root / lab4 / .minix-src / include / clang-3.6 / avx512fintrin.h @ 14
History | View | Annotate | Download (34.5 KB)
| 1 |
/*===---- avx512fintrin.h - AVX2 intrinsics --------------------------------===
|
|---|---|
| 2 |
*
|
| 3 |
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
| 4 |
* of this software and associated documentation files (the "Software"), to deal
|
| 5 |
* in the Software without restriction, including without limitation the rights
|
| 6 |
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
| 7 |
* copies of the Software, and to permit persons to whom the Software is
|
| 8 |
* furnished to do so, subject to the following conditions:
|
| 9 |
*
|
| 10 |
* The above copyright notice and this permission notice shall be included in
|
| 11 |
* all copies or substantial portions of the Software.
|
| 12 |
*
|
| 13 |
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
| 14 |
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
| 15 |
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
| 16 |
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
| 17 |
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
| 18 |
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
| 19 |
* THE SOFTWARE.
|
| 20 |
*
|
| 21 |
*===-----------------------------------------------------------------------===
|
| 22 |
*/
|
| 23 |
#ifndef __IMMINTRIN_H
|
| 24 |
#error "Never use <avx512fintrin.h> directly; include <immintrin.h> instead." |
| 25 |
#endif
|
| 26 |
|
| 27 |
#ifndef __AVX512FINTRIN_H
|
| 28 |
#define __AVX512FINTRIN_H
|
| 29 |
|
| 30 |
typedef double __v8df __attribute__((__vector_size__(64))); |
| 31 |
typedef float __v16sf __attribute__((__vector_size__(64))); |
| 32 |
typedef long long __v8di __attribute__((__vector_size__(64))); |
| 33 |
typedef int __v16si __attribute__((__vector_size__(64))); |
| 34 |
|
| 35 |
typedef float __m512 __attribute__((__vector_size__(64))); |
| 36 |
typedef double __m512d __attribute__((__vector_size__(64))); |
| 37 |
typedef long long __m512i __attribute__((__vector_size__(64))); |
| 38 |
|
| 39 |
typedef unsigned char __mmask8; |
| 40 |
typedef unsigned short __mmask16; |
| 41 |
|
| 42 |
/* Rounding mode macros. */
|
| 43 |
#define _MM_FROUND_TO_NEAREST_INT 0x00 |
| 44 |
#define _MM_FROUND_TO_NEG_INF 0x01 |
| 45 |
#define _MM_FROUND_TO_POS_INF 0x02 |
| 46 |
#define _MM_FROUND_TO_ZERO 0x03 |
| 47 |
#define _MM_FROUND_CUR_DIRECTION 0x04 |
| 48 |
|
| 49 |
/* Create vectors with repeated elements */
|
| 50 |
|
| 51 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 52 |
_mm512_setzero_si512(void)
|
| 53 |
{
|
| 54 |
return (__m512i)(__v8di){ 0, 0, 0, 0, 0, 0, 0, 0 }; |
| 55 |
} |
| 56 |
|
| 57 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 58 |
_mm512_maskz_set1_epi32(__mmask16 __M, int __A)
|
| 59 |
{
|
| 60 |
return (__m512i) __builtin_ia32_pbroadcastd512_gpr_mask (__A,
|
| 61 |
(__v16si) |
| 62 |
_mm512_setzero_si512 (), |
| 63 |
__M); |
| 64 |
} |
| 65 |
|
| 66 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 67 |
_mm512_maskz_set1_epi64(__mmask8 __M, long long __A) |
| 68 |
{
|
| 69 |
#ifdef __x86_64__
|
| 70 |
return (__m512i) __builtin_ia32_pbroadcastq512_gpr_mask (__A,
|
| 71 |
(__v8di) |
| 72 |
_mm512_setzero_si512 (), |
| 73 |
__M); |
| 74 |
#else
|
| 75 |
return (__m512i) __builtin_ia32_pbroadcastq512_mem_mask (__A,
|
| 76 |
(__v8di) |
| 77 |
_mm512_setzero_si512 (), |
| 78 |
__M); |
| 79 |
#endif
|
| 80 |
} |
| 81 |
|
| 82 |
static __inline __m512 __attribute__ ((__always_inline__, __nodebug__))
|
| 83 |
_mm512_setzero_ps(void)
|
| 84 |
{
|
| 85 |
return (__m512){ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, |
| 86 |
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 }; |
| 87 |
} |
| 88 |
static __inline __m512d __attribute__ ((__always_inline__, __nodebug__))
|
| 89 |
_mm512_setzero_pd(void)
|
| 90 |
{
|
| 91 |
return (__m512d){ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 }; |
| 92 |
} |
| 93 |
|
| 94 |
static __inline __m512 __attribute__((__always_inline__, __nodebug__))
|
| 95 |
_mm512_set1_ps(float __w)
|
| 96 |
{
|
| 97 |
return (__m512){ __w, __w, __w, __w, __w, __w, __w, __w,
|
| 98 |
__w, __w, __w, __w, __w, __w, __w, __w }; |
| 99 |
} |
| 100 |
|
| 101 |
static __inline __m512d __attribute__((__always_inline__, __nodebug__))
|
| 102 |
_mm512_set1_pd(double __w)
|
| 103 |
{
|
| 104 |
return (__m512d){ __w, __w, __w, __w, __w, __w, __w, __w };
|
| 105 |
} |
| 106 |
|
| 107 |
static __inline __m512i __attribute__((__always_inline__, __nodebug__))
|
| 108 |
_mm512_set1_epi32(int __s)
|
| 109 |
{
|
| 110 |
