mingw/lib/gcc/mingw32/4.3.3/include/smmintrin.h

   1 /* Copyright (C) 2007 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 2, 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    You should have received a copy of the GNU General Public License
  16    along with GCC; see the file COPYING.  If not, write to
  17    the Free Software Foundation, 59 Temple Place - Suite 330,
  18    Boston, MA 02111-1307, USA.  */
  19
  20 /* As a special exception, if you include this header file into source
  21    files compiled by GCC, this header file does not by itself cause
  22    the resulting executable to be covered by the GNU General Public
  23    License.  This exception does not however invalidate any other
  24    reasons why the executable file might be covered by the GNU General
  25    Public License.  */
  26
  27 /* Implemented from the specification included in the Intel C++ Compiler
  28    User Guide and Reference, version 10.0.  */
  29
  30 #ifndef _SMMINTRIN_H_INCLUDED
  31 #define _SMMINTRIN_H_INCLUDED
  32
  33 #ifndef __SSE4_1__
  34 # error "SSE4.1 instruction set not enabled"
  35 #else
  36
  37 /* We need definitions from the SSSE3, SSE3, SSE2 and SSE header
  38    files.  */
  39 #include <tmmintrin.h>
  40 #include <mmintrin-common.h>
  41
  42 /* SSE4.1 */
  43
  44 /* Integer blend instructions - select data from 2 sources using
  45    constant/variable mask.  */
  46
  47 #ifdef __OPTIMIZE__
  48 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  49 _mm_blend_epi16 (__m128i __X, __m128i __Y, const int __M)
  50 {
  51   return (__m128i) __builtin_ia32_pblendw128 ((__v8hi)__X,
  52                                               (__v8hi)__Y,
  53                                               __M);
  54 }
  55 #else
  56 #define _mm_blend_epi16(X, Y, M)                                        \
  57   ((__m128i) __builtin_ia32_pblendw128 ((__v8hi)(__m128i)(X),           \
  58                                         (__v8hi)(__m128i)(Y), (int)(M)))
  59 #endif
  60
  61 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  62 _mm_blendv_epi8 (__m128i __X, __m128i __Y, __m128i __M)
  63 {
  64   return (__m128i) __builtin_ia32_pblendvb128 ((__v16qi)__X,
  65                                                (__v16qi)__Y,
  66                                                (__v16qi)__M);
  67 }
  68
  69 /* Single precision floating point blend instructions - select data
  70    from 2 sources using constant/variable mask.  */
  71
  72 #ifdef __OPTIMIZE__
  73 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  74 _mm_blend_ps (__m128 __X, __m128 __Y, const int __M)
  75 {
  76   return (__m128) __builtin_ia32_blendps ((__v4sf)__X,
  77                                           (__v4sf)__Y,
  78                                           __M);
  79 }
  80 #else
  81 #define _mm_blend_ps(X, Y, M)                                           \
  82   ((__m128) __builtin_ia32_blendps ((__v4sf)(__m128)(X),                \
  83                                     (__v4sf)(__m128)(Y), (int)(M)))
  84 #endif
  85
  86 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  87 _mm_blendv_ps (__m128 __X, __m128 __Y, __m128 __M)
  88 {
  89   return (__m128) __builtin_ia32_blendvps ((__v4sf)__X,
  90                                            (__v4sf)__Y,
  91                                            (__v4sf)__M);
  92 }
  93
  94 /* Double precision floating point blend instructions - select data
  95    from 2 sources using constant/variable mask.  */
  96
  97 #ifdef __OPTIMIZE__
  98 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  99 _mm_blend_pd (__m128d __X, __m128d __Y, const int __M)
 100 {
 101   return (__m128d) __builtin_ia32_blendpd ((__v2df)__X,
 102                                            (__v2df)__Y,
 103                                            __M);
 104 }
 105 #else
 106 #define _mm_blend_pd(X, Y, M)                                           \
 107   ((__m128d) __builtin_ia32_blendpd ((__v2df)(__m128d)(X),              \
 108                                      (__v2df)(__m128d)(Y), (int)(M)))
