Skip to content
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

Use stdsimd instead of LLVM imports in more instances. #1584

Merged
merged 9 commits into from
Jun 16, 2024
Merged

Use stdsimd instead of LLVM imports in more instances. #1584

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
683868f
Update fma.rs
TDecking Jun 16, 2024
709043b
Update sse.rs
TDecking Jun 16, 2024
b825ce4
Update sse2.rs
TDecking Jun 16, 2024
429243e
Update avx.rs
TDecking Jun 16, 2024
ae27be5
Fix compiler errors
TDecking Jun 16, 2024
f84fdcd
Parenthesis
TDecking Jun 16, 2024
42348e9
Fix SSE sqrt
TDecking Jun 16, 2024
71c890e
Fix SSE2 sqrt
TDecking Jun 16, 2024
b04323f
Replace addsub variations
TDecking Jun 16, 2024
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
4 changes: 1 addition & 3 deletions crates/core_arch/src/x86/avx.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -441,7 +441,7 @@ pub unsafe fn _mm256_floor_ps(a: __m256) -> __m256 {
#[cfg_attr(test, assert_instr(vsqrtps))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_sqrt_ps(a: __m256) -> __m256 {
sqrtps256(a)
simd_fsqrt(a)
}

/// Returns the square root of packed double-precision (64-bit) floating point
Expand Down Expand Up @@ -2961,8 +2961,6 @@ extern "C" {
fn roundpd256(a: __m256d, b: i32) -> __m256d;
#[link_name = "llvm.x86.avx.round.ps.256"]
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.dp.ps.256"]
fn vdpps(a: __m256, b: __m256, imm8: i32) -> __m256;
#[link_name = "llvm.x86.avx.hadd.pd.256"]
Expand Down
167 changes: 78 additions & 89 deletions crates/core_arch/src/x86/fma.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,8 @@
//! [wiki_fma]: https://en.wikipedia.org/wiki/Fused_multiply-accumulate

use crate::core_arch::x86::*;
use crate::intrinsics::simd::simd_fma;
use crate::intrinsics::simd::{simd_fma, simd_insert, simd_neg, simd_shuffle};
use crate::intrinsics::{fmaf32, fmaf64};

#[cfg(test)]
use stdarch_test::assert_instr;
Expand Down Expand Up @@ -83,7 +84,11 @@ pub unsafe fn _mm256_fmadd_ps(a: __m256, b: __m256, c: __m256) -> __m256 {
#[cfg_attr(test, assert_instr(vfmadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fmadd_sd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
vfmaddsd(a, b, c)
simd_insert!(
a,
0,
fmaf64(_mm_cvtsd_f64(a), _mm_cvtsd_f64(b), _mm_cvtsd_f64(c))
)
}

/// Multiplies the lower single-precision (32-bit) floating-point elements in
Expand All @@ -97,7 +102,11 @@ pub unsafe fn _mm_fmadd_sd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
#[cfg_attr(test, assert_instr(vfmadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fmadd_ss(a: __m128, b: __m128, c: __m128) -> __m128 {
vfmaddss(a, b, c)
simd_insert!(
a,
0,
fmaf32(_mm_cvtss_f32(a), _mm_cvtss_f32(b), _mm_cvtss_f32(c))
)
}

/// Multiplies packed double-precision (64-bit) floating-point elements in `a`
Expand All @@ -110,7 +119,9 @@ pub unsafe fn _mm_fmadd_ss(a: __m128, b: __m128, c: __m128) -> __m128 {
#[cfg_attr(test, assert_instr(vfmaddsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fmaddsub_pd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
vfmaddsubpd(a, b, c)
let add = simd_fma(a, b, c);
let sub = simd_fma(a, b, simd_neg(c));
simd_shuffle!(add, sub, [2, 1])
}

/// Multiplies packed double-precision (64-bit) floating-point elements in `a`
Expand All @@ -123,7 +134,9 @@ pub unsafe fn _mm_fmaddsub_pd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
#[cfg_attr(test, assert_instr(vfmaddsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_fmaddsub_pd(a: __m256d, b: __m256d, c: __m256d) -> __m256d {
vfmaddsubpd256(a, b, c)
let add = simd_fma(a, b, c);
let sub = simd_fma(a, b, simd_neg(c));
simd_shuffle!(add, sub, [4, 1, 6, 3])
}

