gcc/config/i386/smmintrin.h

   1 /* Copyright (C) 2007-2024 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 10.0.  */
  26
  27 #ifndef _SMMINTRIN_H_INCLUDED
  28 #define _SMMINTRIN_H_INCLUDED
  29
  30 /* We need definitions from the SSSE3, SSE3, SSE2 and SSE header
  31    files.  */
  32 #include <tmmintrin.h>
  33
  34 #ifndef __SSE4_1__
  35 #pragma GCC push_options
  36 #pragma GCC target("sse4.1")
  37 #define __DISABLE_SSE4_1__
  38 #endif /* __SSE4_1__ */
  39
  40 /* Rounding mode macros. */
  41 #define _MM_FROUND_TO_NEAREST_INT       0x00
  42 #define _MM_FROUND_TO_NEG_INF           0x01
  43 #define _MM_FROUND_TO_POS_INF           0x02
  44 #define _MM_FROUND_TO_ZERO              0x03
  45 #define _MM_FROUND_CUR_DIRECTION        0x04
  46
  47 #define _MM_FROUND_RAISE_EXC            0x00
  48 #define _MM_FROUND_NO_EXC               0x08
  49
  50 #define _MM_FROUND_NINT         \
  51   (_MM_FROUND_TO_NEAREST_INT | _MM_FROUND_RAISE_EXC)
  52 #define _MM_FROUND_FLOOR        \
  53   (_MM_FROUND_TO_NEG_INF | _MM_FROUND_RAISE_EXC)
  54 #define _MM_FROUND_CEIL         \
  55   (_MM_FROUND_TO_POS_INF | _MM_FROUND_RAISE_EXC)
  56 #define _MM_FROUND_TRUNC        \
  57   (_MM_FROUND_TO_ZERO | _MM_FROUND_RAISE_EXC)
  58 #define _MM_FROUND_RINT         \
  59   (_MM_FROUND_CUR_DIRECTION | _MM_FROUND_RAISE_EXC)
  60 #define _MM_FROUND_NEARBYINT    \
  61   (_MM_FROUND_CUR_DIRECTION | _MM_FROUND_NO_EXC)
  62
  63 /* Test Instruction */
  64 /* Packed integer 128-bit bitwise comparison. Return 1 if
  65    (__V & __M) == 0.  */
  66 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  67 _mm_testz_si128 (__m128i __M, __m128i __V)
  68 {
  69   return __builtin_ia32_ptestz128 ((__v2di)__M, (__v2di)__V);
  70 }
  71
  72 /* Packed integer 128-bit bitwise comparison. Return 1 if
  73    (__V & ~__M) == 0.  */
  74 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  75 _mm_testc_si128 (__m128i __M, __m128i __V)
  76 {
  77   return __builtin_ia32_ptestc128 ((__v2di)__M, (__v2di)__V);
  78 }
  79
  80 /* Packed integer 128-bit bitwise comparison. Return 1 if
  81    (__V & __M) != 0 && (__V & ~__M) != 0.  */
  82 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  83 _mm_testnzc_si128 (__m128i __M, __m128i __V)
  84 {
  85   return __builtin_ia32_ptestnzc128 ((__v2di)__M, (__v2di)__V);
  86 }
  87
  88 /* Macros for packed integer 128-bit comparison intrinsics.  */
  89 #define _mm_test_all_zeros(M, V) _mm_testz_si128 ((M), (V))
  90
  91 #define _mm_test_all_ones(V) \
  92   _mm_testc_si128 ((V), _mm_cmpeq_epi32 ((V), (V)))
  93
  94 #define _mm_test_mix_ones_zeros(M, V) _mm_testnzc_si128 ((M), (V))
  95
  96 /* Packed/scalar double precision floating point rounding.  */
  97
  98 #ifdef __OPTIMIZE__
  99 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 100 _mm_round_pd (__m128d __V, const int __M)
 101 {
 102   return (__m128d) __builtin_ia32_roundpd ((__v2df)__V, __M);
 103 }
 104
 105 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 106 _mm_round_sd(__m128d __D, __m128d __V, const int __M)
 107 {
 108   return (__m128d) __builtin_ia32_roundsd ((__v2df)__D,
 109                                            (__v2df)__V,
 110                                            __M);
 111 }
 112 #else
 113 #define _mm_round_pd(V, M) \
 114   ((__m128d) __builtin_ia32_roundpd ((__v2df)(__m128d)(V), (int)(M)))
 115
 116 #define _mm_round_sd(D, V, M)                                           \
 117   ((__m128d) __builtin_ia32_roundsd ((__v2df)(__m128d)(D),              \
 118                                      (__v2df)(__m128d)(V), (int)(M)))