 109 #endif
 110
 111 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 112 _mm_blendv_pd (__m128d __X, __m128d __Y, __m128d __M)
 113 {
 114   return (__m128d) __builtin_ia32_blendvpd ((__v2df)__X,
 115                                             (__v2df)__Y,
 116                                             (__v2df)__M);
 117 }
 118
 119 /* Dot product instructions with mask-defined summing and zeroing parts
 120    of result.  */
 121
 122 #ifdef __OPTIMIZE__
 123 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 124 _mm_dp_ps (__m128 __X, __m128 __Y, const int __M)
 125 {
 126   return (__m128) __builtin_ia32_dpps ((__v4sf)__X,
 127                                        (__v4sf)__Y,
 128                                        __M);
 129 }
 130
 131 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 132 _mm_dp_pd (__m128d __X, __m128d __Y, const int __M)
 133 {
 134   return (__m128d) __builtin_ia32_dppd ((__v2df)__X,
 135                                         (__v2df)__Y,
 136                                         __M);
 137 }
 138 #else
 139 #define _mm_dp_ps(X, Y, M)                                              \
 140   ((__m128) __builtin_ia32_dpps ((__v4sf)(__m128)(X),                   \
 141                                  (__v4sf)(__m128)(Y), (int)(M)))
 142
 143 #define _mm_dp_pd(X, Y, M)                                              \
 144   ((__m128d) __builtin_ia32_dppd ((__v2df)(__m128d)(X),                 \
 145                                   (__v2df)(__m128d)(Y), (int)(M)))
 146 #endif
 147
 148 /* Packed integer 64-bit comparison, zeroing or filling with ones
 149    corresponding parts of result.  */
 150 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 151 _mm_cmpeq_epi64 (__m128i __X, __m128i __Y)
 152 {
 153   return (__m128i) __builtin_ia32_pcmpeqq ((__v2di)__X, (__v2di)__Y);
 154 }
 155
 156 /*  Min/max packed integer instructions.  */
 157
 158 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 159 _mm_min_epi8 (__m128i __X, __m128i __Y)
 160 {
 161   return (__m128i) __builtin_ia32_pminsb128 ((__v16qi)__X, (__v16qi)__Y);
 162 }
 163
 164 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 165 _mm_max_epi8 (__m128i __X, __m128i __Y)
 166 {
 167   return (__m128i) __builtin_ia32_pmaxsb128 ((__v16qi)__X, (__v16qi)__Y);
 168 }
 169
 170 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 171 _mm_min_epu16 (__m128i __X, __m128i __Y)
 172 {
 173   return (__m128i) __builtin_ia32_pminuw128 ((__v8hi)__X, (__v8hi)__Y);
 174 }
 175
 176 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 177 _mm_max_epu16 (__m128i __X, __m128i __Y)
 178 {
 179   return (__m128i) __builtin_ia32_pmaxuw128 ((__v8hi)__X, (__v8hi)__Y);
 180 }
 181
 182 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 183 _mm_min_epi32 (__m128i __X, __m128i __Y)
 184 {
 185   return (__m128i) __builtin_ia32_pminsd128 ((__v4si)__X, (__v4si)__Y);
 186 }
 187
 188 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 189 _mm_max_epi32 (__m128i __X, __m128i __Y)
 190 {
 191   return (__m128i) __builtin_ia32_pmaxsd128 ((__v4si)__X, (__v4si)__Y);
 192 }
 193
 194 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 195 _mm_min_epu32 (__m128i __X, __m128i __Y)
 196 {
 197   return (__m128i) __builtin_ia32_pminud128 ((__v4si)__X, (__v4si)__Y);
 198 }
 199
 200 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 201 _mm_max_epu32 (__m128i __X, __m128i __Y)
 202 {
 203   return (__m128i) __builtin_ia32_pmaxud128 ((__v4si)__X, (__v4si)__Y);
 204 }
 205
 206 /* Packed integer 32-bit multiplication with truncation of upper
 207    halves of results.  */
 208 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 209 _mm_mullo_epi32 (__m128i __X, __m128i __Y)
 210 {
 211   return (__m128i) __builtin_ia32_pmulld128 ((__v4si)__X, (__v4si)__Y);
 212 }
 213
 214 /* Packed integer 32-bit multiplication of 2 pairs of operands
 215    with two 64-bit results.  */
 216 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 217 _mm_mul_epi32 (__m128i __X, __m128i __Y)
 218 {
 219   return (__m128i) __builtin_ia32_pmuldq128 ((__v4si)__X, (__v4si)__Y);
 220 }
