squiid/rust
1
0
Fork 0
Code Issues Pull requests Projects Activity

Enable i586 workaround for both f32 and f64

Browse source
This commit is contained in:
Caleb Zulawski 2021-03-12 20:31:30 -05:00
parent 3cf970fc09
commit e2fa502617
1 changed files with 4 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
crates/core_simd/src/reduction.rs
View file

@ -59,8 +59,8 @@ macro_rules! impl_float_reductions {
/// Produces the sum of the lanes of the vector.
#[inline]
pub fn sum(self) -> $scalar {
// f32 SIMD sum is inaccurate on i586
if cfg!(all(target_arch = "x86", not(target_feature = "sse2"))) && core::mem::size_of::<$scalar>() == 4 {
// LLVM sum is inaccurate on i586
if cfg!(all(target_arch = "x86", not(target_feature = "sse2"))) {
self.as_slice().iter().sum()
} else {
unsafe { crate::intrinsics::simd_reduce_add_ordered(self, 0.) }
@ -70,8 +70,8 @@ macro_rules! impl_float_reductions {
/// Produces the sum of the lanes of the vector.
#[inline]
pub fn product(self) -> $scalar {
// f32 SIMD product is inaccurate on i586
if cfg!(all(target_arch = "x86", not(target_feature = "sse2"))) && core::mem::size_of::<$scalar>() == 4 {
// LLVM product is inaccurate on i586
if cfg!(all(target_arch = "x86", not(target_feature = "sse2"))) {
self.as_slice().iter().product()
} else {
unsafe { crate::intrinsics::simd_reduce_mul_ordered(self, 1.) }