SIZE + (_c0)])
+#define BUCKET_BSTAR(_c0, _c1) (bucket_B[(_c0) * ALPHABET_SIZE + (_c1)])
+#endif
+
+
+/*- Private Functions -*/
+
+static const int lg_table[256]= {
+ -1,0,1,1,2,2,2,2,3,3,3,3,3,3,3,3,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,
+  5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
+  6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
+  6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7
+};
+
+#if (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE)
+
+static INLINE
+int
+ss_ilg(int n) {
+#if SS_BLOCKSIZE == 0
+  return (n & 0xffff0000) ?
+          ((n & 0xff000000) ?
+            24 + lg_table[(n >> 24) & 0xff] :
+            16 + lg_table[(n >> 16) & 0xff]) :
+          ((n & 0x0000ff00) ?
+             8 + lg_table[(n >>  8) & 0xff] :
+             0 + lg_table[(n >>  0) & 0xff]);
+#elif SS_BLOCKSIZE < 256
+  return lg_table[n];
+#else
+  return (n & 0xff00) ?
+          8 + lg_table[(n >> 8) & 0xff] :
+          0 + lg_table[(n >> 0) & 0xff];
+#endif
+}
+
+#endif /* (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE) */
+
+#if SS_BLOCKSIZE != 0
+
+static const int sqq_table[256] = {
+  0,  16,  22,  27,  32,  35,  39,  42,  45,  48,  50,  53,  55,  57,  59,  61,
+ 64,  65,  67,  69,  71,  73,  75,  76,  78,  80,  81,  83,  84,  86,  87,  89,
+ 90,  91,  93,  94,  96,  97,  98,  99, 101, 102, 103, 104, 106, 107, 108, 109,
+110, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126,
+128, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142,
+143, 144, 144, 145, 146, 147, 148, 149, 150, 150, 151, 152, 153, 154, 155, 155,
+156, 157, 158, 159, 160, 160, 161, 162, 163, 163, 164, 165, 166, 167, 167, 168,
+169, 170, 170, 171, 172, 173, 173, 174, 175, 176, 176, 177, 178, 178, 179, 180,
+181, 181, 182, 183, 183, 184, 185, 185, 186, 187, 187, 188, 189, 189, 190, 191,
+192, 192, 193, 193, 194, 195, 195, 196, 197, 197, 198, 199, 199, 200, 201, 201,
+202, 203, 203, 204, 204, 205, 206, 206, 207, 208, 208, 209, 209, 210, 211, 211,
+212, 212, 213, 214, 214, 215, 215, 216, 217, 217, 218, 218, 219, 219, 220, 221,
+221, 222, 222, 223, 224, 224, 225, 225, 226, 226, 227, 227, 228, 229, 229, 230,
+230, 231, 231, 232, 232, 233, 234, 234, 235, 235, 236, 236, 237, 237, 238, 238,
+239, 240, 240, 241, 241, 242, 242, 243, 243, 244, 244, 245, 245, 246, 246, 247,
+247, 248, 248, 249, 249, 250, 250, 251, 251, 252, 252, 253, 253, 254, 254, 255
+};
+
+static INLINE
+int
+ss_isqrt(int x) {
+  int y, e;
+
+  if(x >= (SS_BLOCKSIZE * SS_BLOCKSIZE)) { return SS_BLOCKSIZE; }
+  e = ((unsigned)x & 0xffff0000) ?
+        (((unsigned)x & 0xff000000) ?
+          24 + lg_table[(x >> 24) & 0xff] :
+          16 + lg_table[(x >> 16) & 0xff]) :
+        ((x & 0x0000ff00) ?
+           8 + lg_table[(x >>  8) & 0xff] :
+           0 + lg_table[(x >>  0) & 0xff]);
+
+  if(e >= 16) {
+    y = sqq_table[x >> ((e - 6) - (e & 1))] << ((e >> 1) - 7);
+    if(e >= 24) { y = (y + 1 + x / y) >> 1; }
+    y = (y + 1 + x / y) >> 1;
+  } else if(e >= 8) {
+    y = (sqq_table[x >> ((e - 6) - (e & 1))] >> (7 - (e >> 1))) + 1;
+  } else {
+    return sqq_table[x] >> 4;
+  }
+
+  return (x < (y * y)) ? y - 1 : y;
+}
+
+#endif /* SS_BLOCKSIZE != 0 */
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Compares two suffixes. */
+static INLINE
+int
+ss_compare(const unsigned char *T,
+           const int *p1, const int *p2,
+           int depth) {
+  const unsigned char *U1, *U2, *U1n, *U2n;
+
+  for(U1 = T + depth + *p1,
+      U2 = T + depth + *p2,
+      U1n = T + *(p1 + 1) + 2,
+      U2n = T + *(p2 + 1) + 2;
+      (U1 < U1n) && (U2 < U2n) && (*U1 == *U2);
+      ++U1, ++U2) {
+  }
+
+  return U1 < U1n ?
+        (U2 < U2n ? *U1 - *U2 : 1) :
+        (U2 < U2n ? -1 : 0);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+#if (SS_BLOCKSIZE != 1) && (SS_INSERTIONSORT_THRESHOLD != 1)
+
+/* Insertionsort for small size groups */
+static
+void
+ss_insertionsort(const unsigned char *T, const int *PA,
+                 int *first, int *last, int depth) {
+  int *i, *j;
+  int t;
+  int r;
+
+  for(i = last - 2; first <= i; --i) {
+    for(t = *i, j = i + 1; 0 < (r = ss_compare(T, PA + t, PA + *j, depth));) {
+      do { *(j - 1) = *j; } while((++j < last) && (*j < 0));
+      if(last <= j) { break; }
+    }
+    if(r == 0) { *j = ~*j; }
+    *(j - 1) = t;
+  }
+}
+
+#endif /* (SS_BLOCKSIZE != 1) && (SS_INSERTIONSORT_THRESHOLD != 1) */
+
+
+/*---------------------------------------------------------------------------*/
+
+#if (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE)
+
+static INLINE
+void
+ss_fixdown(const unsigned char *Td, const int *PA,
+           int *SA, int i, int size) {
+  int j, k;
+  int v;
+  int c, d, e;
+
+  for(v = SA[i], c = Td[PA[v]]; (j = 2 * i + 1) < size; SA[i] = SA[k], i = k) {
+    d = Td[PA[SA[k = j++]]];
+    if(d < (e = Td[PA[SA[j]]])) { k = j; d = e; }
+    if(d <= c) { break; }
+  }
+  SA[i] = v;
+}
+
+/* Simple top-down heapsort. */
+static
+void
+ss_heapsort(const unsigned char *Td, const int *PA, int *SA, int size) {
+  int i, m;
+  int t;
+
+  m = size;
+  if((size % 2) == 0) {
+    m--;
+    if(Td[PA[SA[m / 2]]] < Td[PA[SA[m]]]) { SWAP(SA[m], SA[m / 2]); }
+  }
+
+  for(i = m / 2 - 1; 0 <= i; --i) { ss_fixdown(Td, PA, SA, i, m); }
+  if((size % 2) == 0) { SWAP(SA[0], SA[m]); ss_fixdown(Td, PA, SA, 0, m); }
+  for(i = m - 1; 0 < i; --i) {
+    t = SA[0], SA[0] = SA[i];
+    ss_fixdown(Td, PA, SA, 0, i);
+    SA[i] = t;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Returns the median of three elements. */
+static INLINE
+int *
+ss_median3(const unsigned char *Td, const int *PA,
+           int *v1, int *v2, int *v3) {
+  int *t;
+  if(Td[PA[*v1]] > Td[PA[*v2]]) { SWAP(v1, v2); }
+  if(Td[PA[*v2]] > Td[PA[*v3]]) {
+    if(Td[PA[*v1]] > Td[PA[*v3]]) { return v1; }
+    else { return v3; }
+  }
+  return v2;
+}
+
+/* Returns the median of five elements. */
+static INLINE
+int *
+ss_median5(const unsigned char *Td, const int *PA,
+           int *v1, int *v2, int *v3, int *v4, int *v5) {
+  int *t;
+  if(Td[PA[*v2]] > Td[PA[*v3]]) { SWAP(v2, v3); }
+  if(Td[PA[*v4]] > Td[PA[*v5]]) { SWAP(v4, v5); }
+  if(Td[PA[*v2]] > Td[PA[*v4]]) { SWAP(v2, v4); SWAP(v3, v5); }
+  if(Td[PA[*v1]] > Td[PA[*v3]]) { SWAP(v1, v3); }
+  if(Td[PA[*v1]] > Td[PA[*v4]]) { SWAP(v1, v4); SWAP(v3, v5); }
+  if(Td[PA[*v3]] > Td[PA[*v4]]) { return v4; }
+  return v3;
+}
+
+/* Returns the pivot element. */
+static INLINE
+int *
+ss_pivot(const unsigned char *Td, const int *PA, int *first, int *last) {
+  int *middle;
+  int t;
+
+  t = last - first;
+  middle = first + t / 2;
+
+  if(t <= 512) {
+    if(t <= 32) {
+      return ss_median3(Td, PA, first, middle, last - 1);
+    } else {
+      t >>= 2;
+      return ss_median5(Td, PA, first, first + t, middle, last - 1 - t, last - 1);
+    }
+  }
+  t >>= 3;
+  first  = ss_median3(Td, PA, first, first + t, first + (t << 1));
+  middle = ss_median3(Td, PA, middle - t, middle, middle + t);
+  last   = ss_median3(Td, PA, last - 1 - (t << 1), last - 1 - t, last - 1);
+  return ss_median3(Td, PA, first, middle, last);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Binary partition for substrings. */
+static INLINE
+int *
+ss_partition(const int *PA,
+                    int *first, int *last, int depth) {
+  int *a, *b;
+  int t;
+  for(a = first - 1, b = last;;) {
+    for(; (++a < b) && ((PA[*a] + depth) >= (PA[*a + 1] + 1));) { *a = ~*a; }
+    for(; (a < --b) && ((PA[*b] + depth) <  (PA[*b + 1] + 1));) { }
+    if(b <= a) { break; }
+    t = ~*b;
+    *b = *a;
+    *a = t;
+  }
+  if(first < a) { *first = ~*first; }
+  return a;
+}
+
+/* Multikey introsort for medium size groups. */
+static
+void
+ss_mintrosort(const unsigned char *T, const int *PA,
+              int *first, int *last,
+              int depth) {
+#define STACK_SIZE SS_MISORT_STACKSIZE
+  struct { int *a, *b, c; int d; } stack[STACK_SIZE];
+  const unsigned char *Td;
+  int *a, *b, *c, *d, *e, *f;
+  int s, t;
+  int ssize;
+  int limit;
+  int v, x = 0;
+
+  for(ssize = 0, limit = ss_ilg(last - first);;) {
+
+    if((last - first) <= SS_INSERTIONSORT_THRESHOLD) {
+#if 1 < SS_INSERTIONSORT_THRESHOLD
+      if(1 < (last - first)) { ss_insertionsort(T, PA, first, last, depth); }
+#endif
+      STACK_POP(first, last, depth, limit);
+      continue;
+    }
+
+    Td = T + depth;
+    if(limit-- == 0) { ss_heapsort(Td, PA, first, last - first); }
+    if(limit < 0) {
+      for(a = first + 1, v = Td[PA[*first]]; a < last; ++a) {
+        if((x = Td[PA[*a]]) != v) {
+          if(1 < (a - first)) { break; }
+          v = x;
+          first = a;
+        }
+      }
+      if(Td[PA[*first] - 1] < v) {
+        first = ss_partition(PA, first, a, depth);
+      }
+      if((a - first) <= (last - a)) {
+        if(1 < (a - first)) {
+          STACK_PUSH(a, last, depth, -1);
+          last = a, depth += 1, limit = ss_ilg(a - first);
+        } else {
+          first = a, limit = -1;
+        }
+      } else {
+        if(1 < (last - a)) {
+          STACK_PUSH(first, a, depth + 1, ss_ilg(a - first));
+          first = a, limit = -1;
+        } else {
+          last = a, depth += 1, limit = ss_ilg(a - first);
+        }
+      }
+      continue;
+    }
+
+    /* choose pivot */
+    a = ss_pivot(Td, PA, first, last);
+    v = Td[PA[*a]];
+    SWAP(*first, *a);
+
+    /* partition */
+    for(b = first; (++b < last) && ((x = Td[PA[*b]]) == v);) { }
+    if(((a = b) < last) && (x < v)) {
+      for(; (++b < last) && ((x = Td[PA[*b]]) <= v);) {
+        if(x == v) { SWAP(*b, *a); ++a; }
+      }
+    }
+    for(c = last; (b < --c) && ((x = Td[PA[*c]]) == v);) { }
+    if((b < (d = c)) && (x > v)) {
+      for(; (b < --c) && ((x = Td[PA[*c]]) >= v);) {
+        if(x == v) { SWAP(*c, *d); --d; }
+      }
+    }
+    for(; b < c;) {
+      SWAP(*b, *c);
+      for(; (++b < c) && ((x = Td[PA[*b]]) <= v);) {
+        if(x == v) { SWAP(*b, *a); ++a; }
+      }
+      for(; (b < --c) && ((x = Td[PA[*c]]) >= v);) {
+        if(x == v) { SWAP(*c, *d); --d; }
+      }
+    }
+
+    if(a <= d) {
+      c = b - 1;
+
+      if((s = a - first) > (t = b - a)) { s = t; }
+      for(e = first, f = b - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+      if((s = d - c) > (t = last - d - 1)) { s = t; }
+      for(e = b, f = last - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+
+      a = first + (b - a), c = last - (d - c);
+      b = (v <= Td[PA[*a] - 1]) ? a : ss_partition(PA, a, c, depth);
+
+      if((a - first) <= (last - c)) {
+        if((last - c) <= (c - b)) {
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          STACK_PUSH(c, last, depth, limit);
+          last = a;
+        } else if((a - first) <= (c - b)) {
+          STACK_PUSH(c, last, depth, limit);
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          last = a;
+        } else {
+          STACK_PUSH(c, last, depth, limit);
+          STACK_PUSH(first, a, depth, limit);
+          first = b, last = c, depth += 1, limit = ss_ilg(c - b);
+        }
+      } else {
+        if((a - first) <= (c - b)) {
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          STACK_PUSH(first, a, depth, limit);
+          first = c;
+        } else if((last - c) <= (c - b)) {
+          STACK_PUSH(first, a, depth, limit);
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          first = c;
+        } else {
+          STACK_PUSH(first, a, depth, limit);
+          STACK_PUSH(c, last, depth, limit);
+          first = b, last = c, depth += 1, limit = ss_ilg(c - b);
+        }
+      }
+    } else {
+      limit += 1;
+      if(Td[PA[*first] - 1] < v) {
+        first = ss_partition(PA, first, last, depth);
+        limit = ss_ilg(last - first);
+      }
+      depth += 1;
+    }
+  }
+#undef STACK_SIZE
+}
+
+#endif /* (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE) */
+
+
+/*---------------------------------------------------------------------------*/
+
+#if SS_BLOCKSIZE != 0
+
+static INLINE
+void
+ss_blockswap(int *a, int *b, int n) {
+  int t;
+  for(; 0 < n; --n, ++a, ++b) {
+    t = *a, *a = *b, *b = t;
+  }
+}
+
+static INLINE
+void
+ss_rotate(int *first, int *middle, int *last) {
+  int *a, *b, t;
+  int l, r;
+  l = middle - first, r = last - middle;
+  for(; (0 < l) && (0 < r);) {
+    if(l == r) { ss_blockswap(first, middle, l); break; }
+    if(l < r) {
+      a = last - 1, b = middle - 1;
+      t = *a;
+      do {
+        *a-- = *b, *b-- = *a;
+        if(b < first) {
+          *a = t;
+          last = a;
+          if((r -= l + 1) <= l) { break; }
+          a -= 1, b = middle - 1;
+          t = *a;
+        }
+      } while(1);
+    } else {
+      a = first, b = middle;
+      t = *a;
+      do {
+        *a++ = *b, *b++ = *a;
+        if(last <= b) {
+          *a = t;
+          first = a + 1;
+          if((l -= r + 1) <= r) { break; }
+          a += 1, b = middle;
+          t = *a;
+        }
+      } while(1);
+    }
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static
+void
+ss_inplacemerge(const unsigned char *T, const int *PA,
+                int *first, int *middle, int *last,
+                int depth) {
+  const int *p;
+  int *a, *b;
+  int len, half;
+  int q, r;
+  int x;
+
+  for(;;) {
+    if(*(last - 1) < 0) { x = 1; p = PA + ~*(last - 1); }
+    else                { x = 0; p = PA +  *(last - 1); }
+    for(a = first, len = middle - first, half = len >> 1, r = -1;
+        0 < len;
+        len = half, half >>= 1) {
+      b = a + half;
+      q = ss_compare(T, PA + ((0 <= *b) ? *b : ~*b), p, depth);
+      if(q < 0) {
+        a = b + 1;
+        half -= (len & 1) ^ 1;
+      } else {
+        r = q;
+      }
+    }
+    if(a < middle) {
+      if(r == 0) { *a = ~*a; }
+      ss_rotate(a, middle, last);
+      last -= middle - a;
+      middle = a;
+      if(first == middle) { break; }
+    }
+    --last;
+    if(x != 0) { while(*--last < 0) { } }
+    if(middle == last) { break; }
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Merge-forward with internal buffer. */
+static
+void
+ss_mergeforward(const unsigned char *T, const int *PA,
+                int *first, int *middle, int *last,
+                int *buf, int depth) {
+  int *a, *b, *c, *bufend;
+  int t;
+  int r;
+
+  bufend = buf + (middle - first) - 1;
+  ss_blockswap(buf, first, middle - first);
+
+  for(t = *(a = first), b = buf, c = middle;;) {
+    r = ss_compare(T, PA + *b, PA + *c, depth);
+    if(r < 0) {
+      do {
+        *a++ = *b;
+        if(bufend <= b) { *bufend = t; return; }
+        *b++ = *a;
+      } while(*b < 0);
+    } else if(r > 0) {
+      do {
+        *a++ = *c, *c++ = *a;
+        if(last <= c) {
+          while(b < bufend) { *a++ = *b, *b++ = *a; }
+          *a = *b, *b = t;
+          return;
+        }
+      } while(*c < 0);
+    } else {
+      *c = ~*c;
+      do {
+        *a++ = *b;
+        if(bufend <= b) { *bufend = t; return; }
+        *b++ = *a;
+      } while(*b < 0);
+
+      do {
+        *a++ = *c, *c++ = *a;
+        if(last <= c) {
+          while(b < bufend) { *a++ = *b, *b++ = *a; }
+          *a = *b, *b = t;
+          return;
+        }
+      } while(*c < 0);
+    }
+  }
+}
+
+/* Merge-backward with internal buffer. */
+static
+void
+ss_mergebackward(const unsigned char *T, const int *PA,
+                 int *first, int *middle, int *last,
+                 int *buf, int depth) {
+  const int *p1, *p2;
+  int *a, *b, *c, *bufend;
+  int t;
+  int r;
+  int x;
+
+  bufend = buf + (last - middle) - 1;
+  ss_blockswap(buf, middle, last - middle);
+
+  x = 0;
+  if(*bufend < 0)       { p1 = PA + ~*bufend; x |= 1; }
+  else                  { p1 = PA +  *bufend; }
+  if(*(middle - 1) < 0) { p2 = PA + ~*(middle - 1); x |= 2; }
+  else                  { p2 = PA +  *(middle - 1); }
+  for(t = *(a = last - 1), b = bufend, c = middle - 1;;) {
+    r = ss_compare(T, p1, p2, depth);
+    if(0 < r) {
+      if(x & 1) { do { *a-- = *b, *b-- = *a; } while(*b < 0); x ^= 1; }
+      *a-- = *b;
+      if(b <= buf) { *buf = t; break; }
+      *b-- = *a;
+      if(*b < 0) { p1 = PA + ~*b; x |= 1; }
+      else       { p1 = PA +  *b; }
+    } else if(r < 0) {
+      if(x & 2) { do { *a-- = *c, *c-- = *a; } while(*c < 0); x ^= 2; }
+      *a-- = *c, *c-- = *a;
+      if(c < first) {
+        while(buf < b) { *a-- = *b, *b-- = *a; }
+        *a = *b, *b = t;
+        break;
+      }
+      if(*c < 0) { p2 = PA + ~*c; x |= 2; }
+      else       { p2 = PA +  *c; }
+    } else {
+      if(x & 1) { do { *a-- = *b, *b-- = *a; } while(*b < 0); x ^= 1; }
+      *a-- = ~*b;
+      if(b <= buf) { *buf = t; break; }
+      *b-- = *a;
+      if(x & 2) { do { *a-- = *c, *c-- = *a; } while(*c < 0); x ^= 2; }
+      *a-- = *c, *c-- = *a;
+      if(c < first) {
+        while(buf < b) { *a-- = *b, *b-- = *a; }
+        *a = *b, *b = t;
+        break;
+      }
+      if(*b < 0) { p1 = PA + ~*b; x |= 1; }
+      else       { p1 = PA +  *b; }
+      if(*c < 0) { p2 = PA + ~*c; x |= 2; }
+      else       { p2 = PA +  *c; }
+    }
+  }
+}
+
+/* D&C based merge. */
+static
+void
+ss_swapmerge(const unsigned char *T, const int *PA,
+             int *first, int *middle, int *last,
+             int *buf, int bufsize, int depth) {
+#define STACK_SIZE SS_SMERGE_STACKSIZE
+#define GETIDX(a) ((0 <= (a)) ? (a) : (~(a)))
+#define MERGE_CHECK(a, b, c)\
+  do {\
+    if(((c) & 1) ||\
+       (((c) & 2) && (ss_compare(T, PA + GETIDX(*((a) - 1)), PA + *(a), depth) == 0))) {\
+      *(a) = ~*(a);\
+    }\
+    if(((c) & 4) && ((ss_compare(T, PA + GETIDX(*((b) - 1)), PA + *(b), depth) == 0))) {\
+      *(b) = ~*(b);\
+    }\
+  } while(0)
+  struct { int *a, *b, *c; int d; } stack[STACK_SIZE];
+  int *l, *r, *lm, *rm;
+  int m, len, half;
+  int ssize;
+  int check, next;
+
+  for(check = 0, ssize = 0;;) {
+    if((last - middle) <= bufsize) {
+      if((first < middle) && (middle < last)) {
+        ss_mergebackward(T, PA, first, middle, last, buf, depth);
+      }
+      MERGE_CHECK(first, last, check);
+      STACK_POP(first, middle, last, check);
+      continue;
+    }
+
+    if((middle - first) <= bufsize) {
+      if(first < middle) {
+        ss_mergeforward(T, PA, first, middle, last, buf, depth);
+      }
+      MERGE_CHECK(first, last, check);
+      STACK_POP(first, middle, last, check);
+      continue;
+    }
+
+    for(m = 0, len = MIN(middle - first, last - middle), half = len >> 1;
+        0 < len;
+        len = half, half >>= 1) {
+      if(ss_compare(T, PA + GETIDX(*(middle + m + half)),
+                       PA + GETIDX(*(middle - m - half - 1)), depth) < 0) {
+        m += half + 1;
+        half -= (len & 1) ^ 1;
+      }
+    }
+
+    if(0 < m) {
+      lm = middle - m, rm = middle + m;
+      ss_blockswap(lm, middle, m);
+      l = r = middle, next = 0;
+      if(rm < last) {
+        if(*rm < 0) {
+          *rm = ~*rm;
+          if(first < lm) { for(; *--l < 0;) { } next |= 4; }
+          next |= 1;
+        } else if(first < lm) {
+          for(; *r < 0; ++r) { }
+          next |= 2;
+        }
+      }
+
+      if((l - first) <= (last - r)) {
+        STACK_PUSH(r, rm, last, (next & 3) | (check & 4));
+        middle = lm, last = l, check = (check & 3) | (next & 4);
+      } else {
+        if((next & 2) && (r == middle)) { next ^= 6; }
+        STACK_PUSH(first, lm, l, (check & 3) | (next & 4));
+        first = r, middle = rm, check = (next & 3) | (check & 4);
+      }
+    } else {
+      if(ss_compare(T, PA + GETIDX(*(middle - 1)), PA + *middle, depth) == 0) {
+        *middle = ~*middle;
+      }
+      MERGE_CHECK(first, last, check);
+      STACK_POP(first, middle, last, check);
+    }
+  }
+#undef STACK_SIZE
+}
+
+#endif /* SS_BLOCKSIZE != 0 */
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Substring sort */
+static
+void
+sssort(const unsigned char *T, const int *PA,
+       int *first, int *last,
+       int *buf, int bufsize,
+       int depth, int n, int lastsuffix) {
+  int *a;
+#if SS_BLOCKSIZE != 0
+  int *b, *middle, *curbuf;
+  int j, k, curbufsize, limit;
+#endif
+  int i;
+
+  if(lastsuffix != 0) { ++first; }
+
+#if SS_BLOCKSIZE == 0
+  ss_mintrosort(T, PA, first, last, depth);
+#else
+  if((bufsize < SS_BLOCKSIZE) &&
+      (bufsize < (last - first)) &&
+      (bufsize < (limit = ss_isqrt(last - first)))) {
+    if(SS_BLOCKSIZE < limit) { limit = SS_BLOCKSIZE; }
+    buf = middle = last - limit, bufsize = limit;
+  } else {
+    middle = last, limit = 0;
+  }
+  for(a = first, i = 0; SS_BLOCKSIZE < (middle - a); a += SS_BLOCKSIZE, ++i) {
+#if SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE
+    ss_mintrosort(T, PA, a, a + SS_BLOCKSIZE, depth);
+#elif 1 < SS_BLOCKSIZE
+    ss_insertionsort(T, PA, a, a + SS_BLOCKSIZE, depth);
+#endif
+    curbufsize = last - (a + SS_BLOCKSIZE);
+    curbuf = a + SS_BLOCKSIZE;
+    if(curbufsize <= bufsize) { curbufsize = bufsize, curbuf = buf; }
+    for(b = a, k = SS_BLOCKSIZE, j = i; j & 1; b -= k, k <<= 1, j >>= 1) {
+      ss_swapmerge(T, PA, b - k, b, b + k, curbuf, curbufsize, depth);
+    }
+  }
+#if SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE
+  ss_mintrosort(T, PA, a, middle, depth);
+#elif 1 < SS_BLOCKSIZE
+  ss_insertionsort(T, PA, a, middle, depth);
+#endif
+  for(k = SS_BLOCKSIZE; i != 0; k <<= 1, i >>= 1) {
+    if(i & 1) {
+      ss_swapmerge(T, PA, a - k, a, middle, buf, bufsize, depth);
+      a -= k;
+    }
+  }
+  if(limit != 0) {
+#if SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE
+    ss_mintrosort(T, PA, middle, last, depth);
+#elif 1 < SS_BLOCKSIZE
+    ss_insertionsort(T, PA, middle, last, depth);
+#endif
+    ss_inplacemerge(T, PA, first, middle, last, depth);
+  }
+#endif
+
+  if(lastsuffix != 0) {
+    /* Insert last type B* suffix. */
+    int PAi[2]; PAi[0] = PA[*(first - 1)], PAi[1] = n - 2;
+    for(a = first, i = *(first - 1);
+        (a < last) && ((*a < 0) || (0 < ss_compare(T, &(PAi[0]), PA + *a, depth)));
+        ++a) {
+      *(a - 1) = *a;
+    }
+    *(a - 1) = i;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static INLINE
+int
+tr_ilg(int n) {
+  return ((unsigned)n & 0xffff0000) ?
+          (((unsigned)n & 0xff000000) ?
+            24 + lg_table[(n >> 24) & 0xff] :
+            16 + lg_table[(n >> 16) & 0xff]) :
+          ((n & 0x0000ff00) ?
+             8 + lg_table[(n >>  8) & 0xff] :
+             0 + lg_table[(n >>  0) & 0xff]);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Simple insertionsort for small size groups. */
+static
+void
+tr_insertionsort(const int *ISAd, int *first, int *last) {
+  int *a, *b;
+  int t, r;
+
+  for(a = first + 1; a < last; ++a) {
+    for(t = *a, b = a - 1; 0 > (r = ISAd[t] - ISAd[*b]);) {
+      do { *(b + 1) = *b; } while((first <= --b) && (*b < 0));
+      if(b < first) { break; }
+    }
+    if(r == 0) { *b = ~*b; }
+    *(b + 1) = t;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static INLINE
+void
+tr_fixdown(const int *ISAd, int *SA, int i, int size) {
+  int j, k;
+  int v;
+  int c, d, e;
+
+  for(v = SA[i], c = ISAd[v]; (j = 2 * i + 1) < size; SA[i] = SA[k], i = k) {
+    d = ISAd[SA[k = j++]];
+    if(d < (e = ISAd[SA[j]])) { k = j; d = e; }
+    if(d <= c) { break; }
+  }
+  SA[i] = v;
+}
+
+/* Simple top-down heapsort. */
+static
+void
+tr_heapsort(const int *ISAd, int *SA, int size) {
+  int i, m;
+  int t;
+
+  m = size;
+  if((size % 2) == 0) {
+    m--;
+    if(ISAd[SA[m / 2]] < ISAd[SA[m]]) { SWAP(SA[m], SA[m / 2]); }
+  }
+
+  for(i = m / 2 - 1; 0 <= i; --i) { tr_fixdown(ISAd, SA, i, m); }
+  if((size % 2) == 0) { SWAP(SA[0], SA[m]); tr_fixdown(ISAd, SA, 0, m); }
+  for(i = m - 1; 0 < i; --i) {
+    t = SA[0], SA[0] = SA[i];
+    tr_fixdown(ISAd, SA, 0, i);
+    SA[i] = t;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Returns the median of three elements. */
+static INLINE
+int *
+tr_median3(const int *ISAd, int *v1, int *v2, int *v3) {
+  int *t;
+  if(ISAd[*v1] > ISAd[*v2]) { SWAP(v1, v2); }
+  if(ISAd[*v2] > ISAd[*v3]) {
+    if(ISAd[*v1] > ISAd[*v3]) { return v1; }
+    else { return v3; }
+  }
+  return v2;
+}
+
+/* Returns the median of five elements. */
+static INLINE
+int *
+tr_median5(const int *ISAd,
+           int *v1, int *v2, int *v3, int *v4, int *v5) {
+  int *t;
+  if(ISAd[*v2] > ISAd[*v3]) { SWAP(v2, v3); }
+  if(ISAd[*v4] > ISAd[*v5]) { SWAP(v4, v5); }
+  if(ISAd[*v2] > ISAd[*v4]) { SWAP(v2, v4); SWAP(v3, v5); }
+  if(ISAd[*v1] > ISAd[*v3]) { SWAP(v1, v3); }
+  if(ISAd[*v1] > ISAd[*v4]) { SWAP(v1, v4); SWAP(v3, v5); }
+  if(ISAd[*v3] > ISAd[*v4]) { return v4; }
+  return v3;
+}
+
+/* Returns the pivot element. */
+static INLINE
+int *
+tr_pivot(const int *ISAd, int *first, int *last) {
+  int *middle;
+  int t;
+
+  t = last - first;
+  middle = first + t / 2;
+
+  if(t <= 512) {
+    if(t <= 32) {
+      return tr_median3(ISAd, first, middle, last - 1);
+    } else {
+      t >>= 2;
+      return tr_median5(ISAd, first, first + t, middle, last - 1 - t, last - 1);
+    }
+  }
+  t >>= 3;
+  first  = tr_median3(ISAd, first, first + t, first + (t << 1));
+  middle = tr_median3(ISAd, middle - t, middle, middle + t);
+  last   = tr_median3(ISAd, last - 1 - (t << 1), last - 1 - t, last - 1);
+  return tr_median3(ISAd, first, middle, last);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+typedef struct _trbudget_t trbudget_t;
+struct _trbudget_t {
+  int chance;
+  int remain;
+  int incval;
+  int count;
+};
+
+static INLINE
+void
+trbudget_init(trbudget_t *budget, int chance, int incval) {
+  budget->chance = chance;
+  budget->remain = budget->incval = incval;
+}
+
+static INLINE
+int
+trbudget_check(trbudget_t *budget, int size) {
+  if(size <= budget->remain) { budget->remain -= size; return 1; }
+  if(budget->chance == 0) { budget->count += size; return 0; }
+  budget->remain += budget->incval - size;
+  budget->chance -= 1;
+  return 1;
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static INLINE
+void
+tr_partition(const int *ISAd,
+             int *first, int *middle, int *last,
+             int **pa, int **pb, int v) {
+  int *a, *b, *c, *d, *e, *f;
+  int t, s;
+  int x = 0;
+
+  for(b = middle - 1; (++b < last) && ((x = ISAd[*b]) == v);) { }
+  if(((a = b) < last) && (x < v)) {
+    for(; (++b < last) && ((x = ISAd[*b]) <= v);) {
+      if(x == v) { SWAP(*b, *a); ++a; }
+    }
+  }
+  for(c = last; (b < --c) && ((x = ISAd[*c]) == v);) { }
+  if((b < (d = c)) && (x > v)) {
+    for(; (b < --c) && ((x = ISAd[*c]) >= v);) {
+      if(x == v) { SWAP(*c, *d); --d; }
+    }
+  }
+  for(; b < c;) {
+    SWAP(*b, *c);
+    for(; (++b < c) && ((x = ISAd[*b]) <= v);) {
+      if(x == v) { SWAP(*b, *a); ++a; }
+    }
+    for(; (b < --c) && ((x = ISAd[*c]) >= v);) {
+      if(x == v) { SWAP(*c, *d); --d; }
+    }
+  }
+
+  if(a <= d) {
+    c = b - 1;
+    if((s = a - first) > (t = b - a)) { s = t; }
+    for(e = first, f = b - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+    if((s = d - c) > (t = last - d - 1)) { s = t; }
+    for(e = b, f = last - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+    first += (b - a), last -= (d - c);
+  }
+  *pa = first, *pb = last;
+}
+
+static
+void
+tr_copy(int *ISA, const int *SA,
+        int *first, int *a, int *b, int *last,
+        int depth) {
+  /* sort suffixes of middle partition
+     by using sorted order of suffixes of left and right partition. */
+  int *c, *d, *e;
+  int s, v;
+
+  v = b - SA - 1;
+  for(c = first, d = a - 1; c <= d; ++c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *++d = s;
+      ISA[s] = d - SA;
+    }
+  }
+  for(c = last - 1, e = d + 1, d = b; e < d; --c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *--d = s;
+      ISA[s] = d - SA;
+    }
+  }
+}
+
+static
+void
+tr_partialcopy(int *ISA, const int *SA,
+               int *first, int *a, int *b, int *last,
+               int depth) {
+  int *c, *d, *e;
+  int s, v;
+  int rank, lastrank, newrank = -1;
+
+  v = b - SA - 1;
+  lastrank = -1;
+  for(c = first, d = a - 1; c <= d; ++c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *++d = s;
+      rank = ISA[s + depth];
+      if(lastrank != rank) { lastrank = rank; newrank = d - SA; }
+      ISA[s] = newrank;
+    }
+  }
+
+  lastrank = -1;
+  for(e = d; first <= e; --e) {
+    rank = ISA[*e];
+    if(lastrank != rank) { lastrank = rank; newrank = e - SA; }
+    if(newrank != rank) { ISA[*e] = newrank; }
+  }
+
+  lastrank = -1;
+  for(c = last - 1, e = d + 1, d = b; e < d; --c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *--d = s;
+      rank = ISA[s + depth];
+      if(lastrank != rank) { lastrank = rank; newrank = d - SA; }
+      ISA[s] = newrank;
+    }
+  }
+}
+
+static
+void
+tr_introsort(int *ISA, const int *ISAd,
+             int *SA, int *first, int *last,
+             trbudget_t *budget) {
+#define STACK_SIZE TR_STACKSIZE
+  struct { const int *a; int *b, *c; int d, e; }stack[STACK_SIZE];
+  int *a, *b, *c;
+  int t;
+  int v, x = 0;
+  int incr = ISAd - ISA;
+  int limit, next;
+  int ssize, trlink = -1;
+
+  for(ssize = 0, limit = tr_ilg(last - first);;) {
+
+    if(limit < 0) {
+      if(limit == -1) {
+        /* tandem repeat partition */
+        tr_partition(ISAd - incr, first, first, last, &a, &b, last - SA - 1);
+
+        /* update ranks */
+        if(a < last) {
+          for(c = first, v = a - SA - 1; c < a; ++c) { ISA[*c] = v; }
+        }
+        if(b < last) {
+          for(c = a, v = b - SA - 1; c < b; ++c) { ISA[*c] = v; }
+        }
+
+        /* push */
+        if(1 < (b - a)) {
+          STACK_PUSH5(NULL, a, b, 0, 0);
+          STACK_PUSH5(ISAd - incr, first, last, -2, trlink);
+          trlink = ssize - 2;
+        }
+        if((a - first) <= (last - b)) {
+          if(1 < (a - first)) {
+            STACK_PUSH5(ISAd, b, last, tr_ilg(last - b), trlink);
+            last = a, limit = tr_ilg(a - first);
+          } else if(1 < (last - b)) {
+            first = b, limit = tr_ilg(last - b);
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        } else {
+          if(1 < (last - b)) {
+            STACK_PUSH5(ISAd, first, a, tr_ilg(a - first), trlink);
+            first = b, limit = tr_ilg(last - b);
+          } else if(1 < (a - first)) {
+            last = a, limit = tr_ilg(a - first);
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        }
+      } else if(limit == -2) {
+        /* tandem repeat copy */
+        a = stack[--ssize].b, b = stack[ssize].c;
+        if(stack[ssize].d == 0) {
+          tr_copy(ISA, SA, first, a, b, last, ISAd - ISA);
+        } else {
+          if(0 <= trlink) { stack[trlink].d = -1; }
+          tr_partialcopy(ISA, SA, first, a, b, last, ISAd - ISA);
+        }
+        STACK_POP5(ISAd, first, last, limit, trlink);
+      } else {
+        /* sorted partition */
+        if(0 <= *first) {
+          a = first;
+          do { ISA[*a] = a - SA; } while((++a < last) && (0 <= *a));
+          first = a;
+        }
+        if(first < last) {
+          a = first; do { *a = ~*a; } while(*++a < 0);
+          next = (ISA[*a] != ISAd[*a]) ? tr_ilg(a - first + 1) : -1;
+          if(++a < last) { for(b = first, v = a - SA - 1; b < a; ++b) { ISA[*b] = v; } }
+
+          /* push */
+          if(trbudget_check(budget, a - first)) {
+            if((a - first) <= (last - a)) {
+              STACK_PUSH5(ISAd, a, last, -3, trlink);
+              ISAd += incr, last = a, limit = next;
+            } else {
+              if(1 < (last - a)) {
+                STACK_PUSH5(ISAd + incr, first, a, next, trlink);
+                first = a, limit = -3;
+              } else {
+                ISAd += incr, last = a, limit = next;
+              }
+            }
+          } else {
+            if(0 <= trlink) { stack[trlink].d = -1; }
+            if(1 < (last - a)) {
+              first = a, limit = -3;
+            } else {
+              STACK_POP5(ISAd, first, last, limit, trlink);
+            }
+          }
+        } else {
+          STACK_POP5(ISAd, first, last, limit, trlink);
+        }
+      }
+      continue;
+    }
+
+    if((last - first) <= TR_INSERTIONSORT_THRESHOLD) {
+      tr_insertionsort(ISAd, first, last);
+      limit = -3;
+      continue;
+    }
+
+    if(limit-- == 0) {
+      tr_heapsort(ISAd, first, last - first);
+      for(a = last - 1; first < a; a = b) {
+        for(x = ISAd[*a], b = a - 1; (first <= b) && (ISAd[*b] == x); --b) { *b = ~*b; }
+      }
+      limit = -3;
+      continue;
+    }
+
+    /* choose pivot */
+    a = tr_pivot(ISAd, first, last);
+    SWAP(*first, *a);
+    v = ISAd[*first];
+
+    /* partition */
+    tr_partition(ISAd, first, first + 1, last, &a, &b, v);
+    if((last - first) != (b - a)) {
+      next = (ISA[*a] != v) ? tr_ilg(b - a) : -1;
+
+      /* update ranks */
+      for(c = first, v = a - SA - 1; c < a; ++c) { ISA[*c] = v; }
+      if(b < last) { for(c = a, v = b - SA - 1; c < b; ++c) { ISA[*c] = v; } }
+
+      /* push */
+      if((1 < (b - a)) && (trbudget_check(budget, b - a))) {
+        if((a - first) <= (last - b)) {
+          if((last - b) <= (b - a)) {
+            if(1 < (a - first)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              last = a;
+            } else if(1 < (last - b)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              first = b;
+            } else {
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else if((a - first) <= (b - a)) {
+            if(1 < (a - first)) {
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              last = a;
+            } else {
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else {
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            ISAd += incr, first = a, last = b, limit = next;
+          }
+        } else {
+          if((a - first) <= (b - a)) {
+            if(1 < (last - b)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              first = b;
+            } else if(1 < (a - first)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              last = a;
+            } else {
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else if((last - b) <= (b - a)) {
+            if(1 < (last - b)) {
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              first = b;
+            } else {
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else {
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            ISAd += incr, first = a, last = b, limit = next;
+          }
+        }
+      } else {
+        if((1 < (b - a)) && (0 <= trlink)) { stack[trlink].d = -1; }
+        if((a - first) <= (last - b)) {
+          if(1 < (a - first)) {
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            last = a;
+          } else if(1 < (last - b)) {
+            first = b;
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        } else {
+          if(1 < (last - b)) {
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            first = b;
+          } else if(1 < (a - first)) {
+            last = a;
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        }
+      }
+    } else {
+      if(trbudget_check(budget, last - first)) {
+        limit = tr_ilg(last - first), ISAd += incr;
+      } else {
+        if(0 <= trlink) { stack[trlink].d = -1; }
+        STACK_POP5(ISAd, first, last, limit, trlink);
+      }
+    }
+  }
+#undef STACK_SIZE
+}
+
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Tandem repeat sort */
+static
+void
+trsort(int *ISA, int *SA, int n, int depth) {
+  int *ISAd;
+  int *first, *last;
+  trbudget_t budget;
+  int t, skip, unsorted;
+
+  trbudget_init(&budget, tr_ilg(n) * 2 / 3, n);
+/*  trbudget_init(&budget, tr_ilg(n) * 3 / 4, n); */
+  for(ISAd = ISA + depth; -n < *SA; ISAd += ISAd - ISA) {
+    first = SA;
+    skip = 0;
+    unsorted = 0;
+    do {
+      if((t = *first) < 0) { first -= t; skip += t; }
+      else {
+        if(skip != 0) { *(first + skip) = skip; skip = 0; }
+        last = SA + ISA[t] + 1;
+        if(1 < (last - first)) {
+          budget.count = 0;
+          tr_introsort(ISA, ISAd, SA, first, last, &budget);
+          if(budget.count != 0) { unsorted += budget.count; }
+          else { skip = first - last; }
+        } else if((last - first) == 1) {
+          skip = -1;
+        }
+        first = last;
+      }
+    } while(first < (SA + n));
+    if(skip != 0) { *(first + skip) = skip; }
+    if(unsorted == 0) { break; }
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Sorts suffixes of type B*. */
+static
+int
+sort_typeBstar(const unsigned char *T, int *SA,
+               int *bucket_A, int *bucket_B,
+               int n, int openMP) {
+  int *PAb, *ISAb, *buf;
+#ifdef LIBBSC_OPENMP
+  int *curbuf;
+  int l;
+#endif
+  int i, j, k, t, m, bufsize;
+  int c0, c1;
+#ifdef LIBBSC_OPENMP
+  int d0, d1;
+#endif
+  (void)openMP;
+
+  /* Initialize bucket arrays. */
+  for(i = 0; i < BUCKET_A_SIZE; ++i) { bucket_A[i] = 0; }
+  for(i = 0; i < BUCKET_B_SIZE; ++i) { bucket_B[i] = 0; }
+
+  /* Count the number of occurrences of the first one or two characters of each
+     type A, B and B* suffix. Moreover, store the beginning position of all
+     type B* suffixes into the array SA. */
+  for(i = n - 1, m = n, c0 = T[n - 1]; 0 <= i;) {
+    /* type A suffix. */
+    do { ++BUCKET_A(c1 = c0); } while((0 <= --i) && ((c0 = T[i]) >= c1));
+    if(0 <= i) {
+      /* type B* suffix. */
+      ++BUCKET_BSTAR(c0, c1);
+      SA[--m] = i;
+      /* type B suffix. */
+      for(--i, c1 = c0; (0 <= i) && ((c0 = T[i]) <= c1); --i, c1 = c0) {
+        ++BUCKET_B(c0, c1);
+      }
+    }
+  }
+  m = n - m;
+/*
+note:
+  A type B* suffix is lexicographically smaller than a type B suffix that
+  begins with the same first two characters.
+*/
+
+  /* Calculate the index of start/end point of each bucket. */
+  for(c0 = 0, i = 0, j = 0; c0 < ALPHABET_SIZE; ++c0) {
+    t = i + BUCKET_A(c0);
+    BUCKET_A(c0) = i + j; /* start point */
+    i = t + BUCKET_B(c0, c0);
+    for(c1 = c0 + 1; c1 < ALPHABET_SIZE; ++c1) {
+      j += BUCKET_BSTAR(c0, c1);
+      BUCKET_BSTAR(c0, c1) = j; /* end point */
+      i += BUCKET_B(c0, c1);
+    }
+  }
+
+  if(0 < m) {
+    /* Sort the type B* suffixes by their first two characters. */
+    PAb = SA + n - m; ISAb = SA + m;
+    for(i = m - 2; 0 <= i; --i) {
+      t = PAb[i], c0 = T[t], c1 = T[t + 1];
+      SA[--BUCKET_BSTAR(c0, c1)] = i;
+    }
+    t = PAb[m - 1], c0 = T[t], c1 = T[t + 1];
+    SA[--BUCKET_BSTAR(c0, c1)] = m - 1;
+
+    /* Sort the type B* substrings using sssort. */
+#ifdef LIBBSC_OPENMP
+    if (openMP)
+    {
+        buf = SA + m;
+        c0 = ALPHABET_SIZE - 2, c1 = ALPHABET_SIZE - 1, j = m;
+#pragma omp parallel default(shared) private(bufsize, curbuf, k, l, d0, d1)
+        {
+          bufsize = (n - (2 * m)) / omp_get_num_threads();
+          curbuf = buf + omp_get_thread_num() * bufsize;
+          k = 0;
+          for(;;) {
+            #pragma omp critical(sssort_lock)
+            {
+              if(0 < (l = j)) {
+                d0 = c0, d1 = c1;
+                do {
+                  k = BUCKET_BSTAR(d0, d1);
+                  if(--d1 <= d0) {
+                    d1 = ALPHABET_SIZE - 1;
+                    if(--d0 < 0) { break; }
+                  }
+                } while(((l - k) <= 1) && (0 < (l = k)));
+                c0 = d0, c1 = d1, j = k;
+              }
+            }
+            if(l == 0) { break; }
+            sssort(T, PAb, SA + k, SA + l,
+                   curbuf, bufsize, 2, n, *(SA + k) == (m - 1));
+          }
+        }
+    }
+    else
+    {
+        buf = SA + m, bufsize = n - (2 * m);
+        for(c0 = ALPHABET_SIZE - 2, j = m; 0 < j; --c0) {
+          for(c1 = ALPHABET_SIZE - 1; c0 < c1; j = i, --c1) {
+            i = BUCKET_BSTAR(c0, c1);
+            if(1 < (j - i)) {
+              sssort(T, PAb, SA + i, SA + j,
+                     buf, bufsize, 2, n, *(SA + i) == (m - 1));
+            }
+          }
+        }
+    }
+#else
+    buf = SA + m, bufsize = n - (2 * m);
+    for(c0 = ALPHABET_SIZE - 2, j = m; 0 < j; --c0) {
+      for(c1 = ALPHABET_SIZE - 1; c0 < c1; j = i, --c1) {
+        i = BUCKET_BSTAR(c0, c1);
+        if(1 < (j - i)) {
+          sssort(T, PAb, SA + i, SA + j,
+                 buf, bufsize, 2, n, *(SA + i) == (m - 1));
+        }
+      }
+    }
+#endif
+
+    /* Compute ranks of type B* substrings. */
+    for(i = m - 1; 0 <= i; --i) {
+      if(0 <= SA[i]) {
+        j = i;
+        do { ISAb[SA[i]] = i; } while((0 <= --i) && (0 <= SA[i]));
+        SA[i + 1] = i - j;
+        if(i <= 0) { break; }
+      }
+      j = i;
+      do { ISAb[SA[i] = ~SA[i]] = j; } while(SA[--i] < 0);
+      ISAb[SA[i]] = j;
+    }
+
+    /* Construct the inverse suffix array of type B* suffixes using trsort. */
+    trsort(ISAb, SA, m, 1);
+
+    /* Set the sorted order of type B* suffixes. */
+    for(i = n - 1, j = m, c0 = T[n - 1]; 0 <= i;) {
+      for(--i, c1 = c0; (0 <= i) && ((c0 = T[i]) >= c1); --i, c1 = c0) { }
+      if(0 <= i) {
+        t = i;
+        for(--i, c1 = c0; (0 <= i) && ((c0 = T[i]) <= c1); --i, c1 = c0) { }
+        SA[ISAb[--j]] = ((t == 0) || (1 < (t - i))) ? t : ~t;
+      }
+    }
+
+    /* Calculate the index of start/end point of each bucket. */
+    BUCKET_B(ALPHABET_SIZE - 1, ALPHABET_SIZE - 1) = n; /* end point */
+    for(c0 = ALPHABET_SIZE - 2, k = m - 1; 0 <= c0; --c0) {
+      i = BUCKET_A(c0 + 1) - 1;
+      for(c1 = ALPHABET_SIZE - 1; c0 < c1; --c1) {
+        t = i - BUCKET_B(c0, c1);
+        BUCKET_B(c0, c1) = i; /* end point */
+
+        /* Move all type B* suffixes to the correct position. */
+        for(i = t, j = BUCKET_BSTAR(c0, c1);
+            j <= k;
+            --i, --k) { SA[i] = SA[k]; }
+      }
+      BUCKET_BSTAR(c0, c0 + 1) = i - BUCKET_B(c0, c0) + 1; /* start point */
+      BUCKET_B(c0, c0) = i; /* end point */
+    }
+  }
+
+  return m;
+}
+
+/* Constructs the suffix array by using the sorted order of type B* suffixes. */
+static
+void
+construct_SA(const unsigned char *T, int *SA,
+             int *bucket_A, int *bucket_B,
+             int n, int m) {
+  int *i, *j, *k;
+  int s;
+  int c0, c1, c2;
+
+  if(0 < m) {
+    /* Construct the sorted order of type B suffixes by using
+       the sorted order of type B* suffixes. */
+    for(c1 = ALPHABET_SIZE - 2; 0 <= c1; --c1) {
+      /* Scan the suffix array from right to left. */
+      for(i = SA + BUCKET_BSTAR(c1, c1 + 1),
+          j = SA + BUCKET_A(c1 + 1) - 1, k = NULL, c2 = -1;
+          i <= j;
+          --j) {
+        if(0 < (s = *j)) {
+          assert(T[s] == c1);
+          assert(((s + 1) < n) && (T[s] <= T[s + 1]));
+          assert(T[s - 1] <= T[s]);
+          *j = ~s;
+          c0 = T[--s];
+          if((0 < s) && (T[s - 1] > c0)) { s = ~s; }
+          if(c0 != c2) {
+            if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
+            k = SA + BUCKET_B(c2 = c0, c1);
+          }
+          assert(k < j); assert(k != NULL);
+          *k-- = s;
+        } else {
+          assert(((s == 0) && (T[s] == c1)) || (s < 0));
+          *j = ~s;
+        }
+      }
+    }
+  }
+
+  /* Construct the suffix array by using
+     the sorted order of type B suffixes. */
+  k = SA + BUCKET_A(c2 = T[n - 1]);
+  *k++ = (T[n - 2] < c2) ? ~(n - 1) : (n - 1);
+  /* Scan the suffix array from left to right. */
+  for(i = SA, j = SA + n; i < j; ++i) {
+    if(0 < (s = *i)) {
+      assert(T[s - 1] >= T[s]);
+      c0 = T[--s];
+      if((s == 0) || (T[s - 1] < c0)) { s = ~s; }
+      if(c0 != c2) {
+        BUCKET_A(c2) = k - SA;
+        k = SA + BUCKET_A(c2 = c0);
+      }
+      assert(i < k);
+      *k++ = s;
+    } else {
+      assert(s < 0);
+      *i = ~s;
+    }
+  }
+}
+
+/* Constructs the burrows-wheeler transformed string directly
+   by using the sorted order of type B* suffixes. */
+static
+int
+construct_BWT(const unsigned char *T, int *SA,
+              int *bucket_A, int *bucket_B,
+              int n, int m) {
+  int *i, *j, *k, *orig;
+  int s;
+  int c0, c1, c2;
+
+  if(0 < m) {
+    /* Construct the sorted order of type B suffixes by using
+       the sorted order of type B* suffixes. */
+    for(c1 = ALPHABET_SIZE - 2; 0 <= c1; --c1) {
+      /* Scan the suffix array from right to left. */
+      for(i = SA + BUCKET_BSTAR(c1, c1 + 1),
+          j = SA + BUCKET_A(c1 + 1) - 1, k = NULL, c2 = -1;
+          i <= j;
+          --j) {
+        if(0 < (s = *j)) {
+          assert(T[s] == c1);
+          assert(((s + 1) < n) && (T[s] <= T[s + 1]));
+          assert(T[s - 1] <= T[s]);
+          c0 = T[--s];
+          *j = ~((int)c0);
+          if((0 < s) && (T[s - 1] > c0)) { s = ~s; }
+          if(c0 != c2) {
+            if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
+            k = SA + BUCKET_B(c2 = c0, c1);
+          }
+          assert(k < j); assert(k != NULL);
+          *k-- = s;
+        } else if(s != 0) {
+          *j = ~s;
+#ifndef NDEBUG
+        } else {
+          assert(T[s] == c1);
+#endif
+        }
+      }
+    }
+  }
+
+  /* Construct the BWTed string by using
+     the sorted order of type B suffixes. */
+  k = SA + BUCKET_A(c2 = T[n - 1]);
+  *k++ = (T[n - 2] < c2) ? ~((int)T[n - 2]) : (n - 1);
+  /* Scan the suffix array from left to right. */
+  for(i = SA, j = SA + n, orig = SA; i < j; ++i) {
+    if(0 < (s = *i)) {
+      assert(T[s - 1] >= T[s]);
+      c0 = T[--s];
+      *i = c0;
+      if((0 < s) && (T[s - 1] < c0)) { s = ~((int)T[s - 1]); }
+      if(c0 != c2) {
+        BUCKET_A(c2) = k - SA;
+        k = SA + BUCKET_A(c2 = c0);
+      }
+      assert(i < k);
+      *k++ = s;
+    } else if(s != 0) {
+      *i = ~s;
+    } else {
+      orig = i;
+    }
+  }
+
+  return orig - SA;
+}
+
+/* Constructs the burrows-wheeler transformed string directly
+   by using the sorted order of type B* suffixes. */
+static
+int
+construct_BWT_indexes(const unsigned char *T, int *SA,
+                      int *bucket_A, int *bucket_B,
+                      int n, int m,
+                      unsigned char * num_indexes, int * indexes) {
+  int *i, *j, *k, *orig;
+  int s;
+  int c0, c1, c2;
+
+  int mod = n / 8;
+  {
+      mod |= mod >> 1;  mod |= mod >> 2;
+      mod |= mod >> 4;  mod |= mod >> 8;
+      mod |= mod >> 16; mod >>= 1;
+
+      *num_indexes = (unsigned char)((n - 1) / (mod + 1));
+  }
+
+  if(0 < m) {
+    /* Construct the sorted order of type B suffixes by using
+       the sorted order of type B* suffixes. */
+    for(c1 = ALPHABET_SIZE - 2; 0 <= c1; --c1) {
+      /* Scan the suffix array from right to left. */
+      for(i = SA + BUCKET_BSTAR(c1, c1 + 1),
+          j = SA + BUCKET_A(c1 + 1) - 1, k = NULL, c2 = -1;
+          i <= j;
+          --j) {
+        if(0 < (s = *j)) {
+          assert(T[s] == c1);
+          assert(((s + 1) < n) && (T[s] <= T[s + 1]));
+          assert(T[s - 1] <= T[s]);
+
+          if ((s & mod) == 0) indexes[s / (mod + 1) - 1] = j - SA;
+
+          c0 = T[--s];
+          *j = ~((int)c0);
+          if((0 < s) && (T[s - 1] > c0)) { s = ~s; }
+          if(c0 != c2) {
+            if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
+            k = SA + BUCKET_B(c2 = c0, c1);
+          }
+          assert(k < j); assert(k != NULL);
+          *k-- = s;
+        } else if(s != 0) {
+          *j = ~s;
+#ifndef NDEBUG
+        } else {
+          assert(T[s] == c1);
+#endif
+        }
+      }
+    }
+  }
+
+  /* Construct the BWTed string by using
+     the sorted order of type B suffixes. */
+  k = SA + BUCKET_A(c2 = T[n - 1]);
+  if (T[n - 2] < c2) {
+    if (((n - 1) & mod) == 0) indexes[(n - 1) / (mod + 1) - 1] = k - SA;
+    *k++ = ~((int)T[n - 2]);
+  }
+  else {
+    *k++ = n - 1;
+  }
+
+  /* Scan the suffix array from left to right. */
+  for(i = SA, j = SA + n, orig = SA; i < j; ++i) {
+    if(0 < (s = *i)) {
+      assert(T[s - 1] >= T[s]);
+
+      if ((s & mod) == 0) indexes[s / (mod + 1) - 1] = i - SA;
+
+      c0 = T[--s];
+      *i = c0;
+      if(c0 != c2) {
+        BUCKET_A(c2) = k - SA;
+        k = SA + BUCKET_A(c2 = c0);
+      }
+      assert(i < k);
+      if((0 < s) && (T[s - 1] < c0)) {
+          if ((s & mod) == 0) indexes[s / (mod + 1) - 1] = k - SA;
+          *k++ = ~((int)T[s - 1]);
+      } else
+        *k++ = s;
+    } else if(s != 0) {
+      *i = ~s;
+    } else {
+      orig = i;
+    }
+  }
+
+  return orig - SA;
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/*- Function -*/
+
+int
+divsufsort(const unsigned char *T, int *SA, int n, int openMP) {
+  int *bucket_A, *bucket_B;
+  int m;
+  int err = 0;
+
+  /* Check arguments. */
+  if((T == NULL) || (SA == NULL) || (n < 0)) { return -1; }
+  else if(n == 0) { return 0; }
+  else if(n == 1) { SA[0] = 0; return 0; }
+  else if(n == 2) { m = (T[0] < T[1]); SA[m ^ 1] = 0, SA[m] = 1; return 0; }
+
+  bucket_A = (int *)malloc(BUCKET_A_SIZE * sizeof(int));
+  bucket_B = (int *)malloc(BUCKET_B_SIZE * sizeof(int));
+
+  /* Suffixsort. */
+  if((bucket_A != NULL) && (bucket_B != NULL)) {
+    m = sort_typeBstar(T, SA, bucket_A, bucket_B, n, openMP);
+    construct_SA(T, SA, bucket_A, bucket_B, n, m);
+  } else {
+    err = -2;
+  }
+
+  free(bucket_B);
+  free(bucket_A);
+
+  return err;
+}
+
+int
+divbwt(const unsigned char *T, unsigned char *U, int *A, int n, unsigned char * num_indexes, int * indexes, int openMP) {
+  int *B;
+  int *bucket_A, *bucket_B;
+  int m, pidx, i;
+
+  /* Check arguments. */
+  if((T == NULL) || (U == NULL) || (n < 0)) { return -1; }
+  else if(n <= 1) { if(n == 1) { U[0] = T[0]; } return n; }
+
+  if((B = A) == NULL) { B = (int *)malloc((size_t)(n + 1) * sizeof(int)); }
+  bucket_A = (int *)malloc(BUCKET_A_SIZE * sizeof(int));
+  bucket_B = (int *)malloc(BUCKET_B_SIZE * sizeof(int));
+
+  /* Burrows-Wheeler Transform. */
+  if((B != NULL) && (bucket_A != NULL) && (bucket_B != NULL)) {
+    m = sort_typeBstar(T, B, bucket_A, bucket_B, n, openMP);
+
+    if (num_indexes == NULL || indexes == NULL) {
+        pidx = construct_BWT(T, B, bucket_A, bucket_B, n, m);
+    } else {
+        pidx = construct_BWT_indexes(T, B, bucket_A, bucket_B, n, m, num_indexes, indexes);
+    }
+
+    /* Copy to output string. */
+    U[0] = T[n - 1];
+    for(i = 0; i < pidx; ++i) { U[i + 1] = (unsigned char)B[i]; }
+    for(i += 1; i < n; ++i) { U[i] = (unsigned char)B[i]; }
+    pidx += 1;
+  } else {
+    pidx = -2;
+  }
+
+  free(bucket_B);
+  free(bucket_A);
+  if(A == NULL) { free(B); }
+
+  return pidx;
+}
diff --git a/lib/dictBuilder/fastcover.c b/lib/dictBuilder/fastcover.c
new file mode 100644
index 0000000..56a0813
--- /dev/null
+++ b/lib/dictBuilder/fastcover.c
@@ -0,0 +1,765 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stdio.h>  /* fprintf */
+#include <stdlib.h> /* malloc, free, qsort */
+#include <string.h> /* memset */
+#include <time.h>   /* clock */
+
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#  define ZDICT_STATIC_LINKING_ONLY
+#endif
+
+#include "../common/mem.h" /* read */
+#include "../common/pool.h"
+#include "../common/threading.h"
+#include "../common/zstd_internal.h" /* includes zstd.h */
+#include "../compress/zstd_compress_internal.h" /* ZSTD_hash*() */
+#include "../zdict.h"
+#include "cover.h"
+
+
+/*-*************************************
+*  Constants
+***************************************/
+/**
+* There are 32bit indexes used to ref samples, so limit samples size to 4GB
+* on 64bit builds.
+* For 32bit builds we choose 1 GB.
+* Most 32bit platforms have 2GB user-mode addressable space and we allocate a large
+* contiguous buffer, so 1GB is already a high limit.
+*/
+#define FASTCOVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
+#define FASTCOVER_MAX_F 31
+#define FASTCOVER_MAX_ACCEL 10
+#define FASTCOVER_DEFAULT_SPLITPOINT 0.75
+#define DEFAULT_F 20
+#define DEFAULT_ACCEL 1
+
+
+/*-*************************************
+*  Console display
+*
+* Captures the `displayLevel` variable in the local scope.
+***************************************/
+#undef  DISPLAY
+#define DISPLAY(...)                                                           \
+  {                                                                            \
+    fprintf(stderr, __VA_ARGS__);                                              \
+    fflush(stderr);                                                            \
+  }
+#undef  DISPLAYLEVEL
+#define DISPLAYLEVEL(l, ...)                                                   \
+  if (displayLevel >= l) {                                                     \
+    DISPLAY(__VA_ARGS__);                                                      \
+  } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+
+#undef  DISPLAYUPDATE
+#define DISPLAYUPDATE(lastUpdateTime, l, ...)                                  \
+  if (displayLevel >= l) {                                                     \
+    const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;                     \
+    if ((clock() - lastUpdateTime > refreshRate) || (displayLevel >= 4)) {     \
+      lastUpdateTime = clock();                                                \
+      DISPLAY(__VA_ARGS__);                                                    \
+    }                                                                          \
+  }
+
+
+/*-*************************************
+* Hash Functions
+***************************************/
+/**
+ * Hash the d-byte value pointed to by p and mod 2^f into the frequency vector
+ */
+static size_t FASTCOVER_hashPtrToIndex(const void* p, U32 f, unsigned d) {
+  if (d == 6) {
+    return ZSTD_hash6Ptr(p, f);
+  }
+  return ZSTD_hash8Ptr(p, f);
+}
+
+
+/*-*************************************
+* Acceleration
+***************************************/
+typedef struct {
+  unsigned finalize;    /* Percentage of training samples used for ZDICT_finalizeDictionary */
+  unsigned skip;        /* Number of dmer skipped between each dmer counted in computeFrequency */
+} FASTCOVER_accel_t;
+
+
+static const FASTCOVER_accel_t FASTCOVER_defaultAccelParameters[FASTCOVER_MAX_ACCEL+1] = {
+  { 100, 0 },   /* accel = 0, should not happen because accel = 0 defaults to accel = 1 */
+  { 100, 0 },   /* accel = 1 */
+  { 50, 1 },   /* accel = 2 */
+  { 34, 2 },   /* accel = 3 */
+  { 25, 3 },   /* accel = 4 */
+  { 20, 4 },   /* accel = 5 */
+  { 17, 5 },   /* accel = 6 */
+  { 14, 6 },   /* accel = 7 */
+  { 13, 7 },   /* accel = 8 */
+  { 11, 8 },   /* accel = 9 */
+  { 10, 9 },   /* accel = 10 */
+};
+
+
+/*-*************************************
+* Context
+***************************************/
+typedef struct {
+  const BYTE *samples;
+  size_t *offsets;
+  const size_t *samplesSizes;
+  size_t nbSamples;
+  size_t nbTrainSamples;
+  size_t nbTestSamples;
+  size_t nbDmers;
+  U32 *freqs;
+  unsigned d;
+  unsigned f;
+  FASTCOVER_accel_t accelParams;
+  int displayLevel;
+} FASTCOVER_ctx_t;
+
+
+/*-*************************************
+*  Helper functions
+***************************************/
+/**
+ * Selects the best segment in an epoch.
+ * Segments of are scored according to the function:
+ *
+ * Let F(d) be the frequency of all dmers with hash value d.
+ * Let S_i be hash value of the dmer at position i of segment S which has length k.
+ *
+ *     Score(S) = F(S_1) + F(S_2) + ... + F(S_{k-d+1})
+ *
+ * Once the dmer with hash value d is in the dictionary we set F(d) = 0.
+ */
+static COVER_segment_t FASTCOVER_selectSegment(const FASTCOVER_ctx_t *ctx,
+                                              U32 *freqs, U32 begin, U32 end,
+                                              ZDICT_cover_params_t parameters,
+                                              U16* segmentFreqs) {
+  /* Constants */
+  const U32 k = parameters.k;
+  const U32 d = parameters.d;
+  const U32 f = ctx->f;
+  const U32 dmersInK = k - d + 1;
+
+  /* Try each segment (activeSegment) and save the best (bestSegment) */
+  COVER_segment_t bestSegment = {0, 0, 0};
+  COVER_segment_t activeSegment;
+
+  /* Reset the activeDmers in the segment */
+  /* The activeSegment starts at the beginning of the epoch. */
+  activeSegment.begin = begin;
+  activeSegment.end = begin;
+  activeSegment.score = 0;
+
+  /* Slide the activeSegment through the whole epoch.
+   * Save the best segment in bestSegment.
+   */
+  while (activeSegment.end < end) {
+    /* Get hash value of current dmer */
+    const size_t idx = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.end, f, d);
+
+    /* Add frequency of this index to score if this is the first occurrence of index in active segment */
+    if (segmentFreqs[idx] == 0) {
+      activeSegment.score += freqs[idx];
+    }
+    /* Increment end of segment and segmentFreqs*/
+    activeSegment.end += 1;
+    segmentFreqs[idx] += 1;
+    /* If the window is now too large, drop the first position */
+    if (activeSegment.end - activeSegment.begin == dmersInK + 1) {
+      /* Get hash value of the dmer to be eliminated from active segment */
+      const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
+      segmentFreqs[delIndex] -= 1;
+      /* Subtract frequency of this index from score if this is the last occurrence of this index in active segment */
+      if (segmentFreqs[delIndex] == 0) {
+        activeSegment.score -= freqs[delIndex];
+      }
+      /* Increment start of segment */
+      activeSegment.begin += 1;
+    }
+
+    /* If this segment is the best so far save it */
+    if (activeSegment.score > bestSegment.score) {
+      bestSegment = activeSegment;
+    }
+  }
+
+  /* Zero out rest of segmentFreqs array */
+  while (activeSegment.begin < end) {
+    const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
+    segmentFreqs[delIndex] -= 1;
+    activeSegment.begin += 1;
+  }
+
+  {
+    /*  Zero the frequency of hash value of each dmer covered by the chosen segment. */
+    U32 pos;
+    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
+      const size_t i = FASTCOVER_hashPtrToIndex(ctx->samples + pos, f, d);
+      freqs[i] = 0;
+    }
+  }
+
+  return bestSegment;
+}
+
+
+static int FASTCOVER_checkParameters(ZDICT_cover_params_t parameters,
+                                     size_t maxDictSize, unsigned f,
+                                     unsigned accel) {
+  /* k, d, and f are required parameters */
+  if (parameters.d == 0 || parameters.k == 0) {
+    return 0;
+  }
+  /* d has to be 6 or 8 */
+  if (parameters.d != 6 && parameters.d != 8) {
+    return 0;
+  }
+  /* k <= maxDictSize */
+  if (parameters.k > maxDictSize) {
+    return 0;
+  }
+  /* d <= k */
+  if (parameters.d > parameters.k) {
+    return 0;
+  }
+  /* 0 < f <= FASTCOVER_MAX_F*/
+  if (f > FASTCOVER_MAX_F || f == 0) {
+    return 0;
+  }
+  /* 0 < splitPoint <= 1 */
+  if (parameters.splitPoint <= 0 || parameters.splitPoint > 1) {
+    return 0;
+  }
+  /* 0 < accel <= 10 */
+  if (accel > 10 || accel == 0) {
+    return 0;
+  }
+  return 1;
+}
+
+
+/**
+ * Clean up a context initialized with `FASTCOVER_ctx_init()`.
+ */
+static void
+FASTCOVER_ctx_destroy(FASTCOVER_ctx_t* ctx)
+{
+    if (!ctx) return;
+
+    free(ctx->freqs);
+    ctx->freqs = NULL;
+
+    free(ctx->offsets);
+    ctx->offsets = NULL;
+}
+
+
+/**
+ * Calculate for frequency of hash value of each dmer in ctx->samples
+ */
+static void
+FASTCOVER_computeFrequency(U32* freqs, const FASTCOVER_ctx_t* ctx)
+{
+    const unsigned f = ctx->f;
+    const unsigned d = ctx->d;
+    const unsigned skip = ctx->accelParams.skip;
+    const unsigned readLength = MAX(d, 8);
+    size_t i;
+    assert(ctx->nbTrainSamples >= 5);
+    assert(ctx->nbTrainSamples <= ctx->nbSamples);
+    for (i = 0; i < ctx->nbTrainSamples; i++) {
+        size_t start = ctx->offsets[i];  /* start of current dmer */
+        size_t const currSampleEnd = ctx->offsets[i+1];
+        while (start + readLength <= currSampleEnd) {
+            const size_t dmerIndex = FASTCOVER_hashPtrToIndex(ctx->samples + start, f, d);
+            freqs[dmerIndex]++;
+            start = start + skip + 1;
+        }
+    }
+}
+
+
+/**
+ * Prepare a context for dictionary building.
+ * The context is only dependent on the parameter `d` and can be used multiple
+ * times.
+ * Returns 0 on success or error code on error.
+ * The context must be destroyed with `FASTCOVER_ctx_destroy()`.
+ */
+static size_t
+FASTCOVER_ctx_init(FASTCOVER_ctx_t* ctx,
+                   const void* samplesBuffer,
+                   const size_t* samplesSizes, unsigned nbSamples,
+                   unsigned d, double splitPoint, unsigned f,
+                   FASTCOVER_accel_t accelParams,
+                   int displayLevel)
+{
+    const BYTE* const samples = (const BYTE*)samplesBuffer;
+    const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
+    /* Split samples into testing and training sets */
+    const unsigned nbTrainSamples = splitPoint < 1.0 ? (unsigned)((double)nbSamples * splitPoint) : nbSamples;
+    const unsigned nbTestSamples = splitPoint < 1.0 ? nbSamples - nbTrainSamples : nbSamples;
+    const size_t trainingSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes, nbTrainSamples) : totalSamplesSize;
+    const size_t testSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes + nbTrainSamples, nbTestSamples) : totalSamplesSize;
+    ctx->displayLevel = displayLevel;
+
+    /* Checks */
+    if (totalSamplesSize < MAX(d, sizeof(U64)) ||
+        totalSamplesSize >= (size_t)FASTCOVER_MAX_SAMPLES_SIZE) {
+        DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
+                    (unsigned)(totalSamplesSize >> 20), (FASTCOVER_MAX_SAMPLES_SIZE >> 20));
+        return ERROR(srcSize_wrong);
+    }
+
+    /* Check if there are at least 5 training samples */
+    if (nbTrainSamples < 5) {
+        DISPLAYLEVEL(1, "Total number of training samples is %u and is invalid\n", nbTrainSamples);
+        return ERROR(srcSize_wrong);
+    }
+
+    /* Check if there's testing sample */
+    if (nbTestSamples < 1) {
+        DISPLAYLEVEL(1, "Total number of testing samples is %u and is invalid.\n", nbTestSamples);
+        return ERROR(srcSize_wrong);
+    }
+
+    /* Zero the context */
+    memset(ctx, 0, sizeof(*ctx));
+    DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbTrainSamples,
+                    (unsigned)trainingSamplesSize);
+    DISPLAYLEVEL(2, "Testing on %u samples of total size %u\n", nbTestSamples,
+                    (unsigned)testSamplesSize);
+
+    ctx->samples = samples;
+    ctx->samplesSizes = samplesSizes;
+    ctx->nbSamples = nbSamples;
+    ctx->nbTrainSamples = nbTrainSamples;
+    ctx->nbTestSamples = nbTestSamples;
+    ctx->nbDmers = trainingSamplesSize - MAX(d, sizeof(U64)) + 1;
+    ctx->d = d;
+    ctx->f = f;
+    ctx->accelParams = accelParams;
+
+    /* The offsets of each file */
+    ctx->offsets = (size_t*)calloc((nbSamples + 1), sizeof(size_t));
+    if (ctx->offsets == NULL) {
+        DISPLAYLEVEL(1, "Failed to allocate scratch buffers \n");
+        FASTCOVER_ctx_destroy(ctx);
+        return ERROR(memory_allocation);
+    }
+
+    /* Fill offsets from the samplesSizes */
+    {   U32 i;
+        ctx->offsets[0] = 0;
+        assert(nbSamples >= 5);
+        for (i = 1; i <= nbSamples; ++i) {
+            ctx->offsets[i] = ctx->offsets[i - 1] + samplesSizes[i - 1];
+        }
+    }
+
+    /* Initialize frequency array of size 2^f */
+    ctx->freqs = (U32*)calloc(((U64)1 << f), sizeof(U32));
+    if (ctx->freqs == NULL) {
+        DISPLAYLEVEL(1, "Failed to allocate frequency table \n");
+        FASTCOVER_ctx_destroy(ctx);
+        return ERROR(memory_allocation);
+    }
+
+    DISPLAYLEVEL(2, "Computing frequencies\n");
+    FASTCOVER_computeFrequency(ctx->freqs, ctx);
+
+    return 0;
+}
+
+
+/**
+ * Given the prepared context build the dictionary.
+ */
+static size_t
+FASTCOVER_buildDictionary(const FASTCOVER_ctx_t* ctx,
+                          U32* freqs,
+                          void* dictBuffer, size_t dictBufferCapacity,
+                          ZDICT_cover_params_t parameters,
+                          U16* segmentFreqs)
+{
+  BYTE *const dict = (BYTE *)dictBuffer;
+  size_t tail = dictBufferCapacity;
+  /* Divide the data into epochs. We will select one segment from each epoch. */
+  const COVER_epoch_info_t epochs = COVER_computeEpochs(
+      (U32)dictBufferCapacity, (U32)ctx->nbDmers, parameters.k, 1);
+  const size_t maxZeroScoreRun = 10;
+  const int displayLevel = ctx->displayLevel;
+  size_t zeroScoreRun = 0;
+  clock_t lastUpdateTime = 0;
+  size_t epoch;
+  DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n",
+                (U32)epochs.num, (U32)epochs.size);
+  /* Loop through the epochs until there are no more segments or the dictionary
+   * is full.
+   */
+  for (epoch = 0; tail > 0; epoch = (epoch + 1) % epochs.num) {
+    const U32 epochBegin = (U32)(epoch * epochs.size);
+    const U32 epochEnd = epochBegin + epochs.size;
+    size_t segmentSize;
+    /* Select a segment */
+    COVER_segment_t segment = FASTCOVER_selectSegment(
+        ctx, freqs, epochBegin, epochEnd, parameters, segmentFreqs);
+
+    /* If the segment covers no dmers, then we are out of content.
+     * There may be new content in other epochs, for continue for some time.
+     */
+    if (segment.score == 0) {
+      if (++zeroScoreRun >= maxZeroScoreRun) {
+          break;
+      }
+      continue;
+    }
+    zeroScoreRun = 0;
+
+    /* Trim the segment if necessary and if it is too small then we are done */
+    segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
+    if (segmentSize < parameters.d) {
+      break;
+    }
+
+    /* We fill the dictionary from the back to allow the best segments to be
+     * referenced with the smallest offsets.
+     */
+    tail -= segmentSize;
+    memcpy(dict + tail, ctx->samples + segment.begin, segmentSize);
+    DISPLAYUPDATE(
+        lastUpdateTime,
+        2, "\r%u%%       ",
+        (unsigned)(((dictBufferCapacity - tail) * 100) / dictBufferCapacity));
+  }
+  DISPLAYLEVEL(2, "\r%79s\r", "");
+  return tail;
+}
+
+/**
+ * Parameters for FASTCOVER_tryParameters().
+ */
+typedef struct FASTCOVER_tryParameters_data_s {
+    const FASTCOVER_ctx_t* ctx;
+    COVER_best_t* best;
+    size_t dictBufferCapacity;
+    ZDICT_cover_params_t parameters;
+} FASTCOVER_tryParameters_data_t;
+
+
+/**
+ * Tries a set of parameters and updates the COVER_best_t with the results.
+ * This function is thread safe if zstd is compiled with multithreaded support.
+ * It takes its parameters as an *OWNING* opaque pointer to support threading.
+ */
+static void FASTCOVER_tryParameters(void* opaque)
+{
+  /* Save parameters as local variables */
+  FASTCOVER_tryParameters_data_t *const data = (FASTCOVER_tryParameters_data_t*)opaque;
+  const FASTCOVER_ctx_t *const ctx = data->ctx;
+  const ZDICT_cover_params_t parameters = data->parameters;
+  size_t dictBufferCapacity = data->dictBufferCapacity;
+  size_t totalCompressedSize = ERROR(GENERIC);
+  /* Initialize array to keep track of frequency of dmer within activeSegment */
+  U16* segmentFreqs = (U16*)calloc(((U64)1 << ctx->f), sizeof(U16));
+  /* Allocate space for hash table, dict, and freqs */
+  BYTE *const dict = (BYTE*)malloc(dictBufferCapacity);
+  COVER_dictSelection_t selection = COVER_dictSelectionError(ERROR(GENERIC));
+  U32* freqs = (U32*) malloc(((U64)1 << ctx->f) * sizeof(U32));
+  const int displayLevel = ctx->displayLevel;
+  if (!segmentFreqs || !dict || !freqs) {
+    DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
+    goto _cleanup;
+  }
+  /* Copy the frequencies because we need to modify them */
+  memcpy(freqs, ctx->freqs, ((U64)1 << ctx->f) * sizeof(U32));
+  /* Build the dictionary */
+  { const size_t tail = FASTCOVER_buildDictionary(ctx, freqs, dict, dictBufferCapacity,
+                                                    parameters, segmentFreqs);
+
+    const unsigned nbFinalizeSamples = (unsigned)(ctx->nbTrainSamples * ctx->accelParams.finalize / 100);
+    selection = COVER_selectDict(dict + tail, dictBufferCapacity, dictBufferCapacity - tail,
+         ctx->samples, ctx->samplesSizes, nbFinalizeSamples, ctx->nbTrainSamples, ctx->nbSamples, parameters, ctx->offsets,
+         totalCompressedSize);
+
+    if (COVER_dictSelectionIsError(selection)) {
+      DISPLAYLEVEL(1, "Failed to select dictionary\n");
+      goto _cleanup;
+    }
+  }
+_cleanup:
+  free(dict);
+  COVER_best_finish(data->best, parameters, selection);
+  free(data);
+  free(segmentFreqs);
+  COVER_dictSelectionFree(selection);
+  free(freqs);
+}
+
+
+static void
+FASTCOVER_convertToCoverParams(ZDICT_fastCover_params_t fastCoverParams,
+                               ZDICT_cover_params_t* coverParams)
+{
+    coverParams->k = fastCoverParams.k;
+    coverParams->d = fastCoverParams.d;
+    coverParams->steps = fastCoverParams.steps;
+    coverParams->nbThreads = fastCoverParams.nbThreads;
+    coverParams->splitPoint = fastCoverParams.splitPoint;
+    coverParams->zParams = fastCoverParams.zParams;
+    coverParams->shrinkDict = fastCoverParams.shrinkDict;
+}
+
+
+static void
+FASTCOVER_convertToFastCoverParams(ZDICT_cover_params_t coverParams,
+                                   ZDICT_fastCover_params_t* fastCoverParams,
+                                   unsigned f, unsigned accel)
+{
+    fastCoverParams->k = coverParams.k;
+    fastCoverParams->d = coverParams.d;
+    fastCoverParams->steps = coverParams.steps;
+    fastCoverParams->nbThreads = coverParams.nbThreads;
+    fastCoverParams->splitPoint = coverParams.splitPoint;
+    fastCoverParams->f = f;
+    fastCoverParams->accel = accel;
+    fastCoverParams->zParams = coverParams.zParams;
+    fastCoverParams->shrinkDict = coverParams.shrinkDict;
+}
+
+
+ZDICTLIB_STATIC_API size_t
+ZDICT_trainFromBuffer_fastCover(void* dictBuffer, size_t dictBufferCapacity,
+                                const void* samplesBuffer,
+                                const size_t* samplesSizes, unsigned nbSamples,
+                                ZDICT_fastCover_params_t parameters)
+{
+    BYTE* const dict = (BYTE*)dictBuffer;
+    FASTCOVER_ctx_t ctx;
+    ZDICT_cover_params_t coverParams;
+    FASTCOVER_accel_t accelParams;
+    const int displayLevel = (int)parameters.zParams.notificationLevel;
+    /* Assign splitPoint and f if not provided */
+    parameters.splitPoint = 1.0;
+    parameters.f = parameters.f == 0 ? DEFAULT_F : parameters.f;
+    parameters.accel = parameters.accel == 0 ? DEFAULT_ACCEL : parameters.accel;
+    /* Convert to cover parameter */
+    memset(&coverParams, 0 , sizeof(coverParams));
+    FASTCOVER_convertToCoverParams(parameters, &coverParams);
+    /* Checks */
+    if (!FASTCOVER_checkParameters(coverParams, dictBufferCapacity, parameters.f,
+                                   parameters.accel)) {
+      DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
+      return ERROR(parameter_outOfBound);
+    }
+    if (nbSamples == 0) {
+      DISPLAYLEVEL(1, "FASTCOVER must have at least one input file\n");
+      return ERROR(srcSize_wrong);
+    }
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+      DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
+                   ZDICT_DICTSIZE_MIN);
+      return ERROR(dstSize_tooSmall);
+    }
+    /* Assign corresponding FASTCOVER_accel_t to accelParams*/
+    accelParams = FASTCOVER_defaultAccelParameters[parameters.accel];
+    /* Initialize context */
+    {
+      size_t const initVal = FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
+                            coverParams.d, parameters.splitPoint, parameters.f,
+                            accelParams, displayLevel);
+      if (ZSTD_isError(initVal)) {
+        DISPLAYLEVEL(1, "Failed to initialize context\n");
+        return initVal;
+      }
+    }
+    COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, displayLevel);
+    /* Build the dictionary */
+    DISPLAYLEVEL(2, "Building dictionary\n");
+    {
+      /* Initialize array to keep track of frequency of dmer within activeSegment */
+      U16* segmentFreqs = (U16 *)calloc(((U64)1 << parameters.f), sizeof(U16));
+      const size_t tail = FASTCOVER_buildDictionary(&ctx, ctx.freqs, dictBuffer,
+                                                dictBufferCapacity, coverParams, segmentFreqs);
+      const unsigned nbFinalizeSamples = (unsigned)(ctx.nbTrainSamples * ctx.accelParams.finalize / 100);
+      const size_t dictionarySize = ZDICT_finalizeDictionary(
+          dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
+          samplesBuffer, samplesSizes, nbFinalizeSamples, coverParams.zParams);
+      if (!ZSTD_isError(dictionarySize)) {
+          DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
+                      (unsigned)dictionarySize);
+      }
+      FASTCOVER_ctx_destroy(&ctx);
+      free(segmentFreqs);
+      return dictionarySize;
+    }
+}
+
+
+ZDICTLIB_STATIC_API size_t
+ZDICT_optimizeTrainFromBuffer_fastCover(
+                    void* dictBuffer, size_t dictBufferCapacity,
+                    const void* samplesBuffer,
+                    const size_t* samplesSizes, unsigned nbSamples,
+                    ZDICT_fastCover_params_t* parameters)
+{
+    ZDICT_cover_params_t coverParams;
+    FASTCOVER_accel_t accelParams;
+    /* constants */
+    const unsigned nbThreads = parameters->nbThreads;
+    const double splitPoint =
+        parameters->splitPoint <= 0.0 ? FASTCOVER_DEFAULT_SPLITPOINT : parameters->splitPoint;
+    const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
+    const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
+    const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
+    const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
+    const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
+    const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
+    const unsigned kIterations =
+        (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
+    const unsigned f = parameters->f == 0 ? DEFAULT_F : parameters->f;
+    const unsigned accel = parameters->accel == 0 ? DEFAULT_ACCEL : parameters->accel;
+    const unsigned shrinkDict = 0;
+    /* Local variables */
+    const int displayLevel = (int)parameters->zParams.notificationLevel;
+    unsigned iteration = 1;
+    unsigned d;
+    unsigned k;
+    COVER_best_t best;
+    POOL_ctx *pool = NULL;
+    int warned = 0;
+    clock_t lastUpdateTime = 0;
+    /* Checks */
+    if (splitPoint <= 0 || splitPoint > 1) {
+      DISPLAYLEVEL(1, "Incorrect splitPoint\n");
+      return ERROR(parameter_outOfBound);
+    }
+    if (accel == 0 || accel > FASTCOVER_MAX_ACCEL) {
+      DISPLAYLEVEL(1, "Incorrect accel\n");
+      return ERROR(parameter_outOfBound);
+    }
+    if (kMinK < kMaxD || kMaxK < kMinK) {
+      DISPLAYLEVEL(1, "Incorrect k\n");
+      return ERROR(parameter_outOfBound);
+    }
+    if (nbSamples == 0) {
+      DISPLAYLEVEL(1, "FASTCOVER must have at least one input file\n");
+      return ERROR(srcSize_wrong);
+    }
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+      DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
+                   ZDICT_DICTSIZE_MIN);
+      return ERROR(dstSize_tooSmall);
+    }
+    if (nbThreads > 1) {
+      pool = POOL_create(nbThreads, 1);
+      if (!pool) {
+        return ERROR(memory_allocation);
+      }
+    }
+    /* Initialization */
+    COVER_best_init(&best);
+    memset(&coverParams, 0 , sizeof(coverParams));
+    FASTCOVER_convertToCoverParams(*parameters, &coverParams);
+    accelParams = FASTCOVER_defaultAccelParameters[accel];
+    /* Loop through d first because each new value needs a new context */
+    DISPLAYLEVEL(2, "Trying %u different sets of parameters\n", kIterations);
+    for (d = kMinD; d <= kMaxD; d += 2) {
+      /* Initialize the context for this value of d */
+      FASTCOVER_ctx_t ctx;
+      DISPLAYLEVEL(3, "d=%u\n", d);
+      {
+        /* Turn down global display level to clean up display at level 2 and below */
+        const int childDisplayLevel = displayLevel == 0 ? 0 : displayLevel - 1;
+        size_t const initVal = FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint, f, accelParams, childDisplayLevel);
+        if (ZSTD_isError(initVal)) {
+          DISPLAYLEVEL(1, "Failed to initialize context\n");
+          COVER_best_destroy(&best);
+          POOL_free(pool);
+          return initVal;
+        }
+      }
+      if (!warned) {
+        COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, displayLevel);
+        warned = 1;
+      }
+      /* Loop through k reusing the same context */
+      for (k = kMinK; k <= kMaxK; k += kStepSize) {
+        /* Prepare the arguments */
+        FASTCOVER_tryParameters_data_t *data = (FASTCOVER_tryParameters_data_t *)malloc(
+            sizeof(FASTCOVER_tryParameters_data_t));
+        DISPLAYLEVEL(3, "k=%u\n", k);
+        if (!data) {
+          DISPLAYLEVEL(1, "Failed to allocate parameters\n");
+          COVER_best_destroy(&best);
+          FASTCOVER_ctx_destroy(&ctx);
+          POOL_free(pool);
+          return ERROR(memory_allocation);
+        }
+        data->ctx = &ctx;
+        data->best = &best;
+        data->dictBufferCapacity = dictBufferCapacity;
+        data->parameters = coverParams;
+        data->parameters.k = k;
+        data->parameters.d = d;
+        data->parameters.splitPoint = splitPoint;
+        data->parameters.steps = kSteps;
+        data->parameters.shrinkDict = shrinkDict;
+        data->parameters.zParams.notificationLevel = (unsigned)ctx.displayLevel;
+        /* Check the parameters */
+        if (!FASTCOVER_checkParameters(data->parameters, dictBufferCapacity,
+                                       data->ctx->f, accel)) {
+          DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
+          free(data);
+          continue;
+        }
+        /* Call the function and pass ownership of data to it */
+        COVER_best_start(&best);
+        if (pool) {
+          POOL_add(pool, &FASTCOVER_tryParameters, data);
+        } else {
+          FASTCOVER_tryParameters(data);
+        }
+        /* Print status */
+        DISPLAYUPDATE(lastUpdateTime,
+                      2, "\r%u%%       ",
+                      (unsigned)((iteration * 100) / kIterations));
+        ++iteration;
+      }
+      COVER_best_wait(&best);
+      FASTCOVER_ctx_destroy(&ctx);
+    }
+    DISPLAYLEVEL(2, "\r%79s\r", "");
+    /* Fill the output buffer and parameters with output of the best parameters */
+    {
+      const size_t dictSize = best.dictSize;
+      if (ZSTD_isError(best.compressedSize)) {
+        const size_t compressedSize = best.compressedSize;
+        COVER_best_destroy(&best);
+        POOL_free(pool);
+        return compressedSize;
+      }
+      FASTCOVER_convertToFastCoverParams(best.parameters, parameters, f, accel);
+      memcpy(dictBuffer, best.dict, dictSize);
+      COVER_best_destroy(&best);
+      POOL_free(pool);
+      return dictSize;
+    }
+
+}
diff --git a/lib/dictBuilder/zdict.c b/lib/dictBuilder/zdict.c
new file mode 100644
index 0000000..befa616
--- /dev/null
+++ b/lib/dictBuilder/zdict.c
@@ -0,0 +1,1137 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*-**************************************
+*  Tuning parameters
+****************************************/
+#define MINRATIO 4   /* minimum nb of apparition to be selected in dictionary */
+#define ZDICT_MAX_SAMPLES_SIZE (2000U << 20)
+#define ZDICT_MIN_SAMPLES_SIZE (ZDICT_CONTENTSIZE_MIN * MINRATIO)
+
+
+/*-**************************************
+*  Compiler Options
+****************************************/
+/* Unix Large Files support (>4GB) */
+#define _FILE_OFFSET_BITS 64
+#if (defined(__sun__) && (!defined(__LP64__)))   /* Sun Solaris 32-bits requires specific definitions */
+#  ifndef _LARGEFILE_SOURCE
+#  define _LARGEFILE_SOURCE
+#  endif
+#elif ! defined(__LP64__)                        /* No point defining Large file for 64 bit */
+#  ifndef _LARGEFILE64_SOURCE
+#  define _LARGEFILE64_SOURCE
+#  endif
+#endif
+
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stdlib.h>        /* malloc, free */
+#include <string.h>        /* memset */
+#include <stdio.h>         /* fprintf, fopen, ftello64 */
+#include <time.h>          /* clock */
+
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#  define ZDICT_STATIC_LINKING_ONLY
+#endif
+
+#include "../common/mem.h"           /* read */
+#include "../common/fse.h"           /* FSE_normalizeCount, FSE_writeNCount */
+#include "../common/huf.h"           /* HUF_buildCTable, HUF_writeCTable */
+#include "../common/zstd_internal.h" /* includes zstd.h */
+#include "../common/xxhash.h"        /* XXH64 */
+#include "../compress/zstd_compress_internal.h" /* ZSTD_loadCEntropy() */
+#include "../zdict.h"
+#include "divsufsort.h"
+#include "../common/bits.h"          /* ZSTD_NbCommonBytes */
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define DICTLISTSIZE_DEFAULT 10000
+
+#define NOISELENGTH 32
+
+static const U32 g_selectivity_default = 9;
+
+
+/*-*************************************
+*  Console display
+***************************************/
+#undef  DISPLAY
+#define DISPLAY(...)         do { fprintf(stderr, __VA_ARGS__); fflush( stderr ); } while (0)
+#undef  DISPLAYLEVEL
+#define DISPLAYLEVEL(l, ...) do { if (notificationLevel>=l) { DISPLAY(__VA_ARGS__); } } while (0)    /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+
+static clock_t ZDICT_clockSpan(clock_t nPrevious) { return clock() - nPrevious; }
+
+static void ZDICT_printHex(const void* ptr, size_t length)
+{
+    const BYTE* const b = (const BYTE*)ptr;
+    size_t u;
+    for (u=0; u<length; u++) {
+        BYTE c = b[u];
+        if (c<32 || c>126) c = '.';   /* non-printable char */
+        DISPLAY("%c", c);
+    }
+}
+
+
+/*-********************************************************
+*  Helper functions
+**********************************************************/
+unsigned ZDICT_isError(size_t errorCode) { return ERR_isError(errorCode); }
+
+const char* ZDICT_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize)
+{
+    if (dictSize < 8) return 0;
+    if (MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return 0;
+    return MEM_readLE32((const char*)dictBuffer + 4);
+}
+
+size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictSize)
+{
+    size_t headerSize;
+    if (dictSize <= 8 || MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return ERROR(dictionary_corrupted);
+
+    {   ZSTD_compressedBlockState_t* bs = (ZSTD_compressedBlockState_t*)malloc(sizeof(ZSTD_compressedBlockState_t));
+        U32* wksp = (U32*)malloc(HUF_WORKSPACE_SIZE);
+        if (!bs || !wksp) {
+            headerSize = ERROR(memory_allocation);
+        } else {
+            ZSTD_reset_compressedBlockState(bs);
+            headerSize = ZSTD_loadCEntropy(bs, wksp, dictBuffer, dictSize);
+        }
+
+        free(bs);
+        free(wksp);
+    }
+
+    return headerSize;
+}
+
+/*-********************************************************
+*  Dictionary training functions
+**********************************************************/
+/*! ZDICT_count() :
+    Count the nb of common bytes between 2 pointers.
+    Note : this function presumes end of buffer followed by noisy guard band.
+*/
+static size_t ZDICT_count(const void* pIn, const void* pMatch)
+{
+    const char* const pStart = (const char*)pIn;
+    for (;;) {
+        size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+        if (!diff) {
+            pIn = (const char*)pIn+sizeof(size_t);
+            pMatch = (const char*)pMatch+sizeof(size_t);
+            continue;
+        }
+        pIn = (const char*)pIn+ZSTD_NbCommonBytes(diff);
+        return (size_t)((const char*)pIn - pStart);
+    }
+}
+
+
+typedef struct {
+    U32 pos;
+    U32 length;
+    U32 savings;
+} dictItem;
+
+static void ZDICT_initDictItem(dictItem* d)
+{
+    d->pos = 1;
+    d->length = 0;
+    d->savings = (U32)(-1);
+}
+
+
+#define LLIMIT 64          /* heuristic determined experimentally */
+#define MINMATCHLENGTH 7   /* heuristic determined experimentally */
+static dictItem ZDICT_analyzePos(
+                       BYTE* doneMarks,
+                       const unsigned* suffix, U32 start,
+                       const void* buffer, U32 minRatio, U32 notificationLevel)
+{
+    U32 lengthList[LLIMIT] = {0};
+    U32 cumulLength[LLIMIT] = {0};
+    U32 savings[LLIMIT] = {0};
+    const BYTE* b = (const BYTE*)buffer;
+    size_t maxLength = LLIMIT;
+    size_t pos = (size_t)suffix[start];
+    U32 end = start;
+    dictItem solution;
+
+    /* init */
+    memset(&solution, 0, sizeof(solution));
+    doneMarks[pos] = 1;
+
+    /* trivial repetition cases */
+    if ( (MEM_read16(b+pos+0) == MEM_read16(b+pos+2))
+       ||(MEM_read16(b+pos+1) == MEM_read16(b+pos+3))
+       ||(MEM_read16(b+pos+2) == MEM_read16(b+pos+4)) ) {
+        /* skip and mark segment */
+        U16 const pattern16 = MEM_read16(b+pos+4);
+        U32 u, patternEnd = 6;
+        while (MEM_read16(b+pos+patternEnd) == pattern16) patternEnd+=2 ;
+        if (b[pos+patternEnd] == b[pos+patternEnd-1]) patternEnd++;
+        for (u=1; u<patternEnd; u++)
+            doneMarks[pos+u] = 1;
+        return solution;
+    }
+
+    /* look forward */
+    {   size_t length;
+        do {
+            end++;
+            length = ZDICT_count(b + pos, b + suffix[end]);
+        } while (length >= MINMATCHLENGTH);
+    }
+
+    /* look backward */
+    {   size_t length;
+        do {
+            length = ZDICT_count(b + pos, b + *(suffix+start-1));
+            if (length >=MINMATCHLENGTH) start--;
+        } while(length >= MINMATCHLENGTH);
+    }
+
+    /* exit if not found a minimum nb of repetitions */
+    if (end-start < minRatio) {
+        U32 idx;
+        for(idx=start; idx<end; idx++)
+            doneMarks[suffix[idx]] = 1;
+        return solution;
+    }
+
+    {   int i;
+        U32 mml;
+        U32 refinedStart = start;
+        U32 refinedEnd = end;
+
+        DISPLAYLEVEL(4, "\n");
+        DISPLAYLEVEL(4, "found %3u matches of length >= %i at pos %7u  ", (unsigned)(end-start), MINMATCHLENGTH, (unsigned)pos);
+        DISPLAYLEVEL(4, "\n");
+
+        for (mml = MINMATCHLENGTH ; ; mml++) {
+            BYTE currentChar = 0;
+            U32 currentCount = 0;
+            U32 currentID = refinedStart;
+            U32 id;
+            U32 selectedCount = 0;
+            U32 selectedID = currentID;
+            for (id =refinedStart; id < refinedEnd; id++) {
+                if (b[suffix[id] + mml] != currentChar) {
+                    if (currentCount > selectedCount) {
+                        selectedCount = currentCount;
+                        selectedID = currentID;
+                    }
+                    currentID = id;
+                    currentChar = b[ suffix[id] + mml];
+                    currentCount = 0;
+                }
+                currentCount ++;
+            }
+            if (currentCount > selectedCount) {  /* for last */
+                selectedCount = currentCount;
+                selectedID = currentID;
+            }
+
+            if (selectedCount < minRatio)
+                break;
+            refinedStart = selectedID;
+            refinedEnd = refinedStart + selectedCount;
+        }
+
+        /* evaluate gain based on new dict */
+        start = refinedStart;
+        pos = suffix[refinedStart];
+        end = start;
+        memset(lengthList, 0, sizeof(lengthList));
+
+        /* look forward */
+        {   size_t length;
+            do {
+                end++;
+                length = ZDICT_count(b + pos, b + suffix[end]);
+                if (length >= LLIMIT) length = LLIMIT-1;
+                lengthList[length]++;
+            } while (length >=MINMATCHLENGTH);
+        }
+
+        /* look backward */
+        {   size_t length = MINMATCHLENGTH;
+            while ((length >= MINMATCHLENGTH) & (start > 0)) {
+                length = ZDICT_count(b + pos, b + suffix[start - 1]);
+                if (length >= LLIMIT) length = LLIMIT - 1;
+                lengthList[length]++;
+                if (length >= MINMATCHLENGTH) start--;
+            }
+        }
+
+        /* largest useful length */
+        memset(cumulLength, 0, sizeof(cumulLength));
+        cumulLength[maxLength-1] = lengthList[maxLength-1];
+        for (i=(int)(maxLength-2); i>=0; i--)
+            cumulLength[i] = cumulLength[i+1] + lengthList[i];
+
+        {   unsigned u;
+            for (u=LLIMIT-1; u>=MINMATCHLENGTH; u--) if (cumulLength[u]>=minRatio) break;
+            maxLength = u;
+        }
+
+        /* reduce maxLength in case of final into repetitive data */
+        {   U32 l = (U32)maxLength;
+            BYTE const c = b[pos + maxLength-1];
+            while (b[pos+l-2]==c) l--;
+            maxLength = l;
+        }
+        if (maxLength < MINMATCHLENGTH) return solution;   /* skip : no long-enough solution */
+
+        /* calculate savings */
+        savings[5] = 0;
+        {   unsigned u;
+            for (u=MINMATCHLENGTH; u<=maxLength; u++)
+                savings[u] = savings[u-1] + (lengthList[u] * (u-3));
+        }
+
+        DISPLAYLEVEL(4, "Selected dict at position %u, of length %u : saves %u (ratio: %.2f)  \n",
+                     (unsigned)pos, (unsigned)maxLength, (unsigned)savings[maxLength], (double)savings[maxLength] / (double)maxLength);
+
+        solution.pos = (U32)pos;
+        solution.length = (U32)maxLength;
+        solution.savings = savings[maxLength];
+
+        /* mark positions done */
+        {   U32 id;
+            for (id=start; id<end; id++) {
+                U32 p, pEnd, length;
+                U32 const testedPos = (U32)suffix[id];
+                if (testedPos == pos)
+                    length = solution.length;
+                else {
+                    length = (U32)ZDICT_count(b+pos, b+testedPos);
+                    if (length > solution.length) length = solution.length;
+                }
+                pEnd = (U32)(testedPos + length);
+                for (p=testedPos; p<pEnd; p++)
+                    doneMarks[p] = 1;
+    }   }   }
+
+    return solution;
+}
+
+
+static int isIncluded(const void* in, const void* container, size_t length)
+{
+    const char* const ip = (const char*) in;
+    const char* const into = (const char*) container;
+    size_t u;
+
+    for (u=0; u<length; u++) {  /* works because end of buffer is a noisy guard band */
+        if (ip[u] != into[u]) break;
+    }
+
+    return u==length;
+}
+
+/*! ZDICT_tryMerge() :
+    check if dictItem can be merged, do it if possible
+    @return : id of destination elt, 0 if not merged
+*/
+static U32 ZDICT_tryMerge(dictItem* table, dictItem elt, U32 eltNbToSkip, const void* buffer)
+{
+    const U32 tableSize = table->pos;
+    const U32 eltEnd = elt.pos + elt.length;
+    const char* const buf = (const char*) buffer;
+
+    /* tail overlap */
+    U32 u; for (u=1; u<tableSize; u++) {
+        if (u==eltNbToSkip) continue;
+        if ((table[u].pos > elt.pos) && (table[u].pos <= eltEnd)) {  /* overlap, existing > new */
+            /* append */
+            U32 const addedLength = table[u].pos - elt.pos;
+            table[u].length += addedLength;
+            table[u].pos = elt.pos;
+            table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
+            table[u].savings += elt.length / 8;    /* rough approx bonus */
+            elt = table[u];
+            /* sort : improve rank */
+            while ((u>1) && (table[u-1].savings < elt.savings))
+                table[u] = table[u-1], u--;
+            table[u] = elt;
+            return u;
+    }   }
+
+    /* front overlap */
+    for (u=1; u<tableSize; u++) {
+        if (u==eltNbToSkip) continue;
+
+        if ((table[u].pos + table[u].length >= elt.pos) && (table[u].pos < elt.pos)) {  /* overlap, existing < new */
+            /* append */
+            int const addedLength = (int)eltEnd - (int)(table[u].pos + table[u].length); /* note: can be negative */
+            table[u].savings += elt.length / 8;    /* rough approx bonus */
+            if (addedLength > 0) {   /* otherwise, elt fully included into existing */
+                table[u].length += (unsigned)addedLength;
+                table[u].savings += elt.savings * (unsigned)addedLength / elt.length;   /* rough approx */
+            }
+            /* sort : improve rank */
+            elt = table[u];
+            while ((u>1) && (table[u-1].savings < elt.savings))
+                table[u] = table[u-1], u--;
+            table[u] = elt;
+            return u;
+        }
+
+        if (MEM_read64(buf + table[u].pos) == MEM_read64(buf + elt.pos + 1)) {
+            if (isIncluded(buf + table[u].pos, buf + elt.pos + 1, table[u].length)) {
+                size_t const addedLength = MAX( elt.length - table[u].length , 1 );
+                table[u].pos = elt.pos;
+                table[u].savings += (U32)(elt.savings * addedLength / elt.length);
+                table[u].length = MIN(elt.length, table[u].length + 1);
+                return u;
+            }
+        }
+    }
+
+    return 0;
+}
+
+
+static void ZDICT_removeDictItem(dictItem* table, U32 id)
+{
+    /* convention : table[0].pos stores nb of elts */
+    U32 const max = table[0].pos;
+    U32 u;
+    if (!id) return;   /* protection, should never happen */
+    for (u=id; u<max-1; u++)
+        table[u] = table[u+1];
+    table->pos--;
+}
+
+
+static void ZDICT_insertDictItem(dictItem* table, U32 maxSize, dictItem elt, const void* buffer)
+{
+    /* merge if possible */
+    U32 mergeId = ZDICT_tryMerge(table, elt, 0, buffer);
+    if (mergeId) {
+        U32 newMerge = 1;
+        while (newMerge) {
+            newMerge = ZDICT_tryMerge(table, table[mergeId], mergeId, buffer);
+            if (newMerge) ZDICT_removeDictItem(table, mergeId);
+            mergeId = newMerge;
+        }
+        return;
+    }
+
+    /* insert */
+    {   U32 current;
+        U32 nextElt = table->pos;
+        if (nextElt >= maxSize) nextElt = maxSize-1;
+        current = nextElt-1;
+        while (table[current].savings < elt.savings) {
+            table[current+1] = table[current];
+            current--;
+        }
+        table[current+1] = elt;
+        table->pos = nextElt+1;
+    }
+}
+
+
+static U32 ZDICT_dictSize(const dictItem* dictList)
+{
+    U32 u, dictSize = 0;
+    for (u=1; u<dictList[0].pos; u++)
+        dictSize += dictList[u].length;
+    return dictSize;
+}
+
+
+static size_t ZDICT_trainBuffer_legacy(dictItem* dictList, U32 dictListSize,
+                            const void* const buffer, size_t bufferSize,   /* buffer must end with noisy guard band */
+                            const size_t* fileSizes, unsigned nbFiles,
+                            unsigned minRatio, U32 notificationLevel)
+{
+    unsigned* const suffix0 = (unsigned*)malloc((bufferSize+2)*sizeof(*suffix0));
+    unsigned* const suffix = suffix0+1;
+    U32* reverseSuffix = (U32*)malloc((bufferSize)*sizeof(*reverseSuffix));
+    BYTE* doneMarks = (BYTE*)malloc((bufferSize+16)*sizeof(*doneMarks));   /* +16 for overflow security */
+    U32* filePos = (U32*)malloc(nbFiles * sizeof(*filePos));
+    size_t result = 0;
+    clock_t displayClock = 0;
+    clock_t const refreshRate = CLOCKS_PER_SEC * 3 / 10;
+
+#   undef  DISPLAYUPDATE
+#   define DISPLAYUPDATE(l, ...)                                   \
+        do {                                                       \
+            if (notificationLevel>=l) {                            \
+                if (ZDICT_clockSpan(displayClock) > refreshRate) { \
+                    displayClock = clock();                        \
+                    DISPLAY(__VA_ARGS__);                          \
+                }                                                  \
+                if (notificationLevel>=4) fflush(stderr);          \
+            }                                                      \
+        } while (0)
+
+    /* init */
+    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
+    if (!suffix0 || !reverseSuffix || !doneMarks || !filePos) {
+        result = ERROR(memory_allocation);
+        goto _cleanup;
+    }
+    if (minRatio < MINRATIO) minRatio = MINRATIO;
+    memset(doneMarks, 0, bufferSize+16);
+
+    /* limit sample set size (divsufsort limitation)*/
+    if (bufferSize > ZDICT_MAX_SAMPLES_SIZE) DISPLAYLEVEL(3, "sample set too large : reduced to %u MB ...\n", (unsigned)(ZDICT_MAX_SAMPLES_SIZE>>20));
+    while (bufferSize > ZDICT_MAX_SAMPLES_SIZE) bufferSize -= fileSizes[--nbFiles];
+
+    /* sort */
+    DISPLAYLEVEL(2, "sorting %u files of total size %u MB ...\n", nbFiles, (unsigned)(bufferSize>>20));
+    {   int const divSuftSortResult = divsufsort((const unsigned char*)buffer, (int*)suffix, (int)bufferSize, 0);
+        if (divSuftSortResult != 0) { result = ERROR(GENERIC); goto _cleanup; }
+    }
+    suffix[bufferSize] = (unsigned)bufferSize;   /* leads into noise */
+    suffix0[0] = (unsigned)bufferSize;           /* leads into noise */
+    /* build reverse suffix sort */
+    {   size_t pos;
+        for (pos=0; pos < bufferSize; pos++)
+            reverseSuffix[suffix[pos]] = (U32)pos;
+        /* note filePos tracks borders between samples.
