Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 14 | pmbaty | 1 | /*===---- avx2intrin.h - AVX2 intrinsics -----------------------------------=== |
| 2 | * |
||
| 3 | * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
||
| 4 | * See https://llvm.org/LICENSE.txt for license information. |
||
| 5 | * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
||
| 6 | * |
||
| 7 | *===-----------------------------------------------------------------------=== |
||
| 8 | */ |
||
| 9 | |||
| 10 | #ifndef __IMMINTRIN_H |
||
| 11 | #error "Never use <avx2intrin.h> directly; include <immintrin.h> instead." |
||
| 12 | #endif |
||
| 13 | |||
| 14 | #ifndef __AVX2INTRIN_H |
||
| 15 | #define __AVX2INTRIN_H |
||
| 16 | |||
| 17 | /* Define the default attributes for the functions in this file. */ |
||
| 18 | #define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("avx2"), __min_vector_width__(256))) |
||
| 19 | #define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("avx2"), __min_vector_width__(128))) |
||
| 20 | |||
| 21 | /* SSE4 Multiple Packed Sums of Absolute Difference. */ |
||
| 22 | #define _mm256_mpsadbw_epu8(X, Y, M) \ |
||
| 23 | ((__m256i)__builtin_ia32_mpsadbw256((__v32qi)(__m256i)(X), \ |
||
| 24 | (__v32qi)(__m256i)(Y), (int)(M))) |
||
| 25 | |||
| 26 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 27 | _mm256_abs_epi8(__m256i __a) |
||
| 28 | { |
||
| 29 | return (__m256i)__builtin_elementwise_abs((__v32qs)__a); |
||
| 30 | } |
||
| 31 | |||
| 32 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 33 | _mm256_abs_epi16(__m256i __a) |
||
| 34 | { |
||
| 35 | return (__m256i)__builtin_elementwise_abs((__v16hi)__a); |
||
| 36 | } |
||
| 37 | |||
| 38 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 39 | _mm256_abs_epi32(__m256i __a) |
||
| 40 | { |
||
| 41 | return (__m256i)__builtin_elementwise_abs((__v8si)__a); |
||
| 42 | } |
||
| 43 | |||
| 44 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 45 | _mm256_packs_epi16(__m256i __a, __m256i __b) |
||
| 46 | { |
||
| 47 | return (__m256i)__builtin_ia32_packsswb256((__v16hi)__a, (__v16hi)__b); |
||
| 48 | } |
||
| 49 | |||
| 50 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 51 | _mm256_packs_epi32(__m256i __a, __m256i __b) |
||
| 52 | { |
||
| 53 | return (__m256i)__builtin_ia32_packssdw256((__v8si)__a, (__v8si)__b); |
||
| 54 | } |
||
| 55 | |||
| 56 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 57 | _mm256_packus_epi16(__m256i __a, __m256i __b) |
||
| 58 | { |
||
| 59 | return (__m256i)__builtin_ia32_packuswb256((__v16hi)__a, (__v16hi)__b); |
||
| 60 | } |
||
| 61 | |||
| 62 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 63 | _mm256_packus_epi32(__m256i __V1, __m256i __V2) |
||
| 64 | { |
||
| 65 | return (__m256i) __builtin_ia32_packusdw256((__v8si)__V1, (__v8si)__V2); |
||
| 66 | } |
||
| 67 | |||
| 68 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 69 | _mm256_add_epi8(__m256i __a, __m256i __b) |
||
| 70 | { |
||
| 71 | return (__m256i)((__v32qu)__a + (__v32qu)__b); |
||
| 72 | } |
||
| 73 | |||
| 74 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 75 | _mm256_add_epi16(__m256i __a, __m256i __b) |
||
| 76 | { |
||
| 77 | return (__m256i)((__v16hu)__a + (__v16hu)__b); |
||
| 78 | } |
||
| 79 | |||
| 80 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 81 | _mm256_add_epi32(__m256i __a, __m256i __b) |
||
| 82 | { |
||
| 83 | return (__m256i)((__v8su)__a + (__v8su)__b); |
||
| 84 | } |
||
| 85 | |||
| 86 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 87 | _mm256_add_epi64(__m256i __a, __m256i __b) |
||
| 88 | { |
||
| 89 | return (__m256i)((__v4du)__a + (__v4du)__b); |
||
| 90 | } |
||
| 91 | |||
| 92 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 93 | _mm256_adds_epi8(__m256i __a, __m256i __b) |
||
| 94 | { |
||
| 95 | return (__m256i)__builtin_elementwise_add_sat((__v32qs)__a, (__v32qs)__b); |
||
| 96 | } |
||
| 97 | |||
| 98 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 99 | _mm256_adds_epi16(__m256i __a, __m256i __b) |
||
| 100 | { |
||
| 101 | return (__m256i)__builtin_elementwise_add_sat((__v16hi)__a, (__v16hi)__b); |
||
| 102 | } |
||
| 103 | |||
| 104 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 105 | _mm256_adds_epu8(__m256i __a, __m256i __b) |
||
| 106 | { |
||
| 107 | return (__m256i)__builtin_elementwise_add_sat((__v32qu)__a, (__v32qu)__b); |
||
| 108 | } |
||
| 109 | |||
| 110 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 111 | _mm256_adds_epu16(__m256i __a, __m256i __b) |
||
| 112 | { |
||
| 113 | return (__m256i)__builtin_elementwise_add_sat((__v16hu)__a, (__v16hu)__b); |
||
| 114 | } |
||
| 115 | |||
| 116 | #define _mm256_alignr_epi8(a, b, n) \ |
||
| 117 | ((__m256i)__builtin_ia32_palignr256((__v32qi)(__m256i)(a), \ |
||
| 118 | (__v32qi)(__m256i)(b), (n))) |
||
| 119 | |||
| 120 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 121 | _mm256_and_si256(__m256i __a, __m256i __b) |
||
| 122 | { |
||
| 123 | return (__m256i)((__v4du)__a & (__v4du)__b); |
||
| 124 | } |
||
| 125 | |||
| 126 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 127 | _mm256_andnot_si256(__m256i __a, __m256i __b) |
||
| 128 | { |
||
| 129 | return (__m256i)(~(__v4du)__a & (__v4du)__b); |
||
| 130 | } |
||
| 131 | |||
| 132 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 133 | _mm256_avg_epu8(__m256i __a, __m256i __b) |
||
| 134 | { |
||
| 135 | return (__m256i)__builtin_ia32_pavgb256((__v32qi)__a, (__v32qi)__b); |
||
| 136 | } |
||
| 137 | |||
| 138 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 139 | _mm256_avg_epu16(__m256i __a, __m256i __b) |
||
| 140 | { |
||
| 141 | return (__m256i)__builtin_ia32_pavgw256((__v16hi)__a, (__v16hi)__b); |
||
| 142 | } |
||
| 143 | |||
| 144 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 145 | _mm256_blendv_epi8(__m256i __V1, __m256i __V2, __m256i __M) |
||
| 146 | { |
||
| 147 | return (__m256i)__builtin_ia32_pblendvb256((__v32qi)__V1, (__v32qi)__V2, |
||
| 148 | (__v32qi)__M); |
||
| 149 | } |
