rust-lang/rust
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Convert _mm256_round_ps to const generics

Browse Source
This commit is contained in:
Rémy Rakic
2021-03-01 19:14:49 +01:00
committed by Amanieu d'Antras
parent aacd25dd78
commit 1f4cd33f09
1 changed files with 9 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
library/stdarch/crates/core_arch/src/x86/avx.rs
View File

@@ -379,7 +379,7 @@ pub unsafe fn _mm256_floor_pd(a: __m256d) -> __m256d {
} }
/// Rounds packed single-precision (32-bit) floating point elements in `a` /// Rounds packed single-precision (32-bit) floating point elements in `a`
/// according to the flag `b`. The value of `b` may be as follows: /// according to the flag `ROUNDING`. The value of `ROUNDING` may be as follows:
/// ///
/// - `0x00`: Round to the nearest whole number. /// - `0x00`: Round to the nearest whole number.
/// - `0x01`: Round down, toward negative infinity. /// - `0x01`: Round down, toward negative infinity.
@@ -393,16 +393,12 @@ pub unsafe fn _mm256_floor_pd(a: __m256d) -> __m256d {
/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_round_ps) /// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_round_ps)
#[inline] #[inline]
#[target_feature(enable = "avx")] #[target_feature(enable = "avx")]
#[cfg_attr(test, assert_instr(vroundps, b = 0x00))] #[cfg_attr(test, assert_instr(vroundps, ROUNDING = 0x00))]
#[rustc_args_required_const(1)] #[rustc_legacy_const_generics(1)]
#[stable(feature = "simd_x86", since = "1.27.0")] #[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_round_ps(a: __m256, b: i32) -> __m256 { pub unsafe fn _mm256_round_ps<const ROUNDING: i32>(a: __m256) -> __m256 {
macro_rules! call { static_assert_imm4!(ROUNDING);
($imm8:expr) => { roundps256(a, ROUNDING)
roundps256(a, $imm8)
};
}
constify_imm8!(b, call)
} }
/// Rounds packed single-precision (32-bit) floating point elements in `a` /// Rounds packed single-precision (32-bit) floating point elements in `a`
@@ -3471,9 +3467,9 @@ mod tests {
#[simd_test(enable = "avx")] #[simd_test(enable = "avx")]
unsafe fn test_mm256_round_ps() { unsafe fn test_mm256_round_ps() {
let a = _mm256_setr_ps(1.55, 2.2, 3.99, -1.2, 1.55, 2.2, 3.99, -1.2); let a = _mm256_setr_ps(1.55, 2.2, 3.99, -1.2, 1.55, 2.2, 3.99, -1.2);
let result_closest = _mm256_round_ps(a, 0b00000000); let result_closest = _mm256_round_ps::<0b0000>(a);
let result_down = _mm256_round_ps(a, 0b00000001); let result_down = _mm256_round_ps::<0b0001>(a);
let result_up = _mm256_round_ps(a, 0b00000010); let result_up = _mm256_round_ps::<0b0010>(a);
let expected_closest = _mm256_setr_ps(2., 2., 4., -1., 2., 2., 4., -1.); let expected_closest = _mm256_setr_ps(2., 2., 4., -1., 2., 2., 4., -1.);
let expected_down = _mm256_setr_ps(1., 2., 3., -2., 1., 2., 3., -2.); let expected_down = _mm256_setr_ps(1., 2., 3., -2., 1., 2., 3., -2.);
let expected_up = _mm256_setr_ps(2., 3., 4., -1., 2., 3., 4., -1.); let expected_up = _mm256_setr_ps(2., 3., 4., -1., 2., 3., 4., -1.);