xmmintrin.h revision 218893
1/*===---- xmmintrin.h - SSE 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 24#ifndef __XMMINTRIN_H 25#define __XMMINTRIN_H 26 27#ifndef __SSE__ 28#error "SSE instruction set not enabled" 29#else 30 31#include <mmintrin.h> 32 33typedef int __v4si __attribute__((__vector_size__(16))); 34typedef float __v4sf __attribute__((__vector_size__(16))); 35typedef float __m128 __attribute__((__vector_size__(16))); 36 37// This header should only be included in a hosted environment as it depends on 38// a standard library to provide allocation routines. 39#if __STDC_HOSTED__ 40#include <mm_malloc.h> 41#endif 42 43static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 44_mm_add_ss(__m128 a, __m128 b) 45{ 46 a[0] += b[0]; 47 return a; 48} 49 50static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 51_mm_add_ps(__m128 a, __m128 b) 52{ 53 return a + b; 54} 55 56static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 57_mm_sub_ss(__m128 a, __m128 b) 58{ 59 a[0] -= b[0]; 60 return a; 61} 62 63static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 64_mm_sub_ps(__m128 a, __m128 b) 65{ 66 return a - b; 67} 68 69static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 70_mm_mul_ss(__m128 a, __m128 b) 71{ 72 a[0] *= b[0]; 73 return a; 74} 75 76static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 77_mm_mul_ps(__m128 a, __m128 b) 78{ 79 return a * b; 80} 81 82static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 83_mm_div_ss(__m128 a, __m128 b) 84{ 85 a[0] /= b[0]; 86 return a; 87} 88 89static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 90_mm_div_ps(__m128 a, __m128 b) 91{ 92 return a / b; 93} 94 95static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 96_mm_sqrt_ss(__m128 a) 97{ 98 return __builtin_ia32_sqrtss(a); 99} 100 101static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 102_mm_sqrt_ps(__m128 a) 103{ 104 return __builtin_ia32_sqrtps(a); 105} 106 107static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 108_mm_rcp_ss(__m128 a) 109{ 110 return __builtin_ia32_rcpss(a); 111} 112 113static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 114_mm_rcp_ps(__m128 a) 115{ 116 return __builtin_ia32_rcpps(a); 117} 118 119static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 120_mm_rsqrt_ss(__m128 a) 121{ 122 return __builtin_ia32_rsqrtss(a); 123} 124 125static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 126_mm_rsqrt_ps(__m128 a) 127{ 128 return __builtin_ia32_rsqrtps(a); 129} 130 131static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 132_mm_min_ss(__m128 a, __m128 b) 133{ 134 return __builtin_ia32_minss(a, b); 135} 136 137static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 138_mm_min_ps(__m128 a, __m128 b) 139{ 140 return __builtin_ia32_minps(a, b); 141} 142 143static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 144_mm_max_ss(__m128 a, __m128 b) 145{ 146 return __builtin_ia32_maxss(a, b); 147} 148 149static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 150_mm_max_ps(__m128 a, __m128 b) 151{ 152 return __builtin_ia32_maxps(a, b); 153} 154 155static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 156_mm_and_ps(__m128 a, __m128 b) 157{ 158 return (__m128)((__v4si)a & (__v4si)b); 159} 160 161static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 162_mm_andnot_ps(__m128 a, __m128 b) 163{ 164 return (__m128)(~(__v4si)a & (__v4si)b); 165} 166 167static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 168_mm_or_ps(__m128 a, __m128 b) 169{ 170 return (__m128)((__v4si)a | (__v4si)b); 171} 172 173static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 174_mm_xor_ps(__m128 a, __m128 b) 175{ 176 return (__m128)((__v4si)a ^ (__v4si)b); 177} 178 179static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 180_mm_cmpeq_ss(__m128 a, __m128 b) 181{ 182 return (__m128)__builtin_ia32_cmpss(a, b, 0); 183} 184 185static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 186_mm_cmpeq_ps(__m128 a, __m128 b) 187{ 188 return (__m128)__builtin_ia32_cmpps(a, b, 0); 189} 190 191static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 192_mm_cmplt_ss(__m128 a, __m128 b) 193{ 194 return (__m128)__builtin_ia32_cmpss(a, b, 1); 195} 196 197static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 198_mm_cmplt_ps(__m128 a, __m128 b) 199{ 200 return (__m128)__builtin_ia32_cmpps(a, b, 1); 201} 202 203static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 204_mm_cmple_ss(__m128 a, __m128 b) 205{ 206 return (__m128)__builtin_ia32_cmpss(a, b, 2); 207} 208 209static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 210_mm_cmple_ps(__m128 a, __m128 b) 211{ 212 return (__m128)__builtin_ia32_cmpps(a, b, 2); 213} 214 215static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 216_mm_cmpgt_ss(__m128 a, __m128 b) 217{ 218 return (__m128)__builtin_ia32_cmpss(b, a, 1); 219} 220 221static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 222_mm_cmpgt_ps(__m128 a, __m128 b) 223{ 224 return (__m128)__builtin_ia32_cmpps(b, a, 1); 225} 226 227static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 228_mm_cmpge_ss(__m128 a, __m128 b) 229{ 230 return (__m128)__builtin_ia32_cmpss(b, a, 2); 231} 232 233static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 234_mm_cmpge_ps(__m128 a, __m128 b) 235{ 236 return (__m128)__builtin_ia32_cmpps(b, a, 2); 237} 238 239static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 240_mm_cmpneq_ss(__m128 a, __m128 b) 241{ 242 return (__m128)__builtin_ia32_cmpss(a, b, 4); 243} 244 245static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 246_mm_cmpneq_ps(__m128 a, __m128 b) 247{ 248 return (__m128)__builtin_ia32_cmpps(a, b, 4); 249} 250 251static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 252_mm_cmpnlt_ss(__m128 a, __m128 b) 253{ 254 return (__m128)__builtin_ia32_cmpss(a, b, 5); 255} 256 257static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 258_mm_cmpnlt_ps(__m128 a, __m128 b) 259{ 260 return (__m128)__builtin_ia32_cmpps(a, b, 5); 261} 262 263static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 264_mm_cmpnle_ss(__m128 a, __m128 b) 265{ 266 return (__m128)__builtin_ia32_cmpss(a, b, 6); 267} 268 269static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 270_mm_cmpnle_ps(__m128 a, __m128 b) 271{ 272 return (__m128)__builtin_ia32_cmpps(a, b, 6); 273} 274 275static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 276_mm_cmpngt_ss(__m128 a, __m128 b) 277{ 278 return (__m128)__builtin_ia32_cmpss(b, a, 5); 279} 280 281static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 282_mm_cmpngt_ps(__m128 a, __m128 b) 283{ 284 return (__m128)__builtin_ia32_cmpps(b, a, 5); 285} 286 287static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 288_mm_cmpnge_ss(__m128 a, __m128 b) 289{ 290 return (__m128)__builtin_ia32_cmpss(b, a, 6); 291} 292 293static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 294_mm_cmpnge_ps(__m128 a, __m128 b) 295{ 296 return (__m128)__builtin_ia32_cmpps(b, a, 6); 297} 298 299static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 300_mm_cmpord_ss(__m128 