 221
 222 /* Insert single precision float into packed single precision array
 223    element selected by index N.  The bits [7-6] of N define S
 224    index, the bits [5-4] define D index, and bits [3-0] define
 225    zeroing mask for D.  */
 226
 227 #ifdef __OPTIMIZE__
 228 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 229 _mm_insert_ps (__m128 __D, __m128 __S, const int __N)
 230 {
 231   return (__m128) __builtin_ia32_insertps128 ((__v4sf)__D,
 232                                               (__v4sf)__S,
 233                                               __N);
 234 }
 235 #else
 236 #define _mm_insert_ps(D, S, N)                                          \
 237   ((__m128) __builtin_ia32_insertps128 ((__v4sf)(__m128)(D),            \
 238                                         (__v4sf)(__m128)(S), (int)(N)))
 239 #endif
 240
 241 /* Helper macro to create the N value for _mm_insert_ps.  */
 242 #define _MM_MK_INSERTPS_NDX(S, D, M) (((S) << 6) | ((D) << 4) | (M))
 243
 244 /* Extract binary representation of single precision float from packed
 245    single precision array element of X selected by index N.  */
 246
 247 #ifdef __OPTIMIZE__
 248 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 249 _mm_extract_ps (__m128 __X, const int __N)
 250 {
 251   union { int i; float f; } __tmp;
 252   __tmp.f = __builtin_ia32_vec_ext_v4sf ((__v4sf)__X, __N);
 253   return __tmp.i;
 254 }
 255 #else
 256 #define _mm_extract_ps(X, N)                                            \
 257   (__extension__                                                        \
 258    ({                                                                   \
 259      union { int i; float f; } __tmp;                                   \
 260      __tmp.f = __builtin_ia32_vec_ext_v4sf ((__v4sf)(__m128)(X), (int)(N)); \
 261      __tmp.i;                                                           \
 262    }))
 263 #endif
 264
 265 /* Extract binary representation of single precision float into
 266    D from packed single precision array element of S selected
 267    by index N.  */
 268 #define _MM_EXTRACT_FLOAT(D, S, N) \
 269   { (D) = __builtin_ia32_vec_ext_v4sf ((__v4sf)(S), (N)); }
 270
 271 /* Extract specified single precision float element into the lower
 272    part of __m128.  */
 273 #define _MM_PICK_OUT_PS(X, N)                           \
 274   _mm_insert_ps (_mm_setzero_ps (), (X),                \
 275                  _MM_MK_INSERTPS_NDX ((N), 0, 0x0e))
 276
 277 /* Insert integer, S, into packed integer array element of D
 278    selected by index N.  */
 279
 280 #ifdef __OPTIMIZE__
 281 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 282 _mm_insert_epi8 (__m128i __D, int __S, const int __N)
 283 {
 284   return (__m128i) __builtin_ia32_vec_set_v16qi ((__v16qi)__D,
 285                                                  __S, __N);
 286 }
 287
 288 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 289 _mm_insert_epi32 (__m128i __D, int __S, const int __N)
 290 {
 291   return (__m128i) __builtin_ia32_vec_set_v4si ((__v4si)__D,
 292                                                  __S, __N);
 293 }
 294
 295 #ifdef __x86_64__
 296 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 297 _mm_insert_epi64 (__m128i __D, long long __S, const int __N)
 298 {
 299   return (__m128i) __builtin_ia32_vec_set_v2di ((__v2di)__D,
 300                                                  __S, __N);
 301 }
 302 #endif
 303 #else
 304 #define _mm_insert_epi8(D, S, N)                                        \
 305   ((__m128i) __builtin_ia32_vec_set_v16qi ((__v16qi)(__m128i)(D),       \
 306                                            (int)(S), (int)(N)))
 307
 308 #define _mm_insert_epi32(D, S, N)                               \
 309   ((__m128i) __builtin_ia32_vec_set_v4si ((__v4si)(__m128i)(D), \
 310                                           (int)(S), (int)(N)))
 311
 312 #ifdef __x86_64__
 313 #define _mm_insert_epi64(D, S, N)                                       \
 314   ((__m128i) __builtin_ia32_vec_set_v2di ((__v2di)(__m128i)(D),         \