 119 #endif
 120
 121 /* Packed/scalar single precision floating point rounding.  */
 122
 123 #ifdef __OPTIMIZE__
 124 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 125 _mm_round_ps (__m128 __V, const int __M)
 126 {
 127   return (__m128) __builtin_ia32_roundps ((__v4sf)__V, __M);
 128 }
 129
 130 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 131 _mm_round_ss (__m128 __D, __m128 __V, const int __M)
 132 {
 133   return (__m128) __builtin_ia32_roundss ((__v4sf)__D,
 134                                           (__v4sf)__V,
 135                                           __M);
 136 }
 137 #else
 138 #define _mm_round_ps(V, M) \
 139   ((__m128) __builtin_ia32_roundps ((__v4sf)(__m128)(V), (int)(M)))
 140
 141 #define _mm_round_ss(D, V, M)                                           \
 142   ((__m128) __builtin_ia32_roundss ((__v4sf)(__m128)(D),                \
 143                                     (__v4sf)(__m128)(V), (int)(M)))
 144 #endif
 145
 146 /* Macros for ceil/floor intrinsics.  */
 147 #define _mm_ceil_pd(V)     _mm_round_pd ((V), _MM_FROUND_CEIL)
 148 #define _mm_ceil_sd(D, V)  _mm_round_sd ((D), (V), _MM_FROUND_CEIL)
 149
 150 #define _mm_floor_pd(V)    _mm_round_pd((V), _MM_FROUND_FLOOR)
 151 #define _mm_floor_sd(D, V) _mm_round_sd ((D), (V), _MM_FROUND_FLOOR)
 152
 153 #define _mm_ceil_ps(V)     _mm_round_ps ((V), _MM_FROUND_CEIL)
 154 #define _mm_ceil_ss(D, V)  _mm_round_ss ((D), (V), _MM_FROUND_CEIL)
 155
 156 #define _mm_floor_ps(V)    _mm_round_ps ((V), _MM_FROUND_FLOOR)
 157 #define _mm_floor_ss(D, V) _mm_round_ss ((D), (V), _MM_FROUND_FLOOR)
 158
 159 /* SSE4.1 */
 160
 161 /* Integer blend instructions - select data from 2 sources using
 162    constant/variable mask.  */
 163
 164 #ifdef __OPTIMIZE__
 165 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 166 _mm_blend_epi16 (__m128i __X, __m128i __Y, const int __M)
 167 {
 168   return (__m128i) __builtin_ia32_pblendw128 ((__v8hi)__X,
 169                                               (__v8hi)__Y,
 170                                               __M);
 171 }
 172 #else
 173 #define _mm_blend_epi16(X, Y, M)                                        \
 174   ((__m128i) __builtin_ia32_pblendw128 ((__v8hi)(__m128i)(X),           \
 175                                         (__v8hi)(__m128i)(Y), (int)(M)))
 176 #endif
 177
 178 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 179 _mm_blendv_epi8 (__m128i __X, __m128i __Y, __m128i __M)
 180 {
 181   return (__m128i) __builtin_ia32_pblendvb128 ((__v16qi)__X,
 182                                                (__v16qi)__Y,
 183                                                (__v16qi)__M);
 184 }
 185
 186 /* Single precision floating point blend instructions - select data
 187    from 2 sources using constant/variable mask.  */
 188
 189 #ifdef __OPTIMIZE__
 190 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 191 _mm_blend_ps (__m128 __X, __m128 __Y, const int __M)
 192 {
 193   return (__m128) __builtin_ia32_blendps ((__v4sf)__X,
 194                                           (__v4sf)__Y,
 195                                           __M);
 196 }
 197 #else
 198 #define _mm_blend_ps(X, Y, M)                                           \
 199   ((__m128) __builtin_ia32_blendps ((__v4sf)(__m128)(X),                \
 200                                     (__v4sf)(__m128)(Y), (int)(M)))
 201 #endif
 202
 203 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 204 _mm_blendv_ps (__m128 __X, __m128 __Y, __m128 __M)
 205 {
 206   return (__m128) __builtin_ia32_blendvps ((__v4sf)__X,
