Pirama Arumuga Nainar | bb4374f | 2016-10-20 16:43:03 -0700 | [diff] [blame] | 1 | /*===------------- avx512vbmivlintrin.h - VBMI intrinsics ------------------=== |
| 2 | * |
| 3 | * |
| 4 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 5 | * of this software and associated documentation files (the "Software"), to deal |
| 6 | * in the Software without restriction, including without limitation the rights |
| 7 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| 8 | * copies of the Software, and to permit persons to whom the Software is |
| 9 | * furnished to do so, subject to the following conditions: |
| 10 | * |
| 11 | * The above copyright notice and this permission notice shall be included in |
| 12 | * all copies or substantial portions of the Software. |
| 13 | * |
| 14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 17 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 18 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 19 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
| 20 | * THE SOFTWARE. |
| 21 | * |
| 22 | *===-----------------------------------------------------------------------=== |
| 23 | */ |
| 24 | #ifndef __IMMINTRIN_H |
| 25 | #error "Never use <avx512vbmivlintrin.h> directly; include <immintrin.h> instead." |
| 26 | #endif |
| 27 | |
| 28 | #ifndef __VBMIVLINTRIN_H |
| 29 | #define __VBMIVLINTRIN_H |
| 30 | |
| 31 | /* Define the default attributes for the functions in this file. */ |
| 32 | #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi,avx512vl"))) |
| 33 | |
| 34 | |
| 35 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 36 | _mm_mask2_permutex2var_epi8 (__m128i __A, __m128i __I, __mmask16 __U, |
| 37 | __m128i __B) |
| 38 | { |
| 39 | return (__m128i) __builtin_ia32_vpermi2varqi128_mask ((__v16qi) __A, |
| 40 | (__v16qi) __I |
| 41 | /* idx */ , |
| 42 | (__v16qi) __B, |
| 43 | (__mmask16) |
| 44 | __U); |
| 45 | } |
| 46 | |
| 47 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
| 48 | _mm256_mask2_permutex2var_epi8 (__m256i __A, __m256i __I, |
| 49 | __mmask32 __U, __m256i __B) |
| 50 | { |
| 51 | return (__m256i) __builtin_ia32_vpermi2varqi256_mask ((__v32qi) __A, |
| 52 | (__v32qi) __I |
| 53 | /* idx */ , |
| 54 | (__v32qi) __B, |
| 55 | (__mmask32) |
| 56 | __U); |
| 57 | } |
| 58 | |
| 59 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 60 | _mm_permutex2var_epi8 (__m128i __A, __m128i __I, __m128i __B) |
| 61 | { |
| 62 | return (__m128i) __builtin_ia32_vpermt2varqi128_mask ((__v16qi) __I |
| 63 | /* idx */ , |
| 64 | (__v16qi) __A, |
| 65 | (__v16qi) __B, |
| 66 | (__mmask16) - |
| 67 | 1); |
| 68 | } |
| 69 | |
| 70 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 71 | _mm_mask_permutex2var_epi8 (__m128i __A, __mmask16 __U, __m128i __I, |
| 72 | __m128i __B) |
| 73 | { |
| 74 | return (__m128i) __builtin_ia32_vpermt2varqi128_mask ((__v16qi) __I |
| 75 | /* idx */ , |
| 76 | (__v16qi) __A, |
| 77 | (__v16qi) __B, |
| 78 | (__mmask16) |
| 79 | __U); |
| 80 | } |
| 81 | |
| 82 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 83 | _mm_maskz_permutex2var_epi8 (__mmask16 __U, __m128i __A, __m128i __I, |
| 84 | __m128i __B) |
| 85 | { |
| 86 | return (__m128i) __builtin_ia32_vpermt2varqi128_maskz ((__v16qi) __I |
| 87 | /* idx */ , |
| 88 | (__v16qi) __A, |
| 89 | (__v16qi) __B, |
| 90 | (__mmask16) |
| 91 | __U); |
| 92 | } |
| 93 | |
| 94 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
| 95 | _mm256_permutex2var_epi8 (__m256i __A, __m256i __I, __m256i __B) |
| 96 | { |
| 97 | return (__m256i) __builtin_ia32_vpermt2varqi256_mask ((__v32qi) __I |
| 98 | /* idx */ , |
| 99 | (__v32qi) __A, |
| 100 | (__v32qi) __B, |
| 101 | (__mmask32) - |
| 102 | 1); |
| 103 | } |
| 104 | |
| 105 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
| 106 | _mm256_mask_permutex2var_epi8 (__m256i __A, __mmask32 __U, |
| 107 | __m256i __I, __m256i __B) |
| 108 | { |
| 109 | return (__m256i) __builtin_ia32_vpermt2varqi256_mask ((__v32qi) __I |
| 110 | /* idx */ , |
| 111 | (__v32qi) __A, |
| 112 | (__v32qi) __B, |
| 113 | (__mmask32) |
| 114 | __U); |
| 115 | } |
| 116 | |
| 117 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
| 118 | _mm256_maskz_permutex2var_epi8 (__mmask32 __U, __m256i __A, |
| 119 | __m256i __I, __m256i __B) |
| 120 | { |
| 121 | return (__m256i) __builtin_ia32_vpermt2varqi256_maskz ((__v32qi) __I |