return (__m512i)(__v16si){ __s, __s, __s, __s, __s, __s, __s, __s,
|
| 111 |
__s, __s, __s, __s, __s, __s, __s, __s }; |
| 112 |
} |
| 113 |
|
| 114 |
static __inline __m512i __attribute__((__always_inline__, __nodebug__))
|
| 115 |
_mm512_set1_epi64(long long __d) |
| 116 |
{
|
| 117 |
return (__m512i)(__v8di){ __d, __d, __d, __d, __d, __d, __d, __d };
|
| 118 |
} |
| 119 |
|
| 120 |
static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
|
| 121 |
_mm512_broadcastss_ps(__m128 __X) |
| 122 |
{
|
| 123 |
float __f = __X[0]; |
| 124 |
return (__v16sf){ __f, __f, __f, __f,
|
| 125 |
__f, __f, __f, __f, |
| 126 |
__f, __f, __f, __f, |
| 127 |
__f, __f, __f, __f }; |
| 128 |
} |
| 129 |
|
| 130 |
static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
|
| 131 |
_mm512_broadcastsd_pd(__m128d __X) |
| 132 |
{
|
| 133 |
double __d = __X[0]; |
| 134 |
return (__v8df){ __d, __d, __d, __d,
|
| 135 |
__d, __d, __d, __d }; |
| 136 |
} |
| 137 |
|
| 138 |
/* Cast between vector types */
|
| 139 |
|
| 140 |
static __inline __m512d __attribute__((__always_inline__, __nodebug__))
|
| 141 |
_mm512_castpd256_pd512(__m256d __a) |
| 142 |
{
|
| 143 |
return __builtin_shufflevector(__a, __a, 0, 1, 2, 3, -1, -1, -1, -1); |
| 144 |
} |
| 145 |
|
| 146 |
static __inline __m512 __attribute__((__always_inline__, __nodebug__))
|
| 147 |
_mm512_castps256_ps512(__m256 __a) |
| 148 |
{
|
| 149 |
return __builtin_shufflevector(__a, __a, 0, 1, 2, 3, 4, 5, 6, 7, |
| 150 |
-1, -1, -1, -1, -1, -1, -1, -1); |
| 151 |
} |
| 152 |
|
| 153 |
static __inline __m128d __attribute__((__always_inline__, __nodebug__))
|
| 154 |
_mm512_castpd512_pd128(__m512d __a) |
| 155 |
{
|
| 156 |
return __builtin_shufflevector(__a, __a, 0, 1); |
| 157 |
} |
| 158 |
|
| 159 |
static __inline __m128 __attribute__((__always_inline__, __nodebug__))
|
| 160 |
_mm512_castps512_ps128(__m512 __a) |
| 161 |
{
|
| 162 |
return __builtin_shufflevector(__a, __a, 0, 1, 2, 3); |
| 163 |
} |
| 164 |
|
| 165 |
/* Arithmetic */
|
| 166 |
|
| 167 |
static __inline __m512d __attribute__((__always_inline__, __nodebug__))
|
| 168 |
_mm512_add_pd(__m512d __a, __m512d __b) |
| 169 |
{
|
| 170 |
return __a + __b;
|
| 171 |
} |
| 172 |
|
| 173 |
static __inline __m512 __attribute__((__always_inline__, __nodebug__))
|
| 174 |
_mm512_add_ps(__m512 __a, __m512 __b) |
| 175 |
{
|
| 176 |
return __a + __b;
|
| 177 |
} |
| 178 |
|
| 179 |
static __inline __m512d __attribute__((__always_inline__, __nodebug__))
|
| 180 |
_mm512_mul_pd(__m512d __a, __m512d __b) |
| 181 |
{
|
| 182 |
return __a * __b;
|
| 183 |
} |
| 184 |
|
| 185 |
static __inline __m512 __attribute__((__always_inline__, __nodebug__))
|
| 186 |
_mm512_mul_ps(__m512 __a, __m512 __b) |
| 187 |
{
|
| 188 |
return __a * __b;
|
| 189 |
} |
| 190 |
|
| 191 |
static __inline __m512d __attribute__((__always_inline__, __nodebug__))
|
| 192 |
_mm512_sub_pd(__m512d __a, __m512d __b) |
| 193 |
{
|
| 194 |
return __a - __b;
|
| 195 |
} |
| 196 |
|
| 197 |
static __inline __m512 __attribute__((__always_inline__, __nodebug__))
|
| 198 |
_mm512_sub_ps(__m512 __a, __m512 __b) |
| 199 |
{
|
| 200 |
return __a - __b;
|
| 201 |
} |
| 202 |
|
| 203 |
static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
|
| 204 |
_mm512_max_pd(__m512d __A, __m512d __B) |
| 205 |
{
|
| 206 |
return (__m512d) __builtin_ia32_maxpd512_mask ((__v8df) __A,
|
| 207 |
(__v8df) __B, |
| 208 |
(__v8df) |
| 209 |
_mm512_setzero_pd (), |
| 210 |
(__mmask8) -1,
|
| 211 |
_MM_FROUND_CUR_DIRECTION); |
| 212 |
} |
| 213 |
|
| 214 |
static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
|
| 215 |
_mm512_max_ps(__m512 __A, __m512 __B) |
| 216 |
{
|
| 217 |
return (__m512) __builtin_ia32_maxps512_mask ((__v16sf) __A,
|
| 218 |
(__v16sf) __B, |
| 219 |
(__v16sf) |
| 220 |
_mm512_setzero_ps (), |
| 221 |
(__mmask16) -1,
|
| 222 |
_MM_FROUND_CUR_DIRECTION); |
| 223 |
} |
| 224 |
|
| 225 |
static __inline __m512i
|
| 226 |
__attribute__ ((__always_inline__, __nodebug__)) |
| 227 |
_mm512_max_epi32(__m512i __A, __m512i __B) |
| 228 |
{
|
| 229 |
return (__m512i) __builtin_ia32_pmaxsd512_mask ((__v16si) __A,
|
| 230 |
(__v16si) __B, |
| 231 |
(__v16si) |
| 232 |
_mm512_setzero_si512 (), |
| 233 |
(__mmask16) -1);
|
| 234 |
} |
| 235 |
|
| 236 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 237 |
_mm512_max_epu32(__m512i __A, __m512i __B) |
| 238 |
{
|
| 239 |
return (__m512i) __builtin_ia32_pmaxud512_mask ((__v16si) __A,
|
| 240 |
(__v16si) __B, |
| 241 |
(__v16si) |
| 242 |
_mm512_setzero_si512 (), |
| 243 |
(__mmask16) -1);
|
| 244 |
} |
| 245 |
|
| 246 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 247 |
