1/* Copyright (C) 2002-2015 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#ifndef __MMX__ 31#pragma GCC push_options 32#pragma GCC target("mmx") 33#define __DISABLE_MMX__ 34#endif /* __MMX__ */ 35 36/* The Intel API is flexible enough that we must allow aliasing with other 37 vector types, and their scalar components. */ 38typedef int __m64 __attribute__ ((__vector_size__ (8), __may_alias__)); 39 40/* Internal data types for implementing the intrinsics. */ 41typedef int __v2si __attribute__ ((__vector_size__ (8))); 42typedef short __v4hi __attribute__ ((__vector_size__ (8))); 43typedef char __v8qi __attribute__ ((__vector_size__ (8))); 44typedef long long __v1di __attribute__ ((__vector_size__ (8))); 45typedef float __v2sf __attribute__ ((__vector_size__ (8))); 46 47/* Empty the multimedia state. */ 48extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 49_mm_empty (void) 50{ 51 __builtin_ia32_emms (); 52} 53 54extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 55_m_empty (void) 56{ 57 _mm_empty (); 58} 59 60/* Convert I to a __m64 object. The integer is zero-extended to 64-bits. */ 61extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 62_mm_cvtsi32_si64 (int __i) 63{ 64 return (__m64) __builtin_ia32_vec_init_v2si (__i, 0); 65} 66 67extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 68_m_from_int (int __i) 69{ 70 return _mm_cvtsi32_si64 (__i); 71} 72 73#ifdef __x86_64__ 74/* Convert I to a __m64 object. */ 75 76/* Intel intrinsic. */ 77extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 78_m_from_int64 (long long __i) 79{ 80 return (__m64) __i; 81} 82 83extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 84_mm_cvtsi64_m64 (long long __i) 85{ 86 return (__m64) __i; 87} 88 89/* Microsoft intrinsic. */ 90extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 91_mm_cvtsi64x_si64 (long long __i) 92{ 93 return (__m64) __i; 94} 95 96extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 97_mm_set_pi64x (long long __i) 98{ 99 return (__m64) __i; 100} 101#endif 102 103/* Convert the lower 32 bits of the __m64 object into an integer. */ 104extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 105_mm_cvtsi64_si32 (__m64 __i) 106{ 107 return __builtin_ia32_vec_ext_v2si ((__v2si)__i, 0); 108} 109 110extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 111_m_to_int (__m64 __i) 112{ 113 return _mm_cvtsi64_si32 (__i); 114} 115 116#ifdef __x86_64__ 117/* Convert the __m64 object to a 64bit integer. */ 118 119/* Intel intrinsic. */ 120extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 121_m_to_int64 (__m64 __i) 122{ 123 return (long long)__i; 124} 125 126extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 127_mm_cvtm64_si64 (__m64 __i) 128{ 129 return (long long)__i; 130} 131 132/* Microsoft intrinsic. */ 133extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 134_mm_cvtsi64_si64x (__m64 __i) 135{ 136 return (long long)__i; 137} 138#endif 139 140/* Pack the four 16-bit values from M1 into the lower four 8-bit values of 141 the result, and the four 16-bit values from M2 into the upper four 8-bit 142 values of the result, all with signed saturation. */ 143extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 144_mm_packs_pi16 (__m64 __m1, __m64 __m2) 145{ 146 return (__m64) __builtin_ia32_packsswb ((__v4hi)__m1, (__v4hi)__m2); 147} 148 149extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 150_m_packsswb (__m64 __m1, __m64 __m2) 151{ 152 return _mm_packs_pi16 (__m1, __m2); 153} 154 155/* Pack the two 32-bit values from M1 in to the lower two 16-bit values of 156 the result, and the two 32-bit values from M2 into the upper two 16-bit 157 values of the result, all with signed saturation. */ 158extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 159_mm_packs_pi32 (__m64 __m1, __m64 __m2) 160{ 161 return (__m64) __builtin_ia32_packssdw ((__v2si)__m1, (__v2si)__m2); 162} 163 164extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 165_m_packssdw (__m64 __m1, __m64 __m2) 166{ 167 return _mm_packs_pi32 (__m1, __m2); 