||
| 150 | |||
| 151 | #define _mm256_blend_epi16(V1, V2, M) \ |
||
| 152 | ((__m256i)__builtin_ia32_pblendw256((__v16hi)(__m256i)(V1), \ |
||
| 153 | (__v16hi)(__m256i)(V2), (int)(M))) |
||
| 154 | |||
| 155 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 156 | _mm256_cmpeq_epi8(__m256i __a, __m256i __b) |
||
| 157 | { |
||
| 158 | return (__m256i)((__v32qi)__a == (__v32qi)__b); |
||
| 159 | } |
||
| 160 | |||
| 161 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 162 | _mm256_cmpeq_epi16(__m256i __a, __m256i __b) |
||
| 163 | { |
||
| 164 | return (__m256i)((__v16hi)__a == (__v16hi)__b); |
||
| 165 | } |
||
| 166 | |||
| 167 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 168 | _mm256_cmpeq_epi32(__m256i __a, __m256i __b) |
||
| 169 | { |
||
| 170 | return (__m256i)((__v8si)__a == (__v8si)__b); |
||
| 171 | } |
||
| 172 | |||
| 173 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 174 | _mm256_cmpeq_epi64(__m256i __a, __m256i __b) |
||
| 175 | { |
||
| 176 | return (__m256i)((__v4di)__a == (__v4di)__b); |
||
| 177 | } |
||
| 178 | |||
| 179 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 180 | _mm256_cmpgt_epi8(__m256i __a, __m256i __b) |
||
| 181 | { |
||
| 182 | /* This function always performs a signed comparison, but __v32qi is a char |
||
| 183 | which may be signed or unsigned, so use __v32qs. */ |
||
| 184 | return (__m256i)((__v32qs)__a > (__v32qs)__b); |
||
| 185 | } |
||
| 186 | |||
| 187 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 188 | _mm256_cmpgt_epi16(__m256i __a, __m256i __b) |
||
| 189 | { |
||
| 190 | return (__m256i)((__v16hi)__a > (__v16hi)__b); |
||
| 191 | } |
||
| 192 | |||
| 193 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 194 | _mm256_cmpgt_epi32(__m256i __a, __m256i __b) |
||
| 195 | { |
||
| 196 | return (__m256i)((__v8si)__a > (__v8si)__b); |
||
| 197 | } |
||
| 198 | |||
| 199 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 200 | _mm256_cmpgt_epi64(__m256i __a, __m256i __b) |
||
| 201 | { |
||
| 202 | return (__m256i)((__v4di)__a > (__v4di)__b); |
||
| 203 | } |
||
| 204 | |||
| 205 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 206 | _mm256_hadd_epi16(__m256i __a, __m256i __b) |
||
| 207 | { |
||
| 208 | return (__m256i)__builtin_ia32_phaddw256((__v16hi)__a, (__v16hi)__b); |
||
| 209 | } |
||
| 210 | |||
| 211 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 212 | _mm256_hadd_epi32(__m256i __a, __m256i __b) |
||
| 213 | { |
||
| 214 | return (__m256i)__builtin_ia32_phaddd256((__v8si)__a, (__v8si)__b); |
||
| 215 | } |
||
| 216 | |||
| 217 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 218 | _mm256_hadds_epi16(__m256i __a, __m256i __b) |
||
| 219 | { |
||
| 220 | return (__m256i)__builtin_ia32_phaddsw256((__v16hi)__a, (__v16hi)__b); |
||
| 221 | } |
||
| 222 | |||
| 223 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 224 | _mm256_hsub_epi16(__m256i __a, __m256i __b) |
||
| 225 | { |
||
| 226 | return (__m256i)__builtin_ia32_phsubw256((__v16hi)__a, (__v16hi)__b); |
||
| 227 | } |
||
| 228 | |||
| 229 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 230 | _mm256_hsub_epi32(__m256i __a, __m256i __b) |
||
| 231 | { |
||
| 232 | return (__m256i)__builtin_ia32_phsubd256((__v8si)__a, (__v8si)__b); |
||
| 233 | } |
||
| 234 | |||
| 235 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 236 | _mm256_hsubs_epi16(__m256i __a, __m256i __b) |
||
| 237 | { |
||
| 238 | return (__m256i)__builtin_ia32_phsubsw256((__v16hi)__a, (__v16hi)__b); |
||
| 239 | } |
||
| 240 | |||
| 241 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 242 | _mm256_maddubs_epi16(__m256i __a, __m256i __b) |
||
| 243 | { |
||
| 244 | return (__m256i)__builtin_ia32_pmaddubsw256((__v32qi)__a, (__v32qi)__b); |
||
| 245 | } |
||
| 246 | |||
| 247 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 248 | _mm256_madd_epi16(__m256i __a, __m256i __b) |
||
| 249 | { |
||
| 250 | return (__m256i)__builtin_ia32_pmaddwd256((__v16hi)__a, (__v16hi)__b); |
||
| 251 | } |
||
| 252 | |||
| 253 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 254 | _mm256_max_epi8(__m256i __a, __m256i __b) |
||
| 255 | { |
||
| 256 | return (__m256i)__builtin_elementwise_max((__v32qs)__a, (__v32qs)__b); |
||
| 257 | } |
||
| 258 | |||
| 259 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 260 | _mm256_max_epi16(__m256i __a, __m256i __b) |
||
| 261 | { |
||
| 262 | return (__m256i)__builtin_elementwise_max((__v16hi)__a, (__v16hi)__b); |
||
| 263 | } |
||
| 264 | |||
| 265 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 266 | _mm256_max_epi32(__m256i __a, __m256i __b) |
||
| 267 | { |
||
| 268 | return (__m256i)__builtin_elementwise_max((__v8si)__a, (__v8si)__b); |
||
| 269 | } |
||
| 270 | |||
| 271 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 272 | _mm256_max_epu8(__m256i __a, __m256i __b) |
||
| 273 | { |
||
| 274 | return (__m256i)__builtin_elementwise_max((__v32qu)__a, (__v32qu)__b); |
||
| 275 | } |
||
| 276 | |||
| 277 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 278 | _mm256_max_epu16(__m256i __a, __m256i __b) |
||
| 279 | { |
||
| 280 | return (__m256i)__builtin_elementwise_max((__v16hu)__a, (__v16hu)__b); |
||
| 281 | } |
||
| 282 | |||
| 283 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 284 | _mm256_max_epu32(__m256i __a, __m256i __b) |
||
| 285 | { |
||
| 286 | return (__m256i)__builtin_elementwise_max((__v8su)__a, (__v8su)__b); |
||
| 287 | } |
||
| 288 | |||
| 289 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 290 | _mm256_min_epi8(__m256i __a, __m256i __b) |
||
| 291 | { |
||
| 292 | return (__m256i)__builtin_elementwise_min((__v32qs)__a, (__v32qs)__b); |
||
| 293 | } |
||
| 294 | |||
| 295 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 296 | _mm256_min_epi16(__m256i __a, __m256i __b) |
||
| 297 | { |
||
| 298 | return (__m256i)__builtin_elementwise_min((__v16hi)__a, (__v16hi)__b); |
||
| 299 | } |
||
| 300 | |||
| 301 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 302 | _mm256_min_epi32(__m256i __a, __m256i __b) |