/// Multiplies packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -136,7 +149,9 @@ pub unsafe fn _mm256_fmaddsub_pd(a: __m256d, b: __m256d, c: __m256d) -> __m256d
#[cfg_attr(test, assert_instr(vfmaddsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fmaddsub_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
vfmaddsubps(a, b, c)
let add = simd_fma(a, b, c);
let sub = simd_fma(a, b, simd_neg(c));
simd_shuffle!(add, sub, [4, 1, 6, 3])
}

/// Multiplies packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -149,7 +164,9 @@ pub unsafe fn _mm_fmaddsub_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
#[cfg_attr(test, assert_instr(vfmaddsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_fmaddsub_ps(a: __m256, b: __m256, c: __m256) -> __m256 {
vfmaddsubps256(a, b, c)
let add = simd_fma(a, b, c);
let sub = simd_fma(a, b, simd_neg(c));
simd_shuffle!(add, sub, [8, 1, 10, 3, 12, 5, 14, 7])
}

/// Multiplies packed double-precision (64-bit) floating-point elements in `a`
Expand All @@ -161,7 +178,7 @@ pub unsafe fn _mm256_fmaddsub_ps(a: __m256, b: __m256, c: __m256) -> __m256 {
#[cfg_attr(test, assert_instr(vfmsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fmsub_pd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
vfmsubpd(a, b, c)
simd_fma(a, b, simd_neg(c))
}

/// Multiplies packed double-precision (64-bit) floating-point elements in `a`
Expand All @@ -173,7 +190,7 @@ pub unsafe fn _mm_fmsub_pd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
#[cfg_attr(test, assert_instr(vfmsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_fmsub_pd(a: __m256d, b: __m256d, c: __m256d) -> __m256d {
vfmsubpd256(a, b, c)
simd_fma(a, b, simd_neg(c))
}

/// Multiplies packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -185,7 +202,7 @@ pub unsafe fn _mm256_fmsub_pd(a: __m256d, b: __m256d, c: __m256d) -> __m256d {
#[cfg_attr(test, assert_instr(vfmsub213ps))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fmsub_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
vfmsubps(a, b, c)
simd_fma(a, b, simd_neg(c))
}

/// Multiplies packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -197,7 +214,7 @@ pub unsafe fn _mm_fmsub_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
#[cfg_attr(test, assert_instr(vfmsub213ps))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_fmsub_ps(a: __m256, b: __m256, c: __m256) -> __m256 {
vfmsubps256(a, b, c)
simd_fma(a, b, simd_neg(c))
}

/// Multiplies the lower double-precision (64-bit) floating-point elements in
Expand All @@ -211,7 +228,11 @@ pub unsafe fn _mm256_fmsub_ps(a: __m256, b: __m256, c: __m256) -> __m256 {
#[cfg_attr(test, assert_instr(vfmsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fmsub_sd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
vfmsubsd(a, b, c)
simd_insert!(
a,
0,
fmaf64(_mm_cvtsd_f64(a), _mm_cvtsd_f64(b), -_mm_cvtsd_f64(c))
)
}

/// Multiplies the lower single-precision (32-bit) floating-point elements in
Expand All @@ -225,7 +246,11 @@ pub unsafe fn _mm_fmsub_sd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
#[cfg_attr(test, assert_instr(vfmsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fmsub_ss(a: __m128, b: __m128, c: __m128) -> __m128 {
vfmsubss(a, b, c)
simd_insert!(
a,
0,
fmaf32(_mm_cvtss_f32(a), _mm_cvtss_f32(b), -_mm_cvtss_f32(c))
)
}

