Optimize buffer ops

vortex-buffer/Cargo.toml

@@ -24,7 +24,6 @@ warn-copy = ["dep:tracing"]
 
 [dependencies]
 arrow-buffer = { workspace = true }
-bitvec = { workspace = true }
 bytes = { workspace = true }
 itertools = { workspace = true }
 memmap2 = { workspace = true, optional = true }

vortex-buffer/src/alignment.rs

@@ -54,6 +54,9 @@ impl Alignment {
         Self::new(align_of::<T>())
     }
 
+    /// The largest valid alignment: the greatest power of 2 that fits into a `u16`.
+    pub const MAX: Alignment = Alignment::new(1 << 15);
+
     /// Check if `self` alignment is a "larger" than `other` alignment.
     ///
     /// ## Example
@@ -67,10 +70,32 @@ impl Alignment {
     /// assert!(!b.is_aligned_to(a));
     /// ```
     #[inline]
-    pub fn is_aligned_to(&self, other: Alignment) -> bool {
-        // Since we know alignments are powers of 2, we can compare them by checking if the number
-        // of trailing zeros in the binary representation of the alignment is greater or equal.
-        self.0.trailing_zeros() >= other.0.trailing_zeros()
+    pub const fn is_aligned_to(&self, other: Alignment) -> bool {
+        // Since both alignments are powers of 2, divisibility is equivalent to ordering.
+        self.0 >= other.0
+    }
+
+    /// Check if the given byte offset (or length) is a multiple of this alignment.
+    ///
+    /// ## Example
+    ///
+    /// ```
+    /// use vortex_buffer::Alignment;
+    ///
+    /// let a = Alignment::new(4);
+    /// assert!(a.is_offset_aligned(8));
+    /// assert!(!a.is_offset_aligned(2));
+    /// ```
+    #[inline]
+    pub const fn is_offset_aligned(&self, offset: usize) -> bool {
+        // Alignment is always a power of 2, so a mask test is equivalent to `offset % self == 0`.
+        offset & (self.0 - 1) == 0
+    }
+
+    /// Check if the given pointer is aligned to this alignment.
+    #[inline]
+    pub fn is_ptr_aligned<T>(&self, ptr: *const T) -> bool {
+        self.is_offset_aligned(ptr.addr())
     }
 
     /// Returns the log2 of the alignment.

vortex-buffer/src/bit/buf.rs

@@ -228,10 +228,11 @@ impl BitBuffer {
             unsafe { buffer.push_unchecked(packed) }
         }
 
-        buffer.truncate(len.div_ceil(8));
+        let mut bytes = buffer.into_byte_buffer();
+        bytes.truncate(len.div_ceil(8));
 
         Self {
-            buffer: buffer.freeze().into_byte_buffer(),
+            buffer: bytes.freeze(),
             offset: 0,
             len,
         }
@@ -312,7 +313,23 @@ impl BitBuffer {
         assert!(end <= self.len);
         let len = end - start;
 
-        Self::new_with_offset(self.buffer.clone(), len, self.offset + start)
+        let offset = self.offset + start;
+        let byte_offset = offset / 8;
+        let bit_offset = offset % 8;
+
+        // Trim whole bytes off the front directly rather than going through `new_with_offset`,
+        // which would slice (and re-clone) the clone we'd have to pass it.
+        let buffer = if byte_offset != 0 {
+            self.buffer.slice_unaligned(byte_offset..)
+        } else {
+            self.buffer.clone().aligned(Alignment::none())
+        };
+
+        Self {
+            buffer,
+            offset: bit_offset,
+            len,
+        }
     }
 
     /// Slice any full bytes from the buffer, leaving the offset < 8.

vortex-buffer/src/bit/buf_mut.rs

@@ -3,7 +3,7 @@
 
 use std::ops::Not;
 
-use bitvec::view::BitView;
+use arrow_buffer::bit_mask::set_bits;
 
 use crate::BitBuffer;
 use crate::BufferMut;
@@ -267,7 +267,9 @@ impl BitBufferMut {
 