a, __m128 b) 301{ 302 return (__m128)__builtin_ia32_cmpss(a, b, 7); 303} 304 305static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 306_mm_cmpord_ps(__m128 a, __m128 b) 307{ 308 return (__m128)__builtin_ia32_cmpps(a, b, 7); 309} 310 311static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 312_mm_cmpunord_ss(__m128 a, __m128 b) 313{ 314 return (__m128)__builtin_ia32_cmpss(a, b, 3); 315} 316 317static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 318_mm_cmpunord_ps(__m128 a, __m128 b) 319{ 320 return (__m128)__builtin_ia32_cmpps(a, b, 3); 321} 322 323static __inline__ int __attribute__((__always_inline__, __nodebug__)) 324_mm_comieq_ss(__m128 a, __m128 b) 325{ 326 return __builtin_ia32_comieq(a, b); 327} 328 329static __inline__ int __attribute__((__always_inline__, __nodebug__)) 330_mm_comilt_ss(__m128 a, __m128 b) 331{ 332 return __builtin_ia32_comilt(a, b); 333} 334 335static __inline__ int __attribute__((__always_inline__, __nodebug__)) 336_mm_comile_ss(__m128 a, __m128 b) 337{ 338 return __builtin_ia32_comile(a, b); 339} 340 341static __inline__ int __attribute__((__always_inline__, __nodebug__)) 342_mm_comigt_ss(__m128 a, __m128 b) 343{ 344 return __builtin_ia32_comigt(a, b); 345} 346 347static __inline__ int __attribute__((__always_inline__, __nodebug__)) 348_mm_comige_ss(__m128 a, __m128 b) 349{ 350 return __builtin_ia32_comige(a, b); 351} 352 353static __inline__ int __attribute__((__always_inline__, __nodebug__)) 354_mm_comineq_ss(__m128 a, __m128 b) 355{ 356 return __builtin_ia32_comineq(a, b); 357} 358 359static __inline__ int __attribute__((__always_inline__, __nodebug__)) 360_mm_ucomieq_ss(__m128 a, __m128 b) 361{ 362 return __builtin_ia32_ucomieq(a, b); 363} 364 365static __inline__ int __attribute__((__always_inline__, __nodebug__)) 366_mm_ucomilt_ss(__m128 a, __m128 b) 367{ 368 return __builtin_ia32_ucomilt(a, b); 369} 370 371static __inline__ int __attribute__((__always_inline__, __nodebug__)) 372_mm_ucomile_ss(__m128 a, __m128 b) 373{ 374 return __builtin_ia32_ucomile(a, b); 375} 376 377static __inline__ int __attribute__((__always_inline__, __nodebug__)) 378_mm_ucomigt_ss(__m128 a, __m128 b) 379{ 380 return __builtin_ia32_ucomigt(a, b); 381} 382 383static __inline__ int __attribute__((__always_inline__, __nodebug__)) 384_mm_ucomige_ss(__m128 a, __m128 b) 385{ 386 return __builtin_ia32_ucomige(a, b); 387} 388 389static __inline__ int __attribute__((__always_inline__, __nodebug__)) 390_mm_ucomineq_ss(__m128 a, __m128 b) 391{ 392 return __builtin_ia32_ucomineq(a, b); 393} 394 395static __inline__ int __attribute__((__always_inline__, __nodebug__)) 396_mm_cvtss_si32(__m128 a) 397{ 398 return __builtin_ia32_cvtss2si(a); 399} 400 401static __inline__ int __attribute__((__always_inline__, __nodebug__)) 402_mm_cvt_ss2si(__m128 a) 403{ 404 return _mm_cvtss_si32(a); 405} 406 407#ifdef __x86_64__ 408 409static __inline__ long long __attribute__((__always_inline__, __nodebug__)) 410_mm_cvtss_si64(__m128 a) 411{ 412 return __builtin_ia32_cvtss2si64(a); 413} 414 415#endif 416 417static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 418_mm_cvtps_pi32(__m128 a) 419{ 420 return (__m64)__builtin_ia32_cvtps2pi(a); 421} 422 423static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 424_mm_cvt_ps2pi(__m128 a) 425{ 426 return _mm_cvtps_pi32(a); 427} 428 429static __inline__ int __attribute__((__always_inline__, __nodebug__)) 430_mm_cvttss_si32(__m128 a) 431{ 432 return a[0]; 433} 434 435static __inline__ int __attribute__((__always_inline__, __nodebug__)) 