Skip to content

Fix avx512vbmi swizzle_dyn implementation #431

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Aug 27, 2024
Merged

Fix avx512vbmi swizzle_dyn implementation #431

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
30 changes: 24 additions & 6 deletions crates/core_simd/src/swizzle_dyn.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,12 +60,30 @@ where
#[cfg(all(target_feature = "avx2", not(target_feature = "avx512vbmi")))]
32 => transize(avx2_pshufb, self, idxs),
#[cfg(all(target_feature = "avx512vl", target_feature = "avx512vbmi"))]
32 => transize(x86::_mm256_permutexvar_epi8, zeroing_idxs(idxs), self),
// Notable absence: avx512bw shuffle
// If avx512bw is available, odds of avx512vbmi are good
// FIXME: initial AVX512VBMI variant didn't actually pass muster
// #[cfg(target_feature = "avx512vbmi")]
// 64 => transize(x86::_mm512_permutexvar_epi8, self, idxs),
32 => {
// Unlike vpshufb, vpermb doesn't zero out values in the result based on the index high bit
let swizzler = |bytes, idxs| {
let mask = x86::_mm256_cmp_epu8_mask::<{ x86::_MM_CMPINT_LT }>(
idxs,
Simd::<u8, 32>::splat(N as u8).into(),
);
x86::_mm256_maskz_permutexvar_epi8(mask, idxs, bytes)
};
transize(swizzler, self, idxs)
}
// Notable absence: avx512bw pshufb shuffle
#[cfg(all(target_feature = "avx512vl", target_feature = "avx512vbmi"))]
64 => {
// Unlike vpshufb, vpermb doesn't zero out values in the result based on the index high bit
let swizzler = |bytes, idxs| {
let mask = x86::_mm512_cmp_epu8_mask::<{ x86::_MM_CMPINT_LT }>(
idxs,
Simd::<u8, 64>::splat(N as u8).into(),
);
x86::_mm512_maskz_permutexvar_epi8(mask, idxs, bytes)
};
transize(swizzler, self, idxs)
}
_ => {
let mut array = [0; N];
for (i, k) in idxs.to_array().into_iter().enumerate() {
Expand Down
Loading