||
| 303 | { |
||
| 304 | return (__m256i)__builtin_elementwise_min((__v8si)__a, (__v8si)__b); |
||
| 305 | } |
||
| 306 | |||
| 307 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 308 | _mm256_min_epu8(__m256i __a, __m256i __b) |
||
| 309 | { |
||
| 310 | return (__m256i)__builtin_elementwise_min((__v32qu)__a, (__v32qu)__b); |
||
| 311 | } |
||
| 312 | |||
| 313 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 314 | _mm256_min_epu16(__m256i __a, __m256i __b) |
||
| 315 | { |
||
| 316 | return (__m256i)__builtin_elementwise_min((__v16hu)__a, (__v16hu)__b); |
||
| 317 | } |
||
| 318 | |||
| 319 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 320 | _mm256_min_epu32(__m256i __a, __m256i __b) |
||
| 321 | { |
||
| 322 | return (__m256i)__builtin_elementwise_min((__v8su)__a, (__v8su)__b); |
||
| 323 | } |
||
| 324 | |||
| 325 | static __inline__ int __DEFAULT_FN_ATTRS256 |
||
| 326 | _mm256_movemask_epi8(__m256i __a) |
||
| 327 | { |
||
| 328 | return __builtin_ia32_pmovmskb256((__v32qi)__a); |
||
| 329 | } |
||
| 330 | |||
| 331 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 332 | _mm256_cvtepi8_epi16(__m128i __V) |
||
| 333 | { |
||
| 334 | /* This function always performs a signed extension, but __v16qi is a char |
||
| 335 | which may be signed or unsigned, so use __v16qs. */ |
||
| 336 | return (__m256i)__builtin_convertvector((__v16qs)__V, __v16hi); |
||
| 337 | } |
||
| 338 | |||
| 339 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 340 | _mm256_cvtepi8_epi32(__m128i __V) |
||
| 341 | { |
||
| 342 | /* This function always performs a signed extension, but __v16qi is a char |
||
| 343 | which may be signed or unsigned, so use __v16qs. */ |
||
| 344 | return (__m256i)__builtin_convertvector(__builtin_shufflevector((__v16qs)__V, (__v16qs)__V, 0, 1, 2, 3, 4, 5, 6, 7), __v8si); |
||
| 345 | } |
||
| 346 | |||
| 347 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 348 | _mm256_cvtepi8_epi64(__m128i __V) |
||
| 349 | { |
||
| 350 | /* This function always performs a signed extension, but __v16qi is a char |
||
| 351 | which may be signed or unsigned, so use __v16qs. */ |
||
| 352 | return (__m256i)__builtin_convertvector(__builtin_shufflevector((__v16qs)__V, (__v16qs)__V, 0, 1, 2, 3), __v4di); |
||
| 353 | } |
||
| 354 | |||
| 355 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 356 | _mm256_cvtepi16_epi32(__m128i __V) |
||
| 357 | { |
||
| 358 | return (__m256i)__builtin_convertvector((__v8hi)__V, __v8si); |
||
| 359 | } |
||
| 360 | |||
| 361 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 362 | _mm256_cvtepi16_epi64(__m128i __V) |
||
| 363 | { |
||
| 364 | return (__m256i)__builtin_convertvector(__builtin_shufflevector((__v8hi)__V, (__v8hi)__V, 0, 1, 2, 3), __v4di); |
||
| 365 | } |
||
| 366 | |||
| 367 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 368 | _mm256_cvtepi32_epi64(__m128i __V) |
||
| 369 | { |
||
| 370 | return (__m256i)__builtin_convertvector((__v4si)__V, __v4di); |
||
| 371 | } |
||
| 372 | |||
| 373 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 374 | _mm256_cvtepu8_epi16(__m128i __V) |
||
| 375 | { |
||
| 376 | return (__m256i)__builtin_convertvector((__v16qu)__V, __v16hi); |
||
| 377 | } |
||
| 378 | |||
| 379 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 380 | _mm256_cvtepu8_epi32(__m128i __V) |
||
| 381 | { |
||
| 382 | return (__m256i)__builtin_convertvector(__builtin_shufflevector((__v16qu)__V, (__v16qu)__V, 0, 1, 2, 3, 4, 5, 6, 7), __v8si); |
||
| 383 | } |
||
| 384 | |||
| 385 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 386 | _mm256_cvtepu8_epi64(__m128i __V) |
||
| 387 | { |
||
| 388 | return (__m256i)__builtin_convertvector(__builtin_shufflevector((__v16qu)__V, (__v16qu)__V, 0, 1, 2, 3), __v4di); |
||
| 389 | } |
||
| 390 | |||
| 391 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 392 | _mm256_cvtepu16_epi32(__m128i __V) |
||
| 393 | { |
||
| 394 | return (__m256i)__builtin_convertvector((__v8hu)__V, __v8si); |
||
| 395 | } |
||
| 396 | |||
| 397 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 398 | _mm256_cvtepu16_epi64(__m128i __V) |
||
| 399 | { |
||
| 400 | return (__m256i)__builtin_convertvector(__builtin_shufflevector((__v8hu)__V, (__v8hu)__V, 0, 1, 2, 3), __v4di); |
||
| 401 | } |
||
| 402 | |||
| 403 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 404 | _mm256_cvtepu32_epi64(__m128i __V) |
||
| 405 | { |
||
| 406 | return (__m256i)__builtin_convertvector((__v4su)__V, __v4di); |
||
| 407 | } |
||
| 408 | |||
| 409 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 410 | _mm256_mul_epi32(__m256i __a, __m256i __b) |
||
| 411 | { |
||
| 412 | return (__m256i)__builtin_ia32_pmuldq256((__v8si)__a, (__v8si)__b); |
||
| 413 | } |
||
| 414 | |||
| 415 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 416 | _mm256_mulhrs_epi16(__m256i __a, __m256i __b) |
||
| 417 | { |
||
| 418 | return (__m256i)__builtin_ia32_pmulhrsw256((__v16hi)__a, (__v16hi)__b); |
||
| 419 | } |
||
| 420 | |||
| 421 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 422 | _mm256_mulhi_epu16(__m256i __a, __m256i __b) |
||
| 423 | { |
||
| 424 | return (__m256i)__builtin_ia32_pmulhuw256((__v16hi)__a, (__v16hi)__b); |
||
| 425 | } |
||
| 426 | |||
| 427 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 428 | _mm256_mulhi_epi16(__m256i __a, __m256i __b) |
||
| 429 | { |
||
| 430 | return (__m256i)__builtin_ia32_pmulhw256((__v16hi)__a, (__v16hi)__b); |
||
| 431 | } |
||
| 432 | |||
| 433 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 434 | _mm256_mullo_epi16(__m256i __a, __m256i __b) |
||
| 435 | { |
||
| 436 | return (__m256i)((__v16hu)__a * (__v16hu)__b); |
||
| 437 | } |
||
| 438 | |||
| 439 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 440 | _mm256_mullo_epi32 (__m256i __a, __m256i __b) |
||
| 441 | { |
||
| 442 | return (__m256i)((__v8su)__a * (__v8su)__b); |
||
| 443 | } |
||
| 444 | |||
| 445 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 446 | _mm256_mul_epu32(__m256i __a, __m256i __b) |