_mm512_max_epi64(__m512i __A, __m512i __B) |
| 248 |
{
|
| 249 |
return (__m512i) __builtin_ia32_pmaxsq512_mask ((__v8di) __A,
|
| 250 |
(__v8di) __B, |
| 251 |
(__v8di) |
| 252 |
_mm512_setzero_si512 (), |
| 253 |
(__mmask8) -1);
|
| 254 |
} |
| 255 |
|
| 256 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 257 |
_mm512_max_epu64(__m512i __A, __m512i __B) |
| 258 |
{
|
| 259 |
return (__m512i) __builtin_ia32_pmaxuq512_mask ((__v8di) __A,
|
| 260 |
(__v8di) __B, |
| 261 |
(__v8di) |
| 262 |
_mm512_setzero_si512 (), |
| 263 |
(__mmask8) -1);
|
| 264 |
} |
| 265 |
|
| 266 |
static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
|
| 267 |
_mm512_min_pd(__m512d __A, __m512d __B) |
| 268 |
{
|
| 269 |
return (__m512d) __builtin_ia32_minpd512_mask ((__v8df) __A,
|
| 270 |
(__v8df) __B, |
| 271 |
(__v8df) |
| 272 |
_mm512_setzero_pd (), |
| 273 |
(__mmask8) -1,
|
| 274 |
_MM_FROUND_CUR_DIRECTION); |
| 275 |
} |
| 276 |
|
| 277 |
static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
|
| 278 |
_mm512_min_ps(__m512 __A, __m512 __B) |
| 279 |
{
|
| 280 |
return (__m512) __builtin_ia32_minps512_mask ((__v16sf) __A,
|
| 281 |
(__v16sf) __B, |
| 282 |
(__v16sf) |
| 283 |
_mm512_setzero_ps (), |
| 284 |
(__mmask16) -1,
|
| 285 |
_MM_FROUND_CUR_DIRECTION); |
| 286 |
} |
| 287 |
|
| 288 |
static __inline __m512i
|
| 289 |
__attribute__ ((__always_inline__, __nodebug__)) |
| 290 |
_mm512_min_epi32(__m512i __A, __m512i __B) |
| 291 |
{
|
| 292 |
return (__m512i) __builtin_ia32_pminsd512_mask ((__v16si) __A,
|
| 293 |
(__v16si) __B, |
| 294 |
(__v16si) |
| 295 |
_mm512_setzero_si512 (), |
| 296 |
(__mmask16) -1);
|
| 297 |
} |
| 298 |
|
| 299 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 300 |
_mm512_min_epu32(__m512i __A, __m512i __B) |
| 301 |
{
|
| 302 |
return (__m512i) __builtin_ia32_pminud512_mask ((__v16si) __A,
|
| 303 |
(__v16si) __B, |
| 304 |
(__v16si) |
| 305 |
_mm512_setzero_si512 (), |
| 306 |
(__mmask16) -1);
|
| 307 |
} |
| 308 |
|
| 309 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 310 |
_mm512_min_epi64(__m512i __A, __m512i __B) |
| 311 |
{
|
| 312 |
return (__m512i) __builtin_ia32_pminsq512_mask ((__v8di) __A,
|
| 313 |
(__v8di) __B, |
| 314 |
(__v8di) |
| 315 |
_mm512_setzero_si512 (), |
| 316 |
(__mmask8) -1);
|
| 317 |
} |
| 318 |
|
| 319 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 320 |
_mm512_min_epu64(__m512i __A, __m512i __B) |
| 321 |
{
|
| 322 |
return (__m512i) __builtin_ia32_pminuq512_mask ((__v8di) __A,
|
| 323 |
(__v8di) __B, |
| 324 |
(__v8di) |
| 325 |
_mm512_setzero_si512 (), |
| 326 |
(__mmask8) -1);
|
| 327 |
} |
| 328 |
|
| 329 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 330 |
_mm512_mul_epi32(__m512i __X, __m512i __Y) |
| 331 |
{
|
| 332 |
return (__m512i) __builtin_ia32_pmuldq512_mask ((__v16si) __X,
|
| 333 |
(__v16si) __Y, |
| 334 |
(__v8di) |
| 335 |
_mm512_setzero_si512 (), |
| 336 |
(__mmask8) -1);
|
| 337 |
} |
| 338 |
|
| 339 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 340 |
_mm512_mul_epu32(__m512i __X, __m512i __Y) |
| 341 |
{
|
| 342 |
return (__m512i) __builtin_ia32_pmuludq512_mask ((__v16si) __X,
|
| 343 |
(__v16si) __Y, |
| 344 |
(__v8di) |
| 345 |
_mm512_setzero_si512 (), |
| 346 |
(__mmask8) -1);
|
| 347 |
} |
| 348 |
|
| 349 |
static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
|
| 350 |
_mm512_sqrt_pd(__m512d a) |
| 351 |
{
|
| 352 |
return (__m512d)__builtin_ia32_sqrtpd512_mask((__v8df)a,
|
| 353 |
(__v8df) _mm512_setzero_pd (), |
| 354 |
(__mmask8) -1,
|
| 355 |
_MM_FROUND_CUR_DIRECTION); |
| 356 |
} |
| 357 |
|
| 358 |
static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
|
| 359 |
_mm512_sqrt_ps(__m512 a) |
| 360 |
{
|
| 361 |
return (__m512)__builtin_ia32_sqrtps512_mask((__v16sf)a,
|
| 362 |
(__v16sf) _mm512_setzero_ps (), |
| 363 |
(__mmask16) -1,
|
| 364 |
_MM_FROUND_CUR_DIRECTION); |
| 365 |
} |
| 366 |
|
| 367 |
static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
|
| 368 |
_mm512_rsqrt14_pd(__m512d __A) |
| 369 |
{
|
| 370 |
return (__m512d) __builtin_ia32_rsqrt14pd512_mask ((__v8df) __A,
|
| 371 |
(__v8df) |
| 372 |
_mm512_setzero_pd (), |
| 373 |
(__mmask8) -1);}
|
| 374 |
|
| 375 |
static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
|
| 376 |
_mm512_rsqrt14_ps(__m512 __A) |
| 377 |
{
|
| 378 |
return (__m512) __builtin_ia32_rsqrt14ps512_mask ((__v16sf) __A,
|
| 379 |
(__v16sf) |
| 380 |
_mm512_setzero_ps (), |
| 381 |
(__mmask16) -1);
|
| 382 |
} |
| 383 |
|
| 384 |
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
|
| 385 |
_mm_rsqrt14_ss(__m128 __A, __m128 __B) |
| 386 |
{
|
| 387 |