436_mm_cvtt_ss2si(__m128 a) 437{ 438 return _mm_cvttss_si32(a); 439} 440 441static __inline__ long long __attribute__((__always_inline__, __nodebug__)) 442_mm_cvttss_si64(__m128 a) 443{ 444 return a[0]; 445} 446 447static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 448_mm_cvttps_pi32(__m128 a) 449{ 450 return (__m64)__builtin_ia32_cvttps2pi(a); 451} 452 453static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 454_mm_cvtt_ps2pi(__m128 a) 455{ 456 return _mm_cvttps_pi32(a); 457} 458 459static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 460_mm_cvtsi32_ss(__m128 a, int b) 461{ 462 a[0] = b; 463 return a; 464} 465 466static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 467_mm_cvt_si2ss(__m128 a, int b) 468{ 469 return _mm_cvtsi32_ss(a, b); 470} 471 472#ifdef __x86_64__ 473 474static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 475_mm_cvtsi64_ss(__m128 a, long long b) 476{ 477 a[0] = b; 478 return a; 479} 480 481#endif 482 483static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 484_mm_cvtpi32_ps(__m128 a, __m64 b) 485{ 486 return __builtin_ia32_cvtpi2ps(a, (__v2si)b); 487} 488 489static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 490_mm_cvt_pi2ps(__m128 a, __m64 b) 491{ 492 return _mm_cvtpi32_ps(a, b); 493} 494 495static __inline__ float __attribute__((__always_inline__, __nodebug__)) 496_mm_cvtss_f32(__m128 a) 497{ 498 return a[0]; 499} 500 501static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 502_mm_loadh_pi(__m128 a, const __m64 *p) 503{ 504 __m128 b; 505 b[0] = *(float*)p; 506 b[1] = *((float*)p+1); 507 return __builtin_shufflevector(a, b, 0, 1, 4, 5); 508} 509 510static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 511_mm_loadl_pi(__m128 a, const __m64 *p) 512{ 513 __m128 b; 514 b[0] = *(float*)p; 515 b[1] = *((float*)p+1); 516 return __builtin_shufflevector(a, b, 4, 5, 2, 3); 517} 518 519static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 520_mm_load_ss(const float *p) 521{ 522 return (__m128){ *p, 0, 0, 0 }; 523} 524 525static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 526_mm_load1_ps(const float *p) 527{ 528 return (__m128){ *p, *p, *p, *p }; 529} 530 531#define _mm_load_ps1(p) _mm_load1_ps(p) 532 533static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 534_mm_load_ps(const float *p) 535{ 536 return *(__m128*)p; 537} 538 539static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 540_mm_loadu_ps(const float *p) 541{ 542 return __builtin_ia32_loadups(p); 543} 544 545static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 546_mm_loadr_ps(const float *p) 547{ 548 __m128 a = _mm_load_ps(p); 549 return __builtin_shufflevector(a, a, 3, 2, 1, 0); 550} 551 552static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 553_mm_set_ss(float w) 554{ 555 return (__m128){ w, 0, 0, 0 }; 556} 557 558static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 559_mm_set1_ps(float w) 560{ 561 return (__m128){ w, w, w, w }; 562} 563 564// Microsoft specific. 