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

convert _mm_blend_epi32 to const generics

Browse source
This commit is contained in:
Rémy Rakic 2021-03-06 04:39:23 +01:00 • committed by Amanieu d'Antras
parent 8a65e1a0fb
commit 65f6a147de
2 changed files with 26 additions and 34 deletions
Download patch file Download diff file Expand all files Collapse all files

45
library/stdarch/crates/core_arch/src/x86/avx2.rs
View file

@ -358,39 +358,28 @@ pub unsafe fn _mm256_avg_epu8(a: __m256i, b: __m256i) -> __m256i {
transmute(pavgb(a.as_u8x32(), b.as_u8x32()))
}
/// Blends packed 32-bit integers from `a` and `b` using control mask `imm8`.
/// Blends packed 32-bit integers from `a` and `b` using control mask `IMM4`.
///
/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_blend_epi32)
#[inline]
#[target_feature(enable = "avx2")]
#[cfg_attr(test, assert_instr(vblendps, imm8 = 9))]
#[rustc_args_required_const(2)]
#[cfg_attr(test, assert_instr(vblendps, IMM4 = 9))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_blend_epi32(a: __m128i, b: __m128i, imm8: i32) -> __m128i {
let imm8 = (imm8 & 0xFF) as u8;
pub unsafe fn _mm_blend_epi32<const IMM4: i32>(a: __m128i, b: __m128i) -> __m128i {
static_assert_imm4!(IMM4);
let a = a.as_i32x4();
let b = b.as_i32x4();
macro_rules! blend2 {
($a:expr, $b:expr, $c:expr, $d:expr) => {
simd_shuffle4(a, b, [$a, $b, $c, $d])
};
}
macro_rules! blend1 {
($a:expr, $b:expr) => {
match (imm8 >> 2) & 0b11 {
0b00 => blend2!($a, $b, 2, 3),
0b01 => blend2!($a, $b, 6, 3),
0b10 => blend2!($a, $b, 2, 7),
_ => blend2!($a, $b, 6, 7),
}
};
}
let r: i32x4 = match imm8 & 0b11 {
0b00 => blend1!(0, 1),
0b01 => blend1!(4, 1),
0b10 => blend1!(0, 5),
_ => blend1!(4, 5),
};
let r: i32x4 = simd_shuffle4(
a,
b,
[
[0, 4, 0, 4][IMM4 as usize & 0b11],
[1, 1, 5, 5][IMM4 as usize & 0b11],
[2, 6, 2, 6][(IMM4 as usize >> 2) & 0b11],
[3, 3, 7, 7][(IMM4 as usize >> 2) & 0b11],
],
);
transmute(r)
}
@ -4065,10 +4054,10 @@ mod tests {
unsafe fn test_mm_blend_epi32() {
let (a, b) = (_mm_set1_epi32(3), _mm_set1_epi32(9));
let e = _mm_setr_epi32(9, 3, 3, 3);
let r = _mm_blend_epi32(a, b, 0x01 as i32);
let r = _mm_blend_epi32::<0x01>(a, b);
assert_eq_m128i(r, e);
let r = _mm_blend_epi32(b, a, 0x0E as i32);
let r = _mm_blend_epi32::<0x0E>(b, a);
assert_eq_m128i(r, e);
}

15
library/stdarch/crates/core_arch/src/x86/avx512gfni.rs
View file

@ -1057,7 +1057,8 @@ mod tests {
const MASK_WORDS: i32 = 0b01_10;
let expected_result = _mm_gf2p8mul_epi8(left, right);
let result_masked = _mm_maskz_gf2p8mul_epi8(mask_bytes, left, right);
let expected_masked = _mm_blend_epi32(_mm_setzero_si128(), expected_result, MASK_WORDS);
let expected_masked =
_mm_blend_epi32::<MASK_WORDS>(_mm_setzero_si128(), expected_result);
assert_eq_m128i(result_masked, expected_masked);
}
}
@ -1075,7 +1076,7 @@ mod tests {
const MASK_WORDS: i32 = 0b01_10;
let expected_result = _mm_gf2p8mul_epi8(left, right);
let result_masked = _mm_mask_gf2p8mul_epi8(left, mask_bytes, left, right);
let expected_masked = _mm_blend_epi32(left, expected_result, MASK_WORDS);
let expected_masked = _mm_blend_epi32::<MASK_WORDS>(left, expected_result);
assert_eq_m128i(result_masked, expected_masked);
}
}
@ -1281,7 +1282,8 @@ mod tests {
let expected_result = _mm_gf2p8affine_epi64_epi8(vector, matrix, CONSTANT_BYTE);
let result_masked =
_mm_maskz_gf2p8affine_epi64_epi8(mask_bytes, vector, matrix, CONSTANT_BYTE);
let expected_masked = _mm_blend_epi32(_mm_setzero_si128(), expected_result, MASK_WORDS);
let expected_masked =
_mm_blend_epi32::<MASK_WORDS>(_mm_setzero_si128(), expected_result);
assert_eq_m128i(result_masked, expected_masked);
}
}
@ -1301,7 +1303,7 @@ mod tests {
let expected_result = _mm_gf2p8affine_epi64_epi8(left, right, CONSTANT_BYTE);
let result_masked =
_mm_mask_gf2p8affine_epi64_epi8(left, mask_bytes, left, right, CONSTANT_BYTE);
let expected_masked = _mm_blend_epi32(left, expected_result, MASK_WORDS);
let expected_masked = _mm_blend_epi32::<MASK_WORDS>(left, expected_result);
assert_eq_m128i(result_masked, expected_masked);
}
}
@ -1539,7 +1541,8 @@ mod tests {
let expected_result = _mm_gf2p8affineinv_epi64_epi8(vector, matrix, CONSTANT_BYTE);
let result_masked =
_mm_maskz_gf2p8affineinv_epi64_epi8(mask_bytes, vector, matrix, CONSTANT_BYTE);
let expected_masked = _mm_blend_epi32(_mm_setzero_si128(), expected_result, MASK_WORDS);
let expected_masked =
_mm_blend_epi32::<MASK_WORDS>(_mm_setzero_si128(), expected_result);
assert_eq_m128i(result_masked, expected_masked);
}
}
@ -1560,7 +1563,7 @@ mod tests {
let expected_result = _mm_gf2p8affineinv_epi64_epi8(left, right, CONSTANT_BYTE);
let result_masked =
_mm_mask_gf2p8affineinv_epi64_epi8(left, mask_bytes, left, right, CONSTANT_BYTE);
let expected_masked = _mm_blend_epi32(left, expected_result, MASK_WORDS);
let expected_masked = _mm_blend_epi32::<MASK_WORDS>(left, expected_result);
assert_eq_m128i(result_masked, expected_masked);
}
}