return (__m128) __builtin_ia32_rsqrt14ss_mask ((__v4sf) __A,
|
| 388 |
(__v4sf) __B, |
| 389 |
(__v4sf) |
| 390 |
_mm_setzero_ps (), |
| 391 |
(__mmask8) -1);
|
| 392 |
} |
| 393 |
|
| 394 |
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
|
| 395 |
_mm_rsqrt14_sd(__m128d __A, __m128d __B) |
| 396 |
{
|
| 397 |
return (__m128d) __builtin_ia32_rsqrt14sd_mask ((__v2df) __A,
|
| 398 |
(__v2df) __B, |
| 399 |
(__v2df) |
| 400 |
_mm_setzero_pd (), |
| 401 |
(__mmask8) -1);
|
| 402 |
} |
| 403 |
|
| 404 |
static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
|
| 405 |
_mm512_rcp14_pd(__m512d __A) |
| 406 |
{
|
| 407 |
return (__m512d) __builtin_ia32_rcp14pd512_mask ((__v8df) __A,
|
| 408 |
(__v8df) |
| 409 |
_mm512_setzero_pd (), |
| 410 |
(__mmask8) -1);
|
| 411 |
} |
| 412 |
|
| 413 |
static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
|
| 414 |
_mm512_rcp14_ps(__m512 __A) |
| 415 |
{
|
| 416 |
return (__m512) __builtin_ia32_rcp14ps512_mask ((__v16sf) __A,
|
| 417 |
(__v16sf) |
| 418 |
_mm512_setzero_ps (), |
| 419 |
(__mmask16) -1);
|
| 420 |
} |
| 421 |
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
|
| 422 |
_mm_rcp14_ss(__m128 __A, __m128 __B) |
| 423 |
{
|
| 424 |
return (__m128) __builtin_ia32_rcp14ss_mask ((__v4sf) __A,
|
| 425 |
(__v4sf) __B, |
| 426 |
(__v4sf) |
| 427 |
_mm_setzero_ps (), |
| 428 |
(__mmask8) -1);
|
| 429 |
} |
| 430 |
|
| 431 |
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
|
| 432 |
_mm_rcp14_sd(__m128d __A, __m128d __B) |
| 433 |
{
|
| 434 |
return (__m128d) __builtin_ia32_rcp14sd_mask ((__v2df) __A,
|
| 435 |
(__v2df) __B, |
| 436 |
(__v2df) |
| 437 |
_mm_setzero_pd (), |
| 438 |
(__mmask8) -1);
|
| 439 |
} |
| 440 |
|
| 441 |
static __inline __m512 __attribute__ ((__always_inline__, __nodebug__))
|
| 442 |
_mm512_floor_ps(__m512 __A) |
| 443 |
{
|
| 444 |
return (__m512) __builtin_ia32_rndscaleps_mask ((__v16sf) __A,
|
| 445 |
_MM_FROUND_FLOOR, |
| 446 |
(__v16sf) __A, -1,
|
| 447 |
_MM_FROUND_CUR_DIRECTION); |
| 448 |
} |
| 449 |
|
| 450 |
static __inline __m512d __attribute__ ((__always_inline__, __nodebug__))
|
| 451 |
_mm512_floor_pd(__m512d __A) |
| 452 |
{
|
| 453 |
return (__m512d) __builtin_ia32_rndscalepd_mask ((__v8df) __A,
|
| 454 |
_MM_FROUND_FLOOR, |
| 455 |
(__v8df) __A, -1,
|
| 456 |
_MM_FROUND_CUR_DIRECTION); |
| 457 |
} |
| 458 |
|
| 459 |
static __inline __m512 __attribute__ ((__always_inline__, __nodebug__))
|
| 460 |
_mm512_ceil_ps(__m512 __A) |
| 461 |
{
|
| 462 |
return (__m512) __builtin_ia32_rndscaleps_mask ((__v16sf) __A,
|
| 463 |
_MM_FROUND_CEIL, |
| 464 |
(__v16sf) __A, -1,
|
| 465 |
_MM_FROUND_CUR_DIRECTION); |
| 466 |
} |
| 467 |
|
| 468 |
static __inline __m512d __attribute__ ((__always_inline__, __nodebug__))
|
| 469 |
_mm512_ceil_pd(__m512d __A) |
| 470 |
{
|
| 471 |
return (__m512d) __builtin_ia32_rndscalepd_mask ((__v8df) __A,
|
| 472 |
_MM_FROUND_CEIL, |
| 473 |
(__v8df) __A, -1,
|
| 474 |
_MM_FROUND_CUR_DIRECTION); |
| 475 |
} |
| 476 |
|
| 477 |
static __inline __m512i __attribute__ (( __always_inline__, __nodebug__))
|
| 478 |
_mm512_abs_epi64(__m512i __A) |
| 479 |
{
|
| 480 |
return (__m512i) __builtin_ia32_pabsq512_mask ((__v8di) __A,
|
| 481 |
(__v8di) |
| 482 |
_mm512_setzero_si512 (), |
| 483 |
(__mmask8) -1);
|
| 484 |
} |
| 485 |
|
| 486 |
static __inline __m512i __attribute__ (( __always_inline__, __nodebug__))
|
| 487 |
_mm512_abs_epi32(__m512i __A) |
| 488 |
{
|
| 489 |
return (__m512i) __builtin_ia32_pabsd512_mask ((__v16si) __A,
|
| 490 |
(__v16si) |
| 491 |
_mm512_setzero_si512 (), |
| 492 |
(__mmask16) -1);
|
| 493 |
} |
| 494 |
|
| 495 |
static __inline __m512 __attribute__ ((__always_inline__, __nodebug__))
|
| 496 |
_mm512_roundscale_ps(__m512 __A, const int __imm) |
| 497 |
{
|
| 498 |
return (__m512) __builtin_ia32_rndscaleps_mask ((__v16sf) __A, __imm,
|
| 499 |
(__v16sf) __A, -1,
|
| 500 |
_MM_FROUND_CUR_DIRECTION); |
| 501 |
} |
| 502 |
static __inline __m512d __attribute__ ((__always_inline__, __nodebug__))
|
| 503 |
_mm512_roundscale_pd(__m512d __A, const int __imm) |
| 504 |
{
|
| 505 |
return (__m512d) __builtin_ia32_rndscalepd_mask ((__v8df) __A, __imm,
|
| 506 |
(__v8df) __A, -1,
|
| 507 |
_MM_FROUND_CUR_DIRECTION); |
| 508 |
} |
| 509 |
|
| 510 |
static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
|
| 511 |
_mm512_fmadd_pd(__m512d __A, __m512d __B, __m512d __C) |
| 512 |
{
|
| 513 |
return (__m512d)
|
| 514 |
__builtin_ia32_vfmaddpd512_mask(__A, |
| 515 |
__B, |
| 516 |
__C, |
| 517 |
(__mmask8) -1,
|
| 518 |
_MM_FROUND_CUR_DIRECTION); |
| 519 |
} |
| 520 |
|
| 521 |
static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