/// Multiplies packed double-precision (64-bit) floating-point elements in `a`
Expand All @@ -238,7 +263,9 @@ pub unsafe fn _mm_fmsub_ss(a: __m128, b: __m128, c: __m128) -> __m128 {
#[cfg_attr(test, assert_instr(vfmsubadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fmsubadd_pd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
vfmsubaddpd(a, b, c)
let add = simd_fma(a, b, c);
let sub = simd_fma(a, b, simd_neg(c));
simd_shuffle!(add, sub, [0, 3])
}

/// Multiplies packed double-precision (64-bit) floating-point elements in `a`
Expand All @@ -251,7 +278,9 @@ pub unsafe fn _mm_fmsubadd_pd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
#[cfg_attr(test, assert_instr(vfmsubadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_fmsubadd_pd(a: __m256d, b: __m256d, c: __m256d) -> __m256d {
vfmsubaddpd256(a, b, c)
let add = simd_fma(a, b, c);
let sub = simd_fma(a, b, simd_neg(c));
simd_shuffle!(add, sub, [0, 5, 2, 7])
}

/// Multiplies packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -264,7 +293,9 @@ pub unsafe fn _mm256_fmsubadd_pd(a: __m256d, b: __m256d, c: __m256d) -> __m256d
#[cfg_attr(test, assert_instr(vfmsubadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fmsubadd_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
vfmsubaddps(a, b, c)
let add = simd_fma(a, b, c);
let sub = simd_fma(a, b, simd_neg(c));
simd_shuffle!(add, sub, [0, 5, 2, 7])
}

/// Multiplies packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -277,7 +308,9 @@ pub unsafe fn _mm_fmsubadd_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
#[cfg_attr(test, assert_instr(vfmsubadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_fmsubadd_ps(a: __m256, b: __m256, c: __m256) -> __m256 {
vfmsubaddps256(a, b, c)
let add = simd_fma(a, b, c);
let sub = simd_fma(a, b, simd_neg(c));
simd_shuffle!(add, sub, [0, 9, 2, 11, 4, 13, 6, 15])
}

/// Multiplies packed double-precision (64-bit) floating-point elements in `a`
Expand All @@ -289,7 +322,7 @@ pub unsafe fn _mm256_fmsubadd_ps(a: __m256, b: __m256, c: __m256) -> __m256 {
#[cfg_attr(test, assert_instr(vfnmadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fnmadd_pd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
vfnmaddpd(a, b, c)
simd_fma(simd_neg(a), b, c)
}

/// Multiplies packed double-precision (64-bit) floating-point elements in `a`
Expand All @@ -301,7 +334,7 @@ pub unsafe fn _mm_fnmadd_pd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
#[cfg_attr(test, assert_instr(vfnmadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_fnmadd_pd(a: __m256d, b: __m256d, c: __m256d) -> __m256d {
vfnmaddpd256(a, b, c)
simd_fma(simd_neg(a), b, c)
}

/// Multiplies packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -313,7 +346,7 @@ pub unsafe fn _mm256_fnmadd_pd(a: __m256d, b: __m256d, c: __m256d) -> __m256d {
#[cfg_attr(test, assert_instr(vfnmadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fnmadd_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
vfnmaddps(a, b, c)
simd_fma(simd_neg(a), b, c)
}

/// Multiplies packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -325,7 +358,7 @@ pub unsafe fn _mm_fnmadd_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
#[cfg_attr(test, assert_instr(vfnmadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_fnmadd_ps(a: __m256, b: __m256, c: __m256) -> __m256 {
vfnmaddps256(a, b, c)
simd_fma(simd_neg(a), b, c)
}

/// Multiplies the lower double-precision (64-bit) floating-point elements in
Expand All @@ -339,7 +372,11 @@ pub unsafe fn _mm256_fnmadd_ps(a: __m256, b: __m256, c: __m256) -> __m256 {
#[cfg_attr(test, assert_instr(vfnmadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fnmadd_sd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
vfnmaddsd(a, b, c)
simd_insert!(
a,
0,
fmaf64(_mm_cvtsd_f64(a), -_mm_cvtsd_f64(b), _mm_cvtsd_f64(c))
)
}

