Feature/gpu allvsall architecture

src/AntiDupl/AntiDupl.cpp

@@ -80,6 +80,8 @@ typedef ad::TEngine* adEngineHandle;
     if(p == NULL) \
     return AD_ERROR_INVALID_POINTER;
 
+#define AD_DEBUG(msg) OutputDebugStringA(msg)
+
 
 BOOL APIENTRY DllMain(HMODULE hModule, DWORD dwReasonForCall, LPVOID lpReserved)
 {
@@ -153,9 +155,14 @@ DLLAPI adEngineHandle adCreateA(const adCharA * userPath)
 
 DLLAPI adEngineHandle adCreateW(const adCharW * userPath)
 {
+    AD_DEBUG("adCreateW: Starting\n");
     ad::DumpInit();
+    AD_DEBUG("adCreateW: DumpInit done\n");
 
-    return new ad::TEngine(ad::TString(userPath));
+    adEngineHandle handle = new ad::TEngine(ad::TString(userPath));
+    AD_DEBUG("adCreateW: TEngine created\n");
+
+    return handle;
 }
 
 DLLAPI adError adRelease(adEngineHandle handle)

src/AntiDupl/AntiDupl.vcxproj

@@ -23,6 +23,7 @@
   <PropertyGroup Label="Configuration">
     <ConfigurationType>DynamicLibrary</ConfigurationType>
   </PropertyGroup>
+  <Import Project="$(VCTargetsPath)\BuildCustomizations\CUDA 12.8.props" />
   <PropertyGroup Label="Configuration">
     <PlatformToolset>v143</PlatformToolset>
   </PropertyGroup>
@@ -51,14 +52,22 @@
   </PropertyGroup>
   <ItemDefinitionGroup>
     <ClCompile>
-      <PreprocessorDefinitions>_WINDOWS;_USRDLL;ANTIDUPL_EXPORTS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <PreprocessorDefinitions>_WINDOWS;_USRDLL;ANTIDUPL_EXPORTS;ENABLE_CUDA;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <DisableSpecificWarnings Condition="'$(Platform)'=='x64'">4267</DisableSpecificWarnings>
       <AdditionalIncludeDirectories>%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
     </ClCompile>
     <Link>
+      <AdditionalDependencies>cudart.lib;%(AdditionalDependencies)</AdditionalDependencies>
       <SubSystem>Windows</SubSystem>
       <TargetMachine>MachineX64</TargetMachine>
     </Link>
+    <Cudart>
+      <TargetMachinePlatform>64</TargetMachinePlatform>
+    </Cudart>
+    <CudaCompile>
+      <TargetMachinePlatform>64</TargetMachinePlatform>
+      <CodeGeneration>compute_89,sm_89</CodeGeneration>
+    </CudaCompile>
     <PreBuildEvent>
       <Command>"$(ProjectDir)".\adExternal.cmd</Command>
     </PreBuildEvent>
@@ -78,6 +87,7 @@
     <ClCompile Include="adFileUtils.cpp" />
     <ClCompile Include="adGdiplus.cpp" />
     <ClCompile Include="adHeif.cpp" />
+    <ClCompile Include="adGPUManager.cpp" />
     <ClCompile Include="adHintSetter.cpp" />
     <ClCompile Include="adImage.cpp" />
     <ClCompile Include="adImageComparer.cpp" />
@@ -113,6 +123,7 @@
     <ClCompile Include="adUndoRedoEngine.cpp" />
     <ClCompile Include="adUndoRedoTypes.cpp" />
     <ClCompile Include="adWebp.cpp" />
+    <CudaCompile Include="adGPU.cu" />
     <ClCompile Include="AntiDupl.cpp" />
   </ItemGroup>
   <ItemGroup>
@@ -128,6 +139,8 @@
     <ClInclude Include="adFileStream.h" />
     <ClInclude Include="adFileUtils.h" />
     <ClInclude Include="adGdiplus.h" />
+    <ClInclude Include="adGPU.h" />
+    <ClInclude Include="adGPUManager.h" />
     <ClInclude Include="adHeif.h" />
     <ClInclude Include="adHintSetter.h" />
     <ClInclude Include="adImage.h" />
@@ -169,4 +182,7 @@
     <ClInclude Include="AntiDupl.h" />
   </ItemGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+  <ImportGroup Label="ExtensionTargets">
+    <Import Project="$(VCTargetsPath)\BuildCustomizations\CUDA 12.8.targets" />
+  </ImportGroup>
 </Project>

src/AntiDupl/adConfig.h

@@ -45,8 +45,8 @@
 #include <limits>
 
 //#define AD_PERFORMANCE_TEST_ENABLE
-//#define AD_LOGGER_ENABLE
-//#define AD_DUMP_ENABLE
+#define AD_LOGGER_ENABLE
+#define AD_DUMP_ENABLE
 #define AD_TURBO_JPEG_ENABLE
 