|
| 522 |
_mm512_fmsub_pd(__m512d __A, __m512d __B, __m512d __C) |
| 523 |
{
|
| 524 |
return (__m512d)
|
| 525 |
__builtin_ia32_vfmsubpd512_mask(__A, |
| 526 |
__B, |
| 527 |
__C, |
| 528 |
(__mmask8) -1,
|
| 529 |
_MM_FROUND_CUR_DIRECTION); |
| 530 |
} |
| 531 |
|
| 532 |
static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
|
| 533 |
_mm512_fnmadd_pd(__m512d __A, __m512d __B, __m512d __C) |
| 534 |
{
|
| 535 |
return (__m512d)
|
| 536 |
__builtin_ia32_vfnmaddpd512_mask(__A, |
| 537 |
__B, |
| 538 |
__C, |
| 539 |
(__mmask8) -1,
|
| 540 |
_MM_FROUND_CUR_DIRECTION); |
| 541 |
} |
| 542 |
|
| 543 |
static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
|
| 544 |
_mm512_fmadd_ps(__m512 __A, __m512 __B, __m512 __C) |
| 545 |
{
|
| 546 |
return (__m512)
|
| 547 |
__builtin_ia32_vfmaddps512_mask(__A, |
| 548 |
__B, |
| 549 |
__C, |
| 550 |
(__mmask16) -1,
|
| 551 |
_MM_FROUND_CUR_DIRECTION); |
| 552 |
} |
| 553 |
|
| 554 |
static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
|
| 555 |
_mm512_fmsub_ps(__m512 __A, __m512 __B, __m512 __C) |
| 556 |
{
|
| 557 |
return (__m512)
|
| 558 |
__builtin_ia32_vfmsubps512_mask(__A, |
| 559 |
__B, |
| 560 |
__C, |
| 561 |
(__mmask16) -1,
|
| 562 |
_MM_FROUND_CUR_DIRECTION); |
| 563 |
} |
| 564 |
|
| 565 |
static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
|
| 566 |
_mm512_fnmadd_ps(__m512 __A, __m512 __B, __m512 __C) |
| 567 |
{
|
| 568 |
return (__m512)
|
| 569 |
__builtin_ia32_vfnmaddps512_mask(__A, |
| 570 |
__B, |
| 571 |
__C, |
| 572 |
(__mmask16) -1,
|
| 573 |
_MM_FROUND_CUR_DIRECTION); |
| 574 |
} |
| 575 |
|
| 576 |
/* Vector permutations */
|
| 577 |
|
| 578 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 579 |
_mm512_permutex2var_epi32(__m512i __A, __m512i __I, __m512i __B) |
| 580 |
{
|
| 581 |
return (__m512i) __builtin_ia32_vpermt2vard512_mask ((__v16si) __I
|
| 582 |
/* idx */ ,
|
| 583 |
(__v16si) __A, |
| 584 |
(__v16si) __B, |
| 585 |
(__mmask16) -1);
|
| 586 |
} |
| 587 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 588 |
_mm512_permutex2var_epi64(__m512i __A, __m512i __I, __m512i __B) |
| 589 |
{
|
| 590 |
return (__m512i) __builtin_ia32_vpermt2varq512_mask ((__v8di) __I
|
| 591 |
/* idx */ ,
|
| 592 |
(__v8di) __A, |
| 593 |
(__v8di) __B, |
| 594 |
(__mmask8) -1);
|
| 595 |
} |
| 596 |
|
| 597 |
static __inline __m512d __attribute__ ((__always_inline__, __nodebug__))
|
| 598 |
_mm512_permutex2var_pd(__m512d __A, __m512i __I, __m512d __B) |
| 599 |
{
|
| 600 |
return (__m512d) __builtin_ia32_vpermt2varpd512_mask ((__v8di) __I
|
| 601 |
/* idx */ ,
|
| 602 |
(__v8df) __A, |
| 603 |
(__v8df) __B, |
| 604 |
(__mmask8) -1);
|
| 605 |
} |
| 606 |
static __inline __m512 __attribute__ ((__always_inline__, __nodebug__))
|
| 607 |
_mm512_permutex2var_ps(__m512 __A, __m512i __I, __m512 __B) |
| 608 |
{
|
| 609 |
return (__m512) __builtin_ia32_vpermt2varps512_mask ((__v16si) __I
|
| 610 |
/* idx */ ,
|
| 611 |
(__v16sf) __A, |
| 612 |
(__v16sf) __B, |
| 613 |
(__mmask16) -1);
|
| 614 |
} |
| 615 |
|
| 616 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 617 |
_mm512_valign_epi64(__m512i __A, __m512i __B, const int __I) |
| 618 |
{
|
| 619 |
return (__m512i) __builtin_ia32_alignq512_mask((__v8di)__A,
|
| 620 |
(__v8di)__B, |
| 621 |
__I, |
| 622 |
(__v8di)_mm512_setzero_si512(), |
| 623 |
(__mmask8) -1);
|
| 624 |
} |
| 625 |
|
| 626 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 627 |
_mm512_valign_epi32(__m512i __A, __m512i __B, const int __I) |
| 628 |
{
|
| 629 |
return (__m512i)__builtin_ia32_alignd512_mask((__v16si)__A,
|
| 630 |
(__v16si)__B, |
| 631 |
__I, |
| 632 |
(__v16si)_mm512_setzero_si512(), |
| 633 |
(__mmask16) -1);
|
| 634 |
} |
| 635 |
|
| 636 |
/* Vector Blend */
|
| 637 |
|
| 638 |
static __inline __m512d __attribute__ ((__always_inline__, __nodebug__))
|
| 639 |
_mm512_mask_blend_pd(__mmask8 __U, __m512d __A, __m512d __W) |
| 640 |
{
|
| 641 |
return (__m512d) __builtin_ia32_blendmpd_512_mask ((__v8df) __A,
|
| 642 |
(__v8df) __W, |
| 643 |
(__mmask8) __U); |
| 644 |
} |
| 645 |
|
| 646 |
static __inline __m512 __attribute__ ((__always_inline__, __nodebug__))
|
| 647 |
_mm512_mask_blend_ps(__mmask16 __U, __m512 __A, __m512 __W) |
| 648 |
{
|
| 649 |
return (__m512) __builtin_ia32_blendmps_512_mask ((__v16sf) __A,
|
| 650 |
(__v16sf) __W, |
| 651 |
(__mmask16) __U); |
| 652 |
} |
| 653 |
|
| 654 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 655 |
_mm512_mask_blend_epi64(__mmask8 __U, __m512i __A, __m512i __W) |
| 656 |
{
|
| 657 |
return (__m512i) __builtin_ia32_blendmq_512_mask ((__v8di) __A,