||
| 447 | { |
||
| 448 | return __builtin_ia32_pmuludq256((__v8si)__a, (__v8si)__b); |
||
| 449 | } |
||
| 450 | |||
| 451 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 452 | _mm256_or_si256(__m256i __a, __m256i __b) |
||
| 453 | { |
||
| 454 | return (__m256i)((__v4du)__a | (__v4du)__b); |
||
| 455 | } |
||
| 456 | |||
| 457 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 458 | _mm256_sad_epu8(__m256i __a, __m256i __b) |
||
| 459 | { |
||
| 460 | return __builtin_ia32_psadbw256((__v32qi)__a, (__v32qi)__b); |
||
| 461 | } |
||
| 462 | |||
| 463 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 464 | _mm256_shuffle_epi8(__m256i __a, __m256i __b) |
||
| 465 | { |
||
| 466 | return (__m256i)__builtin_ia32_pshufb256((__v32qi)__a, (__v32qi)__b); |
||
| 467 | } |
||
| 468 | |||
| 469 | #define _mm256_shuffle_epi32(a, imm) \ |
||
| 470 | ((__m256i)__builtin_ia32_pshufd256((__v8si)(__m256i)(a), (int)(imm))) |
||
| 471 | |||
| 472 | #define _mm256_shufflehi_epi16(a, imm) \ |
||
| 473 | ((__m256i)__builtin_ia32_pshufhw256((__v16hi)(__m256i)(a), (int)(imm))) |
||
| 474 | |||
| 475 | #define _mm256_shufflelo_epi16(a, imm) \ |
||
| 476 | ((__m256i)__builtin_ia32_pshuflw256((__v16hi)(__m256i)(a), (int)(imm))) |
||
| 477 | |||
| 478 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 479 | _mm256_sign_epi8(__m256i __a, __m256i __b) |
||
| 480 | { |
||
| 481 | return (__m256i)__builtin_ia32_psignb256((__v32qi)__a, (__v32qi)__b); |
||
| 482 | } |
||
| 483 | |||
| 484 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 485 | _mm256_sign_epi16(__m256i __a, __m256i __b) |
||
| 486 | { |
||
| 487 | return (__m256i)__builtin_ia32_psignw256((__v16hi)__a, (__v16hi)__b); |
||
| 488 | } |
||
| 489 | |||
| 490 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 491 | _mm256_sign_epi32(__m256i __a, __m256i __b) |
||
| 492 | { |
||
| 493 | return (__m256i)__builtin_ia32_psignd256((__v8si)__a, (__v8si)__b); |
||
| 494 | } |
||
| 495 | |||
| 496 | #define _mm256_slli_si256(a, imm) \ |
||
| 497 | ((__m256i)__builtin_ia32_pslldqi256_byteshift((__v4di)(__m256i)(a), (int)(imm))) |
||
| 498 | |||
| 499 | #define _mm256_bslli_epi128(a, imm) \ |
||
| 500 | ((__m256i)__builtin_ia32_pslldqi256_byteshift((__v4di)(__m256i)(a), (int)(imm))) |
||
| 501 | |||
| 502 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 503 | _mm256_slli_epi16(__m256i __a, int __count) |
||
| 504 | { |
||
| 505 | return (__m256i)__builtin_ia32_psllwi256((__v16hi)__a, __count); |
||
| 506 | } |
||
| 507 | |||
| 508 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 509 | _mm256_sll_epi16(__m256i __a, __m128i __count) |
||
| 510 | { |
||
| 511 | return (__m256i)__builtin_ia32_psllw256((__v16hi)__a, (__v8hi)__count); |
||
| 512 | } |
||
| 513 | |||
| 514 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 515 | _mm256_slli_epi32(__m256i __a, int __count) |
||
| 516 | { |
||
| 517 | return (__m256i)__builtin_ia32_pslldi256((__v8si)__a, __count); |
||
| 518 | } |
||
| 519 | |||
| 520 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 521 | _mm256_sll_epi32(__m256i __a, __m128i __count) |
||
| 522 | { |
||
| 523 | return (__m256i)__builtin_ia32_pslld256((__v8si)__a, (__v4si)__count); |
||
| 524 | } |
||
| 525 | |||
| 526 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 527 | _mm256_slli_epi64(__m256i __a, int __count) |
||
| 528 | { |
||
| 529 | return __builtin_ia32_psllqi256((__v4di)__a, __count); |
||
| 530 | } |
||
| 531 | |||
| 532 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 533 | _mm256_sll_epi64(__m256i __a, __m128i __count) |
||
| 534 | { |
||
| 535 | return __builtin_ia32_psllq256((__v4di)__a, __count); |
||
| 536 | } |
||
| 537 | |||
| 538 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 539 | _mm256_srai_epi16(__m256i __a, int __count) |
||
| 540 | { |
||
| 541 | return (__m256i)__builtin_ia32_psrawi256((__v16hi)__a, __count); |
||
| 542 | } |
||
| 543 | |||
| 544 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 545 | _mm256_sra_epi16(__m256i __a, __m128i __count) |
||
| 546 | { |
||
| 547 | return (__m256i)__builtin_ia32_psraw256((__v16hi)__a, (__v8hi)__count); |
||
| 548 | } |
||
| 549 | |||
| 550 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 551 | _mm256_srai_epi32(__m256i __a, int __count) |
||
| 552 | { |
||
| 553 | return (__m256i)__builtin_ia32_psradi256((__v8si)__a, __count); |
||
| 554 | } |
||
| 555 | |||
| 556 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 557 | _mm256_sra_epi32(__m256i __a, __m128i __count) |
||
| 558 | { |
||
| 559 | return (__m256i)__builtin_ia32_psrad256((__v8si)__a, (__v4si)__count); |
||
| 560 | } |
||
| 561 | |||
| 562 | #define _mm256_srli_si256(a, imm) \ |
||
| 563 | ((__m256i)__builtin_ia32_psrldqi256_byteshift((__m256i)(a), (int)(imm))) |
||
| 564 | |||
| 565 | #define _mm256_bsrli_epi128(a, imm) \ |
||
| 566 | ((__m256i)__builtin_ia32_psrldqi256_byteshift((__m256i)(a), (int)(imm))) |
||
| 567 | |||
| 568 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 569 | _mm256_srli_epi16(__m256i __a, int __count) |
||
| 570 | { |
||
| 571 | return (__m256i)__builtin_ia32_psrlwi256((__v16hi)__a, __count); |
||
| 572 | } |
||
| 573 | |||
| 574 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 575 | _mm256_srl_epi16(__m256i __a, __m128i __count) |
||
| 576 | { |
||
| 577 | return (__m256i)__builtin_ia32_psrlw256((__v16hi)__a, (__v8hi)__count); |
||
| 578 | } |
||
| 579 | |||
| 580 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 581 | _mm256_srli_epi32(__m256i __a, int __count) |
||
| 582 | { |
||
| 583 | return (__m256i)__builtin_ia32_psrldi256((__v8si)__a, __count); |
||
| 584 | } |
||
| 585 | |||
| 586 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 587 | _mm256_srl_epi32(__m256i __a, __m128i __count) |
||
| 588 | { |
||
| 589 | return (__m256i)__builtin_ia32_psrld256((__v8si)__a, (__v4si)__count); |
||
| 590 | } |
||
| 591 | |||
| 592 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 593 | _mm256_srli_epi64(__m256i __a, int __count) |
