user0/rust
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Use generic simd for avx512 leading zeros

Browse source
This commit is contained in:
Tobias Decking 2024-06-23 15:38:57 +02:00 • committed by Amanieu d'Antras
parent d1004e0abd
commit a56cc86a23
1 changed files with 6 additions and 20 deletions
Download patch file Download diff file Expand all files Collapse all files

26
library/stdarch/crates/core_arch/src/x86/avx512cd.rs
View file

@ -294,7 +294,7 @@ pub unsafe fn _mm_maskz_conflict_epi64(k: __mmask8, a: __m128i) -> __m128i {
#[unstable(feature = "stdarch_x86_avx512", issue = "111137")]
#[cfg_attr(test, assert_instr(vplzcntd))]
pub unsafe fn _mm512_lzcnt_epi32(a: __m512i) -> __m512i {
transmute(vplzcntd(a.as_i32x16(), false))
transmute(simd_ctlz(a.as_i32x16()))
}
/// Counts the number of leading zero bits in each packed 32-bit integer in a, and store the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set).
@ -330,7 +330,7 @@ pub unsafe fn _mm512_maskz_lzcnt_epi32(k: __mmask16, a: __m512i) -> __m512i {
#[unstable(feature = "stdarch_x86_avx512", issue = "111137")]
#[cfg_attr(test, assert_instr(vplzcntd))]
pub unsafe fn _mm256_lzcnt_epi32(a: __m256i) -> __m256i {
transmute(vplzcntd256(a.as_i32x8(), false))
transmute(simd_ctlz(a.as_i32x8()))
}
/// Counts the number of leading zero bits in each packed 32-bit integer in a, and store the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set).
@ -366,7 +366,7 @@ pub unsafe fn _mm256_maskz_lzcnt_epi32(k: __mmask8, a: __m256i) -> __m256i {
#[unstable(feature = "stdarch_x86_avx512", issue = "111137")]
#[cfg_attr(test, assert_instr(vplzcntd))]
pub unsafe fn _mm_lzcnt_epi32(a: __m128i) -> __m128i {
transmute(vplzcntd128(a.as_i32x4(), false))
transmute(simd_ctlz(a.as_i32x4()))
}
/// Counts the number of leading zero bits in each packed 32-bit integer in a, and store the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set).
@ -402,7 +402,7 @@ pub unsafe fn _mm_maskz_lzcnt_epi32(k: __mmask8, a: __m128i) -> __m128i {
#[unstable(feature = "stdarch_x86_avx512", issue = "111137")]
#[cfg_attr(test, assert_instr(vplzcntq))]
pub unsafe fn _mm512_lzcnt_epi64(a: __m512i) -> __m512i {
transmute(vplzcntq(a.as_i64x8(), false))
transmute(simd_ctlz(a.as_i64x8()))
}
/// Counts the number of leading zero bits in each packed 64-bit integer in a, and store the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set).
@ -438,7 +438,7 @@ pub unsafe fn _mm512_maskz_lzcnt_epi64(k: __mmask8, a: __m512i) -> __m512i {
#[unstable(feature = "stdarch_x86_avx512", issue = "111137")]
#[cfg_attr(test, assert_instr(vplzcntq))]
pub unsafe fn _mm256_lzcnt_epi64(a: __m256i) -> __m256i {
transmute(vplzcntq256(a.as_i64x4(), false))
transmute(simd_ctlz(a.as_i64x4()))
}
/// Counts the number of leading zero bits in each packed 64-bit integer in a, and store the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set).
@ -474,7 +474,7 @@ pub unsafe fn _mm256_maskz_lzcnt_epi64(k: __mmask8, a: __m256i) -> __m256i {
#[unstable(feature = "stdarch_x86_avx512", issue = "111137")]
#[cfg_attr(test, assert_instr(vplzcntq))]
pub unsafe fn _mm_lzcnt_epi64(a: __m128i) -> __m128i {
transmute(vplzcntq128(a.as_i64x2(), false))
transmute(simd_ctlz(a.as_i64x2()))
}
/// Counts the number of leading zero bits in each packed 64-bit integer in a, and store the results in dst using writemask k (elements are copied from src when the corresponding mask bit is not set).
@ -517,20 +517,6 @@ extern "C" {
fn vpconflictq256(a: i64x4) -> i64x4;
#[link_name = "llvm.x86.avx512.conflict.q.128"]
fn vpconflictq128(a: i64x2) -> i64x2;
#[link_name = "llvm.ctlz.v16i32"]
fn vplzcntd(a: i32x16, nonzero: bool) -> i32x16;
#[link_name = "llvm.ctlz.v8i32"]
fn vplzcntd256(a: i32x8, nonzero: bool) -> i32x8;
#[link_name = "llvm.ctlz.v4i32"]
fn vplzcntd128(a: i32x4, nonzero: bool) -> i32x4;
#[link_name = "llvm.ctlz.v8i64"]
fn vplzcntq(a: i64x8, nonzero: bool) -> i64x8;
#[link_name = "llvm.ctlz.v4i64"]
fn vplzcntq256(a: i64x4, nonzero: bool) -> i64x4;
#[link_name = "llvm.ctlz.v2i64"]
fn vplzcntq128(a: i64x2, nonzero: bool) -> i64x2;
}
#[cfg(test)]