|
| 658 |
(__v8di) __W, |
| 659 |
(__mmask8) __U); |
| 660 |
} |
| 661 |
|
| 662 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 663 |
_mm512_mask_blend_epi32(__mmask16 __U, __m512i __A, __m512i __W) |
| 664 |
{
|
| 665 |
return (__m512i) __builtin_ia32_blendmd_512_mask ((__v16si) __A,
|
| 666 |
(__v16si) __W, |
| 667 |
(__mmask16) __U); |
| 668 |
} |
| 669 |
|
| 670 |
/* Compare */
|
| 671 |
|
| 672 |
static __inline __mmask16 __attribute__ ((__always_inline__, __nodebug__))
|
| 673 |
_mm512_cmp_ps_mask(__m512 a, __m512 b, const int p) |
| 674 |
{
|
| 675 |
return (__mmask16) __builtin_ia32_cmpps512_mask ((__v16sf) a,
|
| 676 |
(__v16sf) b, p, (__mmask16) -1,
|
| 677 |
_MM_FROUND_CUR_DIRECTION); |
| 678 |
} |
| 679 |
|
| 680 |
static __inline __mmask8 __attribute__ ((__always_inline__, __nodebug__))
|
| 681 |
_mm512_cmp_pd_mask(__m512d __X, __m512d __Y, const int __P) |
| 682 |
{
|
| 683 |
return (__mmask8) __builtin_ia32_cmppd512_mask ((__v8df) __X,
|
| 684 |
(__v8df) __Y, __P, |
| 685 |
(__mmask8) -1,
|
| 686 |
_MM_FROUND_CUR_DIRECTION); |
| 687 |
} |
| 688 |
|
| 689 |
/* Conversion */
|
| 690 |
|
| 691 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 692 |
_mm512_cvttps_epu32(__m512 __A) |
| 693 |
{
|
| 694 |
return (__m512i) __builtin_ia32_cvttps2udq512_mask ((__v16sf) __A,
|
| 695 |
(__v16si) |
| 696 |
_mm512_setzero_si512 (), |
| 697 |
(__mmask16) -1,
|
| 698 |
_MM_FROUND_CUR_DIRECTION); |
| 699 |
} |
| 700 |
|
| 701 |
static __inline __m512 __attribute__ (( __always_inline__, __nodebug__))
|
| 702 |
_mm512_cvt_roundepi32_ps(__m512i __A, const int __R) |
| 703 |
{
|
| 704 |
return (__m512) __builtin_ia32_cvtdq2ps512_mask ((__v16si) __A,
|
| 705 |
(__v16sf) |
| 706 |
_mm512_setzero_ps (), |
| 707 |
(__mmask16) -1,
|
| 708 |
__R); |
| 709 |
} |
| 710 |
|
| 711 |
static __inline __m512 __attribute__ (( __always_inline__, __nodebug__))
|
| 712 |
_mm512_cvt_roundepu32_ps(__m512i __A, const int __R) |
| 713 |
{
|
| 714 |
return (__m512) __builtin_ia32_cvtudq2ps512_mask ((__v16si) __A,
|
| 715 |
(__v16sf) |
| 716 |
_mm512_setzero_ps (), |
| 717 |
(__mmask16) -1,
|
| 718 |
__R); |
| 719 |
} |
| 720 |
|
| 721 |
static __inline __m512d __attribute__ (( __always_inline__, __nodebug__))
|
| 722 |
_mm512_cvtepi32_pd(__m256i __A) |
| 723 |
{
|
| 724 |
return (__m512d) __builtin_ia32_cvtdq2pd512_mask ((__v8si) __A,
|
| 725 |
(__v8df) |
| 726 |
_mm512_setzero_pd (), |
| 727 |
(__mmask8) -1);
|
| 728 |
} |
| 729 |
|
| 730 |
static __inline __m512d __attribute__ (( __always_inline__, __nodebug__))
|
| 731 |
_mm512_cvtepu32_pd(__m256i __A) |
| 732 |
{
|
| 733 |
return (__m512d) __builtin_ia32_cvtudq2pd512_mask ((__v8si) __A,
|
| 734 |
(__v8df) |
| 735 |
_mm512_setzero_pd (), |
| 736 |
(__mmask8) -1);
|
| 737 |
} |
| 738 |
static __inline __m256 __attribute__ (( __always_inline__, __nodebug__))
|
| 739 |
_mm512_cvt_roundpd_ps(__m512d __A, const int __R) |
| 740 |
{
|
| 741 |
return (__m256) __builtin_ia32_cvtpd2ps512_mask ((__v8df) __A,
|
| 742 |
(__v8sf) |
| 743 |
_mm256_setzero_ps (), |
| 744 |
(__mmask8) -1,
|
| 745 |
__R); |
| 746 |
} |
| 747 |
|
| 748 |
static __inline __m256i __attribute__ ((__always_inline__, __nodebug__))
|
| 749 |
_mm512_cvtps_ph(__m512 __A, const int __I) |
| 750 |
{
|
| 751 |
return (__m256i) __builtin_ia32_vcvtps2ph512_mask ((__v16sf) __A,
|
| 752 |
__I, |
| 753 |
(__v16hi) |
| 754 |
_mm256_setzero_si256 (), |
| 755 |
-1);
|
| 756 |
} |
| 757 |
|
| 758 |
static __inline __m512 __attribute__ ((__always_inline__, __nodebug__))
|
| 759 |
_mm512_cvtph_ps(__m256i __A) |
| 760 |
{
|
| 761 |
return (__m512) __builtin_ia32_vcvtph2ps512_mask ((__v16hi) __A,
|
| 762 |
(__v16sf) |
| 763 |
_mm512_setzero_ps (), |
| 764 |
(__mmask16) -1,
|
| 765 |
_MM_FROUND_CUR_DIRECTION); |
| 766 |
} |
| 767 |
|
| 768 |
static __inline __m512i __attribute__((__always_inline__, __nodebug__))
|
| 769 |
_mm512_cvttps_epi32(__m512 a) |
| 770 |
{
|
| 771 |
return (__m512i)
|
| 772 |
__builtin_ia32_cvttps2dq512_mask((__v16sf) a, |
| 773 |
(__v16si) _mm512_setzero_si512 (), |
| 774 |
(__mmask16) -1, _MM_FROUND_CUR_DIRECTION);
|
| 775 |
} |
| 776 |
|
| 777 |
static __inline __m256i __attribute__((__always_inline__, __nodebug__))
|
| 778 |
_mm512_cvttpd_epi32(__m512d a) |
| 779 |
{
|
| 780 |
return (__m256i)__builtin_ia32_cvttpd2dq512_mask((__v8df) a,
|
| 781 |
(__v8si)_mm256_setzero_si256(), |
| 782 |
(__mmask8) -1,
|
| 783 |
_MM_FROUND_CUR_DIRECTION); |
| 784 |
} |
| 785 |
|
| 786 |
static __inline __m256i __attribute__ ((__always_inline__, __nodebug__))
|
| 787 |
_mm512_cvtt_roundpd_epi32(__m512d __A, const int __R) |