+           It's not used at this stage, but planned to become useful in a later update */
+        filePos[0] = 0;
+        for (pos=1; pos<nbFiles; pos++)
+            filePos[pos] = (U32)(filePos[pos-1] + fileSizes[pos-1]);
+    }
+
+    DISPLAYLEVEL(2, "finding patterns ... \n");
+    DISPLAYLEVEL(3, "minimum ratio : %u \n", minRatio);
+
+    {   U32 cursor; for (cursor=0; cursor < bufferSize; ) {
+            dictItem solution;
+            if (doneMarks[cursor]) { cursor++; continue; }
+            solution = ZDICT_analyzePos(doneMarks, suffix, reverseSuffix[cursor], buffer, minRatio, notificationLevel);
+            if (solution.length==0) { cursor++; continue; }
+            ZDICT_insertDictItem(dictList, dictListSize, solution, buffer);
+            cursor += solution.length;
+            DISPLAYUPDATE(2, "\r%4.2f %% \r", (double)cursor / (double)bufferSize * 100.0);
+    }   }
+
+_cleanup:
+    free(suffix0);
+    free(reverseSuffix);
+    free(doneMarks);
+    free(filePos);
+    return result;
+}
+
+
+static void ZDICT_fillNoise(void* buffer, size_t length)
+{
+    unsigned const prime1 = 2654435761U;
+    unsigned const prime2 = 2246822519U;
+    unsigned acc = prime1;
+    size_t p=0;
+    for (p=0; p<length; p++) {
+        acc *= prime2;
+        ((unsigned char*)buffer)[p] = (unsigned char)(acc >> 21);
+    }
+}
+
+
+typedef struct
+{
+    ZSTD_CDict* dict;    /* dictionary */
+    ZSTD_CCtx* zc;     /* working context */
+    void* workPlace;   /* must be ZSTD_BLOCKSIZE_MAX allocated */
+} EStats_ress_t;
+
+#define MAXREPOFFSET 1024
+
+static void ZDICT_countEStats(EStats_ress_t esr, const ZSTD_parameters* params,
+                              unsigned* countLit, unsigned* offsetcodeCount, unsigned* matchlengthCount, unsigned* litlengthCount, U32* repOffsets,
+                              const void* src, size_t srcSize,
+                              U32 notificationLevel)
+{
+    size_t const blockSizeMax = MIN (ZSTD_BLOCKSIZE_MAX, 1 << params->cParams.windowLog);
+    size_t cSize;
+
+    if (srcSize > blockSizeMax) srcSize = blockSizeMax;   /* protection vs large samples */
+    {   size_t const errorCode = ZSTD_compressBegin_usingCDict_deprecated(esr.zc, esr.dict);
+        if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_compressBegin_usingCDict failed \n"); return; }
+
+    }
+    cSize = ZSTD_compressBlock_deprecated(esr.zc, esr.workPlace, ZSTD_BLOCKSIZE_MAX, src, srcSize);
+    if (ZSTD_isError(cSize)) { DISPLAYLEVEL(3, "warning : could not compress sample size %u \n", (unsigned)srcSize); return; }
+
+    if (cSize) {  /* if == 0; block is not compressible */
+        const SeqStore_t* const seqStorePtr = ZSTD_getSeqStore(esr.zc);
+
+        /* literals stats */
+        {   const BYTE* bytePtr;
+            for(bytePtr = seqStorePtr->litStart; bytePtr < seqStorePtr->lit; bytePtr++)
+                countLit[*bytePtr]++;
+        }
+
+        /* seqStats */
+        {   U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+            ZSTD_seqToCodes(seqStorePtr);
+
+            {   const BYTE* codePtr = seqStorePtr->ofCode;
+                U32 u;
+                for (u=0; u<nbSeq; u++) offsetcodeCount[codePtr[u]]++;
+            }
+
+            {   const BYTE* codePtr = seqStorePtr->mlCode;
+                U32 u;
+                for (u=0; u<nbSeq; u++) matchlengthCount[codePtr[u]]++;
+            }
+
+            {   const BYTE* codePtr = seqStorePtr->llCode;
+                U32 u;
+                for (u=0; u<nbSeq; u++) litlengthCount[codePtr[u]]++;
+            }
+
+            if (nbSeq >= 2) { /* rep offsets */
+                const SeqDef* const seq = seqStorePtr->sequencesStart;
+                U32 offset1 = seq[0].offBase - ZSTD_REP_NUM;
+                U32 offset2 = seq[1].offBase - ZSTD_REP_NUM;
+                if (offset1 >= MAXREPOFFSET) offset1 = 0;
+                if (offset2 >= MAXREPOFFSET) offset2 = 0;
+                repOffsets[offset1] += 3;
+                repOffsets[offset2] += 1;
+    }   }   }
+}
+
+static size_t ZDICT_totalSampleSize(const size_t* fileSizes, unsigned nbFiles)
+{
+    size_t total=0;
+    unsigned u;
+    for (u=0; u<nbFiles; u++) total += fileSizes[u];
+    return total;
+}
+
+typedef struct { U32 offset; U32 count; } offsetCount_t;
+
+static void ZDICT_insertSortCount(offsetCount_t table[ZSTD_REP_NUM+1], U32 val, U32 count)
+{
+    U32 u;
+    table[ZSTD_REP_NUM].offset = val;
+    table[ZSTD_REP_NUM].count = count;
+    for (u=ZSTD_REP_NUM; u>0; u--) {
+        offsetCount_t tmp;
+        if (table[u-1].count >= table[u].count) break;
+        tmp = table[u-1];
+        table[u-1] = table[u];
+        table[u] = tmp;
+    }
+}
+
+/* ZDICT_flatLit() :
+ * rewrite `countLit` to contain a mostly flat but still compressible distribution of literals.
+ * necessary to avoid generating a non-compressible distribution that HUF_writeCTable() cannot encode.
+ */
+static void ZDICT_flatLit(unsigned* countLit)
+{
+    int u;
+    for (u=1; u<256; u++) countLit[u] = 2;
+    countLit[0]   = 4;
+    countLit[253] = 1;
+    countLit[254] = 1;
+}
+
+#define OFFCODE_MAX 30  /* only applicable to first block */
+static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
+                                   int compressionLevel,
+                             const void*  srcBuffer, const size_t* fileSizes, unsigned nbFiles,
+                             const void* dictBuffer, size_t  dictBufferSize,
+                                   unsigned notificationLevel)
+{
+    unsigned countLit[256];
+    HUF_CREATE_STATIC_CTABLE(hufTable, 255);
+    unsigned offcodeCount[OFFCODE_MAX+1];
+    short offcodeNCount[OFFCODE_MAX+1];
+    U32 offcodeMax = ZSTD_highbit32((U32)(dictBufferSize + 128 KB));
+    unsigned matchLengthCount[MaxML+1];
+    short matchLengthNCount[MaxML+1];
+    unsigned litLengthCount[MaxLL+1];
+    short litLengthNCount[MaxLL+1];
+    U32 repOffset[MAXREPOFFSET];
+    offsetCount_t bestRepOffset[ZSTD_REP_NUM+1];
+    EStats_ress_t esr = { NULL, NULL, NULL };
+    ZSTD_parameters params;
+    U32 u, huffLog = 11, Offlog = OffFSELog, mlLog = MLFSELog, llLog = LLFSELog, total;
+    size_t pos = 0, errorCode;
+    size_t eSize = 0;
+    size_t const totalSrcSize = ZDICT_totalSampleSize(fileSizes, nbFiles);
+    size_t const averageSampleSize = totalSrcSize / (nbFiles + !nbFiles);
+    BYTE* dstPtr = (BYTE*)dstBuffer;
+    U32 wksp[HUF_CTABLE_WORKSPACE_SIZE_U32];
+
+    /* init */
+    DEBUGLOG(4, "ZDICT_analyzeEntropy");
+    if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionaryCreation_failed); goto _cleanup; }   /* too large dictionary */
+    for (u=0; u<256; u++) countLit[u] = 1;   /* any character must be described */
+    for (u=0; u<=offcodeMax; u++) offcodeCount[u] = 1;
+    for (u=0; u<=MaxML; u++) matchLengthCount[u] = 1;
+    for (u=0; u<=MaxLL; u++) litLengthCount[u] = 1;
+    memset(repOffset, 0, sizeof(repOffset));
+    repOffset[1] = repOffset[4] = repOffset[8] = 1;
+    memset(bestRepOffset, 0, sizeof(bestRepOffset));
+    if (compressionLevel==0) compressionLevel = ZSTD_CLEVEL_DEFAULT;
+    params = ZSTD_getParams(compressionLevel, averageSampleSize, dictBufferSize);
+
+    esr.dict = ZSTD_createCDict_advanced(dictBuffer, dictBufferSize, ZSTD_dlm_byRef, ZSTD_dct_rawContent, params.cParams, ZSTD_defaultCMem);
+    esr.zc = ZSTD_createCCtx();
+    esr.workPlace = malloc(ZSTD_BLOCKSIZE_MAX);
+    if (!esr.dict || !esr.zc || !esr.workPlace) {
+        eSize = ERROR(memory_allocation);
+        DISPLAYLEVEL(1, "Not enough memory \n");
+        goto _cleanup;
+    }
+
+    /* collect stats on all samples */
+    for (u=0; u<nbFiles; u++) {
+        ZDICT_countEStats(esr, &params,
+                          countLit, offcodeCount, matchLengthCount, litLengthCount, repOffset,
+                         (const char*)srcBuffer + pos, fileSizes[u],
+                          notificationLevel);
+        pos += fileSizes[u];
+    }
+
+    if (notificationLevel >= 4) {
+        /* writeStats */
+        DISPLAYLEVEL(4, "Offset Code Frequencies : \n");
+        for (u=0; u<=offcodeMax; u++) {
+            DISPLAYLEVEL(4, "%2u :%7u \n", u, offcodeCount[u]);
+    }   }
+
+    /* analyze, build stats, starting with literals */
+    {   size_t maxNbBits = HUF_buildCTable_wksp(hufTable, countLit, 255, huffLog, wksp, sizeof(wksp));
+        if (HUF_isError(maxNbBits)) {
+            eSize = maxNbBits;
+            DISPLAYLEVEL(1, " HUF_buildCTable error \n");
+            goto _cleanup;
+        }
+        if (maxNbBits==8) {  /* not compressible : will fail on HUF_writeCTable() */
+            DISPLAYLEVEL(2, "warning : pathological dataset : literals are not compressible : samples are noisy or too regular \n");
+            ZDICT_flatLit(countLit);  /* replace distribution by a fake "mostly flat but still compressible" distribution, that HUF_writeCTable() can encode */
+            maxNbBits = HUF_buildCTable_wksp(hufTable, countLit, 255, huffLog, wksp, sizeof(wksp));
+            assert(maxNbBits==9);
+        }
+        huffLog = (U32)maxNbBits;
+    }
+
+    /* looking for most common first offsets */
+    {   U32 offset;
+        for (offset=1; offset<MAXREPOFFSET; offset++)
+            ZDICT_insertSortCount(bestRepOffset, offset, repOffset[offset]);
+    }
+    /* note : the result of this phase should be used to better appreciate the impact on statistics */
+
+    total=0; for (u=0; u<=offcodeMax; u++) total+=offcodeCount[u];
+    errorCode = FSE_normalizeCount(offcodeNCount, Offlog, offcodeCount, total, offcodeMax, /* useLowProbCount */ 1);
+    if (FSE_isError(errorCode)) {
+        eSize = errorCode;
+        DISPLAYLEVEL(1, "FSE_normalizeCount error with offcodeCount \n");
+        goto _cleanup;
+    }
+    Offlog = (U32)errorCode;
+
+    total=0; for (u=0; u<=MaxML; u++) total+=matchLengthCount[u];
+    errorCode = FSE_normalizeCount(matchLengthNCount, mlLog, matchLengthCount, total, MaxML, /* useLowProbCount */ 1);
+    if (FSE_isError(errorCode)) {
+        eSize = errorCode;
+        DISPLAYLEVEL(1, "FSE_normalizeCount error with matchLengthCount \n");
+        goto _cleanup;
+    }
+    mlLog = (U32)errorCode;
+
+    total=0; for (u=0; u<=MaxLL; u++) total+=litLengthCount[u];
+    errorCode = FSE_normalizeCount(litLengthNCount, llLog, litLengthCount, total, MaxLL, /* useLowProbCount */ 1);
+    if (FSE_isError(errorCode)) {
+        eSize = errorCode;
+        DISPLAYLEVEL(1, "FSE_normalizeCount error with litLengthCount \n");
+        goto _cleanup;
+    }
+    llLog = (U32)errorCode;
+
+    /* write result to buffer */
+    {   size_t const hhSize = HUF_writeCTable_wksp(dstPtr, maxDstSize, hufTable, 255, huffLog, wksp, sizeof(wksp));
+        if (HUF_isError(hhSize)) {
+            eSize = hhSize;
+            DISPLAYLEVEL(1, "HUF_writeCTable error \n");
+            goto _cleanup;
+        }
+        dstPtr += hhSize;
+        maxDstSize -= hhSize;
+        eSize += hhSize;
+    }
+
+    {   size_t const ohSize = FSE_writeNCount(dstPtr, maxDstSize, offcodeNCount, OFFCODE_MAX, Offlog);
+        if (FSE_isError(ohSize)) {
+            eSize = ohSize;
+            DISPLAYLEVEL(1, "FSE_writeNCount error with offcodeNCount \n");
+            goto _cleanup;
+        }
+        dstPtr += ohSize;
+        maxDstSize -= ohSize;
+        eSize += ohSize;
+    }
+
+    {   size_t const mhSize = FSE_writeNCount(dstPtr, maxDstSize, matchLengthNCount, MaxML, mlLog);
+        if (FSE_isError(mhSize)) {
+            eSize = mhSize;
+            DISPLAYLEVEL(1, "FSE_writeNCount error with matchLengthNCount \n");
+            goto _cleanup;
+        }
+        dstPtr += mhSize;
+        maxDstSize -= mhSize;
+        eSize += mhSize;
+    }
+
+    {   size_t const lhSize = FSE_writeNCount(dstPtr, maxDstSize, litLengthNCount, MaxLL, llLog);
+        if (FSE_isError(lhSize)) {
+            eSize = lhSize;
+            DISPLAYLEVEL(1, "FSE_writeNCount error with litlengthNCount \n");
+            goto _cleanup;
+        }
+        dstPtr += lhSize;
+        maxDstSize -= lhSize;
+        eSize += lhSize;
+    }
+
+    if (maxDstSize<12) {
+        eSize = ERROR(dstSize_tooSmall);
+        DISPLAYLEVEL(1, "not enough space to write RepOffsets \n");
+        goto _cleanup;
+    }
+# if 0
+    MEM_writeLE32(dstPtr+0, bestRepOffset[0].offset);
+    MEM_writeLE32(dstPtr+4, bestRepOffset[1].offset);
+    MEM_writeLE32(dstPtr+8, bestRepOffset[2].offset);
+#else
+    /* at this stage, we don't use the result of "most common first offset",
+     * as the impact of statistics is not properly evaluated */
+    MEM_writeLE32(dstPtr+0, repStartValue[0]);
+    MEM_writeLE32(dstPtr+4, repStartValue[1]);
+    MEM_writeLE32(dstPtr+8, repStartValue[2]);
+#endif
+    eSize += 12;
+
+_cleanup:
+    ZSTD_freeCDict(esr.dict);
+    ZSTD_freeCCtx(esr.zc);
+    free(esr.workPlace);
+
+    return eSize;
+}
+
+
+/**
+ * @returns the maximum repcode value
+ */
+static U32 ZDICT_maxRep(U32 const reps[ZSTD_REP_NUM])
+{
+    U32 maxRep = reps[0];
+    int r;
+    for (r = 1; r < ZSTD_REP_NUM; ++r)
+        maxRep = MAX(maxRep, reps[r]);
+    return maxRep;
+}
+
+size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
+                          const void* customDictContent, size_t dictContentSize,
+                          const void* samplesBuffer, const size_t* samplesSizes,
+                          unsigned nbSamples, ZDICT_params_t params)
+{
+    size_t hSize;
+#define HBUFFSIZE 256   /* should prove large enough for all entropy headers */
+    BYTE header[HBUFFSIZE];
+    int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
+    U32 const notificationLevel = params.notificationLevel;
+    /* The final dictionary content must be at least as large as the largest repcode */
+    size_t const minContentSize = (size_t)ZDICT_maxRep(repStartValue);
+    size_t paddingSize;
+
+    /* check conditions */
+    DEBUGLOG(4, "ZDICT_finalizeDictionary");
+    if (dictBufferCapacity < dictContentSize) return ERROR(dstSize_tooSmall);
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) return ERROR(dstSize_tooSmall);
+
+    /* dictionary header */
+    MEM_writeLE32(header, ZSTD_MAGIC_DICTIONARY);
+    {   U64 const randomID = XXH64(customDictContent, dictContentSize, 0);
+        U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
+        U32 const dictID = params.dictID ? params.dictID : compliantID;
+        MEM_writeLE32(header+4, dictID);
+    }
+    hSize = 8;
+
+    /* entropy tables */
+    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
+    DISPLAYLEVEL(2, "statistics ... \n");
+    {   size_t const eSize = ZDICT_analyzeEntropy(header+hSize, HBUFFSIZE-hSize,
+                                  compressionLevel,
+                                  samplesBuffer, samplesSizes, nbSamples,
+                                  customDictContent, dictContentSize,
+                                  notificationLevel);
+        if (ZDICT_isError(eSize)) return eSize;
+        hSize += eSize;
+    }
+
+    /* Shrink the content size if it doesn't fit in the buffer */
+    if (hSize + dictContentSize > dictBufferCapacity) {
+        dictContentSize = dictBufferCapacity - hSize;
+    }
+
+    /* Pad the dictionary content with zeros if it is too small */
+    if (dictContentSize < minContentSize) {
+        RETURN_ERROR_IF(hSize + minContentSize > dictBufferCapacity, dstSize_tooSmall,
+                        "dictBufferCapacity too small to fit max repcode");
+        paddingSize = minContentSize - dictContentSize;
+    } else {
+        paddingSize = 0;
+    }
+
+    {
+        size_t const dictSize = hSize + paddingSize + dictContentSize;
+
+        /* The dictionary consists of the header, optional padding, and the content.
+         * The padding comes before the content because the "best" position in the
+         * dictionary is the last byte.
+         */
+        BYTE* const outDictHeader = (BYTE*)dictBuffer;
+        BYTE* const outDictPadding = outDictHeader + hSize;
+        BYTE* const outDictContent = outDictPadding + paddingSize;
+
+        assert(dictSize <= dictBufferCapacity);
+        assert(outDictContent + dictContentSize == (BYTE*)dictBuffer + dictSize);
+
+        /* First copy the customDictContent into its final location.
+         * `customDictContent` and `dictBuffer` may overlap, so we must
+         * do this before any other writes into the output buffer.
+         * Then copy the header & padding into the output buffer.
+         */
+        memmove(outDictContent, customDictContent, dictContentSize);
+        memcpy(outDictHeader, header, hSize);
+        memset(outDictPadding, 0, paddingSize);
+
+        return dictSize;
+    }
+}
+
+
+static size_t ZDICT_addEntropyTablesFromBuffer_advanced(
+        void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
+        const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+        ZDICT_params_t params)
+{
+    int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
+    U32 const notificationLevel = params.notificationLevel;
+    size_t hSize = 8;
+
+    /* calculate entropy tables */
+    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
+    DISPLAYLEVEL(2, "statistics ... \n");
+    {   size_t const eSize = ZDICT_analyzeEntropy((char*)dictBuffer+hSize, dictBufferCapacity-hSize,
+                                  compressionLevel,
+                                  samplesBuffer, samplesSizes, nbSamples,
+                                  (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize,
+                                  notificationLevel);
+        if (ZDICT_isError(eSize)) return eSize;
+        hSize += eSize;
+    }
+
+    /* add dictionary header (after entropy tables) */
+    MEM_writeLE32(dictBuffer, ZSTD_MAGIC_DICTIONARY);
+    {   U64 const randomID = XXH64((char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, 0);
+        U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
+        U32 const dictID = params.dictID ? params.dictID : compliantID;
+        MEM_writeLE32((char*)dictBuffer+4, dictID);
+    }
+
+    if (hSize + dictContentSize < dictBufferCapacity)
+        memmove((char*)dictBuffer + hSize, (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize);
+    return MIN(dictBufferCapacity, hSize+dictContentSize);
+}
+
+/*! ZDICT_trainFromBuffer_unsafe_legacy() :
+*   Warning : `samplesBuffer` must be followed by noisy guard band !!!
+*   @return : size of dictionary, or an error code which can be tested with ZDICT_isError()
+*/
+static size_t ZDICT_trainFromBuffer_unsafe_legacy(
+                            void* dictBuffer, size_t maxDictSize,
+                            const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                            ZDICT_legacy_params_t params)
+{
+    U32 const dictListSize = MAX(MAX(DICTLISTSIZE_DEFAULT, nbSamples), (U32)(maxDictSize/16));
+    dictItem* const dictList = (dictItem*)malloc(dictListSize * sizeof(*dictList));
+    unsigned const selectivity = params.selectivityLevel == 0 ? g_selectivity_default : params.selectivityLevel;
+    unsigned const minRep = (selectivity > 30) ? MINRATIO : nbSamples >> selectivity;
+    size_t const targetDictSize = maxDictSize;
+    size_t const samplesBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
+    size_t dictSize = 0;
+    U32 const notificationLevel = params.zParams.notificationLevel;
+
+    /* checks */
+    if (!dictList) return ERROR(memory_allocation);
+    if (maxDictSize < ZDICT_DICTSIZE_MIN) { free(dictList); return ERROR(dstSize_tooSmall); }   /* requested dictionary size is too small */
+    if (samplesBuffSize < ZDICT_MIN_SAMPLES_SIZE) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* not enough source to create dictionary */
+
+    /* init */
+    ZDICT_initDictItem(dictList);
+
+    /* build dictionary */
+    ZDICT_trainBuffer_legacy(dictList, dictListSize,
+                       samplesBuffer, samplesBuffSize,
+                       samplesSizes, nbSamples,
+                       minRep, notificationLevel);
+
+    /* display best matches */
+    if (params.zParams.notificationLevel>= 3) {
+        unsigned const nb = MIN(25, dictList[0].pos);
+        unsigned const dictContentSize = ZDICT_dictSize(dictList);
+        unsigned u;
+        DISPLAYLEVEL(3, "\n %u segments found, of total size %u \n", (unsigned)dictList[0].pos-1, dictContentSize);
+        DISPLAYLEVEL(3, "list %u best segments \n", nb-1);
+        for (u=1; u<nb; u++) {
+            unsigned const pos = dictList[u].pos;
+            unsigned const length = dictList[u].length;
+            U32 const printedLength = MIN(40, length);
+            if ((pos > samplesBuffSize) || ((pos + length) > samplesBuffSize)) {
+                free(dictList);
+                return ERROR(GENERIC);   /* should never happen */
+            }
+            DISPLAYLEVEL(3, "%3u:%3u bytes at pos %8u, savings %7u bytes |",
+                         u, length, pos, (unsigned)dictList[u].savings);
+            ZDICT_printHex((const char*)samplesBuffer+pos, printedLength);
+            DISPLAYLEVEL(3, "| \n");
+    }   }
+
+
+    /* create dictionary */
+    {   unsigned dictContentSize = ZDICT_dictSize(dictList);
+        if (dictContentSize < ZDICT_CONTENTSIZE_MIN) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* dictionary content too small */
+        if (dictContentSize < targetDictSize/4) {
+            DISPLAYLEVEL(2, "!  warning : selected content significantly smaller than requested (%u < %u) \n", dictContentSize, (unsigned)maxDictSize);
+            if (samplesBuffSize < 10 * targetDictSize)
+                DISPLAYLEVEL(2, "!  consider increasing the number of samples (total size : %u MB)\n", (unsigned)(samplesBuffSize>>20));
+            if (minRep > MINRATIO) {
+                DISPLAYLEVEL(2, "!  consider increasing selectivity to produce larger dictionary (-s%u) \n", selectivity+1);
+                DISPLAYLEVEL(2, "!  note : larger dictionaries are not necessarily better, test its efficiency on samples \n");
+            }
+        }
+
+        if ((dictContentSize > targetDictSize*3) && (nbSamples > 2*MINRATIO) && (selectivity>1)) {
+            unsigned proposedSelectivity = selectivity-1;
+            while ((nbSamples >> proposedSelectivity) <= MINRATIO) { proposedSelectivity--; }
+            DISPLAYLEVEL(2, "!  note : calculated dictionary significantly larger than requested (%u > %u) \n", dictContentSize, (unsigned)maxDictSize);
+            DISPLAYLEVEL(2, "!  consider increasing dictionary size, or produce denser dictionary (-s%u) \n", proposedSelectivity);
+            DISPLAYLEVEL(2, "!  always test dictionary efficiency on real samples \n");
+        }
+
+        /* limit dictionary size */
+        {   U32 const max = dictList->pos;   /* convention : nb of useful elts within dictList */
+            U32 currentSize = 0;
+            U32 n; for (n=1; n<max; n++) {
+                currentSize += dictList[n].length;
+                if (currentSize > targetDictSize) { currentSize -= dictList[n].length; break; }
+            }
+            dictList->pos = n;
+            dictContentSize = currentSize;
+        }
+
+        /* build dict content */
+        {   U32 u;
+            BYTE* ptr = (BYTE*)dictBuffer + maxDictSize;
+            for (u=1; u<dictList->pos; u++) {
+                U32 l = dictList[u].length;
+                ptr -= l;
+                if (ptr<(BYTE*)dictBuffer) { free(dictList); return ERROR(GENERIC); }   /* should not happen */
+                memcpy(ptr, (const char*)samplesBuffer+dictList[u].pos, l);
+        }   }
+
+        dictSize = ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, maxDictSize,
+                                                             samplesBuffer, samplesSizes, nbSamples,
+                                                             params.zParams);
+    }
+
+    /* clean up */
+    free(dictList);
+    return dictSize;
+}
+
+
+/* ZDICT_trainFromBuffer_legacy() :
+ * issue : samplesBuffer need to be followed by a noisy guard band.
+ * work around : duplicate the buffer, and add the noise */
+size_t ZDICT_trainFromBuffer_legacy(void* dictBuffer, size_t dictBufferCapacity,
+                              const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                              ZDICT_legacy_params_t params)
+{
+    size_t result;
+    void* newBuff;
+    size_t const sBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
+    if (sBuffSize < ZDICT_MIN_SAMPLES_SIZE) return 0;   /* not enough content => no dictionary */
+
+    newBuff = malloc(sBuffSize + NOISELENGTH);
+    if (!newBuff) return ERROR(memory_allocation);
+
+    memcpy(newBuff, samplesBuffer, sBuffSize);
+    ZDICT_fillNoise((char*)newBuff + sBuffSize, NOISELENGTH);   /* guard band, for end of buffer condition */
+
+    result =
+        ZDICT_trainFromBuffer_unsafe_legacy(dictBuffer, dictBufferCapacity, newBuff,
+                                            samplesSizes, nbSamples, params);
+    free(newBuff);
+    return result;
+}
+
+
+size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
+                             const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
+{
+    ZDICT_fastCover_params_t params;
+    DEBUGLOG(3, "ZDICT_trainFromBuffer");
+    memset(&params, 0, sizeof(params));
+    params.d = 8;
+    params.steps = 4;
+    /* Use default level since no compression level information is available */
+    params.zParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;
+#if defined(DEBUGLEVEL) && (DEBUGLEVEL>=1)
+    params.zParams.notificationLevel = DEBUGLEVEL;
+#endif
+    return ZDICT_optimizeTrainFromBuffer_fastCover(dictBuffer, dictBufferCapacity,
+                                               samplesBuffer, samplesSizes, nbSamples,
+                                               &params);
+}
+
+size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
+                                  const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
+{
+    ZDICT_params_t params;
+    memset(&params, 0, sizeof(params));
+    return ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, dictBufferCapacity,
+                                                     samplesBuffer, samplesSizes, nbSamples,
+                                                     params);
+}
diff --git a/lib/legacy/zstd_v02.c b/lib/legacy/zstd_v02.c
new file mode 100644
index 0000000..d1e0038
--- /dev/null
+++ b/lib/legacy/zstd_v02.c
@@ -0,0 +1,3465 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include "zstd_v02.h"
+#include "../common/compiler.h"
+#include "../common/error_private.h"
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+
+/****************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/****************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian())
+    {
+        MEM_write16(memPtr, val);
+    }
+    else
+    {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+
+/* ******************************************************************
+   bitstream
+   Part of NewGen Entropy library
+   header file (to include)
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/**********************************************
+*  bitStream decompression API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BIT_DStream_t;
+
+typedef enum { BIT_DStream_unfinished = 0,
+               BIT_DStream_endOfBuffer = 1,
+               BIT_DStream_completed = 2,
+               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
+
+
+/******************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/****************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/**********************************************************
+* bitStream decoding
+**********************************************************/
+
+/*!BIT_initDStream
+*  Initialize a BIT_DStream_t.
+*  @bitD : a pointer to an already allocated BIT_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t))   /* normal case */
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+    }
+    else
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);
+                    /* fallthrough */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);
+                    /* fallthrough */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);
+                    /* fallthrough */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24;
+                    /* fallthrough */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16;
+                    /* fallthrough */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8;
+                    /* fallthrough */
+            default:;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BIT_lookBitsFast :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBits(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BIT_readBitsFast :
+*  unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBitsFast(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BIT_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer))
+    {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BIT_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start)
+    {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
+        return BIT_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BIT_DStream_status result = BIT_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start)
+        {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BIT_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BIT_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   Error codes and messages
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef ERROR_H_MODULE
+#define ERROR_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define ERR_STATIC static inline
+#elif defined(_MSC_VER)
+#  define ERR_STATIC static __inline
+#elif defined(__GNUC__)
+#  define ERR_STATIC static __attribute__((unused))
+#else
+#  define ERR_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/******************************************
+*  Error Management
+******************************************/
+#define PREFIX(name) ZSTD_error_##name
+
+#define ERROR(name) (size_t)-PREFIX(name)
+
+#define ERROR_LIST(ITEM) \
+        ITEM(PREFIX(No_Error)) ITEM(PREFIX(GENERIC)) \
+        ITEM(PREFIX(dstSize_tooSmall)) ITEM(PREFIX(srcSize_wrong)) \
+        ITEM(PREFIX(prefix_unknown)) ITEM(PREFIX(corruption_detected)) \
+        ITEM(PREFIX(tableLog_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooSmall)) \
+        ITEM(PREFIX(maxCode))
+
+#define ERROR_GENERATE_ENUM(ENUM) ENUM,
+typedef enum { ERROR_LIST(ERROR_GENERATE_ENUM) } ERR_codes;  /* enum is exposed, to detect & handle specific errors; compare function result to -enum value */
+
+#define ERROR_CONVERTTOSTRING(STRING) #STRING,
+#define ERROR_GENERATE_STRING(EXPR) ERROR_CONVERTTOSTRING(EXPR)
+static const char* ERR_strings[] = { ERROR_LIST(ERROR_GENERATE_STRING) };
+
+ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
+
+ERR_STATIC const char* ERR_getErrorName(size_t code)
+{
+    static const char* codeError = "Unspecified error code";
+    if (ERR_isError(code)) return ERR_strings[-(int)(code)];
+    return codeError;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ERROR_H_MODULE */
+/*
+Constructor and Destructor of type FSE_CTable
+    Note that its size depends on 'tableLog' and 'maxSymbolValue' */
+typedef unsigned FSE_CTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Static allocation
+******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) (size + (size>>7))
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* You can statically allocate FSE CTable/DTable as a table of unsigned using below macro */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/******************************************
+*  FSE advanced API
+******************************************/
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
+/* build a fake FSE_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
+/* build a fake FSE_DTable, designed to always generate the same symbolValue */
+
+
+/******************************************
+*  FSE symbol decompression API
+******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+
+/******************************************
+*  FSE unsafe API
+******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/******************************************
+*  Implementation of inline functions
+******************************************/
+
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    DStatePtr->state = BIT_readBits(bitD, DTableH.tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Static allocation macros
+******************************************/
+/* Huff0 buffer bounds */
+#define HUF_CTABLEBOUND 129
+#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+#define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* static allocation of Huff0's DTable */
+#define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))  /* nb Cells; use unsigned short for X2, unsigned int for X4 */
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/******************************************
+*  Advanced functions
+******************************************/
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+static size_t HUF_decompress4X6 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* quad-symbols decoder */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+/*
+    zstd - standard compression library
+    Header File
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Version
+***************************************/
+#define ZSTD_VERSION_MAJOR    0    /* for breaking interface changes  */
+#define ZSTD_VERSION_MINOR    2    /* for new (non-breaking) interface capabilities */
+#define ZSTD_VERSION_RELEASE  2    /* for tweaks, bug-fixes, or development */
+#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTD_CCtx_s ZSTD_CCtx;   /* incomplete type */
+
+#if defined (__cplusplus)
+}
+#endif
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Streaming functions
+***************************************/
+
+typedef struct ZSTDv02_Dctx_s ZSTD_DCtx;
+
+/*
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTD_magicNumber 0xFD2FB522   /* v0.2 (current)*/
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/****************************************************************
+*  Tuning parameters
+****************************************************************/
+/* MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSE_MAX_MEMORY_USAGE 14
+#define FSE_DEFAULT_MEMORY_USAGE 13
+
+/* FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSE_MAX_SYMBOL_VALUE 255
+
+
+/****************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+
+
+/****************************************************************
+*  Byte symbol type
+****************************************************************/
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/****************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+
+/****************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+/* Function templates */
+
+#define FSE_DECODE_TYPE FSE_decode_t
+
+static U32 FSE_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+static size_t FSE_buildDTable
+(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* ptr = dt+1;
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*)ptr;
+    FSE_DTableHeader DTableH;
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSE_tableStep(tableSize);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        if (normalizedCounter[s]==-1)
+        {
+            tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        }
+        else
+        {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+        }
+    }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++)
+        {
+            tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }
+    }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++)
+        {
+            FSE_FUNCTION_TYPE symbol = (FSE_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BIT_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+        }
+    }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));   /* memcpy(), to avoid strict aliasing warnings */
+    return 0;
+}
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+/******************************************
+*  FSE helper functions
+******************************************/
+static unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+
+
+/****************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSE_abs(short a)
+{
+    return (short)(a<0 ? -a : a);
+}
+
+static size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr))
+    {
+        if (previous0)
+        {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF)
+            {
+                n0+=24;
+                if (ip < iend-5)
+                {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                }
+                else
+                {
+                    bitStream >>= 16;
+                    bitCount+=16;
+                }
+            }
+            while ((bitStream & 3) == 3)
+            {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+            {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max)
+            {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            }
+            else
+            {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSE_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold)
+            {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            {
+                if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+                {
+                    ip += bitCount>>3;
+                    bitCount &= 7;
+                }
+                else
+                {
+                    bitCount -= (int)(8 * (iend - 4 - ip));
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+            }
+        }
+    }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*********************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const cell = (FSE_decode_t*)(ptr) + 1;   /* because dt is unsigned */
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)(ptr) + 1;   /* because dt is unsigned */
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BIT_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSE_isError(errorCode)) return errorCode;
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) && (op<olimit) ; op+=4)
+    {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
+    while (1)
+    {
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state1);
+
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BIT_endOfDStream(&bitD) && FSE_endOfDState(&state1) && FSE_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+static size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+
+    /* select fast mode (static) */
+    if (DTableH.fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    errorCode = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSE_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/******************************************
+*  Helper functions
+******************************************/
+static unsigned HUF_isError(size_t code) { return ERR_isError(code); }
+
+#define HUF_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUF_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_DEFAULT_TABLELOG  HUF_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUF_MAX_SYMBOL_VALUE 255
+#if (HUF_MAX_TABLELOG > HUF_ABSOLUTEMAX_TABLELOG)
+#  error "HUF_MAX_TABLELOG is too large !"
+#endif
+
+
+
+/*********************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUF_readStats
+    Read compact Huffman tree, saved by HUF_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  /* special header */
+    {
+        if (iSize >= (242))   /* RLE */
+        {
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else   /* Incompressible */
+        {
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2)
+            {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+            }
+        }
+    }
+    else  /* header compressed with FSE (normal case) */
+    {
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSE_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUF_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++)
+    {
+        if (huffWeight[n] >= HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BIT_highbit32(weightTotal) + 1;
+    if (tableLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BIT_highbit32(rest);
+        U32 lastWeight = BIT_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/**************************/
+/* single-symbol decoding */
+/**************************/
+
+static size_t HUF_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* ptr = DTable+1;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)ptr;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(huffWeight, HUF_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof DTable, as allocated, from used size of DTable, in case of DTable re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++)
+    {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++)
+    {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUF_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BIT_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+    {
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd))
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+
+static size_t HUF_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX2* const dt = ((const HUF_DEltX2*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUF_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUF_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/***************************/
+/* double-symbols decoding */
+/***************************/
+
+static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUF_DEltX4 DElt;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)   /* note : sortedSymbols already skipped */
+    {
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUF_ABSOLUTEMAX_TABLELOG][HUF_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)
+    {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits)   /* enough room for a second symbol */
+        {
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            HUF_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+static size_t HUF_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    void* ptr = DTable;
+    HUF_DEltX4* const dt = ((HUF_DEltX4*)ptr) + 1;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        {if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+    HUF_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else
+    {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+        {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+        }
+    }
+    return 1;
+}
+
+
+#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd-7))
+    {
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+
+static size_t HUF_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX4* const dt = ((const HUF_DEltX4*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* quad-symbol decoding           */
+/**********************************/
+typedef struct { BYTE nbBits; BYTE nbBytes; } HUF_DDescX6;
+typedef union { BYTE byte[4]; U32 sequence; } HUF_DSeqX6;
+
+/* recursive, up to level 3; may benefit from <template>-like strategy to nest each level inline */
+static void HUF_fillDTableX6LevelN(HUF_DDescX6* DDescription, HUF_DSeqX6* DSequence, int sizeLog,
+                           const rankVal_t rankValOrigin, const U32 consumed, const int minWeight, const U32 maxWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize, const U32* rankStart,
+                           const U32 nbBitsBaseline, HUF_DSeqX6 baseSeq, HUF_DDescX6 DDesc)
+{
+    const int scaleLog = nbBitsBaseline - sizeLog;   /* note : targetLog >= (nbBitsBaseline-1), hence scaleLog <= 1 */
+    const int minBits  = nbBitsBaseline - maxWeight;
+    const U32 level = DDesc.nbBytes;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 symbolStartPos, s;
+
+    /* local rankVal, will be modified */
+    memcpy(rankVal, rankValOrigin[consumed], sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i;
+        const U32 skipSize = rankVal[minWeight];
+        for (i = 0; i < skipSize; i++)
+        {
+            DSequence[i] = baseSeq;
+            DDescription[i] = DDesc;
+        }
+    }
+
+    /* fill DTable */
+    DDesc.nbBytes++;
+    symbolStartPos = rankStart[minWeight];
+    for (s=symbolStartPos; s<sortedListSize; s++)
+    {
+        const BYTE symbol = sortedSymbols[s].symbol;
+        const U32  weight = sortedSymbols[s].weight;   /* >= 1 (sorted) */
+        const int  nbBits = nbBitsBaseline - weight;   /* >= 1 (by construction) */
+        const int  totalBits = consumed+nbBits;
+        const U32  start  = rankVal[weight];
+        const U32  length = 1 << (sizeLog-nbBits);
+        baseSeq.byte[level] = symbol;
+        DDesc.nbBits = (BYTE)totalBits;
+
+        if ((level<3) && (sizeLog-totalBits >= minBits))   /* enough room for another symbol */
+        {
+            int nextMinWeight = totalBits + scaleLog;
+            if (nextMinWeight < 1) nextMinWeight = 1;
+            HUF_fillDTableX6LevelN(DDescription+start, DSequence+start, sizeLog-nbBits,
+                           rankValOrigin, totalBits, nextMinWeight, maxWeight,
+                           sortedSymbols, sortedListSize, rankStart,
+                           nbBitsBaseline, baseSeq, DDesc);   /* recursive (max : level 3) */
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            for (i = start; i < end; i++)
+            {
+                DDescription[i] = DDesc;
+                DSequence[i] = baseSeq;
+            }
+        }
+        rankVal[weight] += length;
+    }
+}
+
+
+/* note : same preparation as X4 */
+static size_t HUF_readDTableX6 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    rankVal_t rankVal;
+    const U32 memLog = DTable[0];
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        { if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+
+    /* fill tables */
+    {
+        void* ptr = DTable+1;
+        HUF_DDescX6* DDescription = (HUF_DDescX6*)(ptr);
+        void* dSeqStart = DTable + 1 + ((size_t)1<<(memLog-1));
+        HUF_DSeqX6* DSequence = (HUF_DSeqX6*)(dSeqStart);
+        HUF_DSeqX6 DSeq;
+        HUF_DDescX6 DDesc;
+        DSeq.sequence = 0;
+        DDesc.nbBits = 0;
+        DDesc.nbBytes = 0;
+        HUF_fillDTableX6LevelN(DDescription, DSequence, memLog,
+                       (const U32 (*)[HUF_ABSOLUTEMAX_TABLELOG + 1])rankVal, 0, 1, maxW,
+                       sortedSymbol, sizeOfSort, rankStart0,
+                       tableLog+1, DSeq, DDesc);
+    }
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX6(void* op, BIT_DStream_t* DStream, const HUF_DDescX6* dd, const HUF_DSeqX6* ds, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, ds+val, sizeof(HUF_DSeqX6));
+    BIT_skipBits(DStream, dd[val].nbBits);
+    return dd[val].nbBytes;
+}
+
+static U32 HUF_decodeLastSymbolsX6(void* op, const U32 maxL, BIT_DStream_t* DStream,
+                                  const HUF_DDescX6* dd, const HUF_DSeqX6* ds, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    U32 length = dd[val].nbBytes;
+    if (length <= maxL)
+    {
+        memcpy(op, ds+val, length);
+        BIT_skipBits(DStream, dd[val].nbBits);
+        return length;
+    }
+    memcpy(op, ds+val, maxL);
+    if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+    {
+        BIT_skipBits(DStream, dd[val].nbBits);
+        if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+            DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }
+    return maxL;
+}
+
+
+#define HUF_DECODE_SYMBOLX6_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX6(ptr, DStreamPtr, dd, ds, dtLog)
+
+#define HUF_DECODE_SYMBOLX6_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX6_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX6_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX6_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX6(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const U32* DTable, const U32 dtLog)
+{
+    const void* ddPtr = DTable+1;
+    const HUF_DDescX6* dd = (const HUF_DDescX6*)(ddPtr);
+    const void* dsPtr = DTable + 1 + ((size_t)1<<(dtLog-1));
+    const HUF_DSeqX6* ds = (const HUF_DSeqX6*)(dsPtr);
+    BYTE* const pStart = p;
+
+    /* up to 16 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-16))
+    {
+        HUF_DECODE_SYMBOLX6_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX6_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX6_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX6_0(p, bitDPtr);
+    }
+
+    /* closer to the end, up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+        HUF_DECODE_SYMBOLX6_0(p, bitDPtr);
+
+    while (p <= pEnd-4)
+        HUF_DECODE_SYMBOLX6_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    while (p < pEnd)
+        p += HUF_decodeLastSymbolsX6(p, (U32)(pEnd-p), bitDPtr, dd, ds, dtLog);
+
+    return p-pStart;
+}
+
+
+
+static size_t HUF_decompress4X6_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const U32 dtLog = DTable[0];
+        const void* ddPtr = DTable+1;
+        const HUF_DDescX6* dd = (const HUF_DDescX6*)(ddPtr);
+        const void* dsPtr = DTable + 1 + ((size_t)1<<(dtLog-1));
+        const HUF_DSeqX6* ds = (const HUF_DSeqX6*)(dsPtr);
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-64 symbols per loop (4-16 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (op3 <= opStart4) && (endSignal==BIT_DStream_unfinished) && (op4<=(oend-16)) ; )
+        {
+            HUF_DECODE_SYMBOLX6_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX6_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX6_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX6_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX6(op1, &bitD1, opStart2, DTable, dtLog);
+        HUF_decodeStreamX6(op2, &bitD2, opStart3, DTable, dtLog);
+        HUF_decodeStreamX6(op3, &bitD3, opStart4, DTable, dtLog);
+        HUF_decodeStreamX6(op4, &bitD4, oend,     DTable, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X6 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX6(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX6 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X6_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* Generic decompression selector */
+/**********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUF_decompress4X2, HUF_decompress4X4, HUF_decompress4X6 };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+    if (Dtime[2] < Dtime[algoNb]) algoNb = 2;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    //return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUF_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+    //return HUF_decompress4X6(dst, dstSize, cSrc, cSrcSize);   /* multi-streams quad-symbols decoding */
+}
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+*  MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*/
+#define ZSTD_MEMORY_USAGE 17
+
+/*!
+ * HEAPMODE :
+ * Select how default compression functions will allocate memory for their hash table,
+ * in memory stack (0, fastest), or in memory heap (1, requires malloc())
+ * Note that compression context is fairly large, as a consequence heap memory is recommended.
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif /* ZSTD_HEAPMODE */
+
+/*!
+*  LEGACY_SUPPORT :
+*  decompressor can decode older formats (starting from Zstd 0.1+)
+*/
+#ifndef ZSTD_LEGACY_SUPPORT
+#  define ZSTD_LEGACY_SUPPORT 1
+#endif
+
+
+/* *******************************************************
+*  Includes
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug : printf */
+
+
+/* *******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef __AVX2__
+#  include <immintrin.h>   /* AVX2 intrinsics */
+#endif
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/* *******************************************************
+*  Constants
+*********************************************************/
+#define HASH_LOG (ZSTD_MEMORY_USAGE - 2)
+#define HASH_TABLESIZE (1 << HASH_LOG)
+#define HASH_MASK (HASH_TABLESIZE - 1)
+
+#define KNUTH 2654435761
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+#define MIN_SEQUENCES_SIZE (2 /*seqNb*/ + 2 /*dumps*/ + 3 /*seqTables*/ + 1 /*bitStream*/)
+#define MIN_CBLOCK_SIZE (3 /*litCSize*/ + MIN_SEQUENCES_SIZE)
+#define IS_RAW BIT0
+#define IS_RLE BIT1
+
+#define WORKPLACESIZE (BLOCKSIZE*3)
+#define MINMATCH 4
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxML  ((1<<MLbits )-1)
+#define MaxLL  ((1<<LLbits )-1)
+#define MaxOff   31
+#define LitFSELog  11
+#define MLFSELog   10
+#define LLFSELog   10
+#define OffFSELog   9
+#define MAX(a,b) ((a)<(b)?(b):(a))
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define LITERAL_NOENTROPY 63
+#define COMMAND_NOENTROPY 7   /* to remove */
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+static const size_t ZSTD_blockHeaderSize = 3;
+static const size_t ZSTD_frameHeaderSize = 4;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void   ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+static void   ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTD_wildcopy : custom version of memcpy(), can copy up to 7-8 bytes too many */
+static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do COPY8(op, ip) while (op < oend);
+}
+
+
+/* **************************************
+*  Local structures
+****************************************/
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* offCode;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* litLengthStart;
+    BYTE* litLength;
+    BYTE* matchLengthStart;
+    BYTE* matchLength;
+    BYTE* dumpsStart;
+    BYTE* dumps;
+} SeqStore_t;
+
+
+/* *************************************
+*  Error Management
+***************************************/
+/*! ZSTD_isError
+*   tells if a return value is an error code */
+static unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
+
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+struct ZSTDv02_Dctx_s
+{
+    U32 LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
+    U32 OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    U32 MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
+    void* previousDstEnd;
+    void* base;
+    size_t expected;
+    blockType_t bType;
+    U32 phase;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + 8 /* margin for wildcopy */];
+};   /* typedef'd to ZSTD_Dctx within "zstd_static.h" */
+
+
+static size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3) return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+static size_t ZSTD_copyUncompressedBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    if (srcSize > 0) {
+        memcpy(dst, src, srcSize);
+    }
+    return srcSize;
+}
+
+
+/** ZSTD_decompressLiterals
+    @return : nb of bytes read from src, or an error code*/
+static size_t ZSTD_decompressLiterals(void* dst, size_t* maxDstSizePtr,
+                                const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    const size_t litSize = (MEM_readLE32(src) & 0x1FFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+    const size_t litCSize = (MEM_readLE32(ip+2) & 0xFFFFFF) >> 5;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+
+    if (litSize > *maxDstSizePtr) return ERROR(corruption_detected);
+    if (litCSize + 5 > srcSize) return ERROR(corruption_detected);
+
+    if (HUF_isError(HUF_decompress(dst, litSize, ip+5, litCSize))) return ERROR(corruption_detected);
+
+    *maxDstSizePtr = litSize;
+    return litCSize + 5;
+}
+
+
+/** ZSTD_decodeLiteralsBlock
+    @return : nb of bytes read from src (< srcSize )*/
+static size_t ZSTD_decodeLiteralsBlock(void* ctx,
+                          const void* src, size_t srcSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* const istart = (const BYTE* const)src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(*istart & 3)
+    {
+    default:
+    case 0:
+        {
+            size_t litSize = BLOCKSIZE;
+            const size_t readSize = ZSTD_decompressLiterals(dctx->litBuffer, &litSize, src, srcSize);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, 8);
+            return readSize;   /* works if it's an error too */
+        }
+    case IS_RAW:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > srcSize-11)   /* risk of reading too far with wildcopy */
+            {
+                if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+                if (litSize > srcSize-3) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, 8);
+                return litSize+3;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+3;
+            dctx->litSize = litSize;
+            return litSize+3;
+        }
+    case IS_RLE:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[3], litSize + 8);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return 4;
+        }
+    }
+}
+
+
+static size_t ZSTD_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
+                         const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < 5) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = MEM_readLE16(ip); ip+=2;
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2)
+    {
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    }
+    else
+    {
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL and MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case bt_rle :
+            LLlog = 0;
+            FSE_buildDTable_rle(DTableLL, *ip++); break;
+        case bt_raw :
+            LLlog = LLbits;
+            FSE_buildDTable_raw(DTableLL, LLbits); break;
+        default :
+            {   U32 max = MaxLL;
+                headerSize = FSE_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case bt_rle :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case bt_raw :
+            Offlog = Offbits;
+            FSE_buildDTable_raw(DTableOffb, Offbits); break;
+        default :
+            {   U32 max = MaxOff;
+                headerSize = FSE_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case bt_rle :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableML, *ip++); break;
+        case bt_raw :
+            MLlog = MLbits;
+            FSE_buildDTable_raw(DTableML, MLbits); break;
+        default :
+            {   U32 max = MaxML;
+                headerSize = FSE_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t offset;
+    size_t matchLength;
+} seq_t;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    FSE_DState_t stateLL;
+    FSE_DState_t stateOffb;
+    FSE_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    seqState->prevOffset = seq->offset;
+    if (litLength == MaxLL)
+    {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) litLength += add;
+        else if (dumps + 3 <= de)
+        {
+            litLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {
+        static const size_t offsetPrefix[MaxOff+1] = {  /* note : size_t faster than U32 */
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode, nbBits;
+        offsetCode = FSE_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));   /* <= maxOff, by table construction */
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BIT_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* cmove */
+    }
+
+    /* MatchLength */
+    matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML)
+    {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) matchLength += add;
+        else if (dumps + 3 <= de)
+        {
+            matchLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+}
+
+
+static size_t ZSTD_execSequence(BYTE* op,
+                                seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                BYTE* const base, BYTE* const oend)
+{
+    static const int dec32table[] = {0, 1, 2, 1, 4, 4, 4, 4};   /* added */
+    static const int dec64table[] = {8, 8, 8, 7, 8, 9,10,11};   /* subtracted */
+    const BYTE* const ostart = op;
+    BYTE* const oLitEnd = op + sequence.litLength;
+    BYTE* const oMatchEnd = op + sequence.litLength + sequence.matchLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use the pointer check */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+    if (sequence.offset > (U32)(oLitEnd - base)) return ERROR(corruption_detected);
+
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTD_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    {
+        const BYTE* match = op - sequence.offset;
+
+        /* check */
+        if (sequence.offset > (size_t)op) return ERROR(corruption_detected);   /* address space overflow test (this test seems kept by clang optimizer) */
+        //if (match > op) return ERROR(corruption_detected);   /* address space overflow test (is clang optimizer removing this test ?) */
+        if (match < base) return ERROR(corruption_detected);
+
+        /* close range match, overlap */
+        if (sequence.offset < 8)
+        {
+            const int dec64 = dec64table[sequence.offset];
+            op[0] = match[0];
+            op[1] = match[1];
+            op[2] = match[2];
+            op[3] = match[3];
+            match += dec32table[sequence.offset];
+            ZSTD_copy4(op+4, match);
+            match -= dec64;
+        }
+        else
+        {
+            ZSTD_copy8(op, match);
+        }
+        op += 8; match += 8;
+
+        if (oMatchEnd > oend-(16-MINMATCH))
+        {
+            if (op < oend_8)
+            {
+                ZSTD_wildcopy(op, match, oend_8 - op);
+                match += oend_8 - op;
+                op = oend_8;
+            }
+            while (op < oMatchEnd) *op++ = *match++;
+        }
+        else
+        {
+            ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+        }
+    }
+
+    return oMatchEnd - ostart;
+}
+
+static size_t ZSTD_decompressSequences(
+                               void* ctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    BYTE* const base = (BYTE*) (dctx->base);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTD_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, iend-ip);
+    if (ZSTD_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = 1;
+        errorCode = BIT_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (nbSeq>0) ; )
+        {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTD_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTD_execSequence(op, sequence, &litPtr, litEnd, base, oend);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if ( !BIT_endOfDStream(&(seqState.DStream)) ) return ERROR(corruption_detected);   /* requested too much : data is corrupted */
+        if (nbSeq<0) return ERROR(corruption_detected);   /* requested too many sequences : data is corrupted */
+
+        /* last literal segment */
+        {
+            size_t lastLLSize = litEnd - litPtr;
+            if (litPtr > litEnd) return ERROR(corruption_detected);
+            if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+            if (lastLLSize > 0) {
+                if (op != litPtr) memmove(op, litPtr, lastLLSize);
+                op += lastLLSize;
+            }
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static size_t ZSTD_decompressBlock(
+                            void* ctx,
+                            void* dst, size_t maxDstSize,
+                      const void* src, size_t srcSize)
+{
+    /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    /* Decode literals sub-block */
+    size_t litCSize = ZSTD_decodeLiteralsBlock(ctx, src, srcSize);
+    if (ZSTD_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTD_decompressSequences(ctx, dst, maxDstSize, ip, srcSize);
+}
+
+
+static size_t ZSTD_decompressDCtx(void* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTD_decompressBlock(ctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTD_copyUncompressedBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTD_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+static size_t ZSTD_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    ZSTD_DCtx ctx;
+    ctx.base = dst;
+    return ZSTD_decompressDCtx(&ctx, dst, maxDstSize, src, srcSize);
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+static void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv02_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+        return;
+    }
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+        return;
+    }
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * BLOCKSIZE;
+}
+
+/*******************************
+*  Streaming Decompression API
+*******************************/
+
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx)
+{
+    dctx->expected = ZSTD_frameHeaderSize;
+    dctx->phase = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    return 0;
+}
+
+static ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)malloc(sizeof(ZSTD_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTD_resetDCtx(dctx);
+    return dctx;
+}
+
+static size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    free(dctx);
+    return 0;
+}
+
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != ctx->expected) return ERROR(srcSize_wrong);
+    if (dst != ctx->previousDstEnd)  /* not contiguous */
+        ctx->base = dst;
+
+    /* Decompress : frame header */
+    if (ctx->phase == 0)
+    {
+        /* Check frame magic header */
+        U32 magicNumber = MEM_readLE32(src);
+        if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        return 0;
+    }
+
+    /* Decompress : block header */
+    if (ctx->phase == 1)
+    {
+        blockProperties_t bp;
+        size_t blockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+        if (ZSTD_isError(blockSize)) return blockSize;
+        if (bp.blockType == bt_end)
+        {
+            ctx->expected = 0;
+            ctx->phase = 0;
+        }
+        else
+        {
+            ctx->expected = blockSize;
+            ctx->bType = bp.blockType;
+            ctx->phase = 2;
+        }
+
+        return 0;
+    }
+
+    /* Decompress : block content */
+    {
+        size_t rSize;
+        switch(ctx->bType)
+        {
+        case bt_compressed:
+            rSize = ZSTD_decompressBlock(ctx, dst, maxDstSize, src, srcSize);
+            break;
+        case bt_raw :
+            rSize = ZSTD_copyUncompressedBlock(dst, maxDstSize, src, srcSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet handled */
+            break;
+        case bt_end :   /* should never happen (filtered at phase 1) */
+            rSize = 0;
+            break;
+        default:
+            return ERROR(GENERIC);
+        }
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        if (ZSTD_isError(rSize)) return rSize;
+        ctx->previousDstEnd = (void*)( ((char*)dst) + rSize);
+        return rSize;
+    }
+
+}
+
+
+/* wrapper layer */
+
+unsigned ZSTDv02_isError(size_t code)
+{
+    return ZSTD_isError(code);
+}
+
+size_t ZSTDv02_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize)
+{
+    return ZSTD_decompress(dst, maxOriginalSize, src, compressedSize);
+}
+
+ZSTDv02_Dctx* ZSTDv02_createDCtx(void)
+{
+    return (ZSTDv02_Dctx*)ZSTD_createDCtx();
+}
+
+size_t ZSTDv02_freeDCtx(ZSTDv02_Dctx* dctx)
+{
+    return ZSTD_freeDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv02_resetDCtx(ZSTDv02_Dctx* dctx)
+{
+    return ZSTD_resetDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv02_nextSrcSizeToDecompress(ZSTDv02_Dctx* dctx)
+{
+    return ZSTD_nextSrcSizeToDecompress((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv02_decompressContinue(ZSTDv02_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompressContinue((ZSTD_DCtx*)dctx, dst, maxDstSize, src, srcSize);
+}
diff --git a/lib/legacy/zstd_v03.c b/lib/legacy/zstd_v03.c
new file mode 100644
index 0000000..7d82db6
--- /dev/null
+++ b/lib/legacy/zstd_v03.c
@@ -0,0 +1,3105 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include "zstd_v03.h"
+#include "../common/compiler.h"
+#include "../common/error_private.h"
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+
+/****************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/****************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian())
+    {
+        MEM_write16(memPtr, val);
+    }
+    else
+    {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+
+/* ******************************************************************
+   bitstream
+   Part of NewGen Entropy library
+   header file (to include)
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/**********************************************
+*  bitStream decompression API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BIT_DStream_t;
+
+typedef enum { BIT_DStream_unfinished = 0,
+               BIT_DStream_endOfBuffer = 1,
+               BIT_DStream_completed = 2,
+               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
+
+
+
+/******************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/****************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/**********************************************************
+* bitStream decoding
+**********************************************************/
+
+/*!BIT_initDStream
+*  Initialize a BIT_DStream_t.
+*  @bitD : a pointer to an already allocated BIT_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t))   /* normal case */
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+    }
+    else
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);
+                    /* fallthrough */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);
+                    /* fallthrough */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);
+                    /* fallthrough */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24;
+                    /* fallthrough */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16;
+                    /* fallthrough */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8;
+                    /* fallthrough */
+            default:;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BIT_lookBitsFast :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBits(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BIT_readBitsFast :
+*  unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBitsFast(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BIT_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer))
+    {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BIT_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start)
+    {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
+        return BIT_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BIT_DStream_status result = BIT_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start)
+        {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BIT_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BIT_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   Error codes and messages
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef ERROR_H_MODULE
+#define ERROR_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define ERR_STATIC static inline
+#elif defined(_MSC_VER)
+#  define ERR_STATIC static __inline
+#elif defined(__GNUC__)
+#  define ERR_STATIC static __attribute__((unused))
+#else
+#  define ERR_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/******************************************
+*  Error Management
+******************************************/
+#define PREFIX(name) ZSTD_error_##name
+
+#define ERROR(name) (size_t)-PREFIX(name)
+
+#define ERROR_LIST(ITEM) \
+        ITEM(PREFIX(No_Error)) ITEM(PREFIX(GENERIC)) \
+        ITEM(PREFIX(dstSize_tooSmall)) ITEM(PREFIX(srcSize_wrong)) \
+        ITEM(PREFIX(prefix_unknown)) ITEM(PREFIX(corruption_detected)) \
+        ITEM(PREFIX(tableLog_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooSmall)) \
+        ITEM(PREFIX(maxCode))
+
+#define ERROR_GENERATE_ENUM(ENUM) ENUM,
+typedef enum { ERROR_LIST(ERROR_GENERATE_ENUM) } ERR_codes;  /* enum is exposed, to detect & handle specific errors; compare function result to -enum value */
+
+#define ERROR_CONVERTTOSTRING(STRING) #STRING,
+#define ERROR_GENERATE_STRING(EXPR) ERROR_CONVERTTOSTRING(EXPR)
+static const char* ERR_strings[] = { ERROR_LIST(ERROR_GENERATE_STRING) };
+
+ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
+
+ERR_STATIC const char* ERR_getErrorName(size_t code)
+{
+    static const char* codeError = "Unspecified error code";
+    if (ERR_isError(code)) return ERR_strings[-(int)(code)];
+    return codeError;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ERROR_H_MODULE */
+/*
+Constructor and Destructor of type FSE_CTable
+    Note that its size depends on 'tableLog' and 'maxSymbolValue' */
+typedef unsigned FSE_CTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Static allocation
+******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) (size + (size>>7))
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* You can statically allocate FSE CTable/DTable as a table of unsigned using below macro */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/******************************************
+*  FSE advanced API
+******************************************/
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
+/* build a fake FSE_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
+/* build a fake FSE_DTable, designed to always generate the same symbolValue */
+
+
+/******************************************
+*  FSE symbol decompression API
+******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+
+/******************************************
+*  FSE unsafe API
+******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/******************************************
+*  Implementation of inline functions
+******************************************/
+
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    DStatePtr->state = BIT_readBits(bitD, DTableH.tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Static allocation macros
+******************************************/
+/* Huff0 buffer bounds */
+#define HUF_CTABLEBOUND 129
+#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+#define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* static allocation of Huff0's DTable */
+#define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))  /* nb Cells; use unsigned short for X2, unsigned int for X4 */
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/******************************************
+*  Advanced functions
+******************************************/
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+/*
+    zstd - standard compression library
+    Header File
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Version
+***************************************/
+#define ZSTD_VERSION_MAJOR    0    /* for breaking interface changes  */
+#define ZSTD_VERSION_MINOR    2    /* for new (non-breaking) interface capabilities */
+#define ZSTD_VERSION_RELEASE  2    /* for tweaks, bug-fixes, or development */
+#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTD_CCtx_s ZSTD_CCtx;   /* incomplete type */
+
+#if defined (__cplusplus)
+}
+#endif
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Streaming functions
+***************************************/
+
+typedef struct ZSTDv03_Dctx_s ZSTD_DCtx;
+
+/*
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTD_magicNumber 0xFD2FB523   /* v0.3 */
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/****************************************************************
+*  Tuning parameters
+****************************************************************/
+/* MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSE_MAX_MEMORY_USAGE 14
+#define FSE_DEFAULT_MEMORY_USAGE 13
+
+/* FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSE_MAX_SYMBOL_VALUE 255
+
+
+/****************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+
+
+/****************************************************************
+*  Byte symbol type
+****************************************************************/
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/****************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+
+/****************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+/* Function templates */
+
+#define FSE_DECODE_TYPE FSE_decode_t
+
+static U32 FSE_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+static size_t FSE_buildDTable
+(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* ptr = dt+1;
+    FSE_DTableHeader DTableH;
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*)ptr;
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSE_tableStep(tableSize);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        if (normalizedCounter[s]==-1)
+        {
+            tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        }
+        else
+        {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+        }
+    }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++)
+        {
+            tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }
+    }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++)
+        {
+            FSE_FUNCTION_TYPE symbol = (FSE_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BIT_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+        }
+    }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));
+    return 0;
+}
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+/******************************************
+*  FSE helper functions
+******************************************/
+static unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+
+
+/****************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSE_abs(short a)
+{
+    return a<0 ? -a : a;
+}
+
+static size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr))
+    {
+        if (previous0)
+        {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF)
+            {
+                n0+=24;
+                if (ip < iend-5)
+                {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                }
+                else
+                {
+                    bitStream >>= 16;
+                    bitCount+=16;
+                }
+            }
+            while ((bitStream & 3) == 3)
+            {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+            {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max)
+            {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            }
+            else
+            {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSE_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold)
+            {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            {
+                if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+                {
+                    ip += bitCount>>3;
+                    bitCount &= 7;
+                }
+                else
+                {
+                    bitCount -= (int)(8 * (iend - 4 - ip));
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+            }
+        }
+    }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*********************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const cell = (FSE_decode_t*)(ptr) + 1;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)(ptr) + 1;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BIT_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSE_isError(errorCode)) return errorCode;
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) && (op<olimit) ; op+=4)
+    {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
+    while (1)
+    {
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state1);
+
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BIT_endOfDStream(&bitD) && FSE_endOfDState(&state1) && FSE_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+static size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+
+    /* select fast mode (static) */
+    if (DTableH.fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    errorCode = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSE_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/******************************************
+*  Helper functions
+******************************************/
+static unsigned HUF_isError(size_t code) { return ERR_isError(code); }
+
+#define HUF_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUF_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_DEFAULT_TABLELOG  HUF_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUF_MAX_SYMBOL_VALUE 255
+#if (HUF_MAX_TABLELOG > HUF_ABSOLUTEMAX_TABLELOG)
+#  error "HUF_MAX_TABLELOG is too large !"
+#endif
+
+
+
+/*********************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUF_readStats
+    Read compact Huffman tree, saved by HUF_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  /* special header */
+    {
+        if (iSize >= (242))   /* RLE */
+        {
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else   /* Incompressible */
+        {
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2)
+            {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+            }
+        }
+    }
+    else  /* header compressed with FSE (normal case) */
+    {
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSE_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUF_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++)
+    {
+        if (huffWeight[n] >= HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BIT_highbit32(weightTotal) + 1;
+    if (tableLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BIT_highbit32(rest);
+        U32 lastWeight = BIT_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/**************************/
+/* single-symbol decoding */
+/**************************/
+
+static size_t HUF_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* ptr = DTable+1;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)(ptr);
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(huffWeight, HUF_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof DTable, as allocated, from used size of DTable, in case of DTable re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++)
+    {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++)
+    {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUF_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BIT_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+    {
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd))
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+
+static size_t HUF_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX2* const dt = ((const HUF_DEltX2*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUF_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUF_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/***************************/
+/* double-symbols decoding */
+/***************************/
+
+static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUF_DEltX4 DElt;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)   /* note : sortedSymbols already skipped */
+    {
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUF_ABSOLUTEMAX_TABLELOG][HUF_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)
+    {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits)   /* enough room for a second symbol */
+        {
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            HUF_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+static size_t HUF_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    void* ptr = DTable;
+    HUF_DEltX4* const dt = ((HUF_DEltX4*)ptr) + 1;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        { if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+    HUF_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else
+    {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+        {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+        }
+    }
+    return 1;
+}
+
+
+#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd-7))
+    {
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+
+static size_t HUF_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX4* const dt = ((const HUF_DEltX4*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* Generic decompression selector */
+/**********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUF_decompress4X2, HUF_decompress4X4, NULL };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    //return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUF_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+    //return HUF_decompress4X6(dst, dstSize, cSrc, cSrcSize);   /* multi-streams quad-symbols decoding */
+}
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+*  MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*/
+#define ZSTD_MEMORY_USAGE 17
+
+/*!
+ * HEAPMODE :
+ * Select how default compression functions will allocate memory for their hash table,
+ * in memory stack (0, fastest), or in memory heap (1, requires malloc())
+ * Note that compression context is fairly large, as a consequence heap memory is recommended.
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif /* ZSTD_HEAPMODE */
+
+/*!
+*  LEGACY_SUPPORT :
+*  decompressor can decode older formats (starting from Zstd 0.1+)
+*/
+#ifndef ZSTD_LEGACY_SUPPORT
+#  define ZSTD_LEGACY_SUPPORT 1
+#endif
+
+
+/* *******************************************************
+*  Includes
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug : printf */
+
+
+/* *******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef __AVX2__
+#  include <immintrin.h>   /* AVX2 intrinsics */
+#endif
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#else
+#  define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+#endif
+
+
+/* *******************************************************
+*  Constants
+*********************************************************/
+#define HASH_LOG (ZSTD_MEMORY_USAGE - 2)
+#define HASH_TABLESIZE (1 << HASH_LOG)
+#define HASH_MASK (HASH_TABLESIZE - 1)
+
+#define KNUTH 2654435761
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+#define MIN_SEQUENCES_SIZE (2 /*seqNb*/ + 2 /*dumps*/ + 3 /*seqTables*/ + 1 /*bitStream*/)
+#define MIN_CBLOCK_SIZE (3 /*litCSize*/ + MIN_SEQUENCES_SIZE)
+#define IS_RAW BIT0
+#define IS_RLE BIT1
+
+#define WORKPLACESIZE (BLOCKSIZE*3)
+#define MINMATCH 4
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxML  ((1<<MLbits )-1)
+#define MaxLL  ((1<<LLbits )-1)
+#define MaxOff   31
+#define LitFSELog  11
+#define MLFSELog   10
+#define LLFSELog   10
+#define OffFSELog   9
+#define MAX(a,b) ((a)<(b)?(b):(a))
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define LITERAL_NOENTROPY 63
+#define COMMAND_NOENTROPY 7   /* to remove */
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+static const size_t ZSTD_blockHeaderSize = 3;
+static const size_t ZSTD_frameHeaderSize = 4;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void   ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+static void   ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTD_wildcopy : custom version of memcpy(), can copy up to 7-8 bytes too many */
+static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do COPY8(op, ip) while (op < oend);
+}
+
+
+/* **************************************
+*  Local structures
+****************************************/
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* offCode;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* litLengthStart;
+    BYTE* litLength;
+    BYTE* matchLengthStart;
+    BYTE* matchLength;
+    BYTE* dumpsStart;
+    BYTE* dumps;
+} SeqStore_t;
+
+
+/* *************************************
+*  Error Management
+***************************************/
+/*! ZSTD_isError
+*   tells if a return value is an error code */
+static unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
+
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+struct ZSTDv03_Dctx_s
+{
+    U32 LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
+    U32 OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    U32 MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
+    void* previousDstEnd;
+    void* base;
+    size_t expected;
+    blockType_t bType;
+    U32 phase;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + 8 /* margin for wildcopy */];
+};   /* typedef'd to ZSTD_Dctx within "zstd_static.h" */
+
+
+static size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3) return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+static size_t ZSTD_copyUncompressedBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    if (srcSize > 0) {
+        memcpy(dst, src, srcSize);
+    }
+    return srcSize;
+}
+
+
+/** ZSTD_decompressLiterals
+    @return : nb of bytes read from src, or an error code*/
+static size_t ZSTD_decompressLiterals(void* dst, size_t* maxDstSizePtr,
+                                const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    const size_t litSize = (MEM_readLE32(src) & 0x1FFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+    const size_t litCSize = (MEM_readLE32(ip+2) & 0xFFFFFF) >> 5;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+
+    if (litSize > *maxDstSizePtr) return ERROR(corruption_detected);
+    if (litCSize + 5 > srcSize) return ERROR(corruption_detected);
+
+    if (HUF_isError(HUF_decompress(dst, litSize, ip+5, litCSize))) return ERROR(corruption_detected);
+
+    *maxDstSizePtr = litSize;
+    return litCSize + 5;
+}
+
+
+/** ZSTD_decodeLiteralsBlock
+    @return : nb of bytes read from src (< srcSize )*/
+static size_t ZSTD_decodeLiteralsBlock(void* ctx,
+                          const void* src, size_t srcSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* const istart = (const BYTE* const)src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(*istart & 3)
+    {
+    default:
+    case 0:
+        {
+            size_t litSize = BLOCKSIZE;
+            const size_t readSize = ZSTD_decompressLiterals(dctx->litBuffer, &litSize, src, srcSize);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, 8);
+            return readSize;   /* works if it's an error too */
+        }
+    case IS_RAW:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > srcSize-11)   /* risk of reading too far with wildcopy */
+            {
+                if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+                if (litSize > srcSize-3) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, 8);
+                return litSize+3;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+3;
+            dctx->litSize = litSize;
+            return litSize+3;
+        }
+    case IS_RLE:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[3], litSize + 8);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return 4;
+        }
+    }
+}
+
+
+static size_t ZSTD_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
+                         const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < 5) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = MEM_readLE16(ip); ip+=2;
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2)
+    {
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    }
+    else
+    {
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL and MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case bt_rle :
+            LLlog = 0;
+            FSE_buildDTable_rle(DTableLL, *ip++); break;
+        case bt_raw :
+            LLlog = LLbits;
+            FSE_buildDTable_raw(DTableLL, LLbits); break;
+        default :
+            {   U32 max = MaxLL;
+                headerSize = FSE_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case bt_rle :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case bt_raw :
+            Offlog = Offbits;
+            FSE_buildDTable_raw(DTableOffb, Offbits); break;
+        default :
+            {   U32 max = MaxOff;
+                headerSize = FSE_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case bt_rle :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableML, *ip++); break;
+        case bt_raw :
+            MLlog = MLbits;
+            FSE_buildDTable_raw(DTableML, MLbits); break;
+        default :
+            {   U32 max = MaxML;
+                headerSize = FSE_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t offset;
+    size_t matchLength;
+} seq_t;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    FSE_DState_t stateLL;
+    FSE_DState_t stateOffb;
+    FSE_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    seqState->prevOffset = seq->offset;
+    if (litLength == MaxLL)
+    {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) litLength += add;
+        else if (dumps + 3 <= de)
+        {
+            litLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {
+        static const size_t offsetPrefix[MaxOff+1] = {  /* note : size_t faster than U32 */
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode, nbBits;
+        offsetCode = FSE_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));   /* <= maxOff, by table construction */
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BIT_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* cmove */
+    }
+
+    /* MatchLength */
+    matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML)
+    {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) matchLength += add;
+        else if (dumps + 3 <= de)
+        {
+            matchLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+}
+
+
+static size_t ZSTD_execSequence(BYTE* op,
+                                seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                BYTE* const base, BYTE* const oend)
+{
+    static const int dec32table[] = {0, 1, 2, 1, 4, 4, 4, 4};   /* added */
+    static const int dec64table[] = {8, 8, 8, 7, 8, 9,10,11};   /* subtracted */
+    const BYTE* const ostart = op;
+    BYTE* const oLitEnd = op + sequence.litLength;
+    BYTE* const oMatchEnd = op + sequence.litLength + sequence.matchLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+    if (sequence.offset > (U32)(oLitEnd - base)) return ERROR(corruption_detected);
+
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTD_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    {   const BYTE* match = op - sequence.offset;
+
+        /* check */
+        if (sequence.offset > (size_t)op) return ERROR(corruption_detected);   /* address space overflow test (this test seems kept by clang optimizer) */
+        //if (match > op) return ERROR(corruption_detected);   /* address space overflow test (is clang optimizer removing this test ?) */
+        if (match < base) return ERROR(corruption_detected);
+
+        /* close range match, overlap */
+        if (sequence.offset < 8)
+        {
+            const int dec64 = dec64table[sequence.offset];
+            op[0] = match[0];
+            op[1] = match[1];
+            op[2] = match[2];
+            op[3] = match[3];
+            match += dec32table[sequence.offset];
+            ZSTD_copy4(op+4, match);
+            match -= dec64;
+        }
+        else
+        {
+            ZSTD_copy8(op, match);
+        }
+        op += 8; match += 8;
+
+        if (oMatchEnd > oend-(16-MINMATCH))
+        {
+            if (op < oend_8)
+            {
+                ZSTD_wildcopy(op, match, oend_8 - op);
+                match += oend_8 - op;
+                op = oend_8;
+            }
+            while (op < oMatchEnd) *op++ = *match++;
+        }
+        else
+        {
+            ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+        }
+    }
+
+    return oMatchEnd - ostart;
+}
+
+static size_t ZSTD_decompressSequences(
+                               void* ctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    BYTE* const base = (BYTE*) (dctx->base);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTD_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, iend-ip);
+    if (ZSTD_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = sequence.offset = 4;
+        errorCode = BIT_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (nbSeq>0) ; )
+        {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTD_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTD_execSequence(op, sequence, &litPtr, litEnd, base, oend);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if ( !BIT_endOfDStream(&(seqState.DStream)) ) return ERROR(corruption_detected);   /* requested too much : data is corrupted */
+        if (nbSeq<0) return ERROR(corruption_detected);   /* requested too many sequences : data is corrupted */
+
+        /* last literal segment */
+        {
+            size_t lastLLSize = litEnd - litPtr;
+            if (litPtr > litEnd) return ERROR(corruption_detected);
+            if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+            if (lastLLSize > 0) {
+                if (op != litPtr) memmove(op, litPtr, lastLLSize);
+                op += lastLLSize;
+            }
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static size_t ZSTD_decompressBlock(
+                            void* ctx,
+                            void* dst, size_t maxDstSize,
+                      const void* src, size_t srcSize)
+{
+    /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    /* Decode literals sub-block */
+    size_t litCSize = ZSTD_decodeLiteralsBlock(ctx, src, srcSize);
+    if (ZSTD_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTD_decompressSequences(ctx, dst, maxDstSize, ip, srcSize);
+}
+
+
+static size_t ZSTD_decompressDCtx(void* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTD_decompressBlock(ctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTD_copyUncompressedBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTD_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+static size_t ZSTD_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    ZSTD_DCtx ctx;
+    ctx.base = dst;
+    return ZSTD_decompressDCtx(&ctx, dst, maxDstSize, src, srcSize);
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+MEM_STATIC void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv03_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+        return;
+    }
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+        return;
+    }
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * BLOCKSIZE;
+}
+
+
+/*******************************
+*  Streaming Decompression API
+*******************************/
+
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx)
+{
+    dctx->expected = ZSTD_frameHeaderSize;
+    dctx->phase = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    return 0;
+}
+
+static ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)malloc(sizeof(ZSTD_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTD_resetDCtx(dctx);
+    return dctx;
+}
+
+static size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    free(dctx);
+    return 0;
+}
+
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != ctx->expected) return ERROR(srcSize_wrong);
+    if (dst != ctx->previousDstEnd)  /* not contiguous */
+        ctx->base = dst;
+
+    /* Decompress : frame header */
+    if (ctx->phase == 0)
+    {
+        /* Check frame magic header */
+        U32 magicNumber = MEM_readLE32(src);
+        if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        return 0;
+    }
+
+    /* Decompress : block header */
+    if (ctx->phase == 1)
+    {
+        blockProperties_t bp;
+        size_t blockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+        if (ZSTD_isError(blockSize)) return blockSize;
+        if (bp.blockType == bt_end)
+        {
+            ctx->expected = 0;
+            ctx->phase = 0;
+        }
+        else
+        {
+            ctx->expected = blockSize;
+            ctx->bType = bp.blockType;
+            ctx->phase = 2;
+        }
+
+        return 0;
+    }
+
+    /* Decompress : block content */
+    {
+        size_t rSize;
+        switch(ctx->bType)
+        {
+        case bt_compressed:
+            rSize = ZSTD_decompressBlock(ctx, dst, maxDstSize, src, srcSize);
+            break;
+        case bt_raw :
+            rSize = ZSTD_copyUncompressedBlock(dst, maxDstSize, src, srcSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet handled */
+            break;
+        case bt_end :   /* should never happen (filtered at phase 1) */
+            rSize = 0;
+            break;
+        default:
+            return ERROR(GENERIC);
+        }
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        if (ZSTD_isError(rSize)) return rSize;
+        ctx->previousDstEnd = (void*)( ((char*)dst) + rSize);
+        return rSize;
+    }
+
+}
+
+
+/* wrapper layer */
+
+unsigned ZSTDv03_isError(size_t code)
+{
+    return ZSTD_isError(code);
+}
+
+size_t ZSTDv03_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize)
+{
+    return ZSTD_decompress(dst, maxOriginalSize, src, compressedSize);
+}
+
+ZSTDv03_Dctx* ZSTDv03_createDCtx(void)
+{
+    return (ZSTDv03_Dctx*)ZSTD_createDCtx();
+}
+
+size_t ZSTDv03_freeDCtx(ZSTDv03_Dctx* dctx)
+{
+    return ZSTD_freeDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv03_resetDCtx(ZSTDv03_Dctx* dctx)
+{
+    return ZSTD_resetDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv03_nextSrcSizeToDecompress(ZSTDv03_Dctx* dctx)
+{
+    return ZSTD_nextSrcSizeToDecompress((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv03_decompressContinue(ZSTDv03_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompressContinue((ZSTD_DCtx*)dctx, dst, maxDstSize, src, srcSize);
+}
diff --git a/lib/legacy/zstd_v04.c b/lib/legacy/zstd_v04.c
new file mode 100644
index 0000000..0da316c
--- /dev/null
+++ b/lib/legacy/zstd_v04.c
@@ -0,0 +1,3598 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+ /******************************************
+ *  Includes
+ ******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+#include "zstd_v04.h"
+#include "../common/compiler.h"
+#include "../common/error_private.h"
+
+
+/* ******************************************************************
+ *   mem.h
+ *******************************************************************/
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+
+
+/****************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-*************************************
+*  Debug
+***************************************/
+#include "../common/debug.h"
+#ifndef assert
+#  define assert(condition) ((void)0)
+#endif
+
+
+/****************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian())
+    {
+        MEM_write16(memPtr, val);
+    }
+    else
+    {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+/*
+    zstd - standard compression library
+    Header File for static linking only
+*/
+#ifndef ZSTD_STATIC_H
+#define ZSTD_STATIC_H
+
+
+/* *************************************
+*  Types
+***************************************/
+#define ZSTD_WINDOWLOG_ABSOLUTEMIN 11
+
+/** from faster to stronger */
+typedef enum { ZSTD_fast, ZSTD_greedy, ZSTD_lazy, ZSTD_lazy2, ZSTD_btlazy2 } ZSTD_strategy;
+
+typedef struct
+{
+    U64 srcSize;       /* optional : tells how much bytes are present in the frame. Use 0 if not known. */
+    U32 windowLog;     /* largest match distance : larger == more compression, more memory needed during decompression */
+    U32 contentLog;    /* full search segment : larger == more compression, slower, more memory (useless for fast) */
+    U32 hashLog;       /* dispatch table : larger == more memory, faster */
+    U32 searchLog;     /* nb of searches : larger == more compression, slower */
+    U32 searchLength;  /* size of matches : larger == faster decompression, sometimes less compression */
+    ZSTD_strategy strategy;
+} ZSTD_parameters;
+
+typedef ZSTDv04_Dctx ZSTD_DCtx;
+
+/* *************************************
+*  Advanced functions
+***************************************/
+/** ZSTD_decompress_usingDict
+*   Same as ZSTD_decompressDCtx, using a Dictionary content as prefix
+*   Note : dict can be NULL, in which case, it's equivalent to ZSTD_decompressDCtx() */
+static size_t ZSTD_decompress_usingDict(ZSTD_DCtx* ctx,
+                                             void* dst, size_t maxDstSize,
+                                       const void* src, size_t srcSize,
+                                       const void* dict,size_t dictSize);
+
+
+/* **************************************
+*  Streaming functions (direct mode)
+****************************************/
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx);
+static size_t ZSTD_getFrameParams(ZSTD_parameters* params, const void* src, size_t srcSize);
+static void   ZSTD_decompress_insertDictionary(ZSTD_DCtx* ctx, const void* src, size_t srcSize);
+
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
+
+/**
+  Streaming decompression, bufferless mode
+
+  A ZSTD_DCtx object is required to track streaming operations.
+  Use ZSTD_createDCtx() / ZSTD_freeDCtx() to manage it.
+  A ZSTD_DCtx object can be re-used multiple times. Use ZSTD_resetDCtx() to return to fresh status.
+
+  First operation is to retrieve frame parameters, using ZSTD_getFrameParams().
+  This function doesn't consume its input. It needs enough input data to properly decode the frame header.
+  Objective is to retrieve *params.windowlog, to know minimum amount of memory required during decoding.
+  Result : 0 when successful, it means the ZSTD_parameters structure has been filled.
+           >0 : means there is not enough data into src. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTD_isError() (For example, if it's not a ZSTD header)
+
+  Then, you can optionally insert a dictionary.
+  This operation must mimic the compressor behavior, otherwise decompression will fail or be corrupted.
+
+  Then it's possible to start decompression.
+  Use ZSTD_nextSrcSizeToDecompress() and ZSTD_decompressContinue() alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() requires this exact amount of bytes, or it will fail.
+  ZSTD_decompressContinue() needs previous data blocks during decompression, up to (1 << windowlog).
+  They should preferably be located contiguously, prior to current block. Alternatively, a round buffer is also possible.
+
+  @result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+
+  A frame is fully decoded when ZSTD_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+*/
+
+
+
+
+#endif  /* ZSTD_STATIC_H */
+
+
+/*
+    zstd_internal - common functions to include
+    Header File for include
+*/
+#ifndef ZSTD_CCOMMON_H_MODULE
+#define ZSTD_CCOMMON_H_MODULE
+
+/* *************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/* *************************************
+*  Common constants
+***************************************/
+#define ZSTD_MAGICNUMBER 0xFD2FB524   /* v0.4 */
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+
+static const size_t ZSTD_blockHeaderSize = 3;
+static const size_t ZSTD_frameHeaderSize_min = 5;
+#define ZSTD_frameHeaderSize_max 5         /* define, for static allocation */
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define IS_RAW BIT0
+#define IS_RLE BIT1
+
+#define MINMATCH 4
+#define REPCODE_STARTVALUE 4
+
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxML  ((1<<MLbits) - 1)
+#define MaxLL  ((1<<LLbits) - 1)
+#define MaxOff ((1<<Offbits)- 1)
+#define MLFSELog   10
+#define LLFSELog   10
+#define OffFSELog   9
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define MIN_SEQUENCES_SIZE (2 /*seqNb*/ + 2 /*dumps*/ + 3 /*seqTables*/ + 1 /*bitStream*/)
+#define MIN_CBLOCK_SIZE (3 /*litCSize*/ + MIN_SEQUENCES_SIZE)
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+
+/* ******************************************
+*  Shared functions to include for inlining
+********************************************/
+static void ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTD_wildcopy : custom version of memcpy(), can copy up to 7-8 bytes too many */
+static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file
+****************************************************************** */
+#ifndef FSE_H
+#define FSE_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+
+
+/* *****************************************
+*  FSE simple functions
+******************************************/
+static size_t FSE_decompress(void* dst,  size_t maxDstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+FSE_decompress():
+    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'maxDstSize'.
+    return : size of regenerated data (<= maxDstSize)
+             or an error code, which can be tested using FSE_isError()
+
+    ** Important ** : FSE_decompress() doesn't decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+
+
+/* *****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+static unsigned    FSE_isError(size_t code);        /* tells if a return value is an error code */
+
+
+
+/* *****************************************
+*  FSE detailed API
+******************************************/
+/*!
+FSE_compress() does the following:
+1. count symbol occurrence from source[] into table count[]
+2. normalize counters so that sum(count[]) == Power_of_2 (2^tableLog)
+3. save normalized counters to memory buffer using writeNCount()
+4. build encoding table 'CTable' from normalized counters
+5. encode the data stream using encoding table 'CTable'
+
+FSE_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*!
+FSE_readNCount():
+   Read compactly saved 'normalizedCounter' from 'rBuffer'.
+   return : size read from 'rBuffer'
+            or an errorCode, which can be tested using FSE_isError()
+            maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+static  size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*!
+Constructor and Destructor of type FSE_DTable
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+/*!
+FSE_buildDTable():
+   Builds 'dt', which must be already allocated, using FSE_createDTable()
+   return : 0,
+            or an errorCode, which can be tested using FSE_isError() */
+static size_t FSE_buildDTable ( FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*!
+FSE_decompress_usingDTable():
+   Decompress compressed source 'cSrc' of size 'cSrcSize' using 'dt'
+   into 'dst' which must be already allocated.
+   return : size of regenerated data (necessarily <= maxDstSize)
+            or an errorCode, which can be tested using FSE_isError() */
+static  size_t FSE_decompress_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const FSE_DTable* dt);
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSE_readNCount() if it was saved using FSE_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSE_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSE_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+The next step is to build the decompression tables 'FSE_DTable' from 'normalizedCounter'.
+This is performed by the function FSE_buildDTable().
+The space required by 'FSE_DTable' must be already allocated using FSE_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+'FSE_DTable' can then be used to decompress 'cSrc', with FSE_decompress_usingDTable().
+'cSrcSize' must be strictly correct, otherwise decompression will fail.
+FSE_decompress_usingDTable() result will tell how many bytes were regenerated (<=maxDstSize).
+If there is an error, the function will return an error code, which can be tested using FSE_isError(). (ex: dst buffer too small)
+*/
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSE_H */
+
+
+/* ******************************************************************
+   bitstream
+   Part of NewGen Entropy library
+   header file (to include)
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+/**********************************************
+*  bitStream decompression API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BIT_DStream_t;
+
+typedef enum { BIT_DStream_unfinished = 0,
+               BIT_DStream_endOfBuffer = 1,
+               BIT_DStream_completed = 2,
+               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
+
+
+
+
+/******************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/****************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+/**********************************************************
+* bitStream decoding
+**********************************************************/
+
+/*!BIT_initDStream
+*  Initialize a BIT_DStream_t.
+*  @bitD : a pointer to an already allocated BIT_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t))   /* normal case */
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+    }
+    else
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BIT_lookBitsFast :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBits(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BIT_readBitsFast :
+*  unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBitsFast(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BIT_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer))
+    {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BIT_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start)
+    {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
+        return BIT_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BIT_DStream_status result = BIT_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start)
+        {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BIT_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BIT_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSE_STATIC_H
+#define FSE_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) (size + (size>>7))
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of unsigned using below macros */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSE advanced API
+*******************************************/
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
+/* build a fake FSE_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
+/* build a fake FSE_DTable, designed to always generate the same symbolValue */
+
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    DStatePtr->state = BIT_readBits(bitD, DTableH.tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSE_STATIC_H */
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSE_MAX_MEMORY_USAGE 14
+#define FSE_DEFAULT_MEMORY_USAGE 13
+
+/*!FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSE_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+#define FSE_DECODE_TYPE FSE_decode_t
+
+
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Dependencies
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+
+/*-**************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+static U32 FSE_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+
+static size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    FSE_DTableHeader DTableH;
+    void* const tdPtr = dt+1;   /* because dt is unsigned, 32-bits aligned on 32-bits */
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*) (tdPtr);
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSE_tableStep(tableSize);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    memset(tableDecode, 0, sizeof(FSE_DECODE_TYPE) * (maxSymbolValue+1) );   /* useless init, but keep static analyzer happy, and we don't need to performance optimize legacy decoders */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        if (normalizedCounter[s]==-1)
+        {
+            tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        }
+        else
+        {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+        }
+    }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++)
+        {
+            tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }
+    }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++)
+        {
+            FSE_FUNCTION_TYPE symbol = (FSE_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BIT_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+        }
+    }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));
+    return 0;
+}
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+/******************************************
+*  FSE helper functions
+******************************************/
+static unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+
+
+/****************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSE_abs(short a)
+{
+    return a<0 ? -a : a;
+}
+
+static size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr))
+    {
+        if (previous0)
+        {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF)
+            {
+                n0+=24;
+                if (ip < iend-5)
+                {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                }
+                else
+                {
+                    bitStream >>= 16;
+                    bitCount+=16;
+                }
+            }
+            while ((bitStream & 3) == 3)
+            {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+            {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max)
+            {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            }
+            else
+            {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSE_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold)
+            {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            {
+                if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+                {
+                    ip += bitCount>>3;
+                    bitCount &= 7;
+                }
+                else
+                {
+                    bitCount -= (int)(8 * (iend - 4 - ip));
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+            }
+        }
+    }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*********************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSE_decode_t* const cell = (FSE_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BIT_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSE_isError(errorCode)) return errorCode;
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) && (op<olimit) ; op+=4)
+    {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
+    while (1)
+    {
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state1);
+
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BIT_endOfDStream(&bitD) && FSE_endOfDState(&state1) && FSE_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+static size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    U32 fastMode;
+
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    fastMode = DTableH.fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    errorCode = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSE_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
+
+
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef HUFF0_H
+#define HUFF0_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* ****************************************
+*  Dependency
+******************************************/
+#include <stddef.h>    /* size_t */
+
+
+/* ****************************************
+*  Huff0 simple functions
+******************************************/
+static size_t HUF_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+HUF_decompress():
+    Decompress Huff0 data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstSize'.
+    'dstSize' must be the exact size of original (uncompressed) data.
+    Note : in contrast with FSE, HUF_decompress can regenerate RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data, because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize)
+              or an error code, which can be tested using HUF_isError()
+*/
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+static unsigned    HUF_isError(size_t code);        /* tells if a return value is an error code */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUFF0_H */
+
+
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef HUFF0_STATIC_H
+#define HUFF0_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* ****************************************
+*  Static allocation macros
+******************************************/
+/* static allocation of Huff0's DTable */
+#define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))  /* nb Cells; use unsigned short for X2, unsigned int for X4 */
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+/* ****************************************
+*  Huff0 detailed API
+******************************************/
+/*!
+HUF_decompress() does the following:
+1. select the decompression algorithm (X2, X4, X6) based on pre-computed heuristics
+2. build Huffman table from save, using HUF_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUF_decompressSXn_usingDTable
+
+*/
+static size_t HUF_readDTableX2 (unsigned short* DTable, const void* src, size_t srcSize);
+static size_t HUF_readDTableX4 (unsigned* DTable, const void* src, size_t srcSize);
+
+static size_t HUF_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+static size_t HUF_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* HUFF0_STATIC_H */
+
+
+
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+#define HUF_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUF_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_DEFAULT_TABLELOG  HUF_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUF_MAX_SYMBOL_VALUE 255
+#if (HUF_MAX_TABLELOG > HUF_ABSOLUTEMAX_TABLELOG)
+#  error "HUF_MAX_TABLELOG is too large !"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+static unsigned HUF_isError(size_t code) { return ERR_isError(code); }
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+
+/*-*******************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUF_readStats
+    Read compact Huffman tree, saved by HUF_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  /* special header */
+    {
+        if (iSize >= (242))   /* RLE */
+        {
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else   /* Incompressible */
+        {
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2)
+            {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+            }
+        }
+    }
+    else  /* header compressed with FSE (normal case) */
+    {
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSE_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUF_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++)
+    {
+        if (huffWeight[n] >= HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BIT_highbit32(weightTotal) + 1;
+    if (tableLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BIT_highbit32(rest);
+        U32 lastWeight = BIT_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/**************************/
+/* single-symbol decoding */
+/**************************/
+
+static size_t HUF_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    size_t iSize;
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* const dtPtr = DTable + 1;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(huffWeight, HUF_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof DTable, as allocated, from used size of DTable, in case of DTable re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++)
+    {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++)
+    {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUF_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BIT_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+    {
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd))
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+
+static size_t HUF_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUF_DEltX2* const dt = ((const HUF_DEltX2*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUF_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUF_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/***************************/
+/* double-symbols decoding */
+/***************************/
+
+static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUF_DEltX4 DElt;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)   /* note : sortedSymbols already skipped */
+    {
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUF_ABSOLUTEMAX_TABLELOG][HUF_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)
+    {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits)   /* enough room for a second symbol */
+        {
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            HUF_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+static size_t HUF_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    size_t iSize;
+    void* dtPtr = DTable;
+    HUF_DEltX4* const dt = ((HUF_DEltX4*)dtPtr) + 1;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        { if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+    HUF_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else
+    {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+        {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+        }
+    }
+    return 1;
+}
+
+
+#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd-7))
+    {
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+static size_t HUF_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUF_DEltX4* const dt = ((const HUF_DEltX4*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* Generic decompression selector */
+/**********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUF_decompress4X2, HUF_decompress4X4, NULL };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    //return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUF_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+    //return HUF_decompress4X6(dst, dstSize, cSrc, cSrcSize);   /* multi-streams quad-symbols decoding */
+}
+
+
+
+#endif   /* ZSTD_CCOMMON_H_MODULE */
+
+
+/*
+    zstd - decompression module fo v0.4 legacy format
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTD_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif
+
+
+/* *******************************************************
+*  Includes
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug : printf */
+
+
+/* *******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/* *************************************
+*  Local types
+***************************************/
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/* *************************************
+*  Error Management
+***************************************/
+
+/*! ZSTD_isError
+*   tells if a return value is an error code */
+static unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
+
+
+/* *************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock } ZSTD_dStage;
+
+struct ZSTDv04_Dctx_s
+{
+    U32 LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
+    U32 OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    U32 MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    size_t headerSize;
+    ZSTD_parameters params;
+    blockType_t bType;
+    ZSTD_dStage stage;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + 8 /* margin for wildcopy */];
+    BYTE headerBuffer[ZSTD_frameHeaderSize_max];
+};  /* typedef'd to ZSTD_DCtx within "zstd_static.h" */
+
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx)
+{
+    dctx->expected = ZSTD_frameHeaderSize_min;
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    return 0;
+}
+
+static ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)malloc(sizeof(ZSTD_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTD_resetDCtx(dctx);
+    return dctx;
+}
+
+static size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    free(dctx);
+    return 0;
+}
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+/** ZSTD_decodeFrameHeader_Part1
+*   decode the 1st part of the Frame Header, which tells Frame Header size.
+*   srcSize must be == ZSTD_frameHeaderSize_min
+*   @return : the full size of the Frame Header */
+static size_t ZSTD_decodeFrameHeader_Part1(ZSTD_DCtx* zc, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize != ZSTD_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_MAGICNUMBER) return ERROR(prefix_unknown);
+    zc->headerSize = ZSTD_frameHeaderSize_min;
+    return zc->headerSize;
+}
+
+
+static size_t ZSTD_getFrameParams(ZSTD_parameters* params, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize < ZSTD_frameHeaderSize_min) return ZSTD_frameHeaderSize_max;
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_MAGICNUMBER) return ERROR(prefix_unknown);
+    memset(params, 0, sizeof(*params));
+    params->windowLog = (((const BYTE*)src)[4] & 15) + ZSTD_WINDOWLOG_ABSOLUTEMIN;
+    if ((((const BYTE*)src)[4] >> 4) != 0) return ERROR(frameParameter_unsupported);   /* reserved bits */
+    return 0;
+}
+
+/** ZSTD_decodeFrameHeader_Part2
+*   decode the full Frame Header
+*   srcSize must be the size provided by ZSTD_decodeFrameHeader_Part1
+*   @return : 0, or an error code, which can be tested using ZSTD_isError() */
+static size_t ZSTD_decodeFrameHeader_Part2(ZSTD_DCtx* zc, const void* src, size_t srcSize)
+{
+    size_t result;
+    if (srcSize != zc->headerSize) return ERROR(srcSize_wrong);
+    result = ZSTD_getFrameParams(&(zc->params), src, srcSize);
+    if ((MEM_32bits()) && (zc->params.windowLog > 25)) return ERROR(frameParameter_unsupported);
+    return result;
+}
+
+
+static size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3) return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+static size_t ZSTD_copyRawBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    if (srcSize > 0) {
+        memcpy(dst, src, srcSize);
+    }
+    return srcSize;
+}
+
+
+/** ZSTD_decompressLiterals
+    @return : nb of bytes read from src, or an error code*/
+static size_t ZSTD_decompressLiterals(void* dst, size_t* maxDstSizePtr,
+                                const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    const size_t litSize = (MEM_readLE32(src) & 0x1FFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+    const size_t litCSize = (MEM_readLE32(ip+2) & 0xFFFFFF) >> 5;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+
+    if (litSize > *maxDstSizePtr) return ERROR(corruption_detected);
+    if (litCSize + 5 > srcSize) return ERROR(corruption_detected);
+
+    if (HUF_isError(HUF_decompress(dst, litSize, ip+5, litCSize))) return ERROR(corruption_detected);
+
+    *maxDstSizePtr = litSize;
+    return litCSize + 5;
+}
+
+
+/** ZSTD_decodeLiteralsBlock
+    @return : nb of bytes read from src (< srcSize ) */
+static size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(*istart & 3)
+    {
+    /* compressed */
+    case 0:
+        {
+            size_t litSize = BLOCKSIZE;
+            const size_t readSize = ZSTD_decompressLiterals(dctx->litBuffer, &litSize, src, srcSize);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, 8);
+            return readSize;   /* works if it's an error too */
+        }
+    case IS_RAW:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > srcSize-11)   /* risk of reading too far with wildcopy */
+            {
+                if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+                if (litSize > srcSize-3) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, 8);
+                return litSize+3;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+3;
+            dctx->litSize = litSize;
+            return litSize+3;        }
+    case IS_RLE:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[3], litSize + 8);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return 4;
+        }
+    default:
+        return ERROR(corruption_detected);   /* forbidden nominal case */
+    }
+}
+
+
+static size_t ZSTD_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
+                         const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < 5) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = MEM_readLE16(ip); ip+=2;
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2)
+    {
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    }
+    else
+    {
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL >= MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case bt_rle :
+            LLlog = 0;
+            FSE_buildDTable_rle(DTableLL, *ip++); break;
+        case bt_raw :
+            LLlog = LLbits;
+            FSE_buildDTable_raw(DTableLL, LLbits); break;
+        default :
+            {   U32 max = MaxLL;
+                headerSize = FSE_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case bt_rle :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case bt_raw :
+            Offlog = Offbits;
+            FSE_buildDTable_raw(DTableOffb, Offbits); break;
+        default :
+            {   U32 max = MaxOff;
+                headerSize = FSE_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case bt_rle :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableML, *ip++); break;
+        case bt_raw :
+            MLlog = MLbits;
+            FSE_buildDTable_raw(DTableML, MLbits); break;
+        default :
+            {   U32 max = MaxML;
+                headerSize = FSE_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t offset;
+    size_t matchLength;
+} seq_t;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    FSE_DState_t stateLL;
+    FSE_DState_t stateOffb;
+    FSE_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    if (litLength == MaxLL) {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) litLength += add;
+        else if (dumps + 3 <= de) {
+            litLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {   static const U32 offsetPrefix[MaxOff+1] = {
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode, nbBits;
+        offsetCode = FSE_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));   /* <= maxOff, by table construction */
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BIT_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* cmove */
+        if (offsetCode | !litLength) seqState->prevOffset = seq->offset;   /* cmove */
+    }
+
+    /* MatchLength */
+    matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML) {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) matchLength += add;
+        else if (dumps + 3 <= de){
+            matchLength = MEM_readLE24(dumps);
+            dumps += 3;
+        }
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+}
+
+
+static size_t ZSTD_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    static const int dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+    static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+    BYTE* const oLitEnd = op + sequence.litLength;
+    const size_t sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTD_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base))
+    {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase))
+            return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd)
+        {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {
+            size_t length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_8 || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+        }
+    }
+    /* Requirement: op <= oend_8 */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        const int sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTD_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTD_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH))
+    {
+        if (op < oend_8)
+        {
+            ZSTD_wildcopy(op, match, oend_8 - op);
+            match += oend_8 - op;
+            op = oend_8;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    }
+    else
+    {
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8, but must be signed */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTD_decompressSequences(
+                               ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTD_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, iend-ip);
+    if (ZSTD_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        sequence.offset = 4;
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = 4;
+        errorCode = BIT_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && nbSeq ; )
+        {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTD_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if ( !BIT_endOfDStream(&(seqState.DStream)) ) return ERROR(corruption_detected);   /* DStream should be entirely and exactly consumed; otherwise data is corrupted */
+
+        /* last literal segment */
+        {
+            size_t lastLLSize = litEnd - litPtr;
+            if (litPtr > litEnd) return ERROR(corruption_detected);
+            if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+            if (lastLLSize > 0) {
+                if (op != litPtr) memcpy(op, litPtr, lastLLSize);
+                op += lastLLSize;
+            }
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd)   /* not contiguous */
+    {
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
+                            void* dst, size_t maxDstSize,
+                      const void* src, size_t srcSize)
+{
+    /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+    size_t litCSize;
+
+    if (srcSize > BLOCKSIZE) return ERROR(corruption_detected);
+
+    /* Decode literals sub-block */
+    litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize);
+    if (ZSTD_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTD_decompressSequences(dctx, dst, maxDstSize, ip, srcSize);
+}
+
+
+static size_t ZSTD_decompress_usingDict(ZSTD_DCtx* ctx,
+                                 void* dst, size_t maxDstSize,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties;
+
+    /* init */
+    ZSTD_resetDCtx(ctx);
+    if (dict)
+    {
+        ZSTD_decompress_insertDictionary(ctx, dict, dictSize);
+        ctx->dictEnd = ctx->previousDstEnd;
+        ctx->vBase = (const char*)dst - ((const char*)(ctx->previousDstEnd) - (const char*)(ctx->base));
+        ctx->base = dst;
+    }
+    else
+    {
+        ctx->vBase = ctx->base = ctx->dictEnd = dst;
+    }
+
+    /* Frame Header */
+    {
+        size_t frameHeaderSize;
+        if (srcSize < ZSTD_frameHeaderSize_min+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+        frameHeaderSize = ZSTD_decodeFrameHeader_Part1(ctx, src, ZSTD_frameHeaderSize_min);
+        if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+        frameHeaderSize = ZSTD_decodeFrameHeader_Part2(ctx, src, frameHeaderSize);
+        if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTD_decompressBlock_internal(ctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTD_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTD_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+static void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv04_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize_min) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+        return;
+    }
+    if (MEM_readLE32(src) != ZSTD_MAGICNUMBER) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+        return;
+    }
+    ip += ZSTD_frameHeaderSize_min; remainingSize -= ZSTD_frameHeaderSize_min;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * BLOCKSIZE;
+}
+
+/* ******************************
+*  Streaming Decompression API
+********************************/
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != ctx->expected) return ERROR(srcSize_wrong);
+    ZSTD_checkContinuity(ctx, dst);
+
+    /* Decompress : frame header; part 1 */
+    switch (ctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        /* get frame header size */
+        if (srcSize != ZSTD_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        ctx->headerSize = ZSTD_decodeFrameHeader_Part1(ctx, src, ZSTD_frameHeaderSize_min);
+        if (ZSTD_isError(ctx->headerSize)) return ctx->headerSize;
+        memcpy(ctx->headerBuffer, src, ZSTD_frameHeaderSize_min);
+        if (ctx->headerSize > ZSTD_frameHeaderSize_min) return ERROR(GENERIC);   /* impossible */
+        ctx->expected = 0;   /* not necessary to copy more */
+        /* fallthrough */
+    case ZSTDds_decodeFrameHeader:
+        /* get frame header */
+        {   size_t const result = ZSTD_decodeFrameHeader_Part2(ctx, ctx->headerBuffer, ctx->headerSize);
+            if (ZSTD_isError(result)) return result;
+            ctx->expected = ZSTD_blockHeaderSize;
+            ctx->stage = ZSTDds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDds_decodeBlockHeader:
+        /* Decode block header */
+        {   blockProperties_t bp;
+            size_t const blockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+            if (ZSTD_isError(blockSize)) return blockSize;
+            if (bp.blockType == bt_end)
+            {
+                ctx->expected = 0;
+                ctx->stage = ZSTDds_getFrameHeaderSize;
+            }
+            else
+            {
+                ctx->expected = blockSize;
+                ctx->bType = bp.blockType;
+                ctx->stage = ZSTDds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDds_decompressBlock:
+        {
+            /* Decompress : block content */
+            size_t rSize;
+            switch(ctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTD_decompressBlock_internal(ctx, dst, maxDstSize, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTD_copyRawBlock(dst, maxDstSize, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);
+            }
+            ctx->stage = ZSTDds_decodeBlockHeader;
+            ctx->expected = ZSTD_blockHeaderSize;
+            if (ZSTD_isError(rSize)) return rSize;
+            ctx->previousDstEnd = (char*)dst + rSize;
+            return rSize;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static void ZSTD_decompress_insertDictionary(ZSTD_DCtx* ctx, const void* dict, size_t dictSize)
+{
+    ctx->dictEnd = ctx->previousDstEnd;
+    ctx->vBase = (const char*)dict - ((const char*)(ctx->previousDstEnd) - (const char*)(ctx->base));
+    ctx->base = dict;
+    ctx->previousDstEnd = (const char*)dict + dictSize;
+}
+
+
+
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stdlib.h>
+
+
+/** ************************************************
+*  Streaming decompression
+*
+*  A ZBUFF_DCtx object is required to track streaming operation.
+*  Use ZBUFF_createDCtx() and ZBUFF_freeDCtx() to create/release resources.
+*  Use ZBUFF_decompressInit() to start a new decompression operation.
+*  ZBUFF_DCtx objects can be reused multiple times.
+*
+*  Use ZBUFF_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *maxDstSizePtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *maxDstSizePtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to call again the function with remaining input.
+*  The content of dst will be overwritten (up to *maxDstSizePtr) at each function call, so save its content if it matters or change dst .
+*  return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
+*            or 0 when a frame is completely decoded
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory)
+*  output : 128 KB block size is the internal unit, it ensures it's always possible to write a full block when it's decoded.
