Convert _mm256_cmp_ps to const generics and fix imm width

This commit is contained in:
Rémy Rakic 2021-03-01 21:06:57 +01:00 committed by Amanieu d'Antras
parent c7a5963449
commit ea5801037d
2 changed files with 7 additions and 11 deletions

View file

@ -837,16 +837,12 @@ pub unsafe fn _mm_cmp_ps<const IMM8: i32>(a: __m128, b: __m128) -> __m128 {
/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_cmp_ps)
#[inline]
#[target_feature(enable = "avx")]
#[cfg_attr(test, assert_instr(vcmpeqps, imm8 = 0))] // TODO Validate vcmpps
#[rustc_args_required_const(2)]
#[cfg_attr(test, assert_instr(vcmpeqps, IMM8 = 0))] // TODO Validate vcmpps
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_cmp_ps(a: __m256, b: __m256, imm8: i32) -> __m256 {
macro_rules! call {
($imm8:expr) => {
vcmpps256(a, b, $imm8)
};
}
constify_imm6!(imm8, call)
pub unsafe fn _mm256_cmp_ps<const IMM8: i32>(a: __m256, b: __m256) -> __m256 {
static_assert_imm5!(IMM8);
vcmpps256(a, b, IMM8 as u8)
}
/// Compares the lower double-precision (64-bit) floating-point element in
@ -3652,7 +3648,7 @@ mod tests {
unsafe fn test_mm256_cmp_ps() {
let a = _mm256_setr_ps(1., 2., 3., 4., 1., 2., 3., 4.);
let b = _mm256_setr_ps(5., 6., 7., 8., 5., 6., 7., 8.);
let r = _mm256_cmp_ps(a, b, _CMP_GE_OS);
let r = _mm256_cmp_ps::<_CMP_GE_OS>(a, b);
let e = _mm256_set1_ps(0.);
assert_eq_m256(r, e);
}

View file

@ -60,7 +60,7 @@ pub unsafe fn get_m256d(a: __m256d, idx: usize) -> f64 {
#[target_feature(enable = "avx")]
pub unsafe fn assert_eq_m256(a: __m256, b: __m256) {
let cmp = _mm256_cmp_ps(a, b, _CMP_EQ_OQ);
let cmp = _mm256_cmp_ps::<_CMP_EQ_OQ>(a, b);
if _mm256_movemask_ps(cmp) != 0b11111111 {
panic!("{:?} != {:?}", a, b);
}