 315                                           (long long)(S), (int)(N)))
 316 #endif
 317 #endif
 318
 319 /* Extract integer from packed integer array element of X selected by
 320    index N.  */
 321
 322 #ifdef __OPTIMIZE__
 323 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 324 _mm_extract_epi8 (__m128i __X, const int __N)
 325 {
 326    return __builtin_ia32_vec_ext_v16qi ((__v16qi)__X, __N);
 327 }
 328
 329 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 330 _mm_extract_epi32 (__m128i __X, const int __N)
 331 {
 332    return __builtin_ia32_vec_ext_v4si ((__v4si)__X, __N);
 333 }
 334
 335 #ifdef __x86_64__
 336 extern __inline long long  __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 337 _mm_extract_epi64 (__m128i __X, const int __N)
 338 {
 339   return __builtin_ia32_vec_ext_v2di ((__v2di)__X, __N);
 340 }
 341 #endif
 342 #else
 343 #define _mm_extract_epi8(X, N) \
 344   ((int) __builtin_ia32_vec_ext_v16qi ((__v16qi)(__m128i)(X), (int)(N)))
 345 #define _mm_extract_epi32(X, N) \
 346   ((int) __builtin_ia32_vec_ext_v4si ((__v4si)(__m128i)(X), (int)(N)))
 347
 348 #ifdef __x86_64__
 349 #define _mm_extract_epi64(X, N) \
 350   ((long long) __builtin_ia32_vec_ext_v2di ((__v2di)(__m128i)(X), (int)(N)))
 351 #endif
 352 #endif
 353
 354 /* Return horizontal packed word minimum and its index in bits [15:0]
 355    and bits [18:16] respectively.  */
 356 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 357 _mm_minpos_epu16 (__m128i __X)
 358 {
 359   return (__m128i) __builtin_ia32_phminposuw128 ((__v8hi)__X);
 360 }
 361
 362 /* Packed integer sign-extension.  */
 363
 364 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 365 _mm_cvtepi8_epi32 (__m128i __X)
 366 {
 367   return (__m128i) __builtin_ia32_pmovsxbd128 ((__v16qi)__X);
 368 }
 369
 370 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 371 _mm_cvtepi16_epi32 (__m128i __X)
 372 {
 373   return (__m128i) __builtin_ia32_pmovsxwd128 ((__v8hi)__X);
 374 }
 375
 376 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 377 _mm_cvtepi8_epi64 (__m128i __X)
 378 {
 379   return (__m128i) __builtin_ia32_pmovsxbq128 ((__v16qi)__X);
 380 }
 381
 382 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 383 _mm_cvtepi32_epi64 (__m128i __X)
 384 {
 385   return (__m128i) __builtin_ia32_pmovsxdq128 ((__v4si)__X);
 386 }
 387
 388 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 389 _mm_cvtepi16_epi64 (__m128i __X)
 390 {
 391   return (__m128i) __builtin_ia32_pmovsxwq128 ((__v8hi)__X);
 392 }
 393
 394 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 395 _mm_cvtepi8_epi16 (__m128i __X)
 396 {
 397   return (__m128i) __builtin_ia32_pmovsxbw128 ((__v16qi)__X);
 398 }
 399
 400 /* Packed integer zero-extension. */
 401
 402 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 403 _mm_cvtepu8_epi32 (__m128i __X)
 404 {
 405   return (__m128i) __builtin_ia32_pmovzxbd128 ((__v16qi)__X);
 406 }
 407
 408 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 409 _mm_cvtepu16_epi32 (__m128i __X)
 410 {
 411   return (__m128i) __builtin_ia32_pmovzxwd128 ((__v8hi)__X);
 412 }
 413
 414 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 415 _mm_cvtepu8_epi64 (__m128i __X)
 416 {
 417   return (__m128i) __builtin_ia32_pmovzxbq128 ((__v16qi)__X);
 418 }
 419
 420 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 421 _mm_cvtepu32_epi64 (__m128i __X)
 422 {
 423   return (__m128i) __builtin_ia32_pmovzxdq128 ((__v4si)__X);
 424 }
 425
 426 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 427 _mm_cvtepu16_epi64 (__m128i __X)
 428 {
 429   return (__m128i) __builtin_ia32_pmovzxwq128 ((__v8hi)__X);
 430 }
 431
 432 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 433 _mm_cvtepu8_epi16 (__m128i __X)
 434 {
 435   return (__m128i) __builtin_ia32_pmovzxbw128 ((__v16qi)__X);
 436 }
 437
 438 /* Pack 8 double words from 2 operands into 8 words of result with
 439    unsigned saturation. */
 440 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 441 _mm_packus_epi32 (__m128i __X, __m128i __Y)
 442 {
 443   return (__m128i) __builtin_ia32_packusdw128 ((__v4si)__X, (__v4si)__Y);