| 788 |
{
|
| 789 |
return (__m256i) __builtin_ia32_cvttpd2dq512_mask ((__v8df) __A,
|
| 790 |
(__v8si) |
| 791 |
_mm256_setzero_si256 (), |
| 792 |
(__mmask8) -1,
|
| 793 |
__R); |
| 794 |
} |
| 795 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 796 |
_mm512_cvtt_roundps_epi32(__m512 __A, const int __R) |
| 797 |
{
|
| 798 |
return (__m512i) __builtin_ia32_cvttps2dq512_mask ((__v16sf) __A,
|
| 799 |
(__v16si) |
| 800 |
_mm512_setzero_si512 (), |
| 801 |
(__mmask16) -1,
|
| 802 |
__R); |
| 803 |
} |
| 804 |
|
| 805 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 806 |
_mm512_cvt_roundps_epi32(__m512 __A, const int __R) |
| 807 |
{
|
| 808 |
return (__m512i) __builtin_ia32_cvtps2dq512_mask ((__v16sf) __A,
|
| 809 |
(__v16si) |
| 810 |
_mm512_setzero_si512 (), |
| 811 |
(__mmask16) -1,
|
| 812 |
__R); |
| 813 |
} |
| 814 |
static __inline __m256i __attribute__ ((__always_inline__, __nodebug__))
|
| 815 |
_mm512_cvt_roundpd_epi32(__m512d __A, const int __R) |
| 816 |
{
|
| 817 |
return (__m256i) __builtin_ia32_cvtpd2dq512_mask ((__v8df) __A,
|
| 818 |
(__v8si) |
| 819 |
_mm256_setzero_si256 (), |
| 820 |
(__mmask8) -1,
|
| 821 |
__R); |
| 822 |
} |
| 823 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 824 |
_mm512_cvt_roundps_epu32(__m512 __A, const int __R) |
| 825 |
{
|
| 826 |
return (__m512i) __builtin_ia32_cvtps2udq512_mask ((__v16sf) __A,
|
| 827 |
(__v16si) |
| 828 |
_mm512_setzero_si512 (), |
| 829 |
(__mmask16) -1,
|
| 830 |
__R); |
| 831 |
} |
| 832 |
static __inline __m256i __attribute__ ((__always_inline__, __nodebug__))
|
| 833 |
_mm512_cvt_roundpd_epu32(__m512d __A, const int __R) |
| 834 |
{
|
| 835 |
return (__m256i) __builtin_ia32_cvtpd2udq512_mask ((__v8df) __A,
|
| 836 |
(__v8si) |
| 837 |
_mm256_setzero_si256 (), |
| 838 |
(__mmask8) -1,
|
| 839 |
__R); |
| 840 |
} |
| 841 |
|
| 842 |
/* Unpack and Interleave */
|
| 843 |
static __inline __m512d __attribute__((__always_inline__, __nodebug__))
|
| 844 |
_mm512_unpackhi_pd(__m512d __a, __m512d __b) |
| 845 |
{
|
| 846 |
return __builtin_shufflevector(__a, __b, 1, 9, 1+2, 9+2, 1+4, 9+4, 1+6, 9+6); |
| 847 |
} |
| 848 |
|
| 849 |
static __inline __m512d __attribute__((__always_inline__, __nodebug__))
|
| 850 |
_mm512_unpacklo_pd(__m512d __a, __m512d __b) |
| 851 |
{
|
| 852 |
return __builtin_shufflevector(__a, __b, 0, 8, 0+2, 8+2, 0+4, 8+4, 0+6, 8+6); |
| 853 |
} |
| 854 |
|
| 855 |
static __inline __m512 __attribute__((__always_inline__, __nodebug__))
|
| 856 |
_mm512_unpackhi_ps(__m512 __a, __m512 __b) |
| 857 |
{
|
| 858 |
return __builtin_shufflevector(__a, __b,
|
| 859 |
2, 18, 3, 19, |
| 860 |
2+4, 18+4, 3+4, 19+4, |
| 861 |
2+8, 18+8, 3+8, 19+8, |
| 862 |
2+12, 18+12, 3+12, 19+12); |
| 863 |
} |
| 864 |
|
| 865 |
static __inline __m512 __attribute__((__always_inline__, __nodebug__))
|
| 866 |
_mm512_unpacklo_ps(__m512 __a, __m512 __b) |
| 867 |
{
|
| 868 |
return __builtin_shufflevector(__a, __b,
|
| 869 |
0, 16, 1, 17, |
| 870 |
0+4, 16+4, 1+4, 17+4, |
| 871 |
0+8, 16+8, 1+8, 17+8, |
| 872 |
0+12, 16+12, 1+12, 17+12); |
| 873 |
} |
| 874 |
|
| 875 |
/* Bit Test */
|
| 876 |
|
| 877 |
static __inline __mmask16 __attribute__ ((__always_inline__, __nodebug__))
|
| 878 |
_mm512_test_epi32_mask(__m512i __A, __m512i __B) |
| 879 |
{
|
| 880 |
return (__mmask16) __builtin_ia32_ptestmd512 ((__v16si) __A,
|
| 881 |
(__v16si) __B, |
| 882 |
(__mmask16) -1);
|
| 883 |
} |
| 884 |
|
| 885 |
static __inline __mmask8 __attribute__ ((__always_inline__, __nodebug__))
|
| 886 |
_mm512_test_epi64_mask(__m512i __A, __m512i __B) |
| 887 |
{
|
| 888 |
return (__mmask8) __builtin_ia32_ptestmq512 ((__v8di) __A,
|
| 889 |
(__v8di) __B, |
| 890 |
(__mmask8) -1);
|
| 891 |
} |
| 892 |
|
| 893 |
/* SIMD load ops */
|
| 894 |
|
| 895 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 896 |
_mm512_maskz_loadu_epi32(__mmask16 __U, void const *__P) |
| 897 |
{
|
| 898 |
return (__m512i) __builtin_ia32_loaddqusi512_mask ((const __v16si *)__P, |
| 899 |
(__v16si) |
| 900 |
_mm512_setzero_si512 (), |
| 901 |
(__mmask16) __U); |
| 902 |
} |
| 903 |
|
| 904 |
static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
|
| 905 |
_mm512_maskz_loadu_epi64(__mmask8 __U, void const *__P) |
| 906 |
{
|
| 907 |
return (__m512i) __builtin_ia32_loaddqudi512_mask ((const __v8di *)__P, |
| 908 |
(__v8di) |
| 909 |
_mm512_setzero_si512 (), |
| 910 |
(__mmask8) __U); |
| 911 |
} |
| 912 |
|
| 913 |
static __inline __m512 __attribute__ ((__always_inline__, __nodebug__))
|
| 914 |
_mm512_maskz_loadu_ps(__mmask16 __U, void const *__P) |
| 915 |
{
|
| 916 |