168} 169 170/* Pack the four 16-bit values from M1 into the lower four 8-bit values of 171 the result, and the four 16-bit values from M2 into the upper four 8-bit 172 values of the result, all with unsigned saturation. */ 173extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 174_mm_packs_pu16 (__m64 __m1, __m64 __m2) 175{ 176 return (__m64) __builtin_ia32_packuswb ((__v4hi)__m1, (__v4hi)__m2); 177} 178 179extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 180_m_packuswb (__m64 __m1, __m64 __m2) 181{ 182 return _mm_packs_pu16 (__m1, __m2); 183} 184 185/* Interleave the four 8-bit values from the high half of M1 with the four 186 8-bit values from the high half of M2. */ 187extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 188_mm_unpackhi_pi8 (__m64 __m1, __m64 __m2) 189{ 190 return (__m64) __builtin_ia32_punpckhbw ((__v8qi)__m1, (__v8qi)__m2); 191} 192 193extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 194_m_punpckhbw (__m64 __m1, __m64 __m2) 195{ 196 return _mm_unpackhi_pi8 (__m1, __m2); 197} 198 199/* Interleave the two 16-bit values from the high half of M1 with the two 200 16-bit values from the high half of M2. */ 201extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 202_mm_unpackhi_pi16 (__m64 __m1, __m64 __m2) 203{ 204 return (__m64) __builtin_ia32_punpckhwd ((__v4hi)__m1, (__v4hi)__m2); 205} 206 207extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 208_m_punpckhwd (__m64 __m1, __m64 __m2) 209{ 210 return _mm_unpackhi_pi16 (__m1, __m2); 211} 212 213/* Interleave the 32-bit value from the high half of M1 with the 32-bit 214 value from the high half of M2. */ 215extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 216_mm_unpackhi_pi32 (__m64 __m1, __m64 __m2) 217{ 218 return (__m64) __builtin_ia32_punpckhdq ((__v2si)__m1, (__v2si)__m2); 219} 220 221extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 222_m_punpckhdq (__m64 __m1, __m64 __m2) 223{ 224 return _mm_unpackhi_pi32 (__m1, __m2); 225} 226 227/* Interleave the four 8-bit values from the low half of M1 with the four 228 8-bit values from the low half of M2. */ 229extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 230_mm_unpacklo_pi8 (__m64 __m1, __m64 __m2) 231{ 232 return (__m64) __builtin_ia32_punpcklbw ((__v8qi)__m1, (__v8qi)__m2); 233} 234 235extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 236_m_punpcklbw (__m64 __m1, __m64 __m2) 237{ 238 return _mm_unpacklo_pi8 (__m1, __m2); 239} 240 241/* Interleave the two 16-bit values from the low half of M1 with the two 242 16-bit values from the low half of M2. */ 243extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 244_mm_unpacklo_pi16 (__m64 __m1, __m64 __m2) 245{ 246 return (__m64) __builtin_ia32_punpcklwd ((__v4hi)__m1, (__v4hi)__m2); 247} 248 249extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 250_m_punpcklwd (__m64 __m1, __m64 __m2) 251{ 252 return _mm_unpacklo_pi16 (__m1, __m2); 253} 254 255/* Interleave the 32-bit value from the low half of M1 with the 32-bit 256 value from the low half of M2. */ 257extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 258_mm_unpacklo_pi32 (__m64 __m1, __m64 __m2) 259{ 260 return (__m64) __builtin_ia32_punpckldq ((__v2si)__m1, (__v2si)__m2); 261} 262 263extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 264_m_punpckldq (__m64 __m1, __m64 __m2) 265{ 266 return _mm_unpacklo_pi32 (__m1, __m2); 267} 268 269/* Add the 8-bit values in M1 to the 8-bit values in M2. */ 270extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 271_mm_add_pi8 (__m64 __m1, __m64 __m2) 272{ 273 return (__m64) __builtin_ia32_paddb ((__v8qi)__m1, (__v8qi)__m2); 274} 275 276extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 277_m_paddb (__m64 __m1, __m64 __m2) 278{ 279 return _mm_add_pi8 (__m1, __m2); 280} 281 282/* Add the 16-bit values in M1 to the 16-bit values in M2. */ 283extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 284_mm_add_pi16 (__m64 __m1, __m64 __m2) 285{ 286 return (__m64) __builtin_ia32_paddw ((__v4hi)__m1, (__v4hi)__m2); 287} 288 289extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 290_m_paddw (__m64 __m1, __m64 __m2) 291{ 292 return _mm_add_pi16 (__m1, __m2); 293} 294 295/* Add the 32-bit values in M1 to the 32-bit values in M2. */ 296extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 297_mm_add_pi32 (__m64 __m1, __m64 __m2) 298{ 299 return (__m64) __builtin_ia32_paddd ((__v2si)__m1, (__v2si)__m2); 300} 301 302extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 303_m_paddd (__m64 __m1, __m64 __m2) 304{ 305 return _mm_add_pi32 (__m1, __m2); 306} 307 308/* Add the 64-bit values in M1 to the 64-bit values in M2. */ 309#ifndef __SSE2__ 310#pragma GCC push_options 311#pragma GCC target("sse2") 312#define __DISABLE_SSE2__ 313#endif /* __SSE2__ */ 314 315extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 316_mm_add_si64 (__m64 __m1, __m64 __m2) 317{ 318 return (__m64) __builtin_ia32_paddq ((__v1di)__m1, (__v1di)__m2); 319} 320#ifdef __DISABLE_SSE2__ 321#undef __DISABLE_SSE2__ 322#pragma GCC pop_options 323#endif /* __DISABLE_SSE2__ */ 324 325/* Add the 8-bit values in M1 to the 8-bit values in M2 using signed 326 saturated arithmetic. */ 327extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 328_mm_adds_pi8 (__m64 __m1, __m64 __m2) 329{ 330 return (__m64) __builtin_ia32_paddsb ((__v8qi)__m1, (__v8qi)__m2); 331} 332 333extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 334_m_paddsb (__m64 __m1, __m64 __m2) 335{ 336 return _mm_adds_pi8 (__m1, __m2); 337} 338 339/* Add the 16-bit values in M1 to the 16-bit values in M2 using signed 340 saturated arithmetic. */ 341extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 342_mm_adds_pi16 (__m64 __m1, __m64 __m2) 343{ 344 return (__m64) __builtin_ia32_paddsw ((__v4hi)__m1, (__v4hi)__m2); 345} 346 347extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 348_m_paddsw (__m64 __m1, __m64 __m2) 349{ 350 return _mm_adds_pi16 (__m1, __m2); 351} 352 353/* Add the 8-bit values in M1 to the 8-bit values in M2 using unsigned 354 saturated arithmetic. */ 355extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 356_mm_adds_pu8 (__m64 __m1, __m64 __m2) 357{ 358 return (__m64) __builtin_ia32_paddusb ((__v8qi)__m1, (__v8qi)__m2); 359} 360 361extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 362_m_paddusb (__m64 __m1, __m64 __m2) 363{ 364 return _mm_adds_pu8 (__m1, __m2); 365} 366 367/* Add the 16-bit values in M1 to the 16-bit values in M2 using unsigned 368 saturated arithmetic. */ 369extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 370_mm_adds_pu16 (__m64 __m1, __m64 __m2) 371{ 372 return (__m64) __builtin_ia32_paddusw ((__v4hi)__m1, (__v4hi)__m2); 373} 374 375extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 376_m_paddusw (__m64 __m1, __m64 __m2) 377{ 378 return _mm_adds_pu16 (__m1, __m2); 379} 380 381/* Subtract the 8-bit values in M2 from the 8-bit values in M1. */ 382extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 383_mm_sub_pi8 (__m64 __m1, __m64 __m2) 384{ 385 return (__m64) __builtin_ia32_psubb ((__v8qi)__m1, (__v8qi)__m2); 386} 387 388extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 389_m_psubb (__m64 __m1, __m64 __m2) 390{ 391 return _mm_sub_pi8 (__m1, __m2); 392} 393 394/* Subtract the 16-bit values in M2 from the 16-bit values in M1. */ 395extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 396_mm_sub_pi16 (__m64 __m1, __m64 __m2) 397{ 398 return (__m64) __builtin_ia32_psubw ((__v4hi)__m1, (__v4hi)__m2); 399} 400 401extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 402_m_psubw (__m64 __m1, __m64 __m2) 403{ 404 return _mm_sub_pi16 (__m1, __m2); 405} 406 407/* Subtract the 32-bit values in M2 from the 32-bit values in M1. */ 408extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 409_mm_sub_pi32 (__m64 __m1, __m64 __m2) 410{ 411 return (__m64) __builtin_ia32_psubd ((__v2si)__m1, (__v2si)__m2); 412} 413 414extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 415_m_psubd (__m64 __m1, __m64 __m2) 416{ 417 return _mm_sub_pi32 (__m1, __m2); 418} 419 420/* Add the 