 207                                            (__v4sf)__Y,
 208                                            (__v4sf)__M);
 209 }
 210
 211 /* Double precision floating point blend instructions - select data
 212    from 2 sources using constant/variable mask.  */
 213
 214 #ifdef __OPTIMIZE__
 215 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 216 _mm_blend_pd (__m128d __X, __m128d __Y, const int __M)
 217 {
 218   return (__m128d) __builtin_ia32_blendpd ((__v2df)__X,
 219                                            (__v2df)__Y,
 220                                            __M);
 221 }
 222 #else
 223 #define _mm_blend_pd(X, Y, M)                                           \
 224   ((__m128d) __builtin_ia32_blendpd ((__v2df)(__m128d)(X),              \
 225                                      (__v2df)(__m128d)(Y), (int)(M)))
 226 #endif
 227
 228 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 229 _mm_blendv_pd (__m128d __X, __m128d __Y, __m128d __M)
 230 {
 231   return (__m128d) __builtin_ia32_blendvpd ((__v2df)__X,
 232                                             (__v2df)__Y,
 233                                             (__v2df)__M);
 234 }
 235
 236 /* Dot product instructions with mask-defined summing and zeroing parts
 237    of result.  */
 238
 239 #ifdef __OPTIMIZE__
 240 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 241 _mm_dp_ps (__m128 __X, __m128 __Y, const int __M)
 242 {
 243   return (__m128) __builtin_ia32_dpps ((__v4sf)__X,
 244                                        (__v4sf)__Y,
 245                                        __M);
 246 }
 247
 248 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 249 _mm_dp_pd (__m128d __X, __m128d __Y, const int __M)
 250 {
 251   return (__m128d) __builtin_ia32_dppd ((__v2df)__X,
 252                                         (__v2df)__Y,
 253                                         __M);
 254 }
 255 #else
 256 #define _mm_dp_ps(X, Y, M)                                              \
 257   ((__m128) __builtin_ia32_dpps ((__v4sf)(__m128)(X),                   \
 258                                  (__v4sf)(__m128)(Y), (int)(M)))
 259
 260 #define _mm_dp_pd(X, Y, M)                                              \
 261   ((__m128d) __builtin_ia32_dppd ((__v2df)(__m128d)(X),                 \
 262                                   (__v2df)(__m128d)(Y), (int)(M)))
 263 #endif
 264
 265 /* Packed integer 64-bit comparison, zeroing or filling with ones
 266    corresponding parts of result.  */
 267 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 268 _mm_cmpeq_epi64 (__m128i __X, __m128i __Y)
 269 {
 270   return (__m128i) ((__v2di)__X == (__v2di)__Y);
 271 }
 272
 273 /*  Min/max packed integer instructions.  */
 274
 275 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 276 _mm_min_epi8 (__m128i __X, __m128i __Y)
 277 {
 278   return (__m128i) __builtin_ia32_pminsb128 ((__v16qi)__X, (__v16qi)__Y);
 279 }
 280
 281 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 282 _mm_max_epi8 (__m128i __X, __m128i __Y)
 283 {
 284   return (__m128i) __builtin_ia32_pmaxsb128 ((__v16qi)__X, (__v16qi)__Y);
 285 }
 286
 287 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 288 _mm_min_epu16 (__m128i __X, __m128i __Y)
 289 {
 290   return (__m128i) __builtin_ia32_pminuw128 ((__v8hi)__X, (__v8hi)__Y);
 291 }
 292
 293 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 294 _mm_max_epu16 (__m128i __X, __m128i __Y)
 295 {
 296   return (__m128i) __builtin_ia32_pmaxuw128 ((__v8hi)__X, (__v8hi)__Y);
 297 }
 298