return (__m512) __builtin_ia32_loadups512_mask ((const __v16sf *)__P, |
| 917 |
(__v16sf) |
| 918 |
_mm512_setzero_ps (), |
| 919 |
(__mmask16) __U); |
| 920 |
} |
| 921 |
|
| 922 |
static __inline __m512d __attribute__ ((__always_inline__, __nodebug__))
|
| 923 |
_mm512_maskz_loadu_pd(__mmask8 __U, void const *__P) |
| 924 |
{
|
| 925 |
return (__m512d) __builtin_ia32_loadupd512_mask ((const __v8df *)__P, |
| 926 |
(__v8df) |
| 927 |
_mm512_setzero_pd (), |
| 928 |
(__mmask8) __U); |
| 929 |
} |
| 930 |
|
| 931 |
static __inline __m512d __attribute__((__always_inline__, __nodebug__))
|
| 932 |
_mm512_loadu_pd(double const *__p) |
| 933 |
{
|
| 934 |
struct __loadu_pd {
|
| 935 |
__m512d __v; |
| 936 |
} __attribute__((packed, may_alias)); |
| 937 |
return ((struct __loadu_pd*)__p)->__v; |
| 938 |
} |
| 939 |
|
| 940 |
static __inline __m512 __attribute__((__always_inline__, __nodebug__))
|
| 941 |
_mm512_loadu_ps(float const *__p) |
| 942 |
{
|
| 943 |
struct __loadu_ps {
|
| 944 |
__m512 __v; |
| 945 |
} __attribute__((packed, may_alias)); |
| 946 |
return ((struct __loadu_ps*)__p)->__v; |
| 947 |
} |
| 948 |
|
| 949 |
/* SIMD store ops */
|
| 950 |
|
| 951 |
static __inline void __attribute__ ((__always_inline__, __nodebug__)) |
| 952 |
_mm512_mask_storeu_epi64(void *__P, __mmask8 __U, __m512i __A)
|
| 953 |
{
|
| 954 |
__builtin_ia32_storedqudi512_mask ((__v8di *)__P, (__v8di) __A, |
| 955 |
(__mmask8) __U); |
| 956 |
} |
| 957 |
|
| 958 |
static __inline void __attribute__ ((__always_inline__, __nodebug__)) |
| 959 |
_mm512_mask_storeu_epi32(void *__P, __mmask16 __U, __m512i __A)
|
| 960 |
{
|
| 961 |
__builtin_ia32_storedqusi512_mask ((__v16si *)__P, (__v16si) __A, |
| 962 |
(__mmask16) __U); |
| 963 |
} |
| 964 |
|
| 965 |
static __inline void __attribute__ ((__always_inline__, __nodebug__)) |
| 966 |
_mm512_mask_storeu_pd(void *__P, __mmask8 __U, __m512d __A)
|
| 967 |
{
|
| 968 |
__builtin_ia32_storeupd512_mask ((__v8df *)__P, (__v8df) __A, (__mmask8) __U); |
| 969 |
} |
| 970 |
|
| 971 |
static __inline void __attribute__ ((__always_inline__, __nodebug__)) |
| 972 |
_mm512_storeu_pd(void *__P, __m512d __A)
|
| 973 |
{
|
| 974 |
__builtin_ia32_storeupd512_mask((__v8df *)__P, (__v8df)__A, (__mmask8)-1);
|
| 975 |
} |
| 976 |
|
| 977 |
static __inline void __attribute__ ((__always_inline__, __nodebug__)) |
| 978 |
_mm512_mask_storeu_ps(void *__P, __mmask16 __U, __m512 __A)
|
| 979 |
{
|
| 980 |
__builtin_ia32_storeups512_mask ((__v16sf *)__P, (__v16sf) __A, |
| 981 |
(__mmask16) __U); |
| 982 |
} |
| 983 |
|
| 984 |
static __inline void __attribute__ ((__always_inline__, __nodebug__)) |
| 985 |
_mm512_storeu_ps(void *__P, __m512 __A)
|
| 986 |
{
|
| 987 |
__builtin_ia32_storeups512_mask((__v16sf *)__P, (__v16sf)__A, (__mmask16)-1);
|
| 988 |
} |
| 989 |
|
| 990 |
static __inline void __attribute__ ((__always_inline__, __nodebug__)) |
| 991 |
_mm512_store_ps(void *__P, __m512 __A)
|
| 992 |
{
|
| 993 |
*(__m512*)__P = __A; |
| 994 |
} |
| 995 |
|
| 996 |
static __inline void __attribute__ ((__always_inline__, __nodebug__)) |
| 997 |
_mm512_store_pd(void *__P, __m512d __A)
|
| 998 |
{
|
| 999 |
*(__m512d*)__P = __A; |
| 1000 |
} |
| 1001 |
|
| 1002 |
/* Mask ops */
|
| 1003 |
|
| 1004 |
static __inline __mmask16 __attribute__ ((__always_inline__, __nodebug__))
|
| 1005 |
_mm512_knot(__mmask16 __M) |
| 1006 |
{
|
| 1007 |
return __builtin_ia32_knothi(__M);
|
| 1008 |
} |
| 1009 |
|
| 1010 |
/* Integer compare */
|
| 1011 |
|
| 1012 |
static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
|
| 1013 |
_mm512_cmpeq_epi32_mask(__m512i __a, __m512i __b) {
|
| 1014 |
return (__mmask16)__builtin_ia32_pcmpeqd512_mask((__v16si)__a, (__v16si)__b,
|
| 1015 |
(__mmask16)-1);
|
| 1016 |
} |
| 1017 |
|
| 1018 |
static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
|
| 1019 |
_mm512_mask_cmpeq_epi32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
|
| 1020 |
return (__mmask16)__builtin_ia32_pcmpeqd512_mask((__v16si)__a, (__v16si)__b,
|
| 1021 |
__u); |
| 1022 |
} |
| 1023 |
|
| 1024 |
static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
|
| 1025 |
_mm512_mask_cmpeq_epi64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
|
| 1026 |
return (__mmask8)__builtin_ia32_pcmpeqq512_mask((__v8di)__a, (__v8di)__b,
|
| 1027 |
__u); |
| 1028 |
} |
| 1029 |
|
| 1030 |
static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
|
| 1031 |
_mm512_cmpeq_epi64_mask(__m512i __a, __m512i __b) {
|
| 1032 |
return (__mmask8)__builtin_ia32_pcmpeqq512_mask((__v8di)__a, (__v8di)__b,
|
| 1033 |
(__mmask8)-1);
|
| 1034 |
} |
| 1035 |
|
| 1036 |
#endif // __AVX512FINTRIN_H |