64-bit values in M1 to the 64-bit values in M2. */ 421#ifndef __SSE2__ 422#pragma GCC push_options 423#pragma GCC target("sse2") 424#define __DISABLE_SSE2__ 425#endif /* __SSE2__ */ 426 427extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 428_mm_sub_si64 (__m64 __m1, __m64 __m2) 429{ 430 return (__m64) __builtin_ia32_psubq ((__v1di)__m1, (__v1di)__m2); 431} 432#ifdef __DISABLE_SSE2__ 433#undef __DISABLE_SSE2__ 434#pragma GCC pop_options 435#endif /* __DISABLE_SSE2__ */ 436 437/* Subtract the 8-bit values in M2 from the 8-bit values in M1 using signed 438 saturating arithmetic. */ 439extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 440_mm_subs_pi8 (__m64 __m1, __m64 __m2) 441{ 442 return (__m64) __builtin_ia32_psubsb ((__v8qi)__m1, (__v8qi)__m2); 443} 444 445extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 446_m_psubsb (__m64 __m1, __m64 __m2) 447{ 448 return _mm_subs_pi8 (__m1, __m2); 449} 450 451/* Subtract the 16-bit values in M2 from the 16-bit values in M1 using 452 signed saturating arithmetic. */ 453extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 454_mm_subs_pi16 (__m64 __m1, __m64 __m2) 455{ 456 return (__m64) __builtin_ia32_psubsw ((__v4hi)__m1, (__v4hi)__m2); 457} 458 459extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 460_m_psubsw (__m64 __m1, __m64 __m2) 461{ 462 return _mm_subs_pi16 (__m1, __m2); 463} 464 465/* Subtract the 8-bit values in M2 from the 8-bit values in M1 using 466 unsigned saturating arithmetic. */ 467extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 468_mm_subs_pu8 (__m64 __m1, __m64 __m2) 469{ 470 return (__m64) __builtin_ia32_psubusb ((__v8qi)__m1, (__v8qi)__m2); 471} 472 473extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 474_m_psubusb (__m64 __m1, __m64 __m2) 475{ 476 return _mm_subs_pu8 (__m1, __m2); 477} 478 479/* Subtract the 16-bit values in M2 from the 16-bit values in M1 using 480 unsigned saturating arithmetic. */ 481extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 482_mm_subs_pu16 (__m64 __m1, __m64 __m2) 483{ 484 return (__m64) __builtin_ia32_psubusw ((__v4hi)__m1, (__v4hi)__m2); 485} 486 487extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 488_m_psubusw (__m64 __m1, __m64 __m2) 489{ 490 return _mm_subs_pu16 (__m1, __m2); 491} 492 493/* Multiply four 16-bit values in M1 by four 16-bit values in M2 producing 494 four 32-bit intermediate results, which are then summed by pairs to 495 produce two 32-bit results. */ 496extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 497_mm_madd_pi16 (__m64 __m1, __m64 __m2) 498{ 499 return (__m64) __builtin_ia32_pmaddwd ((__v4hi)__m1, (__v4hi)__m2); 500} 501 502extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 503_m_pmaddwd (__m64 __m1, __m64 __m2) 504{ 505 return _mm_madd_pi16 (__m1, __m2); 506} 507 508/* Multiply four signed 16-bit values in M1 by four signed 16-bit values in 509 M2 and produce the high 16 bits of the 32-bit results. */ 510extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 511_mm_mulhi_pi16 (__m64 __m1, __m64 __m2) 512{ 513 return (__m64) __builtin_ia32_pmulhw ((__v4hi)__m1, (__v4hi)__m2); 514} 515 516extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 517_m_pmulhw (__m64 __m1, __m64 __m2) 518{ 519 return _mm_mulhi_pi16 (__m1, __m2); 520} 521 522/* Multiply four 16-bit values in M1 by four 16-bit values in M2 and produce 523 the low 16 bits of the results. */ 524extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 525_mm_mullo_pi16 (__m64 __m1, __m64 __m2) 526{ 527 return (__m64) __builtin_ia32_pmullw ((__v4hi)__m1, (__v4hi)__m2); 528} 529 530extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 531_m_pmullw (__m64 __m1, __m64 __m2) 532{ 533 return _mm_mullo_pi16 (__m1, __m2); 534} 535 536/* Shift four 16-bit values in M left by COUNT. */ 537extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 538_mm_sll_pi16 (__m64 __m, __m64 __count) 539{ 540 return (__m64) __builtin_ia32_psllw ((__v4hi)__m, (__v4hi)__count); 541} 542 543extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 