/// Multiplies the lower single-precision (32-bit) floating-point elements in
Expand All @@ -353,7 +390,11 @@ pub unsafe fn _mm_fnmadd_sd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
#[cfg_attr(test, assert_instr(vfnmadd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fnmadd_ss(a: __m128, b: __m128, c: __m128) -> __m128 {
vfnmaddss(a, b, c)
simd_insert!(
a,
0,
fmaf32(_mm_cvtss_f32(a), -_mm_cvtss_f32(b), _mm_cvtss_f32(c))
)
}

/// Multiplies packed double-precision (64-bit) floating-point elements in `a`
Expand All @@ -366,7 +407,7 @@ pub unsafe fn _mm_fnmadd_ss(a: __m128, b: __m128, c: __m128) -> __m128 {
#[cfg_attr(test, assert_instr(vfnmsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fnmsub_pd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
vfnmsubpd(a, b, c)
simd_fma(simd_neg(a), b, simd_neg(c))
}

/// Multiplies packed double-precision (64-bit) floating-point elements in `a`
Expand All @@ -379,7 +420,7 @@ pub unsafe fn _mm_fnmsub_pd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
#[cfg_attr(test, assert_instr(vfnmsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_fnmsub_pd(a: __m256d, b: __m256d, c: __m256d) -> __m256d {
vfnmsubpd256(a, b, c)
simd_fma(simd_neg(a), b, simd_neg(c))
}

/// Multiplies packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -392,7 +433,7 @@ pub unsafe fn _mm256_fnmsub_pd(a: __m256d, b: __m256d, c: __m256d) -> __m256d {
#[cfg_attr(test, assert_instr(vfnmsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fnmsub_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
vfnmsubps(a, b, c)
simd_fma(simd_neg(a), b, simd_neg(c))
}

/// Multiplies packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -405,7 +446,7 @@ pub unsafe fn _mm_fnmsub_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
#[cfg_attr(test, assert_instr(vfnmsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_fnmsub_ps(a: __m256, b: __m256, c: __m256) -> __m256 {
vfnmsubps256(a, b, c)
simd_fma(simd_neg(a), b, simd_neg(c))
}

/// Multiplies the lower double-precision (64-bit) floating-point elements in
Expand All @@ -420,7 +461,11 @@ pub unsafe fn _mm256_fnmsub_ps(a: __m256, b: __m256, c: __m256) -> __m256 {
#[cfg_attr(test, assert_instr(vfnmsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fnmsub_sd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
vfnmsubsd(a, b, c)
simd_insert!(
a,
0,
fmaf64(_mm_cvtsd_f64(a), -_mm_cvtsd_f64(b), -_mm_cvtsd_f64(c))
)
}

/// Multiplies the lower single-precision (32-bit) floating-point elements in
Expand All @@ -435,67 +480,11 @@ pub unsafe fn _mm_fnmsub_sd(a: __m128d, b: __m128d, c: __m128d) -> __m128d {
#[cfg_attr(test, assert_instr(vfnmsub))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_fnmsub_ss(a: __m128, b: __m128, c: __m128) -> __m128 {
vfnmsubss(a, b, c)
}

