mmintrin.h revision 90075
1327Sjkh/* Copyright (C) 2002 Free Software Foundation, Inc. 216404Sjkh 3327Sjkh This file is part of GNU CC. 4327Sjkh 5327Sjkh GNU CC is free software; you can redistribute it and/or modify 6327Sjkh it under the terms of the GNU General Public License as published by 7327Sjkh the Free Software Foundation; either version 2, or (at your option) 8327Sjkh any later version. 9327Sjkh 10327Sjkh GNU CC is distributed in the hope that it will be useful, 11327Sjkh but WITHOUT ANY WARRANTY; without even the implied warranty of 12327Sjkh MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13327Sjkh GNU General Public License for more details. 14327Sjkh 15327Sjkh You should have received a copy of the GNU General Public License 16327Sjkh along with GNU CC; see the file COPYING. If not, write to 17327Sjkh the Free Software Foundation, 59 Temple Place - Suite 330, 18327Sjkh Boston, MA 02111-1307, USA. */ 19327Sjkh 20327Sjkh/* As a special exception, if you include this header file into source 21327Sjkh files compiled by GCC, this header file does not by itself cause 22327Sjkh the resulting executable to be covered by the GNU General Public 23327Sjkh License. This exception does not however invalidate any other 24327Sjkh reasons why the executable file might be covered by the GNU General 25327Sjkh Public License. */ 26327Sjkh 27327Sjkh/* Implemented from the specification included in the Intel C++ Compiler 28327Sjkh User Guide and Reference, version 5.0. */ 29327Sjkh 30327Sjkh#ifndef _MMINTRIN_H_INCLUDED 31327Sjkh#define _MMINTRIN_H_INCLUDED 32327Sjkh 33327Sjkh/* The data type intended for user use. */ 34327Sjkhtypedef unsigned long long __m64; 35327Sjkh 367937Sjkh/* Internal data types for implementing the intrinsics. */ 37327Sjkhtypedef int __v2si __attribute__ ((__mode__ (__V2SI__))); 38327Sjkhtypedef int __v4hi __attribute__ ((__mode__ (__V4HI__))); 39327Sjkhtypedef int __v8qi __attribute__ ((__mode__ (__V8QI__))); 407937Sjkh 417937Sjkh/* Empty the multimedia state. */ 427937Sjkhstatic __inline void 43327Sjkh_mm_empty (void) 4416404Sjkh{ 4516404Sjkh __builtin_ia32_emms (); 46327Sjkh} 4716404Sjkh 4816404Sjkh/* Convert I to a __m64 object. The integer is zero-extended to 64-bits. */ 4916404Sjkhstatic __inline __m64 5016404Sjkh_mm_cvtsi32_si64 (int __i) 5116404Sjkh{ 5216404Sjkh return (unsigned int) __i; 538857Srgrimes} 5416404Sjkh 5516404Sjkh/* Convert the lower 32 bits of the __m64 object into an integer. */ 5616404Sjkhstatic __inline int 5716404Sjkh_mm_cvtsi64_si32 (__m64 __i) 5816404Sjkh{ 5916404Sjkh return __i; 6016404Sjkh} 6116404Sjkh 6216404Sjkh/* Pack the four 16-bit values from M1 into the lower four 8-bit values of 63327Sjkh the result, and the four 16-bit values from M2 into the upper four 8-bit 6416404Sjkh values of the result, all with signed saturation. */ 6516404Sjkhstatic __inline __m64 6616404Sjkh_mm_packs_pi16 (__m64 __m1, __m64 __m2) 67327Sjkh{ 68327Sjkh return (__m64) __builtin_ia32_packsswb ((__v4hi)__m1, (__v4hi)__m2); 69327Sjkh} 7011780Sjkh 7111780Sjkh/* Pack the two 32-bit