544_m_psllw (__m64 __m, __m64 __count) 545{ 546 return _mm_sll_pi16 (__m, __count); 547} 548 549extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 550_mm_slli_pi16 (__m64 __m, int __count) 551{ 552 return (__m64) __builtin_ia32_psllwi ((__v4hi)__m, __count); 553} 554 555extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 556_m_psllwi (__m64 __m, int __count) 557{ 558 return _mm_slli_pi16 (__m, __count); 559} 560 561/* Shift two 32-bit values in M left by COUNT. */ 562extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 563_mm_sll_pi32 (__m64 __m, __m64 __count) 564{ 565 return (__m64) __builtin_ia32_pslld ((__v2si)__m, (__v2si)__count); 566} 567 568extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 569_m_pslld (__m64 __m, __m64 __count) 570{ 571 return _mm_sll_pi32 (__m, __count); 572} 573 574extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 575_mm_slli_pi32 (__m64 __m, int __count) 576{ 577 return (__m64) __builtin_ia32_pslldi ((__v2si)__m, __count); 578} 579 580extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 581_m_pslldi (__m64 __m, int __count) 582{ 583 return _mm_slli_pi32 (__m, __count); 584} 585 586/* Shift the 64-bit value in M left by COUNT. */ 587extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 588_mm_sll_si64 (__m64 __m, __m64 __count) 589{ 590 return (__m64) __builtin_ia32_psllq ((__v1di)__m, (__v1di)__count); 591} 592 593extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 594_m_psllq (__m64 __m, __m64 __count) 595{ 596 return _mm_sll_si64 (__m, __count); 597} 598 599extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 600_mm_slli_si64 (__m64 __m, int __count) 601{ 602 return (__m64) __builtin_ia32_psllqi ((__v1di)__m, __count); 603} 604 605extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 606_m_psllqi (__m64 __m, int __count) 607{ 608 return _mm_slli_si64 (__m, __count); 609} 610 611/* Shift four 16-bit values in M right by COUNT; shift in the sign bit. */ 612extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 613_mm_sra_pi16 (__m64 __m, __m64 __count) 614{ 615 return (__m64) __builtin_ia32_psraw ((__v4hi)__m, (__v4hi)__count); 616} 617 618extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 619_m_psraw (__m64 __m, __m64 __count) 620{ 621 return _mm_sra_pi16 (__m, __count); 622} 623 624extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 625_mm_srai_pi16 (__m64 __m, int __count) 626{ 627 return (__m64) __builtin_ia32_psrawi ((__v4hi)__m, __count); 628} 629 630extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 631_m_psrawi (__m64 __m, int __count) 632{ 633 return _mm_srai_pi16 (__m, __count); 634} 635 636/* Shift two 32-bit values in M right by COUNT; shift in the sign bit. */ 637extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 638_mm_sra_pi32 (__m64 __m, __m64 __count) 639{ 640 return (__m64) __builtin_ia32_psrad ((__v2si)__m, (__v2si)__count); 641} 642 643extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 644_m_psrad (__m64 __m, __m64 __count) 645{ 646 return _mm_sra_pi32 (__m, __count); 647} 648 649extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 650_mm_srai_pi32 (__m64 __m, int __count) 651{ 652 return (__m64) __builtin_ia32_psradi ((__v2si)__m, __count); 653} 654 655extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 656_m_psradi (__m64 __m, int __count) 657{ 658 return _mm_srai_pi32 (__m, __count); 659} 660 661/* Shift four 16-bit values in M right by COUNT; shift in zeros. */ 662extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 663_mm_srl_pi16 (__m64 __m, __m64 __count) 664{ 665 return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, (__v4hi)__count); 666} 667 668extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 669_m_psrlw (__m64 __m, __m64 __count) 670{ 671 return _mm_srl_pi16 (__m, __count); 672} 673 674extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 675_mm_srli_pi16 (__m64 __m, int __count) 676{ 677 return (__m64) __builtin_ia32_psrlwi ((__v4hi)__m, __count); 678} 679 680extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 