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

add vpmaddwd tests back in

Browse source
This commit is contained in:
Folkert de Vries 2026-02-02 17:01:10 +01:00
parent 821f139a29
commit 54641efc68
No known key found for this signature in database
GPG key ID: 1F17F6FFD112B97C
1 changed files with 20 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

20
library/stdarch/crates/core_arch/src/x86/avx2.rs
View file

@ -4677,6 +4677,26 @@ mod tests {
assert_eq_m256i(r, e);
}
#[target_feature(enable = "avx2")]
#[cfg_attr(test, assert_instr(vpmaddwd))]
unsafe fn test_mm256_madd_epi16_mul_one(v: __m256i) -> __m256i {
// This is a trick used in the adler32 algorithm to get a widening addition. The
// multiplication by 1 is trivial, but must not be optimized out because then the vpmaddwd
// instruction is no longer selected. The assert_instr verifies that this is the case.
let one_v = _mm256_set1_epi16(1);
_mm256_madd_epi16(v, one_v)
}
#[target_feature(enable = "avx2")]
#[cfg_attr(test, assert_instr(vpmaddwd))]
unsafe fn test_mm256_madd_epi16_shl(v: __m256i) -> __m256i {
// This is a trick used in the base64 algorithm to get a widening addition. Instead of a
// multiplication, a vector shl is used. In LLVM 22 that breaks the pattern recognition
// for the automatic optimization to vpmaddwd.
let shift_value = _mm256_set1_epi32(12i32);
_mm256_madd_epi16(v, shift_value)
}
#[simd_test(enable = "avx2")]
const fn test_mm256_inserti128_si256() {
let a = _mm256_setr_epi64x(1, 2, 3, 4);