||
| 594 | { |
||
| 595 | return __builtin_ia32_psrlqi256((__v4di)__a, __count); |
||
| 596 | } |
||
| 597 | |||
| 598 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 599 | _mm256_srl_epi64(__m256i __a, __m128i __count) |
||
| 600 | { |
||
| 601 | return __builtin_ia32_psrlq256((__v4di)__a, __count); |
||
| 602 | } |
||
| 603 | |||
| 604 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 605 | _mm256_sub_epi8(__m256i __a, __m256i __b) |
||
| 606 | { |
||
| 607 | return (__m256i)((__v32qu)__a - (__v32qu)__b); |
||
| 608 | } |
||
| 609 | |||
| 610 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 611 | _mm256_sub_epi16(__m256i __a, __m256i __b) |
||
| 612 | { |
||
| 613 | return (__m256i)((__v16hu)__a - (__v16hu)__b); |
||
| 614 | } |
||
| 615 | |||
| 616 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 617 | _mm256_sub_epi32(__m256i __a, __m256i __b) |
||
| 618 | { |
||
| 619 | return (__m256i)((__v8su)__a - (__v8su)__b); |
||
| 620 | } |
||
| 621 | |||
| 622 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 623 | _mm256_sub_epi64(__m256i __a, __m256i __b) |
||
| 624 | { |
||
| 625 | return (__m256i)((__v4du)__a - (__v4du)__b); |
||
| 626 | } |
||
| 627 | |||
| 628 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 629 | _mm256_subs_epi8(__m256i __a, __m256i __b) |
||
| 630 | { |
||
| 631 | return (__m256i)__builtin_elementwise_sub_sat((__v32qs)__a, (__v32qs)__b); |
||
| 632 | } |
||
| 633 | |||
| 634 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 635 | _mm256_subs_epi16(__m256i __a, __m256i __b) |
||
| 636 | { |
||
| 637 | return (__m256i)__builtin_elementwise_sub_sat((__v16hi)__a, (__v16hi)__b); |
||
| 638 | } |
||
| 639 | |||
| 640 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 641 | _mm256_subs_epu8(__m256i __a, __m256i __b) |
||
| 642 | { |
||
| 643 | return (__m256i)__builtin_elementwise_sub_sat((__v32qu)__a, (__v32qu)__b); |
||
| 644 | } |
||
| 645 | |||
| 646 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 647 | _mm256_subs_epu16(__m256i __a, __m256i __b) |
||
| 648 | { |
||
| 649 | return (__m256i)__builtin_elementwise_sub_sat((__v16hu)__a, (__v16hu)__b); |
||
| 650 | } |
||
| 651 | |||
| 652 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 653 | _mm256_unpackhi_epi8(__m256i __a, __m256i __b) |
||
| 654 | { |
||
| 655 | return (__m256i)__builtin_shufflevector((__v32qi)__a, (__v32qi)__b, 8, 32+8, 9, 32+9, 10, 32+10, 11, 32+11, 12, 32+12, 13, 32+13, 14, 32+14, 15, 32+15, 24, 32+24, 25, 32+25, 26, 32+26, 27, 32+27, 28, 32+28, 29, 32+29, 30, 32+30, 31, 32+31); |
||
| 656 | } |
||
| 657 | |||
| 658 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 659 | _mm256_unpackhi_epi16(__m256i __a, __m256i __b) |
||
| 660 | { |
||
| 661 | return (__m256i)__builtin_shufflevector((__v16hi)__a, (__v16hi)__b, 4, 16+4, 5, 16+5, 6, 16+6, 7, 16+7, 12, 16+12, 13, 16+13, 14, 16+14, 15, 16+15); |
||
| 662 | } |
||
| 663 | |||
| 664 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 665 | _mm256_unpackhi_epi32(__m256i __a, __m256i __b) |
||
| 666 | { |
||
| 667 | return (__m256i)__builtin_shufflevector((__v8si)__a, (__v8si)__b, 2, 8+2, 3, 8+3, 6, 8+6, 7, 8+7); |
||
| 668 | } |
||
| 669 | |||
| 670 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 671 | _mm256_unpackhi_epi64(__m256i __a, __m256i __b) |
||
| 672 | { |
||
| 673 | return (__m256i)__builtin_shufflevector((__v4di)__a, (__v4di)__b, 1, 4+1, 3, 4+3); |
||
| 674 | } |
||
| 675 | |||
| 676 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 677 | _mm256_unpacklo_epi8(__m256i __a, __m256i __b) |
||
| 678 | { |
||
| 679 | return (__m256i)__builtin_shufflevector((__v32qi)__a, (__v32qi)__b, 0, 32+0, 1, 32+1, 2, 32+2, 3, 32+3, 4, 32+4, 5, 32+5, 6, 32+6, 7, 32+7, 16, 32+16, 17, 32+17, 18, 32+18, 19, 32+19, 20, 32+20, 21, 32+21, 22, 32+22, 23, 32+23); |
||
| 680 | } |
||
| 681 | |||
| 682 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 683 | _mm256_unpacklo_epi16(__m256i __a, __m256i __b) |
||
| 684 | { |
||
| 685 | return (__m256i)__builtin_shufflevector((__v16hi)__a, (__v16hi)__b, 0, 16+0, 1, 16+1, 2, 16+2, 3, 16+3, 8, 16+8, 9, 16+9, 10, 16+10, 11, 16+11); |
||
| 686 | } |
||
| 687 | |||
| 688 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 689 | _mm256_unpacklo_epi32(__m256i __a, __m256i __b) |
||
| 690 | { |
||
| 691 | return (__m256i)__builtin_shufflevector((__v8si)__a, (__v8si)__b, 0, 8+0, 1, 8+1, 4, 8+4, 5, 8+5); |
||
| 692 | } |
||
| 693 | |||
| 694 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 695 | _mm256_unpacklo_epi64(__m256i __a, __m256i __b) |
||
| 696 | { |
||
| 697 | return (__m256i)__builtin_shufflevector((__v4di)__a, (__v4di)__b, 0, 4+0, 2, 4+2); |
||
| 698 | } |
||
| 699 | |||
| 700 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 701 | _mm256_xor_si256(__m256i __a, __m256i __b) |
||
| 702 | { |
||
| 703 | return (__m256i)((__v4du)__a ^ (__v4du)__b); |
||
| 704 | } |
||
| 705 | |||
| 706 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 707 | _mm256_stream_load_si256(__m256i const *__V) |
||
| 708 | { |
||
| 709 | typedef __v4di __v4di_aligned __attribute__((aligned(32))); |
||
| 710 | return (__m256i)__builtin_nontemporal_load((const __v4di_aligned *)__V); |
||
| 711 | } |
||
| 712 | |||
| 713 | static __inline__ __m128 __DEFAULT_FN_ATTRS128 |
||
| 714 | _mm_broadcastss_ps(__m128 __X) |
||
| 715 | { |
||
| 716 | return (__m128)__builtin_shufflevector((__v4sf)__X, (__v4sf)__X, 0, 0, 0, 0); |
||
| 717 | } |
||
| 718 | |||
| 719 | static __inline__ __m128d __DEFAULT_FN_ATTRS128 |
||
| 720 | _mm_broadcastsd_pd(__m128d __a) |
||
| 721 | { |
||
| 722 | return __builtin_shufflevector((__v2df)__a, (__v2df)__a, 0, 0); |
||
| 723 | } |
||
| 724 | |||
| 725 | static __inline__ __m256 __DEFAULT_FN_ATTRS256 |
||
| 726 | _mm256_broadcastss_ps(__m128 __X) |
||
| 727 | { |
||
| 728 | return (__m256)__builtin_shufflevector((__v4sf)__X, (__v4sf)__X, 0, 0, 0, 0, 0, 0, 0, 0); |
||
| 729 | } |
||
| 730 | |||
| 731 | static __inline__ __m256d __DEFAULT_FN_ATTRS256 |