     /// Clears the bit buffer (but keeps any allocated memory).
     pub fn clear(&mut self) {
-        // Since there are no items we need to drop, we simply set the length to 0.
+        // Also clear the byte buffer (not just `len`) so the "bits beyond len are zero"
+        // invariant holds; `append_false` and `append_buffer` rely on it.
+        self.buffer.clear();
         self.len = 0;
         self.offset = 0;
     }
@@ -518,17 +520,16 @@ impl BitBufferMut {
                 self.buffer.as_mut_slice()[dst_byte + full_bytes] |= src_bytes[full_bytes] & mask;
             }
         } else {
-            // Use bitvec for unaligned bit copying.
-            let self_slice = self
-                .buffer
-                .as_mut_slice()
-                .view_bits_mut::<bitvec::prelude::Lsb0>();
-            let other_slice = buffer
-                .inner()
-                .as_slice()
-                .view_bits::<bitvec::prelude::Lsb0>();
-            let source_range = src_bit_offset..src_bit_offset + bit_len;
-            self_slice[start_bit_pos..end_bit_pos].copy_from_bitslice(&other_slice[source_range]);
+            // Word-wise bit copy that handles mismatched source/destination bit offsets.
+            // `set_bits` ORs into the destination, which is safe because bits beyond `len`
+            // are guaranteed to be zero.
+            set_bits(
+                self.buffer.as_mut_slice(),
+                buffer.inner().as_slice(),
+                start_bit_pos,
+                src_bit_offset,
+                bit_len,
+            );
         }
 
         self.len += bit_len;
@@ -649,6 +650,8 @@ impl FromIterator<bool> for BitBufferMut {
 
 #[cfg(test)]
 mod tests {
+    use rstest::rstest;
+
     use crate::BufferMut;
     use crate::bit::buf_mut::BitBufferMut;
     use crate::bitbuffer;
@@ -923,7 +926,62 @@ mod tests {
         assert!(frozen.value(7));
     }
 
-    #[cfg_attr(miri, ignore)] // bitvec crate uses a ptr cast that Miri doesn't support
+    #[test]
+    fn append_after_clear_reads_back_false() {
+        // `clear` must not leave stale set bits behind: `append_false` and `append_buffer`
+        // rely on bits beyond `len` being zero.
+        let mut bools = BitBufferMut::new_set(16);
+        bools.clear();
+        bools.append_false();
+        bools.append_buffer(&crate::BitBuffer::new_unset(8));
+
+        let bools = bools.freeze();
+        assert_eq!(bools.len(), 9);
+        assert_eq!(bools.true_count(), 0);
+    }
+
+    #[test]
+    fn test_append_buffer_after_truncate() {
+        // Truncating leaves stale set bits in the last partial byte; an append after that
+        // must overwrite them rather than OR into them.
+        let mut buf = BitBufferMut::new_set(16);
+        buf.truncate(3);
+        buf.append_buffer(&crate::BitBuffer::new_unset(8));
+
+        let frozen = buf.freeze();
+        assert_eq!(frozen.len(), 11);
+        for i in 0..3 {
+            assert!(frozen.value(i), "bit {i} should be set");
+        }
+        for i in 3..11 {
+            assert!(!frozen.value(i), "bit {i} should be unset");
+        }
+    }
+
+    #[rstest]
+    #[case::both_aligned(0, 0)]
+    #[case::dst_unaligned(3, 0)]
+    #[case::src_unaligned(0, 5)]
+    #[case::mismatched(3, 5)]
+    #[case::equal_nonzero(5, 5)]
+    fn test_append_buffer_long(#[case] dst_prefix: usize, #[case] src_start: usize) {
+        // Exercise every alignment combination across many words.
+        let source = crate::BitBuffer::from_iter((0..301).map(|i| i % 3 == 0));
+        let source = source.slice(src_start..301);
+
+        let mut dest = BitBufferMut::with_capacity(512);
+        dest.append_n(true, dst_prefix);
+        dest.append_buffer(&source);
+
+        assert_eq!(dest.len(), dst_prefix + source.len());
+        for i in 0..dst_prefix {
+            assert!(dest.value(i), "prefix bit {i}");
+        }
+        for i in 0..source.len() {
+            assert_eq!(dest.value(dst_prefix + i), source.value(i), "bit {i}");
+        }
+    }
+
     #[test]
     fn test_append_buffer_with_offsets() {
         // Create source buffer with offset

vortex-buffer/src/buffer.rs

@@ -32,10 +32,20 @@ pub struct Buffer<T> {
     pub(crate) _marker: PhantomData<T>,
 }
 
