1 /* Copyright (C) 2002-2018 Free Software Foundation, Inc. 2 3 This file is part of GCC. 4 5 GCC is free software; you can redistribute it and/or modify 6 it under the terms of the GNU General Public License as published by 7 the Free Software Foundation; either version 3, or (at your option) 8 any later version. 9 10 GCC is distributed in the hope that it will be useful, 11 but WITHOUT ANY WARRANTY; without even the implied warranty of 12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 GNU General Public License for more details. 14 15 Under Section 7 of GPL version 3, you are granted additional 16 permissions described in the GCC Runtime Library Exception, version 17 3.1, as published by the Free Software Foundation. 18 19 You should have received a copy of the GNU General Public License and 20 a copy of the GCC Runtime Library Exception along with this program; 21 see the files COPYING3 and COPYING.RUNTIME respectively. If not, see 22 <http://www.gnu.org/licenses/>. */ 23 24 /* Implemented from the specification included in the Intel C++ Compiler 25 User Guide and Reference, version 9.0. */ 26 27 #ifndef _MMINTRIN_H_INCLUDED 28 #define _MMINTRIN_H_INCLUDED 29 30 #if defined __x86_64__ && !defined __SSE__ || !defined __MMX__ 31 #pragma GCC push_options 32 #ifdef __x86_64__ 33 #pragma GCC target("sse,mmx") 34 #else 35 #pragma GCC target("mmx") 36 #endif 37 #define __DISABLE_MMX__ 38 #endif /* __MMX__ */ 39 40 /* The Intel API is flexible enough that we must allow aliasing with other 41 vector types, and their scalar components. */ 42 typedef int __m64 __attribute__ ((__vector_size__ (8), __may_alias__)); 43 44 /* Unaligned version of the same type */ 45 typedef int __m64_u __attribute__ ((__vector_size__ (8), __may_alias__, __aligned__ (1))); 46 47 /* Internal data types for implementing the intrinsics. */ 48 typedef int __v2si __attribute__ ((__vector_size__ (8))); 49 typedef short __v4hi __attribute__ ((__vector_size__ (8))); 50 typedef char __v8qi __attribute__ ((__vector_size__ (8))); 51 typedef long long __v1di __attribute__ ((__vector_size__ (8))); 52 typedef float __v2sf __attribute__ ((__vector_size__ (8))); 53 54 /* Empty the multimedia state. */ 55 extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 56 _mm_empty (void) 57 { 58 __builtin_ia32_emms (); 59 } 60 61 extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 62 _m_empty (void) 63 { 64 _mm_empty (); 65 } 66 67 /* Convert I to a __m64 object. The integer is zero-extended to 64-bits. */ 68 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 69 _mm_cvtsi32_si64 (int __i) 70 { 71 return (__m64) __builtin_ia32_vec_init_v2si (__i, 0); 72 } 73 74 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 75 _m_from_int (int __i) 76 { 77 return _mm_cvtsi32_si64 (__i); 78 } 79 80 #ifdef __x86_64__ 81 /* Convert I to a __m64 object. */ 82 83 /* Intel intrinsic. */ 84 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 85 _m_from_int64 (long long __i) 86 { 87 return (__m64) __i; 88 } 89 90 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 91 _mm_cvtsi64_m64 (long long __i) 92 { 93 return (__m64) __i; 94 } 95 96 /* Microsoft intrinsic. */ 97 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 98 _mm_cvtsi64x_si64 (long long __i) 99 { 100 return (__m64) __i; 101 } 102 103 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 104 _mm_set_pi64x (long long __i) 105 { 106 return (__m64) __i; 107 } 108 #endif 109 110 /* Convert the lower 32 bits of the __m64 object into an integer. */ 111 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 112 _mm_cvtsi64_si32 (__m64 __i) 113 { 114 return __builtin_ia32_vec_ext_v2si ((__v2si)__i, 0); 115 } 116 117 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 