Dandandan commented on code in PR #9297:
URL: https://github.com/apache/arrow-rs/pull/9297#discussion_r2773597338
##########
arrow-buffer/src/buffer/boolean.rs:
##########
@@ -191,67 +189,48 @@ impl BooleanBuffer {
where
F: FnMut(u64) -> u64,
{
- // try fast path for aligned input
- if offset_in_bits & 0x7 == 0 {
- // align to byte boundary
- let aligned = &src.as_ref()[offset_in_bits / 8..];
- if let Some(result) =
- Self::try_from_aligned_bitwise_unary_op(aligned, len_in_bits,
&mut op)
- {
- return result;
+ let end = offset_in_bits + len_in_bits;
+ // Align start and end to 64 bit (8 byte) boundaries if possible to
allow using the
+ // optimized code path as much as possible.
+ let aligned_offset = offset_in_bits & !63;
+ let aligned_end_bytes = bit_util::ceil(end, 64) * 8;
+ let src_len = src.as_ref().len();
+ let slice_end = aligned_end_bytes.min(src_len);
+
+ let aligned_start = &src.as_ref()[aligned_offset / 8..slice_end];
+
+ let (prefix, aligned_u64s, suffix) = unsafe {
aligned_start.as_ref().align_to::<u64>() };
Review Comment:
I *think* previous benchmark results are sometimes noisy because underlying
buffer is not always aligned to 64 bits (not 100% sure but it would explain it)
and then take the slower (unaligned) path.
Now the slower path is not much slower. We probably want to make sure the
array creation path aligns to u64 in most cases and we also keep the alignment.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