 #include "adSimd.h"

src/AntiDupl/adDataCollector.cpp

@@ -32,11 +32,16 @@
 #include "adImageUtils.h"
 #include "adPixelData.h"
 #include "adBlurringDetector.h"
+#include "adGPUManager.h"
+#include <windows.h>
+
+#define AD_DEBUG(msg) OutputDebugStringA(msg)
 
 namespace ad
 {
     TDataCollector::TDataCollector(TEngine *pEngine)
-        :m_pOptions(pEngine->Options()),
+        :m_pEngine(pEngine),
+        m_pOptions(pEngine->Options()),
         m_pResult(pEngine->Result())
     {
         for(int size = INITIAL_REDUCED_IMAGE_SIZE; size > m_pOptions->advanced.reducedImageSize; size >>= 1)
@@ -73,7 +78,7 @@ namespace ad
         TImage *pImage = TImage::Load(pImageData->hGlobal, m_pOptions);
         if(pImage)
         {
-            pImageData->height = (TUInt32)pImage->View()->height; 
+            pImageData->height = (TUInt32)pImage->View()->height;
             pImageData->width = (TUInt32)pImage->View()->width;
             pImageData->type = (TImageType)pImage->Format();
 
@@ -98,13 +103,46 @@ namespace ad
 
 			pImageData->imageExif = pImage->ImageExif();
 
-			Simd::ResizeBilinear(gray, *m_pGrayBuffers.front());
+			Simd::Resize(gray, *m_pGrayBuffers.front());
             for(size_t i = 1; i < m_pGrayBuffers.size(); ++i)
 				Simd::ReduceGray2x2(*m_pGrayBuffers[i - 1], *m_pGrayBuffers[i]);
 			TPixelData & data = *pImageData->data;
             ReduceGray2x2(*m_pGrayBuffers.back(), TView(data.side, data.side, data.side, TView::Gray8, data.main));
             data.filled = true;
 
+            // Upload thumbnail to GPU — only for non-AllVsAll mode
+            // In AllVsAll mode, ExecuteGpuAllVsAllComparison does its own mass upload
+            if (!m_pEngine->SkipComparisonDuringCollection())
+            {
+                // Ensure GPU buffer is initialized before first upload
+                static bool gpuBufferInitialized = false;
+                if (!gpuBufferInitialized)
+                {
+                    AD_DEBUG("FillPixelData: Initializing GPU buffer\n");
+                    size_t estimatedCapacity = 10000; // Start with reasonable estimate
+                    size_t thumbSize = m_pOptions->advanced.reducedImageSize * m_pOptions->advanced.reducedImageSize;
+                    if (m_pEngine->GpuManager()->EnsureCapacity(estimatedCapacity, thumbSize))
+                    {
+                        gpuBufferInitialized = true;
+                        AD_DEBUG("FillPixelData: GPU buffer initialized\n");
+                    }
+                    else
+                    {
+                        AD_DEBUG("FillPixelData: GPU buffer initialization FAILED\n");
+                    }
+                }
+
+                // Upload thumbnail to GPU immediately
+                if (m_pEngine->GpuManager()->UploadThumbnail(pImageData->globalIdx, data.main))
+                {
+                    // Successfully uploaded
+                }
+                else
+                {
+                    AD_DEBUG("FillPixelData: UploadThumbnail FAILED\n");
+                }
+            }
+
 			delete pImage;
         }
         else

src/AntiDupl/adDataCollector.h

@@ -38,6 +38,7 @@ namespace ad
     //-------------------------------------------------------------------------
     class TDataCollector
     {
+        TEngine *m_pEngine;
         TOptions *m_pOptions;
         TResultStorage *m_pResult;
         std::vector<TView*> m_pGrayBuffers;

src/AntiDupl/adDump.cpp

@@ -31,7 +31,7 @@
 
 namespace ad
 {
-    const DWORD STATUS_FATAL_APP_EXIT = 0x40000015;
+    const DWORD AD_STATUS_FATAL_APP_EXIT = 0x40000015;
 
     typedef BOOL (*MiniDumpWriteDumpPtr)(
         HANDLE hProcess,
@@ -107,7 +107,7 @@ namespace ad
     {
         DWORD exception_code = info->ExceptionRecord->ExceptionCode;
         void *address = info->ExceptionRecord->ExceptionAddress;
-        if (exception_code == STATUS_FATAL_APP_EXIT) 
+        if (exception_code == AD_STATUS_FATAL_APP_EXIT) 
         {// abort()
             MessageBox(NULL, TEXT("Program aborted."), 
                 TEXT("Crash report"), MB_OK | MB_ICONERROR);