118 _m_to_int (__m64 __i) 119 { 120 return _mm_cvtsi64_si32 (__i); 121 } 122 123 #ifdef __x86_64__ 124 /* Convert the __m64 object to a 64bit integer. */ 125 126 /* Intel intrinsic. */ 127 extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 128 _m_to_int64 (__m64 __i) 129 { 130 return (long long)__i; 131 } 132 133 extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 134 _mm_cvtm64_si64 (__m64 __i) 135 { 136 return (long long)__i; 137 } 138 139 /* Microsoft intrinsic. */ 140 extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 141 _mm_cvtsi64_si64x (__m64 __i) 142 { 143 return (long long)__i; 144 } 145 #endif 146 147 /* Pack the four 16-bit values from M1 into the lower four 8-bit values of 148 the result, and the four 16-bit values from M2 into the upper four 8-bit 149 values of the result, all with signed saturation. */ 150 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 151 _mm_packs_pi16 (__m64 __m1, __m64 __m2) 152 { 153 return (__m64) __builtin_ia32_packsswb ((__v4hi)__m1, (__v4hi)__m2); 154 } 155 156 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 157 _m_packsswb (__m64 __m1, __m64 __m2) 158 { 159 return _mm_packs_pi16 (__m1, __m2); 160 } 161 162 /* Pack the two 32-bit values from M1 in to the lower two 16-bit values of 163 the result, and the two 32-bit values from M2 into the upper two 16-bit 164 values of the result, all with signed saturation. */ 165 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 166 _mm_packs_pi32 (__m64 __m1, __m64 __m2) 167 { 168 return (__m64) __builtin_ia32_packssdw ((__v2si)__m1, (__v2si)__m2); 169 } 170 171 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 172 _m_packssdw (__m64 __m1, __m64 __m2) 173 { 174 return _mm_packs_pi32 (__m1, __m2); 175 } 176 177 /* Pack the four 16-bit values from M1 into the lower four 8-bit values of 178 the result, and the four 16-bit values from M2 into the upper four 8-bit 179 values of the result, all with unsigned saturation. */ 180 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 181 _mm_packs_pu16 (__m64 __m1, __m64 __m2) 182 { 183 return (__m64) __builtin_ia32_packuswb ((__v4hi)__m1, (__v4hi)__m2); 184 } 185 186 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 187 _m_packuswb (__m64 __m1, __m64 __m2) 188 { 189 return _mm_packs_pu16 (__m1, __m2); 190 } 191 192 /* Interleave the four 8-bit values from the high half of M1 with the four 193 8-bit values from the high half of M2. */ 194 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 195 _mm_unpackhi_pi8 (__m64 __m1, __m64 __m2) 196 { 197 return (__m64) __builtin_ia32_punpckhbw ((__v8qi)__m1, (__v8qi)__m2); 198 } 199 200 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 201 _m_punpckhbw (__m64 __m1, __m64 __m2) 202 { 203 return _mm_unpackhi_pi8 (__m1, __m2); 204 } 205 206 /* Interleave the two 16-bit values from the high half of M1 with the two 207 16-bit values from the high half of M2. */ 208 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 209 _mm_unpackhi_pi16 (__m64 __m1, __m64 __m2) 210 { 211 return (__m64) __builtin_ia32_punpckhwd ((__v4hi)__m1, (__v4hi)__m2); 212 } 213 214 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 215 _m_punpckhwd (__m64 __m1, __m64 __m2) 216 { 217 return _mm_unpackhi_pi16 (__m1, __m2); 218 } 219 220 /* Interleave the 32-bit value from the high half of M1 with the 32-bit 221 value from the high half of M2. */ 222 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 223 _mm_unpackhi_pi32 (__m64 __m1, __m64 __m2) 224 { 225 return (__m64) __builtin_ia32_punpckhdq ((__v2si)__m1, (__v2si)__m2); 226 } 227 228 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 229 _m_punpckhdq (__m64 __m1, __m64 __m2) 230 { 231 return _mm_unpackhi_pi32 (__m1, __m2); 232 } 233 234 /* Interleave the four 8-bit values from the low half of M1 with the four 235 8-bit values from the low half of M2. */ 236 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 237 _mm_unpacklo_pi8 (__m64 __m1, __m64 __m2) 238 { 239 return (__m64) __builtin_ia32_punpcklbw ((__v8qi)__m1, (__v8qi)__m2); 240 } 241 242 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 243 _m_punpcklbw (__m64 __m1, __m64 __m2) 244 { 245 return _mm_unpacklo_pi8 (__m1, __m2); 246 } 247 248 /* Interleave the two 16-bit values from the low half of M1 with the two 249 16-bit values from the low half of M2. */ 250 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 251 _mm_unpacklo_pi16 (__m64 __m1, __m64 __m2) 252 { 253 return (__m64) __builtin_ia32_punpcklwd ((__v4hi)__m1, (__v4hi)__m2); 254 } 255 256 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 257 _m_punpcklwd (__m64 __m1, __m64 __m2) 258 { 259 return _mm_unpacklo_pi16 (__m1, __m2); 260 } 261 262 /* Interleave the 32-bit value from the low half of M1 with the 32-bit 263 value from the low half of M2. */ 264 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 265 _mm_unpacklo_pi32 (__m64 __m1, __m64 __m2) 266 { 267 return (__m64) __builtin_ia32_punpckldq ((__v2si)__m1, (__v2si)__m2); 268 } 269 270 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 271 _m_punpckldq (__m64 __m1, __m64 __m2) 272 { 273 return _mm_unpacklo_pi32 (__m1, __m2); 274 } 275 276 /* Add the 8-bit values in M1 to the 8-bit values in M2. */ 277 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 278 _mm_add_pi8 (__m64 __m1, __m64 __m2) 279 { 280 return (__m64) __builtin_ia32_paddb ((__v8qi)__m1, (__v8qi)__m2); 281 } 282 283 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 284 _m_paddb (__m64 __m1, __m64 __m2) 285 { 286 return _mm_add_pi8 (__m1, __m2); 287 } 288 289 /* Add the 16-bit values in M1 to the 16-bit values in M2. */ 290 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 291 _mm_add_pi16 (__m64 __m1, __m64 __m2) 292 { 293 return (__m64) __builtin_ia32_paddw ((__v4hi)__m1, (__v4hi)__m2); 294 } 295 296 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 297 _m_paddw (__m64 __m1, __m64 __m2) 298 { 299 return _mm_add_pi16 (__m1, __m2); 300 } 301 302 /* Add the 32-bit values in M1 to the 32-bit values in M2. */ 303 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 304 _mm_add_pi32 (__m64 __m1, __m64 __m2) 305 { 306 return (__m64) __builtin_ia32_paddd ((__v2si)__m1, (__v2si)__m2); 307 } 308 309 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 310 _m_paddd (__m64 __m1, __m64 __m2) 311 { 312 return _mm_add_pi32 (__m1, __m2); 313 } 314 315 /* Add the 64-bit values in M1 to the 64-bit values in M2. */ 316 #ifndef __SSE2__ 317 #pragma GCC push_options 318 #pragma GCC target("sse2,mmx") 319 #define __DISABLE_SSE2__ 320 #endif /* __SSE2__ */ 321 322 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 323 _mm_add_si64 (__m64 __m1, __m64 __m2) 324 { 325 return (__m64) __builtin_ia32_paddq ((__v1di)__m1, (__v1di)__m2); 326 } 327 #ifdef __DISABLE_SSE2__ 328 #undef __DISABLE_SSE2__ 329 #pragma GCC pop_options 330 #endif /* __DISABLE_SSE2__ */ 331 332 /* Add the 8-bit values in M1 to the 8-bit values in M2 using signed 333 saturated arithmetic. */ 334 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 335 _mm_adds_pi8 (__m64 __m1, __m64 __m2) 336 { 337 return (__m64) __builtin_ia32_paddsb ((__v8qi)__m1, (__v8qi)__m2); 338 } 339 340 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 341 _m_paddsb (__m64 __m1, __m64 __m2) 342 { 343 return _mm_adds_pi8 (__m1, __m2); 344 } 345 346 /* Add the 16-bit values in M1 to the 16-bit values in M2 using signed 347 saturated arithmetic. */ 348 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 349 _mm_adds_pi16 (__m64 __m1, __m64 __m2) 350 { 351 return (__m64) __builtin_ia32_paddsw ((__v4hi)__m1, (__v4hi)__m2); 352 } 353 354 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 355 _m_paddsw (__m64 __m1, __m64 __m2) 356 { 357 return _mm_adds_pi16 (__m1, __m2); 358 } 359 360 /* Add the 8-bit values in M1 to the 8-bit values in M2 using unsigned 361 saturated arithmetic. */ 362 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 363 _mm_adds_pu8 (__m64 __m1, __m64 __m2) 364 { 365 return (__m64) __builtin_ia32_paddusb ((__v8qi)__m1, (__v8qi)__m2); 366 } 367 368 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 369 _m_paddusb (__m64 __m1, __m64 __m2) 370 { 371 return _mm_adds_pu8 (__m1, __m2); 372 } 373 374 /* Add the 16-bit values in M1 to the 16-bit values in M2 using unsigned 375 saturated arithmetic. */ 376 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 377 _mm_adds_pu16 (__m64 __m1, __m64 __m2) 378 { 379 return (__m64) __builtin_ia32_paddusw ((__v4hi)__m1, (__v4hi)__m2); 380 } 381 382 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 383 _m_paddusw (__m64 __m1, __m64 __m2) 384 { 385 return _mm_adds_pu16 (__m1, __m2); 386 } 387 388 /* Subtract the 8-bit values in M2 from the 8-bit values in M1. */ 389 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 390 _mm_sub_pi8 (__m64 __m1, __m64 __m2) 391 { 392 return (__m64) __builtin_ia32_psubb ((__v8qi)__m1, (__v8qi)__m2); 393 } 394 395 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 396 _m_psubb (__m64 __m1, __m64 __m2) 397 { 398 return _mm_sub_pi8 (__m1, __m2); 399 } 400 401 /* Subtract the 16-bit values in M2 from the 16-bit values in M1. */ 402 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 403 _mm_sub_pi16 (__m64 __m1, __m64 __m2) 404 { 405 return (__m64) __builtin_ia32_psubw ((__v4hi)__m1, (__v4hi)__m2); 406 } 407 408 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 409 _m_psubw (__m64 __m1, __m64 __m2) 410 { 411 return _mm_sub_pi16 (__m1, __m2); 412 } 413 414 /* Subtract the 32-bit values in M2 from the 32-bit values in M1. */ 415 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 416 _mm_sub_pi32 (__m64 __m1, __m64 __m2) 417 { 418 return (__m64) __builtin_ia32_psubd ((__v2si)__m1, (__v2si)__m2); 419 } 420 421 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 422 _m_psubd (__m64 __m1, __m64 __m2) 423 { 424 return _mm_sub_pi32 (__m1, __m2); 425 } 426 427 /* Add the 64-bit values in M1 to the 64-bit values in M2. */ 428 #ifndef __SSE2__ 429 #pragma GCC push_options 430 #pragma GCC target("sse2,mmx") 431 #define __DISABLE_SSE2__ 432 #endif /* __SSE2__ */ 433 434 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 435 _mm_sub_si64 (__m64 __m1, __m64 __m2) 436 { 437 return (__m64) __builtin_ia32_psubq ((__v1di)__m1, (__v1di)__m2); 438 } 439 #ifdef __DISABLE_SSE2__ 440 #undef __DISABLE_SSE2__ 441 #pragma GCC pop_options 442 #endif /* __DISABLE_SSE2__ */ 443 444 /* Subtract the 8-bit values in M2 from the 8-bit values in M1 using signed 445 saturating arithmetic. */ 446 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 447 _mm_subs_pi8 (__m64 __m1, __m64 __m2) 448 { 449 return (__m64) __builtin_ia32_psubsb ((__v8qi)__m1, (__v8qi)__m2); 450 } 451 452 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 453 _m_psubsb (__m64 __m1, __m64 __m2) 454 { 455 return _mm_subs_pi8 (__m1, __m2); 456 } 457 458 /* Subtract the 16-bit values in M2 from the 16-bit values in M1 using 459 signed saturating arithmetic. */ 460 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 461 _mm_subs_pi16 (__m64 __m1, __m64 __m2) 462 { 463 return (__m64) __builtin_ia32_psubsw ((__v4hi)__m1, (__v4hi)__m2); 464 } 465 466 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 467 _m_psubsw (__m64 __m1, __m64 __m2) 468 { 469 return _mm_subs_pi16 (__m1, __m2); 470 } 471 472 /* Subtract the 8-bit values in M2 from the 8-bit values in M1 using 473 unsigned saturating arithmetic. */ 474 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 475 _mm_subs_pu8 (__m64 __m1, __m64 __m2) 476 { 477 return (__m64) __builtin_ia32_psubusb ((__v8qi)__m1, (__v8qi)__m2); 478 } 479 480 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 481 _m_psubusb (__m64 __m1, __m64 __m2) 482 { 483 return _mm_subs_pu8 (__m1, __m2); 484 } 485 486 /* Subtract the 16-bit values in M2 from the 16-bit values in M1 using 487 unsigned saturating arithmetic. */ 488 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 489 _mm_subs_pu16 (__m64 __m1, __m64 __m2) 490 { 491 return (__m64) __builtin_ia32_psubusw ((__v4hi)__m1, (__v4hi)__m2); 492 } 493 494 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 495 _m_psubusw (__m64 __m1, __m64 __m2) 496 { 497 return _mm_subs_pu16 (__m1, __m2); 498 } 499 500 /* Multiply four 16-bit values in M1 by four 16-bit values in M2 producing 501 four 32-bit intermediate results, which are then summed by pairs to 502 produce two 32-bit results. */ 503 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 504 _mm_madd_pi16 (__m64 __m1, __m64 __m2) 505 { 506 return (__m64) __builtin_ia32_pmaddwd ((__v4hi)__m1, (__v4hi)__m2); 507 } 508 509 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 510 _m_pmaddwd (__m64 __m1, __m64 __m2) 511 { 512 return _mm_madd_pi16 (__m1, __m2); 513 } 514 515 /* Multiply four signed 16-bit values in M1 by four signed 16-bit values in 516 M2 and produce the high 16 bits of the 32-bit results. */ 517 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 518 _mm_mulhi_pi16 (__m64 __m1, __m64 __m2) 519 { 520 return (__m64) __builtin_ia32_pmulhw ((__v4hi)__m1, (__v4hi)__m2); 521 } 522 523 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 524 _m_pmulhw (__m64 __m1, __m64 __m2) 525 { 526 return _mm_mulhi_pi16 (__m1, __m2); 527 } 528 529 /* Multiply four 16-bit values in M1 by four 16-bit values in M2 and produce 530 the low 16 bits of the results. */ 531 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 532 _mm_mullo_pi16 (__m64 __m1, __m64 __m2) 533 { 534 return (__m64) __builtin_ia32_pmullw ((__v4hi)__m1, (__v4hi)__m2); 535 } 536 537 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 538 _m_pmullw (__m64 __m1, __m64 __m2) 539 { 540 return _mm_mullo_pi16 (__m1, __m2); 541 } 542 543 /* Shift four 16-bit values in M left by COUNT. */ 544 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 545 _mm_sll_pi16 (__m64 __m, __m64 __count) 546 { 547 return (__m64) __builtin_ia32_psllw ((__v4hi)__m, (__v4hi)__count); 548 } 549 550 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 551 _m_psllw (__m64 __m, __m64 __count) 552 { 553 return _mm_sll_pi16 (__m, __count); 554 } 555 556 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 557 _mm_slli_pi16 (__m64 __m, int __count) 558 { 559 return (__m64) __builtin_ia32_psllwi ((__v4hi)__m, __count); 560 } 561 562 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 563 _m_psllwi (__m64 __m, int __count) 564 { 565 return _mm_slli_pi16 (__m, __count); 566 } 567 568 /* Shift two 32-bit values in M left by COUNT. */ 569 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 570 _mm_sll_pi32 (__m64 __m, __m64 __count) 571 { 572 return (__m64) __builtin_ia32_pslld ((__v2si)__m, (__v2si)__count); 573 } 574 575 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 576 _m_pslld (__m64 __m, __m64 __count) 577 { 578 return _mm_sll_pi32 (__m, __count); 579 } 580 581 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 582 _mm_slli_pi32 (__m64 __m, int __count) 583 { 584 return (__m64) __builtin_ia32_pslldi ((__v2si)__m, __count); 585 } 586 587 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 588 _m_pslldi (__m64 __m, int __count) 589 { 590 return _mm_slli_pi32 (__m, __count); 591 } 592 593 /* Shift the 64-bit value in M left by COUNT. */ 594 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 595 _mm_sll_si64 (__m64 __m, __m64 __count) 596 { 597 return (__m64) __builtin_ia32_psllq ((__v1di)__m, (__v1di)__count); 598 } 599 600 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 601 _m_psllq (__m64 __m, __m64 __count) 602 { 603 return _mm_sll_si64 (__m, __count); 604 } 605 606 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 607 _mm_slli_si64 (__m64 __m, int __count) 608 { 609 return (__m64) __builtin_ia32_psllqi ((__v1di)__m, __count); 610 } 611 612 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 613 _m_psllqi (__m64 __m, int __count) 614 { 615 return _mm_slli_si64 (__m, __count); 616 } 617 618 /* Shift four 16-bit values in M right by COUNT; shift in the sign bit. */ 619 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 620 _mm_sra_pi16 (__m64 __m, __m64 __count) 621 { 622 return (__m64) __builtin_ia32_psraw ((__v4hi)__m, (__v4hi)__count); 623 } 624 625 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 626 _m_psraw (__m64 __m, __m64 __count) 627 { 628 return _mm_sra_pi16 (__m, __count); 629 } 630 631 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 632 _mm_srai_pi16 (__m64 __m, int __count) 633 { 634 return (__m64) __builtin_ia32_psrawi ((__v4hi)__m, __count); 635 } 636 637 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 638 _m_psrawi (__m64 __m, int __count) 639 { 640 return _mm_srai_pi16 (__m, __count); 641 } 642 643 /* Shift two 32-bit values in M right by COUNT; shift in the sign bit. */ 644 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 645 _mm_sra_pi32 (__m64 __m, __m64 __count) 646 { 647 return (__m64) __builtin_ia32_psrad ((__v2si)__m, (__v2si)__count); 648 } 649 650 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 651 _m_psrad (__m64 __m, __m64 __count) 652 { 653 return _mm_sra_pi32 (__m, __count); 654 } 655 656 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 657 _mm_srai_pi32 (__m64 __m, int __count) 658 { 659 return (__m64) __builtin_ia32_psradi ((__v2si)__m, __count); 660 } 661 662 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 663 _m_psradi (__m64 __m, int __count) 664 { 665 return _mm_srai_pi32 (__m, __count); 666 } 667 668 /* Shift four 16-bit values in M right by COUNT; shift in zeros. */ 669 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 670 _mm_srl_pi16 (__m64 __m, __m64 __count) 671 { 672 return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, (__v4hi)__count); 673 } 674 675 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 676 _m_psrlw (__m64 __m, __m64 __count) 677 { 678 return _mm_srl_pi16 (__m, __count); 679 } 680 681 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 682 _mm_srli_pi16 (__m64 __m, int __count) 683 { 684 return (__m64) __builtin_ia32_psrlwi ((__v4hi)__m, __count); 685 } 686 687 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 688 _m_psrlwi (__m64 __m, int __count) 689 { 690 return _mm_srli_pi16 (__m, __count); 691 } 692 693 /* Shift two 32-bit values in M right by COUNT; shift in zeros. */ 694 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 695 _mm_srl_pi32 (__m64 __m, __m64 __count) 696 { 697 return (__m64) __builtin_ia32_psrld ((__v2si)__m, (__v2si)__count); 698 } 699 700 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 701 _m_psrld (__m64 __m, __m64 __count) 702 { 703 return _mm_srl_pi32 (__m, __count); 704 } 705 706 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 707 _mm_srli_pi32 (__m64 __m, int __count) 708 { 709 return (__m64) __builtin_ia32_psrldi ((__v2si)__m, __count); 710 } 711 712 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 713 _m_psrldi (__m64 __m, int __count) 714 { 715 return _mm_srli_pi32 (__m, __count); 716 } 717 718 /* Shift the 64-bit value in M left by COUNT; shift in zeros. */ 719 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 720 _mm_srl_si64 (__m64 __m, __m64 __count) 721 { 722 return (__m64) __builtin_ia32_psrlq ((__v1di)__m, (__v1di)__count); 723 } 724 725 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 726 _m_psrlq (__m64 __m, __m64 __count) 727 { 728 return _mm_srl_si64 (__m, __count); 729 } 730 731 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 732 _mm_srli_si64 (__m64 __m, int __count) 733 { 734 return (__m64) __builtin_ia32_psrlqi ((__v1di)__m, __count); 735 } 736 737 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 738 _m_psrlqi (__m64 __m, int __count) 739 { 740 return _mm_srli_si64 (__m, __count); 741 } 742 743 /* Bit-wise AND the 64-bit values in M1 and M2. */ 744 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 745 _mm_and_si64 (__m64 __m1, __m64 __m2) 746 { 747 return __builtin_ia32_pand (__m1, __m2); 748 } 749 750 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 751 _m_pand (__m64 __m1, __m64 __m2) 752 { 753 return _mm_and_si64 (__m1, __m2); 754 } 755 756 /* Bit-wise complement the 64-bit value in M1 and bit-wise AND it with the 757 64-bit value in M2. */ 758 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 759 _mm_andnot_si64 (__m64 __m1, __m64 __m2) 760 { 761 return __builtin_ia32_pandn (__m1, __m2); 762 } 763 764 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 765 _m_pandn (__m64 __m1, __m64 __m2) 766 { 767 return _mm_andnot_si64 (__m1, __m2); 768 } 769 770 /* Bit-wise inclusive OR the 64-bit values in M1 and M2. */ 771 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 772 _mm_or_si64 (__m64 __m1, __m64 __m2) 773 { 774 return __builtin_ia32_por (__m1, __m2); 775 } 776 777 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 778 _m_por (__m64 __m1, __m64 __m2) 779 { 780 return _mm_or_si64 (__m1, __m2); 781 } 782 783 /* Bit-wise exclusive OR the 64-bit values in M1 and M2. */ 784 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 785 _mm_xor_si64 (__m64 __m1, __m64 __m2) 786 { 787 return __builtin_ia32_pxor (__m1, __m2); 788 } 789 790 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 791 _m_pxor (__m64 __m1, __m64 __m2) 792 { 793 return _mm_xor_si64 (__m1, __m2); 794 } 795 796 /* Compare eight 8-bit values. The result of the comparison is 0xFF if the 797 test is true and zero if false. */ 798 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 799 _mm_cmpeq_pi8 (__m64 __m1, __m64 __m2) 800 { 801 return (__m64) __builtin_ia32_pcmpeqb ((__v8qi)__m1, (__v8qi)__m2); 802 } 803 804 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 805 _m_pcmpeqb (__m64 __m1, __m64 __m2) 806 { 807 return _mm_cmpeq_pi8 (__m1, __m2); 808 } 809 810 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 811 _mm_cmpgt_pi8 (__m64 __m1, __m64 __m2) 812 { 813 return (__m64) __builtin_ia32_pcmpgtb ((__v8qi)__m1, (__v8qi)__m2); 814 } 815 816 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 817 _m_pcmpgtb (__m64 __m1, __m64 __m2) 818 { 819 return _mm_cmpgt_pi8 (__m1, __m2); 820 } 821 822 /* Compare four 16-bit values. The result of the comparison is 0xFFFF if 823 the test is true and zero if false. */ 824 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 825 _mm_cmpeq_pi16 (__m64 __m1, __m64 __m2) 826 { 827 return (__m64) __builtin_ia32_pcmpeqw ((__v4hi)__m1, (__v4hi)__m2); 828 } 829 830 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 831 _m_pcmpeqw (__m64 __m1, __m64 __m2) 832 { 833 return _mm_cmpeq_pi16 (__m1, __m2); 