values from M1 in to the lower two 16-bit values of 72327Sjkh the result, and the two 32-bit values from M2 into the upper two 16-bit 73327Sjkh values of the result, all with signed saturation. */ 74327Sjkhstatic __inline __m64 758086Sjkh_mm_packs_pi32 (__m64 __m1, __m64 __m2) 76327Sjkh{ 778086Sjkh return (__m64) __builtin_ia32_packssdw ((__v2si)__m1, (__v2si)__m2); 78327Sjkh} 79327Sjkh 808086Sjkh/* Pack the four 16-bit values from M1 into the lower four 8-bit values of 818142Sjkh the result, and the four 16-bit values from M2 into the upper four 8-bit 828086Sjkh values of the result, all with unsigned saturation. */ 83327Sjkhstatic __inline __m64 848086Sjkh_mm_packs_pu16 (__m64 __m1, __m64 __m2) 8511780Sjkh{ 868086Sjkh return (__m64) __builtin_ia32_packuswb ((__v4hi)__m1, (__v4hi)__m2); 878086Sjkh} 888086Sjkh 898086Sjkh/* Interleave the four 8-bit values from the high half of M1 with the four 909782Sache 8-bit values from the high half of M2. */ 918086Sjkhstatic __inline __m64 92327Sjkh_mm_unpackhi_pi8 (__m64 __m1, __m64 __m2) 938423Sjkh{ 948423Sjkh return (__m64) __builtin_ia32_punpckhbw ((__v8qi)__m1, (__v8qi)__m2); 958423Sjkh} 968423Sjkh 978423Sjkh/* Interleave the two 16-bit values from the high half of M1 with the two 988423Sjkh 16-bit values from the high half of M2. */ 998423Sjkhstatic __inline __m64 1008423Sjkh_mm_unpackhi_pi16 (__m64 __m1, __m64 __m2) 1018086Sjkh{ 1028086Sjkh return (__m64) __builtin_ia32_punpckhwd ((__v4hi)__m1, (__v4hi)__m2); 1038086Sjkh} 10411780Sjkh 1058086Sjkh/* Interleave the 32-bit value from the high half of M1 with the 32-bit 1068086Sjkh value from the high half of M2. */ 1078086Sjkhstatic __inline __m64 1083364Sjkh_mm_unpackhi_pi32 (__m64 __m1, __m64 __m2) 1093364Sjkh{ 1103364Sjkh return (__m64) __builtin_ia32_punpckhdq ((__v2si)__m1, (__v2si)__m2); 1113577Sjkh} 1123577Sjkh 1133364Sjkh/* Interleave the four 8-bit values from the low half of M1 with the four 1143364Sjkh 8-bit values from the low half of M2. */ 1153364Sjkhstatic __inline __m64 1168086Sjkh_mm_unpacklo_pi8 (__m64 __m1, __m64 __m2) 1178086Sjkh{ 1183364Sjkh return (__m64) __builtin_ia32_punpcklbw ((__v8qi)__m1, (__v8qi)__m2); 11911780Sjkh} 120327Sjkh 121327Sjkh/* Interleave the two 16-bit values from the low half of M1 with the two 1228086Sjkh 16-bit values from the low half of M2. */ 1238086Sjkhstatic __inline __m64 124327Sjkh_mm_unpacklo_pi16 (__m64 __m1, __m64 __m2) 125327Sjkh{ 1268086Sjkh return (__m64) __builtin_ia32_punpcklwd ((__v4hi)__m1, (__v4hi)__m2); 127327Sjkh} 1287937Sjkh 1297937Sjkh/* Interleave the 32-bit value from the low half of M1 with the 32-bit 1307937Sjkh value from the low half of M2. */ 1317937Sjkhstatic __inline __m64 132327Sjkh_mm_unpacklo_pi32 (__m64 __m1, __m64 __m2) 1338086Sjkh{ 134327Sjkh return (__m64) __builtin_ia32_punpckldq ((__v2si)__m1, (__v2si)__m2); 135327Sjkh} 136327Sjkh 137327Sjkh/* Add the 8-bit values in M1 to the 8-bit values in M2. */ 138327Sjkhstatic __inline __m64 139327Sjkh_mm_add_pi8 (__m64 __m1, __m64 __m2) 140327Sjkh{ 141327Sjkh return (__m64) __builtin_ia32_paddb ((__v8qi)__m1, (__v8qi)__m2); 142327Sjkh} 143327Sjkh 144327Sjkh/* Add the 16-bit values in M1 to the 16-bit values in M2. */ 145327Sjkhstatic __inline __m64 146327Sjkh_mm_add_pi16 (__m64 __m1, __m64 __m2) 147327Sjkh{ 1488086Sjkh return (__m64) __builtin_ia32_paddw ((__v4hi)__m1, (__v4hi)__m2); 1498086Sjkh} 150327Sjkh 151327Sjkh/* Add the 32-bit values in M1 to the 32-bit values in M2. */ 152327Sjkhstatic __inline __m64 153327Sjkh_mm_add_pi32 (__m64 __m1, __m64 __m2) 154327Sjkh{ 155327Sjkh return (__m64) __builtin_ia32_paddd ((__v2si)__m1, (__v2si)__m2); 156327Sjkh} 157327Sjkh 158327Sjkh/* Add the 8-bit values in M1 to the 8-bit values in M2 using signed 159327Sjkh saturated arithmetic. */ 1608086Sjkhstatic __inline __m64 161327Sjkh_mm_adds_pi8 (__m64 __m1, __m64 __m2) 1628086Sjkh{ 1638086Sjkh return (__m64) __builtin_ia32_paddsb ((__v8qi)__m1, (__v8qi)__m2); 164327Sjkh} 165327Sjkh 166327Sjkh/* Add the 16-bit values in M1 to the 16-bit values in M2 using signed 167411Sjkh saturated arithmetic. */ 168411Sjkhstatic __inline __m64 169327Sjkh_mm_adds_pi16 (__m64 __m1, __m64 __m2) 170379Sjkh{ 1714996Sjkh return (__m64) __builtin_ia32_paddsw ((__v4hi)__m1, (__v4hi)__m2); 1724996Sjkh} 173327Sjkh 174379Sjkh/* Add the 8-bit values in M1 to the 8-bit values in M2 using unsigned 1754996Sjkh saturated arithmetic. */ 1764996Sjkhstatic __inline __m64 177327Sjkh_mm_adds_pu8 (__m64 __m1, __m64 __m2) 178379Sjkh{ 179327Sjkh return (__m64) __builtin_ia32_paddusb ((__v8qi)__m1, (__v8qi)__m2); 180379Sjkh} 181327Sjkh 182379Sjkh/* Add the 16-bit values in M1 to the 16-bit values in M2 using unsigned 1834996Sjkh saturated arithmetic. */ 1844996Sjkhstatic __inline __m64 185327Sjkh_mm_adds_pu16 (__m64 __m1, __m64 __m2) 186379Sjkh{ 187411Sjkh return (__m64) __builtin_ia32_paddusw ((__v4hi)__m1, (__v4hi)__m2); 188411Sjkh} 189411Sjkh 190411Sjkh/* Subtract the 8-bit values in M2 from the 8-bit values in M1. */ 191327Sjkhstatic __inline __m64 192327Sjkh_mm_sub_pi8 (__m64 __m1, __m64 __m2) 1938086Sjkh{ 19411780Sjkh return (__m64) __builtin_ia32_psubb ((__v8qi)__m1, (__v8qi)__m2); 1958086Sjkh} 1968086Sjkh 1978086Sjkh/* Subtract the 16-bit values in M2 from the 16-bit values in M1. */ 198327Sjkhstatic __inline __m64 199327Sjkh_mm_sub_pi16 (__m64 __m1, __m64 __m2) 200327Sjkh{ 201327Sjkh return (__m64) __builtin_ia32_psubw ((__v4hi)__m1, (__v4hi)__m2); 202327Sjkh} 20311780Sjkh 204327Sjkh/* Subtract the 32-bit values in M2 from the 32-bit values in M1. */ 205static __inline __m64 206_mm_sub_pi32 (__m64 __m1, __m64 __m2) 207{ 208 return (__m64) __builtin_ia32_psubd ((__v2si)__m1, (__v2si)__m2); 209} 210 211/* Subtract the 8-bit values in M2 from the 8-bit values in M1 using signed 212 saturating arithmetic. */ 213static __inline __m64 214_mm_subs_pi8 (__m64 __m1, __m64 __m2) 215{ 216 return (__m64) __builtin_ia32_psubsb ((__v8qi)__m1, (__v8qi)__m2); 217} 218 219/* Subtract the 16-bit values in M2 from the 16-bit values in M1 using 220 signed saturating arithmetic. */ 221static __inline __m64 222_mm_subs_pi16 (__m64 __m1, __m64 __m2) 223{ 224 