simpler

Signed-off-by: Robert Kruszewski <github@robertk.io>

Author: robert3005

vortex-buffer/benches/vortex_bitbuffer.rs

@@ -25,19 +25,7 @@ impl FromIterator<bool> for Arrow<BooleanBuffer> {
 }
 
 const INPUT_SIZE: &[usize] = &[128, 1024, 2048, 16_384, 65_536];
-const TRUE_COUNT_INPUT_SIZE: &[usize] = &[128, 1024, 16_384, 65_536, 1_048_576];
-const TRUE_COUNT_SLICE_CASES: &[(usize, usize)] = &[
-    (128, 1),
-    (128, 7),
-    (1_024, 1),
-    (1_024, 7),
-    (16_384, 1),
-    (16_384, 7),
-    (65_536, 1),
-    (65_536, 7),
-    (1_048_576, 1),
-    (1_048_576, 7),
-];
+const TRUE_COUNT_INPUT_SIZE: &[usize] = &[128, 1024, 16_384, 65_536];
 
 #[inline]
 fn true_count_pattern(i: usize) -> bool {
@@ -194,28 +182,6 @@ fn true_count_arrow_buffer(bencher: Bencher, length: usize) {
         .bench_refs(|buffer| buffer.0.count_set_bits());
 }
 
-#[divan::bench(args = TRUE_COUNT_SLICE_CASES)]
-fn true_count_vortex_buffer_sliced(bencher: Bencher, (length, offset): (usize, usize)) {
-    let buffer = BitBuffer::from_iter((0..length + offset).map(true_count_pattern));
-    let sliced = buffer.slice(offset..offset + length);
-
-    bencher
-        .with_inputs(|| &sliced)
-        .bench_refs(|buffer| buffer.true_count())
-}
-
-#[divan::bench(args = TRUE_COUNT_SLICE_CASES)]
-fn true_count_arrow_buffer_sliced(bencher: Bencher, (length, offset): (usize, usize)) {
-    let buffer = Arrow(BooleanBuffer::from_iter(
-        (0..length + offset).map(true_count_pattern),
-    ));
-    let sliced = Arrow(buffer.0.slice(offset, length));
-
-    bencher
-        .with_inputs(|| &sliced)
-        .bench_refs(|buffer| buffer.0.count_set_bits());
-}
-
 #[divan::bench(args = INPUT_SIZE)]
 fn bitwise_and_vortex_buffer(bencher: Bencher, length: usize) {
     let a = BitBuffer::from_iter((0..length).map(|i| i % 2 == 0));

vortex-buffer/src/bit/buf.rs

@@ -360,9 +360,11 @@ impl BitBuffer {
 
 #[inline]
 fn true_count_impl(bytes: &[u8], offset: usize, len: usize) -> usize {
-    let Some((head, middle, tail)) = byte_aligned_region(bytes, offset, len) else {
+    if bytes.is_empty() {
         return 0;
-    };
+    }
+
+    let (head, middle, tail) = byte_aligned_region(bytes, offset, len);
 
     let mut count = head.map_or(0, |v| v.count_ones() as usize);
 
@@ -374,15 +376,7 @@ fn true_count_impl(bytes: &[u8], offset: usize, len: usize) -> usize {
 }
 
 #[inline]
-fn byte_aligned_region(
-    bytes: &[u8],
-    offset: usize,
-    len: usize,
-) -> Option<(Option<u8>, &[u8], Option<u8>)> {
-    if len == 0 {
-        return None;
-    }
-
+fn byte_aligned_region(bytes: &[u8], offset: usize, len: usize) -> (Option<u8>, &[u8], Option<u8>) {
     let start_byte = offset / 8;
     let start_bit = offset % 8;
     let end_bit = offset + len;
@@ -408,7 +402,7 @@ fn byte_aligned_region(
     let tail_len = len - consumed;
     let tail = (tail_len != 0).then(|| mask_partial_byte(bytes[middle_end], 0, tail_len));
 
-    Some((head, middle, tail))
+    (head, middle, tail)
 }
 
 #[inline]
@@ -428,7 +422,7 @@ fn mask_partial_byte(byte: u8, bit_offset: usize, bit_len: usize) -> u8 {
 
 #[inline]
 fn count_aligned_bytes(bytes: &[u8]) -> usize {
-    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    #[cfg(target_arch = "x86_64")]
     {
         if bytes.len() >= 64
             && is_x86_feature_detected!("avx512f")