Skip to content

Fix swizzle_dyn #395

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
Feb 18, 2024
Merged

Fix swizzle_dyn #395

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
44 changes: 12 additions & 32 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 @@ -44,7 +44,7 @@ where
))]
8 => transize(vtbl1_u8, self, idxs),
#[cfg(target_feature = "ssse3")]
16 => transize(x86::_mm_shuffle_epi8, self, idxs),
16 => transize(x86::_mm_shuffle_epi8, self, zeroing_idxs(idxs)),
#[cfg(target_feature = "simd128")]
16 => transize(wasm::i8x16_swizzle, self, idxs),
#[cfg(all(
Expand All @@ -54,9 +54,9 @@ where
))]
16 => transize(vqtbl1q_u8, self, idxs),
#[cfg(all(target_feature = "avx2", not(target_feature = "avx512vbmi")))]
32 => transize_raw(avx2_pshufb, self, idxs),
32 => transize(avx2_pshufb, self, idxs),
#[cfg(all(target_feature = "avx512vl", target_feature = "avx512vbmi"))]
32 => transize(x86::_mm256_permutexvar_epi8, self, idxs),
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
Expand Down Expand Up @@ -129,45 +129,25 @@ unsafe fn avx2_pshufb(bytes: Simd<u8, 32>, idxs: Simd<u8, 32>) -> Simd<u8, 32> {
#[inline(always)]
unsafe fn transize<T, const N: usize>(
f: unsafe fn(T, T) -> T,
bytes: Simd<u8, N>,
idxs: Simd<u8, N>,
a: Simd<u8, N>,
b: Simd<u8, N>,
) -> Simd<u8, N>
where
LaneCount<N>: SupportedLaneCount,
{
let idxs = zeroing_idxs(idxs);
// SAFETY: Same obligation to use this function as to use mem::transmute_copy.
unsafe { mem::transmute_copy(&f(mem::transmute_copy(&bytes), mem::transmute_copy(&idxs))) }
unsafe { mem::transmute_copy(&f(mem::transmute_copy(&a), mem::transmute_copy(&b))) }
}

/// Make indices that yield 0 for this architecture
/// Make indices that yield 0 for x86
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
#[allow(unused)]
#[inline(always)]
fn zeroing_idxs<const N: usize>(idxs: Simd<u8, N>) -> Simd<u8, N>
where
LaneCount<N>: SupportedLaneCount,
{
// On x86, make sure the top bit is set.
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
let idxs = {
use crate::simd::cmp::SimdPartialOrd;
idxs.simd_lt(Simd::splat(N as u8))
.select(idxs, Simd::splat(u8::MAX))
};
// Simply do nothing on most architectures.
idxs
}

/// As transize but no implicit call to `zeroing_idxs`.
#[allow(dead_code)]
#[inline(always)]
unsafe fn transize_raw<T, const N: usize>(
f: unsafe fn(T, T) -> T,
bytes: Simd<u8, N>,
idxs: Simd<u8, N>,
) -> Simd<u8, N>
where
LaneCount<N>: SupportedLaneCount,
{
// SAFETY: Same obligation to use this function as to use mem::transmute_copy.
unsafe { mem::transmute_copy(&f(mem::transmute_copy(&bytes), mem::transmute_copy(&idxs))) }
use crate::simd::cmp::SimdPartialOrd;
idxs.simd_lt(Simd::splat(N as u8))
.select(idxs, Simd::splat(u8::MAX))
}