return (__m64) __builtin_ia32_psubsw ((__v4hi)__m1, (__v4hi)__m2); 225} 226 227/* Subtract the 8-bit values in M2 from the 8-bit values in M1 using 228 unsigned saturating arithmetic. */ 229static __inline __m64 230_mm_subs_pu8 (__m64 __m1, __m64 __m2) 231{ 232 return (__m64) __builtin_ia32_psubusb ((__v8qi)__m1, (__v8qi)__m2); 233} 234 235/* Subtract the 16-bit values in M2 from the 16-bit values in M1 using 236 unsigned saturating arithmetic. */ 237static __inline __m64 238_mm_subs_pu16 (__m64 __m1, __m64 __m2) 239{ 240 return (__m64) __builtin_ia32_psubusw ((__v4hi)__m1, (__v4hi)__m2); 241} 242 243/* Multiply four 16-bit values in M1 by four 16-bit values in M2 producing 244 four 32-bit intermediate results, which are then summed by pairs to 245 produce two 32-bit results. */ 246static __inline __m64 247_mm_madd_pi16 (__m64 __m1, __m64 __m2) 248{ 249 return (__m64) __builtin_ia32_pmaddwd ((__v4hi)__m1, (__v4hi)__m2); 250} 251 252/* Multiply four signed 16-bit values in M1 by four signed 16-bit values in 253 M2 and produce the high 16 bits of the 32-bit results. */ 254static __inline __m64 255_mm_mulhi_pi16 (__m64 __m1, __m64 __m2) 256{ 257 return (__m64) __builtin_ia32_pmulhw ((__v4hi)__m1, (__v4hi)__m2); 258} 259 260/* Multiply four 16-bit values in M1 by four 16-bit values in M2 and produce 261 the low 16 bits of the results. */ 262static __inline __m64 263_mm_mullo_pi16 (__m64 __m1, __m64 __m2) 264{ 265 return (__m64) __builtin_ia32_pmullw ((__v4hi)__m1, (__v4hi)__m2); 266} 267 268/* Shift four 16-bit values in M left by COUNT. */ 269static __inline __m64 270_mm_sll_pi16 (__m64 __m, __m64 __count) 271{ 272 return (__m64) __builtin_ia32_psllw ((__v4hi)__m, __count); 273} 274 275static __inline __m64 276_mm_slli_pi16 (__m64 __m, int __count) 277{ 278 return (__m64) __builtin_ia32_psllw ((__v4hi)__m, __count); 279} 280 281/* Shift two 32-bit values in M left by COUNT. */ 282static __inline __m64 283_mm_sll_pi32 (__m64 __m, __m64 __count) 284{ 285 return (__m64) __builtin_ia32_pslld ((__v2si)__m, __count); 286} 287 288static __inline __m64 289_mm_slli_pi32 (__m64 __m, int __count) 290{ 291 return (__m64) __builtin_ia32_pslld ((__v2si)__m, __count); 292} 293 294/* Shift the 64-bit value in M left by COUNT. */ 295static __inline __m64 296_mm_sll_pi64 (__m64 __m, __m64 __count) 297{ 298 return (__m64) __builtin_ia32_psllq (__m, __count); 299} 300 301static __inline __m64 302_mm_slli_pi64 (__m64 __m, int __count) 303{ 304 return (__m64) __builtin_ia32_psllq (__m, __count); 305} 306 307/* Shift four 16-bit values in M right by COUNT; shift in the sign bit. */ 308static __inline __m64 309_mm_sra_pi16 (__m64 __m, __m64 __count) 310{ 311 return (__m64) __builtin_ia32_psraw ((__v4hi)__m, __count); 312} 313 314static __inline __m64 315_mm_srai_pi16 (__m64 __m, int __count) 316{ 317 return (__m64) __builtin_ia32_psraw ((__v4hi)__m, __count); 318} 319 320/* Shift two 32-bit values in M right by COUNT; shift in the sign bit. */ 321static __inline __m64 322_mm_sra_pi32 (__m64 __m, __m64 __count) 323{ 324 return (__m64) __builtin_ia32_psrad ((__v2si)__m, __count); 325} 326 327static __inline __m64 328_mm_srai_pi32 (__m64 __m, int __count) 329{ 330 return (__m64) __builtin_ia32_psrad ((__v2si)__m, __count); 331} 332 333/* Shift four 16-bit values in M right by COUNT; shift in zeros. */ 334static __inline __m64 335_mm_srl_pi16 (__m64 __m, __m64 __count) 336{ 337 return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, __count); 338} 339 340static __inline __m64 341_mm_srli_pi16 (__m64 __m, int __count) 342{ 343 return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, __count); 344} 345 346/* Shift two 32-bit values in M right by COUNT; shift in zeros. */ 347static __inline __m64 348_mm_srl_pi32 (__m64 __m, __m64 __count) 349{ 350 return (__m64) __builtin_ia32_psrld ((__v2si)__m, __count); 351} 352 353static __inline __m64 354_mm_srli_pi32 (__m64 __m, int __count) 355{ 356 return (__m64) __builtin_ia32_psrld ((__v2si)__m, __count); 357} 358 359/* Shift the 64-bit value in M left by COUNT; shift in zeros. */ 360static __inline __m64 361_mm_srl_pi64 (__m64 __m, __m64 __count) 362{ 363 return (__m64) __builtin_ia32_psrlq (__m, __count); 364} 365 366static __inline __m64 367_mm_srli_pi64 (__m64 __m, int __count) 368{ 369 return (__m64) __builtin_ia32_psrlq (__m, __count); 370} 371 372/* Bit-wise AND the 64-bit values in M1 and M2. */ 373static __inline __m64 374_mm_and_si64 (__m64 __m1, __m64 __m2) 375{ 376 return __builtin_ia32_pand (__m1, __m2); 377} 378 379/* Bit-wise complement the 64-bit value in M1 and bit-wise AND it with the 380 64-bit value in M2. */ 381static __inline __m64 382_mm_andnot_si64 (__m64 __m1, __m64 __m2) 383{ 384 return __builtin_ia32_pandn (__m1, __m2); 385} 386 387/* Bit-wise inclusive OR the 64-bit values in M1 and M2. */ 388static __inline __m64 389_mm_or_si64 (__m64 __m1, __m64 __m2) 390{ 391 return __builtin_ia32_por (__m1, __m2); 392} 393 394/* Bit-wise exclusive OR the 64-bit values in M1 and M2. */ 395static __inline __m64 396_mm_xor_si64 (__m64 __m1, __m64 __m2) 397{ 398 return __builtin_ia32_pxor (__m1, __m2); 399} 400 401/* Compare eight 8-bit values. The result of the comparison is 0xFF if the 402 test is true and zero if false. */ 403static __inline __m64 404_mm_cmpeq_pi8 (__m64 __m1, __m64 __m2) 405{ 406 return (__m64) __builtin_ia32_pcmpeqb ((__v8qi)__m1, (__v8qi)__m2); 407} 408 409static __inline __m64 410_mm_cmpgt_pi8 (__m64 __m1, __m64 __m2) 411{ 412 return (__m64) __builtin_ia32_pcmpgtb ((__v8qi)__m1, (__v8qi)__m2); 413} 414 415/* Compare four 16-bit values. The result of the comparison is 0xFFFF if 416 the test is true and zero if false. */ 417static __inline __m64 418_mm_cmpeq_pi16 (__m64 __m1, __m64 __m2) 419{ 420 return (__m64) __builtin_ia32_pcmpeqw ((__v4hi)__m1, (__v4hi)__m2); 421} 422 423static __inline __m64 424_mm_cmpgt_pi16 (__m64 __m1, __m64 __m2) 425{ 426 return (__m64) __builtin_ia32_pcmpgtw ((__v4hi)__m1, (__v4hi)__m2); 427} 428 429/* Compare two 32-bit values. The result of the comparison is 0xFFFFFFFF if 430 the test is true and zero if false. */ 431static __inline __m64 