mikros/rust
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Auto merge of #116018 - DianQK:simd-wide-sum-test, r=scottmcm

Browse Source 
Increasing the SIMD size improves the vectorization possibilities

Change the `simd-wide-sum.rs` to pass tests based on the LLVM main branch.

For smaller lengths, we cannot expect to always get vectorized.

A related discussion at https://rust-lang.zulipchat.com/#narrow/stream/187780-t-compiler.2Fwg-llvm/topic/LLVM.20HEAD.3A.20codegen.2Fsimd.2Fsimd-wide-sum.2Ers.20newly.20failing.

r? scottmcm
This commit is contained in:
bors 2023-10-06 08:04:53 +00:00
commit d4ba2b4c7c
1 changed files with 13 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
tests/codegen/simd/simd-wide-sum.rs
View File

@ -11,14 +11,14 @@
#![feature(portable_simd)] #![feature(portable_simd)]
use std::simd::{Simd, SimdUint}; use std::simd::{Simd, SimdUint};
const N: usize = 8; const N: usize = 16;
#[no_mangle] #[no_mangle]
// CHECK-LABEL: @wider_reduce_simd // CHECK-LABEL: @wider_reduce_simd
pub fn wider_reduce_simd(x: Simd<u8, N>) -> u16 { pub fn wider_reduce_simd(x: Simd<u8, N>) -> u16 {
// CHECK: zext <8 x i8> // CHECK: zext <16 x i8>
// CHECK-SAME: to <8 x i16> // CHECK-SAME: to <16 x i16>
// CHECK: call i16 @llvm.vector.reduce.add.v8i16(<8 x i16> // CHECK: call i16 @llvm.vector.reduce.add.v16i16(<16 x i16>
let x: Simd<u16, N> = x.cast(); let x: Simd<u16, N> = x.cast();
x.reduce_sum() x.reduce_sum()
} }
@ -26,9 +26,9 @@ pub fn wider_reduce_simd(x: Simd<u8, N>) -> u16 {
#[no_mangle] #[no_mangle]
// CHECK-LABEL: @wider_reduce_loop // CHECK-LABEL: @wider_reduce_loop
pub fn wider_reduce_loop(x: Simd<u8, N>) -> u16 { pub fn wider_reduce_loop(x: Simd<u8, N>) -> u16 {
// CHECK: zext <8 x i8> // CHECK: zext <16 x i8>
// CHECK-SAME: to <8 x i16> // CHECK-SAME: to <16 x i16>
// CHECK: call i16 @llvm.vector.reduce.add.v8i16(<8 x i16> // CHECK: call i16 @llvm.vector.reduce.add.v16i16(<16 x i16>
let mut sum = 0_u16; let mut sum = 0_u16;
for i in 0..N { for i in 0..N {
sum += u16::from(x[i]); sum += u16::from(x[i]);
@ -39,9 +39,9 @@ pub fn wider_reduce_loop(x: Simd<u8, N>) -> u16 {
#[no_mangle] #[no_mangle]
// CHECK-LABEL: @wider_reduce_iter // CHECK-LABEL: @wider_reduce_iter
pub fn wider_reduce_iter(x: Simd<u8, N>) -> u16 { pub fn wider_reduce_iter(x: Simd<u8, N>) -> u16 {
// CHECK: zext <8 x i8> // CHECK: zext <16 x i8>
// CHECK-SAME: to <8 x i16> // CHECK-SAME: to <16 x i16>
// CHECK: call i16 @llvm.vector.reduce.add.v8i16(<8 x i16> // CHECK: call i16 @llvm.vector.reduce.add.v16i16(<16 x i16>
x.as_array().iter().copied().map(u16::from).sum() x.as_array().iter().copied().map(u16::from).sum()
} }
@ -52,8 +52,8 @@ pub fn wider_reduce_iter(x: Simd<u8, N>) -> u16 {
#[no_mangle] #[no_mangle]
// CHECK-LABEL: @wider_reduce_into_iter // CHECK-LABEL: @wider_reduce_into_iter
pub fn wider_reduce_into_iter(x: Simd<u8, N>) -> u16 { pub fn wider_reduce_into_iter(x: Simd<u8, N>) -> u16 {
// CHECK: zext <8 x i8> // FIXME: It would be nice if this was exactly the same as the above tests,
// CHECK-SAME: to <8 x i16> // but at the time of writing this comment, that didn't happen on LLVM main.
// CHECK: call i16 @llvm.vector.reduce.add.v8i16(<8 x i16> // CHECK: call i16 @llvm.vector.reduce.add
x.to_array().into_iter().map(u16::from).sum() x.to_array().into_iter().map(u16::from).sum()
} }