565static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 566_mm_set_ps1(float w) 567{ 568 return _mm_set1_ps(w); 569} 570 571static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 572_mm_set_ps(float z, float y, float x, float w) 573{ 574 return (__m128){ w, x, y, z }; 575} 576 577static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 578_mm_setr_ps(float z, float y, float x, float w) 579{ 580 return (__m128){ z, y, x, w }; 581} 582 583static __inline__ __m128 __attribute__((__always_inline__)) 584_mm_setzero_ps(void) 585{ 586 return (__m128){ 0, 0, 0, 0 }; 587} 588 589static __inline__ void __attribute__((__always_inline__)) 590_mm_storeh_pi(__m64 *p, __m128 a) 591{ 592 __builtin_ia32_storehps((__v2si *)p, a); 593} 594 595static __inline__ void __attribute__((__always_inline__)) 596_mm_storel_pi(__m64 *p, __m128 a) 597{ 598 __builtin_ia32_storelps((__v2si *)p, a); 599} 600 601static __inline__ void __attribute__((__always_inline__)) 602_mm_store_ss(float *p, __m128 a) 603{ 604 *p = a[0]; 605} 606 607static __inline__ void __attribute__((__always_inline__, __nodebug__)) 608_mm_storeu_ps(float *p, __m128 a) 609{ 610 __builtin_ia32_storeups(p, a); 611} 612 613static __inline__ void __attribute__((__always_inline__, __nodebug__)) 614_mm_store1_ps(float *p, __m128 a) 615{ 616 a = __builtin_shufflevector(a, a, 0, 0, 0, 0); 617 _mm_storeu_ps(p, a); 618} 619 620static __inline__ void __attribute__((__always_inline__, __nodebug__)) 621_mm_store_ps1(float *p, __m128 a) 622{ 623 return _mm_store1_ps(p, a); 624} 625 626static __inline__ void __attribute__((__always_inline__, __nodebug__)) 627_mm_store_ps(float *p, __m128 a) 628{ 629 *(__m128 *)p = a; 630} 631 632static __inline__ void __attribute__((__always_inline__, __nodebug__)) 633_mm_storer_ps(float *p, __m128 a) 634{ 635 a = __builtin_shufflevector(a, a, 3, 2, 1, 0); 636 _mm_store_ps(p, a); 637} 638 639#define _MM_HINT_T0 3 640#define _MM_HINT_T1 2 641#define _MM_HINT_T2 1 642#define _MM_HINT_NTA 0 643 644/* FIXME: We have to #define this because "sel" must be a constant integer, and 645 Sema doesn't do any form of constant propagation yet. */ 646 647#define _mm_prefetch(a, sel) (__builtin_prefetch((void *)(a), 0, sel)) 648 649static __inline__ void __attribute__((__always_inline__, __nodebug__)) 650_mm_stream_pi(__m64 *p, __m64 a) 651{ 652 __builtin_ia32_movntq(p, a); 653} 654 655static __inline__ void __attribute__((__always_inline__, __nodebug__)) 656_mm_stream_ps(float *p, __m128 a) 657{ 658 __builtin_ia32_movntps(p, a); 659} 660 661static __inline__ void __attribute__((__always_inline__, __nodebug__)) 662_mm_sfence(void) 663{ 664 __builtin_ia32_sfence(); 665} 666 667static __inline__ int __attribute__((__always_inline__, __nodebug__)) 668_mm_extract_pi16(__m64 a, int n) 669{ 670 __v4hi b = (__v4hi)a; 671 return (unsigned short)b[n & 3]; 672} 673 674static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 675_mm_insert_pi16(__m64 a, int d, int n) 676{ 677 __v4hi b = (__v4hi)a; 678 b[n & 3] = d; 679 return (__m64)b; 680} 681 682static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 683_mm_max_pi16(__m64 a, __m64 b) 684{ 685 return (__m64)__builtin_ia32_pmaxsw((__v4hi)a, (__v4hi)b); 686} 687 688static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 689_mm_max_pu8(__m64 a, __m64 b) 690{ 691 return (__m64)__builtin_ia32_pmaxub((__v8qi)a, (__v8qi)b); 692} 693 694static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 695_mm_min_pi16(__m64 a, __m64 b) 696{ 697 return (__m64)__builtin_ia32_pminsw((__v4hi)a, (__v4hi)b); 698} 699 700static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 701_mm_min_pu8(__m64 a, __m64 b) 702{ 703 return (__m64)__builtin_ia32_pminub((__v8qi)a, (__v8qi)b); 704} 705 706static __inline__ int __attribute__((__always_inline__, __nodebug__)) 707_mm_movemask_pi8(__m64 a) 708{ 709 return __builtin_ia32_pmovmskb((__v8qi)a); 710} 711 712static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 713_mm_mulhi_pu16(__m64 a, __m64 b) 714{ 715 return (__m64)__builtin_ia32_pmulhuw((__v4hi)a, (__v4hi)b); 