432_mm_cmpeq_pi32 (__m64 __m1, __m64 __m2) 433{ 434 return (__m64) __builtin_ia32_pcmpeqd ((__v2si)__m1, (__v2si)__m2); 435} 436 437static __inline __m64 438_mm_cmpgt_pi32 (__m64 __m1, __m64 __m2) 439{ 440 return (__m64) __builtin_ia32_pcmpgtd ((__v2si)__m1, (__v2si)__m2); 441} 442 443/* Creates a 64-bit zero. */ 444static __inline __m64 445_mm_setzero_si64 (void) 446{ 447 return __builtin_ia32_mmx_zero (); 448} 449 450/* Creates a vector of two 32-bit values; I0 is least significant. */ 451static __inline __m64 452_mm_set_pi32 (int __i1, int __i0) 453{ 454 union { 455 __m64 __q; 456 struct { 457 unsigned int __i0; 458 unsigned int __i1; 459 } __s; 460 } __u; 461 462 __u.__s.__i0 = __i0; 463 __u.__s.__i1 = __i1; 464 465 return __u.__q; 466} 467 468/* Creates a vector of four 16-bit values; W0 is least significant. */ 469static __inline __m64 470_mm_set_pi16 (short __w3, short __w2, short __w1, short __w0) 471{ 472 unsigned int __i1 = (unsigned short)__w3 << 16 | (unsigned short)__w2; 473 unsigned int __i0 = (unsigned short)__w1 << 16 | (unsigned short)__w0; 474 return _mm_set_pi32 (__i1, __i0); 475 476} 477 478/* Creates a vector of eight 8-bit values; B0 is least significant. */ 479static __inline __m64 480_mm_set_pi8 (char __b7, char __b6, char __b5, char __b4, 481 char __b3, char __b2, char __b1, char __b0) 482{ 483 unsigned int __i1, __i0; 484 485 __i1 = (unsigned char)__b7; 486 __i1 = __i1 << 8 | (unsigned char)__b6; 487 __i1 = __i1 << 8 | (unsigned char)__b5; 488 __i1 = __i1 << 8 | (unsigned char)__b4; 489 490 __i0 = (unsigned char)__b3; 491 __i0 = __i0 << 8 | (unsigned char)__b2; 492 __i0 = __i0 << 8 | (unsigned char)__b1; 493 __i0 = __i0 << 8 | (unsigned char)__b0; 494 495 return _mm_set_pi32 (__i1, __i0); 496} 497 498/* Similar, but with the arguments in reverse order. */ 499static __inline __m64 500_mm_setr_pi32 (int __i0, int __i1) 501{ 502 return _mm_set_pi32 (__i1, __i0); 503} 504 505static __inline __m64 506_mm_setr_pi16 (short __w0, short __w1, short __w2, short __w3) 507{ 508 return _mm_set_pi16 (__w3, __w2, __w1, __w0); 509} 510 511static __inline __m64 512_mm_setr_pi8 (char __b0, char __b1, char __b2, char __b3, 513 char __b4, char __b5, char __b6, char __b7) 514{ 515 return _mm_set_pi8 (__b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0); 516} 517 518/* Creates a vector of two 32-bit values, both elements containing I. */ 519static __inline __m64 520_mm_set1_pi32 (int __i) 521{ 522 return _mm_set_pi32 (__i, __i); 523} 524 525/* Creates a vector of four 16-bit values, all elements containing W. */ 526static __inline __m64 527_mm_set1_pi16 (short __w) 528{ 529 unsigned int __i = (unsigned short)__w << 16 | (unsigned short)__w; 530 return _mm_set1_pi32 (__i); 531} 532 533/* Creates a vector of four 16-bit values, all elements containing B. */ 534static __inline __m64 535_mm_set1_pi8 (char __b) 536{ 537 unsigned int __w = (unsigned char)__b << 8 | (unsigned char)__b; 538 unsigned int __i = __w << 16 | __w; 539 return _mm_set1_pi32 (__i); 540} 541 542#endif /* _MMINTRIN_H_INCLUDED */ 543