+/// Zero-length backing for empty buffers, "aligned" to [`Alignment::MAX`] so it satisfies any
+/// valid alignment without allocating. A zero-length slice never reads memory, so it may use a
+/// dangling pointer as long as it is non-null and aligned.
+const EMPTY_BACKING: &[u8] = {
+    let addr = 1usize << 15;
+    assert!(Alignment::MAX.is_offset_aligned(addr));
+    // SAFETY: the pointer is non-null and aligned, and the slice is zero-length.
+    unsafe { std::slice::from_raw_parts(std::ptr::without_provenance(addr), 0) }
+};
+
 impl<T> Default for Buffer<T> {
     fn default() -> Self {
         Self {
-            bytes: Default::default(),
+            bytes: Bytes::from_static(EMPTY_BACKING),
             length: 0,
             alignment: Alignment::of::<T>(),
             _marker: PhantomData,
@@ -101,12 +111,27 @@ impl<T> Buffer<T> {
 
     /// Create a new empty `ByteBuffer` with the provided alignment.
     pub fn empty() -> Self {
-        BufferMut::empty().freeze()
+        Self::empty_aligned(Alignment::of::<T>())
     }
 
     /// Create a new empty `ByteBuffer` with the provided alignment.
+    ///
+    /// This does not allocate: empty buffers are backed by a zero-length `Bytes` that is
+    /// aligned to [`Alignment::MAX`].
     pub fn empty_aligned(alignment: Alignment) -> Self {
-        BufferMut::empty_aligned(alignment).freeze()
+        if !alignment.is_aligned_to(Alignment::of::<T>()) {
+            vortex_panic!(
+                "Alignment {} must align to the scalar type's alignment {}",
+                alignment,
+                Alignment::of::<T>(),
+            );
+        }
+        Self {
+            bytes: Bytes::from_static(EMPTY_BACKING),
+            length: 0,
+            alignment,
+            _marker: PhantomData,
+        }
     }
 
     /// Create a new full `ByteBuffer` with the given value.
@@ -152,7 +177,7 @@ impl<T> Buffer<T> {
                 Alignment::of::<T>(),
             );
         }
-        if bytes.as_ptr().align_offset(*alignment) != 0 {
+        if !alignment.is_ptr_aligned(bytes.as_ptr()) {
             vortex_panic!(
                 "Bytes alignment must align to the requested alignment {}",
                 alignment,
@@ -320,7 +345,7 @@ impl<T> Buffer<T> {
         let begin_byte = begin * size_of::<T>();
         let end_byte = end * size_of::<T>();
 
-        if !begin_byte.is_multiple_of(*alignment) {
+        if !alignment.is_offset_aligned(begin_byte) {
             vortex_panic!(
                 "range start must be aligned to {alignment:?}, byte {}",
                 begin_byte
@@ -369,7 +394,7 @@ impl<T> Buffer<T> {
             vortex_panic!("slice_ref subset alignment must at least align to the buffer alignment")
         }
 
-        if subset.as_ptr().align_offset(*alignment) != 0 {
+        if !alignment.is_ptr_aligned(subset.as_ptr()) {
             vortex_panic!("slice_ref subset must be aligned to {:?}", alignment);
         }
 
@@ -435,17 +460,17 @@ impl<T> Buffer<T> {
     /// Convert self into `BufferMut<T>`, cloning the data if there are multiple strong references.
     pub fn into_mut(self) -> BufferMut<T> {
         self.try_into_mut()
-            .unwrap_or_else(|buffer| BufferMut::<T>::copy_from(&buffer))
+            .unwrap_or_else(|buffer| BufferMut::<T>::copy_from_aligned(&buffer, buffer.alignment))
     }
 