+*  input : just follow indications from ZBUFF_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* **************************************************/
+
+typedef enum { ZBUFFds_init, ZBUFFds_readHeader, ZBUFFds_loadHeader, ZBUFFds_decodeHeader,
+               ZBUFFds_read, ZBUFFds_load, ZBUFFds_flush } ZBUFF_dStage;
+
+/* *** Resource management *** */
+
+#define ZSTD_frameHeaderSize_max 5   /* too magical, should come from reference */
+struct ZBUFFv04_DCtx_s {
+    ZSTD_DCtx* zc;
+    ZSTD_parameters params;
+    char* inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char* outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t hPos;
+    const char* dict;
+    size_t dictSize;
+    ZBUFF_dStage stage;
+    unsigned char headerBuffer[ZSTD_frameHeaderSize_max];
+};   /* typedef'd to ZBUFF_DCtx within "zstd_buffered.h" */
+
+typedef ZBUFFv04_DCtx ZBUFF_DCtx;
+
+
+static ZBUFF_DCtx* ZBUFF_createDCtx(void)
+{
+    ZBUFF_DCtx* zbc = (ZBUFF_DCtx*)malloc(sizeof(ZBUFF_DCtx));
+    if (zbc==NULL) return NULL;
+    memset(zbc, 0, sizeof(*zbc));
+    zbc->zc = ZSTD_createDCtx();
+    zbc->stage = ZBUFFds_init;
+    return zbc;
+}
+
+static size_t ZBUFF_freeDCtx(ZBUFF_DCtx* zbc)
+{
+    if (zbc==NULL) return 0;   /* support free on null */
+    ZSTD_freeDCtx(zbc->zc);
+    free(zbc->inBuff);
+    free(zbc->outBuff);
+    free(zbc);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+static size_t ZBUFF_decompressInit(ZBUFF_DCtx* zbc)
+{
+    zbc->stage = ZBUFFds_readHeader;
+    zbc->hPos = zbc->inPos = zbc->outStart = zbc->outEnd = zbc->dictSize = 0;
+    return ZSTD_resetDCtx(zbc->zc);
+}
+
+
+static size_t ZBUFF_decompressWithDictionary(ZBUFF_DCtx* zbc, const void* src, size_t srcSize)
+{
+    zbc->dict = (const char*)src;
+    zbc->dictSize = srcSize;
+    return 0;
+}
+
+static size_t ZBUFF_limitCopy(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    size_t length = MIN(maxDstSize, srcSize);
+    if (length > 0) {
+        memcpy(dst, src, length);
+    }
+    return length;
+}
+
+/* *** Decompression *** */
+
+static size_t ZBUFF_decompressContinue(ZBUFF_DCtx* zbc, void* dst, size_t* maxDstSizePtr, const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* ip = istart;
+    const char* const iend = istart + *srcSizePtr;
+    char* const ostart = (char*)dst;
+    char* op = ostart;
+    char* const oend = ostart + *maxDstSizePtr;
+    U32 notDone = 1;
+
+    DEBUGLOG(5, "ZBUFF_decompressContinue");
+    while (notDone)
+    {
+        switch(zbc->stage)
+        {
+
+        case ZBUFFds_init :
+            DEBUGLOG(5, "ZBUFF_decompressContinue: stage==ZBUFFds_init => ERROR(init_missing)");
+            return ERROR(init_missing);
+
+        case ZBUFFds_readHeader :
+            /* read header from src */
+            {   size_t const headerSize = ZSTD_getFrameParams(&(zbc->params), src, *srcSizePtr);
+                if (ZSTD_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    memcpy(zbc->headerBuffer+zbc->hPos, src, *srcSizePtr);
+                    zbc->hPos += *srcSizePtr;
+                    *maxDstSizePtr = 0;
+                    zbc->stage = ZBUFFds_loadHeader;
+                    return headerSize - zbc->hPos;
+                }
+                zbc->stage = ZBUFFds_decodeHeader;
+                break;
+            }
+
+        case ZBUFFds_loadHeader:
+            /* complete header from src */
+            {   size_t headerSize = ZBUFF_limitCopy(
+                    zbc->headerBuffer + zbc->hPos, ZSTD_frameHeaderSize_max - zbc->hPos,
+                    src, *srcSizePtr);
+                zbc->hPos += headerSize;
+                ip += headerSize;
+                headerSize = ZSTD_getFrameParams(&(zbc->params), zbc->headerBuffer, zbc->hPos);
+                if (ZSTD_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    *maxDstSizePtr = 0;
+                    return headerSize - zbc->hPos;
+            }   }
+            /* intentional fallthrough */
+
+        case ZBUFFds_decodeHeader:
+                /* apply header to create / resize buffers */
+                {   size_t const neededOutSize = (size_t)1 << zbc->params.windowLog;
+                    size_t const neededInSize = BLOCKSIZE;   /* a block is never > BLOCKSIZE */
+                    if (zbc->inBuffSize < neededInSize) {
+                        free(zbc->inBuff);
+                        zbc->inBuffSize = neededInSize;
+                        zbc->inBuff = (char*)malloc(neededInSize);
+                        if (zbc->inBuff == NULL) return ERROR(memory_allocation);
+                    }
+                    if (zbc->outBuffSize < neededOutSize) {
+                        free(zbc->outBuff);
+                        zbc->outBuffSize = neededOutSize;
+                        zbc->outBuff = (char*)malloc(neededOutSize);
+                        if (zbc->outBuff == NULL) return ERROR(memory_allocation);
+                }   }
+                if (zbc->dictSize)
+                    ZSTD_decompress_insertDictionary(zbc->zc, zbc->dict, zbc->dictSize);
+                if (zbc->hPos) {
+                    /* some data already loaded into headerBuffer : transfer into inBuff */
+                    memcpy(zbc->inBuff, zbc->headerBuffer, zbc->hPos);
+                    zbc->inPos = zbc->hPos;
+                    zbc->hPos = 0;
+                    zbc->stage = ZBUFFds_load;
+                    break;
+                }
+                zbc->stage = ZBUFFds_read;
+		/* fall-through */
+        case ZBUFFds_read:
+            {
+                size_t neededInSize = ZSTD_nextSrcSizeToDecompress(zbc->zc);
+                if (neededInSize==0)   /* end of frame */
+                {
+                    zbc->stage = ZBUFFds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize)
+                {
+                    /* directly decode from src */
+                    size_t decodedSize = ZSTD_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        ip, neededInSize);
+                    if (ZSTD_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize) break;   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbc->stage = ZBUFFds_load;
+            }
+	    /* fall-through */
+        case ZBUFFds_load:
+            {
+                size_t neededInSize = ZSTD_nextSrcSizeToDecompress(zbc->zc);
+                size_t toLoad = neededInSize - zbc->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbc->inBuffSize - zbc->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFF_limitCopy(zbc->inBuff + zbc->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbc->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+                {
+                    size_t decodedSize = ZSTD_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        zbc->inBuff, neededInSize);
+                    if (ZSTD_isError(decodedSize)) return decodedSize;
+                    zbc->inPos = 0;   /* input is consumed */
+                    if (!decodedSize) { zbc->stage = ZBUFFds_read; break; }   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFds_flush;
+                    /* ZBUFFds_flush follows */
+                }
+            }
+	    /* fall-through */
+        case ZBUFFds_flush:
+            {
+                size_t toFlushSize = zbc->outEnd - zbc->outStart;
+                size_t flushedSize = ZBUFF_limitCopy(op, oend-op, zbc->outBuff + zbc->outStart, toFlushSize);
+                op += flushedSize;
+                zbc->outStart += flushedSize;
+                if (flushedSize == toFlushSize)
+                {
+                    zbc->stage = ZBUFFds_read;
+                    if (zbc->outStart + BLOCKSIZE > zbc->outBuffSize)
+                        zbc->outStart = zbc->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+        }
+    }
+
+    *srcSizePtr = ip-istart;
+    *maxDstSizePtr = op-ostart;
+
+    {
+        size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zbc->zc);
+        if (nextSrcSizeHint > 3) nextSrcSizeHint+= 3;   /* get the next block header while at it */
+        nextSrcSizeHint -= zbc->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+unsigned ZBUFFv04_isError(size_t errorCode) { return ERR_isError(errorCode); }
+const char* ZBUFFv04_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+size_t ZBUFFv04_recommendedDInSize(void)  { return BLOCKSIZE + 3; }
+size_t ZBUFFv04_recommendedDOutSize(void) { return BLOCKSIZE; }
+
+
+
+/*- ========================================================================= -*/
+
+/* final wrapping stage */
+
+size_t ZSTDv04_decompressDCtx(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompress_usingDict(dctx, dst, maxDstSize, src, srcSize, NULL, 0);
+}
+
+size_t ZSTDv04_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+#if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE==1)
+    size_t regenSize;
+    ZSTD_DCtx* dctx = ZSTD_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv04_decompressDCtx(dctx, dst, maxDstSize, src, srcSize);
+    ZSTD_freeDCtx(dctx);
+    return regenSize;
+#else
+    ZSTD_DCtx dctx;
+    return ZSTDv04_decompressDCtx(&dctx, dst, maxDstSize, src, srcSize);
+#endif
+}
+
+size_t ZSTDv04_resetDCtx(ZSTDv04_Dctx* dctx) { return ZSTD_resetDCtx(dctx); }
+
+size_t ZSTDv04_nextSrcSizeToDecompress(ZSTDv04_Dctx* dctx)
+{
+    return ZSTD_nextSrcSizeToDecompress(dctx);
+}
+
+size_t ZSTDv04_decompressContinue(ZSTDv04_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompressContinue(dctx, dst, maxDstSize, src, srcSize);
+}
+
+
+
+ZBUFFv04_DCtx* ZBUFFv04_createDCtx(void) { return ZBUFF_createDCtx(); }
+size_t ZBUFFv04_freeDCtx(ZBUFFv04_DCtx* dctx) { return ZBUFF_freeDCtx(dctx); }
+
+size_t ZBUFFv04_decompressInit(ZBUFFv04_DCtx* dctx) { return ZBUFF_decompressInit(dctx); }
+size_t ZBUFFv04_decompressWithDictionary(ZBUFFv04_DCtx* dctx, const void* src, size_t srcSize)
+{ return ZBUFF_decompressWithDictionary(dctx, src, srcSize); }
+
+size_t ZBUFFv04_decompressContinue(ZBUFFv04_DCtx* dctx, void* dst, size_t* maxDstSizePtr, const void* src, size_t* srcSizePtr)
+{
+    DEBUGLOG(5, "ZBUFFv04_decompressContinue");
+    return ZBUFF_decompressContinue(dctx, dst, maxDstSizePtr, src, srcSizePtr);
+}
+
+ZSTD_DCtx* ZSTDv04_createDCtx(void) { return ZSTD_createDCtx(); }
+size_t ZSTDv04_freeDCtx(ZSTD_DCtx* dctx) { return ZSTD_freeDCtx(dctx); }
diff --git a/lib/legacy/zstd_v05.c b/lib/legacy/zstd_v05.c
new file mode 100644
index 0000000..7a3af42
--- /dev/null
+++ b/lib/legacy/zstd_v05.c
@@ -0,0 +1,4005 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*- Dependencies -*/
+#include "zstd_v05.h"
+#include "../common/error_private.h"
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSEv05 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-****************************************
+*  Dependencies
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+
+/*-****************************************
+*  Compiler specifics
+******************************************/
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/*-**************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-**************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC void MEM_write32(void* memPtr, U32 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC void MEM_write64(void* memPtr, U64 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian()) {
+        MEM_write16(memPtr, val);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd
+*/
+#ifndef ZSTD_STATIC_H
+#define ZSTD_STATIC_H
+
+/* The prototypes defined within this file are considered experimental.
+ * They should not be used in the context DLL as they may change in the future.
+ * Prefer static linking if you need them, to control breaking version changes issues.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*-*************************************
+*  Types
+***************************************/
+#define ZSTDv05_WINDOWLOG_ABSOLUTEMIN 11
+
+
+/*-*************************************
+*  Advanced functions
+***************************************/
+/*- Advanced Decompression functions -*/
+
+/*! ZSTDv05_decompress_usingPreparedDCtx() :
+*   Same as ZSTDv05_decompress_usingDict, but using a reference context `preparedDCtx`, where dictionary has been loaded.
+*   It avoids reloading the dictionary each time.
+*   `preparedDCtx` must have been properly initialized using ZSTDv05_decompressBegin_usingDict().
+*   Requires 2 contexts : 1 for reference, which will not be modified, and 1 to run the decompression operation */
+size_t ZSTDv05_decompress_usingPreparedDCtx(
+                                             ZSTDv05_DCtx* dctx, const ZSTDv05_DCtx* preparedDCtx,
+                                             void* dst, size_t dstCapacity,
+                                       const void* src, size_t srcSize);
+
+
+/* **************************************
+*  Streaming functions (direct mode)
+****************************************/
+size_t ZSTDv05_decompressBegin(ZSTDv05_DCtx* dctx);
+
+/*
+  Streaming decompression, direct mode (bufferless)
+
+  A ZSTDv05_DCtx object is required to track streaming operations.
+  Use ZSTDv05_createDCtx() / ZSTDv05_freeDCtx() to manage it.
+  A ZSTDv05_DCtx object can be re-used multiple times.
+
+  First typical operation is to retrieve frame parameters, using ZSTDv05_getFrameParams().
+  This operation is independent, and just needs enough input data to properly decode the frame header.
+  Objective is to retrieve *params.windowlog, to know minimum amount of memory required during decoding.
+  Result : 0 when successful, it means the ZSTDv05_parameters structure has been filled.
+           >0 : means there is not enough data into src. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTDv05_isError()
+
+  Start decompression, with ZSTDv05_decompressBegin() or ZSTDv05_decompressBegin_usingDict()
+  Alternatively, you can copy a prepared context, using ZSTDv05_copyDCtx()
+
+  Then use ZSTDv05_nextSrcSizeToDecompress() and ZSTDv05_decompressContinue() alternatively.
+  ZSTDv05_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTDv05_decompressContinue().
+  ZSTDv05_decompressContinue() requires this exact amount of bytes, or it will fail.
+  ZSTDv05_decompressContinue() needs previous data blocks during decompression, up to (1 << windowlog).
+  They should preferably be located contiguously, prior to current block. Alternatively, a round buffer is also possible.
+
+  @result of ZSTDv05_decompressContinue() is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTDv05_decompressContinue() has decoded some header.
+
+  A frame is fully decoded when ZSTDv05_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+*/
+
+
+/* **************************************
+*  Block functions
+****************************************/
+/*! Block functions produce and decode raw zstd blocks, without frame metadata.
+    User will have to take in charge required information to regenerate data, such as block sizes.
+
+    A few rules to respect :
+    - Uncompressed block size must be <= 128 KB
+    - Compressing or decompressing requires a context structure
+      + Use ZSTDv05_createCCtx() and ZSTDv05_createDCtx()
+    - It is necessary to init context before starting
+      + compression : ZSTDv05_compressBegin()
+      + decompression : ZSTDv05_decompressBegin()
+      + variants _usingDict() are also allowed
+      + copyCCtx() and copyDCtx() work too
+    - When a block is considered not compressible enough, ZSTDv05_compressBlock() result will be zero.
+      In which case, nothing is produced into `dst`.
+      + User must test for such outcome and deal directly with uncompressed data
+      + ZSTDv05_decompressBlock() doesn't accept uncompressed data as input !!
+*/
+
+size_t ZSTDv05_decompressBlock(ZSTDv05_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv05_STATIC_H */
+
+
+/*
+    zstd_internal - common functions to include
+    Header File for include
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+*/
+#ifndef ZSTD_CCOMMON_H_MODULE
+#define ZSTD_CCOMMON_H_MODULE
+
+
+
+/*-*************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/*-*************************************
+*  Common constants
+***************************************/
+#define ZSTDv05_DICT_MAGIC  0xEC30A435
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+
+static const size_t ZSTDv05_blockHeaderSize = 3;
+static const size_t ZSTDv05_frameHeaderSize_min = 5;
+#define ZSTDv05_frameHeaderSize_max 5         /* define, for static allocation */
+
+#define BITv057 128
+#define BITv056  64
+#define BITv055  32
+#define BITv054  16
+#define BITv051   2
+#define BITv050   1
+
+#define IS_HUFv05 0
+#define IS_PCH 1
+#define IS_RAW 2
+#define IS_RLE 3
+
+#define MINMATCH 4
+#define REPCODE_STARTVALUE 1
+
+#define Litbits  8
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxLit ((1<<Litbits) - 1)
+#define MaxML  ((1<<MLbits) - 1)
+#define MaxLL  ((1<<LLbits) - 1)
+#define MaxOff ((1<<Offbits)- 1)
+#define MLFSEv05Log   10
+#define LLFSEv05Log   10
+#define OffFSEv05Log   9
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define FSEv05_ENCODING_RAW     0
+#define FSEv05_ENCODING_RLE     1
+#define FSEv05_ENCODING_STATIC  2
+#define FSEv05_ENCODING_DYNAMIC 3
+
+
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
+
+#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+
+#define WILDCOPY_OVERLENGTH 8
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+
+/*-*******************************************
+*  Shared functions to include for inlining
+*********************************************/
+static void ZSTDv05_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTDv05_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTDv05_wildcopy() :
+*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
+MEM_STATIC void ZSTDv05_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+/*-*******************************************
+*  Private interfaces
+*********************************************/
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* offCode;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* litLengthStart;
+    BYTE* litLength;
+    BYTE* matchLengthStart;
+    BYTE* matchLength;
+    BYTE* dumpsStart;
+    BYTE* dumps;
+    /* opt */
+    U32* matchLengthFreq;
+    U32* litLengthFreq;
+    U32* litFreq;
+    U32* offCodeFreq;
+    U32  matchLengthSum;
+    U32  litLengthSum;
+    U32  litSum;
+    U32  offCodeSum;
+} SeqStore_t;
+
+
+
+#endif   /* ZSTDv05_CCOMMON_H_MODULE */
+/* ******************************************************************
+   FSEv05 : Finite State Entropy coder
+   header file
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSEv05_H
+#define FSEv05_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+
+
+/*-****************************************
+*  FSEv05 simple functions
+******************************************/
+size_t FSEv05_decompress(void* dst,  size_t maxDstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+FSEv05_decompress():
+    Decompress FSEv05 data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'maxDstSize'.
+    return : size of regenerated data (<= maxDstSize)
+             or an error code, which can be tested using FSEv05_isError()
+
+    ** Important ** : FSEv05_decompress() doesn't decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+
+
+/* *****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+unsigned    FSEv05_isError(size_t code);        /* tells if a return value is an error code */
+const char* FSEv05_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+
+
+/* *****************************************
+*  FSEv05 detailed API
+******************************************/
+/* *** DECOMPRESSION *** */
+
+/*!
+FSEv05_readNCount():
+   Read compactly saved 'normalizedCounter' from 'rBuffer'.
+   return : size read from 'rBuffer'
+            or an errorCode, which can be tested using FSEv05_isError()
+            maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+size_t FSEv05_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*!
+Constructor and Destructor of type FSEv05_DTable
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSEv05_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+FSEv05_DTable* FSEv05_createDTable(unsigned tableLog);
+void        FSEv05_freeDTable(FSEv05_DTable* dt);
+
+/*!
+FSEv05_buildDTable():
+   Builds 'dt', which must be already allocated, using FSEv05_createDTable()
+   @return : 0,
+             or an errorCode, which can be tested using FSEv05_isError() */
+size_t FSEv05_buildDTable (FSEv05_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*!
+FSEv05_decompress_usingDTable():
+   Decompress compressed source @cSrc of size @cSrcSize using `dt`
+   into `dst` which must be already allocated.
+   @return : size of regenerated data (necessarily <= @dstCapacity)
+             or an errorCode, which can be tested using FSEv05_isError() */
+size_t FSEv05_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSEv05_DTable* dt);
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv05_H */
+/* ******************************************************************
+   bitstream
+   Part of FSEv05 library
+   header file (to include)
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef BITv05STREAM_H_MODULE
+#define BITv05STREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+
+/*-********************************************
+*  bitStream decoding API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BITv05_DStream_t;
+
+typedef enum { BITv05_DStream_unfinished = 0,
+               BITv05_DStream_endOfBuffer = 1,
+               BITv05_DStream_completed = 2,
+               BITv05_DStream_overflow = 3 } BITv05_DStream_status;  /* result of BITv05_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BITv05_initDStream(BITv05_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BITv05_readBits(BITv05_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BITv05_DStream_status BITv05_reloadDStream(BITv05_DStream_t* bitD);
+MEM_STATIC unsigned BITv05_endOfDStream(const BITv05_DStream_t* bitD);
+
+
+/*-****************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BITv05_readBitsFast(BITv05_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/*-**************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BITv05_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/*-********************************************************
+* bitStream decoding
+**********************************************************/
+/*!BITv05_initDStream
+*  Initialize a BITv05_DStream_t.
+*  @bitD : a pointer to an already allocated BITv05_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BITv05_initDStream(BITv05_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t)) {  /* normal case */
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BITv05_highbit32(contain32);
+    } else {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BITv05_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+MEM_STATIC size_t BITv05_lookBits(BITv05_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BITv05_lookBitsFast :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv05_lookBitsFast(BITv05_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BITv05_skipBits(BITv05_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BITv05_readBits(BITv05_DStream_t* bitD, unsigned nbBits)
+{
+    size_t value = BITv05_lookBits(bitD, nbBits);
+    BITv05_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BITv05_readBitsFast :
+*  unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv05_readBitsFast(BITv05_DStream_t* bitD, unsigned nbBits)
+{
+    size_t value = BITv05_lookBitsFast(bitD, nbBits);
+    BITv05_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BITv05_DStream_status BITv05_reloadDStream(BITv05_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BITv05_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer)) {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BITv05_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start) {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BITv05_DStream_endOfBuffer;
+        return BITv05_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BITv05_DStream_status result = BITv05_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start) {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BITv05_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BITv05_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BITv05_endOfDStream(const BITv05_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITv05STREAM_H_MODULE */
+/* ******************************************************************
+   FSEv05 : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSEv05_STATIC_H
+#define FSEv05_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* It is possible to statically allocate FSEv05 CTable/DTable as a table of unsigned using below macros */
+#define FSEv05_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSEv05 advanced API
+*******************************************/
+size_t FSEv05_buildDTable_raw (FSEv05_DTable* dt, unsigned nbBits);
+/* build a fake FSEv05_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+size_t FSEv05_buildDTable_rle (FSEv05_DTable* dt, unsigned char symbolValue);
+/* build a fake FSEv05_DTable, designed to always generate the same symbolValue */
+
+
+
+/* *****************************************
+*  FSEv05 symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSEv05_DState_t;
+
+
+static void     FSEv05_initDState(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD, const FSEv05_DTable* dt);
+
+static unsigned char FSEv05_decodeSymbol(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD);
+
+static unsigned FSEv05_endOfDState(const FSEv05_DState_t* DStatePtr);
+
+
+
+/* *****************************************
+*  FSEv05 unsafe API
+*******************************************/
+static unsigned char FSEv05_decodeSymbolFast(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSEv05_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSEv05_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSEv05_initDState(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD, const FSEv05_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv05_DTableHeader* const DTableH = (const FSEv05_DTableHeader*)ptr;
+    DStatePtr->state = BITv05_readBits(bitD, DTableH->tableLog);
+    BITv05_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSEv05_peakSymbol(FSEv05_DState_t* DStatePtr)
+{
+    const FSEv05_decode_t DInfo = ((const FSEv05_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC BYTE FSEv05_decodeSymbol(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD)
+{
+    const FSEv05_decode_t DInfo = ((const FSEv05_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BITv05_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSEv05_decodeSymbolFast(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD)
+{
+    const FSEv05_decode_t DInfo = ((const FSEv05_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BITv05_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSEv05_endOfDState(const FSEv05_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv05_STATIC_H */
+/* ******************************************************************
+   FSEv05 : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSEv05 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSEv05_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSEv05_MAX_MEMORY_USAGE 14
+#define FSEv05_DEFAULT_MEMORY_USAGE 13
+
+/*!FSEv05_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSEv05_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSEv05_FUNCTION_TYPE BYTE
+#define FSEv05_FUNCTION_EXTENSION
+#define FSEv05_DECODE_TYPE FSEv05_decode_t
+
+
+#endif   /* !FSEv05_COMMONDEFS_ONLY */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSEv05_MAX_TABLELOG  (FSEv05_MAX_MEMORY_USAGE-2)
+#define FSEv05_MAX_TABLESIZE (1U<<FSEv05_MAX_TABLELOG)
+#define FSEv05_MAXTABLESIZE_MASK (FSEv05_MAX_TABLESIZE-1)
+#define FSEv05_DEFAULT_TABLELOG (FSEv05_DEFAULT_MEMORY_USAGE-2)
+#define FSEv05_MIN_TABLELOG 5
+
+#define FSEv05_TABLELOG_ABSOLUTE_MAX 15
+#if FSEv05_MAX_TABLELOG > FSEv05_TABLELOG_ABSOLUTE_MAX
+#error "FSEv05_MAX_TABLELOG > FSEv05_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSEv05_STATIC_ASSERT(c) { enum { FSEv05_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef unsigned DTable_max_t[FSEv05_DTABLE_SIZE_U32(FSEv05_MAX_TABLELOG)];
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSEv05_FUNCTION_EXTENSION
+#  error "FSEv05_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSEv05_FUNCTION_TYPE
+#  error "FSEv05_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSEv05_CAT(X,Y) X##Y
+#define FSEv05_FUNCTION_NAME(X,Y) FSEv05_CAT(X,Y)
+#define FSEv05_TYPE_NAME(X,Y) FSEv05_CAT(X,Y)
+
+
+/* Function templates */
+static U32 FSEv05_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+
+
+FSEv05_DTable* FSEv05_createDTable (unsigned tableLog)
+{
+    if (tableLog > FSEv05_TABLELOG_ABSOLUTE_MAX) tableLog = FSEv05_TABLELOG_ABSOLUTE_MAX;
+    return (FSEv05_DTable*)malloc( FSEv05_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
+}
+
+void FSEv05_freeDTable (FSEv05_DTable* dt)
+{
+    free(dt);
+}
+
+size_t FSEv05_buildDTable(FSEv05_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    FSEv05_DTableHeader DTableH;
+    void* const tdPtr = dt+1;   /* because dt is unsigned, 32-bits aligned on 32-bits */
+    FSEv05_DECODE_TYPE* const tableDecode = (FSEv05_DECODE_TYPE*) (tdPtr);
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSEv05_tableStep(tableSize);
+    U16 symbolNext[FSEv05_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSEv05_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSEv05_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    memset(tableDecode, 0, sizeof(FSEv05_FUNCTION_TYPE) * (maxSymbolValue+1) );   /* useless init, but keep static analyzer happy, and we don't need to performance optimize legacy decoders */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++) {
+        if (normalizedCounter[s]==-1) {
+            tableDecode[highThreshold--].symbol = (FSEv05_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        } else {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+    }   }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++) {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++) {
+            tableDecode[position].symbol = (FSEv05_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+    }   }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++) {
+            FSEv05_FUNCTION_TYPE symbol = (FSEv05_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BITv05_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+    }   }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));
+    return 0;
+}
+
+
+#ifndef FSEv05_COMMONDEFS_ONLY
+/*-****************************************
+*  FSEv05 helper functions
+******************************************/
+unsigned FSEv05_isError(size_t code) { return ERR_isError(code); }
+
+const char* FSEv05_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/*-**************************************************************
+*  FSEv05 NCount encoding-decoding
+****************************************************************/
+static short FSEv05_abs(short a) { return a<0 ? -a : a; }
+
+
+size_t FSEv05_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSEv05_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSEv05_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr)) {
+        if (previous0) {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF) {
+                n0+=24;
+                if (ip < iend-5) {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                } else {
+                    bitStream >>= 16;
+                    bitCount+=16;
+            }   }
+            while ((bitStream & 3) == 3) {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max) {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            } else {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSEv05_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold) {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+    }   }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+
+/*-*******************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+size_t FSEv05_buildDTable_rle (FSEv05_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSEv05_DTableHeader* const DTableH = (FSEv05_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv05_decode_t* const cell = (FSEv05_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+size_t FSEv05_buildDTable_raw (FSEv05_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSEv05_DTableHeader* const DTableH = (FSEv05_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv05_decode_t* const dinfo = (FSEv05_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++) {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSEv05_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSEv05_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BITv05_DStream_t bitD;
+    FSEv05_DState_t state1;
+    FSEv05_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BITv05_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSEv05_isError(errorCode)) return errorCode;
+
+    FSEv05_initDState(&state1, &bitD, dt);
+    FSEv05_initDState(&state2, &bitD, dt);
+
+#define FSEv05_GETSYMBOL(statePtr) fast ? FSEv05_decodeSymbolFast(statePtr, &bitD) : FSEv05_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BITv05_reloadDStream(&bitD)==BITv05_DStream_unfinished) && (op<olimit) ; op+=4) {
+        op[0] = FSEv05_GETSYMBOL(&state1);
+
+        if (FSEv05_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv05_reloadDStream(&bitD);
+
+        op[1] = FSEv05_GETSYMBOL(&state2);
+
+        if (FSEv05_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BITv05_reloadDStream(&bitD) > BITv05_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSEv05_GETSYMBOL(&state1);
+
+        if (FSEv05_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv05_reloadDStream(&bitD);
+
+        op[3] = FSEv05_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BITv05_reloadDStream(&bitD) >= FSEv05_DStream_partiallyFilled; Ends at exactly BITv05_DStream_completed */
+    while (1) {
+        if ( (BITv05_reloadDStream(&bitD)>BITv05_DStream_completed) || (op==omax) || (BITv05_endOfDStream(&bitD) && (fast || FSEv05_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSEv05_GETSYMBOL(&state1);
+
+        if ( (BITv05_reloadDStream(&bitD)>BITv05_DStream_completed) || (op==omax) || (BITv05_endOfDStream(&bitD) && (fast || FSEv05_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSEv05_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BITv05_endOfDStream(&bitD) && FSEv05_endOfDState(&state1) && FSEv05_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+size_t FSEv05_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSEv05_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv05_DTableHeader* DTableH = (const FSEv05_DTableHeader*)ptr;
+    const U32 fastMode = DTableH->fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSEv05_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSEv05_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+size_t FSEv05_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSEv05_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSEv05_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSEv05 decoding mode */
+    errorCode = FSEv05_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSEv05_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSEv05_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSEv05_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSEv05_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSEv05_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFF0_H
+#define HUFF0_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* ****************************************
+*  Huff0 simple functions
+******************************************/
+size_t HUFv05_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+HUFv05_decompress():
+    Decompress Huff0 data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstSize'.
+    @dstSize : must be the **exact** size of original (uncompressed) data.
+    Note : in contrast with FSEv05, HUFv05_decompress can regenerate
+           RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+           because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize)
+              or an error code, which can be tested using HUFv05_isError()
+*/
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+unsigned    HUFv05_isError(size_t code);        /* tells if a return value is an error code */
+const char* HUFv05_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUF0_H */
+/* ******************************************************************
+   Huff0 : Huffman codec, part of New Generation Entropy library
+   header file, for static linking only
+   Copyright (C) 2013-2016, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUF0_STATIC_H
+#define HUF0_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* ****************************************
+*  Static allocation
+******************************************/
+/* static allocation of Huff0's DTable */
+#define HUFv05_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))
+#define HUFv05_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUFv05_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv05_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUFv05_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv05_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUFv05_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+size_t HUFv05_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv05_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+/* ****************************************
+*  Huff0 detailed API
+******************************************/
+/*!
+HUFv05_decompress() does the following:
+1. select the decompression algorithm (X2, X4, X6) based on pre-computed heuristics
+2. build Huffman table from save, using HUFv05_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUFv05_decompressSXn_usingDTable
+*/
+size_t HUFv05_readDTableX2 (unsigned short* DTable, const void* src, size_t srcSize);
+size_t HUFv05_readDTableX4 (unsigned* DTable, const void* src, size_t srcSize);
+
+size_t HUFv05_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv05_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+/* single stream variants */
+
+size_t HUFv05_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv05_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+
+size_t HUFv05_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv05_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* HUF0_STATIC_H */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSEv05+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+#define HUFv05_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUFv05_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUFv05_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUFv05_ABSOLUTEMAX_TABLELOG */
+#define HUFv05_DEFAULT_TABLELOG  HUFv05_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUFv05_MAX_SYMBOL_VALUE 255
+#if (HUFv05_MAX_TABLELOG > HUFv05_ABSOLUTEMAX_TABLELOG)
+#  error "HUFv05_MAX_TABLELOG is too large !"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+unsigned HUFv05_isError(size_t code) { return ERR_isError(code); }
+const char* HUFv05_getErrorName(size_t code) { return ERR_getErrorName(code); }
+#define HUFv05_STATIC_ASSERT(c) { enum { HUFv05_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* *******************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUFv05_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUFv05_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUFv05_readStats
+    Read compact Huffman tree, saved by HUFv05_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUFv05_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    /* memset(huffWeight, 0, hwSize); */   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  { /* special header */
+        if (iSize >= (242)) {  /* RLE */
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else {   /* Incompressible */
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2) {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+    }   }   }
+    else  {   /* header compressed with FSEv05 (normal case) */
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSEv05_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSEv05_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUFv05_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++) {
+        if (huffWeight[n] >= HUFv05_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BITv05_highbit32(weightTotal) + 1;
+    if (tableLog > HUFv05_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {   /* determine last weight */
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BITv05_highbit32(rest);
+        U32 lastWeight = BITv05_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+
+size_t HUFv05_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUFv05_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUFv05_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    size_t iSize;
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* const dtPtr = DTable + 1;
+    HUFv05_DEltX2* const dt = (HUFv05_DEltX2*)dtPtr;
+
+    HUFv05_STATIC_ASSERT(sizeof(HUFv05_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv05_readStats(huffWeight, HUFv05_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv05_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof allocated DTable, from used size of DTable, in case of re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++) {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++) {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUFv05_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUFv05_decodeSymbolX2(BITv05_DStream_t* Dstream, const HUFv05_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BITv05_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BITv05_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUFv05_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUFv05_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUFv05_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv05_MAX_TABLELOG<=12)) \
+        HUFv05_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUFv05_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUFv05_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUFv05_decodeStreamX2(BYTE* p, BITv05_DStream_t* const bitDPtr, BYTE* const pEnd, const HUFv05_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p <= pEnd-4)) {
+        HUFv05_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p < pEnd))
+        HUFv05_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUFv05_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+size_t HUFv05_decompress1X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + dstSize;
+    const U32 dtLog = DTable[0];
+    const void* dtPtr = DTable;
+    const HUFv05_DEltX2* const dt = ((const HUFv05_DEltX2*)dtPtr)+1;
+    BITv05_DStream_t bitD;
+
+    if (dstSize <= cSrcSize) return ERROR(dstSize_tooSmall);
+    { size_t const errorCode = BITv05_initDStream(&bitD, cSrc, cSrcSize);
+      if (HUFv05_isError(errorCode)) return errorCode; }
+
+    HUFv05_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv05_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+size_t HUFv05_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX2(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUFv05_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv05_decompress1X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+size_t HUFv05_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv05_DEltX2* const dt = ((const HUFv05_DEltX2*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BITv05_DStream_t bitD1;
+        BITv05_DStream_t bitD2;
+        BITv05_DStream_t bitD3;
+        BITv05_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv05_initDStream(&bitD1, istart1, length1);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD2, istart2, length2);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD3, istart3, length3);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD4, istart4, length4);
+        if (HUFv05_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv05_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_0(op4, &bitD4);
+            endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv05_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv05_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv05_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv05_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv05_endOfDStream(&bitD1) & BITv05_endOfDStream(&bitD2) & BITv05_endOfDStream(&bitD3) & BITv05_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv05_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX2(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUFv05_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv05_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+
+static void HUFv05_fillDTableX4Level2(HUFv05_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUFv05_DEltX4 DElt;
+    U32 rankVal[HUFv05_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1) {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {   /* note : sortedSymbols already skipped */
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUFv05_ABSOLUTEMAX_TABLELOG][HUFv05_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUFv05_fillDTableX4(HUFv05_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUFv05_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits) {   /* enough room for a second symbol */
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUFv05_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        } else {
+            U32 i;
+            const U32 end = start + length;
+            HUFv05_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+size_t HUFv05_readDTableX4 (unsigned* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUFv05_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUFv05_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUFv05_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUFv05_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    size_t iSize;
+    void* dtPtr = DTable;
+    HUFv05_DEltX4* const dt = ((HUFv05_DEltX4*)dtPtr) + 1;
+
+    HUFv05_STATIC_ASSERT(sizeof(HUFv05_DEltX4) == sizeof(unsigned));   /* if compilation fails here, assertion is false */
+    if (memLog > HUFv05_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    /* memset(weightList, 0, sizeof(weightList)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv05_readStats(weightList, HUFv05_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv05_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++) {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++) {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++) {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++) {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }
+
+    HUFv05_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUFv05_decodeSymbolX4(void* op, BITv05_DStream_t* DStream, const HUFv05_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv05_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BITv05_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUFv05_decodeLastSymbolX4(void* op, BITv05_DStream_t* DStream, const HUFv05_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv05_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BITv05_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BITv05_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }   }
+    return 1;
+}
+
+
+#define HUFv05_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUFv05_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv05_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv05_MAX_TABLELOG<=12)) \
+        ptr += HUFv05_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv05_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUFv05_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUFv05_decodeStreamX4(BYTE* p, BITv05_DStream_t* bitDPtr, BYTE* const pEnd, const HUFv05_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p < pEnd-7)) {
+        HUFv05_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p <= pEnd-2))
+        HUFv05_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUFv05_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUFv05_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+size_t HUFv05_decompress1X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const unsigned* DTable)
+{
+    const BYTE* const istart = (const BYTE*) cSrc;
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* const oend = ostart + dstSize;
+
+    const U32 dtLog = DTable[0];
+    const void* const dtPtr = DTable;
+    const HUFv05_DEltX4* const dt = ((const HUFv05_DEltX4*)dtPtr) +1;
+    size_t errorCode;
+
+    /* Init */
+    BITv05_DStream_t bitD;
+    errorCode = BITv05_initDStream(&bitD, istart, cSrcSize);
+    if (HUFv05_isError(errorCode)) return errorCode;
+
+    /* finish bitStreams one by one */
+    HUFv05_decodeStreamX4(ostart, &bitD, oend,     dt, dtLog);
+
+    /* check */
+    if (!BITv05_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+size_t HUFv05_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX4(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv05_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv05_decompress1X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+size_t HUFv05_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const unsigned* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv05_DEltX4* const dt = ((const HUFv05_DEltX4*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BITv05_DStream_t bitD1;
+        BITv05_DStream_t bitD2;
+        BITv05_DStream_t bitD3;
+        BITv05_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv05_initDStream(&bitD1, istart1, length1);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD2, istart2, length2);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD3, istart3, length3);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD4, istart4, length4);
+        if (HUFv05_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv05_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv05_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv05_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv05_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv05_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv05_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv05_endOfDStream(&bitD1) & BITv05_endOfDStream(&bitD2) & BITv05_endOfDStream(&bitD3) & BITv05_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv05_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX4(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv05_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv05_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/* ********************************/
+/* Generic decompression selector */
+/* ********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+size_t HUFv05_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUFv05_decompress4X2, HUFv05_decompress4X4, NULL };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize >= dstSize) return ERROR(corruption_detected);   /* invalid, or not compressed, but not compressed already dealt with */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    /* return HUFv05_decompress4X2(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams single-symbol decoding */
+    /* return HUFv05_decompress4X4(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams double-symbols decoding */
+    /* return HUFv05_decompress4X6(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams quad-symbols decoding */
+}
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTDv05_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTDv05_HEAPMODE
+#  define ZSTDv05_HEAPMODE 1
+#endif
+
+
+/*-*******************************************************
+*  Dependencies
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug only : printf */
+
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/*-*************************************
+*  Local types
+***************************************/
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTDv05_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/* *************************************
+*  Error Management
+***************************************/
+/*! ZSTDv05_isError() :
+*   tells if a return value is an error code */
+unsigned ZSTDv05_isError(size_t code) { return ERR_isError(code); }
+
+
+/*! ZSTDv05_getErrorName() :
+*   provides error code string (useful for debugging) */
+const char* ZSTDv05_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* *************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDv05ds_getFrameHeaderSize, ZSTDv05ds_decodeFrameHeader,
+               ZSTDv05ds_decodeBlockHeader, ZSTDv05ds_decompressBlock } ZSTDv05_dStage;
+
+struct ZSTDv05_DCtx_s
+{
+    FSEv05_DTable LLTable[FSEv05_DTABLE_SIZE_U32(LLFSEv05Log)];
+    FSEv05_DTable OffTable[FSEv05_DTABLE_SIZE_U32(OffFSEv05Log)];
+    FSEv05_DTable MLTable[FSEv05_DTABLE_SIZE_U32(MLFSEv05Log)];
+    unsigned   hufTableX4[HUFv05_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    size_t headerSize;
+    ZSTDv05_parameters params;
+    blockType_t bType;   /* used in ZSTDv05_decompressContinue(), to transfer blockType between header decoding and block decoding stages */
+    ZSTDv05_dStage stage;
+    U32 flagStaticTables;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + WILDCOPY_OVERLENGTH];
+    BYTE headerBuffer[ZSTDv05_frameHeaderSize_max];
+};  /* typedef'd to ZSTDv05_DCtx within "zstd_static.h" */
+
+size_t ZSTDv05_sizeofDCtx (void); /* Hidden declaration */
+size_t ZSTDv05_sizeofDCtx (void) { return sizeof(ZSTDv05_DCtx); }
+
+size_t ZSTDv05_decompressBegin(ZSTDv05_DCtx* dctx)
+{
+    dctx->expected = ZSTDv05_frameHeaderSize_min;
+    dctx->stage = ZSTDv05ds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    dctx->hufTableX4[0] = ZSTD_HUFFDTABLE_CAPACITY_LOG;
+    dctx->flagStaticTables = 0;
+    return 0;
+}
+
+ZSTDv05_DCtx* ZSTDv05_createDCtx(void)
+{
+    ZSTDv05_DCtx* dctx = (ZSTDv05_DCtx*)malloc(sizeof(ZSTDv05_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTDv05_decompressBegin(dctx);
+    return dctx;
+}
+
+size_t ZSTDv05_freeDCtx(ZSTDv05_DCtx* dctx)
+{
+    free(dctx);
+    return 0;   /* reserved as a potential error code in the future */
+}
+
+void ZSTDv05_copyDCtx(ZSTDv05_DCtx* dstDCtx, const ZSTDv05_DCtx* srcDCtx)
+{
+    memcpy(dstDCtx, srcDCtx,
+           sizeof(ZSTDv05_DCtx) - (BLOCKSIZE+WILDCOPY_OVERLENGTH + ZSTDv05_frameHeaderSize_max));  /* no need to copy workspace */
+}
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+
+/* Frame format description
+   Frame Header -  [ Block Header - Block ] - Frame End
+   1) Frame Header
+      - 4 bytes - Magic Number : ZSTDv05_MAGICNUMBER (defined within zstd_internal.h)
+      - 1 byte  - Window Descriptor
+   2) Block Header
+      - 3 bytes, starting with a 2-bits descriptor
+                 Uncompressed, Compressed, Frame End, unused
+   3) Block
+      See Block Format Description
+   4) Frame End
+      - 3 bytes, compatible with Block Header
+*/
+
+/* Block format description
+
+   Block = Literal Section - Sequences Section
+   Prerequisite : size of (compressed) block, maximum size of regenerated data
+
+   1) Literal Section
+
+   1.1) Header : 1-5 bytes
+        flags: 2 bits
+            00 compressed by Huff0
+            01 unused
+            10 is Raw (uncompressed)
+            11 is Rle
+            Note : using 01 => Huff0 with precomputed table ?
+            Note : delta map ? => compressed ?
+
+   1.1.1) Huff0-compressed literal block : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+   1.1.2) Raw (uncompressed) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RAW<<6) + (0<<4) + size
+               12 bits: (IS_RAW<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RAW<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.3) Rle (repeated single byte) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RLE<<6) + (0<<4) + size
+               12 bits: (IS_RLE<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RLE<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.4) Huff0-compressed literal block, using precomputed CTables : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+        1- CTable available (stored into workspace ?)
+        2- Small input (fast heuristic ? Full comparison ? depend on clevel ?)
+
+
+   1.2) Literal block content
+
+   1.2.1) Huff0 block, using sizes from header
+        See Huff0 format
+
+   1.2.2) Huff0 block, using prepared table
+
+   1.2.3) Raw content
+
+   1.2.4) single byte
+
+
+   2) Sequences section
+      TO DO
+*/
+
+
+/** ZSTDv05_decodeFrameHeader_Part1() :
+*   decode the 1st part of the Frame Header, which tells Frame Header size.
+*   srcSize must be == ZSTDv05_frameHeaderSize_min.
+*   @return : the full size of the Frame Header */
+static size_t ZSTDv05_decodeFrameHeader_Part1(ZSTDv05_DCtx* zc, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize != ZSTDv05_frameHeaderSize_min)
+        return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTDv05_MAGICNUMBER) return ERROR(prefix_unknown);
+    zc->headerSize = ZSTDv05_frameHeaderSize_min;
+    return zc->headerSize;
+}
+
+
+size_t ZSTDv05_getFrameParams(ZSTDv05_parameters* params, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize < ZSTDv05_frameHeaderSize_min) return ZSTDv05_frameHeaderSize_max;
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTDv05_MAGICNUMBER) return ERROR(prefix_unknown);
+    memset(params, 0, sizeof(*params));
+    params->windowLog = (((const BYTE*)src)[4] & 15) + ZSTDv05_WINDOWLOG_ABSOLUTEMIN;
+    if ((((const BYTE*)src)[4] >> 4) != 0) return ERROR(frameParameter_unsupported);   /* reserved bits */
+    return 0;
+}
+
+/** ZSTDv05_decodeFrameHeader_Part2() :
+*   decode the full Frame Header.
+*   srcSize must be the size provided by ZSTDv05_decodeFrameHeader_Part1().
+*   @return : 0, or an error code, which can be tested using ZSTDv05_isError() */
+static size_t ZSTDv05_decodeFrameHeader_Part2(ZSTDv05_DCtx* zc, const void* src, size_t srcSize)
+{
+    size_t result;
+    if (srcSize != zc->headerSize)
+        return ERROR(srcSize_wrong);
+    result = ZSTDv05_getFrameParams(&(zc->params), src, srcSize);
+    if ((MEM_32bits()) && (zc->params.windowLog > 25)) return ERROR(frameParameter_unsupported);
+    return result;
+}
+
+
+static size_t ZSTDv05_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE*)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3)
+        return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+
+static size_t ZSTDv05_copyRawBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (dst==NULL) return ERROR(dstSize_tooSmall);
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+/*! ZSTDv05_decodeLiteralsBlock() :
+    @return : nb of bytes read from src (< srcSize ) */
+static size_t ZSTDv05_decodeLiteralsBlock(ZSTDv05_DCtx* dctx,
+                                    const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(istart[0]>> 6)
+    {
+    case IS_HUFv05:
+        {
+            size_t litSize, litCSize, singleStream=0;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3 */
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                /* 2 - 2 - 10 - 10 */
+                lhSize=3;
+                singleStream = istart[0] & 16;
+                litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+                litCSize = ((istart[1] &  3) << 8) + istart[2];
+                break;
+            case 2:
+                /* 2 - 2 - 14 - 14 */
+                lhSize=4;
+                litSize  = ((istart[0] & 15) << 10) + (istart[1] << 2) + (istart[2] >> 6);
+                litCSize = ((istart[2] & 63) <<  8) + istart[3];
+                break;
+            case 3:
+                /* 2 - 2 - 18 - 18 */
+                lhSize=5;
+                litSize  = ((istart[0] & 15) << 14) + (istart[1] << 6) + (istart[2] >> 2);
+                litCSize = ((istart[2] &  3) << 16) + (istart[3] << 8) + istart[4];
+                break;
+            }
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            if (HUFv05_isError(singleStream ?
+                            HUFv05_decompress1X2(dctx->litBuffer, litSize, istart+lhSize, litCSize) :
+                            HUFv05_decompress   (dctx->litBuffer, litSize, istart+lhSize, litCSize) ))
+                return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_PCH:
+        {
+            size_t errorCode;
+            size_t litSize, litCSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (lhSize != 1)  /* only case supported for now : small litSize, single stream */
+                return ERROR(corruption_detected);
+            if (!dctx->flagStaticTables)
+                return ERROR(dictionary_corrupted);
+
+            /* 2 - 2 - 10 - 10 */
+            lhSize=3;
+            litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+            litCSize = ((istart[1] &  3) << 8) + istart[2];
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            errorCode = HUFv05_decompress1X4_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->hufTableX4);
+            if (HUFv05_isError(errorCode)) return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_RAW:
+        {
+            size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize=1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                break;
+            }
+
+            if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                return lhSize+litSize;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+lhSize;
+            dctx->litSize = litSize;
+            return lhSize+litSize;
+        }
+    case IS_RLE:
+        {
+            size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize = 1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
+                break;
+            }
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return lhSize+1;
+        }
+    default:
+        return ERROR(corruption_detected);   /* impossible */
+    }
+}
+
+
+static size_t ZSTDv05_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSEv05_DTable* DTableLL, FSEv05_DTable* DTableML, FSEv05_DTable* DTableOffb,
+                         const void* src, size_t srcSize, U32 flagStaticTable)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    unsigned LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < MIN_SEQUENCES_SIZE)
+        return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = *ip++;
+    if (*nbSeq==0) return 1;
+    if (*nbSeq >= 128) {
+        if (ip >= iend) return ERROR(srcSize_wrong);
+        *nbSeq = ((nbSeq[0]-128)<<8) + *ip++;
+    }
+
+    if (ip >= iend) return ERROR(srcSize_wrong);
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2) {
+        if (ip+3 > iend) return ERROR(srcSize_wrong);
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    } else {
+        if (ip+2 > iend) return ERROR(srcSize_wrong);
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL >= MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case FSEv05_ENCODING_RLE :
+            LLlog = 0;
+            FSEv05_buildDTable_rle(DTableLL, *ip++);
+            break;
+        case FSEv05_ENCODING_RAW :
+            LLlog = LLbits;
+            FSEv05_buildDTable_raw(DTableLL, LLbits);
+            break;
+        case FSEv05_ENCODING_STATIC:
+            if (!flagStaticTable) return ERROR(corruption_detected);
+            break;
+        case FSEv05_ENCODING_DYNAMIC :
+        default :   /* impossible */
+            {   unsigned max = MaxLL;
+                headerSize = FSEv05_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSEv05_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSEv05Log) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSEv05_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case FSEv05_ENCODING_RLE :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSEv05_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case FSEv05_ENCODING_RAW :
+            Offlog = Offbits;
+            FSEv05_buildDTable_raw(DTableOffb, Offbits);
+            break;
+        case FSEv05_ENCODING_STATIC:
+            if (!flagStaticTable) return ERROR(corruption_detected);
+            break;
+        case FSEv05_ENCODING_DYNAMIC :
+        default :   /* impossible */
+            {   unsigned max = MaxOff;
+                headerSize = FSEv05_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSEv05_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSEv05Log) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSEv05_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case FSEv05_ENCODING_RLE :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSEv05_buildDTable_rle(DTableML, *ip++);
+            break;
+        case FSEv05_ENCODING_RAW :
+            MLlog = MLbits;
+            FSEv05_buildDTable_raw(DTableML, MLbits);
+            break;
+        case FSEv05_ENCODING_STATIC:
+            if (!flagStaticTable) return ERROR(corruption_detected);
+            break;
+        case FSEv05_ENCODING_DYNAMIC :
+        default :   /* impossible */
+            {   unsigned max = MaxML;
+                headerSize = FSEv05_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSEv05_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSEv05Log) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSEv05_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+} seq_t;
+
+typedef struct {
+    BITv05_DStream_t DStream;
+    FSEv05_DState_t stateLL;
+    FSEv05_DState_t stateOffb;
+    FSEv05_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+
+static void ZSTDv05_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSEv05_peakSymbol(&(seqState->stateLL));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    if (litLength == MaxLL) {
+        const U32 add = *dumps++;
+        if (add < 255) litLength += add;
+        else if (dumps + 2 <= de) {
+            litLength = MEM_readLE16(dumps);
+            dumps += 2;
+            if ((litLength & 1) && dumps < de) {
+                litLength += *dumps << 16;
+                dumps += 1;
+            }
+            litLength>>=1;
+        }
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {
+        static const U32 offsetPrefix[MaxOff+1] = {
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode = FSEv05_peakSymbol(&(seqState->stateOffb));   /* <= maxOff, by table construction */
+        U32 nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BITv05_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BITv05_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* repcode, cmove */
+        if (offsetCode | !litLength) seqState->prevOffset = seq->offset;   /* cmove */
+        FSEv05_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));    /* update */
+    }
+
+    /* Literal length update */
+    FSEv05_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));   /* update */
+    if (MEM_32bits()) BITv05_reloadDStream(&(seqState->DStream));
+
+    /* MatchLength */
+    matchLength = FSEv05_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML) {
+        const U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) matchLength += add;
+        else if (dumps + 2 <= de) {
+            matchLength = MEM_readLE16(dumps);
+            dumps += 2;
+            if ((matchLength & 1) && dumps < de) {
+                matchLength += *dumps << 16;
+                dumps += 1;
+            }
+            matchLength >>= 1;
+        }
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+
+#if 0   /* debug */
+    {
+        static U64 totalDecoded = 0;
+        printf("pos %6u : %3u literals & match %3u bytes at distance %6u \n",
+           (U32)(totalDecoded), (U32)litLength, (U32)matchLength, (U32)offset);
+        totalDecoded += litLength + matchLength;
+    }
+#endif
+}
+
+
+static size_t ZSTDv05_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    static const int dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+    static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+    BYTE* const oLitEnd = op + sequence.litLength;
+    const size_t sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTDv05_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase))
+            return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {
+            size_t length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_8 || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+    }   }
+    /* Requirement: op <= oend_8 */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        const int sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTDv05_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTDv05_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_8) {
+            ZSTDv05_wildcopy(op, match, oend_8 - op);
+            match += oend_8 - op;
+            op = oend_8;
+        }
+        while (op < oMatchEnd)
+            *op++ = *match++;
+    } else {
+        ZSTDv05_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTDv05_decompressSequences(
+                               ZSTDv05_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength=0;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq=0;
+    const BYTE* dumps = NULL;
+    unsigned* DTableLL = dctx->LLTable;
+    unsigned* DTableML = dctx->MLTable;
+    unsigned* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTDv05_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, seqSize, dctx->flagStaticTables);
+    if (ZSTDv05_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        sequence.offset = REPCODE_STARTVALUE;
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = REPCODE_STARTVALUE;
+        errorCode = BITv05_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSEv05_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSEv05_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSEv05_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BITv05_reloadDStream(&(seqState.DStream)) <= BITv05_DStream_completed) && nbSeq ; ) {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTDv05_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTDv05_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+            if (ZSTDv05_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if (nbSeq) return ERROR(corruption_detected);
+    }
+
+    /* last literal segment */
+    {
+        size_t lastLLSize = litEnd - litPtr;
+        if (litPtr > litEnd) return ERROR(corruption_detected);   /* too many literals already used */
+        if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+        if (lastLLSize > 0) {
+            memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTDv05_checkContinuity(ZSTDv05_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTDv05_decompressBlock_internal(ZSTDv05_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+    size_t litCSize;
+
+    if (srcSize >= BLOCKSIZE) return ERROR(srcSize_wrong);
+
+    /* Decode literals sub-block */
+    litCSize = ZSTDv05_decodeLiteralsBlock(dctx, src, srcSize);
+    if (ZSTDv05_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTDv05_decompressSequences(dctx, dst, dstCapacity, ip, srcSize);
+}
+
+
+size_t ZSTDv05_decompressBlock(ZSTDv05_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    ZSTDv05_checkContinuity(dctx, dst);
+    return ZSTDv05_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+/*! ZSTDv05_decompress_continueDCtx
+*   dctx must have been properly initialized */
+static size_t ZSTDv05_decompress_continueDCtx(ZSTDv05_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                                 const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties;
+    memset(&blockProperties, 0, sizeof(blockProperties));
+
+    /* Frame Header */
+    {   size_t frameHeaderSize;
+        if (srcSize < ZSTDv05_frameHeaderSize_min+ZSTDv05_blockHeaderSize) return ERROR(srcSize_wrong);
+        frameHeaderSize = ZSTDv05_decodeFrameHeader_Part1(dctx, src, ZSTDv05_frameHeaderSize_min);
+        if (ZSTDv05_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTDv05_blockHeaderSize) return ERROR(srcSize_wrong);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+        frameHeaderSize = ZSTDv05_decodeFrameHeader_Part2(dctx, src, frameHeaderSize);
+        if (ZSTDv05_isError(frameHeaderSize)) return frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTDv05_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTDv05_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv05_blockHeaderSize;
+        remainingSize -= ZSTDv05_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTDv05_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTDv05_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTDv05_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+
+size_t ZSTDv05_decompress_usingPreparedDCtx(ZSTDv05_DCtx* dctx, const ZSTDv05_DCtx* refDCtx,
+                                         void* dst, size_t maxDstSize,
+                                   const void* src, size_t srcSize)
+{
+    ZSTDv05_copyDCtx(dctx, refDCtx);
+    ZSTDv05_checkContinuity(dctx, dst);
+    return ZSTDv05_decompress_continueDCtx(dctx, dst, maxDstSize, src, srcSize);
+}
+
+
+size_t ZSTDv05_decompress_usingDict(ZSTDv05_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    ZSTDv05_decompressBegin_usingDict(dctx, dict, dictSize);
+    ZSTDv05_checkContinuity(dctx, dst);
+    return ZSTDv05_decompress_continueDCtx(dctx, dst, maxDstSize, src, srcSize);
+}
+
+
+size_t ZSTDv05_decompressDCtx(ZSTDv05_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTDv05_decompress_usingDict(dctx, dst, maxDstSize, src, srcSize, NULL, 0);
+}
+
+size_t ZSTDv05_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+#if defined(ZSTDv05_HEAPMODE) && (ZSTDv05_HEAPMODE==1)
+    size_t regenSize;
+    ZSTDv05_DCtx* dctx = ZSTDv05_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv05_decompressDCtx(dctx, dst, maxDstSize, src, srcSize);
+    ZSTDv05_freeDCtx(dctx);
+    return regenSize;
+#else
+    ZSTDv05_DCtx dctx;
+    return ZSTDv05_decompressDCtx(&dctx, dst, maxDstSize, src, srcSize);
+#endif
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+static void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv05_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTDv05_frameHeaderSize_min) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+        return;
+    }
+    if (MEM_readLE32(src) != ZSTDv05_MAGICNUMBER) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+        return;
+    }
+    ip += ZSTDv05_frameHeaderSize_min; remainingSize -= ZSTDv05_frameHeaderSize_min;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTDv05_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTDv05_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTDv05_blockHeaderSize;
+        remainingSize -= ZSTDv05_blockHeaderSize;
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * BLOCKSIZE;
+}
+
+/* ******************************
+*  Streaming Decompression API
+********************************/
+size_t ZSTDv05_nextSrcSizeToDecompress(ZSTDv05_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+size_t ZSTDv05_decompressContinue(ZSTDv05_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);
+    ZSTDv05_checkContinuity(dctx, dst);
+
+    /* Decompress : frame header; part 1 */
+    switch (dctx->stage)
+    {
+    case ZSTDv05ds_getFrameHeaderSize :
+        /* get frame header size */
+        if (srcSize != ZSTDv05_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        dctx->headerSize = ZSTDv05_decodeFrameHeader_Part1(dctx, src, ZSTDv05_frameHeaderSize_min);
+        if (ZSTDv05_isError(dctx->headerSize)) return dctx->headerSize;
+        memcpy(dctx->headerBuffer, src, ZSTDv05_frameHeaderSize_min);
+        if (dctx->headerSize > ZSTDv05_frameHeaderSize_min) return ERROR(GENERIC); /* should never happen */
+        dctx->expected = 0;   /* not necessary to copy more */
+        /* fallthrough */
+    case ZSTDv05ds_decodeFrameHeader:
+        /* get frame header */
+        {   size_t const result = ZSTDv05_decodeFrameHeader_Part2(dctx, dctx->headerBuffer, dctx->headerSize);
+            if (ZSTDv05_isError(result)) return result;
+            dctx->expected = ZSTDv05_blockHeaderSize;
+            dctx->stage = ZSTDv05ds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDv05ds_decodeBlockHeader:
+        {
+            /* Decode block header */
+            blockProperties_t bp;
+            size_t blockSize = ZSTDv05_getcBlockSize(src, ZSTDv05_blockHeaderSize, &bp);
+            if (ZSTDv05_isError(blockSize)) return blockSize;
+            if (bp.blockType == bt_end) {
+                dctx->expected = 0;
+                dctx->stage = ZSTDv05ds_getFrameHeaderSize;
+            }
+            else {
+                dctx->expected = blockSize;
+                dctx->bType = bp.blockType;
+                dctx->stage = ZSTDv05ds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDv05ds_decompressBlock:
+        {
+            /* Decompress : block content */
+            size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTDv05_decompressBlock_internal(dctx, dst, maxDstSize, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTDv05_copyRawBlock(dst, maxDstSize, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);   /* impossible */
+            }
+            dctx->stage = ZSTDv05ds_decodeBlockHeader;
+            dctx->expected = ZSTDv05_blockHeaderSize;
+            if (ZSTDv05_isError(rSize)) return rSize;
+            dctx->previousDstEnd = (char*)dst + rSize;
+            return rSize;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static void ZSTDv05_refDictContent(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+    dctx->base = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+}
+
+static size_t ZSTDv05_loadEntropy(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t hSize, offcodeHeaderSize, matchlengthHeaderSize, errorCode, litlengthHeaderSize;
+    short offcodeNCount[MaxOff+1];
+    unsigned offcodeMaxValue=MaxOff, offcodeLog;
+    short matchlengthNCount[MaxML+1];
+    unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+    short litlengthNCount[MaxLL+1];
+    unsigned litlengthMaxValue = MaxLL, litlengthLog;
+
+    hSize = HUFv05_readDTableX4(dctx->hufTableX4, dict, dictSize);
+    if (HUFv05_isError(hSize)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + hSize;
+    dictSize -= hSize;
+
+    offcodeHeaderSize = FSEv05_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dict, dictSize);
+    if (FSEv05_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+    if (offcodeLog > OffFSEv05Log) return ERROR(dictionary_corrupted);
+    errorCode = FSEv05_buildDTable(dctx->OffTable, offcodeNCount, offcodeMaxValue, offcodeLog);
+    if (FSEv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + offcodeHeaderSize;
+    dictSize -= offcodeHeaderSize;
+
+    matchlengthHeaderSize = FSEv05_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dict, dictSize);
+    if (FSEv05_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+    if (matchlengthLog > MLFSEv05Log) return ERROR(dictionary_corrupted);
+    errorCode = FSEv05_buildDTable(dctx->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog);
+    if (FSEv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + matchlengthHeaderSize;
+    dictSize -= matchlengthHeaderSize;
+
+    litlengthHeaderSize = FSEv05_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dict, dictSize);
+    if (litlengthLog > LLFSEv05Log) return ERROR(dictionary_corrupted);
+    if (FSEv05_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+    errorCode = FSEv05_buildDTable(dctx->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog);
+    if (FSEv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+
+    dctx->flagStaticTables = 1;
+    return hSize + offcodeHeaderSize + matchlengthHeaderSize + litlengthHeaderSize;
+}
+
+static size_t ZSTDv05_decompress_insertDictionary(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t eSize;
+    U32 magic = MEM_readLE32(dict);
+    if (magic != ZSTDv05_DICT_MAGIC) {
+        /* pure content mode */
+        ZSTDv05_refDictContent(dctx, dict, dictSize);
+        return 0;
+    }
+    /* load entropy tables */
+    dict = (const char*)dict + 4;
+    dictSize -= 4;
+    eSize = ZSTDv05_loadEntropy(dctx, dict, dictSize);
+    if (ZSTDv05_isError(eSize)) return ERROR(dictionary_corrupted);
+
+    /* reference dictionary content */
+    dict = (const char*)dict + eSize;
+    dictSize -= eSize;
+    ZSTDv05_refDictContent(dctx, dict, dictSize);
+
+    return 0;
+}
+
+
+size_t ZSTDv05_decompressBegin_usingDict(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t errorCode;
+    errorCode = ZSTDv05_decompressBegin(dctx);
+    if (ZSTDv05_isError(errorCode)) return errorCode;
+
+    if (dict && dictSize) {
+        errorCode = ZSTDv05_decompress_insertDictionary(dctx, dict, dictSize);
+        if (ZSTDv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+    }
+
+    return 0;
+}
+
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+
+
+/* *************************************
+*  Constants
+***************************************/
+static size_t ZBUFFv05_blockHeaderSize = 3;
+
+
+
+/* *** Compression *** */
+
+static size_t ZBUFFv05_limitCopy(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    size_t length = MIN(maxDstSize, srcSize);
+    if (length > 0) {
+        memcpy(dst, src, length);
+    }
+    return length;
+}
+
+
+
+
+/** ************************************************
+*  Streaming decompression
+*
+*  A ZBUFFv05_DCtx object is required to track streaming operation.
+*  Use ZBUFFv05_createDCtx() and ZBUFFv05_freeDCtx() to create/release resources.
+*  Use ZBUFFv05_decompressInit() to start a new decompression operation.
+*  ZBUFFv05_DCtx objects can be reused multiple times.
+*
+*  Use ZBUFFv05_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *maxDstSizePtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *maxDstSizePtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to call again the function with remaining input.
+*  The content of dst will be overwritten (up to *maxDstSizePtr) at each function call, so save its content if it matters or change dst .
+*  return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
+*            or 0 when a frame is completely decoded
+*            or an error code, which can be tested using ZBUFFv05_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory)
+*  output : 128 KB block size is the internal unit, it ensures it's always possible to write a full block when it's decoded.
+*  input : just follow indications from ZBUFFv05_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* **************************************************/
+
+typedef enum { ZBUFFv05ds_init, ZBUFFv05ds_readHeader, ZBUFFv05ds_loadHeader, ZBUFFv05ds_decodeHeader,
+               ZBUFFv05ds_read, ZBUFFv05ds_load, ZBUFFv05ds_flush } ZBUFFv05_dStage;
+
+/* *** Resource management *** */
+
+#define ZSTDv05_frameHeaderSize_max 5   /* too magical, should come from reference */
+struct ZBUFFv05_DCtx_s {
+    ZSTDv05_DCtx* zc;
+    ZSTDv05_parameters params;
+    char* inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char* outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t hPos;
+    ZBUFFv05_dStage stage;
+    unsigned char headerBuffer[ZSTDv05_frameHeaderSize_max];
+};   /* typedef'd to ZBUFFv05_DCtx within "zstd_buffered.h" */
+
+
+ZBUFFv05_DCtx* ZBUFFv05_createDCtx(void)
+{
+    ZBUFFv05_DCtx* zbc = (ZBUFFv05_DCtx*)malloc(sizeof(ZBUFFv05_DCtx));
+    if (zbc==NULL) return NULL;
+    memset(zbc, 0, sizeof(*zbc));
+    zbc->zc = ZSTDv05_createDCtx();
+    zbc->stage = ZBUFFv05ds_init;
+    return zbc;
+}
+
+size_t ZBUFFv05_freeDCtx(ZBUFFv05_DCtx* zbc)
+{
+    if (zbc==NULL) return 0;   /* support free on null */
+    ZSTDv05_freeDCtx(zbc->zc);
+    free(zbc->inBuff);
+    free(zbc->outBuff);
+    free(zbc);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+size_t ZBUFFv05_decompressInitDictionary(ZBUFFv05_DCtx* zbc, const void* dict, size_t dictSize)
+{
+    zbc->stage = ZBUFFv05ds_readHeader;
+    zbc->hPos = zbc->inPos = zbc->outStart = zbc->outEnd = 0;
+    return ZSTDv05_decompressBegin_usingDict(zbc->zc, dict, dictSize);
+}
+
+size_t ZBUFFv05_decompressInit(ZBUFFv05_DCtx* zbc)
+{
+    return ZBUFFv05_decompressInitDictionary(zbc, NULL, 0);
+}
+
+
+/* *** Decompression *** */
+
+size_t ZBUFFv05_decompressContinue(ZBUFFv05_DCtx* zbc, void* dst, size_t* maxDstSizePtr, const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* ip = istart;
+    const char* const iend = istart + *srcSizePtr;
+    char* const ostart = (char*)dst;
+    char* op = ostart;
+    char* const oend = ostart + *maxDstSizePtr;
+    U32 notDone = 1;
+
+    while (notDone) {
+        switch(zbc->stage)
+        {
+        case ZBUFFv05ds_init :
+            return ERROR(init_missing);
+
+        case ZBUFFv05ds_readHeader :
+            /* read header from src */
+            {
+                size_t headerSize = ZSTDv05_getFrameParams(&(zbc->params), src, *srcSizePtr);
+                if (ZSTDv05_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    memcpy(zbc->headerBuffer+zbc->hPos, src, *srcSizePtr);
+                    zbc->hPos += *srcSizePtr;
+                    *maxDstSizePtr = 0;
+                    zbc->stage = ZBUFFv05ds_loadHeader;
+                    return headerSize - zbc->hPos;
+                }
+                zbc->stage = ZBUFFv05ds_decodeHeader;
+                break;
+            }
+	    /* fall-through */
+        case ZBUFFv05ds_loadHeader:
+            /* complete header from src */
+            {
+                size_t headerSize = ZBUFFv05_limitCopy(
+                    zbc->headerBuffer + zbc->hPos, ZSTDv05_frameHeaderSize_max - zbc->hPos,
+                    src, *srcSizePtr);
+                zbc->hPos += headerSize;
+                ip += headerSize;
+                headerSize = ZSTDv05_getFrameParams(&(zbc->params), zbc->headerBuffer, zbc->hPos);
+                if (ZSTDv05_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    *maxDstSizePtr = 0;
+                    return headerSize - zbc->hPos;
+                }
+                /* zbc->stage = ZBUFFv05ds_decodeHeader; break; */   /* useless : stage follows */
+            }
+	    /* fall-through */
+        case ZBUFFv05ds_decodeHeader:
+                /* apply header to create / resize buffers */
+                {
+                    size_t neededOutSize = (size_t)1 << zbc->params.windowLog;
+                    size_t neededInSize = BLOCKSIZE;   /* a block is never > BLOCKSIZE */
+                    if (zbc->inBuffSize < neededInSize) {
+                        free(zbc->inBuff);
+                        zbc->inBuffSize = neededInSize;
+                        zbc->inBuff = (char*)malloc(neededInSize);
+                        if (zbc->inBuff == NULL) return ERROR(memory_allocation);
+                    }
+                    if (zbc->outBuffSize < neededOutSize) {
+                        free(zbc->outBuff);
+                        zbc->outBuffSize = neededOutSize;
+                        zbc->outBuff = (char*)malloc(neededOutSize);
+                        if (zbc->outBuff == NULL) return ERROR(memory_allocation);
+                }   }
+                if (zbc->hPos) {
+                    /* some data already loaded into headerBuffer : transfer into inBuff */
+                    memcpy(zbc->inBuff, zbc->headerBuffer, zbc->hPos);
+                    zbc->inPos = zbc->hPos;
+                    zbc->hPos = 0;
+                    zbc->stage = ZBUFFv05ds_load;
+                    break;
+                }
+                zbc->stage = ZBUFFv05ds_read;
+		/* fall-through */
+        case ZBUFFv05ds_read:
+            {
+                size_t neededInSize = ZSTDv05_nextSrcSizeToDecompress(zbc->zc);
+                if (neededInSize==0) {  /* end of frame */
+                    zbc->stage = ZBUFFv05ds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {
+                    /* directly decode from src */
+                    size_t decodedSize = ZSTDv05_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        ip, neededInSize);
+                    if (ZSTDv05_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize) break;   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFv05ds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbc->stage = ZBUFFv05ds_load;
+            }
+	    /* fall-through */
+        case ZBUFFv05ds_load:
+            {
+                size_t neededInSize = ZSTDv05_nextSrcSizeToDecompress(zbc->zc);
+                size_t toLoad = neededInSize - zbc->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbc->inBuffSize - zbc->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFFv05_limitCopy(zbc->inBuff + zbc->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbc->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+                {
+                    size_t decodedSize = ZSTDv05_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        zbc->inBuff, neededInSize);
+                    if (ZSTDv05_isError(decodedSize)) return decodedSize;
+                    zbc->inPos = 0;   /* input is consumed */
+                    if (!decodedSize) { zbc->stage = ZBUFFv05ds_read; break; }   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFv05ds_flush;
+                    /* break; */  /* ZBUFFv05ds_flush follows */
+                }
+	    }
+	    /* fall-through */
+        case ZBUFFv05ds_flush:
+            {
+                size_t toFlushSize = zbc->outEnd - zbc->outStart;
+                size_t flushedSize = ZBUFFv05_limitCopy(op, oend-op, zbc->outBuff + zbc->outStart, toFlushSize);
+                op += flushedSize;
+                zbc->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {
+                    zbc->stage = ZBUFFv05ds_read;
+                    if (zbc->outStart + BLOCKSIZE > zbc->outBuffSize)
+                        zbc->outStart = zbc->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+    }   }
+
+    *srcSizePtr = ip-istart;
+    *maxDstSizePtr = op-ostart;
+
+    {   size_t nextSrcSizeHint = ZSTDv05_nextSrcSizeToDecompress(zbc->zc);
+        if (nextSrcSizeHint > ZBUFFv05_blockHeaderSize) nextSrcSizeHint+= ZBUFFv05_blockHeaderSize;   /* get next block header too */
+        nextSrcSizeHint -= zbc->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+unsigned ZBUFFv05_isError(size_t errorCode) { return ERR_isError(errorCode); }
+const char* ZBUFFv05_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+size_t ZBUFFv05_recommendedDInSize(void)  { return BLOCKSIZE + ZBUFFv05_blockHeaderSize /* block header size*/ ; }
+size_t ZBUFFv05_recommendedDOutSize(void) { return BLOCKSIZE; }
diff --git a/lib/legacy/zstd_v06.c b/lib/legacy/zstd_v06.c
new file mode 100644
index 0000000..88a39e2
--- /dev/null
+++ b/lib/legacy/zstd_v06.c
@@ -0,0 +1,4110 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*- Dependencies -*/
+#include "zstd_v06.h"
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+#include <stdlib.h>    /* malloc, free, qsort */
+#include "../common/compiler.h"
+#include "../common/error_private.h"
+
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*-****************************************
+*  Compiler specifics
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+
+
+/*-**************************************************************
+*  Basic Types
+*****************************************************************/
+#if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-**************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U32 MEM_swap32(U32 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_ulong(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap32(in);
+#else
+    return  ((in << 24) & 0xff000000 ) |
+            ((in <<  8) & 0x00ff0000 ) |
+            ((in >>  8) & 0x0000ff00 ) |
+            ((in >> 24) & 0x000000ff );
+#endif
+}
+
+MEM_STATIC U64 MEM_swap64(U64 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_uint64(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap64(in);
+#else
+    return  ((in << 56) & 0xff00000000000000ULL) |
+            ((in << 40) & 0x00ff000000000000ULL) |
+            ((in << 24) & 0x0000ff0000000000ULL) |
+            ((in << 8)  & 0x000000ff00000000ULL) |
+            ((in >> 8)  & 0x00000000ff000000ULL) |
+            ((in >> 24) & 0x0000000000ff0000ULL) |
+            ((in >> 40) & 0x000000000000ff00ULL) |
+            ((in >> 56) & 0x00000000000000ffULL);
+#endif
+}
+
+
+/*=== Little endian r/w ===*/
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian()) {
+        MEM_write16(memPtr, val);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+        return MEM_swap32(MEM_read32(memPtr));
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+        return MEM_swap64(MEM_read64(memPtr));
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd
+*/
+#ifndef ZSTDv06_STATIC_H
+#define ZSTDv06_STATIC_H
+
+/* The prototypes defined within this file are considered experimental.
+ * They should not be used in the context DLL as they may change in the future.
+ * Prefer static linking if you need them, to control breaking version changes issues.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*- Advanced Decompression functions -*/
+
+/*! ZSTDv06_decompress_usingPreparedDCtx() :
+*   Same as ZSTDv06_decompress_usingDict, but using a reference context `preparedDCtx`, where dictionary has been loaded.
+*   It avoids reloading the dictionary each time.
+*   `preparedDCtx` must have been properly initialized using ZSTDv06_decompressBegin_usingDict().
+*   Requires 2 contexts : 1 for reference (preparedDCtx), which will not be modified, and 1 to run the decompression operation (dctx) */
+ZSTDLIBv06_API size_t ZSTDv06_decompress_usingPreparedDCtx(
+                                           ZSTDv06_DCtx* dctx, const ZSTDv06_DCtx* preparedDCtx,
+                                           void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize);
+
+
+
+#define ZSTDv06_FRAMEHEADERSIZE_MAX 13    /* for static allocation */
+static const size_t ZSTDv06_frameHeaderSize_min = 5;
+static const size_t ZSTDv06_frameHeaderSize_max = ZSTDv06_FRAMEHEADERSIZE_MAX;
+
+ZSTDLIBv06_API size_t ZSTDv06_decompressBegin(ZSTDv06_DCtx* dctx);
+
+/*
+  Streaming decompression, direct mode (bufferless)
+
+  A ZSTDv06_DCtx object is required to track streaming operations.
+  Use ZSTDv06_createDCtx() / ZSTDv06_freeDCtx() to manage it.
+  A ZSTDv06_DCtx object can be re-used multiple times.
+
+  First optional operation is to retrieve frame parameters, using ZSTDv06_getFrameParams(), which doesn't consume the input.
+  It can provide the minimum size of rolling buffer required to properly decompress data,
+  and optionally the final size of uncompressed content.
+  (Note : content size is an optional info that may not be present. 0 means : content size unknown)
+  Frame parameters are extracted from the beginning of compressed frame.
+  The amount of data to read is variable, from ZSTDv06_frameHeaderSize_min to ZSTDv06_frameHeaderSize_max (so if `srcSize` >= ZSTDv06_frameHeaderSize_max, it will always work)
+  If `srcSize` is too small for operation to succeed, function will return the minimum size it requires to produce a result.
+  Result : 0 when successful, it means the ZSTDv06_frameParams structure has been filled.
+          >0 : means there is not enough data into `src`. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTDv06_isError()
+
+  Start decompression, with ZSTDv06_decompressBegin() or ZSTDv06_decompressBegin_usingDict().
+  Alternatively, you can copy a prepared context, using ZSTDv06_copyDCtx().
+
+  Then use ZSTDv06_nextSrcSizeToDecompress() and ZSTDv06_decompressContinue() alternatively.
+  ZSTDv06_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTDv06_decompressContinue().
+  ZSTDv06_decompressContinue() requires this exact amount of bytes, or it will fail.
+  ZSTDv06_decompressContinue() needs previous data blocks during decompression, up to (1 << windowlog).
+  They should preferably be located contiguously, prior to current block. Alternatively, a round buffer is also possible.
+
+  @result of ZSTDv06_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity)
+  It can be zero, which is not an error; it just means ZSTDv06_decompressContinue() has decoded some header.
+
+  A frame is fully decoded when ZSTDv06_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+*/
+
+
+/* **************************************
+*  Block functions
+****************************************/
+/*! Block functions produce and decode raw zstd blocks, without frame metadata.
+    User will have to take in charge required information to regenerate data, such as compressed and content sizes.
+
+    A few rules to respect :
+    - Uncompressed block size must be <= ZSTDv06_BLOCKSIZE_MAX (128 KB)
+    - Compressing or decompressing requires a context structure
+      + Use ZSTDv06_createCCtx() and ZSTDv06_createDCtx()
+    - It is necessary to init context before starting
+      + compression : ZSTDv06_compressBegin()
+      + decompression : ZSTDv06_decompressBegin()
+      + variants _usingDict() are also allowed
+      + copyCCtx() and copyDCtx() work too
+    - When a block is considered not compressible enough, ZSTDv06_compressBlock() result will be zero.
+      In which case, nothing is produced into `dst`.
+      + User must test for such outcome and deal directly with uncompressed data
+      + ZSTDv06_decompressBlock() doesn't accept uncompressed data as input !!
+*/
+
+#define ZSTDv06_BLOCKSIZE_MAX (128 * 1024)   /* define, for static allocation */
+ZSTDLIBv06_API size_t ZSTDv06_decompressBlock(ZSTDv06_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv06_STATIC_H */
+/*
+    zstd_internal - common functions to include
+    Header File for include
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://www.zstd.net
+*/
+#ifndef ZSTDv06_CCOMMON_H_MODULE
+#define ZSTDv06_CCOMMON_H_MODULE
+
+
+/*-*************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/*-*************************************
+*  Common constants
+***************************************/
+#define ZSTDv06_DICT_MAGIC  0xEC30A436
+
+#define ZSTDv06_REP_NUM    3
+#define ZSTDv06_REP_INIT   ZSTDv06_REP_NUM
+#define ZSTDv06_REP_MOVE   (ZSTDv06_REP_NUM-1)
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define ZSTDv06_WINDOWLOG_ABSOLUTEMIN 12
+static const size_t ZSTDv06_fcs_fieldSize[4] = { 0, 1, 2, 8 };
+
+#define ZSTDv06_BLOCKHEADERSIZE 3   /* because C standard does not allow a static const value to be defined using another static const value .... :( */
+static const size_t ZSTDv06_blockHeaderSize = ZSTDv06_BLOCKHEADERSIZE;
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
+
+#define IS_HUF 0
+#define IS_PCH 1
+#define IS_RAW 2
+#define IS_RLE 3
+
+#define LONGNBSEQ 0x7F00
+
+#define MINMATCH 3
+#define EQUAL_READ32 4
+#define REPCODE_STARTVALUE 1
+
+#define Litbits  8
+#define MaxLit ((1<<Litbits) - 1)
+#define MaxML  52
+#define MaxLL  35
+#define MaxOff 28
+#define MaxSeq MAX(MaxLL, MaxML)   /* Assumption : MaxOff < MaxLL,MaxML */
+#define MLFSELog    9
+#define LLFSELog    9
+#define OffFSELog   8
+
+#define FSEv06_ENCODING_RAW     0
+#define FSEv06_ENCODING_RLE     1
+#define FSEv06_ENCODING_STATIC  2
+#define FSEv06_ENCODING_DYNAMIC 3
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9,10,11,12,
+                                     13,14,15,16 };
+static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1,
+                                             2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1,
+                                            -1,-1,-1,-1 };
+static const U32 LL_defaultNormLog = 6;
+
+static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 7, 8, 9,10,11,
+                                     12,13,14,15,16 };
+static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,-1,-1,
+                                            -1,-1,-1,-1,-1 };
+static const U32 ML_defaultNormLog = 6;
+
+static const S16 OF_defaultNorm[MaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                              1, 1, 1, 1, 1, 1, 1, 1,-1,-1,-1,-1,-1 };
+static const U32 OF_defaultNormLog = 5;
+
+
+/*-*******************************************
+*  Shared functions to include for inlining
+*********************************************/
+static void ZSTDv06_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+#define COPY8(d,s) { ZSTDv06_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTDv06_wildcopy() :
+*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
+#define WILDCOPY_OVERLENGTH 8
+MEM_STATIC void ZSTDv06_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+
+/*-*******************************************
+*  Private interfaces
+*********************************************/
+typedef struct {
+    U32 off;
+    U32 len;
+} ZSTDv06_match_t;
+
+typedef struct {
+    U32 price;
+    U32 off;
+    U32 mlen;
+    U32 litlen;
+    U32 rep[ZSTDv06_REP_INIT];
+} ZSTDv06_optimal_t;
+
+typedef struct { U32  unused; } ZSTDv06_stats_t;
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* litStart;
+    BYTE* lit;
+    U16*  litLengthStart;
+    U16*  litLength;
+    BYTE* llCodeStart;
+    U16*  matchLengthStart;
+    U16*  matchLength;
+    BYTE* mlCodeStart;
+    U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
+    U32   longLengthPos;
+    /* opt */
+    ZSTDv06_optimal_t* priceTable;
+    ZSTDv06_match_t* matchTable;
+    U32* matchLengthFreq;
+    U32* litLengthFreq;
+    U32* litFreq;
+    U32* offCodeFreq;
+    U32  matchLengthSum;
+    U32  matchSum;
+    U32  litLengthSum;
+    U32  litSum;
+    U32  offCodeSum;
+    U32  log2matchLengthSum;
+    U32  log2matchSum;
+    U32  log2litLengthSum;
+    U32  log2litSum;
+    U32  log2offCodeSum;
+    U32  factor;
+    U32  cachedPrice;
+    U32  cachedLitLength;
+    const BYTE* cachedLiterals;
+    ZSTDv06_stats_t stats;
+} SeqStore_t;
+
+void ZSTDv06_seqToCodes(const SeqStore_t* seqStorePtr, size_t const nbSeq);
+
+
+#endif   /* ZSTDv06_CCOMMON_H_MODULE */
+/* ******************************************************************
+   FSE : Finite State Entropy codec
+   Public Prototypes declaration
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef FSEv06_H
+#define FSEv06_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*-****************************************
+*  FSE simple functions
+******************************************/
+/*! FSEv06_decompress():
+    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstCapacity'.
+    @return : size of regenerated data (<= maxDstSize),
+              or an error code, which can be tested using FSEv06_isError() .
+
+    ** Important ** : FSEv06_decompress() does not decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+size_t FSEv06_decompress(void* dst,  size_t dstCapacity,
+                const void* cSrc, size_t cSrcSize);
+
+
+/*-*****************************************
+*  Tool functions
+******************************************/
+size_t FSEv06_compressBound(size_t size);       /* maximum compressed size */
+
+/* Error Management */
+unsigned    FSEv06_isError(size_t code);        /* tells if a return value is an error code */
+const char* FSEv06_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+
+/*-*****************************************
+*  FSE detailed API
+******************************************/
+/*!
+
+FSEv06_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*! FSEv06_readNCount():
+    Read compactly saved 'normalizedCounter' from 'rBuffer'.
+    @return : size read from 'rBuffer',
+              or an errorCode, which can be tested using FSEv06_isError().
+              maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+size_t FSEv06_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*! Constructor and Destructor of FSEv06_DTable.
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSEv06_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+FSEv06_DTable* FSEv06_createDTable(unsigned tableLog);
+void        FSEv06_freeDTable(FSEv06_DTable* dt);
+
+/*! FSEv06_buildDTable():
+    Builds 'dt', which must be already allocated, using FSEv06_createDTable().
+    return : 0, or an errorCode, which can be tested using FSEv06_isError() */
+size_t FSEv06_buildDTable (FSEv06_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSEv06_decompress_usingDTable():
+    Decompress compressed source `cSrc` of size `cSrcSize` using `dt`
+    into `dst` which must be already allocated.
+    @return : size of regenerated data (necessarily <= `dstCapacity`),
+              or an errorCode, which can be tested using FSEv06_isError() */
+size_t FSEv06_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSEv06_DTable* dt);
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSEv06_readNCount() if it was saved using FSEv06_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSEv06_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSEv06_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSEv06_isError().
+
+The next step is to build the decompression tables 'FSEv06_DTable' from 'normalizedCounter'.
+This is performed by the function FSEv06_buildDTable().
+The space required by 'FSEv06_DTable' must be already allocated using FSEv06_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSEv06_isError().
+
+`FSEv06_DTable` can then be used to decompress `cSrc`, with FSEv06_decompress_usingDTable().
+`cSrcSize` must be strictly correct, otherwise decompression will fail.
+FSEv06_decompress_usingDTable() result will tell how many bytes were regenerated (<=`dstCapacity`).
+If there is an error, the function will return an error code, which can be tested using FSEv06_isError(). (ex: dst buffer too small)
+*/
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv06_H */
+/* ******************************************************************
+   bitstream
+   Part of FSE library
+   header file (to include)
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which must be inlined for best performance.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/*=========================================
+*  Target specific
+=========================================*/
+#if defined(__BMI__) && defined(__GNUC__)
+#  include <immintrin.h>   /* support for bextr (experimental) */
+#endif
+
+
+
+/*-********************************************
+*  bitStream decoding API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BITv06_DStream_t;
+
+typedef enum { BITv06_DStream_unfinished = 0,
+               BITv06_DStream_endOfBuffer = 1,
+               BITv06_DStream_completed = 2,
+               BITv06_DStream_overflow = 3 } BITv06_DStream_status;  /* result of BITv06_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BITv06_initDStream(BITv06_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BITv06_readBits(BITv06_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BITv06_DStream_status BITv06_reloadDStream(BITv06_DStream_t* bitD);
+MEM_STATIC unsigned BITv06_endOfDStream(const BITv06_DStream_t* bitD);
+
+
+
+/*-****************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BITv06_readBitsFast(BITv06_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/*-**************************************************************
+*  Internal functions
+****************************************************************/
+MEM_STATIC unsigned BITv06_highbit32 ( U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/*-********************************************************
+* bitStream decoding
+**********************************************************/
+/*! BITv06_initDStream() :
+*   Initialize a BITv06_DStream_t.
+*   `bitD` : a pointer to an already allocated BITv06_DStream_t structure.
+*   `srcSize` must be the *exact* size of the bitStream, in bytes.
+*   @return : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BITv06_initDStream(BITv06_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(bitD->bitContainer)) {  /* normal case */
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          if (lastByte == 0) return ERROR(GENERIC);   /* endMark not present */
+          bitD->bitsConsumed = 8 - BITv06_highbit32(lastByte); }
+    } else {
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          if (lastByte == 0) return ERROR(GENERIC);   /* endMark not present */
+          bitD->bitsConsumed = 8 - BITv06_highbit32(lastByte); }
+        bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+
+ MEM_STATIC size_t BITv06_lookBits(const BITv06_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BITv06_lookBitsFast() :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv06_lookBitsFast(const BITv06_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BITv06_skipBits(BITv06_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BITv06_readBits(BITv06_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv06_lookBits(bitD, nbBits);
+    BITv06_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*! BITv06_readBitsFast() :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv06_readBitsFast(BITv06_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv06_lookBitsFast(bitD, nbBits);
+    BITv06_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BITv06_DStream_status BITv06_reloadDStream(BITv06_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BITv06_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer)) {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BITv06_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start) {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BITv06_DStream_endOfBuffer;
+        return BITv06_DStream_completed;
+    }
+    {   U32 nbBytes = bitD->bitsConsumed >> 3;
+        BITv06_DStream_status result = BITv06_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start) {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BITv06_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BITv06_endOfDStream() :
+*   @return Tells if DStream has exactly reached its end (all bits consumed).
+*/
+MEM_STATIC unsigned BITv06_endOfDStream(const BITv06_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSEv06_STATIC_H
+#define FSEv06_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSEv06_NCOUNTBOUND 512
+#define FSEv06_BLOCKBOUND(size) (size + (size>>7))
+#define FSEv06_COMPRESSBOUND(size) (FSEv06_NCOUNTBOUND + FSEv06_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of unsigned using below macros */
+#define FSEv06_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSE advanced API
+*******************************************/
+size_t FSEv06_countFast(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+/* same as FSEv06_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr  */
+
+size_t FSEv06_buildDTable_raw (FSEv06_DTable* dt, unsigned nbBits);
+/* build a fake FSEv06_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+size_t FSEv06_buildDTable_rle (FSEv06_DTable* dt, unsigned char symbolValue);
+/* build a fake FSEv06_DTable, designed to always generate the same symbolValue */
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSEv06_DState_t;
+
+
+static void     FSEv06_initDState(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD, const FSEv06_DTable* dt);
+
+static unsigned char FSEv06_decodeSymbol(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD);
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSEv06_decodeSymbolFast(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+
+
+/* ======    Decompression    ====== */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSEv06_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSEv06_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSEv06_initDState(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD, const FSEv06_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv06_DTableHeader* const DTableH = (const FSEv06_DTableHeader*)ptr;
+    DStatePtr->state = BITv06_readBits(bitD, DTableH->tableLog);
+    BITv06_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSEv06_peekSymbol(const FSEv06_DState_t* DStatePtr)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC void FSEv06_updateState(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BITv06_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.newState + lowBits;
+}
+
+MEM_STATIC BYTE FSEv06_decodeSymbol(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv06_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+/*! FSEv06_decodeSymbolFast() :
+    unsafe, only works if no symbol has a probability > 50% */
+MEM_STATIC BYTE FSEv06_decodeSymbolFast(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv06_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+
+
+#ifndef FSEv06_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSEv06_MAX_MEMORY_USAGE 14
+#define FSEv06_DEFAULT_MEMORY_USAGE 13
+
+/*!FSEv06_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSEv06_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSEv06_FUNCTION_TYPE BYTE
+#define FSEv06_FUNCTION_EXTENSION
+#define FSEv06_DECODE_TYPE FSEv06_decode_t
+
+
+#endif   /* !FSEv06_COMMONDEFS_ONLY */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSEv06_MAX_TABLELOG  (FSEv06_MAX_MEMORY_USAGE-2)
+#define FSEv06_MAX_TABLESIZE (1U<<FSEv06_MAX_TABLELOG)
+#define FSEv06_MAXTABLESIZE_MASK (FSEv06_MAX_TABLESIZE-1)
+#define FSEv06_DEFAULT_TABLELOG (FSEv06_DEFAULT_MEMORY_USAGE-2)
+#define FSEv06_MIN_TABLELOG 5
+
+#define FSEv06_TABLELOG_ABSOLUTE_MAX 15
+#if FSEv06_MAX_TABLELOG > FSEv06_TABLELOG_ABSOLUTE_MAX
+#error "FSEv06_MAX_TABLELOG > FSEv06_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+#define FSEv06_TABLESTEP(tableSize) ((tableSize>>1) + (tableSize>>3) + 3)
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv06_STATIC_H */
+/*
+   Common functions of New Generation Entropy library
+   Copyright (C) 2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+*************************************************************************** */
+
+
+/*-****************************************
+*  FSE Error Management
+******************************************/
+unsigned FSEv06_isError(size_t code) { return ERR_isError(code); }
+
+const char* FSEv06_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* **************************************************************
+*  HUF Error Management
+****************************************************************/
+static unsigned HUFv06_isError(size_t code) { return ERR_isError(code); }
+
+
+/*-**************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSEv06_abs(short a) { return a<0 ? -a : a; }
+
+size_t FSEv06_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSEv06_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSEv06_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr)) {
+        if (previous0) {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF) {
+                n0+=24;
+                if (ip < iend-5) {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                } else {
+                    bitStream >>= 16;
+                    bitCount+=16;
+            }   }
+            while ((bitStream & 3) == 3) {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {   short const max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max) {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            } else {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSEv06_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold) {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+    }   }   /* while ((remaining>1) && (charnum<=*maxSVPtr)) */
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+/* ******************************************************************
+   FSE : Finite State Entropy decoder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSEv06_isError ERR_isError
+#define FSEv06_STATIC_ASSERT(c) { enum { FSEv06_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSEv06_DTABLE_SIZE_U32(FSEv06_MAX_TABLELOG)];
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSEv06_FUNCTION_EXTENSION
+#  error "FSEv06_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSEv06_FUNCTION_TYPE
+#  error "FSEv06_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSEv06_CAT(X,Y) X##Y
+#define FSEv06_FUNCTION_NAME(X,Y) FSEv06_CAT(X,Y)
+#define FSEv06_TYPE_NAME(X,Y) FSEv06_CAT(X,Y)
+
+
+/* Function templates */
+FSEv06_DTable* FSEv06_createDTable (unsigned tableLog)
+{
+    if (tableLog > FSEv06_TABLELOG_ABSOLUTE_MAX) tableLog = FSEv06_TABLELOG_ABSOLUTE_MAX;
+    return (FSEv06_DTable*)malloc( FSEv06_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
+}
+
+void FSEv06_freeDTable (FSEv06_DTable* dt)
+{
+    free(dt);
+}
+
+size_t FSEv06_buildDTable(FSEv06_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
+    FSEv06_DECODE_TYPE* const tableDecode = (FSEv06_DECODE_TYPE*) (tdPtr);
+    U16 symbolNext[FSEv06_MAX_SYMBOL_VALUE+1];
+
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+    U32 highThreshold = tableSize-1;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSEv06_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSEv06_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    {   FSEv06_DTableHeader DTableH;
+        DTableH.tableLog = (U16)tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].symbol = (FSEv06_FUNCTION_TYPE)s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    symbolNext[s] = normalizedCounter[s];
+        }   }   }
+        memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    {   U32 const tableMask = tableSize-1;
+        U32 const step = FSEv06_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            for (i=0; i<normalizedCounter[s]; i++) {
+                tableDecode[position].symbol = (FSEv06_FUNCTION_TYPE)s;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+
+        if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {   U32 u;
+        for (u=0; u<tableSize; u++) {
+            FSEv06_FUNCTION_TYPE const symbol = (FSEv06_FUNCTION_TYPE)(tableDecode[u].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BITv06_highbit32 ((U32)nextState) );
+            tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+    }   }
+
+    return 0;
+}
+
+
+
+#ifndef FSEv06_COMMONDEFS_ONLY
+
+/*-*******************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+size_t FSEv06_buildDTable_rle (FSEv06_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSEv06_DTableHeader* const DTableH = (FSEv06_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv06_decode_t* const cell = (FSEv06_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+size_t FSEv06_buildDTable_raw (FSEv06_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSEv06_DTableHeader* const DTableH = (FSEv06_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv06_decode_t* const dinfo = (FSEv06_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSV1 = tableMask+1;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<maxSV1; s++) {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSEv06_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSEv06_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BITv06_DStream_t bitD;
+    FSEv06_DState_t state1;
+    FSEv06_DState_t state2;
+
+    /* Init */
+    { size_t const errorCode = BITv06_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+      if (FSEv06_isError(errorCode)) return errorCode; }
+
+    FSEv06_initDState(&state1, &bitD, dt);
+    FSEv06_initDState(&state2, &bitD, dt);
+
+#define FSEv06_GETSYMBOL(statePtr) fast ? FSEv06_decodeSymbolFast(statePtr, &bitD) : FSEv06_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BITv06_reloadDStream(&bitD)==BITv06_DStream_unfinished) && (op<olimit) ; op+=4) {
+        op[0] = FSEv06_GETSYMBOL(&state1);
+
+        if (FSEv06_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv06_reloadDStream(&bitD);
+
+        op[1] = FSEv06_GETSYMBOL(&state2);
+
+        if (FSEv06_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BITv06_reloadDStream(&bitD) > BITv06_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSEv06_GETSYMBOL(&state1);
+
+        if (FSEv06_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv06_reloadDStream(&bitD);
+
+        op[3] = FSEv06_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BITv06_reloadDStream(&bitD) >= FSEv06_DStream_partiallyFilled; Ends at exactly BITv06_DStream_completed */
+    while (1) {
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv06_GETSYMBOL(&state1);
+
+        if (BITv06_reloadDStream(&bitD)==BITv06_DStream_overflow) {
+            *op++ = FSEv06_GETSYMBOL(&state2);
+            break;
+        }
+
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv06_GETSYMBOL(&state2);
+
+        if (BITv06_reloadDStream(&bitD)==BITv06_DStream_overflow) {
+            *op++ = FSEv06_GETSYMBOL(&state1);
+            break;
+    }   }
+
+    return op-ostart;
+}
+
+
+size_t FSEv06_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSEv06_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv06_DTableHeader* DTableH = (const FSEv06_DTableHeader*)ptr;
+    const U32 fastMode = DTableH->fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSEv06_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSEv06_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+size_t FSEv06_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSEv06_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSEv06_MAX_SYMBOL_VALUE;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    {   size_t const NCountLength = FSEv06_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+        if (FSEv06_isError(NCountLength)) return NCountLength;
+        if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+        ip += NCountLength;
+        cSrcSize -= NCountLength;
+    }
+
+    { size_t const errorCode = FSEv06_buildDTable (dt, counting, maxSymbolValue, tableLog);
+      if (FSEv06_isError(errorCode)) return errorCode; }
+
+    return FSEv06_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);   /* always return, even if it is an error code */
+}
+
+
+
+#endif   /* FSEv06_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFv06_H
+#define HUFv06_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* ****************************************
+*  HUF simple functions
+******************************************/
+size_t HUFv06_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+/*
+HUFv06_decompress() :
+    Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstSize'.
+    `dstSize` : must be the **exact** size of original (uncompressed) data.
+    Note : in contrast with FSE, HUFv06_decompress can regenerate
+           RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+           because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize)
+              or an error code, which can be tested using HUFv06_isError()
+*/
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+size_t HUFv06_compressBound(size_t size);       /**< maximum compressed size */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUFv06_H */
+/* ******************************************************************
+   Huffman codec, part of New Generation Entropy library
+   header file, for static linking only
+   Copyright (C) 2013-2016, Yann Collet
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFv06_STATIC_H
+#define HUFv06_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* ****************************************
+*  Static allocation
+******************************************/
+/* HUF buffer bounds */
+#define HUFv06_CTABLEBOUND 129
+#define HUFv06_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+#define HUFv06_COMPRESSBOUND(size) (HUFv06_CTABLEBOUND + HUFv06_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* static allocation of HUF's DTable */
+#define HUFv06_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))
+#define HUFv06_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUFv06_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv06_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUFv06_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv06_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUFv06_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+size_t HUFv06_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv06_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+
+/*!
+HUFv06_decompress() does the following:
+1. select the decompression algorithm (X2, X4, X6) based on pre-computed heuristics
+2. build Huffman table from save, using HUFv06_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUFv06_decompressSXn_usingDTable
+*/
+size_t HUFv06_readDTableX2 (unsigned short* DTable, const void* src, size_t srcSize);
+size_t HUFv06_readDTableX4 (unsigned* DTable, const void* src, size_t srcSize);
+
+size_t HUFv06_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv06_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+/* single stream variants */
+size_t HUFv06_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv06_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+
+size_t HUFv06_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv06_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+#define HUFv06_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUFv06_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUFv06_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUFv06_ABSOLUTEMAX_TABLELOG */
+#define HUFv06_DEFAULT_TABLELOG  HUFv06_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUFv06_MAX_SYMBOL_VALUE 255
+#if (HUFv06_MAX_TABLELOG > HUFv06_ABSOLUTEMAX_TABLELOG)
+#  error "HUFv06_MAX_TABLELOG is too large !"
+#endif
+
+
+
+/*! HUFv06_readStats() :
+    Read compact Huffman tree, saved by HUFv06_writeCTable().
+    `huffWeight` is destination buffer.
+    @return : size read from `src`
+*/
+MEM_STATIC size_t HUFv06_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    /* memset(huffWeight, 0, hwSize); */   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  { /* special header */
+        if (iSize >= (242)) {  /* RLE */
+            static U32 l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else {   /* Incompressible */
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            {   U32 n;
+                for (n=0; n<oSize; n+=2) {
+                    huffWeight[n]   = ip[n/2] >> 4;
+                    huffWeight[n+1] = ip[n/2] & 15;
+    }   }   }   }
+    else  {   /* header compressed with FSE (normal case) */
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSEv06_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSEv06_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUFv06_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    {   U32 n; for (n=0; n<oSize; n++) {
+            if (huffWeight[n] >= HUFv06_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+            rankStats[huffWeight[n]]++;
+            weightTotal += (1 << huffWeight[n]) >> 1;
+    }   }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    {   U32 const tableLog = BITv06_highbit32(weightTotal) + 1;
+        if (tableLog > HUFv06_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        *tableLogPtr = tableLog;
+        /* determine last weight */
+        {   U32 const total = 1 << tableLog;
+            U32 const rest = total - weightTotal;
+            U32 const verif = 1 << BITv06_highbit32(rest);
+            U32 const lastWeight = BITv06_highbit32(rest) + 1;
+            if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+            huffWeight[oSize] = (BYTE)lastWeight;
+            rankStats[lastWeight]++;
+    }   }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    return iSize+1;
+}
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* HUFv06_STATIC_H */
+/* ******************************************************************
+   Huffman decoder, part of New Generation Entropy library
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUFv06_STATIC_ASSERT(c) { enum { HUFv06_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+
+/* *******************************************************
+*  HUF : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUFv06_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUFv06_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+
+size_t HUFv06_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUFv06_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUFv06_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    size_t iSize;
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* const dtPtr = DTable + 1;
+    HUFv06_DEltX2* const dt = (HUFv06_DEltX2*)dtPtr;
+
+    HUFv06_STATIC_ASSERT(sizeof(HUFv06_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv06_readStats(huffWeight, HUFv06_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv06_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof allocated DTable, from used size of DTable, in case of re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<tableLog+1; n++) {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++) {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUFv06_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+
+static BYTE HUFv06_decodeSymbolX2(BITv06_DStream_t* Dstream, const HUFv06_DEltX2* dt, const U32 dtLog)
+{
+    const size_t val = BITv06_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+    const BYTE c = dt[val].byte;
+    BITv06_skipBits(Dstream, dt[val].nbBits);
+    return c;
+}
+
+#define HUFv06_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUFv06_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUFv06_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv06_MAX_TABLELOG<=12)) \
+        HUFv06_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUFv06_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUFv06_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUFv06_decodeStreamX2(BYTE* p, BITv06_DStream_t* const bitDPtr, BYTE* const pEnd, const HUFv06_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p <= pEnd-4)) {
+        HUFv06_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p < pEnd))
+        HUFv06_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUFv06_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+size_t HUFv06_decompress1X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + dstSize;
+    const U32 dtLog = DTable[0];
+    const void* dtPtr = DTable;
+    const HUFv06_DEltX2* const dt = ((const HUFv06_DEltX2*)dtPtr)+1;
+    BITv06_DStream_t bitD;
+
+    { size_t const errorCode = BITv06_initDStream(&bitD, cSrc, cSrcSize);
+      if (HUFv06_isError(errorCode)) return errorCode; }
+
+    HUFv06_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv06_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+size_t HUFv06_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX2(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const errorCode = HUFv06_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv06_decompress1X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+size_t HUFv06_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);  /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv06_DEltX2* const dt = ((const HUFv06_DEltX2*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BITv06_DStream_t bitD1;
+        BITv06_DStream_t bitD2;
+        BITv06_DStream_t bitD3;
+        BITv06_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv06_initDStream(&bitD1, istart1, length1);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD2, istart2, length2);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD3, istart3, length3);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD4, istart4, length4);
+        if (HUFv06_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv06_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv06_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_0(op4, &bitD4);
+            endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv06_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv06_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv06_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv06_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv06_endOfDStream(&bitD1) & BITv06_endOfDStream(&bitD2) & BITv06_endOfDStream(&bitD3) & BITv06_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv06_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX2(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const errorCode = HUFv06_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv06_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+
+static void HUFv06_fillDTableX4Level2(HUFv06_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUFv06_DEltX4 DElt;
+    U32 rankVal[HUFv06_ABSOLUTEMAX_TABLELOG + 1];
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1) {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    { U32 s; for (s=0; s<sortedListSize; s++) {   /* note : sortedSymbols already skipped */
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }}
+}
+
+typedef U32 rankVal_t[HUFv06_ABSOLUTEMAX_TABLELOG][HUFv06_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUFv06_fillDTableX4(HUFv06_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUFv06_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits) {   /* enough room for a second symbol */
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUFv06_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        } else {
+            HUFv06_DEltX4 DElt;
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits = (BYTE)(nbBits);
+            DElt.length = 1;
+            {   U32 u;
+                const U32 end = start + length;
+                for (u = start; u < end; u++) DTable[u] = DElt;
+        }   }
+        rankVal[weight] += length;
+    }
+}
+
+size_t HUFv06_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUFv06_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUFv06_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUFv06_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUFv06_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    size_t iSize;
+    void* dtPtr = DTable;
+    HUFv06_DEltX4* const dt = ((HUFv06_DEltX4*)dtPtr) + 1;
+
+    HUFv06_STATIC_ASSERT(sizeof(HUFv06_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUFv06_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    /* memset(weightList, 0, sizeof(weightList)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv06_readStats(weightList, HUFv06_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv06_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {   U32 w, nextRankStart = 0;
+        for (w=1; w<maxW+1; w++) {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {   U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 const w = weightList[s];
+            U32 const r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {   U32* const rankVal0 = rankVal[0];
+        {   int const rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+            U32 nextRankVal = 0;
+            U32 w;
+            for (w=1; w<maxW+1; w++) {
+                U32 current = nextRankVal;
+                nextRankVal += rankStats[w] << (w+rescale);
+                rankVal0[w] = current;
+        }   }
+        {   U32 const minBits = tableLog+1 - maxW;
+            U32 consumed;
+            for (consumed = minBits; consumed < memLog - minBits + 1; consumed++) {
+                U32* const rankValPtr = rankVal[consumed];
+                U32 w;
+                for (w = 1; w < maxW+1; w++) {
+                    rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }   }
+
+    HUFv06_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUFv06_decodeSymbolX4(void* op, BITv06_DStream_t* DStream, const HUFv06_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv06_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BITv06_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUFv06_decodeLastSymbolX4(void* op, BITv06_DStream_t* DStream, const HUFv06_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv06_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BITv06_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BITv06_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }   }
+    return 1;
+}
+
+
+#define HUFv06_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUFv06_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv06_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv06_MAX_TABLELOG<=12)) \
+        ptr += HUFv06_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv06_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUFv06_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUFv06_decodeStreamX4(BYTE* p, BITv06_DStream_t* bitDPtr, BYTE* const pEnd, const HUFv06_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p < pEnd-7)) {
+        HUFv06_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p <= pEnd-2))
+        HUFv06_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUFv06_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUFv06_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+size_t HUFv06_decompress1X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    const BYTE* const istart = (const BYTE*) cSrc;
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* const oend = ostart + dstSize;
+
+    const U32 dtLog = DTable[0];
+    const void* const dtPtr = DTable;
+    const HUFv06_DEltX4* const dt = ((const HUFv06_DEltX4*)dtPtr) +1;
+
+    /* Init */
+    BITv06_DStream_t bitD;
+    { size_t const errorCode = BITv06_initDStream(&bitD, istart, cSrcSize);
+      if (HUFv06_isError(errorCode)) return errorCode; }
+
+    /* decode */
+    HUFv06_decodeStreamX4(ostart, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv06_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+size_t HUFv06_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX4(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv06_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv06_decompress1X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+size_t HUFv06_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv06_DEltX4* const dt = ((const HUFv06_DEltX4*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BITv06_DStream_t bitD1;
+        BITv06_DStream_t bitD2;
+        BITv06_DStream_t bitD3;
+        BITv06_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv06_initDStream(&bitD1, istart1, length1);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD2, istart2, length2);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD3, istart3, length3);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD4, istart4, length4);
+        if (HUFv06_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv06_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv06_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv06_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv06_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv06_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv06_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv06_endOfDStream(&bitD1) & BITv06_endOfDStream(&bitD2) & BITv06_endOfDStream(&bitD3) & BITv06_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv06_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX4(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv06_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv06_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+
+
+/* ********************************/
+/* Generic decompression selector */
+/* ********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+size_t HUFv06_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUFv06_decompress4X2, HUFv06_decompress4X4, NULL };
+    U32 Dtime[3];   /* decompression time estimation */
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    {   U32 const Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+        U32 const D256 = (U32)(dstSize >> 8);
+        U32 n; for (n=0; n<3; n++)
+            Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+    }
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    {   U32 algoNb = 0;
+        if (Dtime[1] < Dtime[0]) algoNb = 1;
+        /* if (Dtime[2] < Dtime[algoNb]) algoNb = 2; */   /* current speed of HUFv06_decompress4X6 is not good */
+        return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+    }
+
+    /* return HUFv06_decompress4X2(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams single-symbol decoding */
+    /* return HUFv06_decompress4X4(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams double-symbols decoding */
+    /* return HUFv06_decompress4X6(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams quad-symbols decoding */
+}
+/*
+    Common functions of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd/
+*/
+
+
+/*-****************************************
+*  Version
+******************************************/
+
+/*-****************************************
+*  ZSTD Error Management
+******************************************/
+/*! ZSTDv06_isError() :
+*   tells if a return value is an error code */
+unsigned ZSTDv06_isError(size_t code) { return ERR_isError(code); }
+
+/*! ZSTDv06_getErrorName() :
+*   provides error code string from function result (useful for debugging) */
+const char* ZSTDv06_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* **************************************************************
+*  ZBUFF Error Management
+****************************************************************/
+unsigned ZBUFFv06_isError(size_t errorCode) { return ERR_isError(errorCode); }
+
+const char* ZBUFFv06_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTDv06_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTDv06_HEAPMODE
+#  define ZSTDv06_HEAPMODE 1
+#endif
+
+
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/*-*************************************
+*  Macros
+***************************************/
+#define ZSTDv06_isError ERR_isError   /* for inlining */
+#define FSEv06_isError  ERR_isError
+#define HUFv06_isError  ERR_isError
+
+
+/*_*******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTDv06_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/*-*************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock } ZSTDv06_dStage;
+
+struct ZSTDv06_DCtx_s
+{
+    FSEv06_DTable LLTable[FSEv06_DTABLE_SIZE_U32(LLFSELog)];
+    FSEv06_DTable OffTable[FSEv06_DTABLE_SIZE_U32(OffFSELog)];
+    FSEv06_DTable MLTable[FSEv06_DTABLE_SIZE_U32(MLFSELog)];
+    unsigned   hufTableX4[HUFv06_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    size_t headerSize;
+    ZSTDv06_frameParams fParams;
+    blockType_t bType;   /* used in ZSTDv06_decompressContinue(), to transfer blockType between header decoding and block decoding stages */
+    ZSTDv06_dStage stage;
+    U32 flagRepeatTable;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[ZSTDv06_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH];
+    BYTE headerBuffer[ZSTDv06_FRAMEHEADERSIZE_MAX];
+};  /* typedef'd to ZSTDv06_DCtx within "zstd_static.h" */
+
+size_t ZSTDv06_sizeofDCtx (void); /* Hidden declaration */
+size_t ZSTDv06_sizeofDCtx (void) { return sizeof(ZSTDv06_DCtx); }
+
+size_t ZSTDv06_decompressBegin(ZSTDv06_DCtx* dctx)
+{
+    dctx->expected = ZSTDv06_frameHeaderSize_min;
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    dctx->hufTableX4[0] = ZSTD_HUFFDTABLE_CAPACITY_LOG;
+    dctx->flagRepeatTable = 0;
+    return 0;
+}
+
+ZSTDv06_DCtx* ZSTDv06_createDCtx(void)
+{
+    ZSTDv06_DCtx* dctx = (ZSTDv06_DCtx*)malloc(sizeof(ZSTDv06_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTDv06_decompressBegin(dctx);
+    return dctx;
+}
+
+size_t ZSTDv06_freeDCtx(ZSTDv06_DCtx* dctx)
+{
+    free(dctx);
+    return 0;   /* reserved as a potential error code in the future */
+}
+
+void ZSTDv06_copyDCtx(ZSTDv06_DCtx* dstDCtx, const ZSTDv06_DCtx* srcDCtx)
+{
+    memcpy(dstDCtx, srcDCtx,
+           sizeof(ZSTDv06_DCtx) - (ZSTDv06_BLOCKSIZE_MAX+WILDCOPY_OVERLENGTH + ZSTDv06_frameHeaderSize_max));  /* no need to copy workspace */
+}
+
+
+/*-*************************************************************
+*   Decompression section
+***************************************************************/
+
+/* Frame format description
+   Frame Header -  [ Block Header - Block ] - Frame End
+   1) Frame Header
+      - 4 bytes - Magic Number : ZSTDv06_MAGICNUMBER (defined within zstd_static.h)
+      - 1 byte  - Frame Descriptor
+   2) Block Header
+      - 3 bytes, starting with a 2-bits descriptor
+                 Uncompressed, Compressed, Frame End, unused
+   3) Block
+      See Block Format Description
+   4) Frame End
+      - 3 bytes, compatible with Block Header
+*/
+
+
+/* Frame descriptor
+
+   1 byte, using :
+   bit 0-3 : windowLog - ZSTDv06_WINDOWLOG_ABSOLUTEMIN   (see zstd_internal.h)
+   bit 4   : minmatch 4(0) or 3(1)
+   bit 5   : reserved (must be zero)
+   bit 6-7 : Frame content size : unknown, 1 byte, 2 bytes, 8 bytes
+
+   Optional : content size (0, 1, 2 or 8 bytes)
+   0 : unknown
+   1 : 0-255 bytes
+   2 : 256 - 65535+256
+   8 : up to 16 exa
+*/
+
+
+/* Compressed Block, format description
+
+   Block = Literal Section - Sequences Section
+   Prerequisite : size of (compressed) block, maximum size of regenerated data
+
+   1) Literal Section
+
+   1.1) Header : 1-5 bytes
+        flags: 2 bits
+            00 compressed by Huff0
+            01 unused
+            10 is Raw (uncompressed)
+            11 is Rle
+            Note : using 01 => Huff0 with precomputed table ?