||
| 732 | _mm256_broadcastsd_pd(__m128d __X) |
||
| 733 | { |
||
| 734 | return (__m256d)__builtin_shufflevector((__v2df)__X, (__v2df)__X, 0, 0, 0, 0); |
||
| 735 | } |
||
| 736 | |||
| 737 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 738 | _mm256_broadcastsi128_si256(__m128i __X) |
||
| 739 | { |
||
| 740 | return (__m256i)__builtin_shufflevector((__v2di)__X, (__v2di)__X, 0, 1, 0, 1); |
||
| 741 | } |
||
| 742 | |||
| 743 | #define _mm_broadcastsi128_si256(X) _mm256_broadcastsi128_si256(X) |
||
| 744 | |||
| 745 | #define _mm_blend_epi32(V1, V2, M) \ |
||
| 746 | ((__m128i)__builtin_ia32_pblendd128((__v4si)(__m128i)(V1), \ |
||
| 747 | (__v4si)(__m128i)(V2), (int)(M))) |
||
| 748 | |||
| 749 | #define _mm256_blend_epi32(V1, V2, M) \ |
||
| 750 | ((__m256i)__builtin_ia32_pblendd256((__v8si)(__m256i)(V1), \ |
||
| 751 | (__v8si)(__m256i)(V2), (int)(M))) |
||
| 752 | |||
| 753 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 754 | _mm256_broadcastb_epi8(__m128i __X) |
||
| 755 | { |
||
| 756 | return (__m256i)__builtin_shufflevector((__v16qi)__X, (__v16qi)__X, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0); |
||
| 757 | } |
||
| 758 | |||
| 759 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 760 | _mm256_broadcastw_epi16(__m128i __X) |
||
| 761 | { |
||
| 762 | return (__m256i)__builtin_shufflevector((__v8hi)__X, (__v8hi)__X, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0); |
||
| 763 | } |
||
| 764 | |||
| 765 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 766 | _mm256_broadcastd_epi32(__m128i __X) |
||
| 767 | { |
||
| 768 | return (__m256i)__builtin_shufflevector((__v4si)__X, (__v4si)__X, 0, 0, 0, 0, 0, 0, 0, 0); |
||
| 769 | } |
||
| 770 | |||
| 771 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 772 | _mm256_broadcastq_epi64(__m128i __X) |
||
| 773 | { |
||
| 774 | return (__m256i)__builtin_shufflevector((__v2di)__X, (__v2di)__X, 0, 0, 0, 0); |
||
| 775 | } |
||
| 776 | |||
| 777 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 778 | _mm_broadcastb_epi8(__m128i __X) |
||
| 779 | { |
||
| 780 | return (__m128i)__builtin_shufflevector((__v16qi)__X, (__v16qi)__X, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0); |
||
| 781 | } |
||
| 782 | |||
| 783 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 784 | _mm_broadcastw_epi16(__m128i __X) |
||
| 785 | { |
||
| 786 | return (__m128i)__builtin_shufflevector((__v8hi)__X, (__v8hi)__X, 0, 0, 0, 0, 0, 0, 0, 0); |
||
| 787 | } |
||
| 788 | |||
| 789 | |||
| 790 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 791 | _mm_broadcastd_epi32(__m128i __X) |
||
| 792 | { |
||
| 793 | return (__m128i)__builtin_shufflevector((__v4si)__X, (__v4si)__X, 0, 0, 0, 0); |
||
| 794 | } |
||
| 795 | |||
| 796 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 797 | _mm_broadcastq_epi64(__m128i __X) |
||
| 798 | { |
||
| 799 | return (__m128i)__builtin_shufflevector((__v2di)__X, (__v2di)__X, 0, 0); |
||
| 800 | } |
||
| 801 | |||
| 802 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 803 | _mm256_permutevar8x32_epi32(__m256i __a, __m256i __b) |
||
| 804 | { |
||
| 805 | return (__m256i)__builtin_ia32_permvarsi256((__v8si)__a, (__v8si)__b); |
||
| 806 | } |
||
| 807 | |||
| 808 | #define _mm256_permute4x64_pd(V, M) \ |
||
| 809 | ((__m256d)__builtin_ia32_permdf256((__v4df)(__m256d)(V), (int)(M))) |
||
| 810 | |||
| 811 | static __inline__ __m256 __DEFAULT_FN_ATTRS256 |
||
| 812 | _mm256_permutevar8x32_ps(__m256 __a, __m256i __b) |
||
| 813 | { |
||
| 814 | return (__m256)__builtin_ia32_permvarsf256((__v8sf)__a, (__v8si)__b); |
||
| 815 | } |
||
| 816 | |||
| 817 | #define _mm256_permute4x64_epi64(V, M) \ |
||
| 818 | ((__m256i)__builtin_ia32_permdi256((__v4di)(__m256i)(V), (int)(M))) |
||
| 819 | |||
| 820 | #define _mm256_permute2x128_si256(V1, V2, M) \ |
||
| 821 | ((__m256i)__builtin_ia32_permti256((__m256i)(V1), (__m256i)(V2), (int)(M))) |
||
| 822 | |||
| 823 | #define _mm256_extracti128_si256(V, M) \ |
||
| 824 | ((__m128i)__builtin_ia32_extract128i256((__v4di)(__m256i)(V), (int)(M))) |
||
| 825 | |||
| 826 | #define _mm256_inserti128_si256(V1, V2, M) \ |
||
| 827 | ((__m256i)__builtin_ia32_insert128i256((__v4di)(__m256i)(V1), \ |
||
| 828 | (__v2di)(__m128i)(V2), (int)(M))) |
||
| 829 | |||
| 830 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 831 | _mm256_maskload_epi32(int const *__X, __m256i __M) |
||
| 832 | { |
||
| 833 | return (__m256i)__builtin_ia32_maskloadd256((const __v8si *)__X, (__v8si)__M); |
||
| 834 | } |
||
| 835 | |||
| 836 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 837 | _mm256_maskload_epi64(long long const *__X, __m256i __M) |
||
| 838 | { |
||
| 839 | return (__m256i)__builtin_ia32_maskloadq256((const __v4di *)__X, (__v4di)__M); |
||
| 840 | } |
||
| 841 | |||
| 842 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 843 | _mm_maskload_epi32(int const *__X, __m128i __M) |
||
| 844 | { |
||
| 845 | return (__m128i)__builtin_ia32_maskloadd((const __v4si *)__X, (__v4si)__M); |
||
| 846 | } |
||
| 847 | |||
| 848 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 849 | _mm_maskload_epi64(long long const *__X, __m128i __M) |
||
| 850 | { |
||
| 851 | return (__m128i)__builtin_ia32_maskloadq((const __v2di *)__X, (__v2di)__M); |
||
| 852 | } |
||
| 853 | |||
| 854 | static __inline__ void __DEFAULT_FN_ATTRS256 |
||
| 855 | _mm256_maskstore_epi32(int *__X, __m256i __M, __m256i __Y) |
||
| 856 | { |
||
| 857 | __builtin_ia32_maskstored256((__v8si *)__X, (__v8si)__M, (__v8si)__Y); |
||
| 858 | } |
||
| 859 | |||
| 860 | static __inline__ void __DEFAULT_FN_ATTRS256 |
||
| 861 | _mm256_maskstore_epi64(long long *__X, __m256i __M, __m256i __Y) |
||
| 862 | { |
||
| 863 | __builtin_ia32_maskstoreq256((__v4di *)__X, (__v4di)__M, (__v4di)__Y); |
||
| 864 | } |
||
| 865 | |||
| 866 | static __inline__ void __DEFAULT_FN_ATTRS128 |
||
| 867 | _mm_maskstore_epi32(int *__X, __m128i __M, __m128i __Y) |
||
| 868 | { |
||
| 869 | __builtin_ia32_maskstored((__v4si *)__X, (__v4si)__M, (__v4si)__Y); |
