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

Change Simd::splat to not generate a loop

Browse source 
This fixes poor codegen in some circumstances for `u16x8::splat` on x86_64
https://rust-lang.zulipchat.com/#narrow/stream/257879-project-portable-simd/topic/Very.20bad.20.60u16x8.3A.3Asplat.60.20codegen.20on.20x86_64
This commit is contained in:
Jacob Lifshay 2022-06-06 12:16:17 -07:00
parent f237f133c1
commit c9636158d9
1 changed files with 9 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

13
crates/core_simd/src/vector.rs
View file

@ -9,8 +9,9 @@ pub use uint::*;
// Vectors of pointers are not for public use at the current time.
pub(crate) mod ptr;
use crate::simd::intrinsics;
use crate::simd::{LaneCount, Mask, MaskElement, SimdPartialOrd, SupportedLaneCount};
use crate::simd::{
intrinsics, LaneCount, Mask, MaskElement, SimdPartialOrd, SupportedLaneCount, Swizzle,
};
/// A SIMD vector of `LANES` elements of type `T`. `Simd<T, N>` has the same shape as [`[T; N]`](array), but operates like `T`.
///
@ -123,8 +124,12 @@ where
/// let v = u32x4::splat(8);
/// assert_eq!(v.as_array(), &[8, 8, 8, 8]);
/// ```
pub const fn splat(value: T) -> Self {
Self([value; LANES])
pub fn splat(value: T) -> Self {
struct Splat;
impl<const LANES: usize> Swizzle<1, LANES> for Splat {
const INDEX: [usize; LANES] = [0; LANES];
}
Splat::swizzle(Simd::<T, 1>::from([value]))
}
/// Returns an array reference containing the entire SIMD vector.