716} 717 718#define _mm_shuffle_pi16(a, n) \ 719 ((__m64)__builtin_ia32_pshufw(a, n)) 720 721static __inline__ void __attribute__((__always_inline__, __nodebug__)) 722_mm_maskmove_si64(__m64 d, __m64 n, char *p) 723{ 724 __builtin_ia32_maskmovq((__v8qi)d, (__v8qi)n, p); 725} 726 727static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 728_mm_avg_pu8(__m64 a, __m64 b) 729{ 730 return (__m64)__builtin_ia32_pavgb((__v8qi)a, (__v8qi)b); 731} 732 733static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 734_mm_avg_pu16(__m64 a, __m64 b) 735{ 736 return (__m64)__builtin_ia32_pavgw((__v4hi)a, (__v4hi)b); 737} 738 739static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 740_mm_sad_pu8(__m64 a, __m64 b) 741{ 742 return (__m64)__builtin_ia32_psadbw((__v8qi)a, (__v8qi)b); 743} 744 745static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) 746_mm_getcsr(void) 747{ 748 return __builtin_ia32_stmxcsr(); 749} 750 751static __inline__ void __attribute__((__always_inline__, __nodebug__)) 752_mm_setcsr(unsigned int i) 753{ 754 __builtin_ia32_ldmxcsr(i); 755} 756 757#define _mm_shuffle_ps(a, b, mask) \ 758 (__builtin_shufflevector((__v4sf)(a), (__v4sf)(b), \ 759 (mask) & 0x3, ((mask) & 0xc) >> 2, \ 760 (((mask) & 0x30) >> 4) + 4, \ 761 (((mask) & 0xc0) >> 6) + 4)) 762 763static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 764_mm_unpackhi_ps(__m128 a, __m128 b) 765{ 766 return __builtin_shufflevector(a, b, 2, 6, 3, 7); 767} 768 769static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 770_mm_unpacklo_ps(__m128 a, __m128 b) 771{ 772 return __builtin_shufflevector(a, b, 0, 4, 1, 5); 773} 774 775static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 776_mm_move_ss(__m128 a, __m128 b) 777{ 778 return __builtin_shufflevector(a, b, 4, 1, 2, 3); 779} 780 781static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 782_mm_movehl_ps(__m128 a, __m128 b) 783{ 784 return __builtin_shufflevector(a, b, 6, 7, 2, 3); 785} 786 787static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 788_mm_movelh_ps(__m128 a, __m128 b) 789{ 790 return __builtin_shufflevector(a, b, 0, 1, 4, 5); 791} 792 793static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 794_mm_cvtpi16_ps(__m64 a) 795{ 796 __m64 b, c; 797 __m128 r; 798 799 b = _mm_setzero_si64(); 800 b = _mm_cmpgt_pi16(b, a); 801 c = _mm_unpackhi_pi16(a, b); 802 r = _mm_setzero_ps(); 803 r = _mm_cvtpi32_ps(r, c); 804 r = _mm_movelh_ps(r, r); 805 c = _mm_unpacklo_pi16(a, b); 806 r = _mm_cvtpi32_ps(r, c); 807 808 return r; 809} 810 811static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 812_mm_cvtpu16_ps(__m64 a) 813{ 814 __m64 b, c; 815 __m128 r; 816 817 b = _mm_setzero_si64(); 818 c = _mm_unpackhi_pi16(a, b); 819 r = _mm_setzero_ps(); 820 r = _mm_cvtpi32_ps(r, c); 821 r = _mm_movelh_ps(r, r); 822 c = _mm_unpacklo_pi16(a, b); 823 r = _mm_cvtpi32_ps(r, c); 824 825 return r; 826} 827 828static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 829_mm_cvtpi8_ps(__m64 a) 830{ 831 __m64 b; 832 833 b = _mm_setzero_si64(); 834 b = _mm_cmpgt_pi8(b, a); 835 b = _mm_unpacklo_pi8(a, b); 836 837 return _mm_cvtpi16_ps(b); 838} 839 840static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 841_mm_cvtpu8_ps(__m64 a) 842{ 843 __m64 b; 844 845 b = _mm_setzero_si64(); 846 b = _mm_unpacklo_pi8(a, b); 847 848 return _mm_cvtpi16_ps(b); 849} 850 851static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 852_mm_cvtpi32x2_ps(__m64 a, __m64 b) 853{ 854 __m128 c; 855 856 c = _mm_setzero_ps(); 857 c = _mm_cvtpi32_ps(c, b); 