convert _mm256_mask_i32gather_epi32 to const generics

This commit is contained in:
Rémy Rakic 2021-03-06 00:39:33 +01:00 committed by Amanieu d'Antras
parent a5ca4611ef
commit d69a908608

View file

@ -1158,26 +1158,21 @@ pub unsafe fn _mm256_i32gather_epi32<const SCALE: i32>(
/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_mask_i32gather_epi32)
#[inline]
#[target_feature(enable = "avx2")]
#[cfg_attr(test, assert_instr(vpgatherdd, scale = 1))]
#[rustc_args_required_const(4)]
#[cfg_attr(test, assert_instr(vpgatherdd, SCALE = 1))]
#[rustc_legacy_const_generics(4)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_mask_i32gather_epi32(
pub unsafe fn _mm256_mask_i32gather_epi32<const SCALE: i32>(
src: __m256i,
slice: *const i32,
offsets: __m256i,
mask: __m256i,
scale: i32,
) -> __m256i {
static_assert_imm8_scale!(SCALE);
let src = src.as_i32x8();
let mask = mask.as_i32x8();
let offsets = offsets.as_i32x8();
let slice = slice as *const i8;
macro_rules! call {
($imm8:expr) => {
vpgatherdd(src, slice, offsets, mask, $imm8)
};
}
let r = constify_imm8_gather!(scale, call);
let r = vpgatherdd(src, slice, offsets, mask, SCALE as i8);
transmute(r)
}
@ -5605,12 +5600,11 @@ mod tests {
arr[i as usize] = i;
}
// A multiplier of 4 is word-addressing
let r = _mm256_mask_i32gather_epi32(
let r = _mm256_mask_i32gather_epi32::<4>(
_mm256_set1_epi32(256),
arr.as_ptr(),
_mm256_setr_epi32(0, 16, 64, 96, 0, 0, 0, 0),
_mm256_setr_epi32(-1, -1, -1, 0, 0, 0, 0, 0),
4,
);
assert_eq_m256i(r, _mm256_setr_epi32(0, 16, 64, 256, 256, 256, 256, 256));
}