681_m_psrlwi (__m64 __m, int __count) 682{ 683 return _mm_srli_pi16 (__m, __count); 684} 685 686/* Shift two 32-bit values in M right by COUNT; shift in zeros. */ 687extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 688_mm_srl_pi32 (__m64 __m, __m64 __count) 689{ 690 return (__m64) __builtin_ia32_psrld ((__v2si)__m, (__v2si)__count); 691} 692 693extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 694_m_psrld (__m64 __m, __m64 __count) 695{ 696 return _mm_srl_pi32 (__m, __count); 697} 698 699extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 700_mm_srli_pi32 (__m64 __m, int __count) 701{ 702 return (__m64) __builtin_ia32_psrldi ((__v2si)__m, __count); 703} 704 705extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 706_m_psrldi (__m64 __m, int __count) 707{ 708 return _mm_srli_pi32 (__m, __count); 709} 710 711/* Shift the 64-bit value in M left by COUNT; shift in zeros. */ 712extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 713_mm_srl_si64 (__m64 __m, __m64 __count) 714{ 715 return (__m64) __builtin_ia32_psrlq ((__v1di)__m, (__v1di)__count); 716} 717 718extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 719_m_psrlq (__m64 __m, __m64 __count) 720{ 721 return _mm_srl_si64 (__m, __count); 722} 723 724extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 725_mm_srli_si64 (__m64 __m, int __count) 726{ 727 return (__m64) __builtin_ia32_psrlqi ((__v1di)__m, __count); 728} 729 730extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 731_m_psrlqi (__m64 __m, int __count) 732{ 733 return _mm_srli_si64 (__m, __count); 734} 735 736/* Bit-wise AND the 64-bit values in M1 and M2. */ 737extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 738_mm_and_si64 (__m64 __m1, __m64 __m2) 739{ 740 return __builtin_ia32_pand (__m1, __m2); 741} 742 743extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 744_m_pand (__m64 __m1, __m64 __m2) 745{ 746 return _mm_and_si64 (__m1, __m2); 747} 748 749/* Bit-wise complement the 64-bit value in M1 and bit-wise AND it with the 750 64-bit value in M2. */ 751extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 752_mm_andnot_si64 (__m64 __m1, __m64 __m2) 753{ 754 return __builtin_ia32_pandn (__m1, __m2); 755} 756 757extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 758_m_pandn (__m64 __m1, __m64 __m2) 759{ 760 return _mm_andnot_si64 (__m1, __m2); 761} 762 763/* Bit-wise inclusive OR the 64-bit values in M1 and M2. */ 764extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 765_mm_or_si64 (__m64 __m1, __m64 __m2) 766{ 767 return __builtin_ia32_por (__m1, __m2); 768} 769 770extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 771_m_por (__m64 __m1, __m64 __m2) 772{ 773 return _mm_or_si64 (__m1, __m2); 774} 775 776/* Bit-wise exclusive OR the 64-bit values in M1 and M2. */ 777extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 778_mm_xor_si64 (__m64 __m1, __m64 __m2) 779{ 780 return __builtin_ia32_pxor (__m1, __m2); 781} 782 783extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 784_m_pxor (__m64 __m1, __m64 __m2) 785{ 786 return _mm_xor_si64 (__m1, __m2); 787} 788 789/* Compare eight 8-bit values. The result of the comparison is 0xFF if the 790 test is true and zero if false. */ 791extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 792_mm_cmpeq_pi8 (__m64 __m1, __m64 __m2) 793{ 794 return (__m64) __builtin_ia32_pcmpeqb ((__v8qi)__m1, (__v8qi)__m2); 795} 796 797extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 798_m_pcmpeqb (__m64 __m1, __m64 __m2) 799{ 800 return _mm_cmpeq_pi8 (__m1, __m2); 801} 802 803extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 804_mm_cmpgt_pi8 (__m64 __m1, __m64 __m2) 805{ 806 return (__m64) __builtin_ia32_pcmpgtb ((__v8qi)__m1, (__v8qi)__m2); 807} 808 809extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 810_m_pcmpgtb (__m64 __m1, __m64 __m2) 811{ 812 return _mm_cmpgt_pi8 (__m1, __m2); 813} 814 815/* Compare four 16-bit values. The result of the comparison is 0xFFFF if 816 the test is true and zero if