 444 }
 445
 446 /* Sum absolute 8-bit integer difference of adjacent groups of 4
 447    byte integers in the first 2 operands.  Starting offsets within
 448    operands are determined by the 3rd mask operand.  */
 449
 450 #ifdef __OPTIMIZE__
 451 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 452 _mm_mpsadbw_epu8 (__m128i __X, __m128i __Y, const int __M)
 453 {
 454   return (__m128i) __builtin_ia32_mpsadbw128 ((__v16qi)__X,
 455                                               (__v16qi)__Y, __M);
 456 }
 457 #else
 458 #define _mm_mpsadbw_epu8(X, Y, M)                                       \
 459   ((__m128i) __builtin_ia32_mpsadbw128 ((__v16qi)(__m128i)(X),          \
 460                                         (__v16qi)(__m128i)(Y), (int)(M)))
 461 #endif
 462
 463 /* Load double quadword using non-temporal aligned hint.  */
 464 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 465 _mm_stream_load_si128 (__m128i *__X)
 466 {
 467   return (__m128i) __builtin_ia32_movntdqa ((__v2di *) __X);
 468 }
 469
 470 #ifdef __SSE4_2__
 471
 472 /* These macros specify the source data format.  */
 473 #define _SIDD_UBYTE_OPS                 0x00
 474 #define _SIDD_UWORD_OPS                 0x01
 475 #define _SIDD_SBYTE_OPS                 0x02
 476 #define _SIDD_SWORD_OPS                 0x03
 477
 478 /* These macros specify the comparison operation.  */
 479 #define _SIDD_CMP_EQUAL_ANY             0x00
 480 #define _SIDD_CMP_RANGES                0x04
 481 #define _SIDD_CMP_EQUAL_EACH            0x08
 482 #define _SIDD_CMP_EQUAL_ORDERED         0x0c
 483
 484 /* These macros specify the the polarity.  */
 485 #define _SIDD_POSITIVE_POLARITY         0x00
 486 #define _SIDD_NEGATIVE_POLARITY         0x10
 487 #define _SIDD_MASKED_POSITIVE_POLARITY  0x20
 488 #define _SIDD_MASKED_NEGATIVE_POLARITY  0x30
 489
 490 /* These macros specify the output selection in _mm_cmpXstri ().  */
 491 #define _SIDD_LEAST_SIGNIFICANT         0x00
 492 #define _SIDD_MOST_SIGNIFICANT          0x40
 493
 494 /* These macros specify the output selection in _mm_cmpXstrm ().  */
 495 #define _SIDD_BIT_MASK                  0x00
 496 #define _SIDD_UNIT_MASK                 0x40
 497
 498 /* Intrinsics for text/string processing.  */
 499
 500 #ifdef __OPTIMIZE__
 501 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 502 _mm_cmpistrm (__m128i __X, __m128i __Y, const int __M)
 503 {
 504   return (__m128i) __builtin_ia32_pcmpistrm128 ((__v16qi)__X,
 505                                                 (__v16qi)__Y,
 506                                                 __M);
 507 }
 508
 509 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 510 _mm_cmpistri (__m128i __X, __m128i __Y, const int __M)
 511 {
 512   return __builtin_ia32_pcmpistri128 ((__v16qi)__X,
 513                                       (__v16qi)__Y,
 514                                       __M);
 515 }
 516
 517 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 518 _mm_cmpestrm (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 519 {
 520   return (__m128i) __builtin_ia32_pcmpestrm128 ((__v16qi)__X, __LX,
 521                                                 (__v16qi)__Y, __LY,
 522                                                 __M);
 523 }
 524
 525 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 526 _mm_cmpestri (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 527 {
 528   return __builtin_ia32_pcmpestri128 ((__v16qi)__X, __LX,
 529                                       (__v16qi)__Y, __LY,
 530                                       __M);
 531 }
 532 #else
 533 #define _mm_cmpistrm(X, Y, M)                                           \
 534   ((__m128i) __builtin_ia32_pcmpistrm128 ((__v16qi)(__m128i)(X),        \
 535                                           (__v16qi)(__m128i)(Y), (int)(M)))
 536 #define _mm_cmpistri(X, Y, M)                                           \
 537   ((int) __builtin_ia32_pcmpistri128 ((__v16qi)(__m128i)(X),            \
 538                                       (__v16qi)(__m128i)(Y), (int)(M)))
 539
 540 #define _mm_cmpestrm(X, LX, Y, LY, M)                                   \