||
| 870 | } |
||
| 871 | |||
| 872 | static __inline__ void __DEFAULT_FN_ATTRS128 |
||
| 873 | _mm_maskstore_epi64(long long *__X, __m128i __M, __m128i __Y) |
||
| 874 | { |
||
| 875 | __builtin_ia32_maskstoreq(( __v2di *)__X, (__v2di)__M, (__v2di)__Y); |
||
| 876 | } |
||
| 877 | |||
| 878 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 879 | _mm256_sllv_epi32(__m256i __X, __m256i __Y) |
||
| 880 | { |
||
| 881 | return (__m256i)__builtin_ia32_psllv8si((__v8si)__X, (__v8si)__Y); |
||
| 882 | } |
||
| 883 | |||
| 884 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 885 | _mm_sllv_epi32(__m128i __X, __m128i __Y) |
||
| 886 | { |
||
| 887 | return (__m128i)__builtin_ia32_psllv4si((__v4si)__X, (__v4si)__Y); |
||
| 888 | } |
||
| 889 | |||
| 890 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 891 | _mm256_sllv_epi64(__m256i __X, __m256i __Y) |
||
| 892 | { |
||
| 893 | return (__m256i)__builtin_ia32_psllv4di((__v4di)__X, (__v4di)__Y); |
||
| 894 | } |
||
| 895 | |||
| 896 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 897 | _mm_sllv_epi64(__m128i __X, __m128i __Y) |
||
| 898 | { |
||
| 899 | return (__m128i)__builtin_ia32_psllv2di((__v2di)__X, (__v2di)__Y); |
||
| 900 | } |
||
| 901 | |||
| 902 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 903 | _mm256_srav_epi32(__m256i __X, __m256i __Y) |
||
| 904 | { |
||
| 905 | return (__m256i)__builtin_ia32_psrav8si((__v8si)__X, (__v8si)__Y); |
||
| 906 | } |
||
| 907 | |||
| 908 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 909 | _mm_srav_epi32(__m128i __X, __m128i __Y) |
||
| 910 | { |
||
| 911 | return (__m128i)__builtin_ia32_psrav4si((__v4si)__X, (__v4si)__Y); |
||
| 912 | } |
||
| 913 | |||
| 914 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 915 | _mm256_srlv_epi32(__m256i __X, __m256i __Y) |
||
| 916 | { |
||
| 917 | return (__m256i)__builtin_ia32_psrlv8si((__v8si)__X, (__v8si)__Y); |
||
| 918 | } |
||
| 919 | |||
| 920 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 921 | _mm_srlv_epi32(__m128i __X, __m128i __Y) |
||
| 922 | { |
||
| 923 | return (__m128i)__builtin_ia32_psrlv4si((__v4si)__X, (__v4si)__Y); |
||
| 924 | } |
||
| 925 | |||
| 926 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 927 | _mm256_srlv_epi64(__m256i __X, __m256i __Y) |
||
| 928 | { |
||
| 929 | return (__m256i)__builtin_ia32_psrlv4di((__v4di)__X, (__v4di)__Y); |
||
| 930 | } |
||
| 931 | |||
| 932 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 933 | _mm_srlv_epi64(__m128i __X, __m128i __Y) |
||
| 934 | { |
||
| 935 | return (__m128i)__builtin_ia32_psrlv2di((__v2di)__X, (__v2di)__Y); |
||
| 936 | } |
||
| 937 | |||
| 938 | #define _mm_mask_i32gather_pd(a, m, i, mask, s) \ |
||
| 939 | ((__m128d)__builtin_ia32_gatherd_pd((__v2df)(__m128i)(a), \ |
||
| 940 | (double const *)(m), \ |
||
| 941 | (__v4si)(__m128i)(i), \ |
||
| 942 | (__v2df)(__m128d)(mask), (s))) |
||
| 943 | |||
| 944 | #define _mm256_mask_i32gather_pd(a, m, i, mask, s) \ |
||
| 945 | ((__m256d)__builtin_ia32_gatherd_pd256((__v4df)(__m256d)(a), \ |
||
| 946 | (double const *)(m), \ |
||
| 947 | (__v4si)(__m128i)(i), \ |
||
| 948 | (__v4df)(__m256d)(mask), (s))) |
||
| 949 | |||
| 950 | #define _mm_mask_i64gather_pd(a, m, i, mask, s) \ |
||
| 951 | ((__m128d)__builtin_ia32_gatherq_pd((__v2df)(__m128d)(a), \ |
||
| 952 | (double const *)(m), \ |
||
| 953 | (__v2di)(__m128i)(i), \ |
||
| 954 | (__v2df)(__m128d)(mask), (s))) |
||
| 955 | |||
| 956 | #define _mm256_mask_i64gather_pd(a, m, i, mask, s) \ |
||
| 957 | ((__m256d)__builtin_ia32_gatherq_pd256((__v4df)(__m256d)(a), \ |
||
| 958 | (double const *)(m), \ |
||
| 959 | (__v4di)(__m256i)(i), \ |
||
| 960 | (__v4df)(__m256d)(mask), (s))) |
||
| 961 | |||
| 962 | #define _mm_mask_i32gather_ps(a, m, i, mask, s) \ |
||
| 963 | ((__m128)__builtin_ia32_gatherd_ps((__v4sf)(__m128)(a), \ |
||
| 964 | (float const *)(m), \ |
||
| 965 | (__v4si)(__m128i)(i), \ |
||
| 966 | (__v4sf)(__m128)(mask), (s))) |
||
| 967 | |||
| 968 | #define _mm256_mask_i32gather_ps(a, m, i, mask, s) \ |
||
| 969 | ((__m256)__builtin_ia32_gatherd_ps256((__v8sf)(__m256)(a), \ |
||
| 970 | (float const *)(m), \ |
||
| 971 | (__v8si)(__m256i)(i), \ |
||
| 972 | (__v8sf)(__m256)(mask), (s))) |
||
| 973 | |||
| 974 | #define _mm_mask_i64gather_ps(a, m, i, mask, s) \ |
||
| 975 | ((__m128)__builtin_ia32_gatherq_ps((__v4sf)(__m128)(a), \ |
||
| 976 | (float const *)(m), \ |
||
| 977 | (__v2di)(__m128i)(i), \ |
||
| 978 | (__v4sf)(__m128)(mask), (s))) |
||
| 979 | |||
| 980 | #define _mm256_mask_i64gather_ps(a, m, i, mask, s) \ |
||
| 981 | ((__m128)__builtin_ia32_gatherq_ps256((__v4sf)(__m128)(a), \ |
||
| 982 | (float const *)(m), \ |
||
| 983 | (__v4di)(__m256i)(i), \ |
||
| 984 | (__v4sf)(__m128)(mask), (s))) |
||
| 985 | |||
| 986 | #define _mm_mask_i32gather_epi32(a, m, i, mask, s) \ |
||
| 987 | ((__m128i)__builtin_ia32_gatherd_d((__v4si)(__m128i)(a), \ |
||
| 988 | (int const *)(m), \ |
||
| 989 | (__v4si)(__m128i)(i), \ |
||
| 990 | (__v4si)(__m128i)(mask), (s))) |
||
| 991 | |||
| 992 | #define _mm256_mask_i32gather_epi32(a, m, i, mask, s) \ |
||
| 993 | ((__m256i)__builtin_ia32_gatherd_d256((__v8si)(__m256i)(a), \ |
||
| 994 | (int const *)(m), \ |
||
| 995 | (__v8si)(__m256i)(i), \ |
||
| 996 | (__v8si)(__m256i)(mask), (s))) |
||
| 997 | |||
| 998 | #define _mm_mask_i64gather_epi32(a, m, i, mask, s) \ |
||
| 999 | ((__m128i)__builtin_ia32_gatherq_d((__v4si)(__m128i)(a), \ |
||
| 1000 | (int const *)(m), \ |
||
| 1001 | (__v2di)(__m128i)(i), \ |
||
| 1002 | (__v4si)(__m128i)(mask), (s))) |
||
| 1003 | |||
| 1004 | #define _mm256_mask_i64gather_epi32(a, m, i, mask, s) \ |
||
| 1005 | ((__m128i)__builtin_ia32_gatherq_d256((__v4si)(__m128i)(a), \ |
||
| 1006 | (int const *)(m), \ |
||
| 1007 | (__v4di)(__m256i)(i), \ |
||
| 1008 | (__v4si)(__m128i)(mask), (s))) |
||
| 1009 | |||
| 1010 | #define _mm_mask_i32gather_epi64(a, m, i, mask, s) \ |
||
| 1011 | ((__m128i)__builtin_ia32_gatherd_q((__v2di)(__m128i)(a), \ |
||