     /// Returns whether a `Buffer<T>` is aligned to the given alignment.
     pub fn is_aligned(&self, alignment: Alignment) -> bool {
-        self.bytes.as_ptr().align_offset(*alignment) == 0
+        alignment.is_ptr_aligned(self.bytes.as_ptr())
     }
 
     /// Return a `Buffer<T>` with the given alignment. Where possible, this will be zero-copy.
     pub fn aligned(mut self, alignment: Alignment) -> Self {
-        if self.as_ptr().align_offset(*alignment) == 0 {
+        if alignment.is_ptr_aligned(self.as_ptr()) {
             self.alignment = alignment;
             self
         } else {
@@ -462,7 +487,7 @@ impl<T> Buffer<T> {
 
     /// Return a `Buffer<T>` with the given alignment. Panics if the buffer is not aligned.
     pub fn ensure_aligned(mut self, alignment: Alignment) -> Self {
-        if self.as_ptr().align_offset(*alignment) == 0 {
+        if alignment.is_ptr_aligned(self.as_ptr()) {
             self.alignment = alignment;
             self
         } else {
@@ -634,7 +659,7 @@ impl Buf for ByteBuffer {
 
     #[inline]
     fn advance(&mut self, cnt: usize) {
-        if !cnt.is_multiple_of(*self.alignment) {
+        if !self.alignment.is_offset_aligned(cnt) {
             vortex_panic!(
                 "Cannot advance buffer by {} items, resulting alignment is not {}",
                 cnt,
@@ -786,6 +811,31 @@ mod test {
         assert_eq!(vec, buff.as_ref());
     }
 
+    #[test]
+    fn empty_aligned_max_alignment() {
+        // Empty buffers are backed by a static and must satisfy any valid alignment.
+        let buf = Buffer::<u8>::empty_aligned(Alignment::MAX);
+        assert!(buf.is_empty());
+        assert!(buf.is_aligned(Alignment::MAX));
+    }
+
+    #[test]
+    fn empty_slice_preserves_alignment() {
+        let buf = Buffer::<u64>::zeroed_aligned(8, Alignment::new(64));
+        let sliced = buf.slice(0..0);
+        assert!(sliced.is_empty());
+        assert_eq!(sliced.alignment(), Alignment::new(64));
+        assert!(sliced.is_aligned(Alignment::new(64)));
+    }
+
+    #[test]
+    fn empty_into_mut_preserves_alignment() {
+        let buf = Buffer::<u8>::empty_aligned(Alignment::new(64));
+        let buf_mut = buf.into_mut();
+        assert_eq!(buf_mut.alignment(), Alignment::new(64));
+        assert!(buf_mut.is_empty());
+    }
+
     #[test]
     fn test_slice_unaligned_end_pos() {
         let data = vec![0u8; 2];

vortex-buffer/src/buffer_mut.rs

@@ -358,7 +358,7 @@ impl<T> BufferMut<T> {
         }
 
         let bytes_at = at * size_of::<T>();
-        if !bytes_at.is_multiple_of(*self.alignment) {
+        if !self.alignment.is_offset_aligned(bytes_at) {
             vortex_panic!(
                 "Cannot split buffer at {}, resulting alignment is not {}",
                 at,
@@ -742,7 +742,7 @@ impl Buf for ByteBufferMut {
     }
 
     fn advance(&mut self, cnt: usize) {
-        if !cnt.is_multiple_of(*self.alignment) {
+        if !self.alignment.is_offset_aligned(cnt) {
             vortex_panic!(
                 "Cannot advance buffer by {} items, resulting alignment is not {}",
                 cnt,
@@ -765,7 +765,7 @@ unsafe impl BufMut for ByteBufferMut {
 
     #[inline]
     unsafe fn advance_mut(&mut self, cnt: usize) {
-        if !cnt.is_multiple_of(*self.alignment) {
+        if !self.alignment.is_offset_aligned(cnt) {
             vortex_panic!(
                 "Cannot advance buffer by {} items, resulting alignment is not {}",
                 cnt,