| 1 | // Copyright 2020 Yevhenii Reizner |
| 2 | // |
| 3 | // Use of this source code is governed by a BSD-style license that can be |
| 4 | // found in the LICENSE file. |
| 5 | |
| 6 | // Based on https://github.com/Lokathor/wide (Zlib) |
| 7 | |
| 8 | use super::{f32x8, i32x8}; |
| 9 | |
| 10 | cfg_if::cfg_if! { |
| 11 | if #[cfg(all(feature = "simd" , target_feature = "avx2" ))] { |
| 12 | #[cfg(target_arch = "x86" )] |
| 13 | use core::arch::x86::*; |
| 14 | #[cfg(target_arch = "x86_64" )] |
| 15 | use core::arch::x86_64::*; |
| 16 | |
| 17 | use bytemuck::cast; |
| 18 | |
| 19 | #[derive(Clone, Copy, Debug)] |
| 20 | #[repr(C, align(32))] |
| 21 | pub struct u32x8(__m256i); |
| 22 | } else { |
| 23 | use super::u32x4; |
| 24 | |
| 25 | #[derive (Clone, Copy, Debug)] |
| 26 | #[repr (C, align(32))] |
| 27 | pub struct u32x8(u32x4, u32x4); |
| 28 | } |
| 29 | } |
| 30 | |
| 31 | unsafe impl bytemuck::Zeroable for u32x8 {} |
| 32 | unsafe impl bytemuck::Pod for u32x8 {} |
| 33 | |
| 34 | impl Default for u32x8 { |
| 35 | fn default() -> Self { |
| 36 | Self::splat(0) |
| 37 | } |
| 38 | } |
| 39 | |
| 40 | impl u32x8 { |
| 41 | pub fn splat(n: u32) -> Self { |
| 42 | bytemuck::cast([n, n, n, n, n, n, n, n]) |
| 43 | } |
| 44 | |
| 45 | pub fn to_i32x8_bitcast(self) -> i32x8 { |
| 46 | bytemuck::cast(self) |
| 47 | } |
| 48 | |
| 49 | pub fn to_f32x8_bitcast(self) -> f32x8 { |
| 50 | bytemuck::cast(self) |
| 51 | } |
| 52 | |
| 53 | pub fn cmp_eq(self, rhs: Self) -> Self { |
| 54 | cfg_if::cfg_if! { |
| 55 | if #[cfg(all(feature = "simd" , target_feature = "avx2" ))] { |
| 56 | Self(unsafe { _mm256_cmpeq_epi32(self.0, rhs.0) }) |
| 57 | } else { |
| 58 | Self(self.0.cmp_eq(rhs.0), self.1.cmp_eq(rhs.1)) |
| 59 | } |
| 60 | } |
| 61 | } |
| 62 | |
| 63 | pub fn shl<const RHS: i32>(self) -> Self { |
| 64 | cfg_if::cfg_if! { |
| 65 | if #[cfg(all(feature = "simd" , target_feature = "avx2" ))] { |
| 66 | let shift: __m128i = cast([RHS as u64, 0]); |
| 67 | Self(unsafe { _mm256_sll_epi32(self.0, shift) }) |
| 68 | } else { |
| 69 | Self(self.0.shl::<RHS>(), self.1.shl::<RHS>()) |
| 70 | } |
| 71 | } |
| 72 | } |
| 73 | |
| 74 | pub fn shr<const RHS: i32>(self) -> Self { |
| 75 | cfg_if::cfg_if! { |
| 76 | if #[cfg(all(feature = "simd" , target_feature = "avx2" ))] { |
| 77 | let shift: __m128i = cast([RHS as u64, 0]); |
| 78 | Self(unsafe { _mm256_srl_epi32(self.0, shift) }) |
| 79 | } else { |
| 80 | Self(self.0.shr::<RHS>(), self.1.shr::<RHS>()) |
| 81 | } |
| 82 | } |
| 83 | } |
| 84 | } |
| 85 | |
| 86 | impl core::ops::Not for u32x8 { |
| 87 | type Output = Self; |
| 88 | |
| 89 | fn not(self) -> Self { |
| 90 | cfg_if::cfg_if! { |
| 91 | if #[cfg(all(feature = "simd" , target_feature = "avx2" ))] { |
| 92 | let all_bits = unsafe { _mm256_set1_epi16(-1) }; |
| 93 | Self(unsafe { _mm256_xor_si256(self.0, all_bits) }) |
| 94 | } else { |
| 95 | Self(!self.0, !self.1) |
| 96 | } |
| 97 | } |
| 98 | } |
| 99 | } |
| 100 | |
| 101 | impl core::ops::Add for u32x8 { |
| 102 | type Output = Self; |
| 103 | |
| 104 | fn add(self, rhs: Self) -> Self::Output { |
| 105 | cfg_if::cfg_if! { |
| 106 | if #[cfg(all(feature = "simd" , target_feature = "avx2" ))] { |
| 107 | Self(unsafe { _mm256_add_epi32(self.0, rhs.0) }) |
| 108 | } else { |
| 109 | Self(self.0 + rhs.0, self.1 + rhs.1) |
| 110 | } |
| 111 | } |
| 112 | } |
| 113 | } |
| 114 | |
| 115 | impl core::ops::BitAnd for u32x8 { |
| 116 | type Output = Self; |
| 117 | |
| 118 | fn bitand(self, rhs: Self) -> Self::Output { |
| 119 | cfg_if::cfg_if! { |
| 120 | if #[cfg(all(feature = "simd" , target_feature = "avx2" ))] { |
| 121 | Self(unsafe { _mm256_and_si256(self.0, rhs.0) }) |
| 122 | } else { |
| 123 | Self(self.0 & rhs.0, self.1 & rhs.1) |
| 124 | } |
| 125 | } |
| 126 | } |
| 127 | } |
| 128 | |