false. */ 817extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 818_mm_cmpeq_pi16 (__m64 __m1, __m64 __m2) 819{ 820 return (__m64) __builtin_ia32_pcmpeqw ((__v4hi)__m1, (__v4hi)__m2); 821} 822 823extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 824_m_pcmpeqw (__m64 __m1, __m64 __m2) 825{ 826 return _mm_cmpeq_pi16 (__m1, __m2); 827} 828 829extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 830_mm_cmpgt_pi16 (__m64 __m1, __m64 __m2) 831{ 832 return (__m64) __builtin_ia32_pcmpgtw ((__v4hi)__m1, (__v4hi)__m2); 833} 834 835extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 836_m_pcmpgtw (__m64 __m1, __m64 __m2) 837{ 838 return _mm_cmpgt_pi16 (__m1, __m2); 839} 840 841/* Compare two 32-bit values. The result of the comparison is 0xFFFFFFFF if 842 the test is true and zero if false. */ 843extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 844_mm_cmpeq_pi32 (__m64 __m1, __m64 __m2) 845{ 846 return (__m64) __builtin_ia32_pcmpeqd ((__v2si)__m1, (__v2si)__m2); 847} 848 849extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 850_m_pcmpeqd (__m64 __m1, __m64 __m2) 851{ 852 return _mm_cmpeq_pi32 (__m1, __m2); 853} 854 855extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 856_mm_cmpgt_pi32 (__m64 __m1, __m64 __m2) 857{ 858 return (__m64) __builtin_ia32_pcmpgtd ((__v2si)__m1, (__v2si)__m2); 859} 860 861extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 862_m_pcmpgtd (__m64 __m1, __m64 __m2) 863{ 864 return _mm_cmpgt_pi32 (__m1, __m2); 865} 866 867/* Creates a 64-bit zero. */ 868extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 869_mm_setzero_si64 (void) 870{ 871 return (__m64)0LL; 872} 873 874/* Creates a vector of two 32-bit values; I0 is least significant. */ 875extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 876_mm_set_pi32 (int __i1, int __i0) 877{ 878 return (__m64) __builtin_ia32_vec_init_v2si (__i0, __i1); 879} 880 881/* Creates a vector of four 16-bit values; W0 is least significant. */ 882extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 883_mm_set_pi16 (short __w3, short __w2, short __w1, short __w0) 884{ 885 return (__m64) __builtin_ia32_vec_init_v4hi (__w0, __w1, __w2, __w3); 886} 887 888/* Creates a vector of eight 8-bit values; B0 is least significant. */ 889extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 890_mm_set_pi8 (char __b7, char __b6, char __b5, char __b4, 891 char __b3, char __b2, char __b1, char __b0) 892{ 893 return (__m64) __builtin_ia32_vec_init_v8qi (__b0, __b1, __b2, __b3, 894 __b4, __b5, __b6, __b7); 895} 896 897/* Similar, but with the arguments in reverse order. */ 898extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 899_mm_setr_pi32 (int __i0, int __i1) 900{ 901 return _mm_set_pi32 (__i1, __i0); 902} 903 904extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 905_mm_setr_pi16 (short __w0, short __w1, short __w2, short __w3) 906{ 907 return _mm_set_pi16 (__w3, __w2, __w1, __w0); 908} 909 910extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 911_mm_setr_pi8 (char __b0, char __b1, char __b2, char __b3, 912 char __b4, char __b5, char __b6, char __b7) 913{ 914 return _mm_set_pi8 (__b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0); 915} 916 917/* Creates a vector of two 32-bit values, both elements containing I. */ 918extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 919_mm_set1_pi32 (int __i) 920{ 921 return _mm_set_pi32 (__i, __i); 922} 923 924/* Creates a vector of four 16-bit values, all elements containing W. */ 925extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 926_mm_set1_pi16 (short __w) 927{ 928 return _mm_set_pi16 (__w, __w, __w, __w); 929} 930 931/* Creates a vector of eight 8-bit values, all elements containing B. */ 932extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) 933_mm_set1_pi8 (char __b) 934{ 935 return _mm_set_pi8 (__b, __b, __b, __b, __b, __b, __b, __b); 936} 937#ifdef __DISABLE_MMX__ 938#undef __DISABLE_MMX__ 939#pragma GCC pop_options 940#endif /* __DISABLE_MMX__ */ 941 942#endif /* _MMINTRIN_H_INCLUDED */ 943