834 } 835 836 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 837 _mm_cmpgt_pi16 (__m64 __m1, __m64 __m2) 838 { 839 return (__m64) __builtin_ia32_pcmpgtw ((__v4hi)__m1, (__v4hi)__m2); 840 } 841 842 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 843 _m_pcmpgtw (__m64 __m1, __m64 __m2) 844 { 845 return _mm_cmpgt_pi16 (__m1, __m2); 846 } 847 848 /* Compare two 32-bit values. The result of the comparison is 0xFFFFFFFF if 849 the test is true and zero if false. */ 850 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 851 _mm_cmpeq_pi32 (__m64 __m1, __m64 __m2) 852 { 853 return (__m64) __builtin_ia32_pcmpeqd ((__v2si)__m1, (__v2si)__m2); 854 } 855 856 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 857 _m_pcmpeqd (__m64 __m1, __m64 __m2) 858 { 859 return _mm_cmpeq_pi32 (__m1, __m2); 860 } 861 862 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 863 _mm_cmpgt_pi32 (__m64 __m1, __m64 __m2) 864 { 865 return (__m64) __builtin_ia32_pcmpgtd ((__v2si)__m1, (__v2si)__m2); 866 } 867 868 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 869 _m_pcmpgtd (__m64 __m1, __m64 __m2) 870 { 871 return _mm_cmpgt_pi32 (__m1, __m2); 872 } 873 874 /* Creates a 64-bit zero. */ 875 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 876 _mm_setzero_si64 (void) 877 { 878 return (__m64)0LL; 879 } 880 881 /* Creates a vector of two 32-bit values; I0 is least significant. */ 882 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 883 _mm_set_pi32 (int __i1, int __i0) 884 { 885 return (__m64) __builtin_ia32_vec_init_v2si (__i0, __i1); 886 } 887 888 /* Creates a vector of four 16-bit values; W0 is least significant. */ 889 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 890 _mm_set_pi16 (short __w3, short __w2, short __w1, short __w0) 891 { 892 return (__m64) __builtin_ia32_vec_init_v4hi (__w0, __w1, __w2, __w3); 893 } 894 895 /* Creates a vector of eight 8-bit values; B0 is least significant. */ 896 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 897 _mm_set_pi8 (char __b7, char __b6, char __b5, char __b4, 898 char __b3, char __b2, char __b1, char __b0) 899 { 900 return (__m64) __builtin_ia32_vec_init_v8qi (__b0, __b1, __b2, __b3, 901 __b4, __b5, __b6, __b7); 902 } 903 904 /* Similar, but with the arguments in reverse order. */ 905 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 906 _mm_setr_pi32 (int __i0, int __i1) 907 { 908 return _mm_set_pi32 (__i1, __i0); 909 } 910 911 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 912 _mm_setr_pi16 (short __w0, short __w1, short __w2, short __w3) 913 { 914 return _mm_set_pi16 (__w3, __w2, __w1, __w0); 915 } 916 917 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 918 _mm_setr_pi8 (char __b0, char __b1, char __b2, char __b3, 919 char __b4, char __b5, char __b6, char __b7) 920 { 921 return _mm_set_pi8 (__b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0); 922 } 923 924 /* Creates a vector of two 32-bit values, both elements containing I. */ 925 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 926 _mm_set1_pi32 (int __i) 927 { 928 return _mm_set_pi32 (__i, __i); 929 } 930 931 /* Creates a vector of four 16-bit values, all elements containing W. */ 932 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 933 _mm_set1_pi16 (short __w) 934 { 935 return _mm_set_pi16 (__w, __w, __w, __w); 936 } 937 938 /* Creates a vector of eight 8-bit values, all elements containing B. */ 939 extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 940 _mm_set1_pi8 (char __b) 941 { 942 return _mm_set_pi8 (__b, __b, __b, __b, __b, __b, __b, __b); 943 } 944 #ifdef __DISABLE_MMX__ 945 #undef __DISABLE_MMX__ 946 #pragma GCC pop_options 947 #endif /* __DISABLE_MMX__ */ 948 949 #endif /* _MMINTRIN_H_INCLUDED */ 950