 541   ((__m128i) __builtin_ia32_pcmpestrm128 ((__v16qi)(__m128i)(X),        \
 542                                           (int)(LX), (__v16qi)(__m128i)(Y), \
 543                                           (int)(LY), (int)(M)))
 544 #define _mm_cmpestri(X, LX, Y, LY, M)                                   \
 545   ((int) __builtin_ia32_pcmpestri128 ((__v16qi)(__m128i)(X), (int)(LX), \
 546                                       (__v16qi)(__m128i)(Y), (int)(LY), \
 547                                       (int)(M)))
 548 #endif
 549
 550 /* Intrinsics for text/string processing and reading values of
 551    EFlags.  */
 552
 553 #ifdef __OPTIMIZE__
 554 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 555 _mm_cmpistra (__m128i __X, __m128i __Y, const int __M)
 556 {
 557   return __builtin_ia32_pcmpistria128 ((__v16qi)__X,
 558                                        (__v16qi)__Y,
 559                                        __M);
 560 }
 561
 562 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 563 _mm_cmpistrc (__m128i __X, __m128i __Y, const int __M)
 564 {
 565   return __builtin_ia32_pcmpistric128 ((__v16qi)__X,
 566                                        (__v16qi)__Y,
 567                                        __M);
 568 }
 569
 570 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 571 _mm_cmpistro (__m128i __X, __m128i __Y, const int __M)
 572 {
 573   return __builtin_ia32_pcmpistrio128 ((__v16qi)__X,
 574                                        (__v16qi)__Y,
 575                                        __M);
 576 }
 577
 578 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 579 _mm_cmpistrs (__m128i __X, __m128i __Y, const int __M)
 580 {
 581   return __builtin_ia32_pcmpistris128 ((__v16qi)__X,
 582                                        (__v16qi)__Y,
 583                                        __M);
 584 }
 585
 586 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 587 _mm_cmpistrz (__m128i __X, __m128i __Y, const int __M)
 588 {
 589   return __builtin_ia32_pcmpistriz128 ((__v16qi)__X,
 590                                        (__v16qi)__Y,
 591                                        __M);
 592 }
 593
 594 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 595 _mm_cmpestra (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 596 {
 597   return __builtin_ia32_pcmpestria128 ((__v16qi)__X, __LX,
 598                                        (__v16qi)__Y, __LY,
 599                                        __M);
 600 }
 601
 602 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 603 _mm_cmpestrc (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 604 {
 605   return __builtin_ia32_pcmpestric128 ((__v16qi)__X, __LX,
 606                                        (__v16qi)__Y, __LY,
 607                                        __M);
 608 }
 609
 610 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 611 _mm_cmpestro (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 612 {
 613   return __builtin_ia32_pcmpestrio128 ((__v16qi)__X, __LX,
 614                                        (__v16qi)__Y, __LY,
 615                                        __M);
 616 }
 617
 618 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 619 _mm_cmpestrs (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 620 {
 621   return __builtin_ia32_pcmpestris128 ((__v16qi)__X, __LX,
 622                                        (__v16qi)__Y, __LY,
 623                                        __M);
 624 }
 625
 626 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 627 _mm_cmpestrz (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 628 {
 629   return __builtin_ia32_pcmpestriz128 ((__v16qi)__X, __LX,
 630                                        (__v16qi)__Y, __LY,
 631                                        __M);
 632 }
 633 #else
 634 #define _mm_cmpistra(X, Y, M)                                           \
 635   ((int) __builtin_ia32_pcmpistria128 ((__v16qi)(__m128i)(X),           \
 636                                        (__v16qi)(__m128i)(Y), (int)(M)))
 637 #define _mm_cmpistrc(X, Y, M)                                           \