#[allow(improper_ctypes)]
extern "C" {
#[link_name = "llvm.x86.fma.vfmadd.sd"]
fn vfmaddsd(a: __m128d, b: __m128d, c: __m128d) -> __m128d;
#[link_name = "llvm.x86.fma.vfmadd.ss"]
fn vfmaddss(a: __m128, b: __m128, c: __m128) -> __m128;
#[link_name = "llvm.x86.fma.vfmaddsub.pd"]
fn vfmaddsubpd(a: __m128d, b: __m128d, c: __m128d) -> __m128d;
#[link_name = "llvm.x86.fma.vfmaddsub.pd.256"]
fn vfmaddsubpd256(a: __m256d, b: __m256d, c: __m256d) -> __m256d;
#[link_name = "llvm.x86.fma.vfmaddsub.ps"]
fn vfmaddsubps(a: __m128, b: __m128, c: __m128) -> __m128;
#[link_name = "llvm.x86.fma.vfmaddsub.ps.256"]
fn vfmaddsubps256(a: __m256, b: __m256, c: __m256) -> __m256;
#[link_name = "llvm.x86.fma.vfmsub.pd"]
fn vfmsubpd(a: __m128d, b: __m128d, c: __m128d) -> __m128d;
#[link_name = "llvm.x86.fma.vfmsub.pd.256"]
fn vfmsubpd256(a: __m256d, b: __m256d, c: __m256d) -> __m256d;
#[link_name = "llvm.x86.fma.vfmsub.ps"]
fn vfmsubps(a: __m128, b: __m128, c: __m128) -> __m128;
#[link_name = "llvm.x86.fma.vfmsub.ps.256"]
fn vfmsubps256(a: __m256, b: __m256, c: __m256) -> __m256;
#[link_name = "llvm.x86.fma.vfmsub.sd"]
fn vfmsubsd(a: __m128d, b: __m128d, c: __m128d) -> __m128d;
#[link_name = "llvm.x86.fma.vfmsub.ss"]
fn vfmsubss(a: __m128, b: __m128, c: __m128) -> __m128;
#[link_name = "llvm.x86.fma.vfmsubadd.pd"]
fn vfmsubaddpd(a: __m128d, b: __m128d, c: __m128d) -> __m128d;
#[link_name = "llvm.x86.fma.vfmsubadd.pd.256"]
fn vfmsubaddpd256(a: __m256d, b: __m256d, c: __m256d) -> __m256d;
#[link_name = "llvm.x86.fma.vfmsubadd.ps"]
fn vfmsubaddps(a: __m128, b: __m128, c: __m128) -> __m128;
#[link_name = "llvm.x86.fma.vfmsubadd.ps.256"]
fn vfmsubaddps256(a: __m256, b: __m256, c: __m256) -> __m256;
#[link_name = "llvm.x86.fma.vfnmadd.pd"]
fn vfnmaddpd(a: __m128d, b: __m128d, c: __m128d) -> __m128d;
#[link_name = "llvm.x86.fma.vfnmadd.pd.256"]
fn vfnmaddpd256(a: __m256d, b: __m256d, c: __m256d) -> __m256d;
#[link_name = "llvm.x86.fma.vfnmadd.ps"]
fn vfnmaddps(a: __m128, b: __m128, c: __m128) -> __m128;
#[link_name = "llvm.x86.fma.vfnmadd.ps.256"]
fn vfnmaddps256(a: __m256, b: __m256, c: __m256) -> __m256;
#[link_name = "llvm.x86.fma.vfnmadd.sd"]
fn vfnmaddsd(a: __m128d, b: __m128d, c: __m128d) -> __m128d;
#[link_name = "llvm.x86.fma.vfnmadd.ss"]
fn vfnmaddss(a: __m128, b: __m128, c: __m128) -> __m128;
#[link_name = "llvm.x86.fma.vfnmsub.pd"]
fn vfnmsubpd(a: __m128d, b: __m128d, c: __m128d) -> __m128d;
#[link_name = "llvm.x86.fma.vfnmsub.pd.256"]
fn vfnmsubpd256(a: __m256d, b: __m256d, c: __m256d) -> __m256d;
#[link_name = "llvm.x86.fma.vfnmsub.ps"]
fn vfnmsubps(a: __m128, b: __m128, c: __m128) -> __m128;
#[link_name = "llvm.x86.fma.vfnmsub.ps.256"]
fn vfnmsubps256(a: __m256, b: __m256, c: __m256) -> __m256;
#[link_name = "llvm.x86.fma.vfnmsub.sd"]
fn vfnmsubsd(a: __m128d, b: __m128d, c: __m128d) -> __m128d;
#[link_name = "llvm.x86.fma.vfnmsub.ss"]
fn vfnmsubss(a: __m128, b: __m128, c: __m128) -> __m128;
simd_insert!(
a,
0,
fmaf32(_mm_cvtss_f32(a), -_mm_cvtss_f32(b), -_mm_cvtss_f32(c))
)
}

#[cfg(test)]
Expand Down
Loading
Loading