858 c = _mm_movelh_ps(c, c); 859 860 return _mm_cvtpi32_ps(c, a); 861} 862 863static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 864_mm_cvtps_pi16(__m128 a) 865{ 866 __m64 b, c; 867 868 b = _mm_cvtps_pi32(a); 869 a = _mm_movehl_ps(a, a); 870 c = _mm_cvtps_pi32(a); 871 872 return _mm_packs_pi16(b, c); 873} 874 875static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) 876_mm_cvtps_pi8(__m128 a) 877{ 878 __m64 b, c; 879 880 b = _mm_cvtps_pi16(a); 881 c = _mm_setzero_si64(); 882 883 return _mm_packs_pi16(b, c); 884} 885 886static __inline__ int __attribute__((__always_inline__, __nodebug__)) 887_mm_movemask_ps(__m128 a) 888{ 889 return __builtin_ia32_movmskps(a); 890} 891 892#define _MM_SHUFFLE(z, y, x, w) (((z) << 6) | ((y) << 4) | ((x) << 2) | (w)) 893 894#define _MM_EXCEPT_INVALID (0x0001) 895#define _MM_EXCEPT_DENORM (0x0002) 896#define _MM_EXCEPT_DIV_ZERO (0x0004) 897#define _MM_EXCEPT_OVERFLOW (0x0008) 898#define _MM_EXCEPT_UNDERFLOW (0x0010) 899#define _MM_EXCEPT_INEXACT (0x0020) 900#define _MM_EXCEPT_MASK (0x003f) 901 902#define _MM_MASK_INVALID (0x0080) 903#define _MM_MASK_DENORM (0x0100) 904#define _MM_MASK_DIV_ZERO (0x0200) 905#define _MM_MASK_OVERFLOW (0x0400) 906#define _MM_MASK_UNDERFLOW (0x0800) 907#define _MM_MASK_INEXACT (0x1000) 908#define _MM_MASK_MASK (0x1f80) 909 910#define _MM_ROUND_NEAREST (0x0000) 911#define _MM_ROUND_DOWN (0x2000) 912#define _MM_ROUND_UP (0x4000) 913#define _MM_ROUND_TOWARD_ZERO (0x6000) 914#define _MM_ROUND_MASK (0x6000) 915 916#define _MM_FLUSH_ZERO_MASK (0x8000) 917#define _MM_FLUSH_ZERO_ON (0x8000) 918#define _MM_FLUSH_ZERO_OFF (0x8000) 919 920#define _MM_GET_EXCEPTION_MASK() (_mm_getcsr() & _MM_MASK_MASK) 921#define _MM_GET_EXCEPTION_STATE() (_mm_getcsr() & _MM_EXCEPT_MASK) 922#define _MM_GET_FLUSH_ZERO_MODE() (_mm_getcsr() & _MM_FLUSH_ZERO_MASK) 923#define _MM_GET_ROUNDING_MODE() (_mm_getcsr() & _MM_ROUND_MASK) 924 925#define _MM_SET_EXCEPTION_MASK(x) (_mm_setcsr((_mm_getcsr() & ~_MM_MASK_MASK) | (x))) 926#define _MM_SET_EXCEPTION_STATE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_EXCEPT_MASK) | (x))) 927#define _MM_SET_FLUSH_ZERO_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_FLUSH_ZERO_MASK) | (x))) 928#define _MM_SET_ROUNDING_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_ROUND_MASK) | (x))) 929 930#define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) \ 931do { \ 932 __m128 tmp3, tmp2, tmp1, tmp0; \ 933 tmp0 = _mm_unpacklo_ps((row0), (row1)); \ 934 tmp2 = _mm_unpacklo_ps((row2), (row3)); \ 935 tmp1 = _mm_unpackhi_ps((row0), (row1)); \ 936 tmp3 = _mm_unpackhi_ps((row2), (row3)); \ 937 (row0) = _mm_movelh_ps(tmp0, tmp2); \ 938 (row1) = _mm_movehl_ps(tmp2, tmp0); \ 939 (row2) = _mm_movelh_ps(tmp1, tmp3); \ 940 (row3) = _mm_movehl_ps(tmp3, tmp1); \ 941} while (0) 942 943/* Aliases for compatibility. */ 944#define _m_pextrw _mm_extract_pi16 945#define _m_pinsrw _mm_insert_pi16 946#define _m_pmaxsw _mm_max_pi16 947#define _m_pmaxub _mm_max_pu8 948#define _m_pminsw _mm_min_pi16 949#define _m_pminub _mm_min_pu8 950#define _m_pmovmskb _mm_movemask_pi8 951#define _m_pmulhuw _mm_mulhi_pu16 952#define _m_pshufw _mm_shuffle_pi16 953#define _m_maskmovq _mm_maskmove_si64 954#define _m_pavgb _mm_avg_pu8 955#define _m_pavgw _mm_avg_pu16 956#define _m_psadbw _mm_sad_pu8 957#define _m_ _mm_ 958#define _m_ _mm_ 959 960/* Ugly hack for backwards-compatibility (compatible with gcc) */ 961#ifdef __SSE2__ 962#include <emmintrin.h> 963#endif 964 965#endif /* __SSE__ */ 966 967#endif /* __XMMINTRIN_H */ 968