| 1 | // -------------------------------------------------------------------------- |
| 2 | // |
| 3 | // Copyright |
| 4 | // Markus Wittmann, 2016-2018 |
| 5 | // RRZE, University of Erlangen-Nuremberg, Germany |
| 6 | // markus.wittmann -at- fau.de or hpc -at- rrze.fau.de |
| 7 | // |
| 8 | // Viktor Haag, 2016 |
| 9 | // LSS, University of Erlangen-Nuremberg, Germany |
| 10 | // |
| 11 | // Michael Hussnaetter, 2017-2018 |
| 12 | // University of Erlangen-Nuremberg, Germany |
| 13 | // michael.hussnaetter -at- fau.de |
| 14 | // |
| 15 | // This file is part of the Lattice Boltzmann Benchmark Kernels (LbmBenchKernels). |
| 16 | // |
| 17 | // LbmBenchKernels is free software: you can redistribute it and/or modify |
| 18 | // it under the terms of the GNU General Public License as published by |
| 19 | // the Free Software Foundation, either version 3 of the License, or |
| 20 | // (at your option) any later version. |
| 21 | // |
| 22 | // LbmBenchKernels is distributed in the hope that it will be useful, |
| 23 | // but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 24 | // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 25 | // GNU General Public License for more details. |
| 26 | // |
| 27 | // You should have received a copy of the GNU General Public License |
| 28 | // along with LbmBenchKernels. If not, see <http://www.gnu.org/licenses/>. |
| 29 | // |
| 30 | // -------------------------------------------------------------------------- |
| 31 | #ifndef __VECTOR_H__ |
| 32 | #define __VECTOR_H__ |
| 33 | |
| 34 | #if !defined(VECTOR_AVX512) && !defined(VECTOR_AVX) && !defined(VECTOR_SSE) |
| 35 | #warning Defining VECTOR_AVX as no ISA extension was selected. |
| 36 | #define VECTOR_AVX |
| 37 | #endif |
| 38 | |
| 39 | #if (defined(VECTOR_AVX512) && defined(VECTOR_AVX)) || \ |
| 40 | (defined(VECTOR_AVX512) && defined(VECTOR_SEE)) || \ |
| 41 | (defined(VECTOR_AVX) && defined(VECTOR_SSE)) |
| 42 | #error Only VECTOR_AVX512 or VECTOR_AVX or VECTOR_SSE can be defined at the same time. |
| 43 | #endif |
| 44 | |
| 45 | #if !defined(PRECISION_DP) && !defined(PRECISION_SP) |
| 46 | #error PRECISION_DP or PRECISION_SP must be defined. |
| 47 | #endif |
| 48 | |
| 49 | #if defined(PRECISION_DP) && defined(PRECISION_SP) |
| 50 | #error Only PRECISION_DP or PRECISION_SP can be defined at the same time. |
| 51 | #endif |
| 52 | |
| 53 | #ifdef PRECISION_DP |
| 54 | |
| 55 | #ifdef VECTOR_AVX512 |
| 56 | |
| 57 | #include <immintrin.h> |
| 58 | // Vector size in double-precision floatin-point numbers. |
| 59 | #define VSIZE 8 |
| 60 | |
| 61 | #define VPDFT __m512d |
| 62 | |
| 63 | #define VSET(scalar) _mm512_set1_pd(scalar) |
| 64 | #define VSETI32(scalar) _mm256_set1_epi32(scalar) |
| 65 | |
| 66 | #define VLD(expr) _mm512_load_pd(expr) |
| 67 | #define VLDU(expr) _mm512_loadu_pd(expr) |
| 68 | #define VLIU(expr) _mm256_loadu_si256((__m256i const *)expr) |
| 69 | #define VLI64(expr) _mm512_load_epi64(expr) |
| 70 | |
| 71 | #define VST(dst, src) _mm512_store_pd(dst, src) |
| 72 | #define VSTU(dst, src) _mm512_storeu_pd(dst, src) |
| 73 | #define VSTNT(dst, src) _mm512_stream_pd(dst, src) |
| 74 | |
| 75 | #define VG32(offsets, base, scale) _mm512_i32gather_pd(offsets, base, scale) |
| 76 | #define VG64(offsets, base, scale) _mm512_i64gather_pd(offsets, base, scale) |
| 77 | |
| 78 | #define VPG32(offsets, base, scale, hint) _mm512_prefetch_i32gather_pd(offsets, base, scale, hint) |
| 79 | |
| 80 | #define VS32(dst_base, dst_offsets, src, scale) _mm512_i32scatter_pd(dst_base, dst_offsets, src, scale) |
| 81 | #define VS64(dst_base, dst_offsets, src, scale) _mm512_i64scatter_pd(dst_base, dst_offsets, src, scale) |