+            Note : delta map ? => compressed ?
+
+   1.1.1) Huff0-compressed literal block : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+   1.1.2) Raw (uncompressed) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RAW<<6) + (0<<4) + size
+               12 bits: (IS_RAW<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RAW<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.3) Rle (repeated single byte) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RLE<<6) + (0<<4) + size
+               12 bits: (IS_RLE<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RLE<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.4) Huff0-compressed literal block, using precomputed CTables : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+        1- CTable available (stored into workspace ?)
+        2- Small input (fast heuristic ? Full comparison ? depend on clevel ?)
+
+
+   1.2) Literal block content
+
+   1.2.1) Huff0 block, using sizes from header
+        See Huff0 format
+
+   1.2.2) Huff0 block, using prepared table
+
+   1.2.3) Raw content
+
+   1.2.4) single byte
+
+
+   2) Sequences section
+      TO DO
+*/
+
+/** ZSTDv06_frameHeaderSize() :
+*   srcSize must be >= ZSTDv06_frameHeaderSize_min.
+*   @return : size of the Frame Header */
+static size_t ZSTDv06_frameHeaderSize(const void* src, size_t srcSize)
+{
+    if (srcSize < ZSTDv06_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    { U32 const fcsId = (((const BYTE*)src)[4]) >> 6;
+      return ZSTDv06_frameHeaderSize_min + ZSTDv06_fcs_fieldSize[fcsId]; }
+}
+
+
+/** ZSTDv06_getFrameParams() :
+*   decode Frame Header, or provide expected `srcSize`.
+*   @return : 0, `fparamsPtr` is correctly filled,
+*            >0, `srcSize` is too small, result is expected `srcSize`,
+*             or an error code, which can be tested using ZSTDv06_isError() */
+size_t ZSTDv06_getFrameParams(ZSTDv06_frameParams* fparamsPtr, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize < ZSTDv06_frameHeaderSize_min) return ZSTDv06_frameHeaderSize_min;
+    if (MEM_readLE32(src) != ZSTDv06_MAGICNUMBER) return ERROR(prefix_unknown);
+
+    /* ensure there is enough `srcSize` to fully read/decode frame header */
+    { size_t const fhsize = ZSTDv06_frameHeaderSize(src, srcSize);
+      if (srcSize < fhsize) return fhsize; }
+
+    memset(fparamsPtr, 0, sizeof(*fparamsPtr));
+    {   BYTE const frameDesc = ip[4];
+        fparamsPtr->windowLog = (frameDesc & 0xF) + ZSTDv06_WINDOWLOG_ABSOLUTEMIN;
+        if ((frameDesc & 0x20) != 0) return ERROR(frameParameter_unsupported);   /* reserved 1 bit */
+        switch(frameDesc >> 6)  /* fcsId */
+        {
+            default:   /* impossible */
+            case 0 : fparamsPtr->frameContentSize = 0; break;
+            case 1 : fparamsPtr->frameContentSize = ip[5]; break;
+            case 2 : fparamsPtr->frameContentSize = MEM_readLE16(ip+5)+256; break;
+            case 3 : fparamsPtr->frameContentSize = MEM_readLE64(ip+5); break;
+    }   }
+    return 0;
+}
+
+
+/** ZSTDv06_decodeFrameHeader() :
+*   `srcSize` must be the size provided by ZSTDv06_frameHeaderSize().
+*   @return : 0 if success, or an error code, which can be tested using ZSTDv06_isError() */
+static size_t ZSTDv06_decodeFrameHeader(ZSTDv06_DCtx* zc, const void* src, size_t srcSize)
+{
+    size_t const result = ZSTDv06_getFrameParams(&(zc->fParams), src, srcSize);
+    if ((MEM_32bits()) && (zc->fParams.windowLog > 25)) return ERROR(frameParameter_unsupported);
+    return result;
+}
+
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+/*! ZSTDv06_getcBlockSize() :
+*   Provides the size of compressed block from block header `src` */
+static size_t ZSTDv06_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE*)src;
+    U32 cSize;
+
+    if (srcSize < ZSTDv06_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    bpPtr->blockType = (blockType_t)((*in) >> 6);
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+
+static size_t ZSTDv06_copyRawBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    if (dst==NULL) return ERROR(dstSize_tooSmall);
+    if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+/*! ZSTDv06_decodeLiteralsBlock() :
+    @return : nb of bytes read from src (< srcSize ) */
+static size_t ZSTDv06_decodeLiteralsBlock(ZSTDv06_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(istart[0]>> 6)
+    {
+    case IS_HUF:
+        {   size_t litSize, litCSize, singleStream=0;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for lhSize, + cSize (+nbSeq) */
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                /* 2 - 2 - 10 - 10 */
+                lhSize=3;
+                singleStream = istart[0] & 16;
+                litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+                litCSize = ((istart[1] &  3) << 8) + istart[2];
+                break;
+            case 2:
+                /* 2 - 2 - 14 - 14 */
+                lhSize=4;
+                litSize  = ((istart[0] & 15) << 10) + (istart[1] << 2) + (istart[2] >> 6);
+                litCSize = ((istart[2] & 63) <<  8) + istart[3];
+                break;
+            case 3:
+                /* 2 - 2 - 18 - 18 */
+                lhSize=5;
+                litSize  = ((istart[0] & 15) << 14) + (istart[1] << 6) + (istart[2] >> 2);
+                litCSize = ((istart[2] &  3) << 16) + (istart[3] << 8) + istart[4];
+                break;
+            }
+            if (litSize > ZSTDv06_BLOCKSIZE_MAX) return ERROR(corruption_detected);
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            if (HUFv06_isError(singleStream ?
+                            HUFv06_decompress1X2(dctx->litBuffer, litSize, istart+lhSize, litCSize) :
+                            HUFv06_decompress   (dctx->litBuffer, litSize, istart+lhSize, litCSize) ))
+                return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_PCH:
+        {   size_t litSize, litCSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (lhSize != 1)  /* only case supported for now : small litSize, single stream */
+                return ERROR(corruption_detected);
+            if (!dctx->flagRepeatTable)
+                return ERROR(dictionary_corrupted);
+
+            /* 2 - 2 - 10 - 10 */
+            lhSize=3;
+            litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+            litCSize = ((istart[1] &  3) << 8) + istart[2];
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            {   size_t const errorCode = HUFv06_decompress1X4_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->hufTableX4);
+                if (HUFv06_isError(errorCode)) return ERROR(corruption_detected);
+            }
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_RAW:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize=1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                break;
+            }
+
+            if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                return lhSize+litSize;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+lhSize;
+            dctx->litSize = litSize;
+            return lhSize+litSize;
+        }
+    case IS_RLE:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize = 1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
+                break;
+            }
+            if (litSize > ZSTDv06_BLOCKSIZE_MAX) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return lhSize+1;
+        }
+    default:
+        return ERROR(corruption_detected);   /* impossible */
+    }
+}
+
+
+/*! ZSTDv06_buildSeqTable() :
+    @return : nb bytes read from src,
+              or an error code if it fails, testable with ZSTDv06_isError()
+*/
+static size_t ZSTDv06_buildSeqTable(FSEv06_DTable* DTable, U32 type, U32 max, U32 maxLog,
+                                 const void* src, size_t srcSize,
+                                 const S16* defaultNorm, U32 defaultLog, U32 flagRepeatTable)
+{
+    switch(type)
+    {
+    case FSEv06_ENCODING_RLE :
+        if (!srcSize) return ERROR(srcSize_wrong);
+        if ( (*(const BYTE*)src) > max) return ERROR(corruption_detected);
+        FSEv06_buildDTable_rle(DTable, *(const BYTE*)src);   /* if *src > max, data is corrupted */
+        return 1;
+    case FSEv06_ENCODING_RAW :
+        FSEv06_buildDTable(DTable, defaultNorm, max, defaultLog);
+        return 0;
+    case FSEv06_ENCODING_STATIC:
+        if (!flagRepeatTable) return ERROR(corruption_detected);
+        return 0;
+    default :   /* impossible */
+    case FSEv06_ENCODING_DYNAMIC :
+        {   U32 tableLog;
+            S16 norm[MaxSeq+1];
+            size_t const headerSize = FSEv06_readNCount(norm, &max, &tableLog, src, srcSize);
+            if (FSEv06_isError(headerSize)) return ERROR(corruption_detected);
+            if (tableLog > maxLog) return ERROR(corruption_detected);
+            FSEv06_buildDTable(DTable, norm, max, tableLog);
+            return headerSize;
+    }   }
+}
+
+
+static size_t ZSTDv06_decodeSeqHeaders(int* nbSeqPtr,
+                             FSEv06_DTable* DTableLL, FSEv06_DTable* DTableML, FSEv06_DTable* DTableOffb, U32 flagRepeatTable,
+                             const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip = istart;
+
+    /* check */
+    if (srcSize < MIN_SEQUENCES_SIZE) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    {   int nbSeq = *ip++;
+        if (!nbSeq) { *nbSeqPtr=0; return 1; }
+        if (nbSeq > 0x7F) {
+            if (nbSeq == 0xFF) {
+                if (ip+2 > iend) return ERROR(srcSize_wrong);
+                nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
+            } else {
+                if (ip >= iend) return ERROR(srcSize_wrong);
+                nbSeq = ((nbSeq-0x80)<<8) + *ip++;
+            }
+        }
+        *nbSeqPtr = nbSeq;
+    }
+
+    /* FSE table descriptors */
+    if (ip + 4 > iend) return ERROR(srcSize_wrong); /* min : header byte + all 3 are "raw", hence no header, but at least xxLog bits per type */
+    {   U32 const LLtype  = *ip >> 6;
+        U32 const Offtype = (*ip >> 4) & 3;
+        U32 const MLtype  = (*ip >> 2) & 3;
+        ip++;
+
+        /* Build DTables */
+        {   size_t const bhSize = ZSTDv06_buildSeqTable(DTableLL, LLtype, MaxLL, LLFSELog, ip, iend-ip, LL_defaultNorm, LL_defaultNormLog, flagRepeatTable);
+            if (ZSTDv06_isError(bhSize)) return ERROR(corruption_detected);
+            ip += bhSize;
+        }
+        {   size_t const bhSize = ZSTDv06_buildSeqTable(DTableOffb, Offtype, MaxOff, OffFSELog, ip, iend-ip, OF_defaultNorm, OF_defaultNormLog, flagRepeatTable);
+            if (ZSTDv06_isError(bhSize)) return ERROR(corruption_detected);
+            ip += bhSize;
+        }
+        {   size_t const bhSize = ZSTDv06_buildSeqTable(DTableML, MLtype, MaxML, MLFSELog, ip, iend-ip, ML_defaultNorm, ML_defaultNormLog, flagRepeatTable);
+            if (ZSTDv06_isError(bhSize)) return ERROR(corruption_detected);
+            ip += bhSize;
+    }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+} seq_t;
+
+typedef struct {
+    BITv06_DStream_t DStream;
+    FSEv06_DState_t stateLL;
+    FSEv06_DState_t stateOffb;
+    FSEv06_DState_t stateML;
+    size_t prevOffset[ZSTDv06_REP_INIT];
+} seqState_t;
+
+
+
+static void ZSTDv06_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    /* Literal length */
+    U32 const llCode = FSEv06_peekSymbol(&(seqState->stateLL));
+    U32 const mlCode = FSEv06_peekSymbol(&(seqState->stateML));
+    U32 const ofCode = FSEv06_peekSymbol(&(seqState->stateOffb));   /* <= maxOff, by table construction */
+
+    U32 const llBits = LL_bits[llCode];
+    U32 const mlBits = ML_bits[mlCode];
+    U32 const ofBits = ofCode;
+    U32 const totalBits = llBits+mlBits+ofBits;
+
+    static const U32 LL_base[MaxLL+1] = {
+                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9,   10,    11,    12,    13,    14,     15,
+                            16, 18, 20, 22, 24, 28, 32, 40, 48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                            0x2000, 0x4000, 0x8000, 0x10000 };
+
+    static const U32 ML_base[MaxML+1] = {
+                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10,   11,    12,    13,    14,    15,
+                            16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26,   27,    28,    29,    30,    31,
+                            32, 34, 36, 38, 40, 44, 48, 56, 64, 80, 96, 0x80, 0x100, 0x200, 0x400, 0x800,
+                            0x1000, 0x2000, 0x4000, 0x8000, 0x10000 };
+
+    static const U32 OF_base[MaxOff+1] = {
+                 0,        1,       3,       7,     0xF,     0x1F,     0x3F,     0x7F,
+                 0xFF,   0x1FF,   0x3FF,   0x7FF,   0xFFF,   0x1FFF,   0x3FFF,   0x7FFF,
+                 0xFFFF, 0x1FFFF, 0x3FFFF, 0x7FFFF, 0xFFFFF, 0x1FFFFF, 0x3FFFFF, 0x7FFFFF,
+                 0xFFFFFF, 0x1FFFFFF, 0x3FFFFFF, /*fake*/ 1, 1 };
+
+    /* sequence */
+    {   size_t offset;
+        if (!ofCode)
+            offset = 0;
+        else {
+            offset = OF_base[ofCode] + BITv06_readBits(&(seqState->DStream), ofBits);   /* <=  26 bits */
+            if (MEM_32bits()) BITv06_reloadDStream(&(seqState->DStream));
+        }
+
+        if (offset < ZSTDv06_REP_NUM) {
+            if (llCode == 0 && offset <= 1) offset = 1-offset;
+
+            if (offset != 0) {
+                size_t temp = seqState->prevOffset[offset];
+                if (offset != 1) {
+                    seqState->prevOffset[2] = seqState->prevOffset[1];
+                }
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset = temp;
+
+            } else {
+                offset = seqState->prevOffset[0];
+            }
+        } else {
+            offset -= ZSTDv06_REP_MOVE;
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];
+            seqState->prevOffset[0] = offset;
+        }
+        seq->offset = offset;
+    }
+
+    seq->matchLength = ML_base[mlCode] + MINMATCH + ((mlCode>31) ? BITv06_readBits(&(seqState->DStream), mlBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() && (mlBits+llBits>24)) BITv06_reloadDStream(&(seqState->DStream));
+
+    seq->litLength = LL_base[llCode] + ((llCode>15) ? BITv06_readBits(&(seqState->DStream), llBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() ||
+       (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BITv06_reloadDStream(&(seqState->DStream));
+
+    /* ANS state update */
+    FSEv06_updateState(&(seqState->stateLL), &(seqState->DStream));   /* <=  9 bits */
+    FSEv06_updateState(&(seqState->stateML), &(seqState->DStream));   /* <=  9 bits */
+    if (MEM_32bits()) BITv06_reloadDStream(&(seqState->DStream));     /* <= 18 bits */
+    FSEv06_updateState(&(seqState->stateOffb), &(seqState->DStream)); /* <=  8 bits */
+}
+
+
+static size_t ZSTDv06_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTDv06_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_8 || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+    }   }
+    /* Requirement: op <= oend_8 */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTDv06_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTDv06_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_8) {
+            ZSTDv06_wildcopy(op, match, oend_8 - op);
+            match += oend_8 - op;
+            op = oend_8;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    } else {
+        ZSTDv06_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTDv06_decompressSequences(
+                               ZSTDv06_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    FSEv06_DTable* DTableLL = dctx->LLTable;
+    FSEv06_DTable* DTableML = dctx->MLTable;
+    FSEv06_DTable* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    int nbSeq;
+
+    /* Build Decoding Tables */
+    {   size_t const seqHSize = ZSTDv06_decodeSeqHeaders(&nbSeq, DTableLL, DTableML, DTableOffb, dctx->flagRepeatTable, ip, seqSize);
+        if (ZSTDv06_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+        dctx->flagRepeatTable = 0;
+    }
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        sequence.offset = REPCODE_STARTVALUE;
+        { U32 i; for (i=0; i<ZSTDv06_REP_INIT; i++) seqState.prevOffset[i] = REPCODE_STARTVALUE; }
+        { size_t const errorCode = BITv06_initDStream(&(seqState.DStream), ip, iend-ip);
+          if (ERR_isError(errorCode)) return ERROR(corruption_detected); }
+        FSEv06_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSEv06_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSEv06_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BITv06_reloadDStream(&(seqState.DStream)) <= BITv06_DStream_completed) && nbSeq ; ) {
+            nbSeq--;
+            ZSTDv06_decodeSequence(&sequence, &seqState);
+
+#if 0  /* debug */
+            static BYTE* start = NULL;
+            if (start==NULL) start = op;
+            size_t pos = (size_t)(op-start);
+            if ((pos >= 5810037) && (pos < 5810400))
+                printf("Dpos %6u :%5u literals & match %3u bytes at distance %6u \n",
+                       pos, (U32)sequence.litLength, (U32)sequence.matchLength, (U32)sequence.offset);
+#endif
+
+            {   size_t const oneSeqSize = ZSTDv06_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+                if (ZSTDv06_isError(oneSeqSize)) return oneSeqSize;
+                op += oneSeqSize;
+        }   }
+
+        /* check if reached exact end */
+        if (nbSeq) return ERROR(corruption_detected);
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        if (litPtr > litEnd) return ERROR(corruption_detected);   /* too many literals already used */
+        if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+        if (lastLLSize > 0) {
+            memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTDv06_checkContinuity(ZSTDv06_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTDv06_decompressBlock_internal(ZSTDv06_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize >= ZSTDv06_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);
+
+    /* Decode literals sub-block */
+    {   size_t const litCSize = ZSTDv06_decodeLiteralsBlock(dctx, src, srcSize);
+        if (ZSTDv06_isError(litCSize)) return litCSize;
+        ip += litCSize;
+        srcSize -= litCSize;
+    }
+    return ZSTDv06_decompressSequences(dctx, dst, dstCapacity, ip, srcSize);
+}
+
+
+size_t ZSTDv06_decompressBlock(ZSTDv06_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    ZSTDv06_checkContinuity(dctx, dst);
+    return ZSTDv06_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+/*! ZSTDv06_decompressFrame() :
+*   `dctx` must be properly initialized */
+static size_t ZSTDv06_decompressFrame(ZSTDv06_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + dstCapacity;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties = { bt_compressed, 0 };
+
+    /* check */
+    if (srcSize < ZSTDv06_frameHeaderSize_min+ZSTDv06_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv06_frameHeaderSize(src, ZSTDv06_frameHeaderSize_min);
+        if (ZSTDv06_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTDv06_blockHeaderSize) return ERROR(srcSize_wrong);
+        if (ZSTDv06_decodeFrameHeader(dctx, src, frameHeaderSize)) return ERROR(corruption_detected);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        size_t decodedSize=0;
+        size_t const cBlockSize = ZSTDv06_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTDv06_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv06_blockHeaderSize;
+        remainingSize -= ZSTDv06_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTDv06_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTDv06_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTDv06_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+
+size_t ZSTDv06_decompress_usingPreparedDCtx(ZSTDv06_DCtx* dctx, const ZSTDv06_DCtx* refDCtx,
+                                         void* dst, size_t dstCapacity,
+                                   const void* src, size_t srcSize)
+{
+    ZSTDv06_copyDCtx(dctx, refDCtx);
+    ZSTDv06_checkContinuity(dctx, dst);
+    return ZSTDv06_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv06_decompress_usingDict(ZSTDv06_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    ZSTDv06_decompressBegin_usingDict(dctx, dict, dictSize);
+    ZSTDv06_checkContinuity(dctx, dst);
+    return ZSTDv06_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv06_decompressDCtx(ZSTDv06_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTDv06_decompress_usingDict(dctx, dst, dstCapacity, src, srcSize, NULL, 0);
+}
+
+
+size_t ZSTDv06_decompress(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+#if defined(ZSTDv06_HEAPMODE) && (ZSTDv06_HEAPMODE==1)
+    size_t regenSize;
+    ZSTDv06_DCtx* dctx = ZSTDv06_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv06_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
+    ZSTDv06_freeDCtx(dctx);
+    return regenSize;
+#else   /* stack mode */
+    ZSTDv06_DCtx dctx;
+    return ZSTDv06_decompressDCtx(&dctx, dst, dstCapacity, src, srcSize);
+#endif
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+static void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv06_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+    blockProperties_t blockProperties = { bt_compressed, 0 };
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv06_frameHeaderSize(src, srcSize);
+        if (ZSTDv06_isError(frameHeaderSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, frameHeaderSize);
+            return;
+        }
+        if (MEM_readLE32(src) != ZSTDv06_MAGICNUMBER) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+            return;
+        }
+        if (srcSize < frameHeaderSize+ZSTDv06_blockHeaderSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        size_t const cBlockSize = ZSTDv06_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTDv06_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTDv06_blockHeaderSize;
+        remainingSize -= ZSTDv06_blockHeaderSize;
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * ZSTDv06_BLOCKSIZE_MAX;
+}
+
+/*_******************************
+*  Streaming Decompression API
+********************************/
+size_t ZSTDv06_nextSrcSizeToDecompress(ZSTDv06_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+size_t ZSTDv06_decompressContinue(ZSTDv06_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);
+    if (dstCapacity) ZSTDv06_checkContinuity(dctx, dst);
+
+    /* Decompress : frame header; part 1 */
+    switch (dctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        if (srcSize != ZSTDv06_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        dctx->headerSize = ZSTDv06_frameHeaderSize(src, ZSTDv06_frameHeaderSize_min);
+        if (ZSTDv06_isError(dctx->headerSize)) return dctx->headerSize;
+        memcpy(dctx->headerBuffer, src, ZSTDv06_frameHeaderSize_min);
+        if (dctx->headerSize > ZSTDv06_frameHeaderSize_min) {
+            dctx->expected = dctx->headerSize - ZSTDv06_frameHeaderSize_min;
+            dctx->stage = ZSTDds_decodeFrameHeader;
+            return 0;
+        }
+        dctx->expected = 0;   /* not necessary to copy more */
+	/* fall-through */
+    case ZSTDds_decodeFrameHeader:
+        {   size_t result;
+            memcpy(dctx->headerBuffer + ZSTDv06_frameHeaderSize_min, src, dctx->expected);
+            result = ZSTDv06_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize);
+            if (ZSTDv06_isError(result)) return result;
+            dctx->expected = ZSTDv06_blockHeaderSize;
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDds_decodeBlockHeader:
+        {   blockProperties_t bp;
+            size_t const cBlockSize = ZSTDv06_getcBlockSize(src, ZSTDv06_blockHeaderSize, &bp);
+            if (ZSTDv06_isError(cBlockSize)) return cBlockSize;
+            if (bp.blockType == bt_end) {
+                dctx->expected = 0;
+                dctx->stage = ZSTDds_getFrameHeaderSize;
+            } else {
+                dctx->expected = cBlockSize;
+                dctx->bType = bp.blockType;
+                dctx->stage = ZSTDds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDds_decompressBlock:
+        {   size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTDv06_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTDv06_copyRawBlock(dst, dstCapacity, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);   /* impossible */
+            }
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            dctx->expected = ZSTDv06_blockHeaderSize;
+            if (ZSTDv06_isError(rSize)) return rSize;
+            dctx->previousDstEnd = (char*)dst + rSize;
+            return rSize;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static void ZSTDv06_refDictContent(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+    dctx->base = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+}
+
+static size_t ZSTDv06_loadEntropy(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t hSize, offcodeHeaderSize, matchlengthHeaderSize, litlengthHeaderSize;
+
+    hSize = HUFv06_readDTableX4(dctx->hufTableX4, dict, dictSize);
+    if (HUFv06_isError(hSize)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + hSize;
+    dictSize -= hSize;
+
+    {   short offcodeNCount[MaxOff+1];
+        U32 offcodeMaxValue=MaxOff, offcodeLog;
+        offcodeHeaderSize = FSEv06_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dict, dictSize);
+        if (FSEv06_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv06_buildDTable(dctx->OffTable, offcodeNCount, offcodeMaxValue, offcodeLog);
+          if (FSEv06_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dict = (const char*)dict + offcodeHeaderSize;
+        dictSize -= offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        matchlengthHeaderSize = FSEv06_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dict, dictSize);
+        if (FSEv06_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv06_buildDTable(dctx->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog);
+          if (FSEv06_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dict = (const char*)dict + matchlengthHeaderSize;
+        dictSize -= matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        litlengthHeaderSize = FSEv06_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dict, dictSize);
+        if (FSEv06_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv06_buildDTable(dctx->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog);
+          if (FSEv06_isError(errorCode)) return ERROR(dictionary_corrupted); }
+    }
+
+    dctx->flagRepeatTable = 1;
+    return hSize + offcodeHeaderSize + matchlengthHeaderSize + litlengthHeaderSize;
+}
+
+static size_t ZSTDv06_decompress_insertDictionary(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t eSize;
+    U32 const magic = MEM_readLE32(dict);
+    if (magic != ZSTDv06_DICT_MAGIC) {
+        /* pure content mode */
+        ZSTDv06_refDictContent(dctx, dict, dictSize);
+        return 0;
+    }
+    /* load entropy tables */
+    dict = (const char*)dict + 4;
+    dictSize -= 4;
+    eSize = ZSTDv06_loadEntropy(dctx, dict, dictSize);
+    if (ZSTDv06_isError(eSize)) return ERROR(dictionary_corrupted);
+
+    /* reference dictionary content */
+    dict = (const char*)dict + eSize;
+    dictSize -= eSize;
+    ZSTDv06_refDictContent(dctx, dict, dictSize);
+
+    return 0;
+}
+
+
+size_t ZSTDv06_decompressBegin_usingDict(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    { size_t const errorCode = ZSTDv06_decompressBegin(dctx);
+      if (ZSTDv06_isError(errorCode)) return errorCode; }
+
+    if (dict && dictSize) {
+        size_t const errorCode = ZSTDv06_decompress_insertDictionary(dctx, dict, dictSize);
+        if (ZSTDv06_isError(errorCode)) return ERROR(dictionary_corrupted);
+    }
+
+    return 0;
+}
+
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd/
+*/
+
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFFv06_DCtx object is required to track streaming operations.
+*  Use ZBUFFv06_createDCtx() and ZBUFFv06_freeDCtx() to create/release resources.
+*  Use ZBUFFv06_decompressInit() to start a new decompression operation,
+*   or ZBUFFv06_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv06_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFFv06_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of @dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change @dst.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency),
+*            or 0 when a frame is completely decoded,
+*            or an error code, which can be tested using ZBUFFv06_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv06_recommendedDInSize() and ZBUFFv06_recommendedDOutSize()
+*  output : ZBUFFv06_recommendedDOutSize==128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv06_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFFv06_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+typedef enum { ZBUFFds_init, ZBUFFds_loadHeader,
+               ZBUFFds_read, ZBUFFds_load, ZBUFFds_flush } ZBUFFv06_dStage;
+
+/* *** Resource management *** */
+struct ZBUFFv06_DCtx_s {
+    ZSTDv06_DCtx* zd;
+    ZSTDv06_frameParams fParams;
+    ZBUFFv06_dStage stage;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t blockSize;
+    BYTE headerBuffer[ZSTDv06_FRAMEHEADERSIZE_MAX];
+    size_t lhSize;
+};   /* typedef'd to ZBUFFv06_DCtx within "zstd_buffered.h" */
+
+
+ZBUFFv06_DCtx* ZBUFFv06_createDCtx(void)
+{
+    ZBUFFv06_DCtx* zbd = (ZBUFFv06_DCtx*)malloc(sizeof(ZBUFFv06_DCtx));
+    if (zbd==NULL) return NULL;
+    memset(zbd, 0, sizeof(*zbd));
+    zbd->zd = ZSTDv06_createDCtx();
+    if (zbd->zd==NULL) {
+        ZBUFFv06_freeDCtx(zbd); /* avoid leaking the context */
+        return NULL;
+    }
+    zbd->stage = ZBUFFds_init;
+    return zbd;
+}
+
+size_t ZBUFFv06_freeDCtx(ZBUFFv06_DCtx* zbd)
+{
+    if (zbd==NULL) return 0;   /* support free on null */
+    ZSTDv06_freeDCtx(zbd->zd);
+    free(zbd->inBuff);
+    free(zbd->outBuff);
+    free(zbd);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+size_t ZBUFFv06_decompressInitDictionary(ZBUFFv06_DCtx* zbd, const void* dict, size_t dictSize)
+{
+    zbd->stage = ZBUFFds_loadHeader;
+    zbd->lhSize = zbd->inPos = zbd->outStart = zbd->outEnd = 0;
+    return ZSTDv06_decompressBegin_usingDict(zbd->zd, dict, dictSize);
+}
+
+size_t ZBUFFv06_decompressInit(ZBUFFv06_DCtx* zbd)
+{
+    return ZBUFFv06_decompressInitDictionary(zbd, NULL, 0);
+}
+
+
+
+MEM_STATIC size_t ZBUFFv06_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    size_t length = MIN(dstCapacity, srcSize);
+    if (length > 0) {
+        memcpy(dst, src, length);
+    }
+    return length;
+}
+
+
+/* *** Decompression *** */
+
+size_t ZBUFFv06_decompressContinue(ZBUFFv06_DCtx* zbd,
+                                void* dst, size_t* dstCapacityPtr,
+                          const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* const iend = istart + *srcSizePtr;
+    const char* ip = istart;
+    char* const ostart = (char*)dst;
+    char* const oend = ostart + *dstCapacityPtr;
+    char* op = ostart;
+    U32 notDone = 1;
+
+    while (notDone) {
+        switch(zbd->stage)
+        {
+        case ZBUFFds_init :
+            return ERROR(init_missing);
+
+        case ZBUFFds_loadHeader :
+            {   size_t const hSize = ZSTDv06_getFrameParams(&(zbd->fParams), zbd->headerBuffer, zbd->lhSize);
+                if (hSize != 0) {
+                    size_t const toLoad = hSize - zbd->lhSize;   /* if hSize!=0, hSize > zbd->lhSize */
+                    if (ZSTDv06_isError(hSize)) return hSize;
+                    if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
+                        if (ip != NULL)
+                            memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
+                        zbd->lhSize += iend-ip;
+                        *dstCapacityPtr = 0;
+                        return (hSize - zbd->lhSize) + ZSTDv06_blockHeaderSize;   /* remaining header bytes + next block header */
+                    }
+                    memcpy(zbd->headerBuffer + zbd->lhSize, ip, toLoad); zbd->lhSize = hSize; ip += toLoad;
+                    break;
+            }   }
+
+            /* Consume header */
+            {   size_t const h1Size = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);  /* == ZSTDv06_frameHeaderSize_min */
+                size_t const h1Result = ZSTDv06_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer, h1Size);
+                if (ZSTDv06_isError(h1Result)) return h1Result;
+                if (h1Size < zbd->lhSize) {   /* long header */
+                    size_t const h2Size = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+                    size_t const h2Result = ZSTDv06_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer+h1Size, h2Size);
+                    if (ZSTDv06_isError(h2Result)) return h2Result;
+            }   }
+
+            /* Frame header instruct buffer sizes */
+            {   size_t const blockSize = MIN(1 << zbd->fParams.windowLog, ZSTDv06_BLOCKSIZE_MAX);
+                zbd->blockSize = blockSize;
+                if (zbd->inBuffSize < blockSize) {
+                    free(zbd->inBuff);
+                    zbd->inBuffSize = blockSize;
+                    zbd->inBuff = (char*)malloc(blockSize);
+                    if (zbd->inBuff == NULL) return ERROR(memory_allocation);
+                }
+                {   size_t const neededOutSize = ((size_t)1 << zbd->fParams.windowLog) + blockSize + WILDCOPY_OVERLENGTH * 2;
+                    if (zbd->outBuffSize < neededOutSize) {
+                        free(zbd->outBuff);
+                        zbd->outBuffSize = neededOutSize;
+                        zbd->outBuff = (char*)malloc(neededOutSize);
+                        if (zbd->outBuff == NULL) return ERROR(memory_allocation);
+            }   }   }
+            zbd->stage = ZBUFFds_read;
+	    /* fall-through */
+        case ZBUFFds_read:
+            {   size_t const neededInSize = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+                if (neededInSize==0) {  /* end of frame */
+                    zbd->stage = ZBUFFds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
+                    size_t const decodedSize = ZSTDv06_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, zbd->outBuffSize - zbd->outStart,
+                        ip, neededInSize);
+                    if (ZSTDv06_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize) break;   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbd->stage = ZBUFFds_load;
+            }
+	    /* fall-through */
+        case ZBUFFds_load:
+            {   size_t const neededInSize = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+                size_t const toLoad = neededInSize - zbd->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbd->inBuffSize - zbd->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFFv06_limitCopy(zbd->inBuff + zbd->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbd->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+
+                /* decode loaded input */
+                {   size_t const decodedSize = ZSTDv06_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, zbd->outBuffSize - zbd->outStart,
+                        zbd->inBuff, neededInSize);
+                    if (ZSTDv06_isError(decodedSize)) return decodedSize;
+                    zbd->inPos = 0;   /* input is consumed */
+                    if (!decodedSize) { zbd->stage = ZBUFFds_read; break; }   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    /* break; */ /* ZBUFFds_flush follows */
+                }
+	    }
+	    /* fall-through */
+        case ZBUFFds_flush:
+            {   size_t const toFlushSize = zbd->outEnd - zbd->outStart;
+                size_t const flushedSize = ZBUFFv06_limitCopy(op, oend-op, zbd->outBuff + zbd->outStart, toFlushSize);
+                op += flushedSize;
+                zbd->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {
+                    zbd->stage = ZBUFFds_read;
+                    if (zbd->outStart + zbd->blockSize > zbd->outBuffSize)
+                        zbd->outStart = zbd->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+    }   }
+
+    /* result */
+    *srcSizePtr = ip-istart;
+    *dstCapacityPtr = op-ostart;
+    {   size_t nextSrcSizeHint = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+        if (nextSrcSizeHint > ZSTDv06_blockHeaderSize) nextSrcSizeHint+= ZSTDv06_blockHeaderSize;   /* get following block header too */
+        nextSrcSizeHint -= zbd->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+size_t ZBUFFv06_recommendedDInSize(void)  { return ZSTDv06_BLOCKSIZE_MAX + ZSTDv06_blockHeaderSize /* block header size*/ ; }
+size_t ZBUFFv06_recommendedDOutSize(void) { return ZSTDv06_BLOCKSIZE_MAX; }
diff --git a/lib/legacy/zstd_v07.c b/lib/legacy/zstd_v07.c
new file mode 100644
index 0000000..cdc5628
--- /dev/null
+++ b/lib/legacy/zstd_v07.c
@@ -0,0 +1,4490 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*- Dependencies -*/
+#include <stddef.h>     /* size_t, ptrdiff_t */
+#include <string.h>     /* memcpy */
+#include <stdlib.h>     /* malloc, free, qsort */
+
+#ifndef XXH_STATIC_LINKING_ONLY
+#  define XXH_STATIC_LINKING_ONLY    /* XXH64_state_t */
+#endif
+#include "../common/xxhash.h"                  /* XXH64_* */
+#include "zstd_v07.h"
+
+#define FSEv07_STATIC_LINKING_ONLY   /* FSEv07_MIN_TABLELOG */
+#define HUFv07_STATIC_LINKING_ONLY   /* HUFv07_TABLELOG_ABSOLUTEMAX */
+#define ZSTDv07_STATIC_LINKING_ONLY
+
+#include "../common/compiler.h"
+#include "../common/error_private.h"
+
+
+#ifdef ZSTDv07_STATIC_LINKING_ONLY
+
+/* ====================================================================================
+ * The definitions in this section are considered experimental.
+ * They should never be used with a dynamic library, as they may change in the future.
+ * They are provided for advanced usages.
+ * Use them only in association with static linking.
+ * ==================================================================================== */
+
+/*--- Constants ---*/
+#define ZSTDv07_MAGIC_SKIPPABLE_START  0x184D2A50U
+
+#define ZSTDv07_WINDOWLOG_MAX_32  25
+#define ZSTDv07_WINDOWLOG_MAX_64  27
+#define ZSTDv07_WINDOWLOG_MAX    ((U32)(MEM_32bits() ? ZSTDv07_WINDOWLOG_MAX_32 : ZSTDv07_WINDOWLOG_MAX_64))
+#define ZSTDv07_WINDOWLOG_MIN     18
+#define ZSTDv07_CHAINLOG_MAX     (ZSTDv07_WINDOWLOG_MAX+1)
+#define ZSTDv07_CHAINLOG_MIN       4
+#define ZSTDv07_HASHLOG_MAX       ZSTDv07_WINDOWLOG_MAX
+#define ZSTDv07_HASHLOG_MIN       12
+#define ZSTDv07_HASHLOG3_MAX      17
+#define ZSTDv07_SEARCHLOG_MAX    (ZSTDv07_WINDOWLOG_MAX-1)
+#define ZSTDv07_SEARCHLOG_MIN      1
+#define ZSTDv07_SEARCHLENGTH_MAX   7
+#define ZSTDv07_SEARCHLENGTH_MIN   3
+#define ZSTDv07_TARGETLENGTH_MIN   4
+#define ZSTDv07_TARGETLENGTH_MAX 999
+
+#define ZSTDv07_FRAMEHEADERSIZE_MAX 18    /* for static allocation */
+static const size_t ZSTDv07_frameHeaderSize_min = 5;
+static const size_t ZSTDv07_frameHeaderSize_max = ZSTDv07_FRAMEHEADERSIZE_MAX;
+static const size_t ZSTDv07_skippableHeaderSize = 8;  /* magic number + skippable frame length */
+
+
+/* custom memory allocation functions */
+typedef void* (*ZSTDv07_allocFunction) (void* opaque, size_t size);
+typedef void  (*ZSTDv07_freeFunction) (void* opaque, void* address);
+typedef struct { ZSTDv07_allocFunction customAlloc; ZSTDv07_freeFunction customFree; void* opaque; } ZSTDv07_customMem;
+
+
+/*--- Advanced Decompression functions ---*/
+
+/*! ZSTDv07_estimateDCtxSize() :
+ *  Gives the potential amount of memory allocated to create a ZSTDv07_DCtx */
+ZSTDLIBv07_API size_t ZSTDv07_estimateDCtxSize(void);
+
+/*! ZSTDv07_createDCtx_advanced() :
+ *  Create a ZSTD decompression context using external alloc and free functions */
+ZSTDLIBv07_API ZSTDv07_DCtx* ZSTDv07_createDCtx_advanced(ZSTDv07_customMem customMem);
+
+/*! ZSTDv07_sizeofDCtx() :
+ *  Gives the amount of memory used by a given ZSTDv07_DCtx */
+ZSTDLIBv07_API size_t ZSTDv07_sizeofDCtx(const ZSTDv07_DCtx* dctx);
+
+
+/* ******************************************************************
+*  Buffer-less streaming functions (synchronous mode)
+********************************************************************/
+
+ZSTDLIBv07_API size_t ZSTDv07_decompressBegin(ZSTDv07_DCtx* dctx);
+ZSTDLIBv07_API size_t ZSTDv07_decompressBegin_usingDict(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIBv07_API void   ZSTDv07_copyDCtx(ZSTDv07_DCtx* dctx, const ZSTDv07_DCtx* preparedDCtx);
+
+ZSTDLIBv07_API size_t ZSTDv07_nextSrcSizeToDecompress(ZSTDv07_DCtx* dctx);
+ZSTDLIBv07_API size_t ZSTDv07_decompressContinue(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+/*
+  Buffer-less streaming decompression (synchronous mode)
+
+  A ZSTDv07_DCtx object is required to track streaming operations.
+  Use ZSTDv07_createDCtx() / ZSTDv07_freeDCtx() to manage it.
+  A ZSTDv07_DCtx object can be re-used multiple times.
+
+  First optional operation is to retrieve frame parameters, using ZSTDv07_getFrameParams(), which doesn't consume the input.
+  It can provide the minimum size of rolling buffer required to properly decompress data (`windowSize`),
+  and optionally the final size of uncompressed content.
+  (Note : content size is an optional info that may not be present. 0 means : content size unknown)
+  Frame parameters are extracted from the beginning of compressed frame.
+  The amount of data to read is variable, from ZSTDv07_frameHeaderSize_min to ZSTDv07_frameHeaderSize_max (so if `srcSize` >= ZSTDv07_frameHeaderSize_max, it will always work)
+  If `srcSize` is too small for operation to succeed, function will return the minimum size it requires to produce a result.
+  Result : 0 when successful, it means the ZSTDv07_frameParams structure has been filled.
+          >0 : means there is not enough data into `src`. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTDv07_isError()
+
+  Start decompression, with ZSTDv07_decompressBegin() or ZSTDv07_decompressBegin_usingDict().
+  Alternatively, you can copy a prepared context, using ZSTDv07_copyDCtx().
+
+  Then use ZSTDv07_nextSrcSizeToDecompress() and ZSTDv07_decompressContinue() alternatively.
+  ZSTDv07_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTDv07_decompressContinue().
+  ZSTDv07_decompressContinue() requires this exact amount of bytes, or it will fail.
+
+  @result of ZSTDv07_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
+  It can be zero, which is not an error; it just means ZSTDv07_decompressContinue() has decoded some header.
+
+  ZSTDv07_decompressContinue() needs previous data blocks during decompression, up to `windowSize`.
+  They should preferably be located contiguously, prior to current block.
+  Alternatively, a round buffer of sufficient size is also possible. Sufficient size is determined by frame parameters.
+  ZSTDv07_decompressContinue() is very sensitive to contiguity,
+  if 2 blocks don't follow each other, make sure that either the compressor breaks contiguity at the same place,
+    or that previous contiguous segment is large enough to properly handle maximum back-reference.
+
+  A frame is fully decoded when ZSTDv07_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+
+
+  == Special case : skippable frames ==
+
+  Skippable frames allow the integration of user-defined data into a flow of concatenated frames.
+  Skippable frames will be ignored (skipped) by a decompressor. The format of skippable frame is following:
+  a) Skippable frame ID - 4 Bytes, Little endian format, any value from 0x184D2A50 to 0x184D2A5F
+  b) Frame Size - 4 Bytes, Little endian format, unsigned 32-bits
+  c) Frame Content - any content (User Data) of length equal to Frame Size
+  For skippable frames ZSTDv07_decompressContinue() always returns 0.
+  For skippable frames ZSTDv07_getFrameParams() returns fparamsPtr->windowLog==0 what means that a frame is skippable.
+  It also returns Frame Size as fparamsPtr->frameContentSize.
+*/
+
+
+/* **************************************
+*  Block functions
+****************************************/
+/*! Block functions produce and decode raw zstd blocks, without frame metadata.
+    Frame metadata cost is typically ~18 bytes, which can be non-negligible for very small blocks (< 100 bytes).
+    User will have to take in charge required information to regenerate data, such as compressed and content sizes.
+
+    A few rules to respect :
+    - Compressing and decompressing require a context structure
+      + Use ZSTDv07_createCCtx() and ZSTDv07_createDCtx()
+    - It is necessary to init context before starting
+      + compression : ZSTDv07_compressBegin()
+      + decompression : ZSTDv07_decompressBegin()
+      + variants _usingDict() are also allowed
+      + copyCCtx() and copyDCtx() work too
+    - Block size is limited, it must be <= ZSTDv07_getBlockSizeMax()
+      + If you need to compress more, cut data into multiple blocks
+      + Consider using the regular ZSTDv07_compress() instead, as frame metadata costs become negligible when source size is large.
+    - When a block is considered not compressible enough, ZSTDv07_compressBlock() result will be zero.
+      In which case, nothing is produced into `dst`.
+      + User must test for such outcome and deal directly with uncompressed data
+      + ZSTDv07_decompressBlock() doesn't accept uncompressed data as input !!!
+      + In case of multiple successive blocks, decoder must be informed of uncompressed block existence to follow proper history.
+        Use ZSTDv07_insertBlock() in such a case.
+*/
+
+#define ZSTDv07_BLOCKSIZE_ABSOLUTEMAX (128 * 1024)   /* define, for static allocation */
+ZSTDLIBv07_API size_t ZSTDv07_decompressBlock(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIBv07_API size_t ZSTDv07_insertBlock(ZSTDv07_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert block into `dctx` history. Useful for uncompressed blocks */
+
+
+#endif   /* ZSTDv07_STATIC_LINKING_ONLY */
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-****************************************
+*  Compiler specifics
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+
+
+/*-**************************************************************
+*  Basic Types
+*****************************************************************/
+#if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-**************************************************************
+*  Memory I/O
+*****************************************************************/
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC U32 MEM_swap32(U32 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_ulong(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap32(in);
+#else
+    return  ((in << 24) & 0xff000000 ) |
+            ((in <<  8) & 0x00ff0000 ) |
+            ((in >>  8) & 0x0000ff00 ) |
+            ((in >> 24) & 0x000000ff );
+#endif
+}
+
+MEM_STATIC U64 MEM_swap64(U64 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_uint64(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap64(in);
+#else
+    return  ((in << 56) & 0xff00000000000000ULL) |
+            ((in << 40) & 0x00ff000000000000ULL) |
+            ((in << 24) & 0x0000ff0000000000ULL) |
+            ((in << 8)  & 0x000000ff00000000ULL) |
+            ((in >> 8)  & 0x00000000ff000000ULL) |
+            ((in >> 24) & 0x0000000000ff0000ULL) |
+            ((in >> 40) & 0x000000000000ff00ULL) |
+            ((in >> 56) & 0x00000000000000ffULL);
+#endif
+}
+
+
+/*=== Little endian r/w ===*/
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian()) {
+        MEM_write16(memPtr, val);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+        return MEM_swap32(MEM_read32(memPtr));
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+        return MEM_swap64(MEM_read64(memPtr));
+}
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+/* ******************************************************************
+   bitstream
+   Part of FSE library
+   header file (to include)
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which must be inlined for best performance.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/*=========================================
+*  Target specific
+=========================================*/
+#if defined(__BMI__) && defined(__GNUC__)
+#  include <immintrin.h>   /* support for bextr (experimental) */
+#endif
+
+/*-********************************************
+*  bitStream decoding API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BITv07_DStream_t;
+
+typedef enum { BITv07_DStream_unfinished = 0,
+               BITv07_DStream_endOfBuffer = 1,
+               BITv07_DStream_completed = 2,
+               BITv07_DStream_overflow = 3 } BITv07_DStream_status;  /* result of BITv07_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BITv07_initDStream(BITv07_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BITv07_readBits(BITv07_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BITv07_DStream_status BITv07_reloadDStream(BITv07_DStream_t* bitD);
+MEM_STATIC unsigned BITv07_endOfDStream(const BITv07_DStream_t* bitD);
+
+
+
+/*-****************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BITv07_readBitsFast(BITv07_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/*-**************************************************************
+*  Internal functions
+****************************************************************/
+MEM_STATIC unsigned BITv07_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return __builtin_clz (val) ^ 31;
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    return DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+#   endif
+}
+
+
+
+/*-********************************************************
+* bitStream decoding
+**********************************************************/
+/*! BITv07_initDStream() :
+*   Initialize a BITv07_DStream_t.
+*   `bitD` : a pointer to an already allocated BITv07_DStream_t structure.
+*   `srcSize` must be the *exact* size of the bitStream, in bytes.
+*   @return : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BITv07_initDStream(BITv07_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(bitD->bitContainer)) {  /* normal case */
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          bitD->bitsConsumed = lastByte ? 8 - BITv07_highbit32(lastByte) : 0;
+          if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
+    } else {
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          bitD->bitsConsumed = lastByte ? 8 - BITv07_highbit32(lastByte) : 0;
+          if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
+        bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+
+ MEM_STATIC size_t BITv07_lookBits(const BITv07_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BITv07_lookBitsFast() :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv07_lookBitsFast(const BITv07_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BITv07_skipBits(BITv07_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BITv07_readBits(BITv07_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv07_lookBits(bitD, nbBits);
+    BITv07_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*! BITv07_readBitsFast() :
+*   unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BITv07_readBitsFast(BITv07_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv07_lookBitsFast(bitD, nbBits);
+    BITv07_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BITv07_DStream_status BITv07_reloadDStream(BITv07_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should not happen => corruption detected */
+        return BITv07_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer)) {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BITv07_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start) {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BITv07_DStream_endOfBuffer;
+        return BITv07_DStream_completed;
+    }
+    {   U32 nbBytes = bitD->bitsConsumed >> 3;
+        BITv07_DStream_status result = BITv07_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start) {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BITv07_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BITv07_endOfDStream() :
+*   @return Tells if DStream has exactly reached its end (all bits consumed).
+*/
+MEM_STATIC unsigned BITv07_endOfDStream(const BITv07_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   FSE : Finite State Entropy codec
+   Public Prototypes declaration
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef FSEv07_H
+#define FSEv07_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*-****************************************
+*  FSE simple functions
+******************************************/
+
+/*! FSEv07_decompress():
+    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstCapacity'.
+    @return : size of regenerated data (<= maxDstSize),
+              or an error code, which can be tested using FSEv07_isError() .
+
+    ** Important ** : FSEv07_decompress() does not decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+size_t FSEv07_decompress(void* dst,  size_t dstCapacity,
+                const void* cSrc, size_t cSrcSize);
+
+
+/* Error Management */
+unsigned    FSEv07_isError(size_t code);        /* tells if a return value is an error code */
+const char* FSEv07_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+/*-*****************************************
+*  FSE detailed API
+******************************************/
+/*!
+FSEv07_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*! FSEv07_readNCount():
+    Read compactly saved 'normalizedCounter' from 'rBuffer'.
+    @return : size read from 'rBuffer',
+              or an errorCode, which can be tested using FSEv07_isError().
+              maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+size_t FSEv07_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*! Constructor and Destructor of FSEv07_DTable.
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSEv07_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+FSEv07_DTable* FSEv07_createDTable(unsigned tableLog);
+void        FSEv07_freeDTable(FSEv07_DTable* dt);
+
+/*! FSEv07_buildDTable():
+    Builds 'dt', which must be already allocated, using FSEv07_createDTable().
+    return : 0, or an errorCode, which can be tested using FSEv07_isError() */
+size_t FSEv07_buildDTable (FSEv07_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSEv07_decompress_usingDTable():
+    Decompress compressed source `cSrc` of size `cSrcSize` using `dt`
+    into `dst` which must be already allocated.
+    @return : size of regenerated data (necessarily <= `dstCapacity`),
+              or an errorCode, which can be tested using FSEv07_isError() */
+size_t FSEv07_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSEv07_DTable* dt);
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSEv07_readNCount() if it was saved using FSEv07_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSEv07_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSEv07_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSEv07_isError().
+
+The next step is to build the decompression tables 'FSEv07_DTable' from 'normalizedCounter'.
+This is performed by the function FSEv07_buildDTable().
+The space required by 'FSEv07_DTable' must be already allocated using FSEv07_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSEv07_isError().
+
+`FSEv07_DTable` can then be used to decompress `cSrc`, with FSEv07_decompress_usingDTable().
+`cSrcSize` must be strictly correct, otherwise decompression will fail.
+FSEv07_decompress_usingDTable() result will tell how many bytes were regenerated (<=`dstCapacity`).
+If there is an error, the function will return an error code, which can be tested using FSEv07_isError(). (ex: dst buffer too small)
+*/
+
+
+#ifdef FSEv07_STATIC_LINKING_ONLY
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSEv07_NCOUNTBOUND 512
+#define FSEv07_BLOCKBOUND(size) (size + (size>>7))
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of unsigned using below macros */
+#define FSEv07_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSE advanced API
+*******************************************/
+size_t FSEv07_countFast(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+/**< same as FSEv07_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr  */
+
+unsigned FSEv07_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
+/**< same as FSEv07_optimalTableLog(), which used `minus==2` */
+
+size_t FSEv07_buildDTable_raw (FSEv07_DTable* dt, unsigned nbBits);
+/**< build a fake FSEv07_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+size_t FSEv07_buildDTable_rle (FSEv07_DTable* dt, unsigned char symbolValue);
+/**< build a fake FSEv07_DTable, designed to always generate the same symbolValue */
+
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSEv07_DState_t;
+
+
+static void     FSEv07_initDState(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD, const FSEv07_DTable* dt);
+
+static unsigned char FSEv07_decodeSymbol(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD);
+
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSEv07_decodeSymbolFast(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* ======    Decompression    ====== */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSEv07_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSEv07_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSEv07_initDState(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD, const FSEv07_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv07_DTableHeader* const DTableH = (const FSEv07_DTableHeader*)ptr;
+    DStatePtr->state = BITv07_readBits(bitD, DTableH->tableLog);
+    BITv07_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSEv07_peekSymbol(const FSEv07_DState_t* DStatePtr)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC void FSEv07_updateState(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BITv07_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.newState + lowBits;
+}
+
+MEM_STATIC BYTE FSEv07_decodeSymbol(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv07_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+/*! FSEv07_decodeSymbolFast() :
+    unsafe, only works if no symbol has a probability > 50% */
+MEM_STATIC BYTE FSEv07_decodeSymbolFast(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv07_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+
+
+#ifndef FSEv07_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSEv07_MAX_MEMORY_USAGE 14
+#define FSEv07_DEFAULT_MEMORY_USAGE 13
+
+/*!FSEv07_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSEv07_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSEv07_FUNCTION_TYPE BYTE
+#define FSEv07_FUNCTION_EXTENSION
+#define FSEv07_DECODE_TYPE FSEv07_decode_t
+
+
+#endif   /* !FSEv07_COMMONDEFS_ONLY */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSEv07_MAX_TABLELOG  (FSEv07_MAX_MEMORY_USAGE-2)
+#define FSEv07_MAX_TABLESIZE (1U<<FSEv07_MAX_TABLELOG)
+#define FSEv07_MAXTABLESIZE_MASK (FSEv07_MAX_TABLESIZE-1)
+#define FSEv07_DEFAULT_TABLELOG (FSEv07_DEFAULT_MEMORY_USAGE-2)
+#define FSEv07_MIN_TABLELOG 5
+
+#define FSEv07_TABLELOG_ABSOLUTE_MAX 15
+#if FSEv07_MAX_TABLELOG > FSEv07_TABLELOG_ABSOLUTE_MAX
+#  error "FSEv07_MAX_TABLELOG > FSEv07_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+#define FSEv07_TABLESTEP(tableSize) ((tableSize>>1) + (tableSize>>3) + 3)
+
+
+#endif /* FSEv07_STATIC_LINKING_ONLY */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv07_H */
+/* ******************************************************************
+   Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFv07_H_298734234
+#define HUFv07_H_298734234
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* *** simple functions *** */
+/**
+HUFv07_decompress() :
+    Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated buffer 'dst', of minimum size 'dstSize'.
+    `dstSize` : **must** be the ***exact*** size of original (uncompressed) data.
+    Note : in contrast with FSE, HUFv07_decompress can regenerate
+           RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+           because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize),
+              or an error code, which can be tested using HUFv07_isError()
+*/
+size_t HUFv07_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+#define HUFv07_BLOCKSIZE_MAX (128 * 1024)
+
+/* Error Management */
+unsigned    HUFv07_isError(size_t code);        /**< tells if a return value is an error code */
+const char* HUFv07_getErrorName(size_t code);   /**< provides error code string (useful for debugging) */
+
+
+/* *** Advanced function *** */
+
+
+#ifdef HUFv07_STATIC_LINKING_ONLY
+
+
+/* *** Constants *** */
+#define HUFv07_TABLELOG_ABSOLUTEMAX  16   /* absolute limit of HUFv07_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUFv07_TABLELOG_MAX  12           /* max configured tableLog (for static allocation); can be modified up to HUFv07_ABSOLUTEMAX_TABLELOG */
+#define HUFv07_TABLELOG_DEFAULT  11       /* tableLog by default, when not specified */
+#define HUFv07_SYMBOLVALUE_MAX 255
+#if (HUFv07_TABLELOG_MAX > HUFv07_TABLELOG_ABSOLUTEMAX)
+#  error "HUFv07_TABLELOG_MAX is too large !"
+#endif
+
+
+/* ****************************************
+*  Static allocation
+******************************************/
+/* HUF buffer bounds */
+#define HUFv07_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+
+/* static allocation of HUF's DTable */
+typedef U32 HUFv07_DTable;
+#define HUFv07_DTABLE_SIZE(maxTableLog)   (1 + (1<<(maxTableLog)))
+#define HUFv07_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        HUFv07_DTable DTable[HUFv07_DTABLE_SIZE((maxTableLog)-1)] = { ((U32)((maxTableLog)-1)*0x1000001) }
+#define HUFv07_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        HUFv07_DTable DTable[HUFv07_DTABLE_SIZE(maxTableLog)] = { ((U32)(maxTableLog)*0x1000001) }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+size_t HUFv07_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUFv07_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+
+size_t HUFv07_decompress4X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< decodes RLE and uncompressed */
+size_t HUFv07_decompress4X_hufOnly(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize); /**< considers RLE and uncompressed as errors */
+size_t HUFv07_decompress4X2_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUFv07_decompress4X4_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+
+size_t HUFv07_decompress1X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+size_t HUFv07_decompress1X2_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUFv07_decompress1X4_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+
+
+/* ****************************************
+*  HUF detailed API
+******************************************/
+/*!
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and regenerate 'CTable' using external methods.
+*/
+/* FSEv07_count() : find it within "fse.h" */
+
+/*! HUFv07_readStats() :
+    Read compact Huffman tree, saved by HUFv07_writeCTable().
+    `huffWeight` is destination buffer.
+    @return : size read from `src` , or an error Code .
+    Note : Needed by HUFv07_readCTable() and HUFv07_readDTableXn() . */
+size_t HUFv07_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize);
+
+
+/*
+HUFv07_decompress() does the following:
+1. select the decompression algorithm (X2, X4) based on pre-computed heuristics
+2. build Huffman table from save, using HUFv07_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUFv07_decompressSXn_usingDTable
+*/
+
+/** HUFv07_selectDecoder() :
+*   Tells which decoder is likely to decode faster,
+*   based on a set of pre-determined metrics.
+*   @return : 0==HUFv07_decompress4X2, 1==HUFv07_decompress4X4 .
+*   Assumption : 0 < cSrcSize < dstSize <= 128 KB */
+U32 HUFv07_selectDecoder (size_t dstSize, size_t cSrcSize);
+
+size_t HUFv07_readDTableX2 (HUFv07_DTable* DTable, const void* src, size_t srcSize);
+size_t HUFv07_readDTableX4 (HUFv07_DTable* DTable, const void* src, size_t srcSize);
+
+size_t HUFv07_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+
+
+/* single stream variants */
+size_t HUFv07_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv07_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+
+size_t HUFv07_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+
+
+#endif /* HUFv07_STATIC_LINKING_ONLY */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUFv07_H_298734234 */
+/*
+   Common functions of New Generation Entropy library
+   Copyright (C) 2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+*************************************************************************** */
+
+
+
+/*-****************************************
+*  FSE Error Management
+******************************************/
+unsigned FSEv07_isError(size_t code) { return ERR_isError(code); }
+
+const char* FSEv07_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* **************************************************************
+*  HUF Error Management
+****************************************************************/
+unsigned HUFv07_isError(size_t code) { return ERR_isError(code); }
+
+const char* HUFv07_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/*-**************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSEv07_abs(short a) { return (short)(a<0 ? -a : a); }
+
+size_t FSEv07_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSEv07_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSEv07_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr)) {
+        if (previous0) {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF) {
+                n0+=24;
+                if (ip < iend-5) {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                } else {
+                    bitStream >>= 16;
+                    bitCount+=16;
+            }   }
+            while ((bitStream & 3) == 3) {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {   short const max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max) {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            } else {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSEv07_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold) {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+    }   }   /* while ((remaining>1) && (charnum<=*maxSVPtr)) */
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*! HUFv07_readStats() :
+    Read compact Huffman tree, saved by HUFv07_writeCTable().
+    `huffWeight` is destination buffer.
+    @return : size read from `src` , or an error Code .
+    Note : Needed by HUFv07_readCTable() and HUFv07_readDTableXn() .
+*/
+size_t HUFv07_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    /* memset(huffWeight, 0, hwSize); */   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  { /* special header */
+        if (iSize >= (242)) {  /* RLE */
+            static U32 l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else {   /* Incompressible */
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            {   U32 n;
+                for (n=0; n<oSize; n+=2) {
+                    huffWeight[n]   = ip[n/2] >> 4;
+                    huffWeight[n+1] = ip[n/2] & 15;
+    }   }   }   }
+    else  {   /* header compressed with FSE (normal case) */
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSEv07_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSEv07_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUFv07_TABLELOG_ABSOLUTEMAX + 1) * sizeof(U32));
+    weightTotal = 0;
+    {   U32 n; for (n=0; n<oSize; n++) {
+            if (huffWeight[n] >= HUFv07_TABLELOG_ABSOLUTEMAX) return ERROR(corruption_detected);
+            rankStats[huffWeight[n]]++;
+            weightTotal += (1 << huffWeight[n]) >> 1;
+    }   }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    {   U32 const tableLog = BITv07_highbit32(weightTotal) + 1;
+        if (tableLog > HUFv07_TABLELOG_ABSOLUTEMAX) return ERROR(corruption_detected);
+        *tableLogPtr = tableLog;
+        /* determine last weight */
+        {   U32 const total = 1 << tableLog;
+            U32 const rest = total - weightTotal;
+            U32 const verif = 1 << BITv07_highbit32(rest);
+            U32 const lastWeight = BITv07_highbit32(rest) + 1;
+            if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+            huffWeight[oSize] = (BYTE)lastWeight;
+            rankStats[lastWeight]++;
+    }   }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    return iSize+1;
+}
+/* ******************************************************************
+   FSE : Finite State Entropy decoder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSEv07_isError ERR_isError
+#define FSEv07_STATIC_ASSERT(c) { enum { FSEv07_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSEv07_DTABLE_SIZE_U32(FSEv07_MAX_TABLELOG)];
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSEv07_FUNCTION_EXTENSION
+#  error "FSEv07_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSEv07_FUNCTION_TYPE
+#  error "FSEv07_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSEv07_CAT(X,Y) X##Y
+#define FSEv07_FUNCTION_NAME(X,Y) FSEv07_CAT(X,Y)
+#define FSEv07_TYPE_NAME(X,Y) FSEv07_CAT(X,Y)
+
+
+/* Function templates */
+FSEv07_DTable* FSEv07_createDTable (unsigned tableLog)
+{
+    if (tableLog > FSEv07_TABLELOG_ABSOLUTE_MAX) tableLog = FSEv07_TABLELOG_ABSOLUTE_MAX;
+    return (FSEv07_DTable*)malloc( FSEv07_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
+}
+
+void FSEv07_freeDTable (FSEv07_DTable* dt)
+{
+    free(dt);
+}
+
+size_t FSEv07_buildDTable(FSEv07_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
+    FSEv07_DECODE_TYPE* const tableDecode = (FSEv07_DECODE_TYPE*) (tdPtr);
+    U16 symbolNext[FSEv07_MAX_SYMBOL_VALUE+1];
+
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+    U32 highThreshold = tableSize-1;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSEv07_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSEv07_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    {   FSEv07_DTableHeader DTableH;
+        DTableH.tableLog = (U16)tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].symbol = (FSEv07_FUNCTION_TYPE)s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    symbolNext[s] = normalizedCounter[s];
+        }   }   }
+        memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    {   U32 const tableMask = tableSize-1;
+        U32 const step = FSEv07_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            for (i=0; i<normalizedCounter[s]; i++) {
+                tableDecode[position].symbol = (FSEv07_FUNCTION_TYPE)s;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+
+        if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {   U32 u;
+        for (u=0; u<tableSize; u++) {
+            FSEv07_FUNCTION_TYPE const symbol = (FSEv07_FUNCTION_TYPE)(tableDecode[u].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BITv07_highbit32 ((U32)nextState) );
+            tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+    }   }
+
+    return 0;
+}
+
+
+
+#ifndef FSEv07_COMMONDEFS_ONLY
+
+/*-*******************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+size_t FSEv07_buildDTable_rle (FSEv07_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSEv07_DTableHeader* const DTableH = (FSEv07_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv07_decode_t* const cell = (FSEv07_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+size_t FSEv07_buildDTable_raw (FSEv07_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSEv07_DTableHeader* const DTableH = (FSEv07_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv07_decode_t* const dinfo = (FSEv07_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSV1 = tableMask+1;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<maxSV1; s++) {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSEv07_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSEv07_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BITv07_DStream_t bitD;
+    FSEv07_DState_t state1;
+    FSEv07_DState_t state2;
+
+    /* Init */
+    { size_t const errorCode = BITv07_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+      if (FSEv07_isError(errorCode)) return errorCode; }
+
+    FSEv07_initDState(&state1, &bitD, dt);
+    FSEv07_initDState(&state2, &bitD, dt);
+
+#define FSEv07_GETSYMBOL(statePtr) fast ? FSEv07_decodeSymbolFast(statePtr, &bitD) : FSEv07_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BITv07_reloadDStream(&bitD)==BITv07_DStream_unfinished) && (op<olimit) ; op+=4) {
+        op[0] = FSEv07_GETSYMBOL(&state1);
+
+        if (FSEv07_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv07_reloadDStream(&bitD);
+
+        op[1] = FSEv07_GETSYMBOL(&state2);
+
+        if (FSEv07_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BITv07_reloadDStream(&bitD) > BITv07_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSEv07_GETSYMBOL(&state1);
+
+        if (FSEv07_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv07_reloadDStream(&bitD);
+
+        op[3] = FSEv07_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BITv07_reloadDStream(&bitD) >= FSEv07_DStream_partiallyFilled; Ends at exactly BITv07_DStream_completed */
+    while (1) {
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv07_GETSYMBOL(&state1);
+
+        if (BITv07_reloadDStream(&bitD)==BITv07_DStream_overflow) {
+            *op++ = FSEv07_GETSYMBOL(&state2);
+            break;
+        }
+
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv07_GETSYMBOL(&state2);
+
+        if (BITv07_reloadDStream(&bitD)==BITv07_DStream_overflow) {
+            *op++ = FSEv07_GETSYMBOL(&state1);
+            break;
+    }   }
+
+    return op-ostart;
+}
+
+
+size_t FSEv07_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSEv07_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv07_DTableHeader* DTableH = (const FSEv07_DTableHeader*)ptr;
+    const U32 fastMode = DTableH->fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSEv07_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSEv07_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+size_t FSEv07_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSEv07_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSEv07_MAX_SYMBOL_VALUE;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    {   size_t const NCountLength = FSEv07_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+        if (FSEv07_isError(NCountLength)) return NCountLength;
+        if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+        ip += NCountLength;
+        cSrcSize -= NCountLength;
+    }
+
+    { size_t const errorCode = FSEv07_buildDTable (dt, counting, maxSymbolValue, tableLog);
+      if (FSEv07_isError(errorCode)) return errorCode; }
+
+    return FSEv07_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);   /* always return, even if it is an error code */
+}
+
+
+
+#endif   /* FSEv07_COMMONDEFS_ONLY */
+
+/* ******************************************************************
+   Huffman decoder, part of New Generation Entropy library
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUFv07_STATIC_ASSERT(c) { enum { HUFv07_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/*-***************************/
+/*  generic DTableDesc       */
+/*-***************************/
+
+typedef struct { BYTE maxTableLog; BYTE tableType; BYTE tableLog; BYTE reserved; } DTableDesc;
+
+static DTableDesc HUFv07_getDTableDesc(const HUFv07_DTable* table)
+{
+    DTableDesc dtd;
+    memcpy(&dtd, table, sizeof(dtd));
+    return dtd;
+}
+
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+
+typedef struct { BYTE byte; BYTE nbBits; } HUFv07_DEltX2;   /* single-symbol decoding */
+
+size_t HUFv07_readDTableX2 (HUFv07_DTable* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUFv07_SYMBOLVALUE_MAX + 1];
+    U32 rankVal[HUFv07_TABLELOG_ABSOLUTEMAX + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    U32 nbSymbols = 0;
+    size_t iSize;
+    void* const dtPtr = DTable + 1;
+    HUFv07_DEltX2* const dt = (HUFv07_DEltX2*)dtPtr;
+
+    HUFv07_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUFv07_DTable));
+    /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv07_readStats(huffWeight, HUFv07_SYMBOLVALUE_MAX + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv07_isError(iSize)) return iSize;
+
+    /* Table header */
+    {   DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+        if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge);   /* DTable too small, huffman tree cannot fit in */
+        dtd.tableType = 0;
+        dtd.tableLog = (BYTE)tableLog;
+        memcpy(DTable, &dtd, sizeof(dtd));
+    }
+
+    /* Prepare ranks */
+    {   U32 n, nextRankStart = 0;
+        for (n=1; n<tableLog+1; n++) {
+            U32 current = nextRankStart;
+            nextRankStart += (rankVal[n] << (n-1));
+            rankVal[n] = current;
+    }   }
+
+    /* fill DTable */
+    {   U32 n;
+        for (n=0; n<nbSymbols; n++) {
+            U32 const w = huffWeight[n];
+            U32 const length = (1 << w) >> 1;
+            U32 i;
+            HUFv07_DEltX2 D;
+            D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+            for (i = rankVal[w]; i < rankVal[w] + length; i++)
+                dt[i] = D;
+            rankVal[w] += length;
+    }   }
+
+    return iSize;
+}
+
+
+static BYTE HUFv07_decodeSymbolX2(BITv07_DStream_t* Dstream, const HUFv07_DEltX2* dt, const U32 dtLog)
+{
+    size_t const val = BITv07_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+    BYTE const c = dt[val].byte;
+    BITv07_skipBits(Dstream, dt[val].nbBits);
+    return c;
+}
+
+#define HUFv07_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUFv07_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUFv07_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv07_TABLELOG_MAX<=12)) \
+        HUFv07_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUFv07_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUFv07_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUFv07_decodeStreamX2(BYTE* p, BITv07_DStream_t* const bitDPtr, BYTE* const pEnd, const HUFv07_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p <= pEnd-4)) {
+        HUFv07_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p < pEnd))
+        HUFv07_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUFv07_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+static size_t HUFv07_decompress1X2_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + dstSize;
+    const void* dtPtr = DTable + 1;
+    const HUFv07_DEltX2* const dt = (const HUFv07_DEltX2*)dtPtr;
+    BITv07_DStream_t bitD;
+    DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+    U32 const dtLog = dtd.tableLog;
+
+    { size_t const errorCode = BITv07_initDStream(&bitD, cSrc, cSrcSize);
+      if (HUFv07_isError(errorCode)) return errorCode; }
+
+    HUFv07_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv07_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+size_t HUFv07_decompress1X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 0) return ERROR(GENERIC);
+    return HUFv07_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+size_t HUFv07_decompress1X2_DCtx (HUFv07_DTable* DCtx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv07_readDTableX2 (DCtx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress1X2_usingDTable_internal (dst, dstSize, ip, cSrcSize, DCtx);
+}
+
+size_t HUFv07_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX2(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress1X2_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+
+static size_t HUFv07_decompress4X2_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);  /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable + 1;
+        const HUFv07_DEltX2* const dt = (const HUFv07_DEltX2*)dtPtr;
+
+        /* Init */
+        BITv07_DStream_t bitD1;
+        BITv07_DStream_t bitD2;
+        BITv07_DStream_t bitD3;
+        BITv07_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+        DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        { size_t const errorCode = BITv07_initDStream(&bitD1, istart1, length1);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD2, istart2, length2);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD3, istart3, length3);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD4, istart4, length4);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv07_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv07_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_0(op4, &bitD4);
+            endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv07_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv07_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv07_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv07_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv07_endOfDStream(&bitD1) & BITv07_endOfDStream(&bitD2) & BITv07_endOfDStream(&bitD3) & BITv07_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv07_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 0) return ERROR(GENERIC);
+    return HUFv07_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+
+size_t HUFv07_decompress4X2_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv07_readDTableX2 (dctx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress4X2_usingDTable_internal (dst, dstSize, ip, cSrcSize, dctx);
+}
+
+size_t HUFv07_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX2(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUFv07_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+static void HUFv07_fillDTableX4Level2(HUFv07_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUFv07_DEltX4 DElt;
+    U32 rankVal[HUFv07_TABLELOG_ABSOLUTEMAX + 1];
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1) {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    { U32 s; for (s=0; s<sortedListSize; s++) {   /* note : sortedSymbols already skipped */
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }}
+}
+
+typedef U32 rankVal_t[HUFv07_TABLELOG_ABSOLUTEMAX][HUFv07_TABLELOG_ABSOLUTEMAX + 1];
+
+static void HUFv07_fillDTableX4(HUFv07_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUFv07_TABLELOG_ABSOLUTEMAX + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits) {   /* enough room for a second symbol */
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUFv07_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        } else {
+            HUFv07_DEltX4 DElt;
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits = (BYTE)(nbBits);
+            DElt.length = 1;
+            {   U32 u;
+                const U32 end = start + length;
+                for (u = start; u < end; u++) DTable[u] = DElt;
+        }   }
+        rankVal[weight] += length;
+    }
+}
+
+size_t HUFv07_readDTableX4 (HUFv07_DTable* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUFv07_SYMBOLVALUE_MAX + 1];
+    sortedSymbol_t sortedSymbol[HUFv07_SYMBOLVALUE_MAX + 1];
+    U32 rankStats[HUFv07_TABLELOG_ABSOLUTEMAX + 1] = { 0 };
+    U32 rankStart0[HUFv07_TABLELOG_ABSOLUTEMAX + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    U32 const maxTableLog = dtd.maxTableLog;
+    size_t iSize;
+    void* dtPtr = DTable+1;   /* force compiler to avoid strict-aliasing */
+    HUFv07_DEltX4* const dt = (HUFv07_DEltX4*)dtPtr;
+
+    HUFv07_STATIC_ASSERT(sizeof(HUFv07_DEltX4) == sizeof(HUFv07_DTable));   /* if compilation fails here, assertion is false */
+    if (maxTableLog > HUFv07_TABLELOG_ABSOLUTEMAX) return ERROR(tableLog_tooLarge);
+    /* memset(weightList, 0, sizeof(weightList)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv07_readStats(weightList, HUFv07_SYMBOLVALUE_MAX + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv07_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > maxTableLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {   U32 w, nextRankStart = 0;
+        for (w=1; w<maxW+1; w++) {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {   U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 const w = weightList[s];
+            U32 const r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {   U32* const rankVal0 = rankVal[0];
+        {   int const rescale = (maxTableLog-tableLog) - 1;   /* tableLog <= maxTableLog */
+            U32 nextRankVal = 0;
+            U32 w;
+            for (w=1; w<maxW+1; w++) {
+                U32 current = nextRankVal;
+                nextRankVal += rankStats[w] << (w+rescale);
+                rankVal0[w] = current;
+        }   }
+        {   U32 const minBits = tableLog+1 - maxW;
+            U32 consumed;
+            for (consumed = minBits; consumed < maxTableLog - minBits + 1; consumed++) {
+                U32* const rankValPtr = rankVal[consumed];
+                U32 w;
+                for (w = 1; w < maxW+1; w++) {
+                    rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }   }
+
+    HUFv07_fillDTableX4(dt, maxTableLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    dtd.tableLog = (BYTE)maxTableLog;
+    dtd.tableType = 1;
+    memcpy(DTable, &dtd, sizeof(dtd));
+    return iSize;
+}
+
+
+static U32 HUFv07_decodeSymbolX4(void* op, BITv07_DStream_t* DStream, const HUFv07_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv07_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BITv07_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUFv07_decodeLastSymbolX4(void* op, BITv07_DStream_t* DStream, const HUFv07_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv07_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BITv07_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BITv07_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }   }
+    return 1;
+}
+
+
+#define HUFv07_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUFv07_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv07_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv07_TABLELOG_MAX<=12)) \
+        ptr += HUFv07_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv07_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUFv07_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUFv07_decodeStreamX4(BYTE* p, BITv07_DStream_t* bitDPtr, BYTE* const pEnd, const HUFv07_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p < pEnd-7)) {
+        HUFv07_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to end : up to 2 symbols at a time */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p <= pEnd-2))
+        HUFv07_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUFv07_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUFv07_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+static size_t HUFv07_decompress1X4_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    BITv07_DStream_t bitD;
+
+    /* Init */
+    {   size_t const errorCode = BITv07_initDStream(&bitD, cSrc, cSrcSize);
+        if (HUFv07_isError(errorCode)) return errorCode;
+    }
+
+    /* decode */
+    {   BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
+        const HUFv07_DEltX4* const dt = (const HUFv07_DEltX4*)dtPtr;
+        DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+        HUFv07_decodeStreamX4(ostart, &bitD, oend, dt, dtd.tableLog);
+    }
+
+    /* check */
+    if (!BITv07_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+size_t HUFv07_decompress1X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 1) return ERROR(GENERIC);
+    return HUFv07_decompress1X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+size_t HUFv07_decompress1X4_DCtx (HUFv07_DTable* DCtx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv07_readDTableX4 (DCtx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress1X4_usingDTable_internal (dst, dstSize, ip, cSrcSize, DCtx);
+}
+
+size_t HUFv07_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX4(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress1X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+static size_t HUFv07_decompress4X4_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;
+        const HUFv07_DEltX4* const dt = (const HUFv07_DEltX4*)dtPtr;
+
+        /* Init */
+        BITv07_DStream_t bitD1;
+        BITv07_DStream_t bitD2;
+        BITv07_DStream_t bitD3;
+        BITv07_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+        DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        { size_t const errorCode = BITv07_initDStream(&bitD1, istart1, length1);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD2, istart2, length2);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD3, istart3, length3);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD4, istart4, length4);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv07_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv07_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv07_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv07_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv07_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv07_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        { U32 const endCheck = BITv07_endOfDStream(&bitD1) & BITv07_endOfDStream(&bitD2) & BITv07_endOfDStream(&bitD3) & BITv07_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv07_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 1) return ERROR(GENERIC);
+    return HUFv07_decompress4X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+
+size_t HUFv07_decompress4X4_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv07_readDTableX4 (dctx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress4X4_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx);
+}
+
+size_t HUFv07_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX4(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress4X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+
+/* ********************************/
+/* Generic decompression selector */
+/* ********************************/
+
+size_t HUFv07_decompress1X_usingDTable(void* dst, size_t maxDstSize,
+                                    const void* cSrc, size_t cSrcSize,
+                                    const HUFv07_DTable* DTable)
+{
+    DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+    return dtd.tableType ? HUFv07_decompress1X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable) :
+                           HUFv07_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable);
+}
+
+size_t HUFv07_decompress4X_usingDTable(void* dst, size_t maxDstSize,
+                                    const void* cSrc, size_t cSrcSize,
+                                    const HUFv07_DTable* DTable)
+{
+    DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+    return dtd.tableType ? HUFv07_decompress4X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable) :
+                           HUFv07_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable);
+}
+
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+/** HUFv07_selectDecoder() :
+*   Tells which decoder is likely to decode faster,
+*   based on a set of pre-determined metrics.
+*   @return : 0==HUFv07_decompress4X2, 1==HUFv07_decompress4X4 .
+*   Assumption : 0 < cSrcSize < dstSize <= 128 KB */
+U32 HUFv07_selectDecoder (size_t dstSize, size_t cSrcSize)
+{
+    /* decoder timing evaluation */
+    U32 const Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    U32 const D256 = (U32)(dstSize >> 8);
+    U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
+    U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
+    DTime1 += DTime1 >> 3;  /* advantage to algorithm using less memory, for cache eviction */
+
+    return DTime1 < DTime0;
+}
+
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+size_t HUFv07_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[2] = { HUFv07_decompress4X2, HUFv07_decompress4X4 };
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+    }
+
+    /* return HUFv07_decompress4X2(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams single-symbol decoding */
+    /* return HUFv07_decompress4X4(dst, dstSize, cSrc, cSrcSize); */   /* multi-streams double-symbols decoding */
+}
+
+size_t HUFv07_decompress4X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUFv07_decompress4X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUFv07_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+    }
+}
+
+size_t HUFv07_decompress4X_hufOnly (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if ((cSrcSize >= dstSize) || (cSrcSize <= 1)) return ERROR(corruption_detected);   /* invalid */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUFv07_decompress4X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUFv07_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+    }
+}
+
+size_t HUFv07_decompress1X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUFv07_decompress1X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUFv07_decompress1X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+    }
+}
+/*
+    Common functions of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd/
+*/
+
+
+
+/*-****************************************
+*  ZSTD Error Management
+******************************************/
+/*! ZSTDv07_isError() :
+*   tells if a return value is an error code */
+unsigned ZSTDv07_isError(size_t code) { return ERR_isError(code); }
+
+/*! ZSTDv07_getErrorName() :
+*   provides error code string from function result (useful for debugging) */
+const char* ZSTDv07_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+
+/* **************************************************************
+*  ZBUFF Error Management
+****************************************************************/
+unsigned ZBUFFv07_isError(size_t errorCode) { return ERR_isError(errorCode); }
+
+const char* ZBUFFv07_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+
+
+static void* ZSTDv07_defaultAllocFunction(void* opaque, size_t size)
+{
+    void* address = malloc(size);
+    (void)opaque;
+    /* printf("alloc %p, %d opaque=%p \n", address, (int)size, opaque); */
+    return address;
+}
+
+static void ZSTDv07_defaultFreeFunction(void* opaque, void* address)
+{
+    (void)opaque;
+    /* if (address) printf("free %p opaque=%p \n", address, opaque); */
+    free(address);
+}
+/*
+    zstd_internal - common functions to include
+    Header File for include
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://www.zstd.net
+*/
+#ifndef ZSTDv07_CCOMMON_H_MODULE
+#define ZSTDv07_CCOMMON_H_MODULE
+
+
+/*-*************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/*-*************************************
+*  Common constants
+***************************************/
+#define ZSTDv07_OPT_NUM    (1<<12)
+#define ZSTDv07_DICT_MAGIC  0xEC30A437   /* v0.7 */
+
+#define ZSTDv07_REP_NUM    3
+#define ZSTDv07_REP_INIT   ZSTDv07_REP_NUM
+#define ZSTDv07_REP_MOVE   (ZSTDv07_REP_NUM-1)
+static const U32 repStartValue[ZSTDv07_REP_NUM] = { 1, 4, 8 };
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define ZSTDv07_WINDOWLOG_ABSOLUTEMIN 10
+static const size_t ZSTDv07_fcs_fieldSize[4] = { 0, 2, 4, 8 };
+static const size_t ZSTDv07_did_fieldSize[4] = { 0, 1, 2, 4 };
+
+#define ZSTDv07_BLOCKHEADERSIZE 3   /* C standard doesn't allow `static const` variable to be init using another `static const` variable */
+static const size_t ZSTDv07_blockHeaderSize = ZSTDv07_BLOCKHEADERSIZE;
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
+typedef enum { lbt_huffman, lbt_repeat, lbt_raw, lbt_rle } litBlockType_t;
+
+#define LONGNBSEQ 0x7F00
+
+#define MINMATCH 3
+#define EQUAL_READ32 4
+
+#define Litbits  8
+#define MaxLit ((1<<Litbits) - 1)
+#define MaxML  52
+#define MaxLL  35
+#define MaxOff 28
+#define MaxSeq MAX(MaxLL, MaxML)   /* Assumption : MaxOff < MaxLL,MaxML */
+#define MLFSELog    9
+#define LLFSELog    9
+#define OffFSELog   8
+
+#define FSEv07_ENCODING_RAW     0
+#define FSEv07_ENCODING_RLE     1
+#define FSEv07_ENCODING_STATIC  2
+#define FSEv07_ENCODING_DYNAMIC 3
+
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+
+static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9,10,11,12,
+                                     13,14,15,16 };
+static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1,
+                                             2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1,
+                                            -1,-1,-1,-1 };
+static const U32 LL_defaultNormLog = 6;
+
+static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 7, 8, 9,10,11,
+                                     12,13,14,15,16 };
+static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,-1,-1,
+                                            -1,-1,-1,-1,-1 };
+static const U32 ML_defaultNormLog = 6;
+
+static const S16 OF_defaultNorm[MaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                              1, 1, 1, 1, 1, 1, 1, 1,-1,-1,-1,-1,-1 };
+static const U32 OF_defaultNormLog = 5;
+
+
+/*-*******************************************
+*  Shared functions to include for inlining
+*********************************************/
+static void ZSTDv07_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+#define COPY8(d,s) { ZSTDv07_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTDv07_wildcopy() :
+*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
+#define WILDCOPY_OVERLENGTH 8
+MEM_STATIC void ZSTDv07_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+/*-*******************************************
+*  Private interfaces
+*********************************************/
+typedef struct ZSTDv07_stats_s ZSTDv07_stats_t;
+
+typedef struct {
+    U32 off;
+    U32 len;
+} ZSTDv07_match_t;
+
+typedef struct {
+    U32 price;
+    U32 off;
+    U32 mlen;
+    U32 litlen;
+    U32 rep[ZSTDv07_REP_INIT];
+} ZSTDv07_optimal_t;
+
+struct ZSTDv07_stats_s { U32 unused; };
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* litStart;
+    BYTE* lit;
+    U16*  litLengthStart;
+    U16*  litLength;
+    BYTE* llCodeStart;
+    U16*  matchLengthStart;
+    U16*  matchLength;
+    BYTE* mlCodeStart;
+    U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
+    U32   longLengthPos;
+    /* opt */
+    ZSTDv07_optimal_t* priceTable;
+    ZSTDv07_match_t* matchTable;
+    U32* matchLengthFreq;
+    U32* litLengthFreq;
+    U32* litFreq;
+    U32* offCodeFreq;
+    U32  matchLengthSum;
+    U32  matchSum;
+    U32  litLengthSum;
+    U32  litSum;
+    U32  offCodeSum;
+    U32  log2matchLengthSum;
+    U32  log2matchSum;
+    U32  log2litLengthSum;
+    U32  log2litSum;
+    U32  log2offCodeSum;
+    U32  factor;
+    U32  cachedPrice;
+    U32  cachedLitLength;
+    const BYTE* cachedLiterals;
+    ZSTDv07_stats_t stats;
+} SeqStore_t;
+
+void ZSTDv07_seqToCodes(const SeqStore_t* seqStorePtr, size_t const nbSeq);
+
+/* custom memory allocation functions */
+static const ZSTDv07_customMem defaultCustomMem = { ZSTDv07_defaultAllocFunction, ZSTDv07_defaultFreeFunction, NULL };
+
+#endif   /* ZSTDv07_CCOMMON_H_MODULE */
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTDv07_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTDv07_HEAPMODE
+#  define ZSTDv07_HEAPMODE 1
+#endif
+
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#  pragma warning(disable : 4100)        /* disable: C4100: unreferenced formal parameter */
+#endif
+
+
+/*-*************************************
+*  Macros
+***************************************/
+#define ZSTDv07_isError ERR_isError   /* for inlining */
+#define FSEv07_isError  ERR_isError
+#define HUFv07_isError  ERR_isError
+
+
+/*_*******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTDv07_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/*-*************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock,
+               ZSTDds_decodeSkippableHeader, ZSTDds_skipFrame } ZSTDv07_dStage;
+
+struct ZSTDv07_DCtx_s
+{
+    FSEv07_DTable LLTable[FSEv07_DTABLE_SIZE_U32(LLFSELog)];
+    FSEv07_DTable OffTable[FSEv07_DTABLE_SIZE_U32(OffFSELog)];
+    FSEv07_DTable MLTable[FSEv07_DTABLE_SIZE_U32(MLFSELog)];
+    HUFv07_DTable hufTable[HUFv07_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];  /* can accommodate HUFv07_decompress4X */
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    U32 rep[3];
+    ZSTDv07_frameParams fParams;
+    blockType_t bType;   /* used in ZSTDv07_decompressContinue(), to transfer blockType between header decoding and block decoding stages */
+    ZSTDv07_dStage stage;
+    U32 litEntropy;
+    U32 fseEntropy;
+    XXH64_state_t xxhState;
+    size_t headerSize;
+    U32 dictID;
+    const BYTE* litPtr;
+    ZSTDv07_customMem customMem;
+    size_t litSize;
+    BYTE litBuffer[ZSTDv07_BLOCKSIZE_ABSOLUTEMAX + WILDCOPY_OVERLENGTH];
+    BYTE headerBuffer[ZSTDv07_FRAMEHEADERSIZE_MAX];
+};  /* typedef'd to ZSTDv07_DCtx within "zstd_static.h" */
+
+int ZSTDv07_isSkipFrame(ZSTDv07_DCtx* dctx);
+
+size_t ZSTDv07_sizeofDCtx (const ZSTDv07_DCtx* dctx) { return sizeof(*dctx); }
+
+size_t ZSTDv07_estimateDCtxSize(void) { return sizeof(ZSTDv07_DCtx); }
+
+size_t ZSTDv07_decompressBegin(ZSTDv07_DCtx* dctx)
+{
+    dctx->expected = ZSTDv07_frameHeaderSize_min;
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    dctx->hufTable[0] = (HUFv07_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001);
+    dctx->litEntropy = dctx->fseEntropy = 0;
+    dctx->dictID = 0;
+    { int i; for (i=0; i<ZSTDv07_REP_NUM; i++) dctx->rep[i] = repStartValue[i]; }
+    return 0;
+}
+
+ZSTDv07_DCtx* ZSTDv07_createDCtx_advanced(ZSTDv07_customMem customMem)
+{
+    ZSTDv07_DCtx* dctx;
+
+    if (!customMem.customAlloc && !customMem.customFree)
+        customMem = defaultCustomMem;
+
+    if (!customMem.customAlloc || !customMem.customFree)
+        return NULL;
+
+    dctx = (ZSTDv07_DCtx*) customMem.customAlloc(customMem.opaque, sizeof(ZSTDv07_DCtx));
+    if (!dctx) return NULL;
+    memcpy(&dctx->customMem, &customMem, sizeof(ZSTDv07_customMem));
+    ZSTDv07_decompressBegin(dctx);
+    return dctx;
+}
+
+ZSTDv07_DCtx* ZSTDv07_createDCtx(void)
+{
+    return ZSTDv07_createDCtx_advanced(defaultCustomMem);
+}
+
+size_t ZSTDv07_freeDCtx(ZSTDv07_DCtx* dctx)
+{
+    if (dctx==NULL) return 0;   /* support free on NULL */
+    dctx->customMem.customFree(dctx->customMem.opaque, dctx);
+    return 0;   /* reserved as a potential error code in the future */
+}
+
+void ZSTDv07_copyDCtx(ZSTDv07_DCtx* dstDCtx, const ZSTDv07_DCtx* srcDCtx)
+{
+    memcpy(dstDCtx, srcDCtx,
+           sizeof(ZSTDv07_DCtx) - (ZSTDv07_BLOCKSIZE_ABSOLUTEMAX+WILDCOPY_OVERLENGTH + ZSTDv07_frameHeaderSize_max));  /* no need to copy workspace */
+}
+
+
+/*-*************************************************************
+*   Decompression section
+***************************************************************/
+
+/* Frame format description
+   Frame Header -  [ Block Header - Block ] - Frame End
+   1) Frame Header
+      - 4 bytes - Magic Number : ZSTDv07_MAGICNUMBER (defined within zstd.h)
+      - 1 byte  - Frame Descriptor
+   2) Block Header
+      - 3 bytes, starting with a 2-bits descriptor
+                 Uncompressed, Compressed, Frame End, unused
+   3) Block
+      See Block Format Description
+   4) Frame End
+      - 3 bytes, compatible with Block Header
+*/
+
+
+/* Frame Header :
+
+   1 byte - FrameHeaderDescription :
+   bit 0-1 : dictID (0, 1, 2 or 4 bytes)
+   bit 2   : checksumFlag
+   bit 3   : reserved (must be zero)
+   bit 4   : reserved (unused, can be any value)
+   bit 5   : Single Segment (if 1, WindowLog byte is not present)
+   bit 6-7 : FrameContentFieldSize (0, 2, 4, or 8)
+             if (SkippedWindowLog && !FrameContentFieldsize) FrameContentFieldsize=1;
+
+   Optional : WindowLog (0 or 1 byte)
+   bit 0-2 : octal Fractional (1/8th)
+   bit 3-7 : Power of 2, with 0 = 1 KB (up to 2 TB)
+
+   Optional : dictID (0, 1, 2 or 4 bytes)
+   Automatic adaptation
+   0 : no dictID
+   1 : 1 - 255
+   2 : 256 - 65535
+   4 : all other values
+
+   Optional : content size (0, 1, 2, 4 or 8 bytes)
+   0 : unknown          (fcfs==0 and swl==0)
+   1 : 0-255 bytes      (fcfs==0 and swl==1)
+   2 : 256 - 65535+256  (fcfs==1)
+   4 : 0 - 4GB-1        (fcfs==2)
+   8 : 0 - 16EB-1       (fcfs==3)
+*/
+
+
+/* Compressed Block, format description
+
+   Block = Literal Section - Sequences Section
+   Prerequisite : size of (compressed) block, maximum size of regenerated data
+
+   1) Literal Section
+
+   1.1) Header : 1-5 bytes
+        flags: 2 bits
+            00 compressed by Huff0
+            01 unused
+            10 is Raw (uncompressed)
+            11 is Rle
+            Note : using 01 => Huff0 with precomputed table ?
+            Note : delta map ? => compressed ?
+
+   1.1.1) Huff0-compressed literal block : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+   1.1.2) Raw (uncompressed) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RAW<<6) + (0<<4) + size
+               12 bits: (IS_RAW<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RAW<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.3) Rle (repeated single byte) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RLE<<6) + (0<<4) + size
+               12 bits: (IS_RLE<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RLE<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.4) Huff0-compressed literal block, using precomputed CTables : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+        1- CTable available (stored into workspace ?)
+        2- Small input (fast heuristic ? Full comparison ? depend on clevel ?)
+
+
+   1.2) Literal block content
+
+   1.2.1) Huff0 block, using sizes from header
+        See Huff0 format
+
+   1.2.2) Huff0 block, using prepared table
+
+   1.2.3) Raw content
+
+   1.2.4) single byte
+
+
+   2) Sequences section
+      TO DO
+*/
+
+/** ZSTDv07_frameHeaderSize() :
+*   srcSize must be >= ZSTDv07_frameHeaderSize_min.
+*   @return : size of the Frame Header */
+static size_t ZSTDv07_frameHeaderSize(const void* src, size_t srcSize)
+{
+    if (srcSize < ZSTDv07_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    {   BYTE const fhd = ((const BYTE*)src)[4];
+        U32 const dictID= fhd & 3;
+        U32 const directMode = (fhd >> 5) & 1;
+        U32 const fcsId = fhd >> 6;
+        return ZSTDv07_frameHeaderSize_min + !directMode + ZSTDv07_did_fieldSize[dictID] + ZSTDv07_fcs_fieldSize[fcsId]
+                + (directMode && !ZSTDv07_fcs_fieldSize[fcsId]);
+    }
+}
+
+
+/** ZSTDv07_getFrameParams() :
+*   decode Frame Header, or require larger `srcSize`.
+*   @return : 0, `fparamsPtr` is correctly filled,
+*            >0, `srcSize` is too small, result is expected `srcSize`,
+*             or an error code, which can be tested using ZSTDv07_isError() */
+size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize < ZSTDv07_frameHeaderSize_min) return ZSTDv07_frameHeaderSize_min;
+    memset(fparamsPtr, 0, sizeof(*fparamsPtr));
+    if (MEM_readLE32(src) != ZSTDv07_MAGICNUMBER) {
+        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTDv07_MAGIC_SKIPPABLE_START) {
+            if (srcSize < ZSTDv07_skippableHeaderSize) return ZSTDv07_skippableHeaderSize; /* magic number + skippable frame length */
+            fparamsPtr->frameContentSize = MEM_readLE32((const char *)src + 4);
+            fparamsPtr->windowSize = 0; /* windowSize==0 means a frame is skippable */
+            return 0;
+        }
+        return ERROR(prefix_unknown);
+    }
+
+    /* ensure there is enough `srcSize` to fully read/decode frame header */
+    { size_t const fhsize = ZSTDv07_frameHeaderSize(src, srcSize);
+      if (srcSize < fhsize) return fhsize; }
+
+    {   BYTE const fhdByte = ip[4];
+        size_t pos = 5;
+        U32 const dictIDSizeCode = fhdByte&3;
+        U32 const checksumFlag = (fhdByte>>2)&1;
+        U32 const directMode = (fhdByte>>5)&1;
+        U32 const fcsID = fhdByte>>6;
+        U32 const windowSizeMax = 1U << ZSTDv07_WINDOWLOG_MAX;
+        U32 windowSize = 0;
+        U32 dictID = 0;
+        U64 frameContentSize = 0;
+        if ((fhdByte & 0x08) != 0)   /* reserved bits, which must be zero */
+            return ERROR(frameParameter_unsupported);
+        if (!directMode) {
+            BYTE const wlByte = ip[pos++];
+            U32 const windowLog = (wlByte >> 3) + ZSTDv07_WINDOWLOG_ABSOLUTEMIN;
+            if (windowLog > ZSTDv07_WINDOWLOG_MAX)
+                return ERROR(frameParameter_unsupported);
+            windowSize = (1U << windowLog);
+            windowSize += (windowSize >> 3) * (wlByte&7);
+        }
+
+        switch(dictIDSizeCode)
+        {
+            default:   /* impossible */
+            case 0 : break;
+            case 1 : dictID = ip[pos]; pos++; break;
+            case 2 : dictID = MEM_readLE16(ip+pos); pos+=2; break;
+            case 3 : dictID = MEM_readLE32(ip+pos); pos+=4; break;
+        }
+        switch(fcsID)
+        {
+            default:   /* impossible */
+            case 0 : if (directMode) frameContentSize = ip[pos]; break;
+            case 1 : frameContentSize = MEM_readLE16(ip+pos)+256; break;
+            case 2 : frameContentSize = MEM_readLE32(ip+pos); break;
+            case 3 : frameContentSize = MEM_readLE64(ip+pos); break;
+        }
+        if (!windowSize) windowSize = (U32)frameContentSize;
+        if (windowSize > windowSizeMax)
+            return ERROR(frameParameter_unsupported);
+        fparamsPtr->frameContentSize = frameContentSize;
+        fparamsPtr->windowSize = windowSize;
+        fparamsPtr->dictID = dictID;
+        fparamsPtr->checksumFlag = checksumFlag;
+    }
+    return 0;
+}
+
+
+/** ZSTDv07_getDecompressedSize() :
+*   compatible with legacy mode
+*   @return : decompressed size if known, 0 otherwise
+              note : 0 can mean any of the following :
+                   - decompressed size is not provided within frame header
+                   - frame header unknown / not supported
+                   - frame header not completely provided (`srcSize` too small) */
+unsigned long long ZSTDv07_getDecompressedSize(const void* src, size_t srcSize)
+{
+    ZSTDv07_frameParams fparams;
+    size_t const frResult = ZSTDv07_getFrameParams(&fparams, src, srcSize);
+    if (frResult!=0) return 0;
+    return fparams.frameContentSize;
+}
+
+
+/** ZSTDv07_decodeFrameHeader() :
+*   `srcSize` must be the size provided by ZSTDv07_frameHeaderSize().
+*   @return : 0 if success, or an error code, which can be tested using ZSTDv07_isError() */
+static size_t ZSTDv07_decodeFrameHeader(ZSTDv07_DCtx* dctx, const void* src, size_t srcSize)
+{
+    size_t const result = ZSTDv07_getFrameParams(&(dctx->fParams), src, srcSize);
+    if (dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID)) return ERROR(dictionary_wrong);
+    if (dctx->fParams.checksumFlag) XXH64_reset(&dctx->xxhState, 0);
+    return result;
+}
+
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+/*! ZSTDv07_getcBlockSize() :
+*   Provides the size of compressed block from block header `src` */
+static size_t ZSTDv07_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE*)src;
+    U32 cSize;
+
+    if (srcSize < ZSTDv07_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    bpPtr->blockType = (blockType_t)((*in) >> 6);
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+
+static size_t ZSTDv07_copyRawBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
+    if (srcSize > 0) {
+        memcpy(dst, src, srcSize);
+    }
+    return srcSize;
+}
+
+
+/*! ZSTDv07_decodeLiteralsBlock() :
+    @return : nb of bytes read from src (< srcSize ) */
+static size_t ZSTDv07_decodeLiteralsBlock(ZSTDv07_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch((litBlockType_t)(istart[0]>> 6))
+    {
+    case lbt_huffman:
+        {   size_t litSize, litCSize, singleStream=0;
+            U32 lhSize = (istart[0] >> 4) & 3;
+            if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for lhSize, + cSize (+nbSeq) */
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                /* 2 - 2 - 10 - 10 */
+                lhSize=3;
+                singleStream = istart[0] & 16;
+                litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+                litCSize = ((istart[1] &  3) << 8) + istart[2];
+                break;
+            case 2:
+                /* 2 - 2 - 14 - 14 */
+                lhSize=4;
+                litSize  = ((istart[0] & 15) << 10) + (istart[1] << 2) + (istart[2] >> 6);
+                litCSize = ((istart[2] & 63) <<  8) + istart[3];
+                break;
+            case 3:
+                /* 2 - 2 - 18 - 18 */
+                lhSize=5;
+                litSize  = ((istart[0] & 15) << 14) + (istart[1] << 6) + (istart[2] >> 2);
+                litCSize = ((istart[2] &  3) << 16) + (istart[3] << 8) + istart[4];
+                break;
+            }
+            if (litSize > ZSTDv07_BLOCKSIZE_ABSOLUTEMAX) return ERROR(corruption_detected);
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            if (HUFv07_isError(singleStream ?