| 1012 | (long long const *)(m), \ |
||
| 1013 | (__v4si)(__m128i)(i), \ |
||
| 1014 | (__v2di)(__m128i)(mask), (s))) |
||
| 1015 | |||
| 1016 | #define _mm256_mask_i32gather_epi64(a, m, i, mask, s) \ |
||
| 1017 | ((__m256i)__builtin_ia32_gatherd_q256((__v4di)(__m256i)(a), \ |
||
| 1018 | (long long const *)(m), \ |
||
| 1019 | (__v4si)(__m128i)(i), \ |
||
| 1020 | (__v4di)(__m256i)(mask), (s))) |
||
| 1021 | |||
| 1022 | #define _mm_mask_i64gather_epi64(a, m, i, mask, s) \ |
||
| 1023 | ((__m128i)__builtin_ia32_gatherq_q((__v2di)(__m128i)(a), \ |
||
| 1024 | (long long const *)(m), \ |
||
| 1025 | (__v2di)(__m128i)(i), \ |
||
| 1026 | (__v2di)(__m128i)(mask), (s))) |
||
| 1027 | |||
| 1028 | #define _mm256_mask_i64gather_epi64(a, m, i, mask, s) \ |
||
| 1029 | ((__m256i)__builtin_ia32_gatherq_q256((__v4di)(__m256i)(a), \ |
||
| 1030 | (long long const *)(m), \ |
||
| 1031 | (__v4di)(__m256i)(i), \ |
||
| 1032 | (__v4di)(__m256i)(mask), (s))) |
||
| 1033 | |||
| 1034 | #define _mm_i32gather_pd(m, i, s) \ |
||
| 1035 | ((__m128d)__builtin_ia32_gatherd_pd((__v2df)_mm_undefined_pd(), \ |
||
| 1036 | (double const *)(m), \ |
||
| 1037 | (__v4si)(__m128i)(i), \ |
||
| 1038 | (__v2df)_mm_cmpeq_pd(_mm_setzero_pd(), \ |
||
| 1039 | _mm_setzero_pd()), \ |
||
| 1040 | (s))) |
||
| 1041 | |||
| 1042 | #define _mm256_i32gather_pd(m, i, s) \ |
||
| 1043 | ((__m256d)__builtin_ia32_gatherd_pd256((__v4df)_mm256_undefined_pd(), \ |
||
| 1044 | (double const *)(m), \ |
||
| 1045 | (__v4si)(__m128i)(i), \ |
||
| 1046 | (__v4df)_mm256_cmp_pd(_mm256_setzero_pd(), \ |
||
| 1047 | _mm256_setzero_pd(), \ |
||
| 1048 | _CMP_EQ_OQ), \ |
||
| 1049 | (s))) |
||
| 1050 | |||
| 1051 | #define _mm_i64gather_pd(m, i, s) \ |
||
| 1052 | ((__m128d)__builtin_ia32_gatherq_pd((__v2df)_mm_undefined_pd(), \ |
||
| 1053 | (double const *)(m), \ |
||
| 1054 | (__v2di)(__m128i)(i), \ |
||
| 1055 | (__v2df)_mm_cmpeq_pd(_mm_setzero_pd(), \ |
||
| 1056 | _mm_setzero_pd()), \ |
||
| 1057 | (s))) |
||
| 1058 | |||
| 1059 | #define _mm256_i64gather_pd(m, i, s) \ |
||
| 1060 | ((__m256d)__builtin_ia32_gatherq_pd256((__v4df)_mm256_undefined_pd(), \ |
||
| 1061 | (double const *)(m), \ |
||
| 1062 | (__v4di)(__m256i)(i), \ |
||
| 1063 | (__v4df)_mm256_cmp_pd(_mm256_setzero_pd(), \ |
||
| 1064 | _mm256_setzero_pd(), \ |
||
| 1065 | _CMP_EQ_OQ), \ |
||
| 1066 | (s))) |
||
| 1067 | |||
| 1068 | #define _mm_i32gather_ps(m, i, s) \ |
||
| 1069 | ((__m128)__builtin_ia32_gatherd_ps((__v4sf)_mm_undefined_ps(), \ |
||
| 1070 | (float const *)(m), \ |
||
| 1071 | (__v4si)(__m128i)(i), \ |
||
| 1072 | (__v4sf)_mm_cmpeq_ps(_mm_setzero_ps(), \ |
||
| 1073 | _mm_setzero_ps()), \ |
||
| 1074 | (s))) |
||
| 1075 | |||
| 1076 | #define _mm256_i32gather_ps(m, i, s) \ |
||
| 1077 | ((__m256)__builtin_ia32_gatherd_ps256((__v8sf)_mm256_undefined_ps(), \ |
||
| 1078 | (float const *)(m), \ |
||
| 1079 | (__v8si)(__m256i)(i), \ |
||
| 1080 | (__v8sf)_mm256_cmp_ps(_mm256_setzero_ps(), \ |
||
| 1081 | _mm256_setzero_ps(), \ |
||
| 1082 | _CMP_EQ_OQ), \ |
||
| 1083 | (s))) |
||
| 1084 | |||
| 1085 | #define _mm_i64gather_ps(m, i, s) \ |
||
| 1086 | ((__m128)__builtin_ia32_gatherq_ps((__v4sf)_mm_undefined_ps(), \ |
||
| 1087 | (float const *)(m), \ |
||
| 1088 | (__v2di)(__m128i)(i), \ |
||
| 1089 | (__v4sf)_mm_cmpeq_ps(_mm_setzero_ps(), \ |
||
| 1090 | _mm_setzero_ps()), \ |
||
| 1091 | (s))) |
||
| 1092 | |||
| 1093 | #define _mm256_i64gather_ps(m, i, s) \ |
||
| 1094 | ((__m128)__builtin_ia32_gatherq_ps256((__v4sf)_mm_undefined_ps(), \ |
||
| 1095 | (float const *)(m), \ |
||
| 1096 | (__v4di)(__m256i)(i), \ |
||
| 1097 | (__v4sf)_mm_cmpeq_ps(_mm_setzero_ps(), \ |
||
| 1098 | _mm_setzero_ps()), \ |
||
| 1099 | (s))) |
||
| 1100 | |||
| 1101 | #define _mm_i32gather_epi32(m, i, s) \ |
||
| 1102 | ((__m128i)__builtin_ia32_gatherd_d((__v4si)_mm_undefined_si128(), \ |
||
| 1103 | (int const *)(m), (__v4si)(__m128i)(i), \ |
||
| 1104 | (__v4si)_mm_set1_epi32(-1), (s))) |
||
| 1105 | |||
| 1106 | #define _mm256_i32gather_epi32(m, i, s) \ |
||
| 1107 | ((__m256i)__builtin_ia32_gatherd_d256((__v8si)_mm256_undefined_si256(), \ |
||
| 1108 | (int const *)(m), (__v8si)(__m256i)(i), \ |
||
| 1109 | (__v8si)_mm256_set1_epi32(-1), (s))) |
||
| 1110 | |||
| 1111 | #define _mm_i64gather_epi32(m, i, s) \ |
||
| 1112 | ((__m128i)__builtin_ia32_gatherq_d((__v4si)_mm_undefined_si128(), \ |
||
| 1113 | (int const *)(m), (__v2di)(__m128i)(i), \ |
||
| 1114 | (__v4si)_mm_set1_epi32(-1), (s))) |
||
| 1115 | |||
| 1116 | #define _mm256_i64gather_epi32(m, i, s) \ |
||
| 1117 | ((__m128i)__builtin_ia32_gatherq_d256((__v4si)_mm_undefined_si128(), \ |
||
| 1118 | (int const *)(m), (__v4di)(__m256i)(i), \ |
||
| 1119 | (__v4si)_mm_set1_epi32(-1), (s))) |
||
| 1120 | |||
| 1121 | #define _mm_i32gather_epi64(m, i, s) \ |
||
| 1122 | ((__m128i)__builtin_ia32_gatherd_q((__v2di)_mm_undefined_si128(), \ |
||
| 1123 | (long long const *)(m), \ |
||
| 1124 | (__v4si)(__m128i)(i), \ |
||
| 1125 | (__v2di)_mm_set1_epi64x(-1), (s))) |
||
| 1126 | |||
| 1127 | #define _mm256_i32gather_epi64(m, i, s) \ |
||
| 1128 | ((__m256i)__builtin_ia32_gatherd_q256((__v4di)_mm256_undefined_si256(), \ |
||
| 1129 | (long long const *)(m), \ |
||
| 1130 | (__v4si)(__m128i)(i), \ |
||
| 1131 | (__v4di)_mm256_set1_epi64x(-1), (s))) |
||
| 1132 | |||
| 1133 | #define _mm_i64gather_epi64(m, i, s) \ |
||
| 1134 | ((__m128i)__builtin_ia32_gatherq_q((__v2di)_mm_undefined_si128(), \ |
||
| 1135 | (long long const *)(m), \ |
||
| 1136 | (__v2di)(__m128i)(i), \ |
||
| 1137 | (__v2di)_mm_set1_epi64x(-1), (s))) |
||
| 1138 | |||
| 1139 | #define _mm256_i64gather_epi64(m, i, s) \ |
||
| 1140 | ((__m256i)__builtin_ia32_gatherq_q256((__v4di)_mm256_undefined_si256(), \ |
||
| 1141 | (long long const *)(m), \ |
||
| 1142 | (__v4di)(__m256i)(i), \ |
||
| 1143 | (__v4di)_mm256_set1_epi64x(-1), (s))) |
||
| 1144 | |||
| 1145 | #undef __DEFAULT_FN_ATTRS256 |
||
| 1146 | #undef __DEFAULT_FN_ATTRS128 |
||
| 1147 | |||
| 1148 | #endif /* __AVX2INTRIN_H */ |