 638   ((int) __builtin_ia32_pcmpistric128 ((__v16qi)(__m128i)(X),           \
 639                                        (__v16qi)(__m128i)(Y), (int)(M)))
 640 #define _mm_cmpistro(X, Y, M)                                           \
 641   ((int) __builtin_ia32_pcmpistrio128 ((__v16qi)(__m128i)(X),           \
 642                                        (__v16qi)(__m128i)(Y), (int)(M)))
 643 #define _mm_cmpistrs(X, Y, M)                                           \
 644   ((int) __builtin_ia32_pcmpistris128 ((__v16qi)(__m128i)(X),           \
 645                                        (__v16qi)(__m128i)(Y), (int)(M)))
 646 #define _mm_cmpistrz(X, Y, M)                                           \
 647   ((int) __builtin_ia32_pcmpistriz128 ((__v16qi)(__m128i)(X),           \
 648                                        (__v16qi)(__m128i)(Y), (int)(M)))
 649
 650 #define _mm_cmpestra(X, LX, Y, LY, M)                                   \
 651   ((int) __builtin_ia32_pcmpestria128 ((__v16qi)(__m128i)(X), (int)(LX), \
 652                                        (__v16qi)(__m128i)(Y), (int)(LY), \
 653                                        (int)(M)))
 654 #define _mm_cmpestrc(X, LX, Y, LY, M)                                   \
 655   ((int) __builtin_ia32_pcmpestric128 ((__v16qi)(__m128i)(X), (int)(LX), \
 656                                        (__v16qi)(__m128i)(Y), (int)(LY), \
 657                                        (int)(M)))
 658 #define _mm_cmpestro(X, LX, Y, LY, M)                                   \
 659   ((int) __builtin_ia32_pcmpestrio128 ((__v16qi)(__m128i)(X), (int)(LX), \
 660                                        (__v16qi)(__m128i)(Y), (int)(LY), \
 661                                        (int)(M)))
 662 #define _mm_cmpestrs(X, LX, Y, LY, M)                                   \
 663   ((int) __builtin_ia32_pcmpestris128 ((__v16qi)(__m128i)(X), (int)(LX), \
 664                                        (__v16qi)(__m128i)(Y), (int)(LY), \
 665                                        (int)(M)))
 666 #define _mm_cmpestrz(X, LX, Y, LY, M)                                   \
 667   ((int) __builtin_ia32_pcmpestriz128 ((__v16qi)(__m128i)(X), (int)(LX), \
 668                                        (__v16qi)(__m128i)(Y), (int)(LY), \
 669                                        (int)(M)))
 670 #endif
 671
 672 /* Packed integer 64-bit comparison, zeroing or filling with ones
 673    corresponding parts of result.  */
 674 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 675 _mm_cmpgt_epi64 (__m128i __X, __m128i __Y)
 676 {
 677   return (__m128i) __builtin_ia32_pcmpgtq ((__v2di)__X, (__v2di)__Y);
 678 }
 679
 680 /* Calculate a number of bits set to 1.  */
 681 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 682 _mm_popcnt_u32 (unsigned int __X)
 683 {
 684   return __builtin_popcount (__X);
 685 }
 686
 687 #ifdef __x86_64__
 688 extern __inline long long  __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 689 _mm_popcnt_u64 (unsigned long long __X)
 690 {
 691   return __builtin_popcountll (__X);
 692 }
 693 #endif
 694
 695 /* Accumulate CRC32 (polynomial 0x11EDC6F41) value.  */
 696 extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 697 _mm_crc32_u8 (unsigned int __C, unsigned char __V)
 698 {
 699   return __builtin_ia32_crc32qi (__C, __V);
 700 }
 701
 702 extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 703 _mm_crc32_u16 (unsigned int __C, unsigned short __V)
 704 {
 705   return __builtin_ia32_crc32hi (__C, __V);
 706 }
 707
 708 extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 709 _mm_crc32_u32 (unsigned int __C, unsigned int __V)
 710 {
 711   return __builtin_ia32_crc32si (__C, __V);
 712 }
 713
 714 #ifdef __x86_64__
 715 extern __inline unsigned long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 716 _mm_crc32_u64 (unsigned long long __C, unsigned long long __V)
 717 {
 718   return __builtin_ia32_crc32di (__C, __V);
 719 }
 720 #endif
 721
 722 #endif /* __SSE4_2__ */
 723
 724 #endif /* __SSE4_1__ */
 725
 726 #endif /* _SMMINTRIN_H_INCLUDED */