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

Reimplement _mm_blendv_pd and _mm256_blendv_pd without LLVM intrinsics

Browse source
This commit is contained in:
Eduardo Sánchez Muñoz 2023-10-04 19:47:25 +02:00 • committed by Amanieu d'Antras
parent 81b0d7f279
commit 2534365aee
2 changed files with 4 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

5
library/stdarch/crates/core_arch/src/x86/avx.rs
View file

@ -511,7 +511,8 @@ pub unsafe fn _mm256_blend_ps<const IMM8: i32>(a: __m256, b: __m256) -> __m256 {
#[cfg_attr(test, assert_instr(vblendvpd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_blendv_pd(a: __m256d, b: __m256d, c: __m256d) -> __m256d {
vblendvpd(a, b, c)
let mask: i64x4 = simd_lt(transmute::<_, i64x4>(c), i64x4::splat(0));
transmute(simd_select(mask, b.as_f64x4(), a.as_f64x4()))
}
/// Blends packed single-precision (32-bit) floating-point elements from
@ -2914,8 +2915,6 @@ extern "C" {
fn roundps256(a: __m256, b: i32) -> __m256;
#[link_name = "llvm.x86.avx.sqrt.ps.256"]
fn sqrtps256(a: __m256) -> __m256;
#[link_name = "llvm.x86.avx.blendv.pd.256"]
fn vblendvpd(a: __m256d, b: __m256d, c: __m256d) -> __m256d;
#[link_name = "llvm.x86.avx.blendv.ps.256"]
fn vblendvps(a: __m256, b: __m256, c: __m256) -> __m256;
#[link_name = "llvm.x86.avx.dp.ps.256"]

5
library/stdarch/crates/core_arch/src/x86/sse41.rs
View file

@ -105,7 +105,8 @@ pub unsafe fn _mm_blend_epi16<const IMM8: i32>(a: __m128i, b: __m128i) -> __m128
#[cfg_attr(test, assert_instr(blendvpd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_blendv_pd(a: __m128d, b: __m128d, mask: __m128d) -> __m128d {
blendvpd(a, b, mask)
let mask: i64x2 = simd_lt(transmute::<_, i64x2>(mask), i64x2::splat(0));
transmute(simd_select(mask, b.as_f64x2(), a.as_f64x2()))
}
/// Blend packed single-precision (32-bit) floating-point elements from `a`
@ -1137,8 +1138,6 @@ pub unsafe fn _mm_test_mix_ones_zeros(a: __m128i, mask: __m128i) -> i32 {
#[allow(improper_ctypes)]
extern "C" {
#[link_name = "llvm.x86.sse41.blendvpd"]
fn blendvpd(a: __m128d, b: __m128d, mask: __m128d) -> __m128d;
#[link_name = "llvm.x86.sse41.blendvps"]
fn blendvps(a: __m128, b: __m128, mask: __m128) -> __m128;
#[link_name = "llvm.x86.sse41.blendpd"]