+                            HUFv07_decompress1X2_DCtx(dctx->hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize) :
+                            HUFv07_decompress4X_hufOnly (dctx->hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize) ))
+                return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            dctx->litEntropy = 1;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case lbt_repeat:
+        {   size_t litSize, litCSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (lhSize != 1)  /* only case supported for now : small litSize, single stream */
+                return ERROR(corruption_detected);
+            if (dctx->litEntropy==0)
+                return ERROR(dictionary_corrupted);
+
+            /* 2 - 2 - 10 - 10 */
+            lhSize=3;
+            litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+            litCSize = ((istart[1] &  3) << 8) + istart[2];
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            {   size_t const errorCode = HUFv07_decompress1X4_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->hufTable);
+                if (HUFv07_isError(errorCode)) return ERROR(corruption_detected);
+            }
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case lbt_raw:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize=1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                break;
+            }
+
+            if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                return lhSize+litSize;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+lhSize;
+            dctx->litSize = litSize;
+            return lhSize+litSize;
+        }
+    case lbt_rle:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize = 1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
+                break;
+            }
+            if (litSize > ZSTDv07_BLOCKSIZE_ABSOLUTEMAX) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return lhSize+1;
+        }
+    default:
+        return ERROR(corruption_detected);   /* impossible */
+    }
+}
+
+
+/*! ZSTDv07_buildSeqTable() :
+    @return : nb bytes read from src,
+              or an error code if it fails, testable with ZSTDv07_isError()
+*/
+static size_t ZSTDv07_buildSeqTable(FSEv07_DTable* DTable, U32 type, U32 max, U32 maxLog,
+                                 const void* src, size_t srcSize,
+                                 const S16* defaultNorm, U32 defaultLog, U32 flagRepeatTable)
+{
+    switch(type)
+    {
+    case FSEv07_ENCODING_RLE :
+        if (!srcSize) return ERROR(srcSize_wrong);
+        if ( (*(const BYTE*)src) > max) return ERROR(corruption_detected);
+        FSEv07_buildDTable_rle(DTable, *(const BYTE*)src);   /* if *src > max, data is corrupted */
+        return 1;
+    case FSEv07_ENCODING_RAW :
+        FSEv07_buildDTable(DTable, defaultNorm, max, defaultLog);
+        return 0;
+    case FSEv07_ENCODING_STATIC:
+        if (!flagRepeatTable) return ERROR(corruption_detected);
+        return 0;
+    default :   /* impossible */
+    case FSEv07_ENCODING_DYNAMIC :
+        {   U32 tableLog;
+            S16 norm[MaxSeq+1];
+            size_t const headerSize = FSEv07_readNCount(norm, &max, &tableLog, src, srcSize);
+            if (FSEv07_isError(headerSize)) return ERROR(corruption_detected);
+            if (tableLog > maxLog) return ERROR(corruption_detected);
+            FSEv07_buildDTable(DTable, norm, max, tableLog);
+            return headerSize;
+    }   }
+}
+
+
+static size_t ZSTDv07_decodeSeqHeaders(int* nbSeqPtr,
+                             FSEv07_DTable* DTableLL, FSEv07_DTable* DTableML, FSEv07_DTable* DTableOffb, U32 flagRepeatTable,
+                             const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip = istart;
+
+    /* check */
+    if (srcSize < MIN_SEQUENCES_SIZE) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    {   int nbSeq = *ip++;
+        if (!nbSeq) { *nbSeqPtr=0; return 1; }
+        if (nbSeq > 0x7F) {
+            if (nbSeq == 0xFF) {
+                if (ip+2 > iend) return ERROR(srcSize_wrong);
+                nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
+            } else {
+                if (ip >= iend) return ERROR(srcSize_wrong);
+                nbSeq = ((nbSeq-0x80)<<8) + *ip++;
+            }
+        }
+        *nbSeqPtr = nbSeq;
+    }
+
+    /* FSE table descriptors */
+    if (ip + 4 > iend) return ERROR(srcSize_wrong); /* min : header byte + all 3 are "raw", hence no header, but at least xxLog bits per type */
+    {   U32 const LLtype  = *ip >> 6;
+        U32 const OFtype = (*ip >> 4) & 3;
+        U32 const MLtype  = (*ip >> 2) & 3;
+        ip++;
+
+        /* Build DTables */
+        {   size_t const llhSize = ZSTDv07_buildSeqTable(DTableLL, LLtype, MaxLL, LLFSELog, ip, iend-ip, LL_defaultNorm, LL_defaultNormLog, flagRepeatTable);
+            if (ZSTDv07_isError(llhSize)) return ERROR(corruption_detected);
+            ip += llhSize;
+        }
+        {   size_t const ofhSize = ZSTDv07_buildSeqTable(DTableOffb, OFtype, MaxOff, OffFSELog, ip, iend-ip, OF_defaultNorm, OF_defaultNormLog, flagRepeatTable);
+            if (ZSTDv07_isError(ofhSize)) return ERROR(corruption_detected);
+            ip += ofhSize;
+        }
+        {   size_t const mlhSize = ZSTDv07_buildSeqTable(DTableML, MLtype, MaxML, MLFSELog, ip, iend-ip, ML_defaultNorm, ML_defaultNormLog, flagRepeatTable);
+            if (ZSTDv07_isError(mlhSize)) return ERROR(corruption_detected);
+            ip += mlhSize;
+    }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+} seq_t;
+
+typedef struct {
+    BITv07_DStream_t DStream;
+    FSEv07_DState_t stateLL;
+    FSEv07_DState_t stateOffb;
+    FSEv07_DState_t stateML;
+    size_t prevOffset[ZSTDv07_REP_INIT];
+} seqState_t;
+
+
+static seq_t ZSTDv07_decodeSequence(seqState_t* seqState)
+{
+    seq_t seq;
+
+    U32 const llCode = FSEv07_peekSymbol(&(seqState->stateLL));
+    U32 const mlCode = FSEv07_peekSymbol(&(seqState->stateML));
+    U32 const ofCode = FSEv07_peekSymbol(&(seqState->stateOffb));   /* <= maxOff, by table construction */
+
+    U32 const llBits = LL_bits[llCode];
+    U32 const mlBits = ML_bits[mlCode];
+    U32 const ofBits = ofCode;
+    U32 const totalBits = llBits+mlBits+ofBits;
+
+    static const U32 LL_base[MaxLL+1] = {
+                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9,   10,    11,    12,    13,    14,     15,
+                            16, 18, 20, 22, 24, 28, 32, 40, 48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                            0x2000, 0x4000, 0x8000, 0x10000 };
+
+    static const U32 ML_base[MaxML+1] = {
+                             3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13,   14,    15,    16,    17,    18,
+                            19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29,   30,    31,    32,    33,    34,
+                            35, 37, 39, 41, 43, 47, 51, 59, 67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
+                            0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
+
+    static const U32 OF_base[MaxOff+1] = {
+                 0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
+                 0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
+                 0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
+                 0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
+
+    /* sequence */
+    {   size_t offset;
+        if (!ofCode)
+            offset = 0;
+        else {
+            offset = OF_base[ofCode] + BITv07_readBits(&(seqState->DStream), ofBits);   /* <=  (ZSTDv07_WINDOWLOG_MAX-1) bits */
+            if (MEM_32bits()) BITv07_reloadDStream(&(seqState->DStream));
+        }
+
+        if (ofCode <= 1) {
+            if ((llCode == 0) & (offset <= 1)) offset = 1-offset;
+            if (offset) {
+                size_t const temp = seqState->prevOffset[offset];
+                if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset = temp;
+            } else {
+                offset = seqState->prevOffset[0];
+            }
+        } else {
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];
+            seqState->prevOffset[0] = offset;
+        }
+        seq.offset = offset;
+    }
+
+    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BITv07_readBits(&(seqState->DStream), mlBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() && (mlBits+llBits>24)) BITv07_reloadDStream(&(seqState->DStream));
+
+    seq.litLength = LL_base[llCode] + ((llCode>15) ? BITv07_readBits(&(seqState->DStream), llBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() ||
+       (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BITv07_reloadDStream(&(seqState->DStream));
+
+    /* ANS state update */
+    FSEv07_updateState(&(seqState->stateLL), &(seqState->DStream));   /* <=  9 bits */
+    FSEv07_updateState(&(seqState->stateML), &(seqState->DStream));   /* <=  9 bits */
+    if (MEM_32bits()) BITv07_reloadDStream(&(seqState->DStream));     /* <= 18 bits */
+    FSEv07_updateState(&(seqState->stateOffb), &(seqState->DStream)); /* <=  8 bits */
+
+    return seq;
+}
+
+
+static
+size_t ZSTDv07_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend-WILDCOPY_OVERLENGTH;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* check */
+    assert(oend >= op);
+    if (sequence.litLength + WILDCOPY_OVERLENGTH > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequenceLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    assert(litLimit >= *litPtr);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);;
+
+    /* copy Literals */
+    ZSTDv07_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = (size_t)(dictEnd - match);
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_w || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+    }   }
+    /* Requirement: op <= oend_w */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTDv07_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTDv07_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_w) {
+            ZSTDv07_wildcopy(op, match, oend_w - op);
+            match += oend_w - op;
+            op = oend_w;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    } else {
+        ZSTDv07_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTDv07_decompressSequences(
+                               ZSTDv07_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    FSEv07_DTable* DTableLL = dctx->LLTable;
+    FSEv07_DTable* DTableML = dctx->MLTable;
+    FSEv07_DTable* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    int nbSeq;
+
+    /* Build Decoding Tables */
+    {   size_t const seqHSize = ZSTDv07_decodeSeqHeaders(&nbSeq, DTableLL, DTableML, DTableOffb, dctx->fseEntropy, ip, seqSize);
+        if (ZSTDv07_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+    }
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seqState_t seqState;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i=0; i<ZSTDv07_REP_INIT; i++) seqState.prevOffset[i] = dctx->rep[i]; }
+        { size_t const errorCode = BITv07_initDStream(&(seqState.DStream), ip, iend-ip);
+          if (ERR_isError(errorCode)) return ERROR(corruption_detected); }
+        FSEv07_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSEv07_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSEv07_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BITv07_reloadDStream(&(seqState.DStream)) <= BITv07_DStream_completed) && nbSeq ; ) {
+            nbSeq--;
+            {   seq_t const sequence = ZSTDv07_decodeSequence(&seqState);
+                size_t const oneSeqSize = ZSTDv07_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+                if (ZSTDv07_isError(oneSeqSize)) return oneSeqSize;
+                op += oneSeqSize;
+        }   }
+
+        /* check if reached exact end */
+        if (nbSeq) return ERROR(corruption_detected);
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTDv07_REP_INIT; i++) dctx->rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        /* if (litPtr > litEnd) return ERROR(corruption_detected); */   /* too many literals already used */
+        if (lastLLSize > (size_t)(oend-op)) return ERROR(dstSize_tooSmall);
+        if (lastLLSize > 0) {
+            memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTDv07_checkContinuity(ZSTDv07_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTDv07_decompressBlock_internal(ZSTDv07_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize >= ZSTDv07_BLOCKSIZE_ABSOLUTEMAX) return ERROR(srcSize_wrong);
+
+    /* Decode literals sub-block */
+    {   size_t const litCSize = ZSTDv07_decodeLiteralsBlock(dctx, src, srcSize);
+        if (ZSTDv07_isError(litCSize)) return litCSize;
+        ip += litCSize;
+        srcSize -= litCSize;
+    }
+    return ZSTDv07_decompressSequences(dctx, dst, dstCapacity, ip, srcSize);
+}
+
+
+size_t ZSTDv07_decompressBlock(ZSTDv07_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    size_t dSize;
+    ZSTDv07_checkContinuity(dctx, dst);
+    dSize = ZSTDv07_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+    dctx->previousDstEnd = (char*)dst + dSize;
+    return dSize;
+}
+
+
+/** ZSTDv07_insertBlock() :
+    insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
+ZSTDLIBv07_API size_t ZSTDv07_insertBlock(ZSTDv07_DCtx* dctx, const void* blockStart, size_t blockSize)
+{
+    ZSTDv07_checkContinuity(dctx, blockStart);
+    dctx->previousDstEnd = (const char*)blockStart + blockSize;
+    return blockSize;
+}
+
+
+static size_t ZSTDv07_generateNxBytes(void* dst, size_t dstCapacity, BYTE byte, size_t length)
+{
+    if (length > dstCapacity) return ERROR(dstSize_tooSmall);
+    if (length > 0) {
+        memset(dst, byte, length);
+    }
+    return length;
+}
+
+
+/*! ZSTDv07_decompressFrame() :
+*   `dctx` must be properly initialized */
+static size_t ZSTDv07_decompressFrame(ZSTDv07_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    size_t remainingSize = srcSize;
+
+    /* check */
+    if (srcSize < ZSTDv07_frameHeaderSize_min+ZSTDv07_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv07_frameHeaderSize(src, ZSTDv07_frameHeaderSize_min);
+        if (ZSTDv07_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTDv07_blockHeaderSize) return ERROR(srcSize_wrong);
+        if (ZSTDv07_decodeFrameHeader(dctx, src, frameHeaderSize)) return ERROR(corruption_detected);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        size_t decodedSize;
+        blockProperties_t blockProperties;
+        size_t const cBlockSize = ZSTDv07_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTDv07_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv07_blockHeaderSize;
+        remainingSize -= ZSTDv07_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTDv07_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTDv07_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            decodedSize = ZSTDv07_generateNxBytes(op, oend-op, *ip, blockProperties.origSize);
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            decodedSize = 0;
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (blockProperties.blockType == bt_end) break;   /* bt_end */
+
+        if (ZSTDv07_isError(decodedSize)) return decodedSize;
+        if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, op, decodedSize);
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+
+/*! ZSTDv07_decompress_usingPreparedDCtx() :
+*   Same as ZSTDv07_decompress_usingDict, but using a reference context `preparedDCtx`, where dictionary has been loaded.
+*   It avoids reloading the dictionary each time.
+*   `preparedDCtx` must have been properly initialized using ZSTDv07_decompressBegin_usingDict().
+*   Requires 2 contexts : 1 for reference (preparedDCtx), which will not be modified, and 1 to run the decompression operation (dctx) */
+static size_t ZSTDv07_decompress_usingPreparedDCtx(ZSTDv07_DCtx* dctx, const ZSTDv07_DCtx* refDCtx,
+                                         void* dst, size_t dstCapacity,
+                                   const void* src, size_t srcSize)
+{
+    ZSTDv07_copyDCtx(dctx, refDCtx);
+    ZSTDv07_checkContinuity(dctx, dst);
+    return ZSTDv07_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv07_decompress_usingDict(ZSTDv07_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    ZSTDv07_decompressBegin_usingDict(dctx, dict, dictSize);
+    ZSTDv07_checkContinuity(dctx, dst);
+    return ZSTDv07_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv07_decompressDCtx(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTDv07_decompress_usingDict(dctx, dst, dstCapacity, src, srcSize, NULL, 0);
+}
+
+
+size_t ZSTDv07_decompress(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+#if defined(ZSTDv07_HEAPMODE) && (ZSTDv07_HEAPMODE==1)
+    size_t regenSize;
+    ZSTDv07_DCtx* const dctx = ZSTDv07_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv07_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
+    ZSTDv07_freeDCtx(dctx);
+    return regenSize;
+#else   /* stack mode */
+    ZSTDv07_DCtx dctx;
+    return ZSTDv07_decompressDCtx(&dctx, dst, dstCapacity, src, srcSize);
+#endif
+}
+
+/* ZSTD_errorFrameSizeInfoLegacy() :
+   assumes `cSize` and `dBound` are _not_ NULL */
+static void ZSTD_errorFrameSizeInfoLegacy(size_t* cSize, unsigned long long* dBound, size_t ret)
+{
+    *cSize = ret;
+    *dBound = ZSTD_CONTENTSIZE_ERROR;
+}
+
+void ZSTDv07_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cSize, unsigned long long* dBound)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    size_t nbBlocks = 0;
+
+    /* check */
+    if (srcSize < ZSTDv07_frameHeaderSize_min+ZSTDv07_blockHeaderSize) {
+        ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+        return;
+    }
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv07_frameHeaderSize(src, srcSize);
+        if (ZSTDv07_isError(frameHeaderSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, frameHeaderSize);
+            return;
+        }
+        if (MEM_readLE32(src) != ZSTDv07_MAGICNUMBER) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(prefix_unknown));
+            return;
+        }
+        if (srcSize < frameHeaderSize+ZSTDv07_blockHeaderSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        blockProperties_t blockProperties;
+        size_t const cBlockSize = ZSTDv07_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTDv07_isError(cBlockSize)) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, cBlockSize);
+            return;
+        }
+
+        ip += ZSTDv07_blockHeaderSize;
+        remainingSize -= ZSTDv07_blockHeaderSize;
+
+        if (blockProperties.blockType == bt_end) break;
+
+        if (cBlockSize > remainingSize) {
+            ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, ERROR(srcSize_wrong));
+            return;
+        }
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        nbBlocks++;
+    }
+
+    *cSize = ip - (const BYTE*)src;
+    *dBound = nbBlocks * ZSTDv07_BLOCKSIZE_ABSOLUTEMAX;
+}
+
+/*_******************************
+*  Streaming Decompression API
+********************************/
+size_t ZSTDv07_nextSrcSizeToDecompress(ZSTDv07_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+int ZSTDv07_isSkipFrame(ZSTDv07_DCtx* dctx)
+{
+    return dctx->stage == ZSTDds_skipFrame;
+}
+
+/** ZSTDv07_decompressContinue() :
+*   @return : nb of bytes generated into `dst` (necessarily <= `dstCapacity)
+*             or an error code, which can be tested using ZSTDv07_isError() */
+size_t ZSTDv07_decompressContinue(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);
+    if (dstCapacity) ZSTDv07_checkContinuity(dctx, dst);
+
+    switch (dctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        if (srcSize != ZSTDv07_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTDv07_MAGIC_SKIPPABLE_START) {
+            memcpy(dctx->headerBuffer, src, ZSTDv07_frameHeaderSize_min);
+            dctx->expected = ZSTDv07_skippableHeaderSize - ZSTDv07_frameHeaderSize_min; /* magic number + skippable frame length */
+            dctx->stage = ZSTDds_decodeSkippableHeader;
+            return 0;
+        }
+        dctx->headerSize = ZSTDv07_frameHeaderSize(src, ZSTDv07_frameHeaderSize_min);
+        if (ZSTDv07_isError(dctx->headerSize)) return dctx->headerSize;
+        memcpy(dctx->headerBuffer, src, ZSTDv07_frameHeaderSize_min);
+        if (dctx->headerSize > ZSTDv07_frameHeaderSize_min) {
+            dctx->expected = dctx->headerSize - ZSTDv07_frameHeaderSize_min;
+            dctx->stage = ZSTDds_decodeFrameHeader;
+            return 0;
+        }
+        dctx->expected = 0;   /* not necessary to copy more */
+	/* fall-through */
+    case ZSTDds_decodeFrameHeader:
+        {   size_t result;
+            memcpy(dctx->headerBuffer + ZSTDv07_frameHeaderSize_min, src, dctx->expected);
+            result = ZSTDv07_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize);
+            if (ZSTDv07_isError(result)) return result;
+            dctx->expected = ZSTDv07_blockHeaderSize;
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDds_decodeBlockHeader:
+        {   blockProperties_t bp;
+            size_t const cBlockSize = ZSTDv07_getcBlockSize(src, ZSTDv07_blockHeaderSize, &bp);
+            if (ZSTDv07_isError(cBlockSize)) return cBlockSize;
+            if (bp.blockType == bt_end) {
+                if (dctx->fParams.checksumFlag) {
+                    U64 const h64 = XXH64_digest(&dctx->xxhState);
+                    U32 const h32 = (U32)(h64>>11) & ((1<<22)-1);
+                    const BYTE* const ip = (const BYTE*)src;
+                    U32 const check32 = ip[2] + (ip[1] << 8) + ((ip[0] & 0x3F) << 16);
+                    if (check32 != h32) return ERROR(checksum_wrong);
+                }
+                dctx->expected = 0;
+                dctx->stage = ZSTDds_getFrameHeaderSize;
+            } else {
+                dctx->expected = cBlockSize;
+                dctx->bType = bp.blockType;
+                dctx->stage = ZSTDds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDds_decompressBlock:
+        {   size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTDv07_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTDv07_copyRawBlock(dst, dstCapacity, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);   /* impossible */
+            }
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            dctx->expected = ZSTDv07_blockHeaderSize;
+            if (ZSTDv07_isError(rSize)) return rSize;
+            dctx->previousDstEnd = (char*)dst + rSize;
+            if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, dst, rSize);
+            return rSize;
+        }
+    case ZSTDds_decodeSkippableHeader:
+        {   memcpy(dctx->headerBuffer + ZSTDv07_frameHeaderSize_min, src, dctx->expected);
+            dctx->expected = MEM_readLE32(dctx->headerBuffer + 4);
+            dctx->stage = ZSTDds_skipFrame;
+            return 0;
+        }
+    case ZSTDds_skipFrame:
+        {   dctx->expected = 0;
+            dctx->stage = ZSTDds_getFrameHeaderSize;
+            return 0;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static size_t ZSTDv07_refDictContent(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+    dctx->base = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+    return 0;
+}
+
+static size_t ZSTDv07_loadEntropy(ZSTDv07_DCtx* dctx, const void* const dict, size_t const dictSize)
+{
+    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* const dictEnd = dictPtr + dictSize;
+
+    {   size_t const hSize = HUFv07_readDTableX4(dctx->hufTable, dict, dictSize);
+        if (HUFv07_isError(hSize)) return ERROR(dictionary_corrupted);
+        dictPtr += hSize;
+    }
+
+    {   short offcodeNCount[MaxOff+1];
+        U32 offcodeMaxValue=MaxOff, offcodeLog;
+        size_t const offcodeHeaderSize = FSEv07_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
+        if (FSEv07_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv07_buildDTable(dctx->OffTable, offcodeNCount, offcodeMaxValue, offcodeLog);
+          if (FSEv07_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dictPtr += offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        size_t const matchlengthHeaderSize = FSEv07_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSEv07_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv07_buildDTable(dctx->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog);
+          if (FSEv07_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dictPtr += matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        size_t const litlengthHeaderSize = FSEv07_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSEv07_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv07_buildDTable(dctx->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog);
+          if (FSEv07_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dictPtr += litlengthHeaderSize;
+    }
+
+    if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
+    dctx->rep[0] = MEM_readLE32(dictPtr+0); if (dctx->rep[0] == 0 || dctx->rep[0] >= dictSize) return ERROR(dictionary_corrupted);
+    dctx->rep[1] = MEM_readLE32(dictPtr+4); if (dctx->rep[1] == 0 || dctx->rep[1] >= dictSize) return ERROR(dictionary_corrupted);
+    dctx->rep[2] = MEM_readLE32(dictPtr+8); if (dctx->rep[2] == 0 || dctx->rep[2] >= dictSize) return ERROR(dictionary_corrupted);
+    dictPtr += 12;
+
+    dctx->litEntropy = dctx->fseEntropy = 1;
+    return dictPtr - (const BYTE*)dict;
+}
+
+static size_t ZSTDv07_decompress_insertDictionary(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    if (dictSize < 8) return ZSTDv07_refDictContent(dctx, dict, dictSize);
+    {   U32 const magic = MEM_readLE32(dict);
+        if (magic != ZSTDv07_DICT_MAGIC) {
+            return ZSTDv07_refDictContent(dctx, dict, dictSize);   /* pure content mode */
+    }   }
+    dctx->dictID = MEM_readLE32((const char*)dict + 4);
+
+    /* load entropy tables */
+    dict = (const char*)dict + 8;
+    dictSize -= 8;
+    {   size_t const eSize = ZSTDv07_loadEntropy(dctx, dict, dictSize);
+        if (ZSTDv07_isError(eSize)) return ERROR(dictionary_corrupted);
+        dict = (const char*)dict + eSize;
+        dictSize -= eSize;
+    }
+
+    /* reference dictionary content */
+    return ZSTDv07_refDictContent(dctx, dict, dictSize);
+}
+
+
+size_t ZSTDv07_decompressBegin_usingDict(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    { size_t const errorCode = ZSTDv07_decompressBegin(dctx);
+      if (ZSTDv07_isError(errorCode)) return errorCode; }
+
+    if (dict && dictSize) {
+        size_t const errorCode = ZSTDv07_decompress_insertDictionary(dctx, dict, dictSize);
+        if (ZSTDv07_isError(errorCode)) return ERROR(dictionary_corrupted);
+    }
+
+    return 0;
+}
+
+
+struct ZSTDv07_DDict_s {
+    void* dict;
+    size_t dictSize;
+    ZSTDv07_DCtx* refContext;
+};  /* typedef'd tp ZSTDv07_CDict within zstd.h */
+
+static ZSTDv07_DDict* ZSTDv07_createDDict_advanced(const void* dict, size_t dictSize, ZSTDv07_customMem customMem)
+{
+    if (!customMem.customAlloc && !customMem.customFree)
+        customMem = defaultCustomMem;
+
+    if (!customMem.customAlloc || !customMem.customFree)
+        return NULL;
+
+    {   ZSTDv07_DDict* const ddict = (ZSTDv07_DDict*) customMem.customAlloc(customMem.opaque, sizeof(*ddict));
+        void* const dictContent = customMem.customAlloc(customMem.opaque, dictSize);
+        ZSTDv07_DCtx* const dctx = ZSTDv07_createDCtx_advanced(customMem);
+
+        if (!dictContent || !ddict || !dctx) {
+            customMem.customFree(customMem.opaque, dictContent);
+            customMem.customFree(customMem.opaque, ddict);
+            customMem.customFree(customMem.opaque, dctx);
+            return NULL;
+        }
+
+        memcpy(dictContent, dict, dictSize);
+        {   size_t const errorCode = ZSTDv07_decompressBegin_usingDict(dctx, dictContent, dictSize);
+            if (ZSTDv07_isError(errorCode)) {
+                customMem.customFree(customMem.opaque, dictContent);
+                customMem.customFree(customMem.opaque, ddict);
+                customMem.customFree(customMem.opaque, dctx);
+                return NULL;
+        }   }
+
+        ddict->dict = dictContent;
+        ddict->dictSize = dictSize;
+        ddict->refContext = dctx;
+        return ddict;
+    }
+}
+
+/*! ZSTDv07_createDDict() :
+*   Create a digested dictionary, ready to start decompression without startup delay.
+*   `dict` can be released after `ZSTDv07_DDict` creation */
+ZSTDv07_DDict* ZSTDv07_createDDict(const void* dict, size_t dictSize)
+{
+    ZSTDv07_customMem const allocator = { NULL, NULL, NULL };
+    return ZSTDv07_createDDict_advanced(dict, dictSize, allocator);
+}
+
+size_t ZSTDv07_freeDDict(ZSTDv07_DDict* ddict)
+{
+    ZSTDv07_freeFunction const cFree = ddict->refContext->customMem.customFree;
+    void* const opaque = ddict->refContext->customMem.opaque;
+    ZSTDv07_freeDCtx(ddict->refContext);
+    cFree(opaque, ddict->dict);
+    cFree(opaque, ddict);
+    return 0;
+}
+
+/*! ZSTDv07_decompress_usingDDict() :
+*   Decompression using a pre-digested Dictionary
+*   Use dictionary without significant overhead. */
+ZSTDLIBv07_API size_t ZSTDv07_decompress_usingDDict(ZSTDv07_DCtx* dctx,
+                                           void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize,
+                                     const ZSTDv07_DDict* ddict)
+{
+    return ZSTDv07_decompress_usingPreparedDCtx(dctx, ddict->refContext,
+                                           dst, dstCapacity,
+                                           src, srcSize);
+}
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://facebook.github.io/zstd/
+*/
+
+
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFFv07_DCtx object is required to track streaming operations.
+*  Use ZBUFFv07_createDCtx() and ZBUFFv07_freeDCtx() to create/release resources.
+*  Use ZBUFFv07_decompressInit() to start a new decompression operation,
+*   or ZBUFFv07_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv07_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFFv07_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of @dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change @dst.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency),
+*            or 0 when a frame is completely decoded,
+*            or an error code, which can be tested using ZBUFFv07_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv07_recommendedDInSize() and ZBUFFv07_recommendedDOutSize()
+*  output : ZBUFFv07_recommendedDOutSize==128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv07_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFFv07_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+typedef enum { ZBUFFds_init, ZBUFFds_loadHeader,
+               ZBUFFds_read, ZBUFFds_load, ZBUFFds_flush } ZBUFFv07_dStage;
+
+/* *** Resource management *** */
+struct ZBUFFv07_DCtx_s {
+    ZSTDv07_DCtx* zd;
+    ZSTDv07_frameParams fParams;
+    ZBUFFv07_dStage stage;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t blockSize;
+    BYTE headerBuffer[ZSTDv07_FRAMEHEADERSIZE_MAX];
+    size_t lhSize;
+    ZSTDv07_customMem customMem;
+};   /* typedef'd to ZBUFFv07_DCtx within "zstd_buffered.h" */
+
+ZSTDLIBv07_API ZBUFFv07_DCtx* ZBUFFv07_createDCtx_advanced(ZSTDv07_customMem customMem);
+
+ZBUFFv07_DCtx* ZBUFFv07_createDCtx(void)
+{
+    return ZBUFFv07_createDCtx_advanced(defaultCustomMem);
+}
+
+ZBUFFv07_DCtx* ZBUFFv07_createDCtx_advanced(ZSTDv07_customMem customMem)
+{
+    ZBUFFv07_DCtx* zbd;
+
+    if (!customMem.customAlloc && !customMem.customFree)
+        customMem = defaultCustomMem;
+
+    if (!customMem.customAlloc || !customMem.customFree)
+        return NULL;
+
+    zbd = (ZBUFFv07_DCtx*)customMem.customAlloc(customMem.opaque, sizeof(ZBUFFv07_DCtx));
+    if (zbd==NULL) return NULL;
+    memset(zbd, 0, sizeof(ZBUFFv07_DCtx));
+    memcpy(&zbd->customMem, &customMem, sizeof(ZSTDv07_customMem));
+    zbd->zd = ZSTDv07_createDCtx_advanced(customMem);
+    if (zbd->zd == NULL) { ZBUFFv07_freeDCtx(zbd); return NULL; }
+    zbd->stage = ZBUFFds_init;
+    return zbd;
+}
+
+size_t ZBUFFv07_freeDCtx(ZBUFFv07_DCtx* zbd)
+{
+    if (zbd==NULL) return 0;   /* support free on null */
+    ZSTDv07_freeDCtx(zbd->zd);
+    if (zbd->inBuff) zbd->customMem.customFree(zbd->customMem.opaque, zbd->inBuff);
+    if (zbd->outBuff) zbd->customMem.customFree(zbd->customMem.opaque, zbd->outBuff);
+    zbd->customMem.customFree(zbd->customMem.opaque, zbd);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+size_t ZBUFFv07_decompressInitDictionary(ZBUFFv07_DCtx* zbd, const void* dict, size_t dictSize)
+{
+    zbd->stage = ZBUFFds_loadHeader;
+    zbd->lhSize = zbd->inPos = zbd->outStart = zbd->outEnd = 0;
+    return ZSTDv07_decompressBegin_usingDict(zbd->zd, dict, dictSize);
+}
+
+size_t ZBUFFv07_decompressInit(ZBUFFv07_DCtx* zbd)
+{
+    return ZBUFFv07_decompressInitDictionary(zbd, NULL, 0);
+}
+
+
+/* internal util function */
+MEM_STATIC size_t ZBUFFv07_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    size_t const length = MIN(dstCapacity, srcSize);
+    if (length > 0) {
+        memcpy(dst, src, length);
+    }
+    return length;
+}
+
+
+/* *** Decompression *** */
+
+size_t ZBUFFv07_decompressContinue(ZBUFFv07_DCtx* zbd,
+                                void* dst, size_t* dstCapacityPtr,
+                          const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* const iend = istart + *srcSizePtr;
+    const char* ip = istart;
+    char* const ostart = (char*)dst;
+    char* const oend = ostart + *dstCapacityPtr;
+    char* op = ostart;
+    U32 notDone = 1;
+
+    while (notDone) {
+        switch(zbd->stage)
+        {
+        case ZBUFFds_init :
+            return ERROR(init_missing);
+
+        case ZBUFFds_loadHeader :
+            {   size_t const hSize = ZSTDv07_getFrameParams(&(zbd->fParams), zbd->headerBuffer, zbd->lhSize);
+                if (ZSTDv07_isError(hSize)) return hSize;
+                if (hSize != 0) {
+                    size_t const toLoad = hSize - zbd->lhSize;   /* if hSize!=0, hSize > zbd->lhSize */
+                    if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
+                        if (ip != NULL)
+                            memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
+                        zbd->lhSize += iend-ip;
+                        *dstCapacityPtr = 0;
+                        return (hSize - zbd->lhSize) + ZSTDv07_blockHeaderSize;   /* remaining header bytes + next block header */
+                    }
+                    memcpy(zbd->headerBuffer + zbd->lhSize, ip, toLoad); zbd->lhSize = hSize; ip += toLoad;
+                    break;
+            }   }
+
+            /* Consume header */
+            {   size_t const h1Size = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);  /* == ZSTDv07_frameHeaderSize_min */
+                size_t const h1Result = ZSTDv07_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer, h1Size);
+                if (ZSTDv07_isError(h1Result)) return h1Result;
+                if (h1Size < zbd->lhSize) {   /* long header */
+                    size_t const h2Size = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+                    size_t const h2Result = ZSTDv07_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer+h1Size, h2Size);
+                    if (ZSTDv07_isError(h2Result)) return h2Result;
+            }   }
+
+            zbd->fParams.windowSize = MAX(zbd->fParams.windowSize, 1U << ZSTDv07_WINDOWLOG_ABSOLUTEMIN);
+
+            /* Frame header instruct buffer sizes */
+            {   size_t const blockSize = MIN(zbd->fParams.windowSize, ZSTDv07_BLOCKSIZE_ABSOLUTEMAX);
+                zbd->blockSize = blockSize;
+                if (zbd->inBuffSize < blockSize) {
+                    zbd->customMem.customFree(zbd->customMem.opaque, zbd->inBuff);
+                    zbd->inBuffSize = blockSize;
+                    zbd->inBuff = (char*)zbd->customMem.customAlloc(zbd->customMem.opaque, blockSize);
+                    if (zbd->inBuff == NULL) return ERROR(memory_allocation);
+                }
+                {   size_t const neededOutSize = zbd->fParams.windowSize + blockSize + WILDCOPY_OVERLENGTH * 2;
+                    if (zbd->outBuffSize < neededOutSize) {
+                        zbd->customMem.customFree(zbd->customMem.opaque, zbd->outBuff);
+                        zbd->outBuffSize = neededOutSize;
+                        zbd->outBuff = (char*)zbd->customMem.customAlloc(zbd->customMem.opaque, neededOutSize);
+                        if (zbd->outBuff == NULL) return ERROR(memory_allocation);
+            }   }   }
+            zbd->stage = ZBUFFds_read;
+            /* pass-through */
+	    /* fall-through */
+        case ZBUFFds_read:
+            {   size_t const neededInSize = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+                if (neededInSize==0) {  /* end of frame */
+                    zbd->stage = ZBUFFds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
+                    const int isSkipFrame = ZSTDv07_isSkipFrame(zbd->zd);
+                    size_t const decodedSize = ZSTDv07_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, (isSkipFrame ? 0 : zbd->outBuffSize - zbd->outStart),
+                        ip, neededInSize);
+                    if (ZSTDv07_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize && !isSkipFrame) break;   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbd->stage = ZBUFFds_load;
+            }
+	    /* fall-through */
+        case ZBUFFds_load:
+            {   size_t const neededInSize = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+                size_t const toLoad = neededInSize - zbd->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbd->inBuffSize - zbd->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFFv07_limitCopy(zbd->inBuff + zbd->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbd->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+
+                /* decode loaded input */
+                {  const int isSkipFrame = ZSTDv07_isSkipFrame(zbd->zd);
+                   size_t const decodedSize = ZSTDv07_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, zbd->outBuffSize - zbd->outStart,
+                        zbd->inBuff, neededInSize);
+                    if (ZSTDv07_isError(decodedSize)) return decodedSize;
+                    zbd->inPos = 0;   /* input is consumed */
+                    if (!decodedSize && !isSkipFrame) { zbd->stage = ZBUFFds_read; break; }   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    /* break; */
+                    /* pass-through */
+                }
+	    }
+	    /* fall-through */
+        case ZBUFFds_flush:
+            {   size_t const toFlushSize = zbd->outEnd - zbd->outStart;
+                size_t const flushedSize = ZBUFFv07_limitCopy(op, oend-op, zbd->outBuff + zbd->outStart, toFlushSize);
+                op += flushedSize;
+                zbd->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {
+                    zbd->stage = ZBUFFds_read;
+                    if (zbd->outStart + zbd->blockSize > zbd->outBuffSize)
+                        zbd->outStart = zbd->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+    }   }
+
+    /* result */
+    *srcSizePtr = ip-istart;
+    *dstCapacityPtr = op-ostart;
+    {   size_t nextSrcSizeHint = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+        nextSrcSizeHint -= zbd->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+size_t ZBUFFv07_recommendedDInSize(void)  { return ZSTDv07_BLOCKSIZE_ABSOLUTEMAX + ZSTDv07_blockHeaderSize /* block header size*/ ; }
+size_t ZBUFFv07_recommendedDOutSize(void) { return ZSTDv07_BLOCKSIZE_ABSOLUTEMAX; }
diff --git a/programs/benchzstd.c b/programs/benchzstd.c
new file mode 100644
index 0000000..f55c869
--- /dev/null
+++ b/programs/benchzstd.c
@@ -0,0 +1,1270 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* **************************************
+ *  Tuning parameters
+ ****************************************/
+#ifndef BMK_TIMETEST_DEFAULT_S /* default minimum time per test */
+#    define BMK_TIMETEST_DEFAULT_S 3
+#endif
+
+/* *************************************
+ *  Includes
+ ***************************************/
+/* this must be included first */
+#include "platform.h" /* Large Files support, compiler specifics */
+
+/* then following system includes */
+#include <assert.h> /* assert */
+#include <errno.h>
+#include <stdio.h>    /* fprintf, fopen */
+#include <stdlib.h>   /* malloc, free */
+#include <string.h>   /* memset, strerror */
+#include "util.h"     /* UTIL_getFileSize, UTIL_sleep */
+#include "../lib/common/mem.h"
+#include "benchfn.h"
+#include "timefn.h" /* UTIL_time_t */
+#ifndef ZSTD_STATIC_LINKING_ONLY
+#    define ZSTD_STATIC_LINKING_ONLY
+#endif
+#include "../lib/zstd.h"
+#include "datagen.h" /* RDG_genBuffer */
+#include "lorem.h"   /* LOREM_genBuffer */
+#ifndef XXH_INLINE_ALL
+#    define XXH_INLINE_ALL
+#endif
+#include "../lib/common/xxhash.h"
+#include "../lib/zstd_errors.h"
+#include "benchzstd.h"
+
+/* *************************************
+ *  Constants
+ ***************************************/
+#ifndef ZSTD_GIT_COMMIT
+#    define ZSTD_GIT_COMMIT_STRING ""
+#else
+#    define ZSTD_GIT_COMMIT_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_GIT_COMMIT)
+#endif
+
+#define TIMELOOP_MICROSEC (1 * 1000000ULL)             /* 1 second */
+#define TIMELOOP_NANOSEC (1 * 1000000000ULL)           /* 1 second */
+#define ACTIVEPERIOD_MICROSEC (70 * TIMELOOP_MICROSEC) /* 70 seconds */
+#define COOLPERIOD_SEC 10
+
+#define KB *(1 << 10)
+#define MB *(1 << 20)
+#define GB *(1U << 30)
+
+#define BMK_RUNTEST_DEFAULT_MS 1000
+
+static const size_t maxMemory = (sizeof(size_t) == 4)
+        ?
+        /* 32-bit */ (2 GB - 64 MB)
+        :
+        /* 64-bit */ (size_t)(1ULL << ((sizeof(size_t) * 8) - 31));
+
+/* *************************************
+ *  console display
+ ***************************************/
+#define DISPLAY(...)                  \
+    {                                 \
+        fprintf(stderr, __VA_ARGS__); \
+        fflush(NULL);                 \
+    }
+#define DISPLAYLEVEL(l, ...)  \
+    if (displayLevel >= l) {  \
+        DISPLAY(__VA_ARGS__); \
+    }
+/* 0 : no display;   1: errors;   2 : + result + interaction + warnings;   3 : +
+ * progression;   4 : + information */
+#define OUTPUT(...)                   \
+    {                                 \
+        fprintf(stdout, __VA_ARGS__); \
+        fflush(NULL);                 \
+    }
+#define OUTPUTLEVEL(l, ...)  \
+    if (displayLevel >= l) { \
+        OUTPUT(__VA_ARGS__); \
+    }
+
+/* *************************************
+ *  Exceptions
+ ***************************************/
+#ifndef DEBUG
+#    define DEBUG 0
+#endif
+#define DEBUGOUTPUT(...)          \
+    {                             \
+        if (DEBUG)                \
+            DISPLAY(__VA_ARGS__); \
+    }
+
+#define RETURN_ERROR_INT(errorNum, ...)                \
+    {                                                  \
+        DEBUGOUTPUT("%s: %i: \n", __FILE__, __LINE__); \
+        DISPLAYLEVEL(1, "Error %i : ", errorNum);      \
+        DISPLAYLEVEL(1, __VA_ARGS__);                  \
+        DISPLAYLEVEL(1, " \n");                        \
+        return errorNum;                               \
+    }
+
+#define CHECK_Z(zf)                                                  \
+    {                                                                \
+        size_t const zerr = zf;                                      \
+        if (ZSTD_isError(zerr)) {                                    \
+            DEBUGOUTPUT("%s: %i: \n", __FILE__, __LINE__);           \
+            DISPLAY("Error : ");                                     \
+            DISPLAY("%s failed : %s", #zf, ZSTD_getErrorName(zerr)); \
+            DISPLAY(" \n");                                          \
+            exit(1);                                                 \
+        }                                                            \
+    }
+
+#define RETURN_ERROR(errorNum, retType, ...)           \
+    {                                                  \
+        retType r;                                     \
+        memset(&r, 0, sizeof(retType));                \
+        DEBUGOUTPUT("%s: %i: \n", __FILE__, __LINE__); \
+        DISPLAYLEVEL(1, "Error %i : ", errorNum);      \
+        DISPLAYLEVEL(1, __VA_ARGS__);                  \
+        DISPLAYLEVEL(1, " \n");                        \
+        r.tag = errorNum;                              \
+        return r;                                      \
+    }
+
+static size_t uintSize(unsigned value)
+{
+    size_t size = 1;
+    while (value >= 10) {
+        size++;
+        value /= 10;
+    }
+    return size;
+}
+
+/* Note: presume @buffer is large enough */
+static void writeUint_varLen(char* buffer, size_t capacity, unsigned value)
+{
+    int endPos = (int)uintSize(value) - 1;
+    assert(uintSize(value) >= 1);
+    assert(uintSize(value) < capacity); (void)capacity;
+    while (endPos >= 0) {
+        char c = '0' + (char)(value % 10);
+        buffer[endPos--] = c;
+        value /= 10;
+    }
+}
+
+/* replacement for snprintf(), which is not supported by C89.
+ * sprintf() would be the supported one, but it's labelled unsafe:
+ * modern static analyzer will flag sprintf() as dangerous, making it unusable.
+ * formatString_u() replaces snprintf() for the specific case where there is only one %u argument */
+static int formatString_u(char* buffer, size_t buffer_size, const char* formatString, unsigned int value)
+{
+    size_t const valueSize = uintSize(value);
+    size_t written = 0;
+    int i;
+
+    for (i = 0; formatString[i] != '\0' && written < buffer_size - 1; i++) {
+        if (formatString[i] != '%') {
+            buffer[written++] = formatString[i];
+            continue;
+        }
+
+        i++;
+        if (formatString[i] == 'u') {
+            if (written + valueSize >= buffer_size) abort(); /* buffer not large enough */
+            writeUint_varLen(buffer + written, buffer_size - written, value);
+            written += valueSize;
+        } else if (formatString[i] == '%') { /* Check for escaped percent sign */
+            buffer[written++] = '%';
+        } else {
+            abort(); /* unsupported format */
+        }
+    }
+
+    if (written < buffer_size) {
+        buffer[written] = '\0';
+    } else {
+        abort(); /* buffer not large enough */
+    }
+
+    return (int)written;
+}
+
+/* *************************************
+ *  Benchmark Parameters
+ ***************************************/
+
+BMK_advancedParams_t BMK_initAdvancedParams(void)
+{
+    BMK_advancedParams_t const res = {
+        BMK_both,               /* mode */
+        BMK_TIMETEST_DEFAULT_S, /* nbSeconds */
+        0,                      /* chunkSizeMax */
+        0,               /* targetCBlockSize */
+        0,                      /* nbWorkers */
+        0,                      /* realTime */
+        0,                      /* additionalParam */
+        0,                      /* ldmFlag */
+        0,                      /* ldmMinMatch */
+        0,                      /* ldmHashLog */
+        0,                      /* ldmBuckSizeLog */
+        0,                      /* ldmHashRateLog */
+        ZSTD_ps_auto,           /* literalCompressionMode */
+        0                       /* useRowMatchFinder */
+    };
+    return res;
+}
+
+/* ********************************************************
+ *  Bench functions
+ **********************************************************/
+#undef MIN
+#undef MAX
+#define MIN(a, b) ((a) < (b) ? (a) : (b))
+#define MAX(a, b) ((a) > (b) ? (a) : (b))
+
+static void BMK_initCCtx(
+        ZSTD_CCtx* ctx,
+        const void* dictBuffer,
+        size_t dictBufferSize,
+        int cLevel,
+        const ZSTD_compressionParameters* comprParams,
+        const BMK_advancedParams_t* adv)
+{
+    ZSTD_CCtx_reset(ctx, ZSTD_reset_session_and_parameters);
+    if (adv->nbWorkers == 1) {
+        CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_nbWorkers, 0));
+    } else {
+        CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_nbWorkers, adv->nbWorkers));
+    }
+    CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_compressionLevel, cLevel));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_useRowMatchFinder, adv->useRowMatchFinder));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_enableLongDistanceMatching, adv->ldmFlag));
+    CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_ldmMinMatch, adv->ldmMinMatch));
+    CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_ldmHashLog, adv->ldmHashLog));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_ldmBucketSizeLog, adv->ldmBucketSizeLog));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_ldmHashRateLog, adv->ldmHashRateLog));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_windowLog, (int)comprParams->windowLog));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_hashLog, (int)comprParams->hashLog));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_chainLog, (int)comprParams->chainLog));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_searchLog, (int)comprParams->searchLog));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_minMatch, (int)comprParams->minMatch));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_targetLength, (int)comprParams->targetLength));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx,
+            ZSTD_c_literalCompressionMode,
+            (int)adv->literalCompressionMode));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_strategy, (int)comprParams->strategy));
+    CHECK_Z(ZSTD_CCtx_setParameter(
+            ctx, ZSTD_c_targetCBlockSize, (int)adv->targetCBlockSize));
+    CHECK_Z(ZSTD_CCtx_loadDictionary(ctx, dictBuffer, dictBufferSize));
+}
+
+static void
+BMK_initDCtx(ZSTD_DCtx* dctx, const void* dictBuffer, size_t dictBufferSize)
+{
+    CHECK_Z(ZSTD_DCtx_reset(dctx, ZSTD_reset_session_and_parameters));
+    CHECK_Z(ZSTD_DCtx_loadDictionary(dctx, dictBuffer, dictBufferSize));
+}
+
+typedef struct {
+    ZSTD_CCtx* cctx;
+    const void* dictBuffer;
+    size_t dictBufferSize;
+    int cLevel;
+    const ZSTD_compressionParameters* comprParams;
+    const BMK_advancedParams_t* adv;
+} BMK_initCCtxArgs;
+
+static size_t local_initCCtx(void* payload)
+{
+    BMK_initCCtxArgs* ag = (BMK_initCCtxArgs*)payload;
+    BMK_initCCtx(
+            ag->cctx,
+            ag->dictBuffer,
+            ag->dictBufferSize,
+            ag->cLevel,
+            ag->comprParams,
+            ag->adv);
+    return 0;
+}
+
+typedef struct {
+    ZSTD_DCtx* dctx;
+    const void* dictBuffer;
+    size_t dictBufferSize;
+} BMK_initDCtxArgs;
+
+static size_t local_initDCtx(void* payload)
+{
+    BMK_initDCtxArgs* ag = (BMK_initDCtxArgs*)payload;
+    BMK_initDCtx(ag->dctx, ag->dictBuffer, ag->dictBufferSize);
+    return 0;
+}
+
+/* `addArgs` is the context */
+static size_t local_defaultCompress(
+        const void* srcBuffer,
+        size_t srcSize,
+        void* dstBuffer,
+        size_t dstSize,
+        void* addArgs)
+{
+    ZSTD_CCtx* const cctx = (ZSTD_CCtx*)addArgs;
+    return ZSTD_compress2(cctx, dstBuffer, dstSize, srcBuffer, srcSize);
+}
+
+/* `addArgs` is the context */
+static size_t local_defaultDecompress(
+        const void* srcBuffer,
+        size_t srcSize,
+        void* dstBuffer,
+        size_t dstCapacity,
+        void* addArgs)
+{
+    size_t moreToFlush    = 1;
+    ZSTD_DCtx* const dctx = (ZSTD_DCtx*)addArgs;
+    ZSTD_inBuffer in;
+    ZSTD_outBuffer out;
+    in.src   = srcBuffer;
+    in.size  = srcSize;
+    in.pos   = 0;
+    out.dst  = dstBuffer;
+    out.size = dstCapacity;
+    out.pos  = 0;
+    while (moreToFlush) {
+        if (out.pos == out.size) {
+            return (size_t)-ZSTD_error_dstSize_tooSmall;
+        }
+        moreToFlush = ZSTD_decompressStream(dctx, &out, &in);
+        if (ZSTD_isError(moreToFlush)) {
+            return moreToFlush;
+        }
+    }
+    return out.pos;
+}
+
+/* ================================================================= */
+/*      Benchmark Zstandard, mem-to-mem scenarios                    */
+/* ================================================================= */
+
+int BMK_isSuccessful_benchOutcome(BMK_benchOutcome_t outcome)
+{
+    return outcome.tag == 0;
+}
+
+BMK_benchResult_t BMK_extract_benchResult(BMK_benchOutcome_t outcome)
+{
+    assert(outcome.tag == 0);
+    return outcome.internal_never_use_directly;
+}
+
+static BMK_benchOutcome_t BMK_benchOutcome_error(void)
+{
+    BMK_benchOutcome_t b;
+    memset(&b, 0, sizeof(b));
+    b.tag = 1;
+    return b;
+}
+
+static BMK_benchOutcome_t BMK_benchOutcome_setValidResult(
+        BMK_benchResult_t result)
+{
+    BMK_benchOutcome_t b;
+    b.tag                         = 0;
+    b.internal_never_use_directly = result;
+    return b;
+}
+
+/* benchMem with no allocation */
+static BMK_benchOutcome_t BMK_benchMemAdvancedNoAlloc(
+        const void** srcPtrs,
+        size_t* srcSizes,
+        void** cPtrs,
+        size_t* cCapacities,
+        size_t* cSizes,
+        void** resPtrs,
+        size_t* resSizes,
+        void** resultBufferPtr,
+        void* compressedBuffer,
+        size_t maxCompressedSize,
+        BMK_timedFnState_t* timeStateCompress,
+        BMK_timedFnState_t* timeStateDecompress,
+
+        const void* srcBuffer,
+        size_t srcSize,
+        const size_t* fileSizes,
+        unsigned nbFiles,
+        const int cLevel,
+        const ZSTD_compressionParameters* comprParams,
+        const void* dictBuffer,
+        size_t dictBufferSize,
+        ZSTD_CCtx* cctx,
+        ZSTD_DCtx* dctx,
+        int displayLevel,
+        const char* displayName,
+        const BMK_advancedParams_t* adv)
+{
+    size_t const chunkSizeMax =
+            ((adv->chunkSizeMax >= 32 && (adv->mode != BMK_decodeOnly))
+                     ? adv->chunkSizeMax
+                     : srcSize)
+            + (!srcSize); /* avoid div by 0 */
+    BMK_benchResult_t benchResult;
+    size_t const loadedCompressedSize = srcSize;
+    size_t cSize                      = 0;
+    double ratio                      = 0.;
+    U32 nbChunks = 0;
+
+    assert(cctx != NULL);
+    assert(dctx != NULL);
+
+    /* init */
+    memset(&benchResult, 0, sizeof(benchResult));
+    if (strlen(displayName) > 17)
+        displayName +=
+                strlen(displayName) - 17; /* display last 17 characters */
+    if (adv->mode == BMK_decodeOnly) {
+        /* benchmark only decompression : source must be already compressed */
+        const char* srcPtr = (const char*)srcBuffer;
+        U64 totalDSize64   = 0;
+        U32 fileNb;
+        for (fileNb = 0; fileNb < nbFiles; fileNb++) {
+            U64 const fSize64 =
+                    ZSTD_findDecompressedSize(srcPtr, fileSizes[fileNb]);
+            if (fSize64 == ZSTD_CONTENTSIZE_UNKNOWN) {
+                RETURN_ERROR(
+                        32,
+                        BMK_benchOutcome_t,
+                        "Decompressed size cannot be determined: cannot benchmark");
+            }
+            if (fSize64 == ZSTD_CONTENTSIZE_ERROR) {
+                RETURN_ERROR(
+                        32,
+                        BMK_benchOutcome_t,
+                        "Error while trying to assess decompressed size: data may be invalid");
+            }
+            totalDSize64 += fSize64;
+            srcPtr += fileSizes[fileNb];
+        }
+        {
+            size_t const decodedSize = (size_t)totalDSize64;
+            assert((U64)decodedSize == totalDSize64); /* check overflow */
+            free(*resultBufferPtr);
+            if (totalDSize64 > decodedSize) { /* size_t overflow */
+                RETURN_ERROR(
+                        32,
+                        BMK_benchOutcome_t,
+                        "decompressed size is too large for local system");
+            }
+            *resultBufferPtr = malloc(decodedSize);
+            if (!(*resultBufferPtr)) {
+                RETURN_ERROR(
+                        33,
+                        BMK_benchOutcome_t,
+                        "allocation error: not enough memory");
+            }
+            cSize   = srcSize;
+            srcSize = decodedSize;
+            ratio   = (double)srcSize / (double)cSize;
+        }
+    }
+
+    /* Init data chunks  */
+    {
+        const char* srcPtr = (const char*)srcBuffer;
+        char* cPtr         = (char*)compressedBuffer;
+        char* resPtr       = (char*)(*resultBufferPtr);
+        U32 fileNb, chunkID;
+        for (chunkID = 0, fileNb = 0; fileNb < nbFiles; fileNb++) {
+            size_t remaining              = fileSizes[fileNb];
+            U32 const nbChunksforThisFile = (adv->mode == BMK_decodeOnly)
+                    ? 1
+                    : (U32)((remaining + (chunkSizeMax - 1)) / chunkSizeMax);
+            U32 const chunkIdEnd       = chunkID + nbChunksforThisFile;
+            for (; chunkID < chunkIdEnd; chunkID++) {
+                size_t const chunkSize = MIN(remaining, chunkSizeMax);
+                srcPtrs[chunkID]       = srcPtr;
+                srcSizes[chunkID]      = chunkSize;
+                cPtrs[chunkID]         = cPtr;
+                cCapacities[chunkID]   = (adv->mode == BMK_decodeOnly)
+                             ? chunkSize
+                             : ZSTD_compressBound(chunkSize);
+                resPtrs[chunkID]       = resPtr;
+                resSizes[chunkID]      = (adv->mode == BMK_decodeOnly)
+                                ? (size_t)ZSTD_findDecompressedSize(
+                                srcPtr, chunkSize)
+                                : chunkSize;
+                srcPtr += chunkSize;
+                cPtr += cCapacities[chunkID];
+                resPtr += chunkSize;
+                remaining -= chunkSize;
+                if (adv->mode == BMK_decodeOnly) {
+                    cSizes[chunkID]  = chunkSize;
+                    benchResult.cSize = chunkSize;
+                }
+            }
+        }
+        nbChunks = chunkID;
+    }
+
+    /* warming up `compressedBuffer` */
+    if (adv->mode == BMK_decodeOnly) {
+        memcpy(compressedBuffer, srcBuffer, loadedCompressedSize);
+    } else {
+        RDG_genBuffer(compressedBuffer, maxCompressedSize, 0.10, 0.50, 1);
+    }
+
+    if (!UTIL_support_MT_measurements() && adv->nbWorkers > 1) {
+        OUTPUTLEVEL(
+                2,
+                "Warning : time measurements may be incorrect in multithreading mode... \n")
+    }
+
+    /* Bench */
+    {
+        U64 const crcOrig = (adv->mode == BMK_decodeOnly)
+                ? 0
+                : XXH64(srcBuffer, srcSize, 0);
+#define NB_MARKS 4
+        const char* marks[NB_MARKS] = { " |", " /", " =", " \\" };
+        U32 markNb                  = 0;
+        int compressionCompleted    = (adv->mode == BMK_decodeOnly);
+        int decompressionCompleted  = (adv->mode == BMK_compressOnly);
+        BMK_benchParams_t cbp, dbp;
+        BMK_initCCtxArgs cctxprep;
+        BMK_initDCtxArgs dctxprep;
+
+        cbp.benchFn       = local_defaultCompress; /* ZSTD_compress2 */
+        cbp.benchPayload  = cctx;
+        cbp.initFn        = local_initCCtx; /* BMK_initCCtx */
+        cbp.initPayload   = &cctxprep;
+        cbp.errorFn       = ZSTD_isError;
+        cbp.blockCount    = nbChunks;
+        cbp.srcBuffers    = srcPtrs;
+        cbp.srcSizes      = srcSizes;
+        cbp.dstBuffers    = cPtrs;
+        cbp.dstCapacities = cCapacities;
+        cbp.blockResults  = cSizes;
+
+        cctxprep.cctx           = cctx;
+        cctxprep.dictBuffer     = dictBuffer;
+        cctxprep.dictBufferSize = dictBufferSize;
+        cctxprep.cLevel         = cLevel;
+        cctxprep.comprParams    = comprParams;
+        cctxprep.adv            = adv;
+
+        dbp.benchFn       = local_defaultDecompress;
+        dbp.benchPayload  = dctx;
+        dbp.initFn        = local_initDCtx;
+        dbp.initPayload   = &dctxprep;
+        dbp.errorFn       = ZSTD_isError;
+        dbp.blockCount    = nbChunks;
+        dbp.srcBuffers    = (const void* const*)cPtrs;
+        dbp.srcSizes      = cSizes;
+        dbp.dstBuffers    = resPtrs;
+        dbp.dstCapacities = resSizes;
+        dbp.blockResults  = NULL;
+
+        dctxprep.dctx           = dctx;
+        dctxprep.dictBuffer     = dictBuffer;
+        dctxprep.dictBufferSize = dictBufferSize;
+
+        OUTPUTLEVEL(2, "\r%70s\r", ""); /* blank line */
+        assert(srcSize < UINT_MAX);
+        OUTPUTLEVEL(
+                2,
+                "%2s-%-17.17s :%10u -> \r",
+                marks[markNb],
+                displayName,
+                (unsigned)srcSize);
+
+        while (!(compressionCompleted && decompressionCompleted)) {
+            if (!compressionCompleted) {
+                BMK_runOutcome_t const cOutcome =
+                        BMK_benchTimedFn(timeStateCompress, cbp);
+
+                if (!BMK_isSuccessful_runOutcome(cOutcome)) {
+                    RETURN_ERROR(30, BMK_benchOutcome_t, "compression error");
+                }
+
+                {
+                    BMK_runTime_t const cResult = BMK_extract_runTime(cOutcome);
+                    cSize                       = cResult.sumOfReturn;
+                    ratio = (double)srcSize / (double)cSize;
+                    {
+                        BMK_benchResult_t newResult;
+                        newResult.cSpeed =
+                                (U64)((double)srcSize * TIMELOOP_NANOSEC
+                                      / cResult.nanoSecPerRun);
+                        benchResult.cSize = cSize;
+                        if (newResult.cSpeed > benchResult.cSpeed)
+                            benchResult.cSpeed = newResult.cSpeed;
+                    }
+                }
+
+                {
+                    int const ratioDigits = 1 + (ratio < 100.) + (ratio < 10.);
+                    assert(cSize < UINT_MAX);
+                    OUTPUTLEVEL(
+                            2,
+                            "%2s-%-17.17s :%10u ->%10u (x%5.*f), %6.*f MB/s \r",
+                            marks[markNb],
+                            displayName,
+                            (unsigned)srcSize,
+                            (unsigned)cSize,
+                            ratioDigits,
+                            ratio,
+                            benchResult.cSpeed < (10 * MB_UNIT) ? 2 : 1,
+                            (double)benchResult.cSpeed / MB_UNIT);
+                }
+                compressionCompleted =
+                        BMK_isCompleted_TimedFn(timeStateCompress);
+            }
+
+            if (!decompressionCompleted) {
+                BMK_runOutcome_t const dOutcome =
+                        BMK_benchTimedFn(timeStateDecompress, dbp);
+
+                if (!BMK_isSuccessful_runOutcome(dOutcome)) {
+                    RETURN_ERROR(30, BMK_benchOutcome_t, "decompression error");
+                }
+
+                {
+                    BMK_runTime_t const dResult = BMK_extract_runTime(dOutcome);
+                    U64 const newDSpeed =
+                            (U64)((double)srcSize * TIMELOOP_NANOSEC
+                                  / dResult.nanoSecPerRun);
+                    if (newDSpeed > benchResult.dSpeed)
+                        benchResult.dSpeed = newDSpeed;
+                }
+
+                {
+                    int const ratioDigits = 1 + (ratio < 100.) + (ratio < 10.);
+                    OUTPUTLEVEL(
+                            2,
+                            "%2s-%-17.17s :%10u ->%10u (x%5.*f), %6.*f MB/s, %6.1f MB/s\r",
+                            marks[markNb],
+                            displayName,
+                            (unsigned)srcSize,
+                            (unsigned)cSize,
+                            ratioDigits,
+                            ratio,
+                            benchResult.cSpeed < (10 * MB_UNIT) ? 2 : 1,
+                            (double)benchResult.cSpeed / MB_UNIT,
+                            (double)benchResult.dSpeed / MB_UNIT);
+                }
+                decompressionCompleted =
+                        BMK_isCompleted_TimedFn(timeStateDecompress);
+            }
+            markNb = (markNb + 1) % NB_MARKS;
+        } /* while (!(compressionCompleted && decompressionCompleted)) */
+
+        /* CRC Checking */
+        {   const BYTE* resultBuffer = (const BYTE*)(*resultBufferPtr);
+            U64 const crcCheck       = XXH64(resultBuffer, srcSize, 0);
+            if ((adv->mode == BMK_both) && (crcOrig != crcCheck)) {
+                size_t u;
+                DISPLAY("!!! WARNING !!! %14s : Invalid Checksum : %x != %x   \n",
+                        displayName,
+                        (unsigned)crcOrig,
+                        (unsigned)crcCheck);
+                for (u = 0; u < srcSize; u++) {
+                    if (((const BYTE*)srcBuffer)[u] != resultBuffer[u]) {
+                        unsigned segNb, bNb, pos;
+                        size_t bacc = 0;
+                        DISPLAY("Decoding error at pos %u ", (unsigned)u);
+                        for (segNb = 0; segNb < nbChunks; segNb++) {
+                            if (bacc + srcSizes[segNb] > u)
+                                break;
+                            bacc += srcSizes[segNb];
+                        }
+                        pos = (U32)(u - bacc);
+                        bNb = pos / (128 KB);
+                        DISPLAY("(sample %u, chunk %u, pos %u) \n",
+                                segNb,
+                                bNb,
+                                pos);
+                        {
+                            size_t const lowest = (u > 5) ? 5 : u;
+                            size_t n;
+                            DISPLAY("origin: ");
+                            for (n = lowest; n > 0; n--)
+                                DISPLAY("%02X ",
+                                        ((const BYTE*)srcBuffer)[u - n]);
+                            DISPLAY(" :%02X:  ", ((const BYTE*)srcBuffer)[u]);
+                            for (n = 1; n < 3; n++)
+                                DISPLAY("%02X ",
+                                        ((const BYTE*)srcBuffer)[u + n]);
+                            DISPLAY(" \n");
+                            DISPLAY("decode: ");
+                            for (n = lowest; n > 0; n--)
+                                DISPLAY("%02X ", resultBuffer[u - n]);
+                            DISPLAY(" :%02X:  ", resultBuffer[u]);
+                            for (n = 1; n < 3; n++)
+                                DISPLAY("%02X ", resultBuffer[u + n]);
+                            DISPLAY(" \n");
+                        }
+                        break;
+                    }
+                    if (u == srcSize - 1) { /* should never happen */
+                        DISPLAY("no difference detected\n");
+                    }
+                } /* for (u=0; u<srcSize; u++) */
+            }     /* if ((adv->mode == BMK_both) && (crcOrig!=crcCheck)) */
+        }         /* CRC Checking */
+
+        if (displayLevel
+            == 1) { /* hidden display mode -q, used by python speed benchmark */
+            double const cSpeed = (double)benchResult.cSpeed / MB_UNIT;
+            double const dSpeed = (double)benchResult.dSpeed / MB_UNIT;
+            if (adv->additionalParam) {
+                OUTPUT("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s (param=%d)\n",
+                       cLevel,
+                       (int)cSize,
+                       ratio,
+                       cSpeed,
+                       dSpeed,
+                       displayName,
+                       adv->additionalParam);
+            } else {
+                OUTPUT("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s\n",
+                       cLevel,
+                       (int)cSize,
+                       ratio,
+                       cSpeed,
+                       dSpeed,
+                       displayName);
+            }
+        }
+
+        OUTPUTLEVEL(2, "%2i#\n", cLevel);
+    } /* Bench */
+
+    benchResult.cMem =
+            (1ULL << (comprParams->windowLog)) + ZSTD_sizeof_CCtx(cctx);
+    return BMK_benchOutcome_setValidResult(benchResult);
+}
+
+BMK_benchOutcome_t BMK_benchMemAdvanced(
+        const void* srcBuffer,
+        size_t srcSize,
+        void* dstBuffer,
+        size_t dstCapacity,
+        const size_t* fileSizes,
+        unsigned nbFiles,
+        int cLevel,
+        const ZSTD_compressionParameters* comprParams,
+        const void* dictBuffer,
+        size_t dictBufferSize,
+        int displayLevel,
+        const char* displayName,
+        const BMK_advancedParams_t* adv)
+
+{
+    int const dstParamsError =
+            !dstBuffer ^ !dstCapacity; /* must be both NULL or none */
+
+    size_t const chunkSize =
+            ((adv->chunkSizeMax >= 32 && (adv->mode != BMK_decodeOnly))
+                     ? adv->chunkSizeMax
+                     : srcSize)
+            + (!srcSize) /* avoid div by 0 */;
+    U32 const nbChunksMax =
+            (U32)((srcSize + (chunkSize - 1)) / chunkSize) + nbFiles;
+
+    const void** const srcPtrs =
+            (const void**)malloc(nbChunksMax * sizeof(void*));
+    size_t* const srcSizes = (size_t*)malloc(nbChunksMax * sizeof(size_t));
+
+    void** const cPtrs        = (void**)malloc(nbChunksMax * sizeof(void*));
+    size_t* const cSizes      = (size_t*)malloc(nbChunksMax * sizeof(size_t));
+    size_t* const cCapacities = (size_t*)malloc(nbChunksMax * sizeof(size_t));
+
+    void** const resPtrs   = (void**)malloc(nbChunksMax * sizeof(void*));
+    size_t* const resSizes = (size_t*)malloc(nbChunksMax * sizeof(size_t));
+
+    BMK_timedFnState_t* timeStateCompress = BMK_createTimedFnState(
+            adv->nbSeconds * 1000, BMK_RUNTEST_DEFAULT_MS);
+    BMK_timedFnState_t* timeStateDecompress = BMK_createTimedFnState(
+            adv->nbSeconds * 1000, BMK_RUNTEST_DEFAULT_MS);
+
+    ZSTD_CCtx* const cctx = ZSTD_createCCtx();
+    ZSTD_DCtx* const dctx = ZSTD_createDCtx();
+
+    const size_t maxCompressedSize = dstCapacity
+            ? dstCapacity
+            : ZSTD_compressBound(srcSize) + (nbChunksMax * 1024);
+
+    void* const internalDstBuffer =
+            dstBuffer ? NULL : malloc(maxCompressedSize);
+    void* const compressedBuffer = dstBuffer ? dstBuffer : internalDstBuffer;
+
+    BMK_benchOutcome_t outcome =
+            BMK_benchOutcome_error(); /* error by default */
+
+    void* resultBuffer = srcSize ? malloc(srcSize) : NULL;
+
+    int const allocationincomplete = !srcPtrs || !srcSizes || !cPtrs || !cSizes
+            || !cCapacities || !resPtrs || !resSizes || !timeStateCompress
+            || !timeStateDecompress || !cctx || !dctx || !compressedBuffer
+            || !resultBuffer;
+
+    if (!allocationincomplete && !dstParamsError) {
+        outcome = BMK_benchMemAdvancedNoAlloc(
+                srcPtrs,
+                srcSizes,
+                cPtrs,
+                cCapacities,
+                cSizes,
+                resPtrs,
+                resSizes,
+                &resultBuffer,
+                compressedBuffer,
+                maxCompressedSize,
+                timeStateCompress,
+                timeStateDecompress,
+                srcBuffer,
+                srcSize,
+                fileSizes,
+                nbFiles,
+                cLevel,
+                comprParams,
+                dictBuffer,
+                dictBufferSize,
+                cctx,
+                dctx,
+                displayLevel,
+                displayName,
+                adv);
+    }
+
+    /* clean up */
+    BMK_freeTimedFnState(timeStateCompress);
+    BMK_freeTimedFnState(timeStateDecompress);
+
+    ZSTD_freeCCtx(cctx);
+    ZSTD_freeDCtx(dctx);
+
+    free(internalDstBuffer);
+    free(resultBuffer);
+
+    free((void*)srcPtrs);
+    free(srcSizes);
+    free(cPtrs);
+    free(cSizes);
+    free(cCapacities);
+    free(resPtrs);
+    free(resSizes);
+
+    if (allocationincomplete) {
+        RETURN_ERROR(
+                31, BMK_benchOutcome_t, "allocation error : not enough memory");
+    }
+
+    if (dstParamsError) {
+        RETURN_ERROR(32, BMK_benchOutcome_t, "Dst parameters not coherent");
+    }
+    return outcome;
+}
+
+BMK_benchOutcome_t BMK_benchMem(
+        const void* srcBuffer,
+        size_t srcSize,
+        const size_t* fileSizes,
+        unsigned nbFiles,
+        int cLevel,
+        const ZSTD_compressionParameters* comprParams,
+        const void* dictBuffer,
+        size_t dictBufferSize,
+        int displayLevel,
+        const char* displayName)
+{
+    BMK_advancedParams_t const adv = BMK_initAdvancedParams();
+    return BMK_benchMemAdvanced(
+            srcBuffer,
+            srcSize,
+            NULL,
+            0,
+            fileSizes,
+            nbFiles,
+            cLevel,
+            comprParams,
+            dictBuffer,
+            dictBufferSize,
+            displayLevel,
+            displayName,
+            &adv);
+}
+
+/* @return: 0 on success, !0 if error */
+static int BMK_benchCLevels(
+        const void* srcBuffer,
+        size_t benchedSize,
+        const size_t* fileSizes,
+        unsigned nbFiles,
+        int startCLevel, int endCLevel,
+        const ZSTD_compressionParameters* comprParams,
+        const void* dictBuffer,
+        size_t dictBufferSize,
+        int displayLevel,
+        const char* displayName,
+        BMK_advancedParams_t const* const adv)
+{
+    int level;
+    const char* pch = strrchr(displayName, '\\'); /* Windows */
+    if (!pch)
+        pch = strrchr(displayName, '/'); /* Linux */
+    if (pch)
+        displayName = pch + 1;
+
+    if (endCLevel > ZSTD_maxCLevel()) {
+        DISPLAYLEVEL(1, "Invalid Compression Level \n");
+        return 15;
+    }
+    if (endCLevel < startCLevel) {
+        DISPLAYLEVEL(1, "Invalid Compression Level Range \n");
+        return 15;
+    }
+
+    if (adv->realTime) {
+        DISPLAYLEVEL(2, "Note : switching to real-time priority \n");
+        SET_REALTIME_PRIORITY;
+    }
+
+    if (displayLevel == 1 && !adv->additionalParam) /* --quiet mode */
+        OUTPUT("bench %s %s: input %u bytes, %u seconds, %u KB chunks\n",
+               ZSTD_VERSION_STRING,
+               ZSTD_GIT_COMMIT_STRING,
+               (unsigned)benchedSize,
+               adv->nbSeconds,
+               (unsigned)(adv->chunkSizeMax >> 10));
+
+    for (level = startCLevel; level <= endCLevel; level++) {
+        BMK_benchOutcome_t res = BMK_benchMemAdvanced(
+            srcBuffer,
+            benchedSize,
+            NULL,
+            0,
+            fileSizes,
+            nbFiles,
+            level,
+            comprParams,
+            dictBuffer,
+            dictBufferSize,
+            displayLevel,
+            displayName,
+            adv);
+        if (!BMK_isSuccessful_benchOutcome(res)) return 1;
+    }
+    return 0;
+}
+
+int BMK_syntheticTest(
+        double compressibility,
+        int startingCLevel, int endCLevel,
+        const ZSTD_compressionParameters* compressionParams,
+        int displayLevel,
+        const BMK_advancedParams_t* adv)
+{
+    char nameBuff[20]        = { 0 };
+    const char* name         = nameBuff;
+    size_t const benchedSize = adv->chunkSizeMax ? adv->chunkSizeMax : 10000000;
+
+    /* Memory allocation */
+    void* const srcBuffer = malloc(benchedSize);
+    if (!srcBuffer) {
+        DISPLAYLEVEL(1, "allocation error : not enough memory \n");
+        return 16;
+    }
+
+    /* Fill input buffer */
+    if (compressibility < 0.0) {
+        LOREM_genBuffer(srcBuffer, benchedSize, 0);
+        name = "Lorem ipsum";
+    } else {
+        RDG_genBuffer(srcBuffer, benchedSize, compressibility, 0.0, 0);
+        formatString_u(
+                nameBuff,
+                sizeof(nameBuff),
+                "Synthetic %u%%",
+                (unsigned)(compressibility * 100));
+    }
+
+    /* Bench */
+    {   int res = BMK_benchCLevels(
+                srcBuffer,
+                benchedSize,
+                &benchedSize,
+                1,
+                startingCLevel, endCLevel,
+                compressionParams,
+                NULL,
+                0, /* dictionary */
+                displayLevel,
+                name,
+                adv);
+        free(srcBuffer);
+        return res;
+    }
+}
+
+static size_t BMK_findMaxMem(U64 requiredMem)
+{
+    size_t const step = 64 MB;
+    BYTE* testmem     = NULL;
+
+    requiredMem = (((requiredMem >> 26) + 1) << 26);
+    requiredMem += step;
+    if (requiredMem > maxMemory)
+        requiredMem = maxMemory;
+
+    do {
+        testmem = (BYTE*)malloc((size_t)requiredMem);
+        requiredMem -= step;
+    } while (!testmem && requiredMem > 0);
+
+    free(testmem);
+    return (size_t)(requiredMem);
+}
+
+/*! BMK_loadFiles() :
+ *  Loads `buffer` with content of files listed within `fileNamesTable`.
+ *  At most, fills `buffer` entirely. */
+static int BMK_loadFiles(
+        void* buffer,
+        size_t bufferSize,
+        size_t* fileSizes,
+        const char* const* fileNamesTable,
+        unsigned nbFiles,
+        int displayLevel)
+{
+    size_t pos = 0, totalSize = 0;
+    unsigned n;
+    for (n = 0; n < nbFiles; n++) {
+        const char* const filename = fileNamesTable[n];
+        U64 fileSize = UTIL_getFileSize(
+                filename); /* last file may be shortened */
+        if (UTIL_isDirectory(filename)) {
+            DISPLAYLEVEL(
+                    2, "Ignoring %s directory...       \n", filename);
+            fileSizes[n] = 0;
+            continue;
+        }
+        if (fileSize == UTIL_FILESIZE_UNKNOWN) {
+            DISPLAYLEVEL(
+                    2,
+                    "Cannot evaluate size of %s, ignoring ... \n",
+                    filename);
+            fileSizes[n] = 0;
+            continue;
+        }
+        if (fileSize > bufferSize - pos) {
+            /* buffer too small - limit quantity loaded */
+            fileSize = bufferSize - pos;
+            nbFiles  = n; /* stop after this file */
+        }
+
+        {   FILE* const f = fopen(filename, "rb");
+            if (f == NULL) {
+                RETURN_ERROR_INT(
+                        10, "cannot open file %s", filename);
+            }
+            OUTPUTLEVEL(2, "Loading %s...       \r", filename);
+            {   size_t const readSize =
+                        fread(((char*)buffer) + pos, 1, (size_t)fileSize, f);
+                if (readSize != (size_t)fileSize) {
+                    fclose(f);
+                    RETURN_ERROR_INT(
+                            11, "invalid read %s", filename);
+                }
+                pos += readSize;
+            }
+            fileSizes[n] = (size_t)fileSize;
+            totalSize += (size_t)fileSize;
+            fclose(f);
+        }
+    }
+
+    if (totalSize == 0)
+        RETURN_ERROR_INT(12, "no data to bench");
+    return 0;
+}
+
+int BMK_benchFilesAdvanced(
+        const char* const* fileNamesTable,
+        unsigned nbFiles,
+        const char* dictFileName,
+        int startCLevel, int endCLevel,
+        const ZSTD_compressionParameters* compressionParams,
+        int displayLevel,
+        const BMK_advancedParams_t* adv)
+{
+    void* srcBuffer = NULL;
+    size_t benchedSize;
+    void* dictBuffer      = NULL;
+    size_t dictBufferSize = 0;
+    size_t* fileSizes     = NULL;
+    int res = 1;
+    U64 const totalSizeToLoad = UTIL_getTotalFileSize(fileNamesTable, nbFiles);
+
+    if (!nbFiles) {
+        DISPLAYLEVEL(1, "No Files to Benchmark");
+        return 13;
+    }
+
+    if (endCLevel > ZSTD_maxCLevel()) {
+        DISPLAYLEVEL(1, "Invalid Compression Level");
+        return 14;
+    }
+
+    if (totalSizeToLoad == UTIL_FILESIZE_UNKNOWN) {
+        DISPLAYLEVEL(1, "Error loading files");
+        return 15;
+    }
+
+    fileSizes = (size_t*)calloc(nbFiles, sizeof(size_t));
+    if (!fileSizes) {
+        DISPLAYLEVEL(1, "not enough memory for fileSizes");
+        return 16;
+    }
+
+    /* Load dictionary */
+    if (dictFileName != NULL) {
+        U64 const dictFileSize = UTIL_getFileSize(dictFileName);
+        if (dictFileSize == UTIL_FILESIZE_UNKNOWN) {
+            DISPLAYLEVEL(
+                    1,
+                    "error loading %s : %s \n",
+                    dictFileName,
+                    strerror(errno));
+            free(fileSizes);
+            DISPLAYLEVEL(1, "benchmark aborted");
+            return 17;
+        }
+        if (dictFileSize > 64 MB) {
+            free(fileSizes);
+            DISPLAYLEVEL(1, "dictionary file %s too large", dictFileName);
+            return 18;
+        }
+        dictBufferSize = (size_t)dictFileSize;
+        dictBuffer     = malloc(dictBufferSize);
+        if (dictBuffer == NULL) {
+            free(fileSizes);
+            DISPLAYLEVEL(
+                    1,
+                    "not enough memory for dictionary (%u bytes)",
+                    (unsigned)dictBufferSize);
+            return 19;
+        }
+
+        {
+            int const errorCode = BMK_loadFiles(
+                    dictBuffer,
+                    dictBufferSize,
+                    fileSizes,
+                    &dictFileName /*?*/,
+                    1 /*?*/,
+                    displayLevel);
+            if (errorCode) {
+                goto _cleanUp;
+            }
+        }
+    }
+
+    /* Memory allocation & restrictions */
+    benchedSize = BMK_findMaxMem(totalSizeToLoad * 3) / 3;
+    if ((U64)benchedSize > totalSizeToLoad)
+        benchedSize = (size_t)totalSizeToLoad;
+    if (benchedSize < totalSizeToLoad)
+        DISPLAY("Not enough memory; testing %u MB only...\n",
+                (unsigned)(benchedSize >> 20));
+
+    srcBuffer = benchedSize ? malloc(benchedSize) : NULL;
+    if (!srcBuffer) {
+        free(dictBuffer);
+        free(fileSizes);
+        DISPLAYLEVEL(1, "not enough memory for srcBuffer");
+        return 20;
+    }
+
+    /* Load input buffer */
+    {
+        int const errorCode = BMK_loadFiles(
+                srcBuffer,
+                benchedSize,
+                fileSizes,
+                fileNamesTable,
+                nbFiles,
+                displayLevel);
+        if (errorCode) {
+            goto _cleanUp;
+        }
+    }
+
+    /* Bench */
+    {
+        char mfName[20] = { 0 };
+        formatString_u(mfName, sizeof(mfName), " %u files", nbFiles);
+        {   const char* const displayName =
+                    (nbFiles > 1) ? mfName : fileNamesTable[0];
+            res = BMK_benchCLevels(
+                    srcBuffer,
+                    benchedSize,
+                    fileSizes,
+                    nbFiles,
+                    startCLevel, endCLevel,
+                    compressionParams,
+                    dictBuffer,
+                    dictBufferSize,
+                    displayLevel,
+                    displayName,
+                    adv);
+        }
+    }
+
+_cleanUp:
+    free(srcBuffer);
+    free(dictBuffer);
+    free(fileSizes);
+    return res;
+}
+
+int BMK_benchFiles(
+        const char* const* fileNamesTable,
+        unsigned nbFiles,
+        const char* dictFileName,
+        int cLevel,
+        const ZSTD_compressionParameters* compressionParams,
+        int displayLevel)
+{
+    BMK_advancedParams_t const adv = BMK_initAdvancedParams();
+    return BMK_benchFilesAdvanced(
+            fileNamesTable,
+            nbFiles,
+            dictFileName,
+            cLevel, cLevel,
+            compressionParams,
+            displayLevel,
+            &adv);
+}
diff --git a/programs/dibio.c b/programs/dibio.c
new file mode 100644
index 0000000..63c455a
--- /dev/null
+++ b/programs/dibio.c
@@ -0,0 +1,440 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+
+/* **************************************
+*  Compiler Warnings
+****************************************/
+#ifdef _MSC_VER
+#  pragma warning(disable : 4127)    /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/*-*************************************
+*  Includes
+***************************************/
+#include "platform.h"       /* Large Files support */
+#include "util.h"           /* UTIL_getFileSize, UTIL_getTotalFileSize */
+#include <stdlib.h>         /* malloc, free */
+#include <string.h>         /* memset */
+#include <stdio.h>          /* fprintf, fopen, ftello64 */
+#include <errno.h>          /* errno */
+
+#include "timefn.h"         /* UTIL_time_t, UTIL_clockSpanMicro, UTIL_getTime */
+#include "../lib/common/debug.h" /* assert */
+#include "../lib/common/mem.h"  /* read */
+#include "../lib/zstd_errors.h"
+#include "dibio.h"
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define SAMPLESIZE_MAX (128 KB)
+#define MEMMULT 11    /* rough estimation : memory cost to analyze 1 byte of sample */
+#define COVER_MEMMULT 9    /* rough estimation : memory cost to analyze 1 byte of sample */
+#define FASTCOVER_MEMMULT 1    /* rough estimation : memory cost to analyze 1 byte of sample */
+static const size_t g_maxMemory = (sizeof(size_t) == 4) ? (2 GB - 64 MB) : ((size_t)(512 MB) << sizeof(size_t));
+
+#define NOISELENGTH 32
+#define MAX_SAMPLES_SIZE (2 GB) /* training dataset limited to 2GB */
+
+
+/*-*************************************
+*  Console display
+***************************************/
+#define DISPLAY(...)         fprintf(stderr, __VA_ARGS__)
+#define DISPLAYLEVEL(l, ...) if (displayLevel>=l) { DISPLAY(__VA_ARGS__); }
+
+static const U64 g_refreshRate = SEC_TO_MICRO / 6;
+static UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
+
+#define DISPLAYUPDATE(l, ...) { if (displayLevel>=l) { \
+            if ((UTIL_clockSpanMicro(g_displayClock) > g_refreshRate) || (displayLevel>=4)) \
+            { g_displayClock = UTIL_getTime(); DISPLAY(__VA_ARGS__); \
+            if (displayLevel>=4) fflush(stderr); } } }
+
+/*-*************************************
+*  Exceptions
+***************************************/
+#ifndef DEBUG
+#  define DEBUG 0
+#endif
+#define DEBUGOUTPUT(...) if (DEBUG) DISPLAY(__VA_ARGS__);
+#define EXM_THROW(error, ...)                                             \
+{                                                                         \
+    DEBUGOUTPUT("Error defined at %s, line %i : \n", __FILE__, __LINE__); \
+    DISPLAY("Error %i : ", error);                                        \
+    DISPLAY(__VA_ARGS__);                                                 \
+    DISPLAY("\n");                                                        \
+    exit(error);                                                          \
+}
+
+
+/* ********************************************************
+*  Helper functions
+**********************************************************/
+#undef MIN
+#define MIN(a,b)    ((a) < (b) ? (a) : (b))
+
+/**
+  Returns the size of a file.
+  If error returns -1.
+*/
+static S64 DiB_getFileSize (const char * fileName)
+{
+    U64 const fileSize = UTIL_getFileSize(fileName);
+    return (fileSize == UTIL_FILESIZE_UNKNOWN) ? -1 : (S64)fileSize;
+}
+
+/* ********************************************************
+*  File related operations
+**********************************************************/
+/** DiB_loadFiles() :
+ *  load samples from files listed in fileNamesTable into buffer.
+ *  works even if buffer is too small to load all samples.
+ *  Also provides the size of each sample into sampleSizes table
+ *  which must be sized correctly, using DiB_fileStats().
+ * @return : nb of samples effectively loaded into `buffer`
+ * *bufferSizePtr is modified, it provides the amount data loaded within buffer.
+ *  sampleSizes is filled with the size of each sample.
+ */
+static int DiB_loadFiles(
+    void* buffer, size_t* bufferSizePtr,
+    size_t* sampleSizes, int sstSize,
+    const char** fileNamesTable, int nbFiles,
+    size_t targetChunkSize, int displayLevel )
+{
+    char* const buff = (char*)buffer;
+    size_t totalDataLoaded = 0;
+    int nbSamplesLoaded = 0;
+    int fileIndex = 0;
+    FILE * f = NULL;
+
+    assert(targetChunkSize <= SAMPLESIZE_MAX);
+
+    while ( nbSamplesLoaded < sstSize && fileIndex < nbFiles ) {
+        size_t fileDataLoaded;
+        S64 const fileSize = DiB_getFileSize(fileNamesTable[fileIndex]);
+        if (fileSize <= 0) {
+            /* skip if zero-size or file error */
+            ++fileIndex;
+            continue;
+        }
+
+        f = fopen( fileNamesTable[fileIndex], "rb");
+        if (f == NULL)
+            EXM_THROW(10, "zstd: dictBuilder: %s %s ", fileNamesTable[fileIndex], strerror(errno));
+        DISPLAYUPDATE(2, "Loading %s...       \r", fileNamesTable[fileIndex]);
+
+        /* Load the first chunk of data from the file */
+        fileDataLoaded = targetChunkSize > 0 ?
+                            (size_t)MIN(fileSize, (S64)targetChunkSize) :
+                            (size_t)MIN(fileSize, SAMPLESIZE_MAX );
+        if (totalDataLoaded + fileDataLoaded > *bufferSizePtr)
+            break;
+        if (fread( buff+totalDataLoaded, 1, fileDataLoaded, f ) != fileDataLoaded)
+            EXM_THROW(11, "Pb reading %s", fileNamesTable[fileIndex]);
+        sampleSizes[nbSamplesLoaded++] = fileDataLoaded;
+        totalDataLoaded += fileDataLoaded;
+
+        /* If file-chunking is enabled, load the rest of the file as more samples */
+        if (targetChunkSize > 0) {
+            while( (S64)fileDataLoaded < fileSize && nbSamplesLoaded < sstSize ) {
+                size_t const chunkSize = MIN((size_t)(fileSize-fileDataLoaded), targetChunkSize);
+                if (totalDataLoaded + chunkSize > *bufferSizePtr) /* buffer is full */
+                    break;
+
+                if (fread( buff+totalDataLoaded, 1, chunkSize, f ) != chunkSize)
+                    EXM_THROW(11, "Pb reading %s", fileNamesTable[fileIndex]);
+                sampleSizes[nbSamplesLoaded++] = chunkSize;
+                totalDataLoaded += chunkSize;
+                fileDataLoaded += chunkSize;
+            }
+        }
+        fileIndex += 1;
+        fclose(f); f = NULL;
+    }
+    if (f != NULL)
+        fclose(f);
+
+    DISPLAYLEVEL(2, "\r%79s\r", "");
+    DISPLAYLEVEL(4, "Loaded %d KB total training data, %d nb samples \n",
+        (int)(totalDataLoaded / (1 KB)), nbSamplesLoaded );
+    *bufferSizePtr = totalDataLoaded;
+    return nbSamplesLoaded;
+}
+
+#define DiB_rotl32(x,r) ((x << r) | (x >> (32 - r)))
+static U32 DiB_rand(U32* src)
+{
+    static const U32 prime1 = 2654435761U;
+    static const U32 prime2 = 2246822519U;
+    U32 rand32 = *src;
+    rand32 *= prime1;
+    rand32 ^= prime2;
+    rand32  = DiB_rotl32(rand32, 13);
+    *src = rand32;
+    return rand32 >> 5;
+}
+
+/* DiB_shuffle() :
+ * shuffle a table of file names in a semi-random way
+ * It improves dictionary quality by reducing "locality" impact, so if sample set is very large,
+ * it will load random elements from it, instead of just the first ones. */
+static void DiB_shuffle(const char** fileNamesTable, unsigned nbFiles) {
+    U32 seed = 0xFD2FB528;
+    unsigned i;
+    if (nbFiles == 0)
+        return;
+    for (i = nbFiles - 1; i > 0; --i) {
+        unsigned const j = DiB_rand(&seed) % (i + 1);
+        const char* const tmp = fileNamesTable[j];
+        fileNamesTable[j] = fileNamesTable[i];
+        fileNamesTable[i] = tmp;
+    }
+}
+
+
+/*-********************************************************
+*  Dictionary training functions
+**********************************************************/
+static size_t DiB_findMaxMem(unsigned long long requiredMem)
+{
+    size_t const step = 8 MB;
+    void* testmem = NULL;
+
+    requiredMem = (((requiredMem >> 23) + 1) << 23);
+    requiredMem += step;
+    if (requiredMem > g_maxMemory) requiredMem = g_maxMemory;
+
+    while (!testmem) {
+        testmem = malloc((size_t)requiredMem);
+        requiredMem -= step;
+    }
+
+    free(testmem);
+    return (size_t)requiredMem;
+}
+
+
+static void DiB_fillNoise(void* buffer, size_t length)
+{
+    unsigned const prime1 = 2654435761U;
+    unsigned const prime2 = 2246822519U;
+    unsigned acc = prime1;
+    size_t p=0;
+
+    for (p=0; p<length; p++) {
+        acc *= prime2;
+        ((unsigned char*)buffer)[p] = (unsigned char)(acc >> 21);
+    }
+}
+
+
+static void DiB_saveDict(const char* dictFileName,
+                         const void* buff, size_t buffSize)
+{
+    FILE* const f = fopen(dictFileName, "wb");
+    if (f==NULL) EXM_THROW(3, "cannot open %s ", dictFileName);
+
+    { size_t const n = fwrite(buff, 1, buffSize, f);
+      if (n!=buffSize) EXM_THROW(4, "%s : write error", dictFileName) }
+
+    { size_t const n = (size_t)fclose(f);
+      if (n!=0) EXM_THROW(5, "%s : flush error", dictFileName) }
+}
+
+typedef struct {
+    S64 totalSizeToLoad;
+    int nbSamples;
+    int oneSampleTooLarge;
+} fileStats;
+
+/*! DiB_fileStats() :
+ *  Given a list of files, and a chunkSize (0 == no chunk, whole files)
+ *  provides the amount of data to be loaded and the resulting nb of samples.
+ *  This is useful primarily for allocation purpose => sample buffer, and sample sizes table.
+ */
+static fileStats DiB_fileStats(const char** fileNamesTable, int nbFiles, size_t chunkSize, int displayLevel)
+{
+    fileStats fs;
+    int n;
+    memset(&fs, 0, sizeof(fs));
+
+    /* We assume that if chunking is requested, the chunk size is < SAMPLESIZE_MAX */
+    assert( chunkSize <= SAMPLESIZE_MAX );
+
+    for (n=0; n<nbFiles; n++) {
+      S64 const fileSize = DiB_getFileSize(fileNamesTable[n]);
+      /* TODO: is there a minimum sample size? What if the file is 1-byte? */
+      if (fileSize == 0) {
+        DISPLAYLEVEL(3, "Sample file '%s' has zero size, skipping...\n", fileNamesTable[n]);
+        continue;
+      }
+
+      /* the case where we are breaking up files in sample chunks */
+      if (chunkSize > 0) {
+        /* TODO: is there a minimum sample size? Can we have a 1-byte sample? */
+        fs.nbSamples += (int)((fileSize + chunkSize-1) / chunkSize);
+        fs.totalSizeToLoad += fileSize;
+      }
+      else {
+      /* the case where one file is one sample */
+        if (fileSize > SAMPLESIZE_MAX) {
+          /* flag excessively large sample files */
+          fs.oneSampleTooLarge |= (fileSize > 2*SAMPLESIZE_MAX);
+
+          /* Limit to the first SAMPLESIZE_MAX (128kB) of the file */
+          DISPLAYLEVEL(3, "Sample file '%s' is too large, limiting to %d KB\n",
+              fileNamesTable[n], SAMPLESIZE_MAX / (1 KB));
+        }
+        fs.nbSamples += 1;
+        fs.totalSizeToLoad += MIN(fileSize, SAMPLESIZE_MAX);
+      }
+    }
+    DISPLAYLEVEL(4, "Found training data %d files, %d KB, %d samples\n", nbFiles, (int)(fs.totalSizeToLoad / (1 KB)), fs.nbSamples);
+    return fs;
+}
+
+int DiB_trainFromFiles(const char* dictFileName, size_t maxDictSize,
+                       const char** fileNamesTable, int nbFiles, size_t chunkSize,
+                       ZDICT_legacy_params_t* params, ZDICT_cover_params_t* coverParams,
+                       ZDICT_fastCover_params_t* fastCoverParams, int optimize, unsigned memLimit)
+{
+    fileStats fs;
+    size_t* sampleSizes; /* vector of sample sizes. Each sample can be up to SAMPLESIZE_MAX */
+    int nbSamplesLoaded; /* nb of samples effectively loaded in srcBuffer */
+    size_t loadedSize; /* total data loaded in srcBuffer for all samples */
+    void* srcBuffer /* contiguous buffer with training data/samples */;
+    void* const dictBuffer = malloc(maxDictSize);
+    int result = 0;
+
+    int const displayLevel = params ? params->zParams.notificationLevel :
+        coverParams ? coverParams->zParams.notificationLevel :
+        fastCoverParams ? fastCoverParams->zParams.notificationLevel : 0;
+
+    /* Shuffle input files before we start assessing how much sample datA to load.
+       The purpose of the shuffle is to pick random samples when the sample
+       set is larger than what we can load in memory. */
+    DISPLAYLEVEL(3, "Shuffling input files\n");
+    DiB_shuffle(fileNamesTable, nbFiles);
+
+    /* Figure out how much sample data to load with how many samples */
+    fs = DiB_fileStats(fileNamesTable, nbFiles, chunkSize, displayLevel);
+
+    {
+        int const memMult = params ? MEMMULT :
+                            coverParams ? COVER_MEMMULT:
+                            FASTCOVER_MEMMULT;
+        size_t const maxMem =  DiB_findMaxMem(fs.totalSizeToLoad * memMult) / memMult;
+        /* Limit the size of the training data to the free memory */
+        /* Limit the size of the training data to 2GB */
+        /* TODO: there is opportunity to stop DiB_fileStats() early when the data limit is reached */
+        loadedSize = (size_t)MIN( MIN((S64)maxMem, fs.totalSizeToLoad), MAX_SAMPLES_SIZE );
+        if (memLimit != 0) {
+            DISPLAYLEVEL(2, "!  Warning : setting manual memory limit for dictionary training data at %u MB \n",
+                (unsigned)(memLimit / (1 MB)));
+            loadedSize = (size_t)MIN(loadedSize, memLimit);
+        }
+        srcBuffer = malloc(loadedSize+NOISELENGTH);
+        sampleSizes = (size_t*)malloc(fs.nbSamples * sizeof(size_t));
+    }
+
+    /* Checks */
+    if ((fs.nbSamples && !sampleSizes) || (!srcBuffer) || (!dictBuffer))
+        EXM_THROW(12, "not enough memory for DiB_trainFiles");   /* should not happen */
+    if (fs.oneSampleTooLarge) {
+        DISPLAYLEVEL(2, "!  Warning : some sample(s) are very large \n");
+        DISPLAYLEVEL(2, "!  Note that dictionary is only useful for small samples. \n");
+        DISPLAYLEVEL(2, "!  As a consequence, only the first %u bytes of each sample are loaded \n", SAMPLESIZE_MAX);
+    }
+    if (fs.nbSamples < 5) {
+        DISPLAYLEVEL(2, "!  Warning : nb of samples too low for proper processing !\n");
+        DISPLAYLEVEL(2, "!  Please provide _one file per sample_.\n");
+        DISPLAYLEVEL(2, "!  Alternatively, split file(s) into fixed-size samples, with --split=#\n");
+        EXM_THROW(14, "nb of samples too low");   /* we now clearly forbid this case */
+    }
+    if (fs.totalSizeToLoad < (S64)maxDictSize * 8) {
+        DISPLAYLEVEL(2, "!  Warning : data size of samples too small for target dictionary size \n");
+        DISPLAYLEVEL(2, "!  Samples should be about 100x larger than target dictionary size \n");
+    }
+
+    /* init */
+    if ((S64)loadedSize < fs.totalSizeToLoad)
+        DISPLAYLEVEL(1, "Training samples set too large (%u MB); training on %u MB only...\n",
+            (unsigned)(fs.totalSizeToLoad / (1 MB)),
+            (unsigned)(loadedSize / (1 MB)));
+
+    /* Load input buffer */
+    nbSamplesLoaded = DiB_loadFiles(
+        srcBuffer, &loadedSize, sampleSizes, fs.nbSamples, fileNamesTable,
+        nbFiles, chunkSize, displayLevel);
+
+    {   size_t dictSize = ZSTD_error_GENERIC;
+        if (params) {
+            DiB_fillNoise((char*)srcBuffer + loadedSize, NOISELENGTH);   /* guard band, for end of buffer condition */
+            dictSize = ZDICT_trainFromBuffer_legacy(dictBuffer, maxDictSize,
+                                                    srcBuffer, sampleSizes, nbSamplesLoaded,
+                                                    *params);
+        } else if (coverParams) {
+            if (optimize) {
+              dictSize = ZDICT_optimizeTrainFromBuffer_cover(dictBuffer, maxDictSize,
+                                                             srcBuffer, sampleSizes, nbSamplesLoaded,
+                                                             coverParams);
+              if (!ZDICT_isError(dictSize)) {
+                  unsigned splitPercentage = (unsigned)(coverParams->splitPoint * 100);
+                  DISPLAYLEVEL(2, "k=%u\nd=%u\nsteps=%u\nsplit=%u\n", coverParams->k, coverParams->d,
+                              coverParams->steps, splitPercentage);
+              }
+            } else {
+              dictSize = ZDICT_trainFromBuffer_cover(dictBuffer, maxDictSize, srcBuffer,
+                                                     sampleSizes, nbSamplesLoaded, *coverParams);
+            }
+        } else if (fastCoverParams != NULL) {
+            if (optimize) {
+              dictSize = ZDICT_optimizeTrainFromBuffer_fastCover(dictBuffer, maxDictSize,
+                                                              srcBuffer, sampleSizes, nbSamplesLoaded,
+                                                              fastCoverParams);
+              if (!ZDICT_isError(dictSize)) {
+                unsigned splitPercentage = (unsigned)(fastCoverParams->splitPoint * 100);
+                DISPLAYLEVEL(2, "k=%u\nd=%u\nf=%u\nsteps=%u\nsplit=%u\naccel=%u\n", fastCoverParams->k,
+                            fastCoverParams->d, fastCoverParams->f, fastCoverParams->steps, splitPercentage,
+                            fastCoverParams->accel);
+              }
+            } else {
+              dictSize = ZDICT_trainFromBuffer_fastCover(dictBuffer, maxDictSize, srcBuffer,
+                                                        sampleSizes, nbSamplesLoaded, *fastCoverParams);
+            }
+        } else {
+            assert(0 /* Impossible */);
+        }
+        if (ZDICT_isError(dictSize)) {
+            DISPLAYLEVEL(1, "dictionary training failed : %s \n", ZDICT_getErrorName(dictSize));   /* should not happen */
+            result = 1;
+            goto _cleanup;
+        }
+        /* save dict */
+        DISPLAYLEVEL(2, "Save dictionary of size %u into file %s \n", (unsigned)dictSize, dictFileName);
+        DiB_saveDict(dictFileName, dictBuffer, dictSize);
+    }
+
+    /* clean up */
+_cleanup:
+    free(srcBuffer);
+    free(sampleSizes);
+    free(dictBuffer);
+    return result;
+}
diff --git a/programs/fileio.c b/programs/fileio.c
new file mode 100644
index 0000000..77f3d35
--- /dev/null
+++ b/programs/fileio.c
@@ -0,0 +1,3473 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/* *************************************
+*  Compiler Options
+***************************************/
+#ifdef _MSC_VER   /* Visual */
+#  pragma warning(disable : 4127)  /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4204)  /* non-constant aggregate initializer */
+#endif
+#if defined(__MINGW32__) && !defined(_POSIX_SOURCE)
+#  define _POSIX_SOURCE 1          /* disable %llu warnings with MinGW on Windows */
+#endif
+
+/*-*************************************
+*  Includes
+***************************************/
+#include "platform.h"   /* Large Files support, SET_BINARY_MODE */
+#include "util.h"       /* UTIL_getFileSize, UTIL_isRegularFile, UTIL_isSameFile */
+#include <stdio.h>      /* fprintf, open, fdopen, fread, _fileno, stdin, stdout */
+#include <stdlib.h>     /* malloc, free */
+#include <string.h>     /* strcmp, strlen */
+#include <time.h>       /* clock_t, to measure process time */
+#include <fcntl.h>      /* O_WRONLY */
+#include <assert.h>
+#include <errno.h>      /* errno */
+#include <limits.h>     /* INT_MAX */
+#include <signal.h>
+#include "timefn.h"     /* UTIL_getTime, UTIL_clockSpanMicro */
+
+#if defined (_MSC_VER)
+#  include <sys/stat.h>
+#  include <io.h>
+#endif
+
+#include "fileio.h"
+#include "fileio_asyncio.h"
+#include "fileio_common.h"
+
+FIO_display_prefs_t g_display_prefs = {2, FIO_ps_auto};
+UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
+
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_magicNumber, ZSTD_frameHeaderSize_max */
+#include "../lib/zstd.h"
+#include "../lib/zstd_errors.h"  /* ZSTD_error_frameParameter_windowTooLarge */
+
+#if defined(ZSTD_GZCOMPRESS) || defined(ZSTD_GZDECOMPRESS)
+#  include <zlib.h>
+#  if !defined(z_const)
+#    define z_const
+#  endif
+#endif
+
+#if defined(ZSTD_LZMACOMPRESS) || defined(ZSTD_LZMADECOMPRESS)
+#  include <lzma.h>
+#endif
+
+#define LZ4_MAGICNUMBER 0x184D2204
+#if defined(ZSTD_LZ4COMPRESS) || defined(ZSTD_LZ4DECOMPRESS)
+#  define LZ4F_ENABLE_OBSOLETE_ENUMS
+#  include <lz4frame.h>
+#  include <lz4.h>
+#endif
+
+char const* FIO_zlibVersion(void)
+{
+#if defined(ZSTD_GZCOMPRESS) || defined(ZSTD_GZDECOMPRESS)
+    return zlibVersion();
+#else
+    return "Unsupported";
+#endif
+}
+
+char const* FIO_lz4Version(void)
+{
+#if defined(ZSTD_LZ4COMPRESS) || defined(ZSTD_LZ4DECOMPRESS)
+    /* LZ4_versionString() added in v1.7.3 */
+#   if LZ4_VERSION_NUMBER >= 10703
+        return LZ4_versionString();
+#   else
+#       define ZSTD_LZ4_VERSION LZ4_VERSION_MAJOR.LZ4_VERSION_MINOR.LZ4_VERSION_RELEASE
+#       define ZSTD_LZ4_VERSION_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_LZ4_VERSION)
+        return ZSTD_LZ4_VERSION_STRING;
+#   endif
+#else
+    return "Unsupported";
+#endif
+}
+
+char const* FIO_lzmaVersion(void)
+{
+#if defined(ZSTD_LZMACOMPRESS) || defined(ZSTD_LZMADECOMPRESS)
+    return lzma_version_string();
+#else
+    return "Unsupported";
+#endif
+}
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define ADAPT_WINDOWLOG_DEFAULT 23   /* 8 MB */
+#define DICTSIZE_MAX (32 MB)   /* protection against large input (attack scenario) */
+
+#define FNSPACE 30
+
+/* Default file permissions 0666 (modulated by umask) */
+/* Temporary restricted file permissions are used when we're going to
+ * chmod/chown at the end of the operation. */
+#if !defined(_WIN32)
+/* These macros aren't defined on windows. */
+#define DEFAULT_FILE_PERMISSIONS (S_IRUSR|S_IWUSR|S_IRGRP|S_IWGRP|S_IROTH|S_IWOTH)
+#define TEMPORARY_FILE_PERMISSIONS (S_IRUSR|S_IWUSR)
+#else
+#define DEFAULT_FILE_PERMISSIONS (0666)
+#define TEMPORARY_FILE_PERMISSIONS (0600)
+#endif
+
+/*-************************************
+*  Signal (Ctrl-C trapping)
+**************************************/
+static const char* g_artefact = NULL;
+static void INThandler(int sig)
+{
+    assert(sig==SIGINT); (void)sig;
+#if !defined(_MSC_VER)
+    signal(sig, SIG_IGN);  /* this invocation generates a buggy warning in Visual Studio */
+#endif
+    if (g_artefact) {
+        assert(UTIL_isRegularFile(g_artefact));
+        remove(g_artefact);
+    }
+    DISPLAY("\n");
+    exit(2);
+}
+static void addHandler(char const* dstFileName)
+{
+    if (UTIL_isRegularFile(dstFileName)) {
+        g_artefact = dstFileName;
+        signal(SIGINT, INThandler);
+    } else {
+        g_artefact = NULL;
+    }
+}
+/* Idempotent */
+static void clearHandler(void)
+{
+    if (g_artefact) signal(SIGINT, SIG_DFL);
+    g_artefact = NULL;
+}
+
+
+/*-*********************************************************
+*  Termination signal trapping (Print debug stack trace)
+***********************************************************/
+#if defined(__has_feature) && !defined(BACKTRACE_ENABLE) /* Clang compiler */
+#  if (__has_feature(address_sanitizer))
+#    define BACKTRACE_ENABLE 0
+#  endif /* __has_feature(address_sanitizer) */
+#elif defined(__SANITIZE_ADDRESS__) && !defined(BACKTRACE_ENABLE) /* GCC compiler */
+#  define BACKTRACE_ENABLE 0
+#endif
+
+#if !defined(BACKTRACE_ENABLE)
+/* automatic detector : backtrace enabled by default on linux+glibc and osx */
+#  if (defined(__linux__) && (defined(__GLIBC__) && !defined(__UCLIBC__))) \
+     || (defined(__APPLE__) && defined(__MACH__))
+#    define BACKTRACE_ENABLE 1
+#  else
+#    define BACKTRACE_ENABLE 0
+#  endif
+#endif
+
+/* note : after this point, BACKTRACE_ENABLE is necessarily defined */
+
+
+#if BACKTRACE_ENABLE
+
+#include <execinfo.h>   /* backtrace, backtrace_symbols */
+
+#define MAX_STACK_FRAMES    50
+
+static void ABRThandler(int sig) {
+    const char* name;
+    void* addrlist[MAX_STACK_FRAMES];
+    char** symbollist;
+    int addrlen, i;
+
+    switch (sig) {
+        case SIGABRT: name = "SIGABRT"; break;
+        case SIGFPE: name = "SIGFPE"; break;
+        case SIGILL: name = "SIGILL"; break;
+        case SIGINT: name = "SIGINT"; break;
+        case SIGSEGV: name = "SIGSEGV"; break;
+        default: name = "UNKNOWN";
+    }
+
+    DISPLAY("Caught %s signal, printing stack:\n", name);
+    /* Retrieve current stack addresses. */
+    addrlen = backtrace(addrlist, MAX_STACK_FRAMES);
+    if (addrlen == 0) {
+        DISPLAY("\n");
+        return;
+    }
+    /* Create readable strings to each frame. */
+    symbollist = backtrace_symbols(addrlist, addrlen);
+    /* Print the stack trace, excluding calls handling the signal. */
+    for (i = ZSTD_START_SYMBOLLIST_FRAME; i < addrlen; i++) {
+        DISPLAY("%s\n", symbollist[i]);
+    }
+    free(symbollist);
+    /* Reset and raise the signal so default handler runs. */
+    signal(sig, SIG_DFL);
+    raise(sig);
+}
+#endif
+
+void FIO_addAbortHandler(void)
+{
+#if BACKTRACE_ENABLE
+    signal(SIGABRT, ABRThandler);
+    signal(SIGFPE, ABRThandler);
+    signal(SIGILL, ABRThandler);
+    signal(SIGSEGV, ABRThandler);
+    signal(SIGBUS, ABRThandler);
+#endif
+}
+
+/*-*************************************
+*  Parameters: FIO_ctx_t
+***************************************/
+
+/* typedef'd to FIO_ctx_t within fileio.h */
+struct FIO_ctx_s {
+
+    /* file i/o info */
+    int nbFilesTotal;
+    int hasStdinInput;
+    int hasStdoutOutput;
+
+    /* file i/o state */
+    int currFileIdx;
+    int nbFilesProcessed;
+    size_t totalBytesInput;
+    size_t totalBytesOutput;
+};
+
+static int FIO_shouldDisplayFileSummary(FIO_ctx_t const* fCtx)
+{
+    return fCtx->nbFilesTotal <= 1 || g_display_prefs.displayLevel >= 3;
+}
+
+static int FIO_shouldDisplayMultipleFileSummary(FIO_ctx_t const* fCtx)
+{
+    int const shouldDisplay = (fCtx->nbFilesProcessed >= 1 && fCtx->nbFilesTotal > 1);
+    assert(shouldDisplay || FIO_shouldDisplayFileSummary(fCtx) || fCtx->nbFilesProcessed == 0);
+    return shouldDisplay;
+}
+
+
+/*-*************************************
+*  Parameters: Initialization
+***************************************/
+
+#define FIO_OVERLAP_LOG_NOTSET 9999
+#define FIO_LDM_PARAM_NOTSET 9999
+
+
+FIO_prefs_t* FIO_createPreferences(void)
+{
+    FIO_prefs_t* const ret = (FIO_prefs_t*)malloc(sizeof(FIO_prefs_t));
+    if (!ret) EXM_THROW(21, "Allocation error : not enough memory");
+
+    ret->compressionType = FIO_zstdCompression;
+    ret->overwrite = 0;
+    ret->sparseFileSupport = ZSTD_SPARSE_DEFAULT;
+    ret->dictIDFlag = 1;
+    ret->checksumFlag = 1;
+    ret->removeSrcFile = 0;
+    ret->memLimit = 0;
+    ret->nbWorkers = 1;
+    ret->jobSize = 0;
+    ret->overlapLog = FIO_OVERLAP_LOG_NOTSET;
+    ret->adaptiveMode = 0;
+    ret->rsyncable = 0;
+    ret->minAdaptLevel = -50;   /* initializing this value requires a constant, so ZSTD_minCLevel() doesn't work */
+    ret->maxAdaptLevel = 22;   /* initializing this value requires a constant, so ZSTD_maxCLevel() doesn't work */
+    ret->ldmFlag = 0;
+    ret->ldmHashLog = 0;
+    ret->ldmMinMatch = 0;
+    ret->ldmBucketSizeLog = FIO_LDM_PARAM_NOTSET;
+    ret->ldmHashRateLog = FIO_LDM_PARAM_NOTSET;
+    ret->streamSrcSize = 0;
+    ret->targetCBlockSize = 0;
+    ret->srcSizeHint = 0;
+    ret->testMode = 0;
+    ret->literalCompressionMode = ZSTD_ps_auto;
+    ret->excludeCompressedFiles = 0;
+    ret->allowBlockDevices = 0;
+    ret->asyncIO = AIO_supported();
+    ret->passThrough = -1;
+    return ret;
+}
+
+FIO_ctx_t* FIO_createContext(void)
+{
+    FIO_ctx_t* const ret = (FIO_ctx_t*)malloc(sizeof(FIO_ctx_t));
+    if (!ret) EXM_THROW(21, "Allocation error : not enough memory");
+
+    ret->currFileIdx = 0;
+    ret->hasStdinInput = 0;
+    ret->hasStdoutOutput = 0;
+    ret->nbFilesTotal = 1;
+    ret->nbFilesProcessed = 0;
+    ret->totalBytesInput = 0;
+    ret->totalBytesOutput = 0;
+    return ret;
+}
+
+void FIO_freePreferences(FIO_prefs_t* const prefs)
+{
+    free(prefs);
+}
+
+void FIO_freeContext(FIO_ctx_t* const fCtx)
+{
+    free(fCtx);
+}
+
+
+/*-*************************************
+*  Parameters: Display Options
+***************************************/
+
+void FIO_setNotificationLevel(int level) { g_display_prefs.displayLevel=level; }
+
+void FIO_setProgressSetting(FIO_progressSetting_e setting) { g_display_prefs.progressSetting = setting; }
+
+
+/*-*************************************
+*  Parameters: Setters
+***************************************/
+
+/* FIO_prefs_t functions */
+
+void FIO_setCompressionType(FIO_prefs_t* const prefs, FIO_compressionType_t compressionType) { prefs->compressionType = compressionType; }
+
+void FIO_overwriteMode(FIO_prefs_t* const prefs) { prefs->overwrite = 1; }
+
+void FIO_setSparseWrite(FIO_prefs_t* const prefs, int sparse) { prefs->sparseFileSupport = sparse; }
+
+void FIO_setDictIDFlag(FIO_prefs_t* const prefs, int dictIDFlag) { prefs->dictIDFlag = dictIDFlag; }
+
+void FIO_setChecksumFlag(FIO_prefs_t* const prefs, int checksumFlag) { prefs->checksumFlag = checksumFlag; }
+
+void FIO_setRemoveSrcFile(FIO_prefs_t* const prefs, int flag) { prefs->removeSrcFile = (flag!=0); }
+
+void FIO_setMemLimit(FIO_prefs_t* const prefs, unsigned memLimit) { prefs->memLimit = memLimit; }
+
+void FIO_setNbWorkers(FIO_prefs_t* const prefs, int nbWorkers) {
+#ifndef ZSTD_MULTITHREAD
+    if (nbWorkers > 0) DISPLAYLEVEL(2, "Note : multi-threading is disabled \n");
+#endif
+    prefs->nbWorkers = nbWorkers;
+}
+
+void FIO_setExcludeCompressedFile(FIO_prefs_t* const prefs, int excludeCompressedFiles) { prefs->excludeCompressedFiles = excludeCompressedFiles; }
+
+void FIO_setAllowBlockDevices(FIO_prefs_t* const prefs, int allowBlockDevices) { prefs->allowBlockDevices = allowBlockDevices; }
+
+void FIO_setJobSize(FIO_prefs_t* const prefs, int jobSize) {
+    if (jobSize && prefs->nbWorkers==0)
+        DISPLAYLEVEL(2, "Setting block size is useless in single-thread mode \n");
+    prefs->jobSize = jobSize;
+}
+
+void FIO_setOverlapLog(FIO_prefs_t* const prefs, int overlapLog){
+    if (overlapLog && prefs->nbWorkers==0)
+        DISPLAYLEVEL(2, "Setting overlapLog is useless in single-thread mode \n");
+    prefs->overlapLog = overlapLog;
+}
+
+void FIO_setAdaptiveMode(FIO_prefs_t* const prefs, int adapt) {
+    if ((adapt>0) && (prefs->nbWorkers==0))
+        EXM_THROW(1, "Adaptive mode is not compatible with single thread mode \n");
+    prefs->adaptiveMode = adapt;
+}
+
+void FIO_setUseRowMatchFinder(FIO_prefs_t* const prefs, int useRowMatchFinder) {
+    prefs->useRowMatchFinder = useRowMatchFinder;
+}
+
+void FIO_setRsyncable(FIO_prefs_t* const prefs, int rsyncable) {
+    if ((rsyncable>0) && (prefs->nbWorkers==0))
+        EXM_THROW(1, "Rsyncable mode is not compatible with single thread mode \n");
+    prefs->rsyncable = rsyncable;
+}
+
+void FIO_setStreamSrcSize(FIO_prefs_t* const prefs, size_t streamSrcSize) {
+    prefs->streamSrcSize = streamSrcSize;
+}
+
+void FIO_setTargetCBlockSize(FIO_prefs_t* const prefs, size_t targetCBlockSize) {
+    prefs->targetCBlockSize = targetCBlockSize;
+}
+
+void FIO_setSrcSizeHint(FIO_prefs_t* const prefs, size_t srcSizeHint) {
+    prefs->srcSizeHint = (int)MIN((size_t)INT_MAX, srcSizeHint);
+}
+
+void FIO_setTestMode(FIO_prefs_t* const prefs, int testMode) {
+    prefs->testMode = (testMode!=0);
+}
+
+void FIO_setLiteralCompressionMode(
+        FIO_prefs_t* const prefs,
+        ZSTD_ParamSwitch_e mode) {
+    prefs->literalCompressionMode = mode;
+}
+
+void FIO_setAdaptMin(FIO_prefs_t* const prefs, int minCLevel)
+{
+#ifndef ZSTD_NOCOMPRESS
+    assert(minCLevel >= ZSTD_minCLevel());
+#endif
+    prefs->minAdaptLevel = minCLevel;
+}
+
+void FIO_setAdaptMax(FIO_prefs_t* const prefs, int maxCLevel)
+{
+    prefs->maxAdaptLevel = maxCLevel;
+}
+
+void FIO_setLdmFlag(FIO_prefs_t* const prefs, unsigned ldmFlag) {
+    prefs->ldmFlag = (ldmFlag>0);
+}
+
+void FIO_setLdmHashLog(FIO_prefs_t* const prefs, int ldmHashLog) {
+    prefs->ldmHashLog = ldmHashLog;
+}
+
+void FIO_setLdmMinMatch(FIO_prefs_t* const prefs, int ldmMinMatch) {
+    prefs->ldmMinMatch = ldmMinMatch;
+}
+
+void FIO_setLdmBucketSizeLog(FIO_prefs_t* const prefs, int ldmBucketSizeLog) {
+    prefs->ldmBucketSizeLog = ldmBucketSizeLog;
+}
+
+
+void FIO_setLdmHashRateLog(FIO_prefs_t* const prefs, int ldmHashRateLog) {
+    prefs->ldmHashRateLog = ldmHashRateLog;
+}
+
+void FIO_setPatchFromMode(FIO_prefs_t* const prefs, int value)
+{
+    prefs->patchFromMode = value != 0;
+}
+
+void FIO_setContentSize(FIO_prefs_t* const prefs, int value)
+{
+    prefs->contentSize = value != 0;
+}
+
+void FIO_setAsyncIOFlag(FIO_prefs_t* const prefs, int value) {
+#ifdef ZSTD_MULTITHREAD
+    prefs->asyncIO = value;
+#else
+    (void) prefs;
+    (void) value;
+    DISPLAYLEVEL(2, "Note : asyncio is disabled (lack of multithreading support) \n");
+#endif
+}
+
+void FIO_setPassThroughFlag(FIO_prefs_t* const prefs, int value) {
+    prefs->passThrough = (value != 0);
+}
+
+void FIO_setMMapDict(FIO_prefs_t* const prefs, ZSTD_ParamSwitch_e value)
+{
+    prefs->mmapDict = value;
+}
+
+/* FIO_ctx_t functions */
+
+void FIO_setHasStdoutOutput(FIO_ctx_t* const fCtx, int value) {
+    fCtx->hasStdoutOutput = value;
+}
+
+void FIO_setNbFilesTotal(FIO_ctx_t* const fCtx, int value)
+{
+    fCtx->nbFilesTotal = value;
+}
+
+void FIO_determineHasStdinInput(FIO_ctx_t* const fCtx, const FileNamesTable* const filenames) {
+    size_t i = 0;
+    for ( ; i < filenames->tableSize; ++i) {
+        if (!strcmp(stdinmark, filenames->fileNames[i])) {
+            fCtx->hasStdinInput = 1;
+            return;
+        }
+    }
+}
+
+/*-*************************************
+*  Functions
+***************************************/
+/** FIO_removeFile() :
+ * @result : Unlink `fileName`, even if it's read-only */
+static int FIO_removeFile(const char* path)
+{
+    stat_t statbuf;
+    if (!UTIL_stat(path, &statbuf)) {
+        DISPLAYLEVEL(2, "zstd: Failed to stat %s while trying to remove it\n", path);
+        return 0;
+    }
+    if (!UTIL_isRegularFileStat(&statbuf)) {
+        DISPLAYLEVEL(2, "zstd: Refusing to remove non-regular file %s\n", path);
+        return 0;
+    }
+#if defined(_WIN32)
+    /* windows doesn't allow remove read-only files,
+     * so try to make it writable first */
+    if (!(statbuf.st_mode & _S_IWRITE)) {
+        UTIL_chmod(path, &statbuf, _S_IWRITE);
+    }
+#endif
+    return remove(path);
+}
+
+/** FIO_openSrcFile() :
+ *  condition : `srcFileName` must be non-NULL.