| 122 | /* idx */ , |
| 123 | (__v32qi) __A, |
| 124 | (__v32qi) __B, |
| 125 | (__mmask32) |
| 126 | __U); |
| 127 | } |
| 128 | |
| 129 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 130 | _mm_permutexvar_epi8 (__m128i __A, __m128i __B) |
| 131 | { |
| 132 | return (__m128i) __builtin_ia32_permvarqi128_mask ((__v16qi) __B, |
| 133 | (__v16qi) __A, |
| 134 | (__v16qi) _mm_undefined_si128 (), |
| 135 | (__mmask16) -1); |
| 136 | } |
| 137 | |
| 138 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 139 | _mm_maskz_permutexvar_epi8 (__mmask16 __M, __m128i __A, __m128i __B) |
| 140 | { |
| 141 | return (__m128i) __builtin_ia32_permvarqi128_mask ((__v16qi) __B, |
| 142 | (__v16qi) __A, |
| 143 | (__v16qi) _mm_setzero_si128 (), |
| 144 | (__mmask16) __M); |
| 145 | } |
| 146 | |
| 147 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 148 | _mm_mask_permutexvar_epi8 (__m128i __W, __mmask16 __M, __m128i __A, |
| 149 | __m128i __B) |
| 150 | { |
| 151 | return (__m128i) __builtin_ia32_permvarqi128_mask ((__v16qi) __B, |
| 152 | (__v16qi) __A, |
| 153 | (__v16qi) __W, |
| 154 | (__mmask16) __M); |
| 155 | } |
| 156 | |
| 157 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
| 158 | _mm256_permutexvar_epi8 (__m256i __A, __m256i __B) |
| 159 | { |
| 160 | return (__m256i) __builtin_ia32_permvarqi256_mask ((__v32qi) __B, |
| 161 | (__v32qi) __A, |
| 162 | (__v32qi) _mm256_undefined_si256 (), |
| 163 | (__mmask32) -1); |
| 164 | } |
| 165 | |
| 166 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
| 167 | _mm256_maskz_permutexvar_epi8 (__mmask32 __M, __m256i __A, |
| 168 | __m256i __B) |
| 169 | { |
| 170 | return (__m256i) __builtin_ia32_permvarqi256_mask ((__v32qi) __B, |
| 171 | (__v32qi) __A, |
| 172 | (__v32qi) _mm256_setzero_si256 (), |
| 173 | (__mmask32) __M); |
| 174 | } |
| 175 | |
| 176 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
| 177 | _mm256_mask_permutexvar_epi8 (__m256i __W, __mmask32 __M, __m256i __A, |
| 178 | __m256i __B) |
| 179 | { |
| 180 | return (__m256i) __builtin_ia32_permvarqi256_mask ((__v32qi) __B, |
| 181 | (__v32qi) __A, |
| 182 | (__v32qi) __W, |
| 183 | (__mmask32) __M); |
| 184 | } |
| 185 | |
| 186 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 187 | _mm_mask_multishift_epi64_epi8 (__m128i __W, __mmask16 __M, __m128i __X, __m128i __Y) |
| 188 | { |
| 189 | return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v16qi) __X, |
| 190 | (__v16qi) __Y, |
| 191 | (__v16qi) __W, |
| 192 | (__mmask16) __M); |
| 193 | } |
| 194 | |
| 195 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 196 | _mm_maskz_multishift_epi64_epi8 (__mmask16 __M, __m128i __X, __m128i __Y) |
| 197 | { |
| 198 | return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v16qi) __X, |
| 199 | (__v16qi) __Y, |
| 200 | (__v16qi) |
| 201 | _mm_setzero_si128 (), |
| 202 | (__mmask16) __M); |
| 203 | } |
| 204 | |
| 205 | static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 206 | _mm_multishift_epi64_epi8 (__m128i __X, __m128i __Y) |
| 207 | { |
| 208 | return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v16qi) __X, |
| 209 | (__v16qi) __Y, |
| 210 | (__v16qi) |
| 211 | _mm_undefined_si128 (), |
| 212 | (__mmask16) -1); |
| 213 | } |
| 214 | |
| 215 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
| 216 | _mm256_mask_multishift_epi64_epi8 (__m256i __W, __mmask32 __M, __m256i __X, __m256i __Y) |
| 217 | { |
| 218 | return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v32qi) __X, |
| 219 | (__v32qi) __Y, |
| 220 | (__v32qi) __W, |
| 221 | (__mmask32) __M); |
| 222 | } |
| 223 | |
| 224 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
| 225 | _mm256_maskz_multishift_epi64_epi8 (__mmask32 __M, __m256i __X, __m256i __Y) |
| 226 | { |
| 227 | return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v32qi) __X, |
| 228 | (__v32qi) __Y, |
| 229 | (__v32qi) |
| 230 | _mm256_setzero_si256 (), |
| 231 | (__mmask32) __M); |
| 232 | } |
| 233 | |
| 234 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
| 235 | _mm256_multishift_epi64_epi8 (__m256i __X, __m256i __Y) |
| 236 | { |
| 237 | return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v32qi) __X, |
| 238 | (__v32qi) __Y, |
| 239 | (__v32qi) |
| 240 | _mm256_undefined_si256 (), |
| 241 | (__mmask32) -1); |
| 242 | } |
| 243 | |
| 244 | |
| 245 | #undef __DEFAULT_FN_ATTRS |
| 246 | |
| 247 | #endif |