| 82 | |
| 83 | #define VPS32(dst_base, dst_offsets, scale, hint) _mm512_prefetch_i32scatter_pd(dst_base, dst_offsets, scale, hint) |
| 84 | |
| 85 | #define VMUL(a, b) _mm512_mul_pd(a, b) |
| 86 | #define VADD(a, b) _mm512_add_pd(a, b) |
| 87 | #define VADDI32(a,b) _mm256_add_epi32(a,b) |
| 88 | #define VMULI32(a,b) _mm256_mul_epi32(a,b) |
| 89 | #define VSUB(a, b) _mm512_sub_pd(a, b) |
| 90 | #endif |
| 91 | |
| 92 | #ifdef VECTOR_AVX |
| 93 | |
| 94 | #include <immintrin.h> |
| 95 | // Vector size in double-precision floating-point numbers. |
| 96 | #define VSIZE 4 |
| 97 | |
| 98 | #define VPDFT __m256d |
| 99 | |
| 100 | #define VSET(scalar) _mm256_set1_pd(scalar) |
| 101 | |
| 102 | #define VLD(expr) _mm256_load_pd(expr) |
| 103 | #define VLDU(expr) _mm256_loadu_pd(expr) |
| 104 | |
| 105 | #define VST(dst, src) _mm256_store_pd(dst, src) |
| 106 | #define VSTU(dst, src) _mm256_storeu_pd(dst, src) |
| 107 | #define VSTNT(dst, src) _mm256_stream_pd(dst, src) |
| 108 | |
| 109 | #define VMUL(a, b) _mm256_mul_pd(a, b) |
| 110 | #define VADD(a, b) _mm256_add_pd(a, b) |
| 111 | #define VSUB(a, b) _mm256_sub_pd(a, b) |
| 112 | #endif |
| 113 | |
| 114 | #ifdef VECTOR_SSE |
| 115 | #include <emmintrin.h> |
| 116 | // Vector size in double-precision floating-point numbers. |
| 117 | #define VSIZE 2 |
| 118 | |
| 119 | #define VPDFT __m128d |
| 120 | |
| 121 | #define VSET(scalar) _mm_set1_pd(scalar) |
| 122 | |
| 123 | #define VLD(expr) _mm_load_pd(expr) |
| 124 | #define VLDU(expr) _mm_loadu_pd(expr) |
| 125 | |
| 126 | #define VST(dst, src) _mm_store_pd(dst, src) |
| 127 | #define VSTU(dst, src) _mm_storeu_pd(dst, src) |
| 128 | #define VSTNT(dst, src) _mm_stream_pd(dst, src) |
| 129 | |
| 130 | #define VMUL(a, b) _mm_mul_pd(a, b) |
| 131 | #define VADD(a, b) _mm_add_pd(a, b) |
| 132 | #define VSUB(a, b) _mm_sub_pd(a, b) |
| 133 | #endif |
| 134 | |
| 135 | #elif defined(PRECISION_SP) |
| 136 | |
| 137 | #ifdef VECTOR_AVX512 |
| 138 | #error Single precision intrinsic kernels for AVX512 are currently not implemented. |
| 139 | #endif |
| 140 | |
| 141 | #ifdef VECTOR_AVX |
| 142 | |
| 143 | #include <immintrin.h> |
| 144 | // Vector size in double-precision floating-point numbers. |
| 145 | #define VSIZE 8 |
| 146 | |
| 147 | #define VPDFT __m256 |
| 148 | |
| 149 | #define VSET(scalar) _mm256_set1_ps(scalar) |
| 150 | |
| 151 | #define VLD(expr) _mm256_load_ps(expr) |
| 152 | #define VLDU(expr) _mm256_loadu_ps(expr) |
| 153 | |
| 154 | #define VST(dst, src) _mm256_store_ps(dst, src) |
| 155 | #define VSTU(dst, src) _mm256_storeu_ps(dst, src) |
| 156 | #define VSTNT(dst, src) _mm256_stream_ps(dst, src) |
| 157 | |
| 158 | #define VMUL(a, b) _mm256_mul_ps(a, b) |
| 159 | #define VADD(a, b) _mm256_add_ps(a, b) |
| 160 | #define VSUB(a, b) _mm256_sub_ps(a, b) |
| 161 | #endif |
| 162 | |
| 163 | #ifdef VECTOR_SSE |
| 164 | #include <emmintrin.h> |
| 165 | // Vector size in double-precision floating-point numbers. |
| 166 | #define VSIZE 4 |
| 167 | |
| 168 | #define VPDFT __m128 |
| 169 | |
| 170 | #define VSET(scalar) _mm_set1_ps(scalar) |
| 171 | |
| 172 | #define VLD(expr) _mm_load_ps(expr) |
| 173 | #define VLDU(expr) _mm_loadu_ps(expr) |
| 174 | |
| 175 | #define VST(dst, src) _mm_store_ps(dst, src) |
| 176 | #define VSTU(dst, src) _mm_storeu_ps(dst, src) |
| 177 | #define VSTNT(dst, src) _mm_stream_ps(dst, src) |
| 178 | |
| 179 | #define VMUL(a, b) _mm_mul_ps(a, b) |
| 180 | #define VADD(a, b) _mm_add_ps(a, b) |
| 181 | #define VSUB(a, b) _mm_sub_ps(a, b) |
| 182 | #endif |
| 183 | |
| 184 | #endif // PRECISION |
| 185 | |
| 186 | #endif // __VECTOR_H__ |