+ *  optional: `prefs` may be NULL.
+ * @result : FILE* to `srcFileName`, or NULL if it fails */
+static FILE* FIO_openSrcFile(const FIO_prefs_t* const prefs, const char* srcFileName, stat_t* statbuf)
+{
+    int allowBlockDevices = prefs != NULL ? prefs->allowBlockDevices : 0;
+    assert(srcFileName != NULL);
+    assert(statbuf != NULL);
+
+    if (!strcmp(srcFileName, stdinmark)) {
+        DISPLAYLEVEL(4,"Using stdin for input \n");
+        SET_BINARY_MODE(stdin);
+        return stdin;
+    }
+
+    if (!UTIL_stat(srcFileName, statbuf)) {
+        DISPLAYLEVEL(1, "zstd: can't stat %s : %s -- ignored \n",
+                        srcFileName, strerror(errno));
+        return NULL;
+    }
+
+    /* Accept regular files, FIFOs, and process substitution file descriptors */
+    if (!UTIL_isRegularFileStat(statbuf)
+     && !UTIL_isFIFOStat(statbuf)
+     && !UTIL_isFileDescriptorPipe(srcFileName)  /* Process substitution support */
+     && !(allowBlockDevices && UTIL_isBlockDevStat(statbuf))
+    ) {
+        DISPLAYLEVEL(1, "zstd: %s is not a regular file -- ignored \n",
+                        srcFileName);
+        return NULL;
+    }
+
+    {   FILE* const f = fopen(srcFileName, "rb");
+        if (f == NULL)
+            DISPLAYLEVEL(1, "zstd: %s: %s \n", srcFileName, strerror(errno));
+        return f;
+    }
+}
+
+/** FIO_openDstFile() :
+ *  condition : `dstFileName` must be non-NULL.
+ * @result : FILE* to `dstFileName`, or NULL if it fails */
+static FILE*
+FIO_openDstFile(FIO_ctx_t* fCtx, FIO_prefs_t* const prefs,
+                const char* srcFileName, const char* dstFileName,
+                const int mode)
+{
+    if (prefs->testMode) return NULL;  /* do not open file in test mode */
+
+    assert(dstFileName != NULL);
+    if (!strcmp (dstFileName, stdoutmark)) {
+        DISPLAYLEVEL(4,"Using stdout for output \n");
+        SET_BINARY_MODE(stdout);
+        if (prefs->sparseFileSupport == 1) {
+            prefs->sparseFileSupport = 0;
+            DISPLAYLEVEL(4, "Sparse File Support is automatically disabled on stdout ; try --sparse \n");
+        }
+        return stdout;
+    }
+
+    /* ensure dst is not the same as src */
+    if (srcFileName != NULL && UTIL_isSameFile(srcFileName, dstFileName)) {
+        DISPLAYLEVEL(1, "zstd: Refusing to open an output file which will overwrite the input file \n");
+        return NULL;
+    }
+
+    if (UTIL_isRegularFile(dstFileName)) {
+        /* Check if destination file already exists */
+#if !defined(_WIN32)
+        /* this test does not work on Windows :
+         * `NUL` and `nul` are detected as regular files */
+        if (!strcmp(dstFileName, nulmark)) {
+            EXM_THROW(40, "%s is unexpectedly categorized as a regular file",
+                        dstFileName);
+        }
+#endif
+        if (!prefs->overwrite) {
+            if (g_display_prefs.displayLevel <= 1) {
+                /* No interaction possible */
+                DISPLAYLEVEL(1, "zstd: %s already exists; not overwritten  \n",
+                        dstFileName);
+                return NULL;
+            }
+            DISPLAY("zstd: %s already exists; ", dstFileName);
+            if (UTIL_requireUserConfirmation("overwrite (y/n) ? ", "Not overwritten  \n", "yY", fCtx->hasStdinInput))
+                return NULL;
+        }
+        /* need to unlink */
+        FIO_removeFile(dstFileName);
+    }
+
+    {
+        int isDstRegFile;
+#if defined(_WIN32)
+        /* Windows requires opening the file as a "binary" file to avoid
+         * mangling. This macro doesn't exist on unix. */
+        const int openflags = O_WRONLY|O_CREAT|O_TRUNC|O_BINARY;
+        const int fd = _open(dstFileName, openflags, mode);
+        FILE* f = NULL;
+        if (fd != -1) {
+            f = _fdopen(fd, "wb");
+        }
+#else
+        const int openflags = O_WRONLY|O_CREAT|O_TRUNC;
+        const int fd = open(dstFileName, openflags, mode);
+        FILE* f = NULL;
+        if (fd != -1) {
+            f = fdopen(fd, "wb");
+        }
+#endif
+
+        /* Check regular file after opening with O_CREAT */
+        isDstRegFile = UTIL_isFdRegularFile(fd);
+        if (prefs->sparseFileSupport == 1) {
+            prefs->sparseFileSupport = ZSTD_SPARSE_DEFAULT;
+            if (!isDstRegFile) {
+                prefs->sparseFileSupport = 0;
+                DISPLAYLEVEL(4, "Sparse File Support is disabled when output is not a file \n");
+            }
+        }
+
+        if (f == NULL) {
+            if (UTIL_isFileDescriptorPipe(dstFileName)) {
+                DISPLAYLEVEL(1, "zstd: error: no output specified (use -o or -c). \n");
+            } else {
+                DISPLAYLEVEL(1, "zstd: %s: %s\n", dstFileName, strerror(errno));
+            }
+        } else {
+            /* An increased buffer size can provide a significant performance
+             * boost on some platforms. Note that providing a NULL buf with a
+             * size that's not 0 is not defined in ANSI C, but is defined in an
+             * extension. There are three possibilities here:
+             * 1. Libc supports the extended version and everything is good.
+             * 2. Libc ignores the size when buf is NULL, in which case
+             *    everything will continue as if we didn't call `setvbuf()`.
+             * 3. We fail the call and execution continues but a warning
+             *    message might be shown.
+             * In all cases due execution continues. For now, I believe that
+             * this is a more cost-effective solution than managing the buffers
+             * allocations ourselves (will require an API change).
+             */
+            if (setvbuf(f, NULL, _IOFBF, 1 MB)) {
+                DISPLAYLEVEL(2, "Warning: setvbuf failed for %s\n", dstFileName);
+            }
+        }
+        return f;
+    }
+}
+
+
+/* FIO_getDictFileStat() :
+ */
+static void FIO_getDictFileStat(const char* fileName, stat_t* dictFileStat) {
+    assert(dictFileStat != NULL);
+    if (fileName == NULL) return;
+
+    if (!UTIL_stat(fileName, dictFileStat)) {
+        EXM_THROW(31, "Stat failed on dictionary file %s: %s", fileName, strerror(errno));
+    }
+
+    if (!UTIL_isRegularFileStat(dictFileStat)) {
+        EXM_THROW(32, "Dictionary %s must be a regular file.", fileName);
+    }
+}
+
+/*  FIO_setDictBufferMalloc() :
+ *  allocates a buffer, pointed by `dict->dictBuffer`,
+ *  loads `filename` content into it, up to DICTSIZE_MAX bytes.
+ * @return : loaded size
+ *  if fileName==NULL, returns 0 and a NULL pointer
+ */
+static size_t FIO_setDictBufferMalloc(FIO_Dict_t* dict, const char* fileName, FIO_prefs_t* const prefs, stat_t* dictFileStat)
+{
+    FILE* fileHandle;
+    U64 fileSize;
+    void** bufferPtr = &dict->dictBuffer;
+
+    assert(bufferPtr != NULL);
+    assert(dictFileStat != NULL);
+    *bufferPtr = NULL;
+    if (fileName == NULL) return 0;
+
+    DISPLAYLEVEL(4,"Loading %s as dictionary \n", fileName);
+
+    fileHandle = fopen(fileName, "rb");
+
+    if (fileHandle == NULL) {
+        EXM_THROW(33, "Couldn't open dictionary %s: %s", fileName, strerror(errno));
+    }
+
+    fileSize = UTIL_getFileSizeStat(dictFileStat);
+    {
+        size_t const dictSizeMax = prefs->patchFromMode ? prefs->memLimit : DICTSIZE_MAX;
+        if (fileSize >  dictSizeMax) {
+            EXM_THROW(34, "Dictionary file %s is too large (> %u bytes)",
+                            fileName,  (unsigned)dictSizeMax);   /* avoid extreme cases */
+        }
+    }
+    *bufferPtr = malloc((size_t)fileSize);
+    if (*bufferPtr==NULL) EXM_THROW(34, "%s", strerror(errno));
+    {   size_t const readSize = fread(*bufferPtr, 1, (size_t)fileSize, fileHandle);
+        if (readSize != fileSize) {
+            EXM_THROW(35, "Error reading dictionary file %s : %s",
+                    fileName, strerror(errno));
+        }
+    }
+    fclose(fileHandle);
+    return (size_t)fileSize;
+}
+
+#if (PLATFORM_POSIX_VERSION > 0)
+#include <sys/mman.h>
+static void FIO_munmap(FIO_Dict_t* dict)
+{
+    munmap(dict->dictBuffer, dict->dictBufferSize);
+    dict->dictBuffer = NULL;
+    dict->dictBufferSize = 0;
+}
+static size_t FIO_setDictBufferMMap(FIO_Dict_t* dict, const char* fileName, FIO_prefs_t* const prefs, stat_t* dictFileStat)
+{
+    int fileHandle;
+    U64 fileSize;
+    void** bufferPtr = &dict->dictBuffer;
+
+    assert(bufferPtr != NULL);
+    assert(dictFileStat != NULL);
+    *bufferPtr = NULL;
+    if (fileName == NULL) return 0;
+
+    DISPLAYLEVEL(4,"Loading %s as dictionary \n", fileName);
+
+    fileHandle = open(fileName, O_RDONLY);
+
+    if (fileHandle == -1) {
+        EXM_THROW(33, "Couldn't open dictionary %s: %s", fileName, strerror(errno));
+    }
+
+    fileSize = UTIL_getFileSizeStat(dictFileStat);
+    {
+        size_t const dictSizeMax = prefs->patchFromMode ? prefs->memLimit : DICTSIZE_MAX;
+        if (fileSize >  dictSizeMax) {
+            EXM_THROW(34, "Dictionary file %s is too large (> %u bytes)",
+                            fileName,  (unsigned)dictSizeMax);   /* avoid extreme cases */
+        }
+    }
+
+    *bufferPtr = mmap(NULL, (size_t)fileSize, PROT_READ, MAP_PRIVATE, fileHandle, 0);
+    if (*bufferPtr==NULL) EXM_THROW(34, "%s", strerror(errno));
+
+    close(fileHandle);
+    return (size_t)fileSize;
+}
+#elif defined(_MSC_VER) || defined(_WIN32)
+#include <windows.h>
+static void FIO_munmap(FIO_Dict_t* dict)
+{
+    UnmapViewOfFile(dict->dictBuffer);
+    CloseHandle(dict->dictHandle);
+    dict->dictBuffer = NULL;
+    dict->dictBufferSize = 0;
+}
+static size_t FIO_setDictBufferMMap(FIO_Dict_t* dict, const char* fileName, FIO_prefs_t* const prefs, stat_t* dictFileStat)
+{
+    HANDLE fileHandle, mapping;
+    U64 fileSize;
+    void** bufferPtr = &dict->dictBuffer;
+
+    assert(bufferPtr != NULL);
+    assert(dictFileStat != NULL);
+    *bufferPtr = NULL;
+    if (fileName == NULL) return 0;
+
+    DISPLAYLEVEL(4,"Loading %s as dictionary \n", fileName);
+
+    fileHandle = CreateFileA(fileName, GENERIC_READ, FILE_SHARE_READ, NULL, OPEN_EXISTING, FILE_ATTRIBUTE_READONLY, NULL);
+
+    if (fileHandle == INVALID_HANDLE_VALUE) {
+        EXM_THROW(33, "Couldn't open dictionary %s: %s", fileName, strerror(errno));
+    }
+
+    fileSize = UTIL_getFileSizeStat(dictFileStat);
+    {
+        size_t const dictSizeMax = prefs->patchFromMode ? prefs->memLimit : DICTSIZE_MAX;
+        if (fileSize >  dictSizeMax) {
+            EXM_THROW(34, "Dictionary file %s is too large (> %u bytes)",
+                            fileName,  (unsigned)dictSizeMax);   /* avoid extreme cases */
+        }
+    }
+
+    mapping = CreateFileMapping(fileHandle, NULL, PAGE_READONLY, 0, 0, NULL);
+	if (mapping == NULL) {
+        EXM_THROW(35, "Couldn't map dictionary %s: %s", fileName, strerror(errno));
+    }
+
+	*bufferPtr = MapViewOfFile(mapping, FILE_MAP_READ, 0, 0, (DWORD)fileSize); /* we can only cast to DWORD here because dictSize <= 2GB */
+	if (*bufferPtr==NULL) EXM_THROW(36, "%s", strerror(errno));
+
+    dict->dictHandle = fileHandle;
+    return (size_t)fileSize;
+}
+#else
+static size_t FIO_setDictBufferMMap(FIO_Dict_t* dict, const char* fileName, FIO_prefs_t* const prefs, stat_t* dictFileStat)
+{
+   return FIO_setDictBufferMalloc(dict, fileName, prefs, dictFileStat);
+}
+static void FIO_munmap(FIO_Dict_t* dict) {
+   free(dict->dictBuffer);
+   dict->dictBuffer = NULL;
+   dict->dictBufferSize = 0;
+}
+#endif
+
+static void FIO_freeDict(FIO_Dict_t* dict) {
+    if (dict->dictBufferType == FIO_mallocDict) {
+        free(dict->dictBuffer);
+        dict->dictBuffer = NULL;
+        dict->dictBufferSize = 0;
+    } else if (dict->dictBufferType == FIO_mmapDict)  {
+        FIO_munmap(dict);
+    } else {
+        assert(0); /* Should not reach this case */
+    }
+}
+
+static void FIO_initDict(FIO_Dict_t* dict, const char* fileName, FIO_prefs_t* const prefs, stat_t* dictFileStat, FIO_dictBufferType_t dictBufferType) {
+    dict->dictBufferType = dictBufferType;
+    if (dict->dictBufferType == FIO_mallocDict) {
+        dict->dictBufferSize = FIO_setDictBufferMalloc(dict, fileName, prefs, dictFileStat);
+    } else if (dict->dictBufferType == FIO_mmapDict)  {
+        dict->dictBufferSize = FIO_setDictBufferMMap(dict, fileName, prefs, dictFileStat);
+    } else {
+        assert(0); /* Should not reach this case */
+    }
+}
+
+
+/* FIO_checkFilenameCollisions() :
+ * Checks for and warns if there are any files that would have the same output path
+ */
+int FIO_checkFilenameCollisions(const char** filenameTable, unsigned nbFiles) {
+    const char **filenameTableSorted, *prevElem, *filename;
+    unsigned u;
+
+    filenameTableSorted = (const char**) malloc(sizeof(char*) * nbFiles);
+    if (!filenameTableSorted) {
+        DISPLAYLEVEL(1, "Allocation error during filename collision checking \n");
+        return 1;
+    }
+
+    for (u = 0; u < nbFiles; ++u) {
+        filename = strrchr(filenameTable[u], PATH_SEP);
+        if (filename == NULL) {
+            filenameTableSorted[u] = filenameTable[u];
+        } else {
+            filenameTableSorted[u] = filename+1;
+        }
+    }
+
+    qsort((void*)filenameTableSorted, nbFiles, sizeof(char*), UTIL_compareStr);
+    prevElem = filenameTableSorted[0];
+    for (u = 1; u < nbFiles; ++u) {
+        if (strcmp(prevElem, filenameTableSorted[u]) == 0) {
+            DISPLAYLEVEL(2, "WARNING: Two files have same filename: %s\n", prevElem);
+        }
+        prevElem = filenameTableSorted[u];
+    }
+
+    free((void*)filenameTableSorted);
+    return 0;
+}
+
+static const char*
+extractFilename(const char* path, char separator)
+{
+    const char* search = strrchr(path, separator);
+    if (search == NULL) return path;
+    return search+1;
+}
+
+/* FIO_createFilename_fromOutDir() :
+ * Takes a source file name and specified output directory, and
+ * allocates memory for and returns a pointer to final path.
+ * This function never returns an error (it may abort() in case of pb)
+ */
+static char*
+FIO_createFilename_fromOutDir(const char* path, const char* outDirName, const size_t suffixLen)
+{
+    const char* filenameStart;
+    char separator;
+    char* result;
+
+#if defined(_MSC_VER) || defined(__MINGW32__) || defined (__MSVCRT__) /* windows support */
+    separator = '\\';
+#else
+    separator = '/';
+#endif
+
+    filenameStart = extractFilename(path, separator);
+#if defined(_MSC_VER) || defined(__MINGW32__) || defined (__MSVCRT__) /* windows support */
+    filenameStart = extractFilename(filenameStart, '/');  /* sometimes, '/' separator is also used on Windows (mingw+msys2) */
+#endif
+
+    result = (char*) calloc(1, strlen(outDirName) + 1 + strlen(filenameStart) + suffixLen + 1);
+    if (!result) {
+        EXM_THROW(30, "zstd: FIO_createFilename_fromOutDir: %s", strerror(errno));
+    }
+
+    memcpy(result, outDirName, strlen(outDirName));
+    if (outDirName[strlen(outDirName)-1] == separator) {
+        memcpy(result + strlen(outDirName), filenameStart, strlen(filenameStart));
+    } else {
+        memcpy(result + strlen(outDirName), &separator, 1);
+        memcpy(result + strlen(outDirName) + 1, filenameStart, strlen(filenameStart));
+    }
+
+    return result;
+}
+
+/* FIO_highbit64() :
+ * gives position of highest bit.
+ * note : only works for v > 0 !
+ */
+static unsigned FIO_highbit64(unsigned long long v)
+{
+    unsigned count = 0;
+    assert(v != 0);
+    v >>= 1;
+    while (v) { v >>= 1; count++; }
+    return count;
+}
+
+static void FIO_adjustMemLimitForPatchFromMode(FIO_prefs_t* const prefs,
+                                    unsigned long long const dictSize,
+                                    unsigned long long const maxSrcFileSize)
+{
+    unsigned long long maxSize = MAX(prefs->memLimit, MAX(dictSize, maxSrcFileSize));
+    unsigned const maxWindowSize = (1U << ZSTD_WINDOWLOG_MAX);
+    if (maxSize == UTIL_FILESIZE_UNKNOWN)
+        EXM_THROW(42, "Using --patch-from with stdin requires --stream-size");
+    assert(maxSize != UTIL_FILESIZE_UNKNOWN);
+    if (maxSize > maxWindowSize)
+        EXM_THROW(42, "Can't handle files larger than %u GB\n", maxWindowSize/(1 GB));
+    FIO_setMemLimit(prefs, (unsigned)maxSize);
+}
+
+/* FIO_multiFilesConcatWarning() :
+ * This function handles logic when processing multiple files with -o or -c, displaying the appropriate warnings/prompts.
+ * Returns 1 if the console should abort, 0 if console should proceed.
+ *
+ * If output is stdout or test mode is active, check that `--rm` disabled.
+ *
+ * If there is just 1 file to process, zstd will proceed as usual.
+ * If each file get processed into its own separate destination file, proceed as usual.
+ *
+ * When multiple files are processed into a single output,
+ * display a warning message, then disable --rm if it's set.
+ *
+ * If -f is specified or if output is stdout, just proceed.
+ * If output is set with -o, prompt for confirmation.
+ */
+static int FIO_multiFilesConcatWarning(const FIO_ctx_t* fCtx, FIO_prefs_t* prefs, const char* outFileName, int displayLevelCutoff)
+{
+    if (fCtx->hasStdoutOutput) {
+        if (prefs->removeSrcFile)
+            /* this should not happen ; hard fail, to protect user's data
+             * note: this should rather be an assert(), but we want to be certain that user's data will not be wiped out in case it nonetheless happen */
+            EXM_THROW(43, "It's not allowed to remove input files when processed output is piped to stdout. "
+                "This scenario is not supposed to be possible. "
+                "This is a programming error. File an issue for it to be fixed.");
+    }
+    if (prefs->testMode) {
+        if (prefs->removeSrcFile)
+            /* this should not happen ; hard fail, to protect user's data
+             * note: this should rather be an assert(), but we want to be certain that user's data will not be wiped out in case it nonetheless happen */
+            EXM_THROW(43, "Test mode shall not remove input files! "
+                 "This scenario is not supposed to be possible. "
+                 "This is a programming error. File an issue for it to be fixed.");
+        return 0;
+    }
+
+    if (fCtx->nbFilesTotal == 1) return 0;
+    assert(fCtx->nbFilesTotal > 1);
+
+    if (!outFileName) return 0;
+
+    if (fCtx->hasStdoutOutput) {
+        DISPLAYLEVEL(2, "zstd: WARNING: all input files will be processed and concatenated into stdout. \n");
+    } else {
+        DISPLAYLEVEL(2, "zstd: WARNING: all input files will be processed and concatenated into a single output file: %s \n", outFileName);
+    }
+    DISPLAYLEVEL(2, "The concatenated output CANNOT regenerate original file names nor directory structure. \n")
+
+    /* multi-input into single output : --rm is not allowed */
+    if (prefs->removeSrcFile) {
+        DISPLAYLEVEL(2, "Since it's a destructive operation, input files will not be removed. \n");
+        prefs->removeSrcFile = 0;
+    }
+
+    if (fCtx->hasStdoutOutput) return 0;
+    if (prefs->overwrite) return 0;
+
+    /* multiple files concatenated into single destination file using -o without -f */
+    if (g_display_prefs.displayLevel <= displayLevelCutoff) {
+        /* quiet mode => no prompt => fail automatically */
+        DISPLAYLEVEL(1, "Concatenating multiple processed inputs into a single output loses file metadata. \n");
+        DISPLAYLEVEL(1, "Aborting. \n");
+        return 1;
+    }
+    /* normal mode => prompt */
+    return UTIL_requireUserConfirmation("Proceed? (y/n): ", "Aborting...", "yY", fCtx->hasStdinInput);
+}
+
+static ZSTD_inBuffer setInBuffer(const void* buf, size_t s, size_t pos)
+{
+    ZSTD_inBuffer i;
+    i.src = buf;
+    i.size = s;
+    i.pos = pos;
+    return i;
+}
+
+static ZSTD_outBuffer setOutBuffer(void* buf, size_t s, size_t pos)
+{
+    ZSTD_outBuffer o;
+    o.dst = buf;
+    o.size = s;
+    o.pos = pos;
+    return o;
+}
+
+#ifndef ZSTD_NOCOMPRESS
+
+/* **********************************************************************
+ *  Compression
+ ************************************************************************/
+typedef struct {
+    FIO_Dict_t dict;
+    const char* dictFileName;
+    stat_t dictFileStat;
+    ZSTD_CStream* cctx;
+    WritePoolCtx_t *writeCtx;
+    ReadPoolCtx_t *readCtx;
+} cRess_t;
+
+/** ZSTD_cycleLog() :
+ *  condition for correct operation : hashLog > 1 */
+static U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat)
+{
+    U32 const btScale = ((U32)strat >= (U32)ZSTD_btlazy2);
+    assert(hashLog > 1);
+    return hashLog - btScale;
+}
+
+static void FIO_adjustParamsForPatchFromMode(FIO_prefs_t* const prefs,
+                                    ZSTD_compressionParameters* comprParams,
+                                    unsigned long long const dictSize,
+                                    unsigned long long const maxSrcFileSize,
+                                    int cLevel)
+{
+    unsigned const fileWindowLog = FIO_highbit64(maxSrcFileSize) + 1;
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams(cLevel, (size_t)maxSrcFileSize, (size_t)dictSize);
+    FIO_adjustMemLimitForPatchFromMode(prefs, dictSize, maxSrcFileSize);
+    if (fileWindowLog > ZSTD_WINDOWLOG_MAX)
+        DISPLAYLEVEL(1, "Max window log exceeded by file (compression ratio will suffer)\n");
+    comprParams->windowLog = MAX(ZSTD_WINDOWLOG_MIN, MIN(ZSTD_WINDOWLOG_MAX, fileWindowLog));
+    if (fileWindowLog > ZSTD_cycleLog(cParams.chainLog, cParams.strategy)) {
+        if (!prefs->ldmFlag)
+            DISPLAYLEVEL(2, "long mode automatically triggered\n");
+        FIO_setLdmFlag(prefs, 1);
+    }
+    if (cParams.strategy >= ZSTD_btopt) {
+        DISPLAYLEVEL(4, "[Optimal parser notes] Consider the following to improve patch size at the cost of speed:\n");
+        DISPLAYLEVEL(4, "- Set a larger targetLength (e.g. --zstd=targetLength=4096)\n");
+        DISPLAYLEVEL(4, "- Set a larger chainLog (e.g. --zstd=chainLog=%u)\n", ZSTD_CHAINLOG_MAX);
+        DISPLAYLEVEL(4, "- Set a larger LDM hashLog (e.g. --zstd=ldmHashLog=%u)\n", ZSTD_LDM_HASHLOG_MAX);
+        DISPLAYLEVEL(4, "- Set a smaller LDM rateLog (e.g. --zstd=ldmHashRateLog=%u)\n", ZSTD_LDM_HASHRATELOG_MIN);
+        DISPLAYLEVEL(4, "Also consider playing around with searchLog and hashLog\n");
+    }
+}
+
+static cRess_t FIO_createCResources(FIO_prefs_t* const prefs,
+                                    const char* dictFileName, unsigned long long const maxSrcFileSize,
+                                    int cLevel, ZSTD_compressionParameters comprParams) {
+    int useMMap = prefs->mmapDict == ZSTD_ps_enable;
+    int forceNoUseMMap = prefs->mmapDict == ZSTD_ps_disable;
+    FIO_dictBufferType_t dictBufferType;
+    cRess_t ress;
+    memset(&ress, 0, sizeof(ress));
+
+    DISPLAYLEVEL(6, "FIO_createCResources \n");
+    ress.cctx = ZSTD_createCCtx();
+    if (ress.cctx == NULL)
+        EXM_THROW(30, "allocation error (%s): can't create ZSTD_CCtx",
+                    strerror(errno));
+
+    FIO_getDictFileStat(dictFileName, &ress.dictFileStat);
+
+    /* need to update memLimit before calling createDictBuffer
+     * because of memLimit check inside it */
+    if (prefs->patchFromMode) {
+        U64 const dictSize = UTIL_getFileSizeStat(&ress.dictFileStat);
+        unsigned long long const ssSize = (unsigned long long)prefs->streamSrcSize;
+        useMMap |= dictSize > prefs->memLimit;
+        FIO_adjustParamsForPatchFromMode(prefs, &comprParams, dictSize, ssSize > 0 ? ssSize : maxSrcFileSize, cLevel);
+    }
+
+    dictBufferType = (useMMap && !forceNoUseMMap) ? FIO_mmapDict : FIO_mallocDict;
+    FIO_initDict(&ress.dict, dictFileName, prefs, &ress.dictFileStat, dictBufferType);   /* works with dictFileName==NULL */
+
+    ress.writeCtx = AIO_WritePool_create(prefs, ZSTD_CStreamOutSize());
+    ress.readCtx = AIO_ReadPool_create(prefs, ZSTD_CStreamInSize());
+
+    /* Advanced parameters, including dictionary */
+    if (dictFileName && (ress.dict.dictBuffer==NULL))
+        EXM_THROW(32, "allocation error : can't create dictBuffer");
+    ress.dictFileName = dictFileName;
+
+    if (prefs->adaptiveMode && !prefs->ldmFlag && !comprParams.windowLog)
+        comprParams.windowLog = ADAPT_WINDOWLOG_DEFAULT;
+
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_contentSizeFlag, prefs->contentSize) );  /* always enable content size when available (note: supposed to be default) */
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_dictIDFlag, prefs->dictIDFlag) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_checksumFlag, prefs->checksumFlag) );
+    /* compression level */
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_compressionLevel, cLevel) );
+    /* max compressed block size */
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_targetCBlockSize, (int)prefs->targetCBlockSize) );
+    /* source size hint */
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_srcSizeHint, (int)prefs->srcSizeHint) );
+    /* long distance matching */
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_enableLongDistanceMatching, prefs->ldmFlag) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_ldmHashLog, prefs->ldmHashLog) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_ldmMinMatch, prefs->ldmMinMatch) );
+    if (prefs->ldmBucketSizeLog != FIO_LDM_PARAM_NOTSET) {
+        CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_ldmBucketSizeLog, prefs->ldmBucketSizeLog) );
+    }
+    if (prefs->ldmHashRateLog != FIO_LDM_PARAM_NOTSET) {
+        CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_ldmHashRateLog, prefs->ldmHashRateLog) );
+    }
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_useRowMatchFinder, prefs->useRowMatchFinder));
+    /* compression parameters */
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_windowLog, (int)comprParams.windowLog) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_chainLog, (int)comprParams.chainLog) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_hashLog, (int)comprParams.hashLog) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_searchLog, (int)comprParams.searchLog) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_minMatch, (int)comprParams.minMatch) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_targetLength, (int)comprParams.targetLength) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_strategy, (int)comprParams.strategy) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_literalCompressionMode, (int)prefs->literalCompressionMode) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_enableDedicatedDictSearch, 1) );
+    /* multi-threading */
+#ifdef ZSTD_MULTITHREAD
+    DISPLAYLEVEL(5,"set nb workers = %u \n", prefs->nbWorkers);
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_nbWorkers, prefs->nbWorkers) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_jobSize, prefs->jobSize) );
+    if (prefs->overlapLog != FIO_OVERLAP_LOG_NOTSET) {
+        DISPLAYLEVEL(3,"set overlapLog = %u \n", prefs->overlapLog);
+        CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_overlapLog, prefs->overlapLog) );
+    }
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_rsyncable, prefs->rsyncable) );
+#endif
+    /* dictionary */
+    if (prefs->patchFromMode) {
+        CHECK( ZSTD_CCtx_refPrefix(ress.cctx, ress.dict.dictBuffer, ress.dict.dictBufferSize) );
+    } else {
+        CHECK( ZSTD_CCtx_loadDictionary_byReference(ress.cctx, ress.dict.dictBuffer, ress.dict.dictBufferSize) );
+    }
+
+    return ress;
+}
+
+static void FIO_freeCResources(cRess_t* const ress)
+{
+    FIO_freeDict(&(ress->dict));
+    AIO_WritePool_free(ress->writeCtx);
+    AIO_ReadPool_free(ress->readCtx);
+    ZSTD_freeCStream(ress->cctx);   /* never fails */
+}
+
+
+#ifdef ZSTD_GZCOMPRESS
+static unsigned long long
+FIO_compressGzFrame(const cRess_t* ress,  /* buffers & handlers are used, but not changed */
+                    const char* srcFileName, U64 const srcFileSize,
+                    int compressionLevel, U64* readsize)
+{
+    unsigned long long inFileSize = 0, outFileSize = 0;
+    z_stream strm;
+    IOJob_t *writeJob = NULL;
+
+    if (compressionLevel > Z_BEST_COMPRESSION)
+        compressionLevel = Z_BEST_COMPRESSION;
+
+    strm.zalloc = Z_NULL;
+    strm.zfree = Z_NULL;
+    strm.opaque = Z_NULL;
+
+    {   int const ret = deflateInit2(&strm, compressionLevel, Z_DEFLATED,
+                        15 /* maxWindowLogSize */ + 16 /* gzip only */,
+                        8, Z_DEFAULT_STRATEGY); /* see https://www.zlib.net/manual.html */
+        if (ret != Z_OK) {
+            EXM_THROW(71, "zstd: %s: deflateInit2 error %d \n", srcFileName, ret);
+    }   }
+
+    writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
+    strm.next_in = 0;
+    strm.avail_in = 0;
+    strm.next_out = (Bytef*)writeJob->buffer;
+    strm.avail_out = (uInt)writeJob->bufferSize;
+
+    while (1) {
+        int ret;
+        if (strm.avail_in == 0) {
+            AIO_ReadPool_fillBuffer(ress->readCtx, ZSTD_CStreamInSize());
+            if (ress->readCtx->srcBufferLoaded == 0) break;
+            inFileSize += ress->readCtx->srcBufferLoaded;
+            strm.next_in = (z_const unsigned char*)ress->readCtx->srcBuffer;
+            strm.avail_in = (uInt)ress->readCtx->srcBufferLoaded;
+        }
+
+        {
+            size_t const availBefore = strm.avail_in;
+            ret = deflate(&strm, Z_NO_FLUSH);
+            AIO_ReadPool_consumeBytes(ress->readCtx, availBefore - strm.avail_in);
+        }
+
+        if (ret != Z_OK)
+            EXM_THROW(72, "zstd: %s: deflate error %d \n", srcFileName, ret);
+        {   size_t const cSize = writeJob->bufferSize - strm.avail_out;
+            if (cSize) {
+                writeJob->usedBufferSize = cSize;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+                outFileSize += cSize;
+                strm.next_out = (Bytef*)writeJob->buffer;
+                strm.avail_out = (uInt)writeJob->bufferSize;
+            }   }
+        if (srcFileSize == UTIL_FILESIZE_UNKNOWN) {
+            DISPLAYUPDATE_PROGRESS(
+                    "\rRead : %u MB ==> %.2f%% ",
+                    (unsigned)(inFileSize>>20),
+                    (double)outFileSize/(double)inFileSize*100)
+        } else {
+            DISPLAYUPDATE_PROGRESS(
+                    "\rRead : %u / %u MB ==> %.2f%% ",
+                    (unsigned)(inFileSize>>20), (unsigned)(srcFileSize>>20),
+                    (double)outFileSize/(double)inFileSize*100);
+    }   }
+
+    while (1) {
+        int const ret = deflate(&strm, Z_FINISH);
+        {   size_t const cSize = writeJob->bufferSize - strm.avail_out;
+            if (cSize) {
+                writeJob->usedBufferSize = cSize;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+                outFileSize += cSize;
+                strm.next_out = (Bytef*)writeJob->buffer;
+                strm.avail_out = (uInt)writeJob->bufferSize;
+            }   }
+        if (ret == Z_STREAM_END) break;
+        if (ret != Z_BUF_ERROR)
+            EXM_THROW(77, "zstd: %s: deflate error %d \n", srcFileName, ret);
+    }
+
+    {   int const ret = deflateEnd(&strm);
+        if (ret != Z_OK) {
+            EXM_THROW(79, "zstd: %s: deflateEnd error %d \n", srcFileName, ret);
+    }   }
+    *readsize = inFileSize;
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
+    return outFileSize;
+}
+#endif
+
+
+#ifdef ZSTD_LZMACOMPRESS
+static unsigned long long
+FIO_compressLzmaFrame(cRess_t* ress,
+                      const char* srcFileName, U64 const srcFileSize,
+                      int compressionLevel, U64* readsize, int plain_lzma)
+{
+    unsigned long long inFileSize = 0, outFileSize = 0;
+    lzma_stream strm = LZMA_STREAM_INIT;
+    lzma_action action = LZMA_RUN;
+    lzma_ret ret;
+    IOJob_t *writeJob = NULL;
+
+    if (compressionLevel < 0) compressionLevel = 0;
+    if (compressionLevel > 9) compressionLevel = 9;
+
+    if (plain_lzma) {
+        lzma_options_lzma opt_lzma;
+        if (lzma_lzma_preset(&opt_lzma, compressionLevel))
+            EXM_THROW(81, "zstd: %s: lzma_lzma_preset error", srcFileName);
+        ret = lzma_alone_encoder(&strm, &opt_lzma); /* LZMA */
+        if (ret != LZMA_OK)
+            EXM_THROW(82, "zstd: %s: lzma_alone_encoder error %d", srcFileName, ret);
+    } else {
+        ret = lzma_easy_encoder(&strm, compressionLevel, LZMA_CHECK_CRC64); /* XZ */
+        if (ret != LZMA_OK)
+            EXM_THROW(83, "zstd: %s: lzma_easy_encoder error %d", srcFileName, ret);
+    }
+
+    writeJob =AIO_WritePool_acquireJob(ress->writeCtx);
+    strm.next_out = (BYTE*)writeJob->buffer;
+    strm.avail_out = writeJob->bufferSize;
+    strm.next_in = 0;
+    strm.avail_in = 0;
+
+    while (1) {
+        if (strm.avail_in == 0) {
+            size_t const inSize = AIO_ReadPool_fillBuffer(ress->readCtx, ZSTD_CStreamInSize());
+            if (ress->readCtx->srcBufferLoaded == 0) action = LZMA_FINISH;
+            inFileSize += inSize;
+            strm.next_in = (BYTE const*)ress->readCtx->srcBuffer;
+            strm.avail_in = ress->readCtx->srcBufferLoaded;
+        }
+
+        {
+            size_t const availBefore = strm.avail_in;
+            ret = lzma_code(&strm, action);
+            AIO_ReadPool_consumeBytes(ress->readCtx, availBefore - strm.avail_in);
+        }
+
+
+        if (ret != LZMA_OK && ret != LZMA_STREAM_END)
+            EXM_THROW(84, "zstd: %s: lzma_code encoding error %d", srcFileName, ret);
+        {   size_t const compBytes = writeJob->bufferSize - strm.avail_out;
+            if (compBytes) {
+                writeJob->usedBufferSize = compBytes;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+                outFileSize += compBytes;
+                strm.next_out = (BYTE*)writeJob->buffer;
+                strm.avail_out = writeJob->bufferSize;
+        }   }
+        if (srcFileSize == UTIL_FILESIZE_UNKNOWN)
+            DISPLAYUPDATE_PROGRESS("\rRead : %u MB ==> %.2f%%",
+                            (unsigned)(inFileSize>>20),
+                            (double)outFileSize/(double)inFileSize*100)
+        else
+            DISPLAYUPDATE_PROGRESS("\rRead : %u / %u MB ==> %.2f%%",
+                            (unsigned)(inFileSize>>20), (unsigned)(srcFileSize>>20),
+                            (double)outFileSize/(double)inFileSize*100);
+        if (ret == LZMA_STREAM_END) break;
+    }
+
+    lzma_end(&strm);
+    *readsize = inFileSize;
+
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
+
+    return outFileSize;
+}
+#endif
+
+#ifdef ZSTD_LZ4COMPRESS
+
+#if LZ4_VERSION_NUMBER <= 10600
+#define LZ4F_blockLinked blockLinked
+#define LZ4F_max64KB max64KB
+#endif
+
+static int FIO_LZ4_GetBlockSize_FromBlockId (int id) { return (1 << (8 + (2 * id))); }
+
+static unsigned long long
+FIO_compressLz4Frame(cRess_t* ress,
+                     const char* srcFileName, U64 const srcFileSize,
+                     int compressionLevel, int checksumFlag,
+                     U64* readsize)
+{
+    const size_t blockSize = FIO_LZ4_GetBlockSize_FromBlockId(LZ4F_max64KB);
+    unsigned long long inFileSize = 0, outFileSize = 0;
+
+    LZ4F_preferences_t prefs;
+    LZ4F_compressionContext_t ctx;
+
+    IOJob_t* writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
+
+    LZ4F_errorCode_t const errorCode = LZ4F_createCompressionContext(&ctx, LZ4F_VERSION);
+    if (LZ4F_isError(errorCode))
+        EXM_THROW(31, "zstd: failed to create lz4 compression context");
+
+    memset(&prefs, 0, sizeof(prefs));
+
+    assert(blockSize <= ress->readCtx->base.jobBufferSize);
+
+    /* autoflush off to mitigate a bug in lz4<=1.9.3 for compression level 12 */
+    prefs.autoFlush = 0;
+    prefs.compressionLevel = compressionLevel;
+    prefs.frameInfo.blockMode = LZ4F_blockLinked;
+    prefs.frameInfo.blockSizeID = LZ4F_max64KB;
+    prefs.frameInfo.contentChecksumFlag = (contentChecksum_t)checksumFlag;
+#if LZ4_VERSION_NUMBER >= 10600
+    prefs.frameInfo.contentSize = (srcFileSize==UTIL_FILESIZE_UNKNOWN) ? 0 : srcFileSize;
+#endif
+    assert(LZ4F_compressBound(blockSize, &prefs) <= writeJob->bufferSize);
+
+    {
+        size_t headerSize = LZ4F_compressBegin(ctx, writeJob->buffer, writeJob->bufferSize, &prefs);
+        if (LZ4F_isError(headerSize))
+            EXM_THROW(33, "File header generation failed : %s",
+                            LZ4F_getErrorName(headerSize));
+        writeJob->usedBufferSize = headerSize;
+        AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+        outFileSize += headerSize;
+
+        /* Read first block */
+        inFileSize += AIO_ReadPool_fillBuffer(ress->readCtx, blockSize);
+
+        /* Main Loop */
+        while (ress->readCtx->srcBufferLoaded) {
+            size_t inSize = MIN(blockSize, ress->readCtx->srcBufferLoaded);
+            size_t const outSize = LZ4F_compressUpdate(ctx, writeJob->buffer, writeJob->bufferSize,
+                                                       ress->readCtx->srcBuffer, inSize, NULL);
+            if (LZ4F_isError(outSize))
+                EXM_THROW(35, "zstd: %s: lz4 compression failed : %s",
+                            srcFileName, LZ4F_getErrorName(outSize));
+            outFileSize += outSize;
+            if (srcFileSize == UTIL_FILESIZE_UNKNOWN) {
+                DISPLAYUPDATE_PROGRESS("\rRead : %u MB ==> %.2f%%",
+                                (unsigned)(inFileSize>>20),
+                                (double)outFileSize/(double)inFileSize*100)
+            } else {
+                DISPLAYUPDATE_PROGRESS("\rRead : %u / %u MB ==> %.2f%%",
+                                (unsigned)(inFileSize>>20), (unsigned)(srcFileSize>>20),
+                                (double)outFileSize/(double)inFileSize*100);
+            }
+
+            /* Write Block */
+            writeJob->usedBufferSize = outSize;
+            AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+
+            /* Read next block */
+            AIO_ReadPool_consumeBytes(ress->readCtx, inSize);
+            inFileSize += AIO_ReadPool_fillBuffer(ress->readCtx, blockSize);
+        }
+
+        /* End of Stream mark */
+        headerSize = LZ4F_compressEnd(ctx, writeJob->buffer, writeJob->bufferSize, NULL);
+        if (LZ4F_isError(headerSize))
+            EXM_THROW(38, "zstd: %s: lz4 end of file generation failed : %s",
+                        srcFileName, LZ4F_getErrorName(headerSize));
+
+        writeJob->usedBufferSize = headerSize;
+        AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+        outFileSize += headerSize;
+    }
+
+    *readsize = inFileSize;
+    LZ4F_freeCompressionContext(ctx);
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
+
+    return outFileSize;
+}
+#endif
+
+static unsigned long long
+FIO_compressZstdFrame(FIO_ctx_t* const fCtx,
+                      FIO_prefs_t* const prefs,
+                      const cRess_t* ressPtr,
+                      const char* srcFileName, U64 fileSize,
+                      int compressionLevel, U64* readsize)
+{
+    cRess_t const ress = *ressPtr;
+    IOJob_t* writeJob = AIO_WritePool_acquireJob(ressPtr->writeCtx);
+
+    U64 compressedfilesize = 0;
+    ZSTD_EndDirective directive = ZSTD_e_continue;
+    U64 pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+
+    /* stats */
+    ZSTD_frameProgression previous_zfp_update = { 0, 0, 0, 0, 0, 0 };
+    ZSTD_frameProgression previous_zfp_correction = { 0, 0, 0, 0, 0, 0 };
+    typedef enum { noChange, slower, faster } speedChange_e;
+    speedChange_e speedChange = noChange;
+    unsigned flushWaiting = 0;
+    unsigned inputPresented = 0;
+    unsigned inputBlocked = 0;
+    unsigned lastJobID = 0;
+    UTIL_time_t lastAdaptTime = UTIL_getTime();
+    U64 const adaptEveryMicro = REFRESH_RATE;
+
+    UTIL_HumanReadableSize_t const file_hrs = UTIL_makeHumanReadableSize(fileSize);
+
+    DISPLAYLEVEL(6, "compression using zstd format \n");
+
+    /* init */
+    if (fileSize != UTIL_FILESIZE_UNKNOWN) {
+        pledgedSrcSize = fileSize;
+        CHECK(ZSTD_CCtx_setPledgedSrcSize(ress.cctx, fileSize));
+    } else if (prefs->streamSrcSize > 0) {
+      /* unknown source size; use the declared stream size */
+      pledgedSrcSize = prefs->streamSrcSize;
+      CHECK( ZSTD_CCtx_setPledgedSrcSize(ress.cctx, prefs->streamSrcSize) );
+    }
+
+    {   int windowLog;
+        UTIL_HumanReadableSize_t windowSize;
+        CHECK(ZSTD_CCtx_getParameter(ress.cctx, ZSTD_c_windowLog, &windowLog));
+        if (windowLog == 0) {
+            if (prefs->ldmFlag) {
+                /* If long mode is set without a window size libzstd will set this size internally */
+                windowLog = ZSTD_WINDOWLOG_LIMIT_DEFAULT;
+            } else {
+                const ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, fileSize, 0);
+                windowLog = (int)cParams.windowLog;
+            }
+        }
+        windowSize = UTIL_makeHumanReadableSize(MAX(1ULL, MIN(1ULL << windowLog, pledgedSrcSize)));
+        DISPLAYLEVEL(4, "Decompression will require %.*f%s of memory\n", windowSize.precision, windowSize.value, windowSize.suffix);
+    }
+
+    /* Main compression loop */
+    do {
+        size_t stillToFlush;
+        /* Fill input Buffer */
+        size_t const inSize = AIO_ReadPool_fillBuffer(ress.readCtx, ZSTD_CStreamInSize());
+        ZSTD_inBuffer inBuff = setInBuffer( ress.readCtx->srcBuffer, ress.readCtx->srcBufferLoaded, 0 );
+        DISPLAYLEVEL(6, "fread %u bytes from source \n", (unsigned)inSize);
+        *readsize += inSize;
+
+        if ((ress.readCtx->srcBufferLoaded == 0) || (*readsize == fileSize))
+            directive = ZSTD_e_end;
+
+        stillToFlush = 1;
+        while ((inBuff.pos != inBuff.size)   /* input buffer must be entirely ingested */
+            || (directive == ZSTD_e_end && stillToFlush != 0) ) {
+
+            size_t const oldIPos = inBuff.pos;
+            ZSTD_outBuffer outBuff = setOutBuffer( writeJob->buffer, writeJob->bufferSize, 0 );
+            size_t const toFlushNow = ZSTD_toFlushNow(ress.cctx);
+            CHECK_V(stillToFlush, ZSTD_compressStream2(ress.cctx, &outBuff, &inBuff, directive));
+            AIO_ReadPool_consumeBytes(ress.readCtx, inBuff.pos - oldIPos);
+
+            /* count stats */
+            inputPresented++;
+            if (oldIPos == inBuff.pos) inputBlocked++;  /* input buffer is full and can't take any more : input speed is faster than consumption rate */
+            if (!toFlushNow) flushWaiting = 1;
+
+            /* Write compressed stream */
+            DISPLAYLEVEL(6, "ZSTD_compress_generic(end:%u) => input pos(%u)<=(%u)size ; output generated %u bytes \n",
+                         (unsigned)directive, (unsigned)inBuff.pos, (unsigned)inBuff.size, (unsigned)outBuff.pos);
+            if (outBuff.pos) {
+                writeJob->usedBufferSize = outBuff.pos;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+                compressedfilesize += outBuff.pos;
+            }
+
+            /* adaptive mode : statistics measurement and speed correction */
+            if (prefs->adaptiveMode && UTIL_clockSpanMicro(lastAdaptTime) > adaptEveryMicro) {
+                ZSTD_frameProgression const zfp = ZSTD_getFrameProgression(ress.cctx);
+
+                lastAdaptTime = UTIL_getTime();
+
+                /* check output speed */
+                if (zfp.currentJobID > 1) {  /* only possible if nbWorkers >= 1 */
+
+                    unsigned long long newlyProduced = zfp.produced - previous_zfp_update.produced;
+                    unsigned long long newlyFlushed = zfp.flushed - previous_zfp_update.flushed;
+                    assert(zfp.produced >= previous_zfp_update.produced);
+                    assert(prefs->nbWorkers >= 1);
+
+                    /* test if compression is blocked
+                        * either because output is slow and all buffers are full
+                        * or because input is slow and no job can start while waiting for at least one buffer to be filled.
+                        * note : exclude starting part, since currentJobID > 1 */
+                    if ( (zfp.consumed == previous_zfp_update.consumed)   /* no data compressed : no data available, or no more buffer to compress to, OR compression is really slow (compression of a single block is slower than update rate)*/
+                        && (zfp.nbActiveWorkers == 0)                       /* confirmed : no compression ongoing */
+                        ) {
+                        DISPLAYLEVEL(6, "all buffers full : compression stopped => slow down \n")
+                        speedChange = slower;
+                    }
+
+                    previous_zfp_update = zfp;
+
+                    if ( (newlyProduced > (newlyFlushed * 9 / 8))   /* compression produces more data than output can flush (though production can be spiky, due to work unit : (N==4)*block sizes) */
+                        && (flushWaiting == 0)                        /* flush speed was never slowed by lack of production, so it's operating at max capacity */
+                        ) {
+                        DISPLAYLEVEL(6, "compression faster than flush (%llu > %llu), and flushed was never slowed down by lack of production => slow down \n", newlyProduced, newlyFlushed);
+                        speedChange = slower;
+                    }
+                    flushWaiting = 0;
+                }
+
+                /* course correct only if there is at least one new job completed */
+                if (zfp.currentJobID > lastJobID) {
+                    DISPLAYLEVEL(6, "compression level adaptation check \n")
+
+                    /* check input speed */
+                    if (zfp.currentJobID > (unsigned)(prefs->nbWorkers+1)) {   /* warm up period, to fill all workers */
+                        if (inputBlocked <= 0) {
+                            DISPLAYLEVEL(6, "input is never blocked => input is slower than ingestion \n");
+                            speedChange = slower;
+                        } else if (speedChange == noChange) {
+                            unsigned long long newlyIngested = zfp.ingested - previous_zfp_correction.ingested;
+                            unsigned long long newlyConsumed = zfp.consumed - previous_zfp_correction.consumed;
+                            unsigned long long newlyProduced = zfp.produced - previous_zfp_correction.produced;
+                            unsigned long long newlyFlushed  = zfp.flushed  - previous_zfp_correction.flushed;
+                            previous_zfp_correction = zfp;
+                            assert(inputPresented > 0);
+                            DISPLAYLEVEL(6, "input blocked %u/%u(%.2f) - ingested:%u vs %u:consumed - flushed:%u vs %u:produced \n",
+                                            inputBlocked, inputPresented, (double)inputBlocked/inputPresented*100,
+                                            (unsigned)newlyIngested, (unsigned)newlyConsumed,
+                                            (unsigned)newlyFlushed, (unsigned)newlyProduced);
+                            if ( (inputBlocked > inputPresented / 8)     /* input is waiting often, because input buffers is full : compression or output too slow */
+                                && (newlyFlushed * 33 / 32 > newlyProduced)  /* flush everything that is produced */
+                                && (newlyIngested * 33 / 32 > newlyConsumed) /* input speed as fast or faster than compression speed */
+                            ) {
+                                DISPLAYLEVEL(6, "recommend faster as in(%llu) >= (%llu)comp(%llu) <= out(%llu) \n",
+                                                newlyIngested, newlyConsumed, newlyProduced, newlyFlushed);
+                                speedChange = faster;
+                            }
+                        }
+                        inputBlocked = 0;
+                        inputPresented = 0;
+                    }
+
+                    if (speedChange == slower) {
+                        DISPLAYLEVEL(6, "slower speed , higher compression \n")
+                        compressionLevel ++;
+                        if (compressionLevel > ZSTD_maxCLevel()) compressionLevel = ZSTD_maxCLevel();
+                        if (compressionLevel > prefs->maxAdaptLevel) compressionLevel = prefs->maxAdaptLevel;
+                        compressionLevel += (compressionLevel == 0);   /* skip 0 */
+                        ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_compressionLevel, compressionLevel);
+                    }
+                    if (speedChange == faster) {
+                        DISPLAYLEVEL(6, "faster speed , lighter compression \n")
+                        compressionLevel --;
+                        if (compressionLevel < prefs->minAdaptLevel) compressionLevel = prefs->minAdaptLevel;
+                        compressionLevel -= (compressionLevel == 0);   /* skip 0 */
+                        ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_compressionLevel, compressionLevel);
+                    }
+                    speedChange = noChange;
+
+                    lastJobID = zfp.currentJobID;
+                }  /* if (zfp.currentJobID > lastJobID) */
+            } /* if (prefs->adaptiveMode && UTIL_clockSpanMicro(lastAdaptTime) > adaptEveryMicro) */
+
+            /* display notification */
+            if (SHOULD_DISPLAY_PROGRESS() && READY_FOR_UPDATE()) {
+                ZSTD_frameProgression const zfp = ZSTD_getFrameProgression(ress.cctx);
+                double const cShare = (double)zfp.produced / (double)(zfp.consumed + !zfp.consumed/*avoid div0*/) * 100;
+                UTIL_HumanReadableSize_t const buffered_hrs = UTIL_makeHumanReadableSize(zfp.ingested - zfp.consumed);
+                UTIL_HumanReadableSize_t const consumed_hrs = UTIL_makeHumanReadableSize(zfp.consumed);
+                UTIL_HumanReadableSize_t const produced_hrs = UTIL_makeHumanReadableSize(zfp.produced);
+
+                DELAY_NEXT_UPDATE();
+
+                /* display progress notifications */
+                DISPLAY_PROGRESS("\r%79s\r", "");    /* Clear out the current displayed line */
+                if (g_display_prefs.displayLevel >= 3) {
+                    /* Verbose progress update */
+                    DISPLAY_PROGRESS(
+                            "(L%i) Buffered:%5.*f%s - Consumed:%5.*f%s - Compressed:%5.*f%s => %.2f%% ",
+                            compressionLevel,
+                            buffered_hrs.precision, buffered_hrs.value, buffered_hrs.suffix,
+                            consumed_hrs.precision, consumed_hrs.value, consumed_hrs.suffix,
+                            produced_hrs.precision, produced_hrs.value, produced_hrs.suffix,
+                            cShare );
+                } else {
+                    /* Require level 2 or forcibly displayed progress counter for summarized updates */
+                    if (fCtx->nbFilesTotal > 1) {
+                        size_t srcFileNameSize = strlen(srcFileName);
+                        /* Ensure that the string we print is roughly the same size each time */
+                        if (srcFileNameSize > 18) {
+                            const char* truncatedSrcFileName = srcFileName + srcFileNameSize - 15;
+                            DISPLAY_PROGRESS("Compress: %u/%u files. Current: ...%s ",
+                                        fCtx->currFileIdx+1, fCtx->nbFilesTotal, truncatedSrcFileName);
+                        } else {
+                            DISPLAY_PROGRESS("Compress: %u/%u files. Current: %*s ",
+                                        fCtx->currFileIdx+1, fCtx->nbFilesTotal, (int)(18-srcFileNameSize), srcFileName);
+                        }
+                    }
+                    DISPLAY_PROGRESS("Read:%6.*f%4s ", consumed_hrs.precision, consumed_hrs.value, consumed_hrs.suffix);
+                    if (fileSize != UTIL_FILESIZE_UNKNOWN)
+                        DISPLAY_PROGRESS("/%6.*f%4s", file_hrs.precision, file_hrs.value, file_hrs.suffix);
+                    DISPLAY_PROGRESS(" ==> %2.f%%", cShare);
+                }
+            }  /* if (SHOULD_DISPLAY_PROGRESS() && READY_FOR_UPDATE()) */
+        }  /* while ((inBuff.pos != inBuff.size) */
+    } while (directive != ZSTD_e_end);
+
+    if (fileSize != UTIL_FILESIZE_UNKNOWN && *readsize != fileSize) {
+        EXM_THROW(27, "Read error : Incomplete read : %llu / %llu B",
+                (unsigned long long)*readsize, (unsigned long long)fileSize);
+    }
+
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ressPtr->writeCtx);
+
+    return compressedfilesize;
+}
+
+/*! FIO_compressFilename_internal() :
+ *  same as FIO_compressFilename_extRess(), with `ress.desFile` already opened.
+ *  @return : 0 : compression completed correctly,
+ *            1 : missing or pb opening srcFileName
+ */
+static int
+FIO_compressFilename_internal(FIO_ctx_t* const fCtx,
+                              FIO_prefs_t* const prefs,
+                              cRess_t ress,
+                              const char* dstFileName, const char* srcFileName,
+                              int compressionLevel)
+{
+    UTIL_time_t const timeStart = UTIL_getTime();
+    clock_t const cpuStart = clock();
+    U64 readsize = 0;
+    U64 compressedfilesize = 0;
+    U64 const fileSize = UTIL_getFileSize(srcFileName);
+    DISPLAYLEVEL(5, "%s: %llu bytes \n", srcFileName, (unsigned long long)fileSize);
+
+    /* compression format selection */
+    switch (prefs->compressionType) {
+        default:
+        case FIO_zstdCompression:
+            compressedfilesize = FIO_compressZstdFrame(fCtx, prefs, &ress, srcFileName, fileSize, compressionLevel, &readsize);
+            break;
+
+        case FIO_gzipCompression:
+#ifdef ZSTD_GZCOMPRESS
+            compressedfilesize = FIO_compressGzFrame(&ress, srcFileName, fileSize, compressionLevel, &readsize);
+#else
+            (void)compressionLevel;
+            EXM_THROW(20, "zstd: %s: file cannot be compressed as gzip (zstd compiled without ZSTD_GZCOMPRESS) -- ignored \n",
+                            srcFileName);
+#endif
+            break;
+
+        case FIO_xzCompression:
+        case FIO_lzmaCompression:
+#ifdef ZSTD_LZMACOMPRESS
+            compressedfilesize = FIO_compressLzmaFrame(&ress, srcFileName, fileSize, compressionLevel, &readsize, prefs->compressionType==FIO_lzmaCompression);
+#else
+            (void)compressionLevel;
+            EXM_THROW(20, "zstd: %s: file cannot be compressed as xz/lzma (zstd compiled without ZSTD_LZMACOMPRESS) -- ignored \n",
+                            srcFileName);
+#endif
+            break;
+
+        case FIO_lz4Compression:
+#ifdef ZSTD_LZ4COMPRESS
+            compressedfilesize = FIO_compressLz4Frame(&ress, srcFileName, fileSize, compressionLevel, prefs->checksumFlag, &readsize);
+#else
+            (void)compressionLevel;
+            EXM_THROW(20, "zstd: %s: file cannot be compressed as lz4 (zstd compiled without ZSTD_LZ4COMPRESS) -- ignored \n",
+                            srcFileName);
+#endif
+            break;
+    }
+
+    /* Status */
+    fCtx->totalBytesInput += (size_t)readsize;
+    fCtx->totalBytesOutput += (size_t)compressedfilesize;
+    DISPLAY_PROGRESS("\r%79s\r", "");
+    if (FIO_shouldDisplayFileSummary(fCtx)) {
+        UTIL_HumanReadableSize_t hr_isize = UTIL_makeHumanReadableSize((U64) readsize);
+        UTIL_HumanReadableSize_t hr_osize = UTIL_makeHumanReadableSize((U64) compressedfilesize);
+        if (readsize == 0) {
+            DISPLAY_SUMMARY("%-20s :  (%6.*f%s => %6.*f%s, %s) \n",
+                srcFileName,
+                hr_isize.precision, hr_isize.value, hr_isize.suffix,
+                hr_osize.precision, hr_osize.value, hr_osize.suffix,
+                dstFileName);
+        } else {
+            DISPLAY_SUMMARY("%-20s :%6.2f%%   (%6.*f%s => %6.*f%s, %s) \n",
+                srcFileName,
+                (double)compressedfilesize / (double)readsize * 100,
+                hr_isize.precision, hr_isize.value, hr_isize.suffix,
+                hr_osize.precision, hr_osize.value, hr_osize.suffix,
+                dstFileName);
+        }
+    }
+
+    /* Elapsed Time and CPU Load */
+    {   clock_t const cpuEnd = clock();
+        double const cpuLoad_s = (double)(cpuEnd - cpuStart) / CLOCKS_PER_SEC;
+        U64 const timeLength_ns = UTIL_clockSpanNano(timeStart);
+        double const timeLength_s = (double)timeLength_ns / 1000000000;
+        double const cpuLoad_pct = (cpuLoad_s / timeLength_s) * 100;
+        DISPLAYLEVEL(4, "%-20s : Completed in %.2f sec  (cpu load : %.0f%%)\n",
+                        srcFileName, timeLength_s, cpuLoad_pct);
+    }
+    return 0;
+}
+
+
+/*! FIO_compressFilename_dstFile() :
+ *  open dstFileName, or pass-through if ress.file != NULL,
+ *  then start compression with FIO_compressFilename_internal().
+ *  Manages source removal (--rm) and file permissions transfer.
+ *  note : ress.srcFile must be != NULL,
+ *  so reach this function through FIO_compressFilename_srcFile().
+ *  @return : 0 : compression completed correctly,
+ *            1 : pb
+ */
+static int FIO_compressFilename_dstFile(FIO_ctx_t* const fCtx,
+                                        FIO_prefs_t* const prefs,
+                                        cRess_t ress,
+                                        const char* dstFileName,
+                                        const char* srcFileName,
+                                        const stat_t* srcFileStat,
+                                        int compressionLevel)
+{
+    int closeDstFile = 0;
+    int result;
+    int transferStat = 0;
+    int dstFd = -1;
+
+    assert(AIO_ReadPool_getFile(ress.readCtx) != NULL);
+    if (AIO_WritePool_getFile(ress.writeCtx) == NULL) {
+        int dstFileInitialPermissions = DEFAULT_FILE_PERMISSIONS;
+        if ( strcmp (srcFileName, stdinmark)
+          && strcmp (dstFileName, stdoutmark)
+          && UTIL_isRegularFileStat(srcFileStat) ) {
+            transferStat = 1;
+            dstFileInitialPermissions = TEMPORARY_FILE_PERMISSIONS;
+        }
+
+        closeDstFile = 1;
+        DISPLAYLEVEL(6, "FIO_compressFilename_dstFile: opening dst: %s \n", dstFileName);
+        {   FILE *dstFile = FIO_openDstFile(fCtx, prefs, srcFileName, dstFileName, dstFileInitialPermissions);
+            if (dstFile==NULL) return 1;  /* could not open dstFileName */
+            dstFd = fileno(dstFile);
+            AIO_WritePool_setFile(ress.writeCtx, dstFile);
+        }
+        /* Must only be added after FIO_openDstFile() succeeds.
+         * Otherwise we may delete the destination file if it already exists,
+         * and the user presses Ctrl-C when asked if they wish to overwrite.
+         */
+        addHandler(dstFileName);
+    }
+
+    result = FIO_compressFilename_internal(fCtx, prefs, ress, dstFileName, srcFileName, compressionLevel);
+
+    if (closeDstFile) {
+        clearHandler();
+
+        if (transferStat) {
+            UTIL_setFDStat(dstFd, dstFileName, srcFileStat);
+        }
+
+        DISPLAYLEVEL(6, "FIO_compressFilename_dstFile: closing dst: %s \n", dstFileName);
+        if (AIO_WritePool_closeFile(ress.writeCtx)) { /* error closing file */
+            DISPLAYLEVEL(1, "zstd: %s: %s \n", dstFileName, strerror(errno));
+            result=1;
+        }
+
+        if (transferStat) {
+            UTIL_utime(dstFileName, srcFileStat);
+        }
+
+        if ( (result != 0)  /* operation failure */
+          && strcmp(dstFileName, stdoutmark)  /* special case : don't remove() stdout */
+          ) {
+            FIO_removeFile(dstFileName); /* remove compression artefact; note don't do anything special if remove() fails */
+        }
+    }
+
+    return result;
+}
+
+/* List used to compare file extensions (used with --exclude-compressed flag)
+* Different from the suffixList and should only apply to ZSTD compress operationResult
+*/
+static const char *compressedFileExtensions[] = {
+    ZSTD_EXTENSION,
+    TZSTD_EXTENSION,
+    GZ_EXTENSION,
+    TGZ_EXTENSION,
+    LZMA_EXTENSION,
+    XZ_EXTENSION,
+    TXZ_EXTENSION,
+    LZ4_EXTENSION,
+    TLZ4_EXTENSION,
+    ".7z",
+    ".aa3",
+    ".aac",
+    ".aar",
+    ".ace",
+    ".alac",
+    ".ape",
+    ".apk",
+    ".apng",
+    ".arc",
+    ".archive",
+    ".arj",
+    ".ark",
+    ".asf",
+    ".avi",
+    ".avif",
+    ".ba",
+    ".br",
+    ".bz2",
+    ".cab",
+    ".cdx",
+    ".chm",
+    ".cr2",
+    ".divx",
+    ".dmg",
+    ".dng",
+    ".docm",
+    ".docx",
+    ".dotm",
+    ".dotx",
+    ".dsft",
+    ".ear",
+    ".eftx",
+    ".emz",
+    ".eot",
+    ".epub",
+    ".f4v",
+    ".flac",
+    ".flv",
+    ".gho",
+    ".gif",
+    ".gifv",
+    ".gnp",
+    ".iso",
+    ".jar",
+    ".jpeg",
+    ".jpg",
+    ".jxl",
+    ".lz",
+    ".lzh",
+    ".m4a",
+    ".m4v",
+    ".mkv",
+    ".mov",
+    ".mp2",
+    ".mp3",
+    ".mp4",
+    ".mpa",
+    ".mpc",
+    ".mpe",
+    ".mpeg",
+    ".mpg",
+    ".mpl",
+    ".mpv",
+    ".msi",
+    ".odp",
+    ".ods",
+    ".odt",
+    ".ogg",
+    ".ogv",
+    ".otp",
+    ".ots",
+    ".ott",
+    ".pea",
+    ".png",
+    ".pptx",
+    ".qt",
+    ".rar",
+    ".s7z",
+    ".sfx",
+    ".sit",
+    ".sitx",
+    ".sqx",
+    ".svgz",
+    ".swf",
+    ".tbz2",
+    ".tib",
+    ".tlz",
+    ".vob",
+    ".war",
+    ".webm",
+    ".webp",
+    ".wma",
+    ".wmv",
+    ".woff",
+    ".woff2",
+    ".wvl",
+    ".xlsx",
+    ".xpi",
+    ".xps",
+    ".zip",
+    ".zipx",
+    ".zoo",
+    ".zpaq",
+    NULL
+};
+
+/*! FIO_compressFilename_srcFile() :
+ *  @return : 0 : compression completed correctly,
+ *            1 : missing or pb opening srcFileName
+ */
+static int
+FIO_compressFilename_srcFile(FIO_ctx_t* const fCtx,
+                             FIO_prefs_t* const prefs,
+                             cRess_t ress,
+                             const char* dstFileName,
+                             const char* srcFileName,
+                             int compressionLevel)
+{
+    int result;
+    FILE* srcFile;
+    stat_t srcFileStat;
+    U64 fileSize = UTIL_FILESIZE_UNKNOWN;
+    DISPLAYLEVEL(6, "FIO_compressFilename_srcFile: %s \n", srcFileName);
+
+    if (strcmp(srcFileName, stdinmark)) {
+        if (UTIL_stat(srcFileName, &srcFileStat)) {
+            /* failure to stat at all is handled during opening */
+
+            /* ensure src is not a directory */
+            if (UTIL_isDirectoryStat(&srcFileStat)) {
+                DISPLAYLEVEL(1, "zstd: %s is a directory -- ignored \n", srcFileName);
+                return 1;
+            }
+
+            /* ensure src is not the same as dict (if present) */
+            if (ress.dictFileName != NULL && UTIL_isSameFileStat(srcFileName, ress.dictFileName, &srcFileStat, &ress.dictFileStat)) {
+                DISPLAYLEVEL(1, "zstd: cannot use %s as an input file and dictionary \n", srcFileName);
+                return 1;
+            }
+        }
+    }
+
+    /* Check if "srcFile" is compressed. Only done if --exclude-compressed flag is used
+    * YES => ZSTD will skip compression of the file and will return 0.
+    * NO => ZSTD will resume with compress operation.
+    */
+    if (prefs->excludeCompressedFiles == 1 && UTIL_isCompressedFile(srcFileName, compressedFileExtensions)) {
+        DISPLAYLEVEL(4, "File is already compressed : %s \n", srcFileName);
+        return 0;
+    }
+
+    srcFile = FIO_openSrcFile(prefs, srcFileName, &srcFileStat);
+    if (srcFile == NULL) return 1;   /* srcFile could not be opened */
+
+    /* Don't use AsyncIO for small files */
+    if (strcmp(srcFileName, stdinmark)) /* Stdin doesn't have stats */
+        fileSize = UTIL_getFileSizeStat(&srcFileStat);
+    if(fileSize != UTIL_FILESIZE_UNKNOWN && fileSize < ZSTD_BLOCKSIZE_MAX * 3) {
+        AIO_ReadPool_setAsync(ress.readCtx, 0);
+        AIO_WritePool_setAsync(ress.writeCtx, 0);
+    } else {
+        AIO_ReadPool_setAsync(ress.readCtx, 1);
+        AIO_WritePool_setAsync(ress.writeCtx, 1);
+    }
+
+    AIO_ReadPool_setFile(ress.readCtx, srcFile);
+    result = FIO_compressFilename_dstFile(
+            fCtx, prefs, ress,
+            dstFileName, srcFileName,
+            &srcFileStat, compressionLevel);
+    AIO_ReadPool_closeFile(ress.readCtx);
+
+    if ( prefs->removeSrcFile  /* --rm */
+      && result == 0           /* success */
+      && strcmp(srcFileName, stdinmark)  /* exception : don't erase stdin */
+      ) {
+        /* We must clear the handler, since after this point calling it would
+         * delete both the source and destination files.
+         */
+        clearHandler();
+        if (FIO_removeFile(srcFileName))
+            EXM_THROW(1, "zstd: %s: %s", srcFileName, strerror(errno));
+    }
+    return result;
+}
+
+static const char*
+checked_index(const char* options[], size_t length, size_t index) {
+    assert(index < length);
+    /* Necessary to avoid warnings since -O3 will omit the above `assert` */
+    (void) length;
+    return options[index];
+}
+
+#define INDEX(options, index) checked_index((options), sizeof(options)  / sizeof(char*), (size_t)(index))
+
+void FIO_displayCompressionParameters(const FIO_prefs_t* prefs)
+{
+    static const char* formatOptions[5] = {ZSTD_EXTENSION, GZ_EXTENSION, XZ_EXTENSION,
+                                           LZMA_EXTENSION, LZ4_EXTENSION};
+    static const char* sparseOptions[3] = {" --no-sparse", "", " --sparse"};
+    static const char* checkSumOptions[3] = {" --no-check", "", " --check"};
+    static const char* rowMatchFinderOptions[3] = {"", " --no-row-match-finder", " --row-match-finder"};
+    static const char* compressLiteralsOptions[3] = {"", " --compress-literals", " --no-compress-literals"};
+
+    assert(g_display_prefs.displayLevel >= 4);
+
+    DISPLAY("--format=%s", formatOptions[prefs->compressionType]);
+    DISPLAY("%s", INDEX(sparseOptions, prefs->sparseFileSupport));
+    DISPLAY("%s", prefs->dictIDFlag ? "" : " --no-dictID");
+    DISPLAY("%s", INDEX(checkSumOptions, prefs->checksumFlag));
+    DISPLAY(" --jobsize=%d", prefs->jobSize);
+    if (prefs->adaptiveMode)
+        DISPLAY(" --adapt=min=%d,max=%d", prefs->minAdaptLevel, prefs->maxAdaptLevel);
+    DISPLAY("%s", INDEX(rowMatchFinderOptions, prefs->useRowMatchFinder));
+    DISPLAY("%s", prefs->rsyncable ? " --rsyncable" : "");
+    if (prefs->streamSrcSize)
+        DISPLAY(" --stream-size=%u", (unsigned) prefs->streamSrcSize);
+    if (prefs->srcSizeHint)
+        DISPLAY(" --size-hint=%d", prefs->srcSizeHint);
+    if (prefs->targetCBlockSize)
+        DISPLAY(" --target-compressed-block-size=%u", (unsigned) prefs->targetCBlockSize);
+    DISPLAY("%s", INDEX(compressLiteralsOptions, prefs->literalCompressionMode));
+    DISPLAY(" --memory=%u", prefs->memLimit ? prefs->memLimit : 128 MB);
+    DISPLAY(" --threads=%d", prefs->nbWorkers);
+    DISPLAY("%s", prefs->excludeCompressedFiles ? " --exclude-compressed" : "");
+    DISPLAY(" --%scontent-size", prefs->contentSize ? "" : "no-");
+    DISPLAY("\n");
+}
+
+#undef INDEX
+
+int FIO_compressFilename(FIO_ctx_t* const fCtx, FIO_prefs_t* const prefs, const char* dstFileName,
+                         const char* srcFileName, const char* dictFileName,
+                         int compressionLevel, ZSTD_compressionParameters comprParams)
+{
+    cRess_t ress = FIO_createCResources(prefs, dictFileName, UTIL_getFileSize(srcFileName), compressionLevel, comprParams);
+    int const result = FIO_compressFilename_srcFile(fCtx, prefs, ress, dstFileName, srcFileName, compressionLevel);
+
+#define DISPLAY_LEVEL_DEFAULT 2
+
+    FIO_freeCResources(&ress);
+    return result;
+}
+
+/* FIO_determineCompressedName() :
+ * create a destination filename for compressed srcFileName.
+ * @return a pointer to it.
+ * This function never returns an error (it may abort() in case of pb)
+ */
+static const char*
+FIO_determineCompressedName(const char* srcFileName, const char* outDirName, const char* suffix)
+{
+    static size_t dfnbCapacity = 0;
+    static char* dstFileNameBuffer = NULL;   /* using static allocation : this function cannot be multi-threaded */
+    char* outDirFilename = NULL;
+    size_t sfnSize = strlen(srcFileName);
+    size_t const srcSuffixLen = strlen(suffix);
+
+    if(!strcmp(srcFileName, stdinmark)) {
+        return stdoutmark;
+    }
+
+    if (outDirName) {
+        outDirFilename = FIO_createFilename_fromOutDir(srcFileName, outDirName, srcSuffixLen);
+        sfnSize = strlen(outDirFilename);
+        assert(outDirFilename != NULL);
+    }
+
+    if (dfnbCapacity <= sfnSize+srcSuffixLen+1) {
+        /* resize buffer for dstName */
+        free(dstFileNameBuffer);
+        dfnbCapacity = sfnSize + srcSuffixLen + 30;
+        dstFileNameBuffer = (char*)malloc(dfnbCapacity);
+        if (!dstFileNameBuffer) {
+            EXM_THROW(30, "zstd: %s", strerror(errno));
+        }
+    }
+    assert(dstFileNameBuffer != NULL);
+
+    if (outDirFilename) {
+        memcpy(dstFileNameBuffer, outDirFilename, sfnSize);
+        free(outDirFilename);
+    } else {
+        memcpy(dstFileNameBuffer, srcFileName, sfnSize);
+    }
+    memcpy(dstFileNameBuffer+sfnSize, suffix, srcSuffixLen+1 /* Include terminating null */);
+    return dstFileNameBuffer;
+}
+
+static unsigned long long FIO_getLargestFileSize(const char** inFileNames, unsigned nbFiles)
+{
+    size_t i;
+    unsigned long long fileSize, maxFileSize = 0;
+    for (i = 0; i < nbFiles; i++) {
+        fileSize = UTIL_getFileSize(inFileNames[i]);
+        maxFileSize = fileSize > maxFileSize ? fileSize : maxFileSize;
+    }
+    return maxFileSize;
+}
+
+/* FIO_compressMultipleFilenames() :
+ * compress nbFiles files
+ * into either one destination (outFileName),
+ * or into one file each (outFileName == NULL, but suffix != NULL),
+ * or into a destination folder (specified with -O)
+ */
+int FIO_compressMultipleFilenames(FIO_ctx_t* const fCtx,
+                                  FIO_prefs_t* const prefs,
+                                  const char** inFileNamesTable,
+                                  const char* outMirroredRootDirName,
+                                  const char* outDirName,
+                                  const char* outFileName, const char* suffix,
+                                  const char* dictFileName, int compressionLevel,
+                                  ZSTD_compressionParameters comprParams)
+{
+    int status;
+    int error = 0;
+    cRess_t ress = FIO_createCResources(prefs, dictFileName,
+        FIO_getLargestFileSize(inFileNamesTable, (unsigned)fCtx->nbFilesTotal),
+        compressionLevel, comprParams);
+
+    /* init */
+    assert(outFileName != NULL || suffix != NULL);
+    if (outFileName != NULL) {   /* output into a single destination (stdout typically) */
+        FILE *dstFile;
+        if (FIO_multiFilesConcatWarning(fCtx, prefs, outFileName, 1 /* displayLevelCutoff */)) {
+            FIO_freeCResources(&ress);
+            return 1;
+        }
+        dstFile = FIO_openDstFile(fCtx, prefs, NULL, outFileName, DEFAULT_FILE_PERMISSIONS);
+        if (dstFile == NULL) {  /* could not open outFileName */
+            error = 1;
+        } else {
+            AIO_WritePool_setFile(ress.writeCtx, dstFile);
+            for (; fCtx->currFileIdx < fCtx->nbFilesTotal; ++fCtx->currFileIdx) {
+                status = FIO_compressFilename_srcFile(fCtx, prefs, ress, outFileName, inFileNamesTable[fCtx->currFileIdx], compressionLevel);
+                if (!status) fCtx->nbFilesProcessed++;
+                error |= status;
+            }
+            if (AIO_WritePool_closeFile(ress.writeCtx))
+                EXM_THROW(29, "Write error (%s) : cannot properly close %s",
+                            strerror(errno), outFileName);
+        }
+    } else {
+        if (outMirroredRootDirName)
+            UTIL_mirrorSourceFilesDirectories(inFileNamesTable, (unsigned)fCtx->nbFilesTotal, outMirroredRootDirName);
+
+        for (; fCtx->currFileIdx < fCtx->nbFilesTotal; ++fCtx->currFileIdx) {
+            const char* const srcFileName = inFileNamesTable[fCtx->currFileIdx];
+            const char* dstFileName = NULL;
+            if (outMirroredRootDirName) {
+                char* validMirroredDirName = UTIL_createMirroredDestDirName(srcFileName, outMirroredRootDirName);
+                if (validMirroredDirName) {
+                    dstFileName = FIO_determineCompressedName(srcFileName, validMirroredDirName, suffix);
+                    free(validMirroredDirName);
+                } else {
+                    DISPLAYLEVEL(2, "zstd: --output-dir-mirror cannot compress '%s' into '%s' \n", srcFileName, outMirroredRootDirName);
+                    error=1;
+                    continue;
+                }
+            } else {
+                dstFileName = FIO_determineCompressedName(srcFileName, outDirName, suffix);  /* cannot fail */
+            }
+            status = FIO_compressFilename_srcFile(fCtx, prefs, ress, dstFileName, srcFileName, compressionLevel);
+            if (!status) fCtx->nbFilesProcessed++;
+            error |= status;
+        }
+
+        if (outDirName)
+            FIO_checkFilenameCollisions(inFileNamesTable , (unsigned)fCtx->nbFilesTotal);
+    }
+
+    if (FIO_shouldDisplayMultipleFileSummary(fCtx)) {
+        UTIL_HumanReadableSize_t hr_isize = UTIL_makeHumanReadableSize((U64) fCtx->totalBytesInput);
+        UTIL_HumanReadableSize_t hr_osize = UTIL_makeHumanReadableSize((U64) fCtx->totalBytesOutput);
+
+        DISPLAY_PROGRESS("\r%79s\r", "");
+        if (fCtx->totalBytesInput == 0) {
+            DISPLAY_SUMMARY("%3d files compressed : (%6.*f%4s => %6.*f%4s)\n",
+                            fCtx->nbFilesProcessed,
+                            hr_isize.precision, hr_isize.value, hr_isize.suffix,
+                            hr_osize.precision, hr_osize.value, hr_osize.suffix);
+        } else {
+            DISPLAY_SUMMARY("%3d files compressed : %.2f%% (%6.*f%4s => %6.*f%4s)\n",
+                            fCtx->nbFilesProcessed,
+                            (double)fCtx->totalBytesOutput/((double)fCtx->totalBytesInput)*100,
+                            hr_isize.precision, hr_isize.value, hr_isize.suffix,
+                            hr_osize.precision, hr_osize.value, hr_osize.suffix);
+        }
+    }
+
+    FIO_freeCResources(&ress);
+    return error;
+}
+
+#endif /* #ifndef ZSTD_NOCOMPRESS */
+
+
+
+#ifndef ZSTD_NODECOMPRESS
+
+/* **************************************************************************
+ *  Decompression
+ ***************************************************************************/
+typedef struct {
+    FIO_Dict_t dict;
+    ZSTD_DStream* dctx;
+    WritePoolCtx_t *writeCtx;
+    ReadPoolCtx_t *readCtx;
+} dRess_t;
+
+static dRess_t FIO_createDResources(FIO_prefs_t* const prefs, const char* dictFileName)
+{
+    int useMMap = prefs->mmapDict == ZSTD_ps_enable;
+    int forceNoUseMMap = prefs->mmapDict == ZSTD_ps_disable;
+    stat_t statbuf;
+    dRess_t ress;
+    memset(&statbuf, 0, sizeof(statbuf));
+    memset(&ress, 0, sizeof(ress));
+
+    FIO_getDictFileStat(dictFileName, &statbuf);
+
+    if (prefs->patchFromMode){
+        U64 const dictSize = UTIL_getFileSizeStat(&statbuf);
+        useMMap |= dictSize > prefs->memLimit;
+        FIO_adjustMemLimitForPatchFromMode(prefs, dictSize, 0 /* just use the dict size */);
+    }
+
+    /* Allocation */
+    ress.dctx = ZSTD_createDStream();
+    if (ress.dctx==NULL)
+        EXM_THROW(60, "Error: %s : can't create ZSTD_DStream", strerror(errno));
+    CHECK( ZSTD_DCtx_setMaxWindowSize(ress.dctx, prefs->memLimit) );
+    CHECK( ZSTD_DCtx_setParameter(ress.dctx, ZSTD_d_forceIgnoreChecksum, !prefs->checksumFlag));
+
+    /* dictionary */
+    {
+        FIO_dictBufferType_t dictBufferType = (useMMap && !forceNoUseMMap) ? FIO_mmapDict : FIO_mallocDict;
+        FIO_initDict(&ress.dict, dictFileName, prefs, &statbuf, dictBufferType);
+
+        CHECK(ZSTD_DCtx_reset(ress.dctx, ZSTD_reset_session_only) );
+
+        if (prefs->patchFromMode){
+            CHECK(ZSTD_DCtx_refPrefix(ress.dctx, ress.dict.dictBuffer, ress.dict.dictBufferSize));
+        } else {
+            CHECK(ZSTD_DCtx_loadDictionary_byReference(ress.dctx, ress.dict.dictBuffer, ress.dict.dictBufferSize));
+        }
+    }
+
+    ress.writeCtx = AIO_WritePool_create(prefs, ZSTD_DStreamOutSize());
+    ress.readCtx = AIO_ReadPool_create(prefs, ZSTD_DStreamInSize());
+    return ress;
+}
+
+static void FIO_freeDResources(dRess_t ress)
+{
+    FIO_freeDict(&(ress.dict));
+    CHECK( ZSTD_freeDStream(ress.dctx) );
+    AIO_WritePool_free(ress.writeCtx);
+    AIO_ReadPool_free(ress.readCtx);
+}
+
+/* FIO_passThrough() : just copy input into output, for compatibility with gzip -df mode
+ * @return : 0 (no error) */
+static int FIO_passThrough(dRess_t *ress)
+{
+    size_t const blockSize = MIN(MIN(64 KB, ZSTD_DStreamInSize()), ZSTD_DStreamOutSize());
+    IOJob_t *writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
+    AIO_ReadPool_fillBuffer(ress->readCtx, blockSize);
+
+    while(ress->readCtx->srcBufferLoaded) {
+        size_t writeSize;
+        writeSize = MIN(blockSize, ress->readCtx->srcBufferLoaded);
+        assert(writeSize <= writeJob->bufferSize);
+        memcpy(writeJob->buffer, ress->readCtx->srcBuffer, writeSize);
+        writeJob->usedBufferSize = writeSize;
+        AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+        AIO_ReadPool_consumeBytes(ress->readCtx, writeSize);
+        AIO_ReadPool_fillBuffer(ress->readCtx, blockSize);
+    }
+    assert(ress->readCtx->reachedEof);
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
+    return 0;
+}
+
+/* FIO_zstdErrorHelp() :
+ * detailed error message when requested window size is too large */
+static void
+FIO_zstdErrorHelp(const FIO_prefs_t* const prefs,
+                  const dRess_t* ress,
+                  size_t err,
+                  const char* srcFileName)
+{
+    ZSTD_FrameHeader header;
+
+    /* Help message only for one specific error */
+    if (ZSTD_getErrorCode(err) != ZSTD_error_frameParameter_windowTooLarge)
+        return;
+
+    /* Try to decode the frame header */
+    err = ZSTD_getFrameHeader(&header, ress->readCtx->srcBuffer, ress->readCtx->srcBufferLoaded);
+    if (err == 0) {
+        unsigned long long const windowSize = header.windowSize;
+        unsigned const windowLog = FIO_highbit64(windowSize) + ((windowSize & (windowSize - 1)) != 0);
+        assert(prefs->memLimit > 0);
+        DISPLAYLEVEL(1, "%s : Window size larger than maximum : %llu > %u \n",
+                        srcFileName, windowSize, prefs->memLimit);
+        if (windowLog <= ZSTD_WINDOWLOG_MAX) {
+            unsigned const windowMB = (unsigned)((windowSize >> 20) + ((windowSize & ((1 MB) - 1)) != 0));
+            assert(windowSize < (U64)(1ULL << 52));   /* ensure now overflow for windowMB */
+            DISPLAYLEVEL(1, "%s : Use --long=%u or --memory=%uMB \n",
+                            srcFileName, windowLog, windowMB);
+            return;
+    }   }
+    DISPLAYLEVEL(1, "%s : Window log larger than ZSTD_WINDOWLOG_MAX=%u; not supported \n",
+                    srcFileName, ZSTD_WINDOWLOG_MAX);
+}
+
+/** FIO_decompressFrame() :
+ *  @return : size of decoded zstd frame, or an error code
+ */
+#define FIO_ERROR_FRAME_DECODING   ((unsigned long long)(-2))
+static unsigned long long
+FIO_decompressZstdFrame(FIO_ctx_t* const fCtx, dRess_t* ress,
+                        const FIO_prefs_t* const prefs,
+                        const char* srcFileName,
+                        U64 alreadyDecoded)  /* for multi-frames streams */
+{
+    U64 frameSize = 0;
+    const char* srcFName20 = srcFileName;
+    IOJob_t* writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
+    assert(writeJob);
+
+    /* display last 20 characters only when not --verbose */
+    {   size_t const srcFileLength = strlen(srcFileName);
+        if ((srcFileLength>20) && (g_display_prefs.displayLevel<3))
+            srcFName20 += srcFileLength-20;
+    }
+
+    ZSTD_DCtx_reset(ress->dctx, ZSTD_reset_session_only);
+
+    /* Header loading : ensures ZSTD_getFrameHeader() will succeed */
+    AIO_ReadPool_fillBuffer(ress->readCtx, ZSTD_FRAMEHEADERSIZE_MAX);
+
+    /* Main decompression Loop */
+    while (1) {
+        ZSTD_inBuffer  inBuff = setInBuffer( ress->readCtx->srcBuffer, ress->readCtx->srcBufferLoaded, 0 );
+        ZSTD_outBuffer outBuff= setOutBuffer( writeJob->buffer, writeJob->bufferSize, 0 );
+        size_t const readSizeHint = ZSTD_decompressStream(ress->dctx, &outBuff, &inBuff);
+        UTIL_HumanReadableSize_t const hrs = UTIL_makeHumanReadableSize(alreadyDecoded+frameSize);
+        if (ZSTD_isError(readSizeHint)) {
+            DISPLAYLEVEL(1, "%s : Decoding error (36) : %s \n",
+                            srcFileName, ZSTD_getErrorName(readSizeHint));
+            FIO_zstdErrorHelp(prefs, ress, readSizeHint, srcFileName);
+            AIO_WritePool_releaseIoJob(writeJob);
+            return FIO_ERROR_FRAME_DECODING;
+        }
+
+        /* Write block */
+        writeJob->usedBufferSize = outBuff.pos;
+        AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+        frameSize += outBuff.pos;
+        if (fCtx->nbFilesTotal > 1) {
+            DISPLAYUPDATE_PROGRESS(
+                        "\rDecompress: %2u/%2u files. Current: %s : %.*f%s...    ",
+                        fCtx->currFileIdx+1, fCtx->nbFilesTotal, srcFName20, hrs.precision, hrs.value, hrs.suffix);
+        } else {
+            DISPLAYUPDATE_PROGRESS("\r%-20.20s : %.*f%s...     ",
+                            srcFName20, hrs.precision, hrs.value, hrs.suffix);
+        }
+
+        AIO_ReadPool_consumeBytes(ress->readCtx, inBuff.pos);
+
+        if (readSizeHint == 0) break;   /* end of frame */
+
+        /* Fill input buffer */
+        {   size_t const toDecode = MIN(readSizeHint, ZSTD_DStreamInSize());  /* support large skippable frames */
+            if (ress->readCtx->srcBufferLoaded < toDecode) {
+                size_t const readSize = AIO_ReadPool_fillBuffer(ress->readCtx, toDecode);
+                if (readSize==0) {
+                    DISPLAYLEVEL(1, "%s : Read error (39) : premature end \n",
+                                 srcFileName);
+                    AIO_WritePool_releaseIoJob(writeJob);
+                    return FIO_ERROR_FRAME_DECODING;
+                }
+            }   }   }
+
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
+
+    return frameSize;
+}
+
+
+#ifdef ZSTD_GZDECOMPRESS
+static unsigned long long
+FIO_decompressGzFrame(dRess_t* ress, const char* srcFileName)
+{
+    unsigned long long outFileSize = 0;
+    z_stream strm;
+    int flush = Z_NO_FLUSH;
+    int decodingError = 0;
+    IOJob_t *writeJob = NULL;
+
+    strm.zalloc = Z_NULL;
+    strm.zfree = Z_NULL;
+    strm.opaque = Z_NULL;
+    strm.next_in = 0;
+    strm.avail_in = 0;
+    /* see https://www.zlib.net/manual.html */
+    if (inflateInit2(&strm, 15 /* maxWindowLogSize */ + 16 /* gzip only */) != Z_OK)
+        return FIO_ERROR_FRAME_DECODING;
+
+    writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
+    strm.next_out = (Bytef*)writeJob->buffer;
+    strm.avail_out = (uInt)writeJob->bufferSize;
+    strm.avail_in = (uInt)ress->readCtx->srcBufferLoaded;
+    strm.next_in = (z_const unsigned char*)ress->readCtx->srcBuffer;
+
+    for ( ; ; ) {
+        int ret;
+        if (strm.avail_in == 0) {
+            AIO_ReadPool_consumeAndRefill(ress->readCtx);
+            if (ress->readCtx->srcBufferLoaded == 0) flush = Z_FINISH;
+            strm.next_in = (z_const unsigned char*)ress->readCtx->srcBuffer;
+            strm.avail_in = (uInt)ress->readCtx->srcBufferLoaded;
+        }
+        ret = inflate(&strm, flush);
+        if (ret == Z_BUF_ERROR) {
+            DISPLAYLEVEL(1, "zstd: %s: premature gz end \n", srcFileName);
+            decodingError = 1; break;
+        }
+        if (ret != Z_OK && ret != Z_STREAM_END) {
+            DISPLAYLEVEL(1, "zstd: %s: inflate error %d \n", srcFileName, ret);
+            decodingError = 1; break;
+        }
+        {   size_t const decompBytes = writeJob->bufferSize - strm.avail_out;
+            if (decompBytes) {
+                writeJob->usedBufferSize = decompBytes;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+                outFileSize += decompBytes;
+                strm.next_out = (Bytef*)writeJob->buffer;
+                strm.avail_out = (uInt)writeJob->bufferSize;
+            }
+        }
+        if (ret == Z_STREAM_END) break;
+    }
+
+    AIO_ReadPool_consumeBytes(ress->readCtx, ress->readCtx->srcBufferLoaded - strm.avail_in);
+
+    if ( (inflateEnd(&strm) != Z_OK)  /* release resources ; error detected */
+      && (decodingError==0) ) {
+        DISPLAYLEVEL(1, "zstd: %s: inflateEnd error \n", srcFileName);
+        decodingError = 1;
+    }
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
+    return decodingError ? FIO_ERROR_FRAME_DECODING : outFileSize;
+}
+#endif
+
+#ifdef ZSTD_LZMADECOMPRESS
+static unsigned long long
+FIO_decompressLzmaFrame(dRess_t* ress,
+                        const char* srcFileName, int plain_lzma)
+{
+    unsigned long long outFileSize = 0;
+    lzma_stream strm = LZMA_STREAM_INIT;
+    lzma_action action = LZMA_RUN;
+    lzma_ret initRet;
+    int decodingError = 0;
+    IOJob_t *writeJob = NULL;
+
+    strm.next_in = 0;
+    strm.avail_in = 0;
+    if (plain_lzma) {
+        initRet = lzma_alone_decoder(&strm, UINT64_MAX); /* LZMA */
+    } else {
+        initRet = lzma_stream_decoder(&strm, UINT64_MAX, 0); /* XZ */
+    }
+
+    if (initRet != LZMA_OK) {
+        DISPLAYLEVEL(1, "zstd: %s: %s error %d \n",
+                        plain_lzma ? "lzma_alone_decoder" : "lzma_stream_decoder",
+                        srcFileName, initRet);
+        return FIO_ERROR_FRAME_DECODING;
+    }
+
+    writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
+    strm.next_out = (BYTE*)writeJob->buffer;
+    strm.avail_out = writeJob->bufferSize;
+    strm.next_in = (BYTE const*)ress->readCtx->srcBuffer;
+    strm.avail_in = ress->readCtx->srcBufferLoaded;
+
+    for ( ; ; ) {
+        lzma_ret ret;
+        if (strm.avail_in == 0) {
+            AIO_ReadPool_consumeAndRefill(ress->readCtx);
+            if (ress->readCtx->srcBufferLoaded == 0) action = LZMA_FINISH;
+            strm.next_in = (BYTE const*)ress->readCtx->srcBuffer;
+            strm.avail_in = ress->readCtx->srcBufferLoaded;
+        }
+        ret = lzma_code(&strm, action);
+
+        if (ret == LZMA_BUF_ERROR) {
+            DISPLAYLEVEL(1, "zstd: %s: premature lzma end \n", srcFileName);
+            decodingError = 1; break;
+        }
+        if (ret != LZMA_OK && ret != LZMA_STREAM_END) {
+            DISPLAYLEVEL(1, "zstd: %s: lzma_code decoding error %d \n",
+                            srcFileName, ret);
+            decodingError = 1; break;
+        }
+        {   size_t const decompBytes = writeJob->bufferSize - strm.avail_out;
+            if (decompBytes) {
+                writeJob->usedBufferSize = decompBytes;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+                outFileSize += decompBytes;
+                strm.next_out = (BYTE*)writeJob->buffer;
+                strm.avail_out = writeJob->bufferSize;
+        }   }
+        if (ret == LZMA_STREAM_END) break;
+    }
+
+    AIO_ReadPool_consumeBytes(ress->readCtx, ress->readCtx->srcBufferLoaded - strm.avail_in);
+    lzma_end(&strm);
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
+    return decodingError ? FIO_ERROR_FRAME_DECODING : outFileSize;
+}
+#endif
+
+#ifdef ZSTD_LZ4DECOMPRESS
+static unsigned long long
+FIO_decompressLz4Frame(dRess_t* ress, const char* srcFileName)
+{
+    unsigned long long filesize = 0;
+    LZ4F_errorCode_t nextToLoad = 4;
+    LZ4F_decompressionContext_t dCtx;
+    LZ4F_errorCode_t const errorCode = LZ4F_createDecompressionContext(&dCtx, LZ4F_VERSION);
+    int decodingError = 0;
+    IOJob_t *writeJob = NULL;
+
+    if (LZ4F_isError(errorCode)) {
+        DISPLAYLEVEL(1, "zstd: failed to create lz4 decompression context \n");
+        return FIO_ERROR_FRAME_DECODING;
+    }
+
+    writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
+
+    /* Main Loop */
+    for (;nextToLoad;) {
+        size_t pos = 0;
+        size_t decodedBytes = writeJob->bufferSize;
+        int fullBufferDecoded = 0;
+
+        /* Read input */
+        AIO_ReadPool_fillBuffer(ress->readCtx, nextToLoad);
+        if(!ress->readCtx->srcBufferLoaded) break; /* reached end of file */
+
+        while ((pos < ress->readCtx->srcBufferLoaded) || fullBufferDecoded) {  /* still to read, or still to flush */
+            /* Decode Input (at least partially) */
+            size_t remaining = ress->readCtx->srcBufferLoaded - pos;
+            decodedBytes = writeJob->bufferSize;
+            nextToLoad = LZ4F_decompress(dCtx, writeJob->buffer, &decodedBytes, (char*)(ress->readCtx->srcBuffer)+pos,
+                                         &remaining, NULL);
+            if (LZ4F_isError(nextToLoad)) {
+                DISPLAYLEVEL(1, "zstd: %s: lz4 decompression error : %s \n",
+                                srcFileName, LZ4F_getErrorName(nextToLoad));
+                decodingError = 1; nextToLoad = 0; break;
+            }
+            pos += remaining;
+            assert(pos <= ress->readCtx->srcBufferLoaded);
+            fullBufferDecoded = decodedBytes == writeJob->bufferSize;
+
+            /* Write Block */
+            if (decodedBytes) {
+                UTIL_HumanReadableSize_t hrs;
+                writeJob->usedBufferSize = decodedBytes;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+                filesize += decodedBytes;
+                hrs = UTIL_makeHumanReadableSize(filesize);
+                DISPLAYUPDATE_PROGRESS("\rDecompressed : %.*f%s  ", hrs.precision, hrs.value, hrs.suffix);
+            }
+
+            if (!nextToLoad) break;
+        }
+        AIO_ReadPool_consumeBytes(ress->readCtx, pos);
+    }
+    if (nextToLoad!=0) {
+        DISPLAYLEVEL(1, "zstd: %s: unfinished lz4 stream \n", srcFileName);
+        decodingError=1;
+    }
+
+    LZ4F_freeDecompressionContext(dCtx);
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
+
+    return decodingError ? FIO_ERROR_FRAME_DECODING : filesize;
+}
+#endif
+
+
+
+/** FIO_decompressFrames() :
+ *  Find and decode frames inside srcFile
+ *  srcFile presumed opened and valid
+ * @return : 0 : OK
+ *           1 : error
+ */
+static int FIO_decompressFrames(FIO_ctx_t* const fCtx,
+                                dRess_t ress, const FIO_prefs_t* const prefs,
+                                const char* dstFileName, const char* srcFileName)
+{
+    unsigned readSomething = 0;
+    unsigned long long filesize = 0;
+    int passThrough = prefs->passThrough;
+
+    if (passThrough == -1) {
+        /* If pass-through mode is not explicitly enabled or disabled,
+         * default to the legacy behavior of enabling it if we are writing
+         * to stdout with the overwrite flag enabled.
+         */
+        passThrough = prefs->overwrite && !strcmp(dstFileName, stdoutmark);
+    }
+    assert(passThrough == 0 || passThrough == 1);
+
+    /* for each frame */
+    for ( ; ; ) {
+        /* check magic number -> version */
+        size_t const toRead = 4;
+        const BYTE* buf;
+        AIO_ReadPool_fillBuffer(ress.readCtx, toRead);
+        buf = (const BYTE*)ress.readCtx->srcBuffer;
+        if (ress.readCtx->srcBufferLoaded==0) {
+            if (readSomething==0) {  /* srcFile is empty (which is invalid) */
+                DISPLAYLEVEL(1, "zstd: %s: unexpected end of file \n", srcFileName);
+                return 1;
+            }  /* else, just reached frame boundary */
+            break;   /* no more input */
+        }
+        readSomething = 1;   /* there is at least 1 byte in srcFile */
+        if (ress.readCtx->srcBufferLoaded < toRead) { /* not enough input to check magic number */
+            if (passThrough) {
+                return FIO_passThrough(&ress);
+            }
+            DISPLAYLEVEL(1, "zstd: %s: unknown header \n", srcFileName);
+            return 1;
+        }
+        if (ZSTD_isFrame(buf, ress.readCtx->srcBufferLoaded)) {
+            unsigned long long const frameSize = FIO_decompressZstdFrame(fCtx, &ress, prefs, srcFileName, filesize);
+            if (frameSize == FIO_ERROR_FRAME_DECODING) return 1;
+            filesize += frameSize;
+        } else if (buf[0] == 31 && buf[1] == 139) { /* gz magic number */
+#ifdef ZSTD_GZDECOMPRESS
+            unsigned long long const frameSize = FIO_decompressGzFrame(&ress, srcFileName);
+            if (frameSize == FIO_ERROR_FRAME_DECODING) return 1;
+            filesize += frameSize;
+#else
+            DISPLAYLEVEL(1, "zstd: %s: gzip file cannot be uncompressed (zstd compiled without HAVE_ZLIB) -- ignored \n", srcFileName);
+            return 1;
+#endif
+        } else if ((buf[0] == 0xFD && buf[1] == 0x37)  /* xz magic number */
+                || (buf[0] == 0x5D && buf[1] == 0x00)) { /* lzma header (no magic number) */
+#ifdef ZSTD_LZMADECOMPRESS
+            unsigned long long const frameSize = FIO_decompressLzmaFrame(&ress, srcFileName, buf[0] != 0xFD);
+            if (frameSize == FIO_ERROR_FRAME_DECODING) return 1;
+            filesize += frameSize;
+#else
+            DISPLAYLEVEL(1, "zstd: %s: xz/lzma file cannot be uncompressed (zstd compiled without HAVE_LZMA) -- ignored \n", srcFileName);
+            return 1;
+#endif
+        } else if (MEM_readLE32(buf) == LZ4_MAGICNUMBER) {
+#ifdef ZSTD_LZ4DECOMPRESS
+            unsigned long long const frameSize = FIO_decompressLz4Frame(&ress, srcFileName);
+            if (frameSize == FIO_ERROR_FRAME_DECODING) return 1;
+            filesize += frameSize;
+#else
+            DISPLAYLEVEL(1, "zstd: %s: lz4 file cannot be uncompressed (zstd compiled without HAVE_LZ4) -- ignored \n", srcFileName);
+            return 1;
+#endif
+        } else if (passThrough) {
+            return FIO_passThrough(&ress);
+        } else {
+            DISPLAYLEVEL(1, "zstd: %s: unsupported format \n", srcFileName);
+            return 1;
+    }   }  /* for each frame */
+
+    /* Final Status */
+    fCtx->totalBytesOutput += (size_t)filesize;
+    DISPLAY_PROGRESS("\r%79s\r", "");
+    if (FIO_shouldDisplayFileSummary(fCtx))
+        DISPLAY_SUMMARY("%-20s: %llu bytes \n", srcFileName, filesize);
+
+    return 0;
+}
+
+/** FIO_decompressDstFile() :
+    open `dstFileName`, or pass-through if writeCtx's file is already != 0,
+    then start decompression process (FIO_decompressFrames()).
+    @return : 0 : OK
+              1 : operation aborted
+*/
+static int FIO_decompressDstFile(FIO_ctx_t* const fCtx,
+                                 FIO_prefs_t* const prefs,
+                                 dRess_t ress,
+                                 const char* dstFileName,
+                                 const char* srcFileName,
+                                 const stat_t* srcFileStat)
+{
+    int result;
+    int releaseDstFile = 0;
+    int transferStat = 0;
+    int dstFd = 0;
+
+    if ((AIO_WritePool_getFile(ress.writeCtx) == NULL) && (prefs->testMode == 0)) {
+        FILE *dstFile;
+        int dstFilePermissions = DEFAULT_FILE_PERMISSIONS;
+        if ( strcmp(srcFileName, stdinmark)   /* special case : don't transfer permissions from stdin */
+          && strcmp(dstFileName, stdoutmark)
+          && UTIL_isRegularFileStat(srcFileStat) ) {
+            transferStat = 1;
+            dstFilePermissions = TEMPORARY_FILE_PERMISSIONS;
+        }
+
+        releaseDstFile = 1;
+
+        dstFile = FIO_openDstFile(fCtx, prefs, srcFileName, dstFileName, dstFilePermissions);
+        if (dstFile==NULL) return 1;
+        dstFd = fileno(dstFile);
+        AIO_WritePool_setFile(ress.writeCtx, dstFile);
+
+        /* Must only be added after FIO_openDstFile() succeeds.
+         * Otherwise we may delete the destination file if it already exists,
+         * and the user presses Ctrl-C when asked if they wish to overwrite.
+         */
+        addHandler(dstFileName);
+    }
+
+    result = FIO_decompressFrames(fCtx, ress, prefs, dstFileName, srcFileName);
+
+    if (releaseDstFile) {
+        clearHandler();
+
+        if (transferStat) {
+            UTIL_setFDStat(dstFd, dstFileName, srcFileStat);
+        }
+
+        if (AIO_WritePool_closeFile(ress.writeCtx)) {
+            DISPLAYLEVEL(1, "zstd: %s: %s \n", dstFileName, strerror(errno));
+            result = 1;
+        }
+
+        if (transferStat) {
+            UTIL_utime(dstFileName, srcFileStat);
+        }
+
+        if ( (result != 0)  /* operation failure */
+          && strcmp(dstFileName, stdoutmark)  /* special case : don't remove() stdout */
+          ) {
+            FIO_removeFile(dstFileName);  /* remove decompression artefact; note: don't do anything special if remove() fails */
+        }
+    }
+
+    return result;
+}
+
+
+/** FIO_decompressSrcFile() :
+    Open `srcFileName`, transfer control to decompressDstFile()
+    @return : 0 : OK
+              1 : error
+*/
+static int FIO_decompressSrcFile(FIO_ctx_t* const fCtx, FIO_prefs_t* const prefs, dRess_t ress, const char* dstFileName, const char* srcFileName)
+{
+    FILE* srcFile;
+    stat_t srcFileStat;
+    int result;
+    U64 fileSize = UTIL_FILESIZE_UNKNOWN;
+
+    if (UTIL_isDirectory(srcFileName)) {
+        DISPLAYLEVEL(1, "zstd: %s is a directory -- ignored \n", srcFileName);
+        return 1;
+    }
+
+    srcFile = FIO_openSrcFile(prefs, srcFileName, &srcFileStat);
+    if (srcFile==NULL) return 1;
+
+    /* Don't use AsyncIO for small files */
+    if (strcmp(srcFileName, stdinmark)) /* Stdin doesn't have stats */
+        fileSize = UTIL_getFileSizeStat(&srcFileStat);
+    if(fileSize != UTIL_FILESIZE_UNKNOWN && fileSize < ZSTD_BLOCKSIZE_MAX * 3) {
+        AIO_ReadPool_setAsync(ress.readCtx, 0);
+        AIO_WritePool_setAsync(ress.writeCtx, 0);
+    } else {
+        AIO_ReadPool_setAsync(ress.readCtx, 1);
+        AIO_WritePool_setAsync(ress.writeCtx, 1);
+    }
+
+    AIO_ReadPool_setFile(ress.readCtx, srcFile);
+
+    result = FIO_decompressDstFile(fCtx, prefs, ress, dstFileName, srcFileName, &srcFileStat);
+
+    AIO_ReadPool_setFile(ress.readCtx, NULL);
+
+    /* Close file */
+    if (fclose(srcFile)) {
+        DISPLAYLEVEL(1, "zstd: %s: %s \n", srcFileName, strerror(errno));  /* error should not happen */
+        return 1;
+    }
+    if ( prefs->removeSrcFile  /* --rm */
+      && (result==0)      /* decompression successful */
+      && strcmp(srcFileName, stdinmark) ) /* not stdin */ {
+        /* We must clear the handler, since after this point calling it would
+         * delete both the source and destination files.
+         */
+        clearHandler();
+        if (FIO_removeFile(srcFileName)) {
+            /* failed to remove src file */
+            DISPLAYLEVEL(1, "zstd: %s: %s \n", srcFileName, strerror(errno));
+            return 1;
+    }   }
+    return result;
+}
+
+
+
+int FIO_decompressFilename(FIO_ctx_t* const fCtx, FIO_prefs_t* const prefs,
+                           const char* dstFileName, const char* srcFileName,
+                           const char* dictFileName)
+{
+    dRess_t const ress = FIO_createDResources(prefs, dictFileName);
+
+    int const decodingError = FIO_decompressSrcFile(fCtx, prefs, ress, dstFileName, srcFileName);
+
+
+
+    FIO_freeDResources(ress);
+    return decodingError;
+}
+
+static const char *suffixList[] = {
+    ZSTD_EXTENSION,
+    TZSTD_EXTENSION,
+#ifndef ZSTD_NODECOMPRESS
+    ZSTD_ALT_EXTENSION,
+#endif
+#ifdef ZSTD_GZDECOMPRESS
+    GZ_EXTENSION,
+    TGZ_EXTENSION,
+#endif
+#ifdef ZSTD_LZMADECOMPRESS
+    LZMA_EXTENSION,
+    XZ_EXTENSION,
+    TXZ_EXTENSION,
+#endif
+#ifdef ZSTD_LZ4DECOMPRESS
+    LZ4_EXTENSION,
+    TLZ4_EXTENSION,
+#endif
+    NULL
+};
+
+static const char *suffixListStr =
+    ZSTD_EXTENSION "/" TZSTD_EXTENSION
+#ifdef ZSTD_GZDECOMPRESS
+    "/" GZ_EXTENSION "/" TGZ_EXTENSION
+#endif
+#ifdef ZSTD_LZMADECOMPRESS
+    "/" LZMA_EXTENSION "/" XZ_EXTENSION "/" TXZ_EXTENSION
+#endif
+#ifdef ZSTD_LZ4DECOMPRESS
+    "/" LZ4_EXTENSION "/" TLZ4_EXTENSION
+#endif
+;
+
+/* FIO_determineDstName() :
+ * create a destination filename from a srcFileName.
+ * @return a pointer to it.
+ * @return == NULL if there is an error */
+static const char*
+FIO_determineDstName(const char* srcFileName, const char* outDirName)
+{
+    static size_t dfnbCapacity = 0;
+    static char* dstFileNameBuffer = NULL;   /* using static allocation : this function cannot be multi-threaded */
+    size_t dstFileNameEndPos;
+    char* outDirFilename = NULL;
+    const char* dstSuffix = "";
+    size_t dstSuffixLen = 0;
+
+    size_t sfnSize = strlen(srcFileName);
+
+    size_t srcSuffixLen;
+    const char* const srcSuffix = strrchr(srcFileName, '.');
+
+    if(!strcmp(srcFileName, stdinmark)) {
+        return stdoutmark;
+    }
+
+    if (srcSuffix == NULL) {
+        DISPLAYLEVEL(1,
+            "zstd: %s: unknown suffix (%s expected). "
+            "Can't derive the output file name. "
+            "Specify it with -o dstFileName. Ignoring.\n",
+            srcFileName, suffixListStr);
+        return NULL;
+    }
+    srcSuffixLen = strlen(srcSuffix);
+
+    {
+        const char** matchedSuffixPtr;
+        for (matchedSuffixPtr = suffixList; *matchedSuffixPtr != NULL; matchedSuffixPtr++) {
+            if (!strcmp(*matchedSuffixPtr, srcSuffix)) {
+                break;
+            }
+        }
+
+        /* check suffix is authorized */
+        if (sfnSize <= srcSuffixLen || *matchedSuffixPtr == NULL) {
+            DISPLAYLEVEL(1,
+                "zstd: %s: unknown suffix (%s expected). "
+                "Can't derive the output file name. "
+                "Specify it with -o dstFileName. Ignoring.\n",
+                srcFileName, suffixListStr);
+            return NULL;
+        }
+
+        if ((*matchedSuffixPtr)[1] == 't') {
+            dstSuffix = ".tar";
+            dstSuffixLen = strlen(dstSuffix);
+        }
+    }
+
+    if (outDirName) {
+        outDirFilename = FIO_createFilename_fromOutDir(srcFileName, outDirName, 0);
+        sfnSize = strlen(outDirFilename);
+        assert(outDirFilename != NULL);
+    }
+
+    if (dfnbCapacity+srcSuffixLen <= sfnSize+1+dstSuffixLen) {
+        /* allocate enough space to write dstFilename into it */
+        free(dstFileNameBuffer);
+        dfnbCapacity = sfnSize + 20;
+        dstFileNameBuffer = (char*)malloc(dfnbCapacity);
+        if (dstFileNameBuffer==NULL)
+            EXM_THROW(74, "%s : not enough memory for dstFileName",
+                      strerror(errno));
+    }
+
+    /* return dst name == src name truncated from suffix */
+    assert(dstFileNameBuffer != NULL);
+    dstFileNameEndPos = sfnSize - srcSuffixLen;
+    if (outDirFilename) {
+        memcpy(dstFileNameBuffer, outDirFilename, dstFileNameEndPos);
+        free(outDirFilename);
+    } else {
+        memcpy(dstFileNameBuffer, srcFileName, dstFileNameEndPos);
+    }
+
+    /* The short tar extensions tzst, tgz, txz and tlz4 files should have "tar"
+     * extension on decompression. Also writes terminating null. */
+    strcpy(dstFileNameBuffer + dstFileNameEndPos, dstSuffix);
+    return dstFileNameBuffer;
+
+    /* note : dstFileNameBuffer memory is not going to be free */
+}
+
+int
+FIO_decompressMultipleFilenames(FIO_ctx_t* const fCtx,
+                                FIO_prefs_t* const prefs,
+                                const char** srcNamesTable,
+                                const char* outMirroredRootDirName,
+                                const char* outDirName, const char* outFileName,
+                                const char* dictFileName)
+{
+    int status;
+    int error = 0;
+    dRess_t ress = FIO_createDResources(prefs, dictFileName);
+
+    if (outFileName) {
+        if (FIO_multiFilesConcatWarning(fCtx, prefs, outFileName, 1 /* displayLevelCutoff */)) {
+            FIO_freeDResources(ress);
+            return 1;
+        }
+        if (!prefs->testMode) {
+            FILE* dstFile = FIO_openDstFile(fCtx, prefs, NULL, outFileName, DEFAULT_FILE_PERMISSIONS);
+            if (dstFile == 0) EXM_THROW(19, "cannot open %s", outFileName);
+            AIO_WritePool_setFile(ress.writeCtx, dstFile);
+        }
+        for (; fCtx->currFileIdx < fCtx->nbFilesTotal; fCtx->currFileIdx++) {
+            status = FIO_decompressSrcFile(fCtx, prefs, ress, outFileName, srcNamesTable[fCtx->currFileIdx]);
+            if (!status) fCtx->nbFilesProcessed++;
+            error |= status;
+        }
+        if ((!prefs->testMode) && (AIO_WritePool_closeFile(ress.writeCtx)))
+            EXM_THROW(72, "Write error : %s : cannot properly close output file",
+                        strerror(errno));
+    } else {
+        if (outMirroredRootDirName)
+            UTIL_mirrorSourceFilesDirectories(srcNamesTable, (unsigned)fCtx->nbFilesTotal, outMirroredRootDirName);
+
+        for (; fCtx->currFileIdx < fCtx->nbFilesTotal; fCtx->currFileIdx++) {   /* create dstFileName */
+            const char* const srcFileName = srcNamesTable[fCtx->currFileIdx];
+            const char* dstFileName = NULL;
+            if (outMirroredRootDirName) {
+                char* validMirroredDirName = UTIL_createMirroredDestDirName(srcFileName, outMirroredRootDirName);
+                if (validMirroredDirName) {
+                    dstFileName = FIO_determineDstName(srcFileName, validMirroredDirName);
+                    free(validMirroredDirName);
+                } else {
+                    DISPLAYLEVEL(2, "zstd: --output-dir-mirror cannot decompress '%s' into '%s'\n", srcFileName, outMirroredRootDirName);
+                }
+            } else {
+                dstFileName = FIO_determineDstName(srcFileName, outDirName);
+            }
+            if (dstFileName == NULL) { error=1; continue; }
+            status = FIO_decompressSrcFile(fCtx, prefs, ress, dstFileName, srcFileName);
+            if (!status) fCtx->nbFilesProcessed++;
+            error |= status;
+        }
+        if (outDirName)
+            FIO_checkFilenameCollisions(srcNamesTable , (unsigned)fCtx->nbFilesTotal);
+    }
+
+    if (FIO_shouldDisplayMultipleFileSummary(fCtx)) {
+        DISPLAY_PROGRESS("\r%79s\r", "");
+        DISPLAY_SUMMARY("%d files decompressed : %6llu bytes total \n",
+            fCtx->nbFilesProcessed, (unsigned long long)fCtx->totalBytesOutput);
+    }
+
+    FIO_freeDResources(ress);
+    return error;
+}
+
+/* **************************************************************************
+ *  .zst file info (--list command)
+ ***************************************************************************/
+
+typedef struct {
+    U64 decompressedSize;
+    U64 compressedSize;
+    U64 windowSize;
+    int numActualFrames;
+    int numSkippableFrames;
+    int decompUnavailable;
+    int usesCheck;
+    BYTE checksum[4];
+    U32 nbFiles;
+    unsigned dictID;
+} fileInfo_t;
+
+typedef enum {
+  info_success=0,
+  info_frame_error=1,
+  info_not_zstd=2,
+  info_file_error=3,
+  info_truncated_input=4
+} InfoError;
+
+#define ERROR_IF(c,n,...) {             \
+    if (c) {                           \
+        DISPLAYLEVEL(1, __VA_ARGS__);  \
+        DISPLAYLEVEL(1, " \n");        \
+        return n;                      \
+    }                                  \
+}
+
+static InfoError
+FIO_analyzeFrames(fileInfo_t* info, FILE* const srcFile)
+{
+    /* begin analyzing frame */
+    for ( ; ; ) {
+        BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
+        size_t const numBytesRead = fread(headerBuffer, 1, sizeof(headerBuffer), srcFile);
+        if (numBytesRead < ZSTD_FRAMEHEADERSIZE_MIN(ZSTD_f_zstd1)) {
+            if ( feof(srcFile)
+              && (numBytesRead == 0)
+              && (info->compressedSize > 0)
+              && (info->compressedSize != UTIL_FILESIZE_UNKNOWN) ) {
+                unsigned long long file_position = (unsigned long long) LONG_TELL(srcFile);
+                unsigned long long file_size = (unsigned long long) info->compressedSize;
+                ERROR_IF(file_position != file_size, info_truncated_input,
+                  "Error: seeked to position %llu, which is beyond file size of %llu\n",
+                  file_position,
+                  file_size);
+                break;  /* correct end of file => success */
+            }
+            ERROR_IF(feof(srcFile), info_not_zstd, "Error: reached end of file with incomplete frame");
+            ERROR_IF(1, info_frame_error, "Error: did not reach end of file but ran out of frames");
+        }
+        {   U32 const magicNumber = MEM_readLE32(headerBuffer);
+            /* Zstandard frame */
+            if (magicNumber == ZSTD_MAGICNUMBER) {
+                ZSTD_FrameHeader header;
+                U64 const frameContentSize = ZSTD_getFrameContentSize(headerBuffer, numBytesRead);
+                if ( frameContentSize == ZSTD_CONTENTSIZE_ERROR
+                  || frameContentSize == ZSTD_CONTENTSIZE_UNKNOWN ) {
+                    info->decompUnavailable = 1;
+                } else {
+                    info->decompressedSize += frameContentSize;
+                }
+                ERROR_IF(ZSTD_getFrameHeader(&header, headerBuffer, numBytesRead) != 0,
+                        info_frame_error, "Error: could not decode frame header");
+                if (info->dictID != 0 && info->dictID != header.dictID) {
+                    DISPLAY("WARNING: File contains multiple frames with different dictionary IDs. Showing dictID 0 instead");
+                    info->dictID = 0;
+                } else {
+                    info->dictID = header.dictID;
+                }
+                info->windowSize = header.windowSize;
+                /* move to the end of the frame header */
+                {   size_t const headerSize = ZSTD_frameHeaderSize(headerBuffer, numBytesRead);
+                    ERROR_IF(ZSTD_isError(headerSize), info_frame_error, "Error: could not determine frame header size");
+                    ERROR_IF(fseek(srcFile, ((long)headerSize)-((long)numBytesRead), SEEK_CUR) != 0,
+                            info_frame_error, "Error: could not move to end of frame header");
+                }
+
+                /* skip all blocks in the frame */
+                {   int lastBlock = 0;
+                    do {
+                        BYTE blockHeaderBuffer[3];
+                        ERROR_IF(fread(blockHeaderBuffer, 1, 3, srcFile) != 3,
+                                info_frame_error, "Error while reading block header");
+                        {   U32 const blockHeader = MEM_readLE24(blockHeaderBuffer);
+                            U32 const blockTypeID = (blockHeader >> 1) & 3;
+                            U32 const isRLE = (blockTypeID == 1);
+                            U32 const isWrongBlock = (blockTypeID == 3);
+                            long const blockSize = isRLE ? 1 : (long)(blockHeader >> 3);
+                            ERROR_IF(isWrongBlock, info_frame_error, "Error: unsupported block type");
+                            lastBlock = blockHeader & 1;
+                            ERROR_IF(fseek(srcFile, blockSize, SEEK_CUR) != 0,
+                                    info_frame_error, "Error: could not skip to end of block");
+                        }
+                    } while (lastBlock != 1);
+                }
+
+                /* check if checksum is used */
+                {   BYTE const frameHeaderDescriptor = headerBuffer[4];
+                    int const contentChecksumFlag = (frameHeaderDescriptor & (1 << 2)) >> 2;
+                    if (contentChecksumFlag) {
+                        info->usesCheck = 1;
+                        ERROR_IF(fread(info->checksum, 1, 4, srcFile) != 4,
+                                info_frame_error, "Error: could not read checksum");
+                }   }
+                info->numActualFrames++;
+            }
+            /* Skippable frame */
+            else if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+                U32 const frameSize = MEM_readLE32(headerBuffer + 4);
+                long const seek = (long)(8 + frameSize - numBytesRead);
+                ERROR_IF(LONG_SEEK(srcFile, seek, SEEK_CUR) != 0,
+                        info_frame_error, "Error: could not find end of skippable frame");
+                info->numSkippableFrames++;
+            }
+            /* unknown content */
+            else {
+                return info_not_zstd;
+            }
+        }  /* magic number analysis */
+    }  /* end analyzing frames */
+    return info_success;
+}
+
+
+static InfoError
+getFileInfo_fileConfirmed(fileInfo_t* info, const char* inFileName)
+{
+    InfoError status;
+    stat_t srcFileStat;
+    FILE* const srcFile = FIO_openSrcFile(NULL, inFileName, &srcFileStat);
+    ERROR_IF(srcFile == NULL, info_file_error, "Error: could not open source file %s", inFileName);
+
+    info->compressedSize = UTIL_getFileSizeStat(&srcFileStat);
+    status = FIO_analyzeFrames(info, srcFile);
+
+    fclose(srcFile);
+    info->nbFiles = 1;
+    return status;
+}
+
+
+/** getFileInfo() :
+ *  Reads information from file, stores in *info
+ * @return : InfoError status
+ */
+static InfoError
+getFileInfo(fileInfo_t* info, const char* srcFileName)
+{
+    ERROR_IF(!UTIL_isRegularFile(srcFileName),
+            info_file_error, "Error : %s is not a file", srcFileName);
+    return getFileInfo_fileConfirmed(info, srcFileName);
+}
+
+
+static void
+displayInfo(const char* inFileName, const fileInfo_t* info, int displayLevel)
+{
+    UTIL_HumanReadableSize_t const window_hrs = UTIL_makeHumanReadableSize(info->windowSize);
+    UTIL_HumanReadableSize_t const compressed_hrs = UTIL_makeHumanReadableSize(info->compressedSize);
+    UTIL_HumanReadableSize_t const decompressed_hrs = UTIL_makeHumanReadableSize(info->decompressedSize);
+    double const ratio = (info->compressedSize == 0) ? 0 : ((double)info->decompressedSize)/(double)info->compressedSize;
+    const char* const checkString = (info->usesCheck ? "XXH64" : "None");
+    if (displayLevel <= 2) {
+        if (!info->decompUnavailable) {
+            DISPLAYOUT("%6d  %5d  %6.*f%4s  %8.*f%4s  %5.3f  %5s  %s\n",
+                    info->numSkippableFrames + info->numActualFrames,
+                    info->numSkippableFrames,
+                    compressed_hrs.precision, compressed_hrs.value, compressed_hrs.suffix,
+                    decompressed_hrs.precision, decompressed_hrs.value, decompressed_hrs.suffix,
+                    ratio, checkString, inFileName);
+        } else {
+            DISPLAYOUT("%6d  %5d  %6.*f%4s                       %5s  %s\n",
+                    info->numSkippableFrames + info->numActualFrames,
+                    info->numSkippableFrames,
+                    compressed_hrs.precision, compressed_hrs.value, compressed_hrs.suffix,
+                    checkString, inFileName);
+        }
+    } else {
+        DISPLAYOUT("%s \n", inFileName);
+        DISPLAYOUT("# Zstandard Frames: %d\n", info->numActualFrames);
+        if (info->numSkippableFrames)
+            DISPLAYOUT("# Skippable Frames: %d\n", info->numSkippableFrames);
+        DISPLAYOUT("DictID: %u\n", info->dictID);
+        DISPLAYOUT("Window Size: %.*f%s (%llu B)\n",
+                   window_hrs.precision, window_hrs.value, window_hrs.suffix,
+                   (unsigned long long)info->windowSize);
+        DISPLAYOUT("Compressed Size: %.*f%s (%llu B)\n",
+                    compressed_hrs.precision, compressed_hrs.value, compressed_hrs.suffix,
+                    (unsigned long long)info->compressedSize);
+        if (!info->decompUnavailable) {
+            DISPLAYOUT("Decompressed Size: %.*f%s (%llu B)\n",
+                    decompressed_hrs.precision, decompressed_hrs.value, decompressed_hrs.suffix,
+                    (unsigned long long)info->decompressedSize);
+            DISPLAYOUT("Ratio: %.4f\n", ratio);
+        }
+
+        if (info->usesCheck && info->numActualFrames == 1) {
+            DISPLAYOUT("Check: %s %02x%02x%02x%02x\n", checkString,
+                info->checksum[3], info->checksum[2],
+                info->checksum[1], info->checksum[0]
+            );
+        } else {
+            DISPLAYOUT("Check: %s\n", checkString);
+        }
+
+        DISPLAYOUT("\n");
+    }
+}
+
+static fileInfo_t FIO_addFInfo(fileInfo_t fi1, fileInfo_t fi2)
+{
+    fileInfo_t total;
+    memset(&total, 0, sizeof(total));
+    total.numActualFrames = fi1.numActualFrames + fi2.numActualFrames;
+    total.numSkippableFrames = fi1.numSkippableFrames + fi2.numSkippableFrames;
+    total.compressedSize = fi1.compressedSize + fi2.compressedSize;
+    total.decompressedSize = fi1.decompressedSize + fi2.decompressedSize;
+    total.decompUnavailable = fi1.decompUnavailable | fi2.decompUnavailable;
+    total.usesCheck = fi1.usesCheck & fi2.usesCheck;
+    total.nbFiles = fi1.nbFiles + fi2.nbFiles;
+    return total;
+}
+
+static int
+FIO_listFile(fileInfo_t* total, const char* inFileName, int displayLevel)
+{
+    fileInfo_t info;
+    memset(&info, 0, sizeof(info));
+    {   InfoError const error = getFileInfo(&info, inFileName);
+        switch (error) {
+            case info_frame_error:
+                /* display error, but provide output */
+                DISPLAYLEVEL(1, "Error while parsing \"%s\" \n", inFileName);
+                break;
+            case info_not_zstd:
+                DISPLAYOUT("File \"%s\" not compressed by zstd \n", inFileName);
+                if (displayLevel > 2) DISPLAYOUT("\n");
+                return 1;
+            case info_file_error:
+                /* error occurred while opening the file */
+                if (displayLevel > 2) DISPLAYOUT("\n");
+                return 1;
+            case info_truncated_input:
+                DISPLAYOUT("File \"%s\" is truncated \n", inFileName);
+                if (displayLevel > 2) DISPLAYOUT("\n");
+                return 1;
+            case info_success:
+            default:
+                break;
+        }
+
+        displayInfo(inFileName, &info, displayLevel);
+        *total = FIO_addFInfo(*total, info);
+        assert(error == info_success || error == info_frame_error);
+        return (int)error;
+    }
+}
+
+int FIO_listMultipleFiles(unsigned numFiles, const char** filenameTable, int displayLevel)
+{
+    /* ensure no specified input is stdin (needs fseek() capability) */
+    {   unsigned u;
+        for (u=0; u<numFiles;u++) {
+            ERROR_IF(!strcmp (filenameTable[u], stdinmark),
+                    1, "zstd: --list does not support reading from standard input");
+    }   }
+
+    if (numFiles == 0) {
+        if (!UTIL_isConsole(stdin)) {
+            DISPLAYLEVEL(1, "zstd: --list does not support reading from standard input \n");
+        }
+        DISPLAYLEVEL(1, "No files given \n");
+        return 1;
+    }
+
+    if (displayLevel <= 2) {
+        DISPLAYOUT("Frames  Skips  Compressed  Uncompressed  Ratio  Check  Filename\n");
+    }
+    {   int error = 0;
+        fileInfo_t total;
+        memset(&total, 0, sizeof(total));
+        total.usesCheck = 1;
+        /* --list each file, and check for any error */
+        {   unsigned u;
+            for (u=0; u<numFiles;u++) {
+                error |= FIO_listFile(&total, filenameTable[u], displayLevel);
+        }   }
+        if (numFiles > 1 && displayLevel <= 2) {   /* display total */
+            UTIL_HumanReadableSize_t const compressed_hrs = UTIL_makeHumanReadableSize(total.compressedSize);
+            UTIL_HumanReadableSize_t const decompressed_hrs = UTIL_makeHumanReadableSize(total.decompressedSize);
+            double const ratio = (total.compressedSize == 0) ? 0 : ((double)total.decompressedSize)/(double)total.compressedSize;
+            const char* const checkString = (total.usesCheck ? "XXH64" : "");
+            DISPLAYOUT("----------------------------------------------------------------- \n");
+            if (total.decompUnavailable) {
+                DISPLAYOUT("%6d  %5d  %6.*f%4s                       %5s  %u files\n",
+                        total.numSkippableFrames + total.numActualFrames,
+                        total.numSkippableFrames,
+                        compressed_hrs.precision, compressed_hrs.value, compressed_hrs.suffix,
+                        checkString, (unsigned)total.nbFiles);
+            } else {
+                DISPLAYOUT("%6d  %5d  %6.*f%4s  %8.*f%4s  %5.3f  %5s  %u files\n",
+                        total.numSkippableFrames + total.numActualFrames,
+                        total.numSkippableFrames,
+                        compressed_hrs.precision, compressed_hrs.value, compressed_hrs.suffix,
+                        decompressed_hrs.precision, decompressed_hrs.value, decompressed_hrs.suffix,
+                        ratio, checkString, (unsigned)total.nbFiles);
+        }   }
+        return error;
+    }
+}
+
+
+#endif /* #ifndef ZSTD_NODECOMPRESS */
diff --git a/programs/fileio_asyncio.c b/programs/fileio_asyncio.c
new file mode 100644
index 0000000..42a4720
--- /dev/null
+++ b/programs/fileio_asyncio.c
@@ -0,0 +1,663 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "platform.h"
+#include <stdio.h>      /* fprintf, open, fdopen, fread, _fileno, stdin, stdout */
+#include <stdlib.h>     /* malloc, free */
+#include <assert.h>
+#include <errno.h>      /* errno */
+
+#if defined (_MSC_VER)
+#  include <sys/stat.h>
+#  include <io.h>
+#endif
+
+#include "fileio_asyncio.h"
+#include "fileio_common.h"
+
+/* **********************************************************************
+ *  Sparse write
+ ************************************************************************/
+
+/** AIO_fwriteSparse() :
+*  @return : storedSkips,
+*            argument for next call to AIO_fwriteSparse() or AIO_fwriteSparseEnd() */
+static unsigned
+AIO_fwriteSparse(FILE* file,
+                 const void* buffer, size_t bufferSize,
+                 const FIO_prefs_t* const prefs,
+                 unsigned storedSkips)
+{
+    const size_t* const bufferT = (const size_t*)buffer;   /* Buffer is supposed malloc'ed, hence aligned on size_t */
+    size_t bufferSizeT = bufferSize / sizeof(size_t);
+    const size_t* const bufferTEnd = bufferT + bufferSizeT;
+    const size_t* ptrT = bufferT;
+    static const size_t segmentSizeT = (32 KB) / sizeof(size_t);   /* check every 32 KB */
+
+    if (prefs->testMode) return 0;  /* do not output anything in test mode */
+
+    if (!prefs->sparseFileSupport) {  /* normal write */
+        size_t const sizeCheck = fwrite(buffer, 1, bufferSize, file);
+        if (sizeCheck != bufferSize)
+            EXM_THROW(70, "Write error : cannot write block : %s",
+                      strerror(errno));
+        return 0;
+    }
+
+    /* avoid int overflow */
+    if (storedSkips > 1 GB) {
+        if (LONG_SEEK(file, 1 GB, SEEK_CUR) != 0)
+        EXM_THROW(91, "1 GB skip error (sparse file support)");
+        storedSkips -= 1 GB;
+    }
+
+    while (ptrT < bufferTEnd) {
+        size_t nb0T;
+
+        /* adjust last segment if < 32 KB */
+        size_t seg0SizeT = segmentSizeT;
+        if (seg0SizeT > bufferSizeT) seg0SizeT = bufferSizeT;
+        bufferSizeT -= seg0SizeT;
+
+        /* count leading zeroes */
+        for (nb0T=0; (nb0T < seg0SizeT) && (ptrT[nb0T] == 0); nb0T++) ;
+        storedSkips += (unsigned)(nb0T * sizeof(size_t));
+
+        if (nb0T != seg0SizeT) {   /* not all 0s */
+            size_t const nbNon0ST = seg0SizeT - nb0T;
+            /* skip leading zeros */
+            if (LONG_SEEK(file, storedSkips, SEEK_CUR) != 0)
+                EXM_THROW(92, "Sparse skip error ; try --no-sparse");
+            storedSkips = 0;
+            /* write the rest */
+            if (fwrite(ptrT + nb0T, sizeof(size_t), nbNon0ST, file) != nbNon0ST)
+                EXM_THROW(93, "Write error : cannot write block : %s",
+                          strerror(errno));
+        }
+        ptrT += seg0SizeT;
+    }
+
+    {   static size_t const maskT = sizeof(size_t)-1;
+        if (bufferSize & maskT) {
+            /* size not multiple of sizeof(size_t) : implies end of block */
+            const char* const restStart = (const char*)bufferTEnd;
+            const char* restPtr = restStart;
+            const char* const restEnd = (const char*)buffer + bufferSize;
+            assert(restEnd > restStart && restEnd < restStart + sizeof(size_t));
+            for ( ; (restPtr < restEnd) && (*restPtr == 0); restPtr++) ;
+            storedSkips += (unsigned) (restPtr - restStart);
+            if (restPtr != restEnd) {
+                /* not all remaining bytes are 0 */
+                size_t const restSize = (size_t)(restEnd - restPtr);
+                if (LONG_SEEK(file, storedSkips, SEEK_CUR) != 0)
+                    EXM_THROW(92, "Sparse skip error ; try --no-sparse");
+                if (fwrite(restPtr, 1, restSize, file) != restSize)
+                    EXM_THROW(95, "Write error : cannot write end of decoded block : %s",
+                              strerror(errno));
+                storedSkips = 0;
+            }   }   }
+
+    return storedSkips;
+}
+
+static void
+AIO_fwriteSparseEnd(const FIO_prefs_t* const prefs, FILE* file, unsigned storedSkips)
+{
+    if (prefs->testMode) assert(storedSkips == 0);
+    if (storedSkips>0) {
+        assert(prefs->sparseFileSupport > 0);  /* storedSkips>0 implies sparse support is enabled */
+        (void)prefs;   /* assert can be disabled, in which case prefs becomes unused */
+        if (LONG_SEEK(file, storedSkips-1, SEEK_CUR) != 0)
+            EXM_THROW(69, "Final skip error (sparse file support)");
+        /* last zero must be explicitly written,
+         * so that skipped ones get implicitly translated as zero by FS */
+        {   const char lastZeroByte[1] = { 0 };
+            if (fwrite(lastZeroByte, 1, 1, file) != 1)
+                EXM_THROW(69, "Write error : cannot write last zero : %s", strerror(errno));
+        }   }
+}
+
+
+/* **********************************************************************
+ *  AsyncIO functionality
+ ************************************************************************/
+
+/* AIO_supported:
+ * Returns 1 if AsyncIO is supported on the system, 0 otherwise. */
+int AIO_supported(void) {
+#ifdef ZSTD_MULTITHREAD
+    return 1;
+#else
+    return 0;
+#endif
+}
+
+/* ***********************************
+ *  Generic IoPool implementation
+ *************************************/
+
+static IOJob_t *AIO_IOPool_createIoJob(IOPoolCtx_t *ctx, size_t bufferSize) {
+    IOJob_t* const job  = (IOJob_t*) malloc(sizeof(IOJob_t));
+    void* const buffer = malloc(bufferSize);
+    if(!job || !buffer)
+        EXM_THROW(101, "Allocation error : not enough memory");
+    job->buffer = buffer;
+    job->bufferSize = bufferSize;
+    job->usedBufferSize = 0;
+    job->file = NULL;
+    job->ctx = ctx;
+    job->offset = 0;
+    return job;
+}
+
+
+/* AIO_IOPool_createThreadPool:
+ * Creates a thread pool and a mutex for threaded IO pool.
+ * Displays warning if asyncio is requested but MT isn't available. */
+static void AIO_IOPool_createThreadPool(IOPoolCtx_t* ctx, const FIO_prefs_t* prefs) {
+    ctx->threadPool = NULL;
+    ctx->threadPoolActive = 0;
+    if(prefs->asyncIO) {
+        if (ZSTD_pthread_mutex_init(&ctx->ioJobsMutex, NULL))
+            EXM_THROW(102,"Failed creating ioJobsMutex mutex");
+        /* We want MAX_IO_JOBS-2 queue items because we need to always have 1 free buffer to
+         * decompress into and 1 buffer that's actively written to disk and owned by the writing thread. */
+        assert(MAX_IO_JOBS >= 2);
+        ctx->threadPool = POOL_create(1, MAX_IO_JOBS - 2);
+        ctx->threadPoolActive = 1;
+        if (!ctx->threadPool)
+            EXM_THROW(104, "Failed creating I/O thread pool");
+    }
+}
+
+/* AIO_IOPool_init:
+ * Allocates and sets and a new I/O thread pool including its included availableJobs. */
+static void AIO_IOPool_init(IOPoolCtx_t* ctx, const FIO_prefs_t* prefs, POOL_function poolFunction, size_t bufferSize) {
+    int i;
+    AIO_IOPool_createThreadPool(ctx, prefs);
+    ctx->prefs = prefs;
+    ctx->poolFunction = poolFunction;
+    ctx->totalIoJobs = ctx->threadPool ? MAX_IO_JOBS : 2;
+    ctx->availableJobsCount = ctx->totalIoJobs;
+    for(i=0; i < ctx->availableJobsCount; i++) {
+        ctx->availableJobs[i] = AIO_IOPool_createIoJob(ctx, bufferSize);
+    }
+    ctx->jobBufferSize = bufferSize;
+    ctx->file = NULL;
+}
+
+
+/* AIO_IOPool_threadPoolActive:
+ * Check if current operation uses thread pool.
+ * Note that in some cases we have a thread pool initialized but choose not to use it. */
+static int AIO_IOPool_threadPoolActive(IOPoolCtx_t* ctx) {
+    return ctx->threadPool && ctx->threadPoolActive;
+}
+
+
+/* AIO_IOPool_lockJobsMutex:
+ * Locks the IO jobs mutex if threading is active */
+static void AIO_IOPool_lockJobsMutex(IOPoolCtx_t* ctx) {
+    if(AIO_IOPool_threadPoolActive(ctx))
+        ZSTD_pthread_mutex_lock(&ctx->ioJobsMutex);
+}
+
+/* AIO_IOPool_unlockJobsMutex:
+ * Unlocks the IO jobs mutex if threading is active */
+static void AIO_IOPool_unlockJobsMutex(IOPoolCtx_t* ctx) {
+    if(AIO_IOPool_threadPoolActive(ctx))
+        ZSTD_pthread_mutex_unlock(&ctx->ioJobsMutex);
+}
+
+/* AIO_IOPool_releaseIoJob:
+ * Releases an acquired job back to the pool. Doesn't execute the job. */
+static void AIO_IOPool_releaseIoJob(IOJob_t* job) {
+    IOPoolCtx_t* const ctx = (IOPoolCtx_t *) job->ctx;
+    AIO_IOPool_lockJobsMutex(ctx);
+    assert(ctx->availableJobsCount < ctx->totalIoJobs);
+    ctx->availableJobs[ctx->availableJobsCount++] = job;
+    AIO_IOPool_unlockJobsMutex(ctx);
+}
+
+/* AIO_IOPool_join:
+ * Waits for all tasks in the pool to finish executing. */
+static void AIO_IOPool_join(IOPoolCtx_t* ctx) {
+    if(AIO_IOPool_threadPoolActive(ctx))
+        POOL_joinJobs(ctx->threadPool);
+}
+
+/* AIO_IOPool_setThreaded:
+ * Allows (de)activating threaded mode, to be used when the expected overhead
+ * of threading costs more than the expected gains. */
+static void AIO_IOPool_setThreaded(IOPoolCtx_t* ctx, int threaded) {
+    assert(threaded == 0 || threaded == 1);
+    assert(ctx != NULL);
+    if(ctx->threadPoolActive != threaded) {
+        AIO_IOPool_join(ctx);
+        ctx->threadPoolActive = threaded;
+    }
+}
+
+/* AIO_IOPool_free:
+ * Release a previously allocated IO thread pool. Makes sure all tasks are done and released. */
+static void AIO_IOPool_destroy(IOPoolCtx_t* ctx) {
+    int i;
+    if(ctx->threadPool) {
+        /* Make sure we finish all tasks and then free the resources */
+        AIO_IOPool_join(ctx);
+        /* Make sure we are not leaking availableJobs */
+        assert(ctx->availableJobsCount == ctx->totalIoJobs);
+        POOL_free(ctx->threadPool);
+        ZSTD_pthread_mutex_destroy(&ctx->ioJobsMutex);
+    }
+    assert(ctx->file == NULL);
+    for(i=0; i<ctx->availableJobsCount; i++) {
+        IOJob_t* job = (IOJob_t*) ctx->availableJobs[i];
+        free(job->buffer);
+        free(job);
+    }
+}
+
+/* AIO_IOPool_acquireJob:
+ * Returns an available io job to be used for a future io. */
+static IOJob_t* AIO_IOPool_acquireJob(IOPoolCtx_t* ctx) {
+    IOJob_t* job;
+    assert(ctx->file != NULL || ctx->prefs->testMode);
+    AIO_IOPool_lockJobsMutex(ctx);
+    assert(ctx->availableJobsCount > 0);
+    job = (IOJob_t*) ctx->availableJobs[--ctx->availableJobsCount];
+    AIO_IOPool_unlockJobsMutex(ctx);
+    job->usedBufferSize = 0;
+    job->file = ctx->file;
+    job->offset = 0;
+    return job;
+}
+
+
+/* AIO_IOPool_setFile:
+ * Sets the destination file for future files in the pool.
+ * Requires completion of all queued jobs and release of all otherwise acquired jobs. */
+static void AIO_IOPool_setFile(IOPoolCtx_t* ctx, FILE* file) {
+    assert(ctx!=NULL);
+    AIO_IOPool_join(ctx);
+    assert(ctx->availableJobsCount == ctx->totalIoJobs);
+    ctx->file = file;
+}
+
+static FILE* AIO_IOPool_getFile(const IOPoolCtx_t* ctx) {
+    return ctx->file;
+}
+
+/* AIO_IOPool_enqueueJob:
+ * Enqueues an io job for execution.
+ * The queued job shouldn't be used directly after queueing it. */
+static void AIO_IOPool_enqueueJob(IOJob_t* job) {
+    IOPoolCtx_t* const ctx = (IOPoolCtx_t *)job->ctx;
+    if(AIO_IOPool_threadPoolActive(ctx))
+        POOL_add(ctx->threadPool, ctx->poolFunction, job);
+    else
+        ctx->poolFunction(job);
+}
+
+/* ***********************************
+ *  WritePool implementation
+ *************************************/
+
+/* AIO_WritePool_acquireJob:
+ * Returns an available write job to be used for a future write. */
+IOJob_t* AIO_WritePool_acquireJob(WritePoolCtx_t* ctx) {
+    return AIO_IOPool_acquireJob(&ctx->base);
+}
+
+/* AIO_WritePool_enqueueAndReacquireWriteJob:
+ * Queues a write job for execution and acquires a new one.
+ * After execution `job`'s pointed value would change to the newly acquired job.
+ * Make sure to set `usedBufferSize` to the wanted length before call.
+ * The queued job shouldn't be used directly after queueing it. */
+void AIO_WritePool_enqueueAndReacquireWriteJob(IOJob_t **job) {
+    AIO_IOPool_enqueueJob(*job);
+    *job = AIO_IOPool_acquireJob((IOPoolCtx_t *)(*job)->ctx);
+}
+
+/* AIO_WritePool_sparseWriteEnd:
+ * Ends sparse writes to the current file.
+ * Blocks on completion of all current write jobs before executing. */
+void AIO_WritePool_sparseWriteEnd(WritePoolCtx_t* ctx) {
+    assert(ctx != NULL);
+    AIO_IOPool_join(&ctx->base);
+    AIO_fwriteSparseEnd(ctx->base.prefs, ctx->base.file, ctx->storedSkips);
+    ctx->storedSkips = 0;
+}
+
+/* AIO_WritePool_setFile:
+ * Sets the destination file for future writes in the pool.
+ * Requires completion of all queues write jobs and release of all otherwise acquired jobs.
+ * Also requires ending of sparse write if a previous file was used in sparse mode. */
+void AIO_WritePool_setFile(WritePoolCtx_t* ctx, FILE* file) {
+    AIO_IOPool_setFile(&ctx->base, file);
+    assert(ctx->storedSkips == 0);
+}
+
+/* AIO_WritePool_getFile:
+ * Returns the file the writePool is currently set to write to. */
+FILE* AIO_WritePool_getFile(const WritePoolCtx_t* ctx) {
+    return AIO_IOPool_getFile(&ctx->base);
+}
+
+/* AIO_WritePool_releaseIoJob:
+ * Releases an acquired job back to the pool. Doesn't execute the job. */
+void AIO_WritePool_releaseIoJob(IOJob_t* job) {
+    AIO_IOPool_releaseIoJob(job);
+}
+
+/* AIO_WritePool_closeFile:
+ * Ends sparse write and closes the writePool's current file and sets the file to NULL.
+ * Requires completion of all queues write jobs and release of all otherwise acquired jobs.  */
+int AIO_WritePool_closeFile(WritePoolCtx_t* ctx) {
+    FILE* const dstFile = ctx->base.file;
+    assert(dstFile!=NULL || ctx->base.prefs->testMode!=0);
+    AIO_WritePool_sparseWriteEnd(ctx);
+    AIO_IOPool_setFile(&ctx->base, NULL);
+    return fclose(dstFile);
+}
+
+/* AIO_WritePool_executeWriteJob:
+ * Executes a write job synchronously. Can be used as a function for a thread pool. */
+static void AIO_WritePool_executeWriteJob(void* opaque){
+    IOJob_t* const job = (IOJob_t*) opaque;
+    WritePoolCtx_t* const ctx = (WritePoolCtx_t*) job->ctx;
+    ctx->storedSkips = AIO_fwriteSparse(job->file, job->buffer, job->usedBufferSize, ctx->base.prefs, ctx->storedSkips);
+    AIO_IOPool_releaseIoJob(job);
+}
+
+/* AIO_WritePool_create:
+ * Allocates and sets and a new write pool including its included jobs. */
+WritePoolCtx_t* AIO_WritePool_create(const FIO_prefs_t* prefs, size_t bufferSize) {
+    WritePoolCtx_t* const ctx = (WritePoolCtx_t*) malloc(sizeof(WritePoolCtx_t));
+    if(!ctx) EXM_THROW(100, "Allocation error : not enough memory");
+    AIO_IOPool_init(&ctx->base, prefs, AIO_WritePool_executeWriteJob, bufferSize);
+    ctx->storedSkips = 0;
+    return ctx;
+}
+
+/* AIO_WritePool_free:
+ * Frees and releases a writePool and its resources. Closes destination file if needs to. */
+void AIO_WritePool_free(WritePoolCtx_t* ctx) {
+    /* Make sure we finish all tasks and then free the resources */
+    if(AIO_WritePool_getFile(ctx))
+        AIO_WritePool_closeFile(ctx);
+    AIO_IOPool_destroy(&ctx->base);
+    assert(ctx->storedSkips==0);
+    free(ctx);
+}
+
+/* AIO_WritePool_setAsync:
+ * Allows (de)activating async mode, to be used when the expected overhead
+ * of asyncio costs more than the expected gains. */
+void AIO_WritePool_setAsync(WritePoolCtx_t* ctx, int async) {
+    AIO_IOPool_setThreaded(&ctx->base, async);
+}
+
+
+/* ***********************************
+ *  ReadPool implementation
+ *************************************/
+static void AIO_ReadPool_releaseAllCompletedJobs(ReadPoolCtx_t* ctx) {
+    int i;
+    for(i=0; i<ctx->completedJobsCount; i++) {
+        IOJob_t* job = (IOJob_t*) ctx->completedJobs[i];
+        AIO_IOPool_releaseIoJob(job);
+    }
+    ctx->completedJobsCount = 0;
+}
+
+static void AIO_ReadPool_addJobToCompleted(IOJob_t* job) {
+    ReadPoolCtx_t* const ctx = (ReadPoolCtx_t *)job->ctx;
+    AIO_IOPool_lockJobsMutex(&ctx->base);
+    assert(ctx->completedJobsCount < MAX_IO_JOBS);
+    ctx->completedJobs[ctx->completedJobsCount++] = job;
+    if(AIO_IOPool_threadPoolActive(&ctx->base)) {
+        ZSTD_pthread_cond_signal(&ctx->jobCompletedCond);
+    }
+    AIO_IOPool_unlockJobsMutex(&ctx->base);
+}
+
+/* AIO_ReadPool_findNextWaitingOffsetCompletedJob_locked:
+ * Looks through the completed jobs for a job matching the waitingOnOffset and returns it,
+ * if job wasn't found returns NULL.
+ * IMPORTANT: assumes ioJobsMutex is locked. */
+static IOJob_t* AIO_ReadPool_findNextWaitingOffsetCompletedJob_locked(ReadPoolCtx_t* ctx) {
+    IOJob_t *job = NULL;
+    int i;
+    /* This implementation goes through all completed jobs and looks for the one matching the next offset.
+     * While not strictly needed for a single threaded reader implementation (as in such a case we could expect
+     * reads to be completed in order) this implementation was chosen as it better fits other asyncio
+     * interfaces (such as io_uring) that do not provide promises regarding order of completion. */
+    for (i=0; i<ctx->completedJobsCount; i++) {
+        job = (IOJob_t *) ctx->completedJobs[i];
+        if (job->offset == ctx->waitingOnOffset) {
+            ctx->completedJobs[i] = ctx->completedJobs[--ctx->completedJobsCount];
+            return job;
+        }
+    }
+    return NULL;
+}
+
+/* AIO_ReadPool_numReadsInFlight:
+ * Returns the number of IO read jobs currently in flight. */
+static size_t AIO_ReadPool_numReadsInFlight(ReadPoolCtx_t* ctx) {
+    const int jobsHeld = (ctx->currentJobHeld==NULL ? 0 : 1);
+    return (size_t)(ctx->base.totalIoJobs - (ctx->base.availableJobsCount + ctx->completedJobsCount + jobsHeld));
+}
+
+/* AIO_ReadPool_getNextCompletedJob:
+ * Returns a completed IOJob_t for the next read in line based on waitingOnOffset and advances waitingOnOffset.
+ * Would block. */
+static IOJob_t* AIO_ReadPool_getNextCompletedJob(ReadPoolCtx_t* ctx) {
+    IOJob_t *job = NULL;
+    AIO_IOPool_lockJobsMutex(&ctx->base);
+
+    job = AIO_ReadPool_findNextWaitingOffsetCompletedJob_locked(ctx);
+
+    /* As long as we didn't find the job matching the next read, and we have some reads in flight continue waiting */
+    while (!job && (AIO_ReadPool_numReadsInFlight(ctx) > 0)) {
+        assert(ctx->base.threadPool != NULL); /* we shouldn't be here if we work in sync mode */
+        ZSTD_pthread_cond_wait(&ctx->jobCompletedCond, &ctx->base.ioJobsMutex);
+        job = AIO_ReadPool_findNextWaitingOffsetCompletedJob_locked(ctx);
+    }
+
+    if(job) {
+        assert(job->offset == ctx->waitingOnOffset);
+        ctx->waitingOnOffset += job->usedBufferSize;
+    }
+
+    AIO_IOPool_unlockJobsMutex(&ctx->base);
+    return job;
+}
+
+
+/* AIO_ReadPool_executeReadJob:
+ * Executes a read job synchronously. Can be used as a function for a thread pool. */
+static void AIO_ReadPool_executeReadJob(void* opaque){
+    IOJob_t* const job = (IOJob_t*) opaque;
+    ReadPoolCtx_t* const ctx = (ReadPoolCtx_t *)job->ctx;
+    if(ctx->reachedEof) {
+        job->usedBufferSize = 0;
+        AIO_ReadPool_addJobToCompleted(job);
+        return;
+    }
+    job->usedBufferSize = fread(job->buffer, 1, job->bufferSize, job->file);
+    if(job->usedBufferSize < job->bufferSize) {
+        if(ferror(job->file)) {
+            EXM_THROW(37, "Read error");
+        } else if(feof(job->file)) {
+            ctx->reachedEof = 1;
+        } else {
+            EXM_THROW(37, "Unexpected short read");
+        }
+    }
+    AIO_ReadPool_addJobToCompleted(job);
+}
+
+static void AIO_ReadPool_enqueueRead(ReadPoolCtx_t* ctx) {
+    IOJob_t* const job = AIO_IOPool_acquireJob(&ctx->base);
+    job->offset = ctx->nextReadOffset;
+    ctx->nextReadOffset += job->bufferSize;
+    AIO_IOPool_enqueueJob(job);
+}
+
+static void AIO_ReadPool_startReading(ReadPoolCtx_t* ctx) {
+    while(ctx->base.availableJobsCount) {
+        AIO_ReadPool_enqueueRead(ctx);
+    }
+}
+
+/* AIO_ReadPool_setFile:
+ * Sets the source file for future read in the pool. Initiates reading immediately if file is not NULL.
+ * Waits for all current enqueued tasks to complete if a previous file was set. */
+void AIO_ReadPool_setFile(ReadPoolCtx_t* ctx, FILE* file) {
+    assert(ctx!=NULL);
+    AIO_IOPool_join(&ctx->base);
+    AIO_ReadPool_releaseAllCompletedJobs(ctx);
+    if (ctx->currentJobHeld) {
+        AIO_IOPool_releaseIoJob((IOJob_t *)ctx->currentJobHeld);
+        ctx->currentJobHeld = NULL;
+    }
+    AIO_IOPool_setFile(&ctx->base, file);
+    ctx->nextReadOffset = 0;
+    ctx->waitingOnOffset = 0;
+    ctx->srcBuffer = ctx->coalesceBuffer;
+    ctx->srcBufferLoaded = 0;
+    ctx->reachedEof = 0;
+    if(file != NULL)
+        AIO_ReadPool_startReading(ctx);
+}
+
+/* AIO_ReadPool_create:
+ * Allocates and sets and a new readPool including its included jobs.
+ * bufferSize should be set to the maximal buffer we want to read at a time, will also be used
+ * as our basic read size. */
+ReadPoolCtx_t* AIO_ReadPool_create(const FIO_prefs_t* prefs, size_t bufferSize) {
+    ReadPoolCtx_t* const ctx = (ReadPoolCtx_t*) malloc(sizeof(ReadPoolCtx_t));
+    if(!ctx) EXM_THROW(100, "Allocation error : not enough memory");
+    AIO_IOPool_init(&ctx->base, prefs, AIO_ReadPool_executeReadJob, bufferSize);
+
+    ctx->coalesceBuffer = (U8*) malloc(bufferSize * 2);
+    if(!ctx->coalesceBuffer) EXM_THROW(100, "Allocation error : not enough memory");
+    ctx->srcBuffer = ctx->coalesceBuffer;
+    ctx->srcBufferLoaded = 0;
+    ctx->completedJobsCount = 0;
+    ctx->currentJobHeld = NULL;
+
+    if(ctx->base.threadPool)
+        if (ZSTD_pthread_cond_init(&ctx->jobCompletedCond, NULL))
+            EXM_THROW(103,"Failed creating jobCompletedCond cond");
+
+    return ctx;
+}
+
+/* AIO_ReadPool_free:
+ * Frees and releases a readPool and its resources. Closes source file. */
+void AIO_ReadPool_free(ReadPoolCtx_t* ctx) {
+    if(AIO_ReadPool_getFile(ctx))
+        AIO_ReadPool_closeFile(ctx);
+    if(ctx->base.threadPool)
+        ZSTD_pthread_cond_destroy(&ctx->jobCompletedCond);
+    AIO_IOPool_destroy(&ctx->base);
+    free(ctx->coalesceBuffer);
+    free(ctx);
+}
+
+/* AIO_ReadPool_consumeBytes:
+ * Consumes byes from srcBuffer's beginning and updates srcBufferLoaded accordingly. */
+void AIO_ReadPool_consumeBytes(ReadPoolCtx_t* ctx, size_t n) {
+    assert(n <= ctx->srcBufferLoaded);
+    ctx->srcBufferLoaded -= n;
+    ctx->srcBuffer += n;
+}
+
+/* AIO_ReadPool_releaseCurrentlyHeldAndGetNext:
+ * Release the current held job and get the next one, returns NULL if no next job available. */
+static IOJob_t* AIO_ReadPool_releaseCurrentHeldAndGetNext(ReadPoolCtx_t* ctx) {
+    if (ctx->currentJobHeld) {
+        AIO_IOPool_releaseIoJob((IOJob_t *)ctx->currentJobHeld);
+        ctx->currentJobHeld = NULL;
+        AIO_ReadPool_enqueueRead(ctx);
+    }
+    ctx->currentJobHeld = AIO_ReadPool_getNextCompletedJob(ctx);
+    return (IOJob_t*) ctx->currentJobHeld;
+}
+
+/* AIO_ReadPool_fillBuffer:
+ * Tries to fill the buffer with at least n or jobBufferSize bytes (whichever is smaller).
+ * Returns if srcBuffer has at least the expected number of bytes loaded or if we've reached the end of the file.
+ * Return value is the number of bytes added to the buffer.
+ * Note that srcBuffer might have up to 2 times jobBufferSize bytes. */
+size_t AIO_ReadPool_fillBuffer(ReadPoolCtx_t* ctx, size_t n) {
+    IOJob_t *job;
+    int useCoalesce = 0;
+    if(n > ctx->base.jobBufferSize)
+        n = ctx->base.jobBufferSize;
+
+    /* We are good, don't read anything */
+    if (ctx->srcBufferLoaded >= n)
+        return 0;
+
+    /* We still have bytes loaded, but not enough to satisfy caller. We need to get the next job
+     * and coalesce the remaining bytes with the next job's buffer */
+    if (ctx->srcBufferLoaded > 0) {
+        useCoalesce = 1;
+        memcpy(ctx->coalesceBuffer, ctx->srcBuffer, ctx->srcBufferLoaded);
+        ctx->srcBuffer = ctx->coalesceBuffer;
+    }
+
+    /* Read the next chunk */
+    job = AIO_ReadPool_releaseCurrentHeldAndGetNext(ctx);
+    if(!job)
+        return 0;
+    if(useCoalesce) {
+        assert(ctx->srcBufferLoaded + job->usedBufferSize <= 2*ctx->base.jobBufferSize);
+        memcpy(ctx->coalesceBuffer + ctx->srcBufferLoaded, job->buffer, job->usedBufferSize);
+        ctx->srcBufferLoaded += job->usedBufferSize;
+    }
+    else {
+        ctx->srcBuffer = (U8 *) job->buffer;
+        ctx->srcBufferLoaded = job->usedBufferSize;
+    }
+    return job->usedBufferSize;
+}
+
+/* AIO_ReadPool_consumeAndRefill:
+ * Consumes the current buffer and refills it with bufferSize bytes. */
+size_t AIO_ReadPool_consumeAndRefill(ReadPoolCtx_t* ctx) {
+    AIO_ReadPool_consumeBytes(ctx, ctx->srcBufferLoaded);
+    return AIO_ReadPool_fillBuffer(ctx, ctx->base.jobBufferSize);
+}
+
+/* AIO_ReadPool_getFile:
+ * Returns the current file set for the read pool. */
+FILE* AIO_ReadPool_getFile(const ReadPoolCtx_t* ctx) {
+    return AIO_IOPool_getFile(&ctx->base);
+}
+
+/* AIO_ReadPool_closeFile:
+ * Closes the current set file. Waits for all current enqueued tasks to complete and resets state. */
+int AIO_ReadPool_closeFile(ReadPoolCtx_t* ctx) {
+    FILE* const file = AIO_ReadPool_getFile(ctx);
+    AIO_ReadPool_setFile(ctx, NULL);
+    return fclose(file);
+}
+
+/* AIO_ReadPool_setAsync:
+ * Allows (de)activating async mode, to be used when the expected overhead
+ * of asyncio costs more than the expected gains. */
+void AIO_ReadPool_setAsync(ReadPoolCtx_t* ctx, int async) {
+    AIO_IOPool_setThreaded(&ctx->base, async);
+}
diff --git a/programs/lorem.c b/programs/lorem.c
new file mode 100644
index 0000000..79030c9
--- /dev/null
+++ b/programs/lorem.c
@@ -0,0 +1,285 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* Implementation notes:
+ *
+ * This is a very simple lorem ipsum generator
+ * which features a static list of words
+ * and print them one after another randomly
+ * with a fake sentence / paragraph structure.
+ *
+ * The goal is to generate a printable text
+ * that can be used to fake a text compression scenario.
+ * The resulting compression / ratio curve of the lorem ipsum generator
+ * is more satisfying than the previous statistical generator,
+ * which was initially designed for entropy compression,
+ * and lacks a regularity more representative of text.
+ *
+ * The compression ratio achievable on the generated lorem ipsum
+ * is still a bit too good, presumably because the dictionary is a bit too
+ * small. It would be possible to create some more complex scheme, notably by
+ * enlarging the dictionary with a word generator, and adding grammatical rules
+ * (composition) and syntax rules. But that's probably overkill for the intended
+ * goal.
+ */
+
+#include "lorem.h"
+#include <assert.h>
+#include <limits.h> /* INT_MAX */
+#include <string.h> /* memcpy */
+
+#define WORD_MAX_SIZE 20
+
+/* Define the word pool */
+static const char* kWords[] = {
+    "lorem",        "ipsum",      "dolor",       "sit",          "amet",
+    "consectetur",  "adipiscing", "elit",        "sed",          "do",
+    "eiusmod",      "tempor",     "incididunt",  "ut",           "labore",
+    "et",           "dolore",     "magna",       "aliqua",       "dis",
+    "lectus",       "vestibulum", "mattis",      "ullamcorper",  "velit",
+    "commodo",      "a",          "lacus",       "arcu",         "magnis",
+    "parturient",   "montes",     "nascetur",    "ridiculus",    "mus",
+    "mauris",       "nulla",      "malesuada",   "pellentesque", "eget",
+    "gravida",      "in",         "dictum",      "non",          "erat",
+    "nam",          "voluptat",   "maecenas",    "blandit",      "aliquam",
+    "etiam",        "enim",       "lobortis",    "scelerisque",  "fermentum",
+    "dui",          "faucibus",   "ornare",      "at",           "elementum",
+    "eu",           "facilisis",  "odio",        "morbi",        "quis",
+    "eros",         "donec",      "ac",          "orci",         "purus",
+    "turpis",       "cursus",     "leo",         "vel",          "porta",
+    "consequat",    "interdum",   "varius",      "vulputate",    "aliquet",
+    "pharetra",     "nunc",       "auctor",      "urna",         "id",
+    "metus",        "viverra",    "nibh",        "cras",         "mi",
+    "unde",         "omnis",      "iste",        "natus",        "error",
+    "perspiciatis", "voluptatem", "accusantium", "doloremque",   "laudantium",
+    "totam",        "rem",        "aperiam",     "eaque",        "ipsa",
+    "quae",         "ab",         "illo",        "inventore",    "veritatis",
+    "quasi",        "architecto", "beatae",      "vitae",        "dicta",
+    "sunt",         "explicabo",  "nemo",        "ipsam",        "quia",
+    "voluptas",     "aspernatur", "aut",         "odit",         "fugit",
+    "consequuntur", "magni",      "dolores",     "eos",          "qui",
+    "ratione",      "sequi",      "nesciunt",    "neque",        "porro",
+    "quisquam",     "est",        "dolorem",     "adipisci",     "numquam",
+    "eius",         "modi",       "tempora",     "incidunt",     "magnam",
+    "quaerat",      "ad",         "minima",      "veniam",       "nostrum",
+    "ullam",        "corporis",   "suscipit",    "laboriosam",   "nisi",
+    "aliquid",      "ex",         "ea",          "commodi",      "consequatur",
+    "autem",        "eum",        "iure",        "voluptate",    "esse",
+    "quam",         "nihil",      "molestiae",   "illum",        "fugiat",
+    "quo",          "pariatur",   "vero",        "accusamus",    "iusto",
+    "dignissimos",  "ducimus",    "blanditiis",  "praesentium",  "voluptatum",
+    "deleniti",     "atque",      "corrupti",    "quos",         "quas",
+    "molestias",    "excepturi",  "sint",        "occaecati",    "cupiditate",
+    "provident",    "similique",  "culpa",       "officia",      "deserunt",
+    "mollitia",     "animi",      "laborum",     "dolorum",      "fuga",
+    "harum",        "quidem",     "rerum",       "facilis",      "expedita",
+    "distinctio",   "libero",     "tempore",     "cum",          "soluta",
+    "nobis",        "eligendi",   "optio",       "cumque",       "impedit",
+    "minus",        "quod",       "maxime",      "placeat",      "facere",
+    "possimus",     "assumenda",  "repellendus", "temporibus",   "quibusdam",
+    "officiis",     "debitis",    "saepe",       "eveniet",      "voluptates",
+    "repudiandae",  "recusandae", "itaque",      "earum",        "hic",
+    "tenetur",      "sapiente",   "delectus",    "reiciendis",   "cillum",
+    "maiores",      "alias",      "perferendis", "doloribus",    "asperiores",
+    "repellat",     "minim",      "nostrud",     "exercitation", "ullamco",
+    "laboris",      "aliquip",    "duis",        "aute",         "irure",
+};
+static const unsigned kNbWords = sizeof(kWords) / sizeof(kWords[0]);
+
+/* simple 1-dimension distribution, based on word's length, favors small words
+ */
+static const int kWeights[]    = { 0, 8, 6, 4, 3, 2 };
+static const size_t kNbWeights = sizeof(kWeights) / sizeof(kWeights[0]);
+
+#define DISTRIB_SIZE_MAX 650
+static int g_distrib[DISTRIB_SIZE_MAX] = { 0 };
+static unsigned g_distribCount         = 0;
+
+static void countFreqs(
+        const char* words[],
+        size_t nbWords,
+        const int* weights,
+        size_t nbWeights)
+{
+    unsigned total = 0;
+    size_t w;
+    for (w = 0; w < nbWords; w++) {
+        size_t len = strlen(words[w]);
+        int lmax;
+        if (len >= nbWeights)
+            len = nbWeights - 1;
+        lmax = weights[len];
+        total += (unsigned)lmax;
+    }
+    g_distribCount = total;
+    assert(g_distribCount <= DISTRIB_SIZE_MAX);
+}
+
+static void init_word_distrib(
+        const char* words[],
+        size_t nbWords,
+        const int* weights,
+        size_t nbWeights)
+{
+    size_t w, d = 0;
+    countFreqs(words, nbWords, weights, nbWeights);
+    for (w = 0; w < nbWords; w++) {
+        size_t len = strlen(words[w]);
+        int l, lmax;
+        if (len >= nbWeights)
+            len = nbWeights - 1;
+        lmax = weights[len];
+        for (l = 0; l < lmax; l++) {
+            g_distrib[d++] = (int)w;
+        }
+    }
+}
+
+/* Note: this unit only works when invoked sequentially.
+ * No concurrent access is allowed */
+static char* g_ptr         = NULL;
+static size_t g_nbChars    = 0;
+static size_t g_maxChars   = 10000000;
+static unsigned g_randRoot = 0;
+
+#define RDG_rotl32(x, r) ((x << r) | (x >> (32 - r)))
+static unsigned LOREM_rand(unsigned range)
+{
+    static const unsigned prime1 = 2654435761U;
+    static const unsigned prime2 = 2246822519U;
+    unsigned rand32              = g_randRoot;
+    rand32 *= prime1;
+    rand32 ^= prime2;
+    rand32     = RDG_rotl32(rand32, 13);
+    g_randRoot = rand32;
+    return (unsigned)(((unsigned long long)rand32 * range) >> 32);
+}
+
+static void writeLastCharacters(void)
+{
+    size_t lastChars = g_maxChars - g_nbChars;
+    assert(g_maxChars >= g_nbChars);
+    if (lastChars == 0)
+        return;
+    g_ptr[g_nbChars++] = '.';
+    if (lastChars > 2) {
+        memset(g_ptr + g_nbChars, ' ', lastChars - 2);
+    }
+    if (lastChars > 1) {
+        g_ptr[g_maxChars - 1] = '\n';
+    }
+    g_nbChars = g_maxChars;
+}
+
+static void generateWord(const char* word, const char* separator, int upCase)
+{
+    size_t const len = strlen(word) + strlen(separator);
+    if (g_nbChars + len > g_maxChars) {
+        writeLastCharacters();
+        return;
+    }
+    memcpy(g_ptr + g_nbChars, word, strlen(word));
+    if (upCase) {
+        static const char toUp = 'A' - 'a';
+        g_ptr[g_nbChars]       = (char)(g_ptr[g_nbChars] + toUp);
+    }
+    g_nbChars += strlen(word);
+    memcpy(g_ptr + g_nbChars, separator, strlen(separator));
+    g_nbChars += strlen(separator);
+}
+
+static int about(unsigned target)
+{
+    return (int)(LOREM_rand(target) + LOREM_rand(target) + 1);
+}
+
+/* Function to generate a random sentence */
+static void generateSentence(int nbWords)
+{
+    int commaPos       = about(9);
+    int comma2         = commaPos + about(7);
+    int qmark          = (LOREM_rand(11) == 7);
+    const char* endSep = qmark ? "? " : ". ";
+    int i;
+    for (i = 0; i < nbWords; i++) {
+        int const wordID       = g_distrib[LOREM_rand(g_distribCount)];
+        const char* const word = kWords[wordID];
+        const char* sep        = " ";
+        if (i == commaPos)
+            sep = ", ";
+        if (i == comma2)
+            sep = ", ";
+        if (i == nbWords - 1)
+            sep = endSep;
+        generateWord(word, sep, i == 0);
+    }
+}
+
+static void generateParagraph(int nbSentences)
+{
+    int i;
+    for (i = 0; i < nbSentences; i++) {
+        int wordsPerSentence = about(11);
+        generateSentence(wordsPerSentence);
+    }
+    if (g_nbChars < g_maxChars) {
+        g_ptr[g_nbChars++] = '\n';
+    }
+    if (g_nbChars < g_maxChars) {
+        g_ptr[g_nbChars++] = '\n';
+    }
+}
+
+/* It's "common" for lorem ipsum generators to start with the same first
+ * pre-defined sentence */
+static void generateFirstSentence(void)
+{
+    int i;
+    for (i = 0; i < 18; i++) {
+        const char* word      = kWords[i];
+        const char* separator = " ";
+        if (i == 4)
+            separator = ", ";
+        if (i == 7)
+            separator = ", ";
+        generateWord(word, separator, i == 0);
+    }
+    generateWord(kWords[18], ". ", 0);
+}
+
+size_t
+LOREM_genBlock(void* buffer, size_t size, unsigned seed, int first, int fill)
+{
+    g_ptr = (char*)buffer;
+    assert(size < INT_MAX);
+    g_maxChars = size;
+    g_nbChars  = 0;
+    g_randRoot = seed;
+    if (g_distribCount == 0) {
+        init_word_distrib(kWords, kNbWords, kWeights, kNbWeights);
+    }
+
+    if (first) {
+        generateFirstSentence();
+    }
+    while (g_nbChars < g_maxChars) {
+        int sentencePerParagraph = about(7);
+        generateParagraph(sentencePerParagraph);
+        if (!fill)
+            break; /* only generate one paragraph in not-fill mode */
+    }
+    g_ptr = NULL;
+    return g_nbChars;
+}
+
+void LOREM_genBuffer(void* buffer, size_t size, unsigned seed)
+{
+    LOREM_genBlock(buffer, size, seed, 1, 1);
+}
diff --git a/programs/util.c b/programs/util.c
new file mode 100644
index 0000000..d11f13c
--- /dev/null
+++ b/programs/util.c
@@ -0,0 +1,1730 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/*-****************************************
+*  Dependencies
+******************************************/
+#include "util.h"       /* note : ensure that platform.h is included first ! */
+#include <stdlib.h>     /* malloc, realloc, free */
+#include <stdio.h>      /* fprintf */
+#include <time.h>       /* clock_t, clock, CLOCKS_PER_SEC, nanosleep */
+#include <errno.h>
+#include <assert.h>
+
+#if defined(__FreeBSD__)
+#include <sys/param.h> /* __FreeBSD_version */
+#endif /* #ifdef __FreeBSD__ */
+
+#if defined(_WIN32)
+#  include <sys/utime.h>  /* utime */
+#  include <io.h>         /* _chmod */
+#  define ZSTD_USE_UTIMENSAT 0
+#else
+#  include <unistd.h>     /* chown, stat */
+#  include <sys/stat.h>   /* utimensat, st_mtime */
+#  if (PLATFORM_POSIX_VERSION >= 200809L && defined(st_mtime)) \
+      || (defined(__FreeBSD__) && __FreeBSD_version >= 1100056)
+#    define ZSTD_USE_UTIMENSAT 1
+#  else
+#    define ZSTD_USE_UTIMENSAT 0
+#  endif
+#  if ZSTD_USE_UTIMENSAT
+#    include <fcntl.h>    /* AT_FDCWD */
+#  else
+#    include <utime.h>    /* utime */
+#  endif
+#endif
+
+#if defined(_MSC_VER) || defined(__MINGW32__) || defined (__MSVCRT__)
+#include <direct.h>     /* needed for _mkdir in windows */
+#endif
+
+#if defined(__linux__) || (PLATFORM_POSIX_VERSION >= 200112L)  /* opendir, readdir require POSIX.1-2001 */
+#  include <dirent.h>       /* opendir, readdir */
+#  include <string.h>       /* strerror, memcpy */
+#endif /* #ifdef _WIN32 */
+
+/*-****************************************
+*  Internal Macros
+******************************************/
+
+/* CONTROL is almost like an assert(), but is never disabled.
+ * It's designed for failures that may happen rarely,
+ * but we don't want to maintain a specific error code path for them,
+ * such as a malloc() returning NULL for example.
+ * Since it's always active, this macro can trigger side effects.
+ */
+#define CONTROL(c)  {         \
+    if (!(c)) {               \
+        UTIL_DISPLAYLEVEL(1, "Error : %s, %i : %s",  \
+                          __FILE__, __LINE__, #c);   \
+        exit(1);              \
+}   }
+
+/* console log */
+#define UTIL_DISPLAY(...)         fprintf(stderr, __VA_ARGS__)
+#define UTIL_DISPLAYLEVEL(l, ...) { if (g_utilDisplayLevel>=l) { UTIL_DISPLAY(__VA_ARGS__); } }
+
+static int g_traceDepth = 0;
+int g_traceFileStat = 0;
+
+#define UTIL_TRACE_CALL(...)                                         \
+    {                                                                \
+        if (g_traceFileStat) {                                       \
+            UTIL_DISPLAY("Trace:FileStat: %*s> ", g_traceDepth, ""); \
+            UTIL_DISPLAY(__VA_ARGS__);                               \
+            UTIL_DISPLAY("\n");                                      \
+            ++g_traceDepth;                                          \
+        }                                                            \
+    }
+
+#define UTIL_TRACE_RET(ret)                                                     \
+    {                                                                           \
+        if (g_traceFileStat) {                                                  \
+            --g_traceDepth;                                                     \
+            UTIL_DISPLAY("Trace:FileStat: %*s< %d\n", g_traceDepth, "", (ret)); \
+        }                                                                      \
+    }
+
+/* A modified version of realloc().
+ * If UTIL_realloc() fails the original block is freed.
+ */
+UTIL_STATIC void* UTIL_realloc(void *ptr, size_t size)
+{
+    void *newptr = realloc(ptr, size);
+    if (newptr) return newptr;
+    free(ptr);
+    return NULL;
+}
+
+#if defined(_MSC_VER)
+    #define chmod _chmod
+#endif
+
+#ifndef ZSTD_HAVE_FCHMOD
+#if PLATFORM_POSIX_VERSION >= 199309L
+#define ZSTD_HAVE_FCHMOD
+#endif
+#endif
+
+#ifndef ZSTD_HAVE_FCHOWN
+#if PLATFORM_POSIX_VERSION >= 200809L
+#define ZSTD_HAVE_FCHOWN
+#endif
+#endif
+
+/*-****************************************
+*  Console log
+******************************************/
+int g_utilDisplayLevel;
+
+int UTIL_requireUserConfirmation(const char* prompt, const char* abortMsg,
+                                 const char* acceptableLetters, int hasStdinInput) {
+    int ch, result;
+
+    if (hasStdinInput) {
+        UTIL_DISPLAY("stdin is an input - not proceeding.\n");
+        return 1;
+    }
+
+    UTIL_DISPLAY("%s", prompt);
+    ch = getchar();
+    result = 0;
+    if (strchr(acceptableLetters, ch) == NULL) {
+        UTIL_DISPLAY("%s \n", abortMsg);
+        result = 1;
+    }
+    /* flush the rest */
+    while ((ch!=EOF) && (ch!='\n'))
+        ch = getchar();
+    return result;
+}
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define KB * (1 << 10)
+#define LIST_SIZE_INCREASE   (8 KB)
+#define MAX_FILE_OF_FILE_NAMES_SIZE (1<<20)*50
+
+
+/*-*************************************
+*  Functions
+***************************************/
+
+void UTIL_traceFileStat(void)
+{
+    g_traceFileStat = 1;
+}
+
+int UTIL_fstat(const int fd, const char* filename, stat_t* statbuf)
+{
+    int ret;
+    UTIL_TRACE_CALL("UTIL_stat(%d, %s)", fd, filename);
+#if defined(_MSC_VER)
+    if (fd >= 0) {
+        ret = !_fstat64(fd, statbuf);
+    } else {
+        ret = !_stat64(filename, statbuf);
+    }
+#elif defined(__MINGW32__) && defined (__MSVCRT__)
+    if (fd >= 0) {
+        ret = !_fstati64(fd, statbuf);
+    } else {
+        ret = !_stati64(filename, statbuf);
+    }
+#else
+    if (fd >= 0) {
+        ret = !fstat(fd, statbuf);
+    } else {
+        ret = !stat(filename, statbuf);
+    }
+#endif
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+int UTIL_stat(const char* filename, stat_t* statbuf)
+{
+    return UTIL_fstat(-1, filename, statbuf);
+}
+
+int UTIL_isFdRegularFile(int fd)
+{
+    stat_t statbuf;
+    int ret;
+    UTIL_TRACE_CALL("UTIL_isFdRegularFile(%d)", fd);
+    ret = fd >= 0 && UTIL_fstat(fd, "", &statbuf) && UTIL_isRegularFileStat(&statbuf);
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+int UTIL_isRegularFile(const char* infilename)
+{
+    stat_t statbuf;
+    int ret;
+    UTIL_TRACE_CALL("UTIL_isRegularFile(%s)", infilename);
+    ret = UTIL_stat(infilename, &statbuf) && UTIL_isRegularFileStat(&statbuf);
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+int UTIL_isRegularFileStat(const stat_t* statbuf)
+{
+#if defined(_MSC_VER)
+    return (statbuf->st_mode & S_IFREG) != 0;
+#else
+    return S_ISREG(statbuf->st_mode) != 0;
+#endif
+}
+
+/* like chmod, but avoid changing permission of /dev/null */
+int UTIL_chmod(char const* filename, const stat_t* statbuf, mode_t permissions)
+{
+    return UTIL_fchmod(-1, filename, statbuf, permissions);
+}
+
+int UTIL_fchmod(const int fd, char const* filename, const stat_t* statbuf, mode_t permissions)
+{
+    stat_t localStatBuf;
+    UTIL_TRACE_CALL("UTIL_chmod(%s, %#4o)", filename, (unsigned)permissions);
+    if (statbuf == NULL) {
+        if (!UTIL_fstat(fd, filename, &localStatBuf)) {
+            UTIL_TRACE_RET(0);
+            return 0;
+        }
+        statbuf = &localStatBuf;
+    }
+    if (!UTIL_isRegularFileStat(statbuf)) {
+        UTIL_TRACE_RET(0);
+        return 0; /* pretend success, but don't change anything */
+    }
+#ifdef ZSTD_HAVE_FCHMOD
+    if (fd >= 0) {
+        int ret;
+        UTIL_TRACE_CALL("fchmod");
+        ret = fchmod(fd, permissions);
+        UTIL_TRACE_RET(ret);
+        UTIL_TRACE_RET(ret);
+        return ret;
+    } else
+#endif
+    {
+        int ret;
+        UTIL_TRACE_CALL("chmod");
+        ret = chmod(filename, permissions);
+        UTIL_TRACE_RET(ret);
+        UTIL_TRACE_RET(ret);
+        return ret;
+    }
+}
+
+/* set access and modification times */
+int UTIL_utime(const char* filename, const stat_t *statbuf)
+{
+    int ret;
+    UTIL_TRACE_CALL("UTIL_utime(%s)", filename);
+    /* We check that st_mtime is a macro here in order to give us confidence
+     * that struct stat has a struct timespec st_mtim member. We need this
+     * check because there are some platforms that claim to be POSIX 2008
+     * compliant but which do not have st_mtim... */
+    /* FreeBSD has implemented POSIX 2008 for a long time but still only
+     * advertises support for POSIX 2001. They have a version macro that
+     * lets us safely gate them in.
+     * See https://docs.freebsd.org/en/books/porters-handbook/versions/.
+     */
+#if ZSTD_USE_UTIMENSAT
+    {
+        /* (atime, mtime) */
+        struct timespec timebuf[2] = { {0, UTIME_NOW} };
+        timebuf[1] = statbuf->st_mtim;
+        ret = utimensat(AT_FDCWD, filename, timebuf, 0);
+    }
+#else
+    {
+        struct utimbuf timebuf;
+        timebuf.actime = time(NULL);
+        timebuf.modtime = statbuf->st_mtime;
+        ret = utime(filename, &timebuf);
+    }
+#endif
+    errno = 0;
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+int UTIL_setFileStat(const char *filename, const stat_t *statbuf)
+{
+    return UTIL_setFDStat(-1, filename, statbuf);
+}
+
+int UTIL_setFDStat(const int fd, const char *filename, const stat_t *statbuf)
+{
+    int res = 0;
+    stat_t curStatBuf;
+    UTIL_TRACE_CALL("UTIL_setFileStat(%d, %s)", fd, filename);
+
+    if (!UTIL_fstat(fd, filename, &curStatBuf) || !UTIL_isRegularFileStat(&curStatBuf)) {
+        UTIL_TRACE_RET(-1);
+        return -1;
+    }
+
+    /* Mimic gzip's behavior:
+     *
+     * "Change the group first, then the permissions, then the owner.
+     * That way, the permissions will be correct on systems that allow
+     * users to give away files, without introducing a security hole.
+     * Security depends on permissions not containing the setuid or
+     * setgid bits." */
+
+#if !defined(_WIN32)
+#ifdef ZSTD_HAVE_FCHOWN
+    if (fd >= 0) {
+        res += fchown(fd, -1, statbuf->st_gid);  /* Apply group ownership */
+    } else
+#endif
+    {
+        res += chown(filename, -1, statbuf->st_gid);  /* Apply group ownership */
+    }
+#endif
+
+    res += UTIL_fchmod(fd, filename, &curStatBuf, statbuf->st_mode & 0777);  /* Copy file permissions */
+
+#if !defined(_WIN32)
+#ifdef ZSTD_HAVE_FCHOWN
+    if (fd >= 0) {
+        res += fchown(fd, statbuf->st_uid, -1);  /* Apply user ownership */
+    } else
+#endif
+    {
+        res += chown(filename, statbuf->st_uid, -1);  /* Apply user ownership */
+    }
+#endif
+
+    errno = 0;
+    UTIL_TRACE_RET(-res);
+    return -res; /* number of errors is returned */
+}
+
+int UTIL_isDirectory(const char* infilename)
+{
+    stat_t statbuf;
+    int ret;
+    UTIL_TRACE_CALL("UTIL_isDirectory(%s)", infilename);
+    ret = UTIL_stat(infilename, &statbuf) && UTIL_isDirectoryStat(&statbuf);
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+int UTIL_isDirectoryStat(const stat_t* statbuf)
+{
+    int ret;
+    UTIL_TRACE_CALL("UTIL_isDirectoryStat()");
+#if defined(_MSC_VER)
+    ret = (statbuf->st_mode & _S_IFDIR) != 0;
+#else
+    ret = S_ISDIR(statbuf->st_mode) != 0;
+#endif
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+int UTIL_compareStr(const void *p1, const void *p2) {
+    return strcmp(* (char * const *) p1, * (char * const *) p2);
+}
+
+int UTIL_isSameFile(const char* fName1, const char* fName2)
+{
+    int ret;
+    assert(fName1 != NULL); assert(fName2 != NULL);
+    UTIL_TRACE_CALL("UTIL_isSameFile(%s, %s)", fName1, fName2);
+#if defined(_MSC_VER) || defined(_WIN32)
+    /* note : Visual does not support file identification by inode.
+     *        inode does not work on Windows, even with a posix layer, like msys2.
+     *        The following work-around is limited to detecting exact name repetition only,
+     *        aka `filename` is considered different from `subdir/../filename` */
+    ret = !strcmp(fName1, fName2);
+#else
+    {   stat_t file1Stat;
+        stat_t file2Stat;
+        ret =  UTIL_stat(fName1, &file1Stat)
+            && UTIL_stat(fName2, &file2Stat)
+            && UTIL_isSameFileStat(fName1, fName2, &file1Stat, &file2Stat);
+    }
+#endif
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+int UTIL_isSameFileStat(
+        const char* fName1, const char* fName2,
+        const stat_t* file1Stat, const stat_t* file2Stat)
+{
+    int ret;
+    assert(fName1 != NULL); assert(fName2 != NULL);
+    UTIL_TRACE_CALL("UTIL_isSameFileStat(%s, %s)", fName1, fName2);
+#if defined(_MSC_VER) || defined(_WIN32)
+    /* note : Visual does not support file identification by inode.
+     *        inode does not work on Windows, even with a posix layer, like msys2.
+     *        The following work-around is limited to detecting exact name repetition only,
+     *        aka `filename` is considered different from `subdir/../filename` */
+    (void)file1Stat;
+    (void)file2Stat;
+    ret = !strcmp(fName1, fName2);
+#else
+    {
+        ret =  (file1Stat->st_dev == file2Stat->st_dev)
+            && (file1Stat->st_ino == file2Stat->st_ino);
+    }
+#endif
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+/* UTIL_isFIFO : distinguish named pipes */
+int UTIL_isFIFO(const char* infilename)
+{
+    UTIL_TRACE_CALL("UTIL_isFIFO(%s)", infilename);
+/* macro guards, as defined in : https://linux.die.net/man/2/lstat */
+#if PLATFORM_POSIX_VERSION >= 200112L
+    {
+        stat_t statbuf;
+        if (UTIL_stat(infilename, &statbuf) && UTIL_isFIFOStat(&statbuf)) {
+            UTIL_TRACE_RET(1);
+            return 1;
+        }
+    }
+#endif
+    (void)infilename;
+    UTIL_TRACE_RET(0);
+    return 0;
+}
+
+/* UTIL_isFIFO : distinguish named pipes */
+int UTIL_isFIFOStat(const stat_t* statbuf)
+{
+/* macro guards, as defined in : https://linux.die.net/man/2/lstat */
+#if PLATFORM_POSIX_VERSION >= 200112L
+    if (S_ISFIFO(statbuf->st_mode)) return 1;
+#endif
+    (void)statbuf;
+    return 0;
+}
+
+/* process substitution */
+int UTIL_isFileDescriptorPipe(const char* filename)
+{
+    UTIL_TRACE_CALL("UTIL_isFileDescriptorPipe(%s)", filename);
+    /* Check if the filename is a /dev/fd/ path which indicates a file descriptor */
+    if (filename[0] == '/' && strncmp(filename, "/dev/fd/", 8) == 0) {
+        UTIL_TRACE_RET(1);
+        return 1;
+    }
+
+    /* Check for alternative process substitution formats on different systems */
+    if (filename[0] == '/' && strncmp(filename, "/proc/self/fd/", 14) == 0) {
+        UTIL_TRACE_RET(1);
+        return 1;
+    }
+
+    UTIL_TRACE_RET(0);
+    return 0; /* Not recognized as a file descriptor pipe */
+}
+
+/* UTIL_isBlockDevStat : distinguish named pipes */
+int UTIL_isBlockDevStat(const stat_t* statbuf)
+{
+/* macro guards, as defined in : https://linux.die.net/man/2/lstat */
+#if PLATFORM_POSIX_VERSION >= 200112L
+    if (S_ISBLK(statbuf->st_mode)) return 1;
+#endif
+    (void)statbuf;
+    return 0;
+}
+
+int UTIL_isLink(const char* infilename)
+{
+    UTIL_TRACE_CALL("UTIL_isLink(%s)", infilename);
+/* macro guards, as defined in : https://linux.die.net/man/2/lstat */
+#if PLATFORM_POSIX_VERSION >= 200112L
+    {
+        stat_t statbuf;
+        int const r = lstat(infilename, &statbuf);
+        if (!r && S_ISLNK(statbuf.st_mode)) {
+            UTIL_TRACE_RET(1);
+            return 1;
+        }
+    }
+#endif
+    (void)infilename;
+    UTIL_TRACE_RET(0);
+    return 0;
+}
+
+static int g_fakeStdinIsConsole = 0;
+static int g_fakeStderrIsConsole = 0;
+static int g_fakeStdoutIsConsole = 0;
+
+int UTIL_isConsole(FILE* file)
+{
+    int ret;
+    UTIL_TRACE_CALL("UTIL_isConsole(%d)", fileno(file));
+    if (file == stdin && g_fakeStdinIsConsole)
+        ret = 1;
+    else if (file == stderr && g_fakeStderrIsConsole)
+        ret = 1;
+    else if (file == stdout && g_fakeStdoutIsConsole)
+        ret = 1;
+    else
+        ret = IS_CONSOLE(file);
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+void UTIL_fakeStdinIsConsole(void)
+{
+    g_fakeStdinIsConsole = 1;
+}
+void UTIL_fakeStdoutIsConsole(void)
+{
+    g_fakeStdoutIsConsole = 1;
+}
+void UTIL_fakeStderrIsConsole(void)
+{
+    g_fakeStderrIsConsole = 1;
+}
+
+U64 UTIL_getFileSize(const char* infilename)
+{
+    stat_t statbuf;
+    UTIL_TRACE_CALL("UTIL_getFileSize(%s)", infilename);
+    if (!UTIL_stat(infilename, &statbuf)) {
+        UTIL_TRACE_RET(-1);
+        return UTIL_FILESIZE_UNKNOWN;
+    }
+    {
+        U64 const size = UTIL_getFileSizeStat(&statbuf);
+        UTIL_TRACE_RET((int)size);
+        return size;
+    }
+}
+
+U64 UTIL_getFileSizeStat(const stat_t* statbuf)
+{
+    if (!UTIL_isRegularFileStat(statbuf)) return UTIL_FILESIZE_UNKNOWN;
+#if defined(_MSC_VER)
+    if (!(statbuf->st_mode & S_IFREG)) return UTIL_FILESIZE_UNKNOWN;
+#elif defined(__MINGW32__) && defined (__MSVCRT__)
+    if (!(statbuf->st_mode & S_IFREG)) return UTIL_FILESIZE_UNKNOWN;
+#else
+    if (!S_ISREG(statbuf->st_mode)) return UTIL_FILESIZE_UNKNOWN;
+#endif
+    return (U64)statbuf->st_size;
+}
+
+UTIL_HumanReadableSize_t UTIL_makeHumanReadableSize(U64 size)
+{
+    UTIL_HumanReadableSize_t hrs;
+
+    if (g_utilDisplayLevel > 3) {
+        /* In verbose mode, do not scale sizes down, except in the case of
+         * values that exceed the integral precision of a double. */
+        if (size >= (1ull << 53)) {
+            hrs.value = (double)size / (1ull << 20);
+            hrs.suffix = " MiB";
+            /* At worst, a double representation of a maximal size will be
+             * accurate to better than tens of kilobytes. */
+            hrs.precision = 2;
+        } else {
+            hrs.value = (double)size;
+            hrs.suffix = " B";
+            hrs.precision = 0;
+        }
+    } else {
+        /* In regular mode, scale sizes down and use suffixes. */
+        if (size >= (1ull << 60)) {
+            hrs.value = (double)size / (1ull << 60);
+            hrs.suffix = " EiB";
+        } else if (size >= (1ull << 50)) {
+            hrs.value = (double)size / (1ull << 50);
+            hrs.suffix = " PiB";
+        } else if (size >= (1ull << 40)) {
+            hrs.value = (double)size / (1ull << 40);
+            hrs.suffix = " TiB";
+        } else if (size >= (1ull << 30)) {
+            hrs.value = (double)size / (1ull << 30);
+            hrs.suffix = " GiB";
+        } else if (size >= (1ull << 20)) {
+            hrs.value = (double)size / (1ull << 20);
+            hrs.suffix = " MiB";
+        } else if (size >= (1ull << 10)) {
+            hrs.value = (double)size / (1ull << 10);
+            hrs.suffix = " KiB";
+        } else {
+            hrs.value = (double)size;
+            hrs.suffix = " B";
+        }
+
+        if (hrs.value >= 100 || (U64)hrs.value == size) {
+            hrs.precision = 0;
+        } else if (hrs.value >= 10) {
+            hrs.precision = 1;
+        } else if (hrs.value > 1) {
+            hrs.precision = 2;
+        } else {
+            hrs.precision = 3;
+        }
+    }
+
+    return hrs;
+}
+
+U64 UTIL_getTotalFileSize(const char* const * fileNamesTable, unsigned nbFiles)
+{
+    U64 total = 0;
+    unsigned n;
+    UTIL_TRACE_CALL("UTIL_getTotalFileSize(%u)", nbFiles);
+    for (n=0; n<nbFiles; n++) {
+        U64 const size = UTIL_getFileSize(fileNamesTable[n]);
+        if (size == UTIL_FILESIZE_UNKNOWN) {
+            UTIL_TRACE_RET(-1);
+            return UTIL_FILESIZE_UNKNOWN;
+        }
+        total += size;
+    }
+    UTIL_TRACE_RET((int)total);
+    return total;
+}
+
+
+/* Read the entire content of a file into a buffer with progressive resizing */
+static char* UTIL_readFileContent(FILE* inFile, size_t* totalReadPtr)
+{
+    size_t bufSize = 64 KB;  /* Start with a reasonable buffer size */
+    size_t totalRead = 0;
+    size_t bytesRead = 0;
+    char* buf = (char*)malloc(bufSize);
+    if (buf == NULL) return NULL;
+
+
+    /* Read the file incrementally */
+    while ((bytesRead = fread(buf + totalRead, 1, bufSize - totalRead - 1, inFile)) > 0) {
+        totalRead += bytesRead;
+
+        /* If buffer is nearly full, expand it */
+        if (bufSize - totalRead < 1 KB) {
+            if (bufSize >= MAX_FILE_OF_FILE_NAMES_SIZE) {
+                /* Too large, abort */
+                free(buf);
+                return NULL;
+            }
+
+            {   size_t newBufSize = bufSize * 2;
+                if (newBufSize > MAX_FILE_OF_FILE_NAMES_SIZE)
+                    newBufSize = MAX_FILE_OF_FILE_NAMES_SIZE;
+
+                {   char* newBuf = (char*)realloc(buf, newBufSize);
+                    if (newBuf == NULL) {
+                        free(buf);
+                        return NULL;
+                    }
+
+                    buf = newBuf;
+                    bufSize = newBufSize;
+        }   }   }
+    }
+
+    /* Add null terminator to the end */
+    buf[totalRead] = '\0';
+    *totalReadPtr = totalRead;
+
+    return buf;
+}
+
+/* Process a buffer containing multiple lines and count the number of lines */
+static size_t UTIL_processLines(char* buffer, size_t bufferSize)
+{
+    size_t lineCount = 0;
+    size_t i = 0;
+
+    /* Convert newlines to null terminators and count lines */
+    while (i < bufferSize) {
+        if (buffer[i] == '\n') {
+            buffer[i] = '\0';  /* Replace newlines with null terminators */
+            lineCount++;
+        }
+        i++;
+    }
+
+    /* Count the last line if it doesn't end with a newline */
+    if (bufferSize > 0 && (i == 0 || buffer[i-1] != '\0')) {
+        lineCount++;
+    }
+
+    return lineCount;
+}
+
+/* Create an array of pointers to the lines in a buffer */
+static const char** UTIL_createLinePointers(char* buffer, size_t numLines, size_t bufferSize)
+{
+    size_t lineIndex = 0;
+    size_t pos = 0;
+    void* const bufferPtrs = malloc(numLines * sizeof(const char**));
+    const char** const linePointers = (const char**)bufferPtrs;
+    if (bufferPtrs == NULL) return NULL;
+
+    while (lineIndex < numLines && pos < bufferSize) {
+        size_t len = 0;
+        linePointers[lineIndex++] = buffer+pos;
+
+        /* Find the next null terminator, being careful not to go past the buffer */
+        while ((pos + len < bufferSize) && buffer[pos + len] != '\0') {
+            len++;
+        }
+
+        /* Move past this string and its null terminator */
+        pos += len;
+        if (pos < bufferSize) pos++;  /* Skip the null terminator if we're not at buffer end */
+    }
+
+    /* Verify we processed the expected number of lines */
+    if (lineIndex != numLines) {
+        /* Something went wrong - we didn't find as many lines as expected */
+        free(bufferPtrs);
+        return NULL;
+    }
+
+    return linePointers;
+}
+
+FileNamesTable*
+UTIL_createFileNamesTable_fromFileList(const char* fileList)
+{
+    stat_t statbuf;
+    char* buffer = NULL;
+    size_t numLines = 0;
+    size_t bufferSize = 0;
+
+    /* Check if the input is a valid file */
+    if (!UTIL_stat(fileList, &statbuf)) {
+        return NULL;
+    }
+
+    /* Check if the input is a supported type */
+    if (!UTIL_isRegularFileStat(&statbuf) &&
+        !UTIL_isFIFOStat(&statbuf) &&
+        !UTIL_isFileDescriptorPipe(fileList)) {
+        return NULL;
+    }
+
+    /* Open the input file */
+    {   FILE* const inFile = fopen(fileList, "rb");
+        if (inFile == NULL) return NULL;
+
+        /* Read the file content */
+        buffer = UTIL_readFileContent(inFile, &bufferSize);
+        fclose(inFile);
+    }
+
+    if (buffer == NULL) return NULL;
+
+    /* Process lines */
+    numLines = UTIL_processLines(buffer, bufferSize);
+    if (numLines == 0) {
+        free(buffer);
+        return NULL;
+    }
+
+    /* Create line pointers */
+    {   const char** linePointers = UTIL_createLinePointers(buffer, numLines, bufferSize);
+        if (linePointers == NULL) {
+            free(buffer);
+            return NULL;
+        }
+
+        /* Create the final table */
+        return UTIL_assembleFileNamesTable(linePointers, numLines, buffer);
+    }
+}
+
+
+static FileNamesTable*
+UTIL_assembleFileNamesTable2(const char** filenames, size_t tableSize, size_t tableCapacity, char* buf)
+{
+    FileNamesTable* const table = (FileNamesTable*) malloc(sizeof(*table));
+    CONTROL(table != NULL);
+    table->fileNames = filenames;
+    table->buf = buf;
+    table->tableSize = tableSize;
+    table->tableCapacity = tableCapacity;
+    return table;
+}
+
+FileNamesTable*
+UTIL_assembleFileNamesTable(const char** filenames, size_t tableSize, char* buf)
+{
+    return UTIL_assembleFileNamesTable2(filenames, tableSize, tableSize, buf);
+}
+
+void UTIL_freeFileNamesTable(FileNamesTable* table)
+{
+    if (table==NULL) return;
+    free((void*)table->fileNames);
+    free(table->buf);
+    free(table);
+}
+
+FileNamesTable* UTIL_allocateFileNamesTable(size_t tableSize)
+{
+    const char** const fnTable = (const char**)malloc(tableSize * sizeof(*fnTable));
+    FileNamesTable* fnt;
+    if (fnTable==NULL) return NULL;
+    fnt = UTIL_assembleFileNamesTable(fnTable, tableSize, NULL);
+    fnt->tableSize = 0;   /* the table is empty */
+    return fnt;
+}
+
+int UTIL_searchFileNamesTable(FileNamesTable* table, char const* name) {
+    size_t i;
+    for(i=0 ;i < table->tableSize; i++) {
+        if(!strcmp(table->fileNames[i], name)) {
+            return (int)i;
+        }
+    }
+    return -1;
+}
+
+void UTIL_refFilename(FileNamesTable* fnt, const char* filename)
+{
+    assert(fnt->tableSize < fnt->tableCapacity);
+    fnt->fileNames[fnt->tableSize] = filename;
+    fnt->tableSize++;
+}
+
+static size_t getTotalTableSize(FileNamesTable* table)
+{
+    size_t fnb, totalSize = 0;
+    for(fnb = 0 ; fnb < table->tableSize && table->fileNames[fnb] ; ++fnb) {
+        totalSize += strlen(table->fileNames[fnb]) + 1; /* +1 to add '\0' at the end of each fileName */
+    }
+    return totalSize;
+}
+
+FileNamesTable*
+UTIL_mergeFileNamesTable(FileNamesTable* table1, FileNamesTable* table2)
+{
+    unsigned newTableIdx = 0;
+    size_t pos = 0;
+    size_t newTotalTableSize;
+    char* buf;
+
+    FileNamesTable* const newTable = UTIL_assembleFileNamesTable(NULL, 0, NULL);
+    CONTROL( newTable != NULL );
+
+    newTotalTableSize = getTotalTableSize(table1) + getTotalTableSize(table2);
+
+    buf = (char*) calloc(newTotalTableSize, sizeof(*buf));
+    CONTROL ( buf != NULL );
+
+    newTable->buf = buf;
+    newTable->tableSize = table1->tableSize + table2->tableSize;
+    newTable->fileNames = (const char **) calloc(newTable->tableSize, sizeof(*(newTable->fileNames)));
+    CONTROL ( newTable->fileNames != NULL );
+
+    {   unsigned idx1;
+        for( idx1=0 ; (idx1 < table1->tableSize) && table1->fileNames[idx1] && (pos < newTotalTableSize); ++idx1, ++newTableIdx) {
+            size_t const curLen = strlen(table1->fileNames[idx1]);
+            memcpy(buf+pos, table1->fileNames[idx1], curLen);
+            assert(newTableIdx <= newTable->tableSize);
+            newTable->fileNames[newTableIdx] = buf+pos;
+            pos += curLen+1;
+    }   }
+
+    {   unsigned idx2;
+        for( idx2=0 ; (idx2 < table2->tableSize) && table2->fileNames[idx2] && (pos < newTotalTableSize) ; ++idx2, ++newTableIdx) {
+            size_t const curLen = strlen(table2->fileNames[idx2]);
+            memcpy(buf+pos, table2->fileNames[idx2], curLen);
+            assert(newTableIdx < newTable->tableSize);
+            newTable->fileNames[newTableIdx] = buf+pos;
+            pos += curLen+1;
+    }   }
+    assert(pos <= newTotalTableSize);
+    newTable->tableSize = newTableIdx;
+
+    UTIL_freeFileNamesTable(table1);
+    UTIL_freeFileNamesTable(table2);
+
+    return newTable;
+}
+
+#ifdef _WIN32
+static int UTIL_prepareFileList(const char* dirName,
+                                char** bufStart, size_t* pos,
+                                char** bufEnd, int followLinks)
+{
+    char* path;
+    size_t dirLength, pathLength;
+    int nbFiles = 0;
+    WIN32_FIND_DATAA cFile;
+    HANDLE hFile;
+
+    dirLength = strlen(dirName);
+    path = (char*) malloc(dirLength + 3);
+    if (!path) return 0;
+
+    memcpy(path, dirName, dirLength);
+    path[dirLength] = '\\';
+    path[dirLength+1] = '*';
+    path[dirLength+2] = 0;
+
+    hFile=FindFirstFileA(path, &cFile);
+    if (hFile == INVALID_HANDLE_VALUE) {
+        UTIL_DISPLAYLEVEL(1, "Cannot open directory '%s'\n", dirName);
+        return 0;
+    }
+    free(path);
+
+    do {
+        size_t const fnameLength = strlen(cFile.cFileName);
+        path = (char*) malloc(dirLength + fnameLength + 2);
+        if (!path) { FindClose(hFile); return 0; }
+        memcpy(path, dirName, dirLength);
+        path[dirLength] = '\\';
+        memcpy(path+dirLength+1, cFile.cFileName, fnameLength);
+        pathLength = dirLength+1+fnameLength;
+        path[pathLength] = 0;
+        if (cFile.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) {
+            if ( strcmp (cFile.cFileName, "..") == 0
+              || strcmp (cFile.cFileName, ".") == 0 )
+                continue;
+            /* Recursively call "UTIL_prepareFileList" with the new path. */
+            nbFiles += UTIL_prepareFileList(path, bufStart, pos, bufEnd, followLinks);
+            if (*bufStart == NULL) { free(path); FindClose(hFile); return 0; }
+        } else if ( (cFile.dwFileAttributes & FILE_ATTRIBUTE_NORMAL)
+                 || (cFile.dwFileAttributes & FILE_ATTRIBUTE_ARCHIVE)
+                 || (cFile.dwFileAttributes & FILE_ATTRIBUTE_COMPRESSED) ) {
+            if (*bufStart + *pos + pathLength >= *bufEnd) {
+                ptrdiff_t const newListSize = (*bufEnd - *bufStart) + LIST_SIZE_INCREASE;
+                *bufStart = (char*)UTIL_realloc(*bufStart, newListSize);
+                if (*bufStart == NULL) { free(path); FindClose(hFile); return 0; }
+                *bufEnd = *bufStart + newListSize;
+            }
+            if (*bufStart + *pos + pathLength < *bufEnd) {
+                memcpy(*bufStart + *pos, path, pathLength+1 /* include final \0 */);
+                *pos += pathLength + 1;
+                nbFiles++;
+        }   }
+        free(path);
+    } while (FindNextFileA(hFile, &cFile));
+
+    FindClose(hFile);
+    return nbFiles;
+}
+
+#elif defined(__linux__) || (PLATFORM_POSIX_VERSION >= 200112L)  /* opendir, readdir require POSIX.1-2001 */
+
+static int UTIL_prepareFileList(const char *dirName,
+                                char** bufStart, size_t* pos,
+                                char** bufEnd, int followLinks)
+{
+    DIR* dir;
+    struct dirent * entry;
+    size_t dirLength;
+    int nbFiles = 0;
+
+    if (!(dir = opendir(dirName))) {
+        UTIL_DISPLAYLEVEL(1, "Cannot open directory '%s': %s\n", dirName, strerror(errno));
+        return 0;
+    }
+
+    dirLength = strlen(dirName);
+    errno = 0;
+    while ((entry = readdir(dir)) != NULL) {
+        char* path;
+        size_t fnameLength, pathLength;
+        if (strcmp (entry->d_name, "..") == 0 ||
+            strcmp (entry->d_name, ".") == 0) continue;
+        fnameLength = strlen(entry->d_name);
+        path = (char*) malloc(dirLength + fnameLength + 2);
+        if (!path) { closedir(dir); return 0; }
+        memcpy(path, dirName, dirLength);
+
+        path[dirLength] = '/';
+        memcpy(path+dirLength+1, entry->d_name, fnameLength);
+        pathLength = dirLength+1+fnameLength;
+        path[pathLength] = 0;
+
+        if (!followLinks && UTIL_isLink(path)) {
+            UTIL_DISPLAYLEVEL(2, "Warning : %s is a symbolic link, ignoring\n", path);
+            free(path);
+            continue;
+        }
+
+        if (UTIL_isDirectory(path)) {
+            nbFiles += UTIL_prepareFileList(path, bufStart, pos, bufEnd, followLinks);  /* Recursively call "UTIL_prepareFileList" with the new path. */
+            if (*bufStart == NULL) { free(path); closedir(dir); return 0; }
+        } else {
+            if (*bufStart + *pos + pathLength >= *bufEnd) {
+                ptrdiff_t newListSize = (*bufEnd - *bufStart) + LIST_SIZE_INCREASE;
+                assert(newListSize >= 0);
+                *bufStart = (char*)UTIL_realloc(*bufStart, (size_t)newListSize);
+                if (*bufStart != NULL) {
+                    *bufEnd = *bufStart + newListSize;
+                } else {
+                    free(path); closedir(dir); return 0;
+                }
+            }
+            if (*bufStart + *pos + pathLength < *bufEnd) {
+                memcpy(*bufStart + *pos, path, pathLength + 1);  /* with final \0 */
+                *pos += pathLength + 1;
+                nbFiles++;
+        }   }
+        free(path);
+        errno = 0; /* clear errno after UTIL_isDirectory, UTIL_prepareFileList */
+    }
+
+    if (errno != 0) {
+        UTIL_DISPLAYLEVEL(1, "readdir(%s) error: %s \n", dirName, strerror(errno));
+        free(*bufStart);
+        *bufStart = NULL;
+    }
+    closedir(dir);
+    return nbFiles;
+}
+
+#else
+
+static int UTIL_prepareFileList(const char *dirName,
+                                char** bufStart, size_t* pos,
+                                char** bufEnd, int followLinks)
+{
+    (void)bufStart; (void)bufEnd; (void)pos; (void)followLinks;
+    UTIL_DISPLAYLEVEL(1, "Directory %s ignored (compiled without _WIN32 or _POSIX_C_SOURCE) \n", dirName);
+    return 0;
+}
+
+#endif /* #ifdef _WIN32 */
+
+int UTIL_isCompressedFile(const char *inputName, const char *extensionList[])
+{
+  const char* ext = UTIL_getFileExtension(inputName);
+  while(*extensionList!=NULL)
+  {
+    const int isCompressedExtension = strcmp(ext,*extensionList);
+    if(isCompressedExtension==0)
+      return 1;
+    ++extensionList;
+  }
+   return 0;
+}
+
+/*Utility function to get file extension from file */
+const char* UTIL_getFileExtension(const char* infilename)
+{
+   const char* extension = strrchr(infilename, '.');
+   if(!extension || extension==infilename) return "";
+   return extension;
+}
+
+static int pathnameHas2Dots(const char *pathname)
+{
+    /* We need to figure out whether any ".." present in the path is a whole
+     * path token, which is the case if it is bordered on both sides by either
+     * the beginning/end of the path or by a directory separator.
+     */
+    const char *needle = pathname;
+    while (1) {
+        needle = strstr(needle, "..");
+
+        if (needle == NULL) {
+            return 0;
+        }
+
+        if ((needle == pathname || needle[-1] == PATH_SEP)
+         && (needle[2] == '\0' || needle[2] == PATH_SEP)) {
+            return 1;
+        }
+
+        /* increment so we search for the next match */
+        needle++;
+    };
+    return 0;
+}
+
+static int isFileNameValidForMirroredOutput(const char *filename)
+{
+    return !pathnameHas2Dots(filename);
+}
+
+
+#define DIR_DEFAULT_MODE 0755
+static mode_t getDirMode(const char *dirName)
+{
+    stat_t st;
+    if (!UTIL_stat(dirName, &st)) {
+        UTIL_DISPLAY("zstd: failed to get DIR stats %s: %s\n", dirName, strerror(errno));
+        return DIR_DEFAULT_MODE;
+    }
+    if (!UTIL_isDirectoryStat(&st)) {
+        UTIL_DISPLAY("zstd: expected directory: %s\n", dirName);
+        return DIR_DEFAULT_MODE;
+    }
+    return st.st_mode;
+}
+
+static int makeDir(const char *dir, mode_t mode)
+{
+#if defined(_MSC_VER) || defined(__MINGW32__) || defined (__MSVCRT__)
+    int ret = _mkdir(dir);
+    (void) mode;
+#else
+    int ret = mkdir(dir, mode);
+#endif
+    if (ret != 0) {
+        if (errno == EEXIST)
+            return 0;
+        UTIL_DISPLAY("zstd: failed to create DIR %s: %s\n", dir, strerror(errno));
+    }
+    return ret;
+}
+
+/* this function requires a mutable input string */
+static void convertPathnameToDirName(char *pathname)
+{
+    size_t len = 0;
+    char* pos = NULL;
+    /* get dir name from pathname similar to 'dirname()' */
+    assert(pathname != NULL);
+
+    /* remove trailing '/' chars */
+    len = strlen(pathname);
+    assert(len > 0);
+    while (pathname[len] == PATH_SEP) {
+        pathname[len] = '\0';
+        len--;
+    }
+    if (len == 0) return;
+
+    /* if input is a single file, return '.' instead. i.e.
+     * "xyz/abc/file.txt" => "xyz/abc"
+       "./file.txt"       => "."
+       "file.txt"         => "."
+     */
+    pos = strrchr(pathname, PATH_SEP);
+    if (pos == NULL) {
+        pathname[0] = '.';
+        pathname[1] = '\0';
+    } else {
+        *pos = '\0';
+    }
+}
+
+/* pathname must be valid */
+static const char* trimLeadingRootChar(const char *pathname)
+{
+    assert(pathname != NULL);
+    if (pathname[0] == PATH_SEP)
+        return pathname + 1;
+    return pathname;
+}
+
+/* pathname must be valid */
+static const char* trimLeadingCurrentDirConst(const char *pathname)
+{
+    assert(pathname != NULL);
+    if ((pathname[0] == '.') && (pathname[1] == PATH_SEP))
+        return pathname + 2;
+    return pathname;
+}
+
+static char*
+trimLeadingCurrentDir(char *pathname)
+{
+    /* 'union charunion' can do const-cast without compiler warning */
+    union charunion {
+        char *chr;
+        const char* cchr;
+    } ptr;
+    ptr.cchr = trimLeadingCurrentDirConst(pathname);
+    return ptr.chr;
+}
+
+/* remove leading './' or '/' chars here */
+static const char * trimPath(const char *pathname)
+{
+    return trimLeadingRootChar(
+            trimLeadingCurrentDirConst(pathname));
+}
+
+static char* mallocAndJoin2Dir(const char *dir1, const char *dir2)
+{
+    assert(dir1 != NULL && dir2 != NULL);
+    {   const size_t dir1Size = strlen(dir1);
+        const size_t dir2Size = strlen(dir2);
+        char *outDirBuffer, *buffer;
+
+        outDirBuffer = (char *) malloc(dir1Size + dir2Size + 2);
+        CONTROL(outDirBuffer != NULL);
+
+        memcpy(outDirBuffer, dir1, dir1Size);
+        outDirBuffer[dir1Size] = '\0';
+
+        buffer = outDirBuffer + dir1Size;
+        if (dir1Size > 0 && *(buffer - 1) != PATH_SEP) {
+            *buffer = PATH_SEP;
+            buffer++;
+        }
+        memcpy(buffer, dir2, dir2Size);
+        buffer[dir2Size] = '\0';
+
+        return outDirBuffer;
+    }
+}
+
+/* this function will return NULL if input srcFileName is not valid name for mirrored output path */
+char* UTIL_createMirroredDestDirName(const char* srcFileName, const char* outDirRootName)
+{
+    char* pathname = NULL;
+    if (!isFileNameValidForMirroredOutput(srcFileName))
+        return NULL;
+
+    pathname = mallocAndJoin2Dir(outDirRootName, trimPath(srcFileName));
+
+    convertPathnameToDirName(pathname);
+    return pathname;
+}
+
+static int
+mirrorSrcDir(char* srcDirName, const char* outDirName)
+{
+    mode_t srcMode;
+    int status = 0;
+    char* newDir = mallocAndJoin2Dir(outDirName, trimPath(srcDirName));
+    if (!newDir)
+        return -ENOMEM;
+
+    srcMode = getDirMode(srcDirName);
+    status = makeDir(newDir, srcMode);
+    free(newDir);
+    return status;
+}
+
+static int
+mirrorSrcDirRecursive(char* srcDirName, const char* outDirName)
+{
+    int status = 0;
+    char* pp = trimLeadingCurrentDir(srcDirName);
+    char* sp = NULL;
+
+    while ((sp = strchr(pp, PATH_SEP)) != NULL) {
+        if (sp != pp) {
+            *sp = '\0';
+            status = mirrorSrcDir(srcDirName, outDirName);
+            if (status != 0)
+                return status;
+            *sp = PATH_SEP;
+        }
+        pp = sp + 1;
+    }
+    status = mirrorSrcDir(srcDirName, outDirName);
+    return status;
+}
+
+static void
+makeMirroredDestDirsWithSameSrcDirMode(char** srcDirNames, unsigned nbFile, const char* outDirName)
+{
+    unsigned int i = 0;
+    for (i = 0; i < nbFile; i++)
+        mirrorSrcDirRecursive(srcDirNames[i], outDirName);
+}
+
+static int
+firstIsParentOrSameDirOfSecond(const char* firstDir, const char* secondDir)
+{
+    size_t firstDirLen  = strlen(firstDir),
+           secondDirLen = strlen(secondDir);
+    return firstDirLen <= secondDirLen &&
+           (secondDir[firstDirLen] == PATH_SEP || secondDir[firstDirLen] == '\0') &&
+           0 == strncmp(firstDir, secondDir, firstDirLen);
+}
+
+static int compareDir(const void* pathname1, const void* pathname2) {
+    /* sort it after remove the leading '/'  or './'*/
+    const char* s1 = trimPath(*(char * const *) pathname1);
+    const char* s2 = trimPath(*(char * const *) pathname2);
+    return strcmp(s1, s2);
+}
+
+static void
+makeUniqueMirroredDestDirs(char** srcDirNames, unsigned nbFile, const char* outDirName)
+{
+    unsigned int i = 0, uniqueDirNr = 0;
+    char** uniqueDirNames = NULL;
+
+    if (nbFile == 0)
+        return;
+
+    uniqueDirNames = (char** ) malloc(nbFile * sizeof (char *));
+    CONTROL(uniqueDirNames != NULL);
+
+    /* if dirs is "a/b/c" and "a/b/c/d", we only need call:
+     * we just need "a/b/c/d" */
+    qsort((void *)srcDirNames, nbFile, sizeof(char*), compareDir);
+
+    uniqueDirNr = 1;
+    uniqueDirNames[uniqueDirNr - 1] = srcDirNames[0];
+    for (i = 1; i < nbFile; i++) {
+        char* prevDirName = srcDirNames[i - 1];
+        char* currDirName = srcDirNames[i];
+
+        /* note: we always compare trimmed path, i.e.:
+         * src dir of "./foo" and "/foo" will be both saved into:
+         * "outDirName/foo/" */
+        if (!firstIsParentOrSameDirOfSecond(trimPath(prevDirName),
+                                            trimPath(currDirName)))
+            uniqueDirNr++;
+
+        /* we need to maintain original src dir name instead of trimmed
+         * dir, so we can retrieve the original src dir's mode_t */
+        uniqueDirNames[uniqueDirNr - 1] = currDirName;
+    }
+
+    makeMirroredDestDirsWithSameSrcDirMode(uniqueDirNames, uniqueDirNr, outDirName);
+
+    free(uniqueDirNames);
+}
+
+static void
+makeMirroredDestDirs(char** srcFileNames, unsigned nbFile, const char* outDirName)
+{
+    unsigned int i = 0;
+    for (i = 0; i < nbFile; ++i)
+        convertPathnameToDirName(srcFileNames[i]);
+    makeUniqueMirroredDestDirs(srcFileNames, nbFile, outDirName);
+}
+
+void UTIL_mirrorSourceFilesDirectories(const char** inFileNames, unsigned int nbFile, const char* outDirName)
+{
+    unsigned int i = 0, validFilenamesNr = 0;
+    char** srcFileNames = (char **) malloc(nbFile * sizeof (char *));
+    CONTROL(srcFileNames != NULL);
+
+    /* check input filenames is valid */
+    for (i = 0; i < nbFile; ++i) {
+        if (isFileNameValidForMirroredOutput(inFileNames[i])) {
+            char* fname = STRDUP(inFileNames[i]);
+            CONTROL(fname != NULL);
+            srcFileNames[validFilenamesNr++] = fname;
+        }
+    }
+
+    if (validFilenamesNr > 0) {
+        makeDir(outDirName, DIR_DEFAULT_MODE);
+        makeMirroredDestDirs(srcFileNames, validFilenamesNr, outDirName);
+    }
+
+    for (i = 0; i < validFilenamesNr; i++)
+        free(srcFileNames[i]);
+    free(srcFileNames);
+}
+
+FileNamesTable*
+UTIL_createExpandedFNT(const char* const* inputNames, size_t nbIfns, int followLinks)
+{
+    unsigned nbFiles;
+    char* buf = (char*)malloc(LIST_SIZE_INCREASE);
+    char* bufend = buf + LIST_SIZE_INCREASE;
+
+    if (!buf) return NULL;
+
+    {   size_t ifnNb, pos;
+        for (ifnNb=0, pos=0, nbFiles=0; ifnNb<nbIfns; ifnNb++) {
+            if (!UTIL_isDirectory(inputNames[ifnNb])) {
+                size_t const len = strlen(inputNames[ifnNb]);
+                if (buf + pos + len >= bufend) {
+                    ptrdiff_t newListSize = (bufend - buf) + LIST_SIZE_INCREASE;
+                    assert(newListSize >= 0);
+                    buf = (char*)UTIL_realloc(buf, (size_t)newListSize);
+                    if (!buf) return NULL;
+                    bufend = buf + newListSize;
+                }
+                if (buf + pos + len < bufend) {
+                    memcpy(buf+pos, inputNames[ifnNb], len+1);  /* including final \0 */
+                    pos += len + 1;
+                    nbFiles++;
+                }
+            } else {
+                nbFiles += (unsigned)UTIL_prepareFileList(inputNames[ifnNb], &buf, &pos, &bufend, followLinks);
+                if (buf == NULL) return NULL;
+    }   }   }
+
+    /* note : even if nbFiles==0, function returns a valid, though empty, FileNamesTable* object */
+
+    {   size_t ifnNb, pos;
+        size_t const fntCapacity = nbFiles + 1;  /* minimum 1, allows adding one reference, typically stdin */
+        const char** const fileNamesTable = (const char**)malloc(fntCapacity * sizeof(*fileNamesTable));
+        if (!fileNamesTable) { free(buf); return NULL; }
+
+        for (ifnNb = 0, pos = 0; ifnNb < nbFiles; ifnNb++) {
+            fileNamesTable[ifnNb] = buf + pos;
+            if (buf + pos > bufend) { free(buf); free((void*)fileNamesTable); return NULL; }
+            pos += strlen(fileNamesTable[ifnNb]) + 1;
+        }
+        return UTIL_assembleFileNamesTable2(fileNamesTable, nbFiles, fntCapacity, buf);
+    }
+}
+
+
+void UTIL_expandFNT(FileNamesTable** fnt, int followLinks)
+{
+    FileNamesTable* const newFNT = UTIL_createExpandedFNT((*fnt)->fileNames, (*fnt)->tableSize, followLinks);
+    CONTROL(newFNT != NULL);
+    UTIL_freeFileNamesTable(*fnt);
+    *fnt = newFNT;
+}
+
+FileNamesTable* UTIL_createFNT_fromROTable(const char** filenames, size_t nbFilenames)
+{
+    size_t const sizeof_FNTable = nbFilenames * sizeof(*filenames);
+    const char** const newFNTable = (const char**)malloc(sizeof_FNTable);
+    if (newFNTable==NULL) return NULL;
+    memcpy((void*)newFNTable, filenames, sizeof_FNTable);  /* void* : mitigate a Visual compiler bug or limitation */
+    return UTIL_assembleFileNamesTable(newFNTable, nbFilenames, NULL);
+}
+
+
+/*-****************************************
+*  count the number of cores
+******************************************/
+
+#if defined(_WIN32) || defined(WIN32)
+
+#include <windows.h>
+
+typedef BOOL(WINAPI* LPFN_GLPI)(PSYSTEM_LOGICAL_PROCESSOR_INFORMATION, PDWORD);
+
+DWORD CountSetBits(ULONG_PTR bitMask)
+{
+    DWORD LSHIFT = sizeof(ULONG_PTR)*8 - 1;
+    DWORD bitSetCount = 0;
+    ULONG_PTR bitTest = (ULONG_PTR)1 << LSHIFT;
+    DWORD i;
+
+    for (i = 0; i <= LSHIFT; ++i)
+    {
+        bitSetCount += ((bitMask & bitTest)?1:0);
+        bitTest/=2;
+    }
+
+    return bitSetCount;
+}
+
+int UTIL_countCores(int logical)
+{
+    static int numCores = 0;
+    if (numCores != 0) return numCores;
+
+    {   LPFN_GLPI glpi;
+        BOOL done = FALSE;
+        PSYSTEM_LOGICAL_PROCESSOR_INFORMATION buffer = NULL;
+        PSYSTEM_LOGICAL_PROCESSOR_INFORMATION ptr = NULL;
+        DWORD returnLength = 0;
+        size_t byteOffset = 0;
+
+#if defined(_MSC_VER)
+/* Visual Studio does not like the following cast */
+#   pragma warning( disable : 4054 )  /* conversion from function ptr to data ptr */
+#   pragma warning( disable : 4055 )  /* conversion from data ptr to function ptr */
+#endif
+        glpi = (LPFN_GLPI)(void*)GetProcAddress(GetModuleHandle(TEXT("kernel32")),
+                                               "GetLogicalProcessorInformation");
+
+        if (glpi == NULL) {
+            goto failed;
+        }
+
+        while(!done) {
+            DWORD rc = glpi(buffer, &returnLength);
+            if (FALSE == rc) {
+                if (GetLastError() == ERROR_INSUFFICIENT_BUFFER) {
+                    if (buffer)
+                        free(buffer);
+                    buffer = (PSYSTEM_LOGICAL_PROCESSOR_INFORMATION)malloc(returnLength);
+
+                    if (buffer == NULL) {
+                        perror("zstd");
+                        exit(1);
+                    }
+                } else {
+                    /* some other error */
+                    goto failed;
+                }
+            } else {
+                done = TRUE;
+        }   }
+
+        ptr = buffer;
+
+        while (byteOffset + sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION) <= returnLength) {
+
+            if (ptr->Relationship == RelationProcessorCore) {
+                if (logical)
+                    numCores += CountSetBits(ptr->ProcessorMask);
+                else
+                    numCores++;
+            }
+
+            ptr++;
+            byteOffset += sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION);
+        }
+
+        free(buffer);
+
+        return numCores;
+    }
+
+failed:
+    /* try to fall back on GetSystemInfo */
+    {   SYSTEM_INFO sysinfo;
+        GetSystemInfo(&sysinfo);
+        numCores = sysinfo.dwNumberOfProcessors;
+        if (numCores == 0) numCores = 1; /* just in case */
+    }
+    return numCores;
+}
+
+#elif defined(__APPLE__)
+
+#include <sys/sysctl.h>
+
+/* Use apple-provided syscall
+ * see: man 3 sysctl */
+int UTIL_countCores(int logical)
+{
+    static S32 numCores = 0; /* apple specifies int32_t */
+    if (numCores != 0) return numCores;
+
+    {   size_t size = sizeof(S32);
+        int const ret = sysctlbyname(logical ? "hw.logicalcpu" : "hw.physicalcpu", &numCores, &size, NULL, 0);
+        if (ret != 0) {
+            if (errno == ENOENT) {
+                /* entry not present, fall back on 1 */
+                numCores = 1;
+            } else {
+                perror("zstd: can't get number of cpus");
+                exit(1);
+            }
+        }
+
+        return numCores;
+    }
+}
+
+#elif defined(__linux__)
+
+/* parse /proc/cpuinfo
+ * siblings / cpu cores should give hyperthreading ratio
+ * otherwise fall back on sysconf */
+int UTIL_countCores(int logical)
+{
+    static int numCores = 0;
+
+    if (numCores != 0) return numCores;
+
+    numCores = (int)sysconf(_SC_NPROCESSORS_ONLN);
+    if (numCores == -1) {
+        /* value not queryable, fall back on 1 */
+        return numCores = 1;
+    }
+
+    /* try to determine if there's hyperthreading */
+    {   FILE* const cpuinfo = fopen("/proc/cpuinfo", "r");
+#define BUF_SIZE 80
+        char buff[BUF_SIZE];
+
+        int siblings = 0;
+        int cpu_cores = 0;
+        int ratio = 1;
+
+        if (cpuinfo == NULL) {
+            /* fall back on the sysconf value */
+            return numCores;
+        }
+
+        /* assume the cpu cores/siblings values will be constant across all
+         * present processors */
+        while (!feof(cpuinfo)) {
+            if (fgets(buff, BUF_SIZE, cpuinfo) != NULL) {
+                if (strncmp(buff, "siblings", 8) == 0) {
+                    const char* const sep = strchr(buff, ':');
+                    if (sep == NULL || *sep == '\0') {
+                        /* formatting was broken? */
+                        goto failed;
+                    }
+
+                    siblings = atoi(sep + 1);
+                }
+                if (strncmp(buff, "cpu cores", 9) == 0) {
+                    const char* const sep = strchr(buff, ':');
+                    if (sep == NULL || *sep == '\0') {
+                        /* formatting was broken? */
+                        goto failed;
+                    }
+
+                    cpu_cores = atoi(sep + 1);
+                }
+            } else if (ferror(cpuinfo)) {
+                /* fall back on the sysconf value */
+                goto failed;
+        }   }
+        if (siblings && cpu_cores && siblings > cpu_cores) {
+            ratio = siblings / cpu_cores;
+        }
+
+        if (ratio && numCores > ratio && !logical) {
+            numCores = numCores / ratio;
+        }
+
+failed:
+        fclose(cpuinfo);
+        return numCores;
+    }
+}
+
+#elif defined(__FreeBSD__)
+
+#include <sys/sysctl.h>
+
+/* Use physical core sysctl when available
+ * see: man 4 smp, man 3 sysctl */
+int UTIL_countCores(int logical)
+{
+    static int numCores = 0; /* freebsd sysctl is native int sized */
+#if __FreeBSD_version >= 1300008
+    static int perCore = 1;
+#endif
+    if (numCores != 0) return numCores;
+
+#if __FreeBSD_version >= 1300008
+    {   size_t size = sizeof(numCores);
+        int ret = sysctlbyname("kern.smp.cores", &numCores, &size, NULL, 0);
+        if (ret == 0) {
+            if (logical) {
+                ret = sysctlbyname("kern.smp.threads_per_core", &perCore, &size, NULL, 0);
+                /* default to physical cores if logical cannot be read */
+                if (ret == 0)
+                    numCores *= perCore;
+            }
+
+            return numCores;
+        }
+        if (errno != ENOENT) {
+            perror("zstd: can't get number of cpus");
+            exit(1);
+        }
+        /* sysctl not present, fall through to older sysconf method */
+    }
+#else
+    /* suppress unused parameter warning */
+    (void) logical;
+#endif
+
+    numCores = (int)sysconf(_SC_NPROCESSORS_ONLN);
+    if (numCores == -1) {
+        /* value not queryable, fall back on 1 */
+        numCores = 1;
+    }
+    return numCores;
+}
+
+#elif defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) || defined(__CYGWIN__)
+
+/* Use POSIX sysconf
+ * see: man 3 sysconf */
+int UTIL_countCores(int logical)
+{
+    static int numCores = 0;
+
+    /* suppress unused parameter warning */
+    (void)logical;
+
+    if (numCores != 0) return numCores;
+
+    numCores = (int)sysconf(_SC_NPROCESSORS_ONLN);
+    if (numCores == -1) {
+        /* value not queryable, fall back on 1 */
+        return numCores = 1;
+    }
+    return numCores;
+}
+
+#else
+
+int UTIL_countCores(int logical)
+{
+    /* suppress unused parameter warning */
+    (void)logical;
+
+    /* assume 1 */
+    return 1;
+}
+
+#endif
+
+int UTIL_countPhysicalCores(void)
+{
+    return UTIL_countCores(0);
+}
+
+int UTIL_countLogicalCores(void)
+{
+    return UTIL_countCores(1);
+}
diff --git a/programs/zstdcli.c b/programs/zstdcli.c
new file mode 100644
index 0000000..fa7ea37
--- /dev/null
+++ b/programs/zstdcli.c
@@ -0,0 +1,1668 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/*-************************************
+*  Dependencies
+**************************************/
+#include "platform.h" /* PLATFORM_POSIX_VERSION */
+#include "util.h"     /* UTIL_HAS_CREATEFILELIST, UTIL_createFileList, UTIL_isConsole */
+#include <stdlib.h>   /* getenv */
+#include <string.h>   /* strcmp, strlen */
+#include <stdio.h>    /* fprintf(), stdin, stdout, stderr */
+#include <assert.h>   /* assert */
+
+#include "fileio.h"   /* stdinmark, stdoutmark, ZSTD_EXTENSION */
+#ifndef ZSTD_NOBENCH
+#  include "benchzstd.h"  /* BMK_benchFilesAdvanced */
+#endif
+#ifndef ZSTD_NODICT
+#  include "dibio.h"  /* ZDICT_cover_params_t, DiB_trainFromFiles() */
+#endif
+#ifndef ZSTD_NOTRACE
+#  include "zstdcli_trace.h"
+#endif
+#include "../lib/zstd.h"  /* ZSTD_VERSION_STRING, ZSTD_minCLevel, ZSTD_maxCLevel */
+#include "fileio_asyncio.h"
+#include "fileio_common.h"
+
+/*-************************************
+*  Tuning parameters
+**************************************/
+#ifndef ZSTDCLI_CLEVEL_DEFAULT
+#  define ZSTDCLI_CLEVEL_DEFAULT 3
+#endif
+
+#ifndef ZSTDCLI_CLEVEL_MAX
+#  define ZSTDCLI_CLEVEL_MAX 19   /* without using --ultra */
+#endif
+
+#ifndef ZSTDCLI_NBTHREADS_DEFAULT
+#define ZSTDCLI_NBTHREADS_DEFAULT (unsigned)(MAX(1, MIN(4, UTIL_countLogicalCores() / 4)))
+#endif
+
+
+
+/*-************************************
+*  Constants
+**************************************/
+#define COMPRESSOR_NAME "Zstandard CLI"
+#ifndef ZSTD_VERSION
+#  define ZSTD_VERSION "v" ZSTD_VERSION_STRING
+#endif
+#define AUTHOR "Yann Collet"
+#define WELCOME_MESSAGE "*** %s (%i-bit) %s, by %s ***\n", COMPRESSOR_NAME, (int)(sizeof(size_t)*8), ZSTD_VERSION, AUTHOR
+
+#define ZSTD_ZSTDMT "zstdmt"
+#define ZSTD_UNZSTD "unzstd"
+#define ZSTD_CAT "zstdcat"
+#define ZSTD_ZCAT "zcat"
+#define ZSTD_GZ "gzip"
+#define ZSTD_GUNZIP "gunzip"
+#define ZSTD_GZCAT "gzcat"
+#define ZSTD_LZMA "lzma"
+#define ZSTD_UNLZMA "unlzma"
+#define ZSTD_XZ "xz"
+#define ZSTD_UNXZ "unxz"
+#define ZSTD_LZ4 "lz4"
+#define ZSTD_UNLZ4 "unlz4"
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define DISPLAY_LEVEL_DEFAULT 2
+
+static const char*    g_defaultDictName = "dictionary";
+static const unsigned g_defaultMaxDictSize = 110 KB;
+static const int      g_defaultDictCLevel = 3;
+static const unsigned g_defaultSelectivityLevel = 9;
+static const unsigned g_defaultMaxWindowLog = 27;
+#define OVERLAP_LOG_DEFAULT 9999
+#define LDM_PARAM_DEFAULT 9999  /* Default for parameters where 0 is valid */
+static U32 g_overlapLog = OVERLAP_LOG_DEFAULT;
+static U32 g_ldmHashLog = 0;
+static U32 g_ldmMinMatch = 0;
+static U32 g_ldmHashRateLog = LDM_PARAM_DEFAULT;
+static U32 g_ldmBucketSizeLog = LDM_PARAM_DEFAULT;
+
+
+#define DEFAULT_ACCEL 1
+#define NBWORKERS_AUTOCPU 0
+
+typedef enum { cover, fastCover, legacy } dictType;
+
+/*-************************************
+*  Display Macros
+**************************************/
+#undef DISPLAYLEVEL
+#define DISPLAYLEVEL(l, ...) { if (g_displayLevel>=l) { DISPLAY(__VA_ARGS__); } }
+static int g_displayLevel = DISPLAY_LEVEL_DEFAULT;   /* 0 : no display,  1: errors,  2 : + result + interaction + warnings,  3 : + progression,  4 : + information */
+
+
+/*-************************************
+*  Check Version (when CLI linked to dynamic library)
+**************************************/
+
+/* Due to usage of experimental symbols and capabilities by the CLI,
+ * the CLI must be linked against a dynamic library of same version */
+static void checkLibVersion(void)
+{
+    if (strcmp(ZSTD_VERSION_STRING, ZSTD_versionString())) {
+        DISPLAYLEVEL(1, "Error : incorrect library version (expecting : %s ; actual : %s ) \n",
+                    ZSTD_VERSION_STRING, ZSTD_versionString());
+        DISPLAYLEVEL(1, "Please update library to version %s, or use stand-alone zstd binary \n",
+                    ZSTD_VERSION_STRING);
+        exit(1);
+    }
+}
+
+
+/*! exeNameMatch() :
+    @return : a non-zero value if exeName matches test, excluding the extension
+   */
+static int exeNameMatch(const char* exeName, const char* test)
+{
+    return !strncmp(exeName, test, strlen(test)) &&
+        (exeName[strlen(test)] == '\0' || exeName[strlen(test)] == '.');
+}
+
+/*-************************************
+*  Command Line
+**************************************/
+/* print help either in `stderr` or `stdout` depending on originating request
+ * error (badUsage) => stderr
+ * help (usageAdvanced) => stdout
+ */
+static void usage(FILE* f, const char* programName)
+{
+    DISPLAY_F(f, "Compress or decompress the INPUT file(s); reads from STDIN if INPUT is `-` or not provided.\n\n");
+    DISPLAY_F(f, "Usage: %s [OPTIONS...] [INPUT... | -] [-o OUTPUT]\n\n", programName);
+    DISPLAY_F(f, "Options:\n");
+    DISPLAY_F(f, "  -o OUTPUT                     Write output to a single file, OUTPUT.\n");
+    DISPLAY_F(f, "  -k, --keep                    Preserve INPUT file(s). [Default] \n");
+    DISPLAY_F(f, "  --rm                          Remove INPUT file(s) after successful (de)compression.\n");
+#ifdef ZSTD_GZCOMPRESS
+    if (exeNameMatch(programName, ZSTD_GZ)) {     /* behave like gzip */
+        DISPLAY_F(f, "  -n, --no-name                 Do not store original filename when compressing.\n\n");
+    }
+#endif
+    DISPLAY_F(f, "\n");
+#ifndef ZSTD_NOCOMPRESS
+    DISPLAY_F(f, "  -#                            Desired compression level, where `#` is a number between 1 and %d;\n", ZSTDCLI_CLEVEL_MAX);
+    DISPLAY_F(f, "                                lower numbers provide faster compression, higher numbers yield\n");
+    DISPLAY_F(f, "                                better compression ratios. [Default: %d]\n\n", ZSTDCLI_CLEVEL_DEFAULT);
+#endif
+#ifndef ZSTD_NODECOMPRESS
+    DISPLAY_F(f, "  -d, --decompress              Perform decompression.\n");
+#endif
+    DISPLAY_F(f, "  -D DICT                       Use DICT as the dictionary for compression or decompression.\n\n");
+    DISPLAY_F(f, "  -f, --force                   Disable input and output checks. Allows overwriting existing files,\n");
+    DISPLAY_F(f, "                                receiving input from the console, printing output to STDOUT, and\n");
+    DISPLAY_F(f, "                                operating on links, block devices, etc. Unrecognized formats will be\n");
+    DISPLAY_F(f, "                                passed-through through as-is.\n\n");
+
+    DISPLAY_F(f, "  -h                            Display short usage and exit.\n");
+    DISPLAY_F(f, "  -H, --help                    Display full help and exit.\n");
+    DISPLAY_F(f, "  -V, --version                 Display the program version and exit.\n");
+    DISPLAY_F(f, "\n");
+}
+
+static void usageAdvanced(const char* programName)
+{
+    DISPLAYOUT(WELCOME_MESSAGE);
+    DISPLAYOUT("\n");
+    usage(stdout, programName);
+    DISPLAYOUT("Advanced options:\n");
+    DISPLAYOUT("  -c, --stdout                  Write to STDOUT (even if it is a console) and keep the INPUT file(s).\n\n");
+
+    DISPLAYOUT("  -v, --verbose                 Enable verbose output; pass multiple times to increase verbosity.\n");
+    DISPLAYOUT("  -q, --quiet                   Suppress warnings; pass twice to suppress errors.\n");
+#ifndef ZSTD_NOTRACE
+    DISPLAYOUT("  --trace LOG                   Log tracing information to LOG.\n");
+#endif
+    DISPLAYOUT("\n");
+    DISPLAYOUT("  --[no-]progress               Forcibly show/hide the progress counter. NOTE: Any (de)compressed\n");
+    DISPLAYOUT("                                output to terminal will mix with progress counter text.\n\n");
+
+#ifdef UTIL_HAS_CREATEFILELIST
+    DISPLAYOUT("  -r                            Operate recursively on directories.\n");
+    DISPLAYOUT("  --filelist LIST               Read a list of files to operate on from LIST.\n");
+    DISPLAYOUT("  --output-dir-flat DIR         Store processed files in DIR.\n");
+#endif
+
+#ifdef UTIL_HAS_MIRRORFILELIST
+    DISPLAYOUT("  --output-dir-mirror DIR       Store processed files in DIR, respecting original directory structure.\n");
+#endif
+    if (AIO_supported())
+        DISPLAYOUT("  --[no-]asyncio                Use asynchronous IO. [Default: Enabled]\n");
+
+    DISPLAYOUT("\n");
+#ifndef ZSTD_NOCOMPRESS
+    DISPLAYOUT("  --[no-]check                  Add XXH64 integrity checksums during compression. [Default: Add, Validate]\n");
+#ifndef ZSTD_NODECOMPRESS
+    DISPLAYOUT("                                If `-d` is present, ignore/validate checksums during decompression.\n");
+#endif
+#else
+#ifdef ZSTD_NOCOMPRESS
+    DISPLAYOUT("  --[no-]check                  Ignore/validate checksums during decompression. [Default: Validate]");
+#endif
+#endif /* ZSTD_NOCOMPRESS */
+
+    DISPLAYOUT("\n");
+    DISPLAYOUT("  --                            Treat remaining arguments after `--` as files.\n");
+
+#ifndef ZSTD_NOCOMPRESS
+    DISPLAYOUT("\n");
+    DISPLAYOUT("Advanced compression options:\n");
+    DISPLAYOUT("  --ultra                       Enable levels beyond %i, up to %i; requires more memory.\n", ZSTDCLI_CLEVEL_MAX, ZSTD_maxCLevel());
+    DISPLAYOUT("  --fast[=#]                    Use to very fast compression levels. [Default: %u]\n", 1);
+#ifdef ZSTD_GZCOMPRESS
+    if (exeNameMatch(programName, ZSTD_GZ)) {     /* behave like gzip */
+        DISPLAYOUT("  --best                        Compatibility alias for `-9`.\n");
+    }
+#endif
+    DISPLAYOUT("  --adapt                       Dynamically adapt compression level to I/O conditions.\n");
+    DISPLAYOUT("  --long[=#]                    Enable long distance matching with window log #. [Default: %u]\n", g_defaultMaxWindowLog);
+    DISPLAYOUT("  --patch-from=REF              Use REF as the reference point for Zstandard's diff engine. \n");
+    DISPLAYOUT("  --patch-apply                 Equivalent for `-d --patch-from` \n\n");
+# ifdef ZSTD_MULTITHREAD
+    DISPLAYOUT("  -T#                           Spawn # compression threads. [Default: 1; pass 0 for core count.]\n");
+    DISPLAYOUT("  --single-thread               Share a single thread for I/O and compression (slightly different than `-T1`).\n");
+    DISPLAYOUT("  --auto-threads={physical|logical}\n");
+    DISPLAYOUT("                                Use physical/logical cores when using `-T0`. [Default: Physical]\n\n");
+    DISPLAYOUT("  --jobsize=#                   Set job size to #. [Default: 0 (automatic)]\n");
+    DISPLAYOUT("  --rsyncable                   Compress using a rsync-friendly method (`--jobsize=#` sets unit size). \n");
+    DISPLAYOUT("\n");
+# endif
+    DISPLAYOUT("  --exclude-compressed          Only compress files that are not already compressed.\n\n");
+
+    DISPLAYOUT("  --stream-size=#               Specify size of streaming input from STDIN.\n");
+    DISPLAYOUT("  --size-hint=#                 Optimize compression parameters for streaming input of approximately size #.\n");
+    DISPLAYOUT("  --target-compressed-block-size=#\n");
+    DISPLAYOUT("                                Generate compressed blocks of approximately # size.\n\n");
+    DISPLAYOUT("  --no-dictID                   Don't write `dictID` into the header (dictionary compression only).\n");
+    DISPLAYOUT("  --[no-]compress-literals      Force (un)compressed literals.\n");
+    DISPLAYOUT("  --[no-]row-match-finder       Explicitly enable/disable the fast, row-based matchfinder for\n");
+    DISPLAYOUT("                                the 'greedy', 'lazy', and 'lazy2' strategies.\n");
+
+    DISPLAYOUT("\n");
+    DISPLAYOUT("  --format=zstd                 Compress files to the `.zst` format. [Default]\n");
+    DISPLAYOUT("  --[no-]mmap-dict              Memory-map dictionary file rather than mallocing and loading all at once\n");
+#ifdef ZSTD_GZCOMPRESS
+    DISPLAYOUT("  --format=gzip                 Compress files to the `.gz` format.\n");
+#endif
+#ifdef ZSTD_LZMACOMPRESS
+    DISPLAYOUT("  --format=xz                   Compress files to the `.xz` format.\n");
+    DISPLAYOUT("  --format=lzma                 Compress files to the `.lzma` format.\n");
+#endif
+#ifdef ZSTD_LZ4COMPRESS
+    DISPLAYOUT( "  --format=lz4                 Compress files to the `.lz4` format.\n");
+#endif
+#endif  /* !ZSTD_NOCOMPRESS */
+
+#ifndef ZSTD_NODECOMPRESS
+    DISPLAYOUT("\n");
+    DISPLAYOUT("Advanced decompression options:\n");
+    DISPLAYOUT("  -l                            Print information about Zstandard-compressed files.\n");
+    DISPLAYOUT("  --test                        Test compressed file integrity.\n");
+    DISPLAYOUT("  -M#                           Set the memory usage limit to # megabytes.\n");
+# if ZSTD_SPARSE_DEFAULT
+    DISPLAYOUT("  --[no-]sparse                 Enable sparse mode. [Default: Enabled for files, disabled for STDOUT.]\n");
+# else
+    DISPLAYOUT("  --[no-]sparse                 Enable sparse mode. [Default: Disabled]\n");
+# endif
+    {
+        char const* passThroughDefault = "Disabled";
+        if (exeNameMatch(programName, ZSTD_CAT) ||
+            exeNameMatch(programName, ZSTD_ZCAT) ||
+            exeNameMatch(programName, ZSTD_GZCAT)) {
+            passThroughDefault = "Enabled";
+        }
+        DISPLAYOUT("  --[no-]pass-through           Pass through uncompressed files as-is. [Default: %s]\n", passThroughDefault);
+    }
+#endif  /* ZSTD_NODECOMPRESS */
+
+#ifndef ZSTD_NODICT
+    DISPLAYOUT("\n");
+    DISPLAYOUT("Dictionary builder:\n");
+    DISPLAYOUT("  --train                       Create a dictionary from a training set of files.\n\n");
+    DISPLAYOUT("  --train-cover[=k=#,d=#,steps=#,split=#,shrink[=#]]\n");
+    DISPLAYOUT("                                Use the cover algorithm (with optional arguments).\n");
+    DISPLAYOUT("  --train-fastcover[=k=#,d=#,f=#,steps=#,split=#,accel=#,shrink[=#]]\n");
+    DISPLAYOUT("                                Use the fast cover algorithm (with optional arguments).\n\n");
+    DISPLAYOUT("  --train-legacy[=s=#]          Use the legacy algorithm with selectivity #. [Default: %u]\n", g_defaultSelectivityLevel);
+    DISPLAYOUT("  -o NAME                       Use NAME as dictionary name. [Default: %s]\n", g_defaultDictName);
+    DISPLAYOUT("  --maxdict=#                   Limit dictionary to specified size #. [Default: %u]\n", g_defaultMaxDictSize);
+    DISPLAYOUT("  --dictID=#                    Force dictionary ID to #. [Default: Random]\n");
+#endif
+
+#ifndef ZSTD_NOBENCH
+    DISPLAYOUT("\n");
+    DISPLAYOUT("Benchmark options:\n");
+    DISPLAYOUT("  -b#                           Perform benchmarking with compression level #. [Default: %d]\n", ZSTDCLI_CLEVEL_DEFAULT);
+    DISPLAYOUT("  -e#                           Test all compression levels up to #; starting level is `-b#`. [Default: 1]\n");
+    DISPLAYOUT("  -i#                           Set the minimum evaluation to time # seconds. [Default: 3]\n");
+    DISPLAYOUT("  --split=#                     Split input into independent chunks of size #. [Default: No chunking]\n");
+    DISPLAYOUT("  -S                            Output one benchmark result per input file. [Default: Consolidated result]\n");
+    DISPLAYOUT("  -D dictionary                 Benchmark using dictionary \n");
+    DISPLAYOUT("  --priority=rt                 Set process priority to real-time.\n");
+#endif
+
+}
+
+static void badUsage(const char* programName, const char* parameter)
+{
+    DISPLAYLEVEL(1, "Incorrect parameter: %s \n", parameter);
+    if (g_displayLevel >= 2) usage(stderr, programName);
+}
+
+static void waitEnter(void)
+{
+    int unused;
+    DISPLAY("Press enter to continue... \n");
+    unused = getchar();
+    (void)unused;
+}
+
+static const char* lastNameFromPath(const char* path)
+{
+    const char* name = path;
+    if (strrchr(name, '/')) name = strrchr(name, '/') + 1;
+    if (strrchr(name, '\\')) name = strrchr(name, '\\') + 1; /* windows */
+    return name;
+}
+
+static void errorOut(const char* msg)
+{
+    DISPLAYLEVEL(1, "%s \n", msg); exit(1);
+}
+
+/*! readU32FromCharChecked() :
+ * @return 0 if success, and store the result in *value.
+ *  allows and interprets K, KB, KiB, M, MB and MiB suffix.
+ *  Will also modify `*stringPtr`, advancing it to position where it stopped reading.
+ * @return 1 if an overflow error occurs */
+static int readU32FromCharChecked(const char** stringPtr, unsigned* value)
+{
+    unsigned result = 0;
+    while ((**stringPtr >='0') && (**stringPtr <='9')) {
+        unsigned const max = ((unsigned)(-1)) / 10;
+        unsigned last = result;
+        if (result > max) return 1; /* overflow error */
+        result *= 10;
+        result += (unsigned)(**stringPtr - '0');
+        if (result < last) return 1; /* overflow error */
+        (*stringPtr)++ ;
+    }
+    if ((**stringPtr=='K') || (**stringPtr=='M')) {
+        unsigned const maxK = ((unsigned)(-1)) >> 10;
+        if (result > maxK) return 1; /* overflow error */
+        result <<= 10;
+        if (**stringPtr=='M') {
+            if (result > maxK) return 1; /* overflow error */
+            result <<= 10;
+        }
+        (*stringPtr)++;  /* skip `K` or `M` */
+        if (**stringPtr=='i') (*stringPtr)++;
+        if (**stringPtr=='B') (*stringPtr)++;
+    }
+    *value = result;
+    return 0;
+}
+
+/*! readU32FromChar() :
+ * @return : unsigned integer value read from input in `char` format.
+ *  allows and interprets K, KB, KiB, M, MB and MiB suffix.
+ *  Will also modify `*stringPtr`, advancing it to position where it stopped reading.
+ *  Note : function will exit() program if digit sequence overflows */
+static unsigned readU32FromChar(const char** stringPtr) {
+    static const char errorMsg[] = "error: numeric value overflows 32-bit unsigned int";
+    unsigned result;
+    if (readU32FromCharChecked(stringPtr, &result)) { errorOut(errorMsg); }
+    return result;
+}
+
+/*! readIntFromChar() :
+ * @return : signed integer value read from input in `char` format.
+ *  allows and interprets K, KB, KiB, M, MB and MiB suffix.
+ *  Will also modify `*stringPtr`, advancing it to position where it stopped reading.
+ *  Note : function will exit() program if digit sequence overflows */
+static int readIntFromChar(const char** stringPtr) {
+    static const char errorMsg[] = "error: numeric value overflows 32-bit int";
+    int sign = 1;
+    unsigned result;
+    if (**stringPtr=='-') {
+        (*stringPtr)++;
+        sign = -1;
+    }
+    if (readU32FromCharChecked(stringPtr, &result)) { errorOut(errorMsg); }
+    return (int) result * sign;
+}
+
+/*! readSizeTFromCharChecked() :
+ * @return 0 if success, and store the result in *value.
+ *  allows and interprets K, KB, KiB, M, MB and MiB suffix.
+ *  Will also modify `*stringPtr`, advancing it to position where it stopped reading.
+ * @return 1 if an overflow error occurs */
+static int readSizeTFromCharChecked(const char** stringPtr, size_t* value)
+{
+    size_t result = 0;
+    while ((**stringPtr >='0') && (**stringPtr <='9')) {
+        size_t const max = ((size_t)(-1)) / 10;
+        size_t last = result;
+        if (result > max) return 1; /* overflow error */
+        result *= 10;
+        result += (size_t)(**stringPtr - '0');
+        if (result < last) return 1; /* overflow error */
+        (*stringPtr)++ ;
+    }
+    if ((**stringPtr=='K') || (**stringPtr=='M')) {
+        size_t const maxK = ((size_t)(-1)) >> 10;
+        if (result > maxK) return 1; /* overflow error */
+        result <<= 10;
+        if (**stringPtr=='M') {
+            if (result > maxK) return 1; /* overflow error */
+            result <<= 10;
+        }
+        (*stringPtr)++;  /* skip `K` or `M` */
+        if (**stringPtr=='i') (*stringPtr)++;
+        if (**stringPtr=='B') (*stringPtr)++;
+    }
+    *value = result;
+    return 0;
+}
+
+/*! readSizeTFromChar() :
+ * @return : size_t value read from input in `char` format.
+ *  allows and interprets K, KB, KiB, M, MB and MiB suffix.
+ *  Will also modify `*stringPtr`, advancing it to position where it stopped reading.
+ *  Note : function will exit() program if digit sequence overflows */
+static size_t readSizeTFromChar(const char** stringPtr) {
+    static const char errorMsg[] = "error: numeric value overflows size_t";
+    size_t result;
+    if (readSizeTFromCharChecked(stringPtr, &result)) { errorOut(errorMsg); }
+    return result;
+}
+
+/** longCommandWArg() :
+ *  check if *stringPtr is the same as longCommand.
+ *  If yes, @return 1 and advances *stringPtr to the position which immediately follows longCommand.
+ * @return 0 and doesn't modify *stringPtr otherwise.
+ */
+static int longCommandWArg(const char** stringPtr, const char* longCommand)
+{
+    size_t const comSize = strlen(longCommand);
+    int const result = !strncmp(*stringPtr, longCommand, comSize);
+    if (result) *stringPtr += comSize;
+    return result;
+}
+
+
+#ifndef ZSTD_NODICT
+
+static const unsigned kDefaultRegression = 1;
+/**
+ * parseCoverParameters() :
+ * reads cover parameters from *stringPtr (e.g. "--train-cover=k=48,d=8,steps=32") into *params
+ * @return 1 means that cover parameters were correct
+ * @return 0 in case of malformed parameters
+ */
+static unsigned parseCoverParameters(const char* stringPtr, ZDICT_cover_params_t* params)
+{
+    memset(params, 0, sizeof(*params));
+    for (; ;) {
+        if (longCommandWArg(&stringPtr, "k=")) { params->k = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "d=")) { params->d = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "steps=")) { params->steps = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "split=")) {
+          unsigned splitPercentage = readU32FromChar(&stringPtr);
+          params->splitPoint = (double)splitPercentage / 100.0;
+          if (stringPtr[0]==',') { stringPtr++; continue; } else break;
+        }
+        if (longCommandWArg(&stringPtr, "shrink")) {
+          params->shrinkDictMaxRegression = kDefaultRegression;
+          params->shrinkDict = 1;
+          if (stringPtr[0]=='=') {
+            stringPtr++;
+            params->shrinkDictMaxRegression = readU32FromChar(&stringPtr);
+          }
+          if (stringPtr[0]==',') {
+            stringPtr++;
+            continue;
+          }
+          else break;
+        }
+        return 0;
+    }
+    if (stringPtr[0] != 0) return 0;
+    DISPLAYLEVEL(4, "cover: k=%u\nd=%u\nsteps=%u\nsplit=%u\nshrink%u\n", params->k, params->d, params->steps, (unsigned)(params->splitPoint * 100), params->shrinkDictMaxRegression);
+    return 1;
+}
+
+/**
+ * parseFastCoverParameters() :
+ * reads fastcover parameters from *stringPtr (e.g. "--train-fastcover=k=48,d=8,f=20,steps=32,accel=2") into *params
+ * @return 1 means that fastcover parameters were correct
+ * @return 0 in case of malformed parameters
+ */
+static unsigned parseFastCoverParameters(const char* stringPtr, ZDICT_fastCover_params_t* params)
+{
+    memset(params, 0, sizeof(*params));
+    for (; ;) {
+        if (longCommandWArg(&stringPtr, "k=")) { params->k = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "d=")) { params->d = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "f=")) { params->f = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "steps=")) { params->steps = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "accel=")) { params->accel = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "split=")) {
+          unsigned splitPercentage = readU32FromChar(&stringPtr);
+          params->splitPoint = (double)splitPercentage / 100.0;
+          if (stringPtr[0]==',') { stringPtr++; continue; } else break;
+        }
+        if (longCommandWArg(&stringPtr, "shrink")) {
+          params->shrinkDictMaxRegression = kDefaultRegression;
+          params->shrinkDict = 1;
+          if (stringPtr[0]=='=') {
+            stringPtr++;
+            params->shrinkDictMaxRegression = readU32FromChar(&stringPtr);
+          }
+          if (stringPtr[0]==',') {
+            stringPtr++;
+            continue;
+          }
+          else break;
+        }
+        return 0;
+    }
+    if (stringPtr[0] != 0) return 0;
+    DISPLAYLEVEL(4, "cover: k=%u\nd=%u\nf=%u\nsteps=%u\nsplit=%u\naccel=%u\nshrink=%u\n", params->k, params->d, params->f, params->steps, (unsigned)(params->splitPoint * 100), params->accel, params->shrinkDictMaxRegression);
+    return 1;
+}
+
+/**
+ * parseLegacyParameters() :
+ * reads legacy dictionary builder parameters from *stringPtr (e.g. "--train-legacy=selectivity=8") into *selectivity
+ * @return 1 means that legacy dictionary builder parameters were correct
+ * @return 0 in case of malformed parameters
+ */
+static unsigned parseLegacyParameters(const char* stringPtr, unsigned* selectivity)
+{
+    if (!longCommandWArg(&stringPtr, "s=") && !longCommandWArg(&stringPtr, "selectivity=")) { return 0; }
+    *selectivity = readU32FromChar(&stringPtr);
+    if (stringPtr[0] != 0) return 0;
+    DISPLAYLEVEL(4, "legacy: selectivity=%u\n", *selectivity);
+    return 1;
+}
+
+static ZDICT_cover_params_t defaultCoverParams(void)
+{
+    ZDICT_cover_params_t params;
+    memset(&params, 0, sizeof(params));
+    params.d = 8;
+    params.steps = 4;
+    params.splitPoint = 1.0;
+    params.shrinkDict = 0;
+    params.shrinkDictMaxRegression = kDefaultRegression;
+    return params;
+}
+
+static ZDICT_fastCover_params_t defaultFastCoverParams(void)
+{
+    ZDICT_fastCover_params_t params;
+    memset(&params, 0, sizeof(params));
+    params.d = 8;
+    params.f = 20;
+    params.steps = 4;
+    params.splitPoint = 0.75; /* different from default splitPoint of cover */
+    params.accel = DEFAULT_ACCEL;
+    params.shrinkDict = 0;
+    params.shrinkDictMaxRegression = kDefaultRegression;
+    return params;
+}
+#endif
+
+
+/** parseAdaptParameters() :
+ *  reads adapt parameters from *stringPtr (e.g. "--adapt=min=1,max=19) and store them into adaptMinPtr and adaptMaxPtr.
+ *  Both adaptMinPtr and adaptMaxPtr must be already allocated and correctly initialized.
+ *  There is no guarantee that any of these values will be updated.
+ *  @return 1 means that parsing was successful,
+ *  @return 0 in case of malformed parameters
+ */
+static unsigned parseAdaptParameters(const char* stringPtr, int* adaptMinPtr, int* adaptMaxPtr)
+{
+    for ( ; ;) {
+        if (longCommandWArg(&stringPtr, "min=")) { *adaptMinPtr = readIntFromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "max=")) { *adaptMaxPtr = readIntFromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        DISPLAYLEVEL(4, "invalid compression parameter \n");
+        return 0;
+    }
+    if (stringPtr[0] != 0) return 0; /* check the end of string */
+    if (*adaptMinPtr > *adaptMaxPtr) {
+        DISPLAYLEVEL(4, "incoherent adaptation limits \n");
+        return 0;
+    }
+    return 1;
+}
+
+
+/** parseCompressionParameters() :
+ *  reads compression parameters from *stringPtr (e.g. "--zstd=wlog=23,clog=23,hlog=22,slog=6,mml=3,tlen=48,strat=6") into *params
+ *  @return 1 means that compression parameters were correct
+ *  @return 0 in case of malformed parameters
+ */
+static unsigned parseCompressionParameters(const char* stringPtr, ZSTD_compressionParameters* params)
+{
+    for ( ; ;) {
+        if (longCommandWArg(&stringPtr, "windowLog=") || longCommandWArg(&stringPtr, "wlog=")) { params->windowLog = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "chainLog=") || longCommandWArg(&stringPtr, "clog=")) { params->chainLog = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "hashLog=") || longCommandWArg(&stringPtr, "hlog=")) { params->hashLog = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "searchLog=") || longCommandWArg(&stringPtr, "slog=")) { params->searchLog = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "minMatch=") || longCommandWArg(&stringPtr, "mml=")) { params->minMatch = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "targetLength=") || longCommandWArg(&stringPtr, "tlen=")) { params->targetLength = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "strategy=") || longCommandWArg(&stringPtr, "strat=")) { params->strategy = (ZSTD_strategy)(readU32FromChar(&stringPtr)); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "overlapLog=") || longCommandWArg(&stringPtr, "ovlog=")) { g_overlapLog = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "ldmHashLog=") || longCommandWArg(&stringPtr, "lhlog=")) { g_ldmHashLog = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "ldmMinMatch=") || longCommandWArg(&stringPtr, "lmml=")) { g_ldmMinMatch = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "ldmBucketSizeLog=") || longCommandWArg(&stringPtr, "lblog=")) { g_ldmBucketSizeLog = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "ldmHashRateLog=") || longCommandWArg(&stringPtr, "lhrlog=")) { g_ldmHashRateLog = readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        DISPLAYLEVEL(4, "invalid compression parameter \n");
+        return 0;
+    }
+
+    if (stringPtr[0] != 0) return 0; /* check the end of string */
+    return 1;
+}
+
+static void setMaxCompression(ZSTD_compressionParameters* params)
+{
+    params->windowLog = ZSTD_WINDOWLOG_MAX;
+    params->chainLog = ZSTD_CHAINLOG_MAX;
+    params->hashLog = ZSTD_HASHLOG_MAX;
+    params->searchLog = ZSTD_SEARCHLOG_MAX;
+    params->minMatch = ZSTD_MINMATCH_MIN;
+    params->targetLength = ZSTD_TARGETLENGTH_MAX;
+    params->strategy = ZSTD_STRATEGY_MAX;
+    g_overlapLog = ZSTD_OVERLAPLOG_MAX;
+    g_ldmHashLog = ZSTD_LDM_HASHLOG_MAX;
+    g_ldmHashRateLog = 0; /* automatically derived */
+    g_ldmMinMatch = 16; /* heuristic */
+    g_ldmBucketSizeLog = ZSTD_LDM_BUCKETSIZELOG_MAX;
+}
+
+static void printVersion(void)
+{
+    if (g_displayLevel < DISPLAY_LEVEL_DEFAULT) {
+        DISPLAYOUT("%s\n", ZSTD_VERSION_STRING);
+        return;
+    }
+
+    DISPLAYOUT(WELCOME_MESSAGE);
+    if (g_displayLevel >= 3) {
+    /* format support */
+        DISPLAYOUT("*** supports: zstd");
+    #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>0) && (ZSTD_LEGACY_SUPPORT<8)
+        DISPLAYOUT(", zstd legacy v0.%d+", ZSTD_LEGACY_SUPPORT);
+    #endif
+    #ifdef ZSTD_GZCOMPRESS
+        DISPLAYOUT(", gzip");
+    #endif
+    #ifdef ZSTD_LZ4COMPRESS
+        DISPLAYOUT(", lz4");
+    #endif
+    #ifdef ZSTD_LZMACOMPRESS
+        DISPLAYOUT(", lzma, xz ");
+    #endif
+        DISPLAYOUT("\n");
+        if (g_displayLevel >= 4) {
+            /* library versions */
+            DISPLAYOUT("zlib version %s\n", FIO_zlibVersion());
+            DISPLAYOUT("lz4 version %s\n", FIO_lz4Version());
+            DISPLAYOUT("lzma version %s\n", FIO_lzmaVersion());
+
+        #ifdef ZSTD_MULTITHREAD
+            DISPLAYOUT("supports Multithreading \n");
+        #else
+            DISPLAYOUT("single-thread operations only \n");
+        #endif
+
+            /* posix support */
+        #ifdef _POSIX_C_SOURCE
+            DISPLAYOUT("_POSIX_C_SOURCE defined: %ldL\n", (long) _POSIX_C_SOURCE);
+        #endif
+        #ifdef _POSIX_VERSION
+            DISPLAYOUT("_POSIX_VERSION defined: %ldL \n", (long) _POSIX_VERSION);
+        #endif
+        #ifdef PLATFORM_POSIX_VERSION
+            DISPLAYOUT("PLATFORM_POSIX_VERSION defined: %ldL\n", (long) PLATFORM_POSIX_VERSION);
+        #endif
+
+            if (!ZSTD_isDeterministicBuild()) {
+                DISPLAYOUT("non-deterministic build\n");
+    }   }   }
+}
+
+#define ZSTD_NB_STRATEGIES 9
+static const char* ZSTD_strategyMap[ZSTD_NB_STRATEGIES + 1] = { "", "ZSTD_fast",
+                "ZSTD_dfast", "ZSTD_greedy", "ZSTD_lazy", "ZSTD_lazy2", "ZSTD_btlazy2",
+                "ZSTD_btopt", "ZSTD_btultra", "ZSTD_btultra2"};
+
+#ifndef ZSTD_NOCOMPRESS
+
+static void printDefaultCParams(const char* filename, const char* dictFileName, int cLevel) {
+    unsigned long long fileSize = UTIL_getFileSize(filename);
+    const size_t dictSize = dictFileName != NULL ? (size_t)UTIL_getFileSize(dictFileName) : 0;
+    const ZSTD_compressionParameters cParams = ZSTD_getCParams(cLevel, fileSize, dictSize);
+    if (fileSize != UTIL_FILESIZE_UNKNOWN) DISPLAY("%s (%llu bytes)\n", filename, fileSize);
+    else DISPLAY("%s (src size unknown)\n", filename);
+    DISPLAY(" - windowLog     : %u\n", cParams.windowLog);
+    DISPLAY(" - chainLog      : %u\n", cParams.chainLog);
+    DISPLAY(" - hashLog       : %u\n", cParams.hashLog);
+    DISPLAY(" - searchLog     : %u\n", cParams.searchLog);
+    DISPLAY(" - minMatch      : %u\n", cParams.minMatch);
+    DISPLAY(" - targetLength  : %u\n", cParams.targetLength);
+    assert(cParams.strategy < ZSTD_NB_STRATEGIES + 1);
+    DISPLAY(" - strategy      : %s (%u)\n", ZSTD_strategyMap[(int)cParams.strategy], (unsigned)cParams.strategy);
+}
+
+static void printActualCParams(const char* filename, const char* dictFileName, int cLevel, const ZSTD_compressionParameters* cParams) {
+    unsigned long long fileSize = UTIL_getFileSize(filename);
+    const size_t dictSize = dictFileName != NULL ? (size_t)UTIL_getFileSize(dictFileName) : 0;
+    ZSTD_compressionParameters actualCParams = ZSTD_getCParams(cLevel, fileSize, dictSize);
+    assert(g_displayLevel >= 4);
+    actualCParams.windowLog = cParams->windowLog == 0 ? actualCParams.windowLog : cParams->windowLog;
+    actualCParams.chainLog = cParams->chainLog == 0 ? actualCParams.chainLog : cParams->chainLog;
+    actualCParams.hashLog = cParams->hashLog == 0 ? actualCParams.hashLog : cParams->hashLog;
+    actualCParams.searchLog = cParams->searchLog == 0 ? actualCParams.searchLog : cParams->searchLog;
+    actualCParams.minMatch = cParams->minMatch == 0 ? actualCParams.minMatch : cParams->minMatch;
+    actualCParams.targetLength = cParams->targetLength == 0 ? actualCParams.targetLength : cParams->targetLength;
+    actualCParams.strategy = cParams->strategy == 0 ? actualCParams.strategy : cParams->strategy;
+    DISPLAY("--zstd=wlog=%d,clog=%d,hlog=%d,slog=%d,mml=%d,tlen=%d,strat=%d\n",
+            actualCParams.windowLog, actualCParams.chainLog, actualCParams.hashLog, actualCParams.searchLog,
+            actualCParams.minMatch, actualCParams.targetLength, actualCParams.strategy);
+}
+
+#endif
+
+/* Environment variables for parameter setting */
+#define ENV_CLEVEL "ZSTD_CLEVEL"
+#define ENV_NBWORKERS "ZSTD_NBTHREADS"    /* takes lower precedence than directly specifying -T# in the CLI */
+
+/* pick up environment variable */
+static int init_cLevel(void) {
+    const char* const env = getenv(ENV_CLEVEL);
+    if (env != NULL) {
+        const char* ptr = env;
+        int sign = 1;
+        if (*ptr == '-') {
+            sign = -1;
+            ptr++;
+        } else if (*ptr == '+') {
+            ptr++;
+        }
+
+        if ((*ptr>='0') && (*ptr<='9')) {
+            unsigned absLevel;
+            if (readU32FromCharChecked(&ptr, &absLevel)) {
+                DISPLAYLEVEL(2, "Ignore environment variable setting %s=%s: numeric value too large \n", ENV_CLEVEL, env);
+                return ZSTDCLI_CLEVEL_DEFAULT;
+            } else if (*ptr == 0) {
+                return sign * (int)absLevel;
+        }   }
+
+        DISPLAYLEVEL(2, "Ignore environment variable setting %s=%s: not a valid integer value \n", ENV_CLEVEL, env);
+    }
+
+    return ZSTDCLI_CLEVEL_DEFAULT;
+}
+
+static unsigned init_nbWorkers(void) {
+#ifdef ZSTD_MULTITHREAD
+    const char* const env = getenv(ENV_NBWORKERS);
+    if (env != NULL) {
+        const char* ptr = env;
+        if ((*ptr>='0') && (*ptr<='9')) {
+            unsigned nbThreads;
+            if (readU32FromCharChecked(&ptr, &nbThreads)) {
+                DISPLAYLEVEL(2, "Ignore environment variable setting %s=%s: numeric value too large \n", ENV_NBWORKERS, env);
+                return ZSTDCLI_NBTHREADS_DEFAULT;
+            } else if (*ptr == 0) {
+                return nbThreads;
+            }
+        }
+        DISPLAYLEVEL(2, "Ignore environment variable setting %s=%s: not a valid unsigned value \n", ENV_NBWORKERS, env);
+    }
+
+    return ZSTDCLI_NBTHREADS_DEFAULT;
+#else
+    return 1;
+#endif
+}
+
+#define NEXT_FIELD(ptr) {         \
+    if (*argument == '=') {       \
+        ptr = ++argument;         \
+        argument += strlen(ptr);  \
+    } else {                      \
+        argNb++;                  \
+        if (argNb >= argCount) {  \
+            DISPLAYLEVEL(1, "error: missing command argument \n"); \
+            CLEAN_RETURN(1);      \
+        }                         \
+        ptr = argv[argNb];        \
+        assert(ptr != NULL);      \
+        if (ptr[0]=='-') {        \
+            DISPLAYLEVEL(1, "error: command cannot be separated from its argument by another command \n"); \
+            CLEAN_RETURN(1);      \
+}   }   }
+
+#define NEXT_UINT32(_varu32) {        \
+    const char* __nb;                 \
+    NEXT_FIELD(__nb);                 \
+    _varu32 = readU32FromChar(&__nb); \
+    if(*__nb != 0) {                  \
+        errorOut("error: only numeric values with optional suffixes K, KB, KiB, M, MB, MiB are allowed"); \
+    }                                 \
+}
+
+#define NEXT_TSIZE(_varTsize) {           \
+    const char* __nb;                     \
+    NEXT_FIELD(__nb);                     \
+    _varTsize = readSizeTFromChar(&__nb); \
+    if(*__nb != 0) {                      \
+        errorOut("error: only numeric values with optional suffixes K, KB, KiB, M, MB, MiB are allowed"); \
+    }                                     \
+}
+
+typedef enum { zom_compress, zom_decompress, zom_test, zom_bench, zom_train, zom_list } zstd_operation_mode;
+
+#define CLEAN_RETURN(i) { operationResult = (i); goto _end; }
+
+#ifdef ZSTD_NOCOMPRESS
+/* symbols from compression library are not defined and should not be invoked */
+# define MINCLEVEL  -99
+# define MAXCLEVEL   22
+#else
+# define MINCLEVEL  ZSTD_minCLevel()
+# define MAXCLEVEL  ZSTD_maxCLevel()
+#endif
+
+int main(int argCount, const char* argv[])
+{
+    int argNb,
+        followLinks = 0,
+        allowBlockDevices = 0,
+        forceStdin = 0,
+        forceStdout = 0,
+        hasStdout = 0,
+        ldmFlag = 0,
+        main_pause = 0,
+        adapt = 0,
+        adaptMin = MINCLEVEL,
+        adaptMax = MAXCLEVEL,
+        rsyncable = 0,
+        nextArgumentsAreFiles = 0,
+        operationResult = 0,
+        separateFiles = 0,
+        setRealTimePrio = 0,
+        singleThread = 0,
+        defaultLogicalCores = 0,
+        showDefaultCParams = 0,
+        contentSize = 1,
+        removeSrcFile = 0,
+        cLevel = init_cLevel(),
+        ultra = 0,
+        cLevelLast = MINCLEVEL - 1, /* for benchmark range */
+        setThreads_non1 = 0;
+    unsigned nbWorkers = init_nbWorkers();
+    ZSTD_ParamSwitch_e mmapDict = ZSTD_ps_auto;
+    ZSTD_ParamSwitch_e useRowMatchFinder = ZSTD_ps_auto;
+    FIO_compressionType_t cType = FIO_zstdCompression;
+    double compressibility = -1.0;  /* lorem ipsum generator */
+    unsigned bench_nbSeconds = 3;   /* would be better if this value was synchronized from bench */
+    size_t chunkSize = 0;
+
+    FIO_prefs_t* const prefs = FIO_createPreferences();
+    FIO_ctx_t* const fCtx = FIO_createContext();
+    FIO_progressSetting_e progress = FIO_ps_auto;
+    zstd_operation_mode operation = zom_compress;
+    ZSTD_compressionParameters compressionParams;
+    unsigned recursive = 0;
+    unsigned memLimit = 0;
+    FileNamesTable* filenames = UTIL_allocateFileNamesTable((size_t)argCount);  /* argCount >= 1 */
+    FileNamesTable* file_of_names = UTIL_allocateFileNamesTable((size_t)argCount);  /* argCount >= 1 */
+    const char* programName = argv[0];
+    const char* outFileName = NULL;
+    const char* outDirName = NULL;
+    const char* outMirroredDirName = NULL;
+    const char* dictFileName = NULL;
+    const char* patchFromDictFileName = NULL;
+    const char* suffix = ZSTD_EXTENSION;
+    unsigned maxDictSize = g_defaultMaxDictSize;
+    unsigned dictID = 0;
+    size_t streamSrcSize = 0;
+    size_t targetCBlockSize = 0;
+    size_t srcSizeHint = 0;
+    size_t nbInputFileNames = 0;
+    int dictCLevel = g_defaultDictCLevel;
+    unsigned dictSelect = g_defaultSelectivityLevel;
+#ifndef ZSTD_NODICT
+    ZDICT_cover_params_t coverParams = defaultCoverParams();
+    ZDICT_fastCover_params_t fastCoverParams = defaultFastCoverParams();
+    dictType dict = fastCover;
+#endif
+#ifndef ZSTD_NOBENCH
+    BMK_advancedParams_t benchParams = BMK_initAdvancedParams();
+#endif
+    ZSTD_ParamSwitch_e literalCompressionMode = ZSTD_ps_auto;
+
+    /* init */
+    checkLibVersion();
+    (void)recursive; (void)cLevelLast;    /* not used when ZSTD_NOBENCH set */
+    (void)memLimit;
+    assert(argCount >= 1);
+    if ((filenames==NULL) || (file_of_names==NULL)) { DISPLAYLEVEL(1, "zstd: allocation error \n"); exit(1); }
+    programName = lastNameFromPath(programName);
+
+    /* preset behaviors */
+    if (exeNameMatch(programName, ZSTD_ZSTDMT)) nbWorkers=NBWORKERS_AUTOCPU, singleThread=0;
+    if (exeNameMatch(programName, ZSTD_UNZSTD)) operation=zom_decompress;
+    if (exeNameMatch(programName, ZSTD_CAT)) { operation=zom_decompress; FIO_overwriteMode(prefs); forceStdout=1; followLinks=1; FIO_setPassThroughFlag(prefs, 1); outFileName=stdoutmark; g_displayLevel=1; }     /* supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_ZCAT)) { operation=zom_decompress; FIO_overwriteMode(prefs); forceStdout=1; followLinks=1; FIO_setPassThroughFlag(prefs, 1); outFileName=stdoutmark; g_displayLevel=1; }    /* behave like zcat, also supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_GZ)) {   /* behave like gzip */
+        suffix = GZ_EXTENSION; cType = FIO_gzipCompression; removeSrcFile=1;
+        dictCLevel = cLevel = 6;  /* gzip default is -6 */
+    }
+    if (exeNameMatch(programName, ZSTD_GUNZIP)) { operation=zom_decompress; removeSrcFile=1; }                              /* behave like gunzip, also supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_GZCAT)) { operation=zom_decompress; FIO_overwriteMode(prefs); forceStdout=1; followLinks=1; FIO_setPassThroughFlag(prefs, 1); outFileName=stdoutmark; g_displayLevel=1; }   /* behave like gzcat, also supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_LZMA)) { suffix = LZMA_EXTENSION; cType = FIO_lzmaCompression; removeSrcFile=1; }    /* behave like lzma */
+    if (exeNameMatch(programName, ZSTD_UNLZMA)) { operation=zom_decompress; cType = FIO_lzmaCompression; removeSrcFile=1; } /* behave like unlzma, also supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_XZ)) { suffix = XZ_EXTENSION; cType = FIO_xzCompression; removeSrcFile=1; }          /* behave like xz */
+    if (exeNameMatch(programName, ZSTD_UNXZ)) { operation=zom_decompress; cType = FIO_xzCompression; removeSrcFile=1; }     /* behave like unxz, also supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_LZ4)) { suffix = LZ4_EXTENSION; cType = FIO_lz4Compression; }                        /* behave like lz4 */
+    if (exeNameMatch(programName, ZSTD_UNLZ4)) { operation=zom_decompress; cType = FIO_lz4Compression; }                    /* behave like unlz4, also supports multiple formats */
+    memset(&compressionParams, 0, sizeof(compressionParams));
+
+    /* init crash handler */
+    FIO_addAbortHandler();
+
+    /* command switches */
+    for (argNb=1; argNb<argCount; argNb++) {
+        const char* argument = argv[argNb];
+        const char* const originalArgument = argument;
+        if (!argument) continue;   /* Protection if argument empty */
+
+        if (nextArgumentsAreFiles) {
+            UTIL_refFilename(filenames, argument);
+            continue;
+        }
+
+        /* "-" means stdin/stdout */
+        if (!strcmp(argument, "-")){
+            UTIL_refFilename(filenames, stdinmark);
+            continue;
+        }
+
+        /* Decode commands (note : aggregated commands are allowed) */
+        if (argument[0]=='-') {
+
+            if (argument[1]=='-') {
+                /* long commands (--long-word) */
+                if (!strcmp(argument, "--")) { nextArgumentsAreFiles=1; continue; }   /* only file names allowed from now on */
+                if (!strcmp(argument, "--list")) { operation=zom_list; continue; }
+                if (!strcmp(argument, "--compress")) { operation=zom_compress; continue; }
+                if (!strcmp(argument, "--decompress")) { operation=zom_decompress; continue; }
+                if (!strcmp(argument, "--uncompress")) { operation=zom_decompress; continue; }
+                if (!strcmp(argument, "--force")) { FIO_overwriteMode(prefs); forceStdin=1; forceStdout=1; followLinks=1; allowBlockDevices=1; continue; }
+                if (!strcmp(argument, "--version")) { printVersion(); CLEAN_RETURN(0); }
+                if (!strcmp(argument, "--help")) { usageAdvanced(programName); CLEAN_RETURN(0); }
+                if (!strcmp(argument, "--verbose")) { g_displayLevel++; continue; }
+                if (!strcmp(argument, "--quiet")) { g_displayLevel--; continue; }
+                if (!strcmp(argument, "--stdout")) { forceStdout=1; outFileName=stdoutmark; continue; }
+                if (!strcmp(argument, "--ultra")) { ultra=1; continue; }
+                if (!strcmp(argument, "--check")) { FIO_setChecksumFlag(prefs, 2); continue; }
+                if (!strcmp(argument, "--no-check")) { FIO_setChecksumFlag(prefs, 0); continue; }
+                if (!strcmp(argument, "--sparse")) { FIO_setSparseWrite(prefs, 2); continue; }
+                if (!strcmp(argument, "--no-sparse")) { FIO_setSparseWrite(prefs, 0); continue; }
+                if (!strcmp(argument, "--pass-through")) { FIO_setPassThroughFlag(prefs, 1); continue; }
+                if (!strcmp(argument, "--no-pass-through")) { FIO_setPassThroughFlag(prefs, 0); continue; }
+                if (!strcmp(argument, "--test")) { operation=zom_test; continue; }
+                if (!strcmp(argument, "--asyncio")) { FIO_setAsyncIOFlag(prefs, 1); continue;}
+                if (!strcmp(argument, "--no-asyncio")) { FIO_setAsyncIOFlag(prefs, 0); continue;}
+                if (!strcmp(argument, "--train")) { operation=zom_train; if (outFileName==NULL) outFileName=g_defaultDictName; continue; }
+                if (!strcmp(argument, "--no-dictID")) { FIO_setDictIDFlag(prefs, 0); continue; }
+                if (!strcmp(argument, "--keep")) { removeSrcFile=0; continue; }
+                if (!strcmp(argument, "--rm")) { removeSrcFile=1; continue; }
+                if (!strcmp(argument, "--priority=rt")) { setRealTimePrio = 1; continue; }
+                if (!strcmp(argument, "--show-default-cparams")) { showDefaultCParams = 1; continue; }
+                if (!strcmp(argument, "--content-size")) { contentSize = 1; continue; }
+                if (!strcmp(argument, "--no-content-size")) { contentSize = 0; continue; }
+                if (!strcmp(argument, "--adapt")) { adapt = 1; continue; }
+                if (!strcmp(argument, "--no-row-match-finder")) { useRowMatchFinder = ZSTD_ps_disable; continue; }
+                if (!strcmp(argument, "--row-match-finder")) { useRowMatchFinder = ZSTD_ps_enable; continue; }
+                if (longCommandWArg(&argument, "--adapt=")) { adapt = 1; if (!parseAdaptParameters(argument, &adaptMin, &adaptMax)) { badUsage(programName, originalArgument); CLEAN_RETURN(1); } continue; }
+                if (!strcmp(argument, "--single-thread")) { nbWorkers = 0; singleThread = 1; continue; }
+                if (!strcmp(argument, "--format=zstd")) { suffix = ZSTD_EXTENSION; cType = FIO_zstdCompression; continue; }
+                if (!strcmp(argument, "--mmap-dict")) { mmapDict = ZSTD_ps_enable; continue; }
+                if (!strcmp(argument, "--no-mmap-dict")) { mmapDict = ZSTD_ps_disable; continue; }
+#ifdef ZSTD_GZCOMPRESS
+                if (!strcmp(argument, "--format=gzip")) { suffix = GZ_EXTENSION; cType = FIO_gzipCompression; continue; }
+                if (exeNameMatch(programName, ZSTD_GZ)) {     /* behave like gzip */
+                    if (!strcmp(argument, "--best")) { dictCLevel = cLevel = 9; continue; }
+                    if (!strcmp(argument, "--no-name")) { /* ignore for now */; continue; }
+                }
+#endif
+#ifdef ZSTD_LZMACOMPRESS
+                if (!strcmp(argument, "--format=lzma")) { suffix = LZMA_EXTENSION; cType = FIO_lzmaCompression; continue; }
+                if (!strcmp(argument, "--format=xz")) { suffix = XZ_EXTENSION; cType = FIO_xzCompression; continue; }
+#endif
+#ifdef ZSTD_LZ4COMPRESS
+                if (!strcmp(argument, "--format=lz4")) { suffix = LZ4_EXTENSION; cType = FIO_lz4Compression; continue; }
+#endif
+                if (!strcmp(argument, "--rsyncable")) { rsyncable = 1; continue; }
+                if (!strcmp(argument, "--compress-literals")) { literalCompressionMode = ZSTD_ps_enable; continue; }
+                if (!strcmp(argument, "--no-compress-literals")) { literalCompressionMode = ZSTD_ps_disable; continue; }
+                if (!strcmp(argument, "--no-progress")) { progress = FIO_ps_never; continue; }
+                if (!strcmp(argument, "--progress")) { progress = FIO_ps_always; continue; }
+                if (!strcmp(argument, "--exclude-compressed")) { FIO_setExcludeCompressedFile(prefs, 1); continue; }
+                if (!strcmp(argument, "--fake-stdin-is-console")) { UTIL_fakeStdinIsConsole(); continue; }
+                if (!strcmp(argument, "--fake-stdout-is-console")) { UTIL_fakeStdoutIsConsole(); continue; }
+                if (!strcmp(argument, "--fake-stderr-is-console")) { UTIL_fakeStderrIsConsole(); continue; }
+                if (!strcmp(argument, "--trace-file-stat")) { UTIL_traceFileStat(); continue; }
+
+                if (!strcmp(argument, "--max")) {
+                    if (sizeof(void*)==4) {
+                        DISPLAYLEVEL(2, "--max is incompatible with 32-bit mode \n");
+                        badUsage(programName, originalArgument);
+                        CLEAN_RETURN(1);
+                    }
+                    ultra=1; ldmFlag = 1; setMaxCompression(&compressionParams);
+                    continue;
+                }
+
+                /* long commands with arguments */
+#ifndef ZSTD_NODICT
+                if (longCommandWArg(&argument, "--train-cover")) {
+                  operation = zom_train;
+                  if (outFileName == NULL)
+                      outFileName = g_defaultDictName;
+                  dict = cover;
+                  /* Allow optional arguments following an = */
+                  if (*argument == 0) { memset(&coverParams, 0, sizeof(coverParams)); }
+                  else if (*argument++ != '=') { badUsage(programName, originalArgument); CLEAN_RETURN(1); }
+                  else if (!parseCoverParameters(argument, &coverParams)) { badUsage(programName, originalArgument); CLEAN_RETURN(1); }
+                  continue;
+                }
+                if (longCommandWArg(&argument, "--train-fastcover")) {
+                  operation = zom_train;
+                  if (outFileName == NULL)
+                      outFileName = g_defaultDictName;
+                  dict = fastCover;
+                  /* Allow optional arguments following an = */
+                  if (*argument == 0) { memset(&fastCoverParams, 0, sizeof(fastCoverParams)); }
+                  else if (*argument++ != '=') { badUsage(programName, originalArgument); CLEAN_RETURN(1); }
+                  else if (!parseFastCoverParameters(argument, &fastCoverParams)) { badUsage(programName, originalArgument); CLEAN_RETURN(1); }
+                  continue;
+                }
+                if (longCommandWArg(&argument, "--train-legacy")) {
+                  operation = zom_train;
+                  if (outFileName == NULL)
+                      outFileName = g_defaultDictName;
+                  dict = legacy;
+                  /* Allow optional arguments following an = */
+                  if (*argument == 0) { continue; }
+                  else if (*argument++ != '=') { badUsage(programName, originalArgument); CLEAN_RETURN(1); }
+                  else if (!parseLegacyParameters(argument, &dictSelect)) { badUsage(programName, originalArgument); CLEAN_RETURN(1); }
+                  continue;
+                }
+#endif
+                if (longCommandWArg(&argument, "--threads")) { NEXT_UINT32(nbWorkers); setThreads_non1 = (nbWorkers != 1); continue; }
+                if (longCommandWArg(&argument, "--memlimit")) { NEXT_UINT32(memLimit); continue; }
+                if (longCommandWArg(&argument, "--memory")) { NEXT_UINT32(memLimit); continue; }
+                if (longCommandWArg(&argument, "--memlimit-decompress")) { NEXT_UINT32(memLimit); continue; }
+                if (longCommandWArg(&argument, "--block-size")) { NEXT_TSIZE(chunkSize); continue; } /* hidden command, prefer --split below */
+                if (longCommandWArg(&argument, "--split")) { NEXT_TSIZE(chunkSize); continue; }
+                if (longCommandWArg(&argument, "--jobsize")) { NEXT_TSIZE(chunkSize); continue; } /* note: overloaded variable */
+                if (longCommandWArg(&argument, "--maxdict")) { NEXT_UINT32(maxDictSize); continue; }
+                if (longCommandWArg(&argument, "--dictID")) { NEXT_UINT32(dictID); continue; }
+                if (longCommandWArg(&argument, "--zstd=")) { if (!parseCompressionParameters(argument, &compressionParams)) { badUsage(programName, originalArgument); CLEAN_RETURN(1); } ; cType = FIO_zstdCompression; continue; }
+                if (longCommandWArg(&argument, "--stream-size")) { NEXT_TSIZE(streamSrcSize); continue; }
+                if (longCommandWArg(&argument, "--target-compressed-block-size")) { NEXT_TSIZE(targetCBlockSize); continue; }
+                if (longCommandWArg(&argument, "--size-hint")) { NEXT_TSIZE(srcSizeHint); continue; }
+                if (longCommandWArg(&argument, "--output-dir-flat")) {
+                    NEXT_FIELD(outDirName);
+                    if (strlen(outDirName) == 0) {
+                        DISPLAYLEVEL(1, "error: output dir cannot be empty string (did you mean to pass '.' instead?)\n");
+                        CLEAN_RETURN(1);
+                    }
+                    continue;
+                }
+                if (longCommandWArg(&argument, "--auto-threads")) {
+                    const char* threadDefault = NULL;
+                    NEXT_FIELD(threadDefault);
+                    if (strcmp(threadDefault, "logical") == 0)
+                        defaultLogicalCores = 1;
+                    continue;
+                }
+#ifdef UTIL_HAS_MIRRORFILELIST
+                if (longCommandWArg(&argument, "--output-dir-mirror")) {
+                    NEXT_FIELD(outMirroredDirName);
+                    if (strlen(outMirroredDirName) == 0) {
+                        DISPLAYLEVEL(1, "error: output dir cannot be empty string (did you mean to pass '.' instead?)\n");
+                        CLEAN_RETURN(1);
+                    }
+                    continue;
+                }
+#endif
+#ifndef ZSTD_NOTRACE
+                if (longCommandWArg(&argument, "--trace")) { char const* traceFile; NEXT_FIELD(traceFile); TRACE_enable(traceFile); continue; }
+#endif
+                if (longCommandWArg(&argument, "--patch-from")) { NEXT_FIELD(patchFromDictFileName); ultra = 1; continue; }
+                if (longCommandWArg(&argument, "--patch-apply")) { operation=zom_decompress; NEXT_FIELD(patchFromDictFileName); memLimit= 1U << ZSTD_WINDOWLOG_MAX; continue; }
+                if (longCommandWArg(&argument, "--long")) {
+                    unsigned ldmWindowLog = 0;
+                    ldmFlag = 1;
+                    ultra = 1;
+                    /* Parse optional window log */
+                    if (*argument == '=') {
+                        ++argument;
+                        ldmWindowLog = readU32FromChar(&argument);
+                    } else if (*argument != 0) {
+                        /* Invalid character following --long */
+                        badUsage(programName, originalArgument);
+                        CLEAN_RETURN(1);
+                    } else {
+                        ldmWindowLog = g_defaultMaxWindowLog;
+                    }
+                    /* Only set windowLog if not already set by --zstd */
+                    if (compressionParams.windowLog == 0)
+                        compressionParams.windowLog = ldmWindowLog;
+                    continue;
+                }
+#ifndef ZSTD_NOCOMPRESS   /* linking ZSTD_minCLevel() requires compression support */
+                if (longCommandWArg(&argument, "--fast")) {
+                    /* Parse optional acceleration factor */
+                    if (*argument == '=') {
+                        U32 const maxFast = (U32)-ZSTD_minCLevel();
+                        U32 fastLevel;
+                        ++argument;
+                        fastLevel = readU32FromChar(&argument);
+                        if (fastLevel > maxFast) fastLevel = maxFast;
+                        if (fastLevel) {
+                            dictCLevel = cLevel = -(int)fastLevel;
+                        } else {
+                            badUsage(programName, originalArgument);
+                            CLEAN_RETURN(1);
+                        }
+                    } else if (*argument != 0) {
+                        /* Invalid character following --fast */
+                        badUsage(programName, originalArgument);
+                        CLEAN_RETURN(1);
+                    } else {
+                        cLevel = -1;  /* default for --fast */
+                    }
+                    continue;
+                }
+#endif
+
+                if (longCommandWArg(&argument, "--filelist")) {
+                    const char* listName;
+                    NEXT_FIELD(listName);
+                    UTIL_refFilename(file_of_names, listName);
+                    continue;
+                }
+
+                badUsage(programName, originalArgument);
+                CLEAN_RETURN(1);
+            }
+
+            argument++;
+            while (argument[0]!=0) {
+
+#ifndef ZSTD_NOCOMPRESS
+                /* compression Level */
+                if ((*argument>='0') && (*argument<='9')) {
+                    dictCLevel = cLevel = (int)readU32FromChar(&argument);
+                    continue;
+                }
+#endif
+
+                switch(argument[0])
+                {
+                    /* Display help */
+                case 'V': printVersion(); CLEAN_RETURN(0);   /* Version Only */
+                case 'H': usageAdvanced(programName); CLEAN_RETURN(0);
+                case 'h': usage(stdout, programName); CLEAN_RETURN(0);
+
+                     /* Compress */
+                case 'z': operation=zom_compress; argument++; break;
+
+                     /* Decoding */
+                case 'd':
+#ifndef ZSTD_NOBENCH
+                        benchParams.mode = BMK_decodeOnly;
+                        if (operation==zom_bench) { argument++; break; }  /* benchmark decode (hidden option) */
+#endif
+                        operation=zom_decompress; argument++; break;
+
+                    /* Force stdout, even if stdout==console */
+                case 'c': forceStdout=1; outFileName=stdoutmark; argument++; break;
+
+                    /* destination file name */
+                case 'o': argument++; NEXT_FIELD(outFileName); break;
+
+                    /* do not store filename - gzip compatibility - nothing to do */
+                case 'n': argument++; break;
+
+                    /* Use file content as dictionary */
+                case 'D': argument++; NEXT_FIELD(dictFileName); break;
+
+                    /* Overwrite */
+                case 'f': FIO_overwriteMode(prefs); forceStdin=1; forceStdout=1; followLinks=1; allowBlockDevices=1; argument++; break;
+
+                    /* Verbose mode */
+                case 'v': g_displayLevel++; argument++; break;
+
+                    /* Quiet mode */
+                case 'q': g_displayLevel--; argument++; break;
+
+                    /* keep source file (default) */
+                case 'k': removeSrcFile=0; argument++; break;
+
+                    /* Checksum */
+                case 'C': FIO_setChecksumFlag(prefs, 2); argument++; break;
+
+                    /* test compressed file */
+                case 't': operation=zom_test; argument++; break;
+
+                    /* limit memory */
+                case 'M':
+                    argument++;
+                    memLimit = readU32FromChar(&argument);
+                    break;
+                case 'l': operation=zom_list; argument++; break;
+#ifdef UTIL_HAS_CREATEFILELIST
+                    /* recursive */
+                case 'r': recursive=1; argument++; break;
+#endif
+
+#ifndef ZSTD_NOBENCH
+                    /* Benchmark */
+                case 'b':
+                    operation=zom_bench;
+                    argument++;
+                    break;
+
+                    /* range bench (benchmark only) */
+                case 'e':
+                    /* compression Level */
+                    argument++;
+                    cLevelLast = (int)readU32FromChar(&argument);
+                    break;
+
+                    /* Modify Nb Iterations (benchmark only) */
+                case 'i':
+                    argument++;
+                    bench_nbSeconds = readU32FromChar(&argument);
+                    break;
+
+                    /* hidden shortcut for --split=# and --jobsize=# */
+                case 'B':
+                    argument++;
+                    chunkSize = readU32FromChar(&argument);
+                    break;
+
+                    /* benchmark files separately (hidden option) */
+                case 'S':
+                    argument++;
+                    separateFiles = 1;
+                    break;
+
+#endif   /* ZSTD_NOBENCH */
+
+                    /* nb of threads (hidden option) */
+                case 'T':
+                    argument++;
+                    nbWorkers = readU32FromChar(&argument);
+                    setThreads_non1 = (nbWorkers != 1);
+                    break;
+
+                    /* Dictionary Selection level */
+                case 's':
+                    argument++;
+                    dictSelect = readU32FromChar(&argument);
+                    break;
+
+                    /* Pause at the end (-p) or set an additional param (-p#) (hidden option) */
+                case 'p': argument++;
+#ifndef ZSTD_NOBENCH
+                    if ((*argument>='0') && (*argument<='9')) {
+                        benchParams.additionalParam = (int)readU32FromChar(&argument);
+                    } else
+#endif
+                        main_pause=1;
+                    break;
+
+                    /* Select compressibility of synthetic sample */
+                case 'P':
+                    argument++;
+                    compressibility = (double)readU32FromChar(&argument) / 100;
+                    break;
+
+                    /* unknown command */
+                default :
+                    {   char shortArgument[3] = {'-', 0, 0};
+                        shortArgument[1] = argument[0];
+                        badUsage(programName, shortArgument);
+                        CLEAN_RETURN(1);
+                    }
+                }
+            }
+            continue;
+        }   /* if (argument[0]=='-') */
+
+        /* none of the above : add filename to list */
+        UTIL_refFilename(filenames, argument);
+    }
+
+    /* Welcome message (if verbose) */
+    DISPLAYLEVEL(3, WELCOME_MESSAGE);
+
+#ifdef ZSTD_MULTITHREAD
+    if ((operation==zom_decompress) && (setThreads_non1)) {
+        DISPLAYLEVEL(2, "Warning : decompression does not support multi-threading\n");
+    }
+    if ((nbWorkers==NBWORKERS_AUTOCPU) && (!singleThread)) {
+        /* automatically set # workers based on # of reported cpu cores */
+        if (defaultLogicalCores) {
+            nbWorkers = (unsigned)UTIL_countLogicalCores();
+            DISPLAYLEVEL(3, "Note: %d logical core(s) detected \n", nbWorkers);
+        } else {
+            nbWorkers = (unsigned)UTIL_countPhysicalCores();
+            DISPLAYLEVEL(3, "Note: %d physical core(s) detected \n", nbWorkers);
+        }
+    }
+    if (operation == zom_compress)
+        DISPLAYLEVEL(4, "Compressing with %u worker threads \n", nbWorkers);
+#else
+    (void)singleThread; (void)nbWorkers; (void)defaultLogicalCores; (void)setThreads_non1;
+#endif
+
+    g_utilDisplayLevel = g_displayLevel;
+
+#ifdef UTIL_HAS_CREATEFILELIST
+    if (!followLinks) {
+        unsigned u, fileNamesNb;
+        unsigned const nbFilenames = (unsigned)filenames->tableSize;
+        for (u=0, fileNamesNb=0; u<nbFilenames; u++) {
+            if ( UTIL_isLink(filenames->fileNames[u])
+             && !UTIL_isFIFO(filenames->fileNames[u])
+            ) {
+                DISPLAYLEVEL(2, "Warning : %s is a symbolic link, ignoring \n", filenames->fileNames[u]);
+            } else {
+                filenames->fileNames[fileNamesNb++] = filenames->fileNames[u];
+        }   }
+        if (fileNamesNb == 0 && nbFilenames > 0)  /* all names are eliminated */
+            CLEAN_RETURN(1);
+        filenames->tableSize = fileNamesNb;
+    }   /* if (!followLinks) */
+
+    /* read names from a file */
+    if (file_of_names->tableSize) {
+        size_t const nbFileLists = file_of_names->tableSize;
+        size_t flNb;
+        for (flNb=0; flNb < nbFileLists; flNb++) {
+            FileNamesTable* const fnt = UTIL_createFileNamesTable_fromFileList(file_of_names->fileNames[flNb]);
+            if (fnt==NULL) {
+                DISPLAYLEVEL(1, "zstd: error reading %s \n", file_of_names->fileNames[flNb]);
+                CLEAN_RETURN(1);
+            }
+            filenames = UTIL_mergeFileNamesTable(filenames, fnt);
+        }
+    }
+
+    nbInputFileNames = filenames->tableSize; /* saving number of input files */
+
+    if (recursive) {  /* at this stage, filenameTable is a list of paths, which can contain both files and directories */
+        UTIL_expandFNT(&filenames, followLinks);
+    }
+#else
+    (void)followLinks;
+#endif
+
+    if (operation == zom_list) {
+#ifndef ZSTD_NODECOMPRESS
+        int const ret = FIO_listMultipleFiles((unsigned)filenames->tableSize, filenames->fileNames, g_displayLevel);
+        CLEAN_RETURN(ret);
+#else
+        DISPLAYLEVEL(1, "file information is not supported \n");
+        CLEAN_RETURN(1);
+#endif
+    }
+
+    /* Check if benchmark is selected */
+    if (operation==zom_bench) {
+#ifndef ZSTD_NOBENCH
+        if (cType != FIO_zstdCompression) {
+            DISPLAYLEVEL(1, "benchmark mode is only compatible with zstd format \n");
+            CLEAN_RETURN(1);
+        }
+        benchParams.chunkSizeMax = chunkSize;
+        benchParams.targetCBlockSize = targetCBlockSize;
+        benchParams.nbWorkers = (int)nbWorkers;
+        benchParams.realTime = (unsigned)setRealTimePrio;
+        benchParams.nbSeconds = bench_nbSeconds;
+        benchParams.ldmFlag = ldmFlag;
+        benchParams.ldmMinMatch = (int)g_ldmMinMatch;
+        benchParams.ldmHashLog = (int)g_ldmHashLog;
+        benchParams.useRowMatchFinder = (int)useRowMatchFinder;
+        if (g_ldmBucketSizeLog != LDM_PARAM_DEFAULT) {
+            benchParams.ldmBucketSizeLog = (int)g_ldmBucketSizeLog;
+        }
+        if (g_ldmHashRateLog != LDM_PARAM_DEFAULT) {
+            benchParams.ldmHashRateLog = (int)g_ldmHashRateLog;
+        }
+        benchParams.literalCompressionMode = literalCompressionMode;
+
+        if (benchParams.mode == BMK_decodeOnly) cLevel = cLevelLast = 0;
+        if (cLevel > ZSTD_maxCLevel()) cLevel = ZSTD_maxCLevel();
+        if (cLevelLast > ZSTD_maxCLevel()) cLevelLast = ZSTD_maxCLevel();
+        if (cLevelLast < cLevel) cLevelLast = cLevel;
+        DISPLAYLEVEL(3, "Benchmarking ");
+        if (filenames->tableSize > 1)
+            DISPLAYLEVEL(3, "%u files ", (unsigned)filenames->tableSize);
+        if (cLevelLast > cLevel) {
+            DISPLAYLEVEL(3, "from level %d to %d ", cLevel, cLevelLast);
+        } else {
+            DISPLAYLEVEL(3, "at level %d ", cLevel);
+        }
+        DISPLAYLEVEL(3, "using %i threads \n", nbWorkers);
+        if (filenames->tableSize > 0) {
+            if(separateFiles) {
+                unsigned i;
+                for(i = 0; i < filenames->tableSize; i++) {
+                    operationResult = BMK_benchFilesAdvanced(&filenames->fileNames[i], 1, dictFileName, cLevel, cLevelLast, &compressionParams, g_displayLevel, &benchParams);
+                }
+            } else {
+                operationResult = BMK_benchFilesAdvanced(filenames->fileNames, (unsigned)filenames->tableSize, dictFileName, cLevel, cLevelLast, &compressionParams, g_displayLevel, &benchParams);
+            }
+        } else {
+            operationResult = BMK_syntheticTest(compressibility, cLevel, cLevelLast, &compressionParams, g_displayLevel, &benchParams);
+        }
+
+#else
+        (void)bench_nbSeconds; (void)chunkSize; (void)setRealTimePrio; (void)separateFiles; (void)compressibility;
+#endif
+        goto _end;
+    }
+
+    /* Check if dictionary builder is selected */
+    if (operation==zom_train) {
+#ifndef ZSTD_NODICT
+        ZDICT_params_t zParams;
+        zParams.compressionLevel = dictCLevel;
+        zParams.notificationLevel = (unsigned)g_displayLevel;
+        zParams.dictID = dictID;
+        if (dict == cover) {
+            int const optimize = !coverParams.k || !coverParams.d;
+            coverParams.nbThreads = (unsigned)nbWorkers;
+            coverParams.zParams = zParams;
+            operationResult = DiB_trainFromFiles(outFileName, maxDictSize, filenames->fileNames, (int)filenames->tableSize, chunkSize, NULL, &coverParams, NULL, optimize, memLimit);
+        } else if (dict == fastCover) {
+            int const optimize = !fastCoverParams.k || !fastCoverParams.d;
+            fastCoverParams.nbThreads = (unsigned)nbWorkers;
+            fastCoverParams.zParams = zParams;
+            operationResult = DiB_trainFromFiles(outFileName, maxDictSize, filenames->fileNames, (int)filenames->tableSize, chunkSize, NULL, NULL, &fastCoverParams, optimize, memLimit);
+        } else {
+            ZDICT_legacy_params_t dictParams;
+            memset(&dictParams, 0, sizeof(dictParams));
+            dictParams.selectivityLevel = dictSelect;
+            dictParams.zParams = zParams;
+            operationResult = DiB_trainFromFiles(outFileName, maxDictSize, filenames->fileNames, (int)filenames->tableSize, chunkSize, &dictParams, NULL, NULL, 0, memLimit);
+        }
+#else
+        (void)dictCLevel; (void)dictSelect; (void)dictID;  (void)maxDictSize; /* not used when ZSTD_NODICT set */
+        DISPLAYLEVEL(1, "training mode not available \n");
+        operationResult = 1;
+#endif
+        goto _end;
+    }
+
+#ifndef ZSTD_NODECOMPRESS
+    if (operation==zom_test) { FIO_setTestMode(prefs, 1); outFileName=nulmark; removeSrcFile=0; }  /* test mode */
+#endif
+
+    /* No input filename ==> use stdin and stdout */
+    if (filenames->tableSize == 0) {
+      /* It is possible that the input
+       was a number of empty directories. In this case
+       stdin and stdout should not be used */
+       if (nbInputFileNames > 0 ){
+        DISPLAYLEVEL(1, "please provide correct input file(s) or non-empty directories -- ignored \n");
+        CLEAN_RETURN(0);
+       }
+       UTIL_refFilename(filenames, stdinmark);
+    }
+
+    if (filenames->tableSize == 1 && !strcmp(filenames->fileNames[0], stdinmark) && !outFileName)
+        outFileName = stdoutmark;  /* when input is stdin, default output is stdout */
+
+    /* Check if input/output defined as console; trigger an error in this case */
+    if (!forceStdin
+     && (UTIL_searchFileNamesTable(filenames, stdinmark) != -1)
+     && UTIL_isConsole(stdin) ) {
+        DISPLAYLEVEL(1, "stdin is a console, aborting\n");
+        CLEAN_RETURN(1);
+    }
+    if ( (!outFileName || !strcmp(outFileName, stdoutmark))
+      && UTIL_isConsole(stdout)
+      && (UTIL_searchFileNamesTable(filenames, stdinmark) != -1)
+      && !forceStdout
+      && operation!=zom_decompress ) {
+        DISPLAYLEVEL(1, "stdout is a console, aborting\n");
+        CLEAN_RETURN(1);
+    }
+
+#ifndef ZSTD_NOCOMPRESS
+    /* check compression level limits */
+    {   int const maxCLevel = ultra ? ZSTD_maxCLevel() : ZSTDCLI_CLEVEL_MAX;
+        if (cLevel > maxCLevel) {
+            DISPLAYLEVEL(2, "Warning : compression level higher than max, reduced to %i \n", maxCLevel);
+            cLevel = maxCLevel;
+    }   }
+#endif
+
+    if (showDefaultCParams) {
+        if (operation == zom_decompress) {
+            DISPLAYLEVEL(1, "error : can't use --show-default-cparams in decompression mode \n");
+            CLEAN_RETURN(1);
+        }
+    }
+
+    if (dictFileName != NULL && patchFromDictFileName != NULL) {
+        DISPLAYLEVEL(1, "error : can't use -D and --patch-from=# at the same time \n");
+        CLEAN_RETURN(1);
+    }
+
+    if (patchFromDictFileName != NULL && filenames->tableSize > 1) {
+        DISPLAYLEVEL(1, "error : can't use --patch-from=# on multiple files \n");
+        CLEAN_RETURN(1);
+    }
+
+    /* No status message by default when output is stdout */
+    hasStdout = outFileName && !strcmp(outFileName,stdoutmark);
+    if (hasStdout && (g_displayLevel==2)) g_displayLevel=1;
+
+    /* when stderr is not the console, do not pollute it with progress updates (unless requested) */
+    if (!UTIL_isConsole(stderr) && (progress!=FIO_ps_always)) progress=FIO_ps_never;
+    FIO_setProgressSetting(progress);
+
+    /* don't remove source files when output is stdout */;
+    if (hasStdout && removeSrcFile) {
+        DISPLAYLEVEL(3, "Note: src files are not removed when output is stdout \n");
+        removeSrcFile = 0;
+    }
+    FIO_setRemoveSrcFile(prefs, removeSrcFile);
+
+    /* IO Stream/File */
+    FIO_setHasStdoutOutput(fCtx, hasStdout);
+    FIO_setNbFilesTotal(fCtx, (int)filenames->tableSize);
+    FIO_determineHasStdinInput(fCtx, filenames);
+    FIO_setNotificationLevel(g_displayLevel);
+    FIO_setAllowBlockDevices(prefs, allowBlockDevices);
+    FIO_setPatchFromMode(prefs, patchFromDictFileName != NULL);
+    FIO_setMMapDict(prefs, mmapDict);
+    if (memLimit == 0) {
+        if (compressionParams.windowLog == 0) {
+            memLimit = (U32)1 << g_defaultMaxWindowLog;
+        } else {
+            memLimit = (U32)1 << (compressionParams.windowLog & 31);
+    }   }
+    if (patchFromDictFileName != NULL)
+        dictFileName = patchFromDictFileName;
+    FIO_setMemLimit(prefs, memLimit);
+    if (operation==zom_compress) {
+#ifndef ZSTD_NOCOMPRESS
+        FIO_setCompressionType(prefs, cType);
+        FIO_setContentSize(prefs, contentSize);
+        FIO_setNbWorkers(prefs, (int)nbWorkers);
+        FIO_setJobSize(prefs, (int)chunkSize);
+        if (g_overlapLog!=OVERLAP_LOG_DEFAULT) FIO_setOverlapLog(prefs, (int)g_overlapLog);
+        FIO_setLdmFlag(prefs, (unsigned)ldmFlag);
+        FIO_setLdmHashLog(prefs, (int)g_ldmHashLog);
+        FIO_setLdmMinMatch(prefs, (int)g_ldmMinMatch);
+        if (g_ldmBucketSizeLog != LDM_PARAM_DEFAULT) FIO_setLdmBucketSizeLog(prefs, (int)g_ldmBucketSizeLog);
+        if (g_ldmHashRateLog != LDM_PARAM_DEFAULT) FIO_setLdmHashRateLog(prefs, (int)g_ldmHashRateLog);
+        FIO_setAdaptiveMode(prefs, adapt);
+        FIO_setUseRowMatchFinder(prefs, (int)useRowMatchFinder);
+        FIO_setAdaptMin(prefs, adaptMin);
+        FIO_setAdaptMax(prefs, adaptMax);
+        FIO_setRsyncable(prefs, rsyncable);
+        FIO_setStreamSrcSize(prefs, streamSrcSize);
+        FIO_setTargetCBlockSize(prefs, targetCBlockSize);
+        FIO_setSrcSizeHint(prefs, srcSizeHint);
+        FIO_setLiteralCompressionMode(prefs, literalCompressionMode);
+        FIO_setSparseWrite(prefs, 0);
+        if (adaptMin > cLevel) cLevel = adaptMin;
+        if (adaptMax < cLevel) cLevel = adaptMax;
+
+        /* Compare strategies constant with the ground truth */
+        { ZSTD_bounds strategyBounds = ZSTD_cParam_getBounds(ZSTD_c_strategy);
+          assert(ZSTD_NB_STRATEGIES == strategyBounds.upperBound);
+          (void)strategyBounds; }
+
+        if (showDefaultCParams || g_displayLevel >= 4) {
+            size_t fileNb;
+            for (fileNb = 0; fileNb < (size_t)filenames->tableSize; fileNb++) {
+                if (showDefaultCParams)
+                    printDefaultCParams(filenames->fileNames[fileNb], dictFileName, cLevel);
+                if (g_displayLevel >= 4)
+                    printActualCParams(filenames->fileNames[fileNb], dictFileName, cLevel, &compressionParams);
+            }
+        }
+
+        if (g_displayLevel >= 4)
+            FIO_displayCompressionParameters(prefs);
+        if ((filenames->tableSize==1) && outFileName)
+            operationResult = FIO_compressFilename(fCtx, prefs, outFileName, filenames->fileNames[0], dictFileName, cLevel, compressionParams);
+        else
+            operationResult = FIO_compressMultipleFilenames(fCtx, prefs, filenames->fileNames, outMirroredDirName, outDirName, outFileName, suffix, dictFileName, cLevel, compressionParams);
+#else
+        /* these variables are only used when compression mode is enabled */
+        (void)contentSize; (void)suffix; (void)adapt; (void)rsyncable;
+        (void)ultra; (void)cLevel; (void)ldmFlag; (void)literalCompressionMode;
+        (void)targetCBlockSize; (void)streamSrcSize; (void)srcSizeHint;
+        (void)ZSTD_strategyMap; (void)useRowMatchFinder; (void)cType;
+        DISPLAYLEVEL(1, "Compression not supported \n");
+#endif
+    } else {  /* decompression or test */
+#ifndef ZSTD_NODECOMPRESS
+        if (filenames->tableSize == 1 && outFileName) {
+            operationResult = FIO_decompressFilename(fCtx, prefs, outFileName, filenames->fileNames[0], dictFileName);
+        } else {
+            operationResult = FIO_decompressMultipleFilenames(fCtx, prefs, filenames->fileNames, outMirroredDirName, outDirName, outFileName, dictFileName);
+        }
+#else
+        DISPLAYLEVEL(1, "Decompression not supported \n");
+#endif
+    }
+
+_end:
+    FIO_freePreferences(prefs);
+    FIO_freeContext(fCtx);
+    if (main_pause) waitEnter();
+    UTIL_freeFileNamesTable(filenames);
+    UTIL_freeFileNamesTable(file_of_names);
+#ifndef ZSTD_NOTRACE
+    TRACE_finish();
+#endif
+
+    return operationResult;
+}