 299 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 300 _mm_min_epi32 (__m128i __X, __m128i __Y)
 301 {
 302   return (__m128i) __builtin_ia32_pminsd128 ((__v4si)__X, (__v4si)__Y);
 303 }
 304
 305 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 306 _mm_max_epi32 (__m128i __X, __m128i __Y)
 307 {
 308   return (__m128i) __builtin_ia32_pmaxsd128 ((__v4si)__X, (__v4si)__Y);
 309 }
 310
 311 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 312 _mm_min_epu32 (__m128i __X, __m128i __Y)
 313 {
 314   return (__m128i) __builtin_ia32_pminud128 ((__v4si)__X, (__v4si)__Y);
 315 }
 316
 317 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 318 _mm_max_epu32 (__m128i __X, __m128i __Y)
 319 {
 320   return (__m128i) __builtin_ia32_pmaxud128 ((__v4si)__X, (__v4si)__Y);
 321 }
 322
 323 /* Packed integer 32-bit multiplication with truncation of upper
 324    halves of results.  */
 325 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 326 _mm_mullo_epi32 (__m128i __X, __m128i __Y)
 327 {
 328   return (__m128i) ((__v4su)__X * (__v4su)__Y);
 329 }
 330
 331 /* Packed integer 32-bit multiplication of 2 pairs of operands
 332    with two 64-bit results.  */
 333 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 334 _mm_mul_epi32 (__m128i __X, __m128i __Y)
 335 {
 336   return (__m128i) __builtin_ia32_pmuldq128 ((__v4si)__X, (__v4si)__Y);
 337 }
 338
 339 /* Insert single precision float into packed single precision array
 340    element selected by index N.  The bits [7-6] of N define S
 341    index, the bits [5-4] define D index, and bits [3-0] define
 342    zeroing mask for D.  */
 343
 344 #ifdef __OPTIMIZE__
 345 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 346 _mm_insert_ps (__m128 __D, __m128 __S, const int __N)
 347 {
 348   return (__m128) __builtin_ia32_insertps128 ((__v4sf)__D,
 349                                               (__v4sf)__S,
 350                                               __N);
 351 }
 352 #else
 353 #define _mm_insert_ps(D, S, N)                                          \
 354   ((__m128) __builtin_ia32_insertps128 ((__v4sf)(__m128)(D),            \
 355                                         (__v4sf)(__m128)(S), (int)(N)))
 356 #endif
 357
 358 /* Helper macro to create the N value for _mm_insert_ps.  */
 359 #define _MM_MK_INSERTPS_NDX(S, D, M) (((S) << 6) | ((D) << 4) | (M))
 360
 361 /* Extract binary representation of single precision float from packed
 362    single precision array element of X selected by index N.  */
 363
 364 #ifdef __OPTIMIZE__
 365 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 366 _mm_extract_ps (__m128 __X, const int __N)
 367 {
 368   union { int __i; float __f; } __tmp;
 369   __tmp.__f = __builtin_ia32_vec_ext_v4sf ((__v4sf)__X, __N);
 370   return __tmp.__i;
 371 }
 372 #else
 373 #define _mm_extract_ps(X, N)                                            \
 374   (__extension__                                                        \
 375    ({                                                                   \
 376      union { int __i; float __f; } __tmp;                               \
 377      __tmp.__f = __builtin_ia32_vec_ext_v4sf ((__v4sf)(__m128)(X),      \
 378                                               (int)(N));                \
 379      __tmp.__i;                                                         \
 380    }))
 381 #endif
 382
 383 /* Extract binary representation of single precision float into
 384    D from packed single precision array element of S selected
 385    by index N.  */
 386 #define _MM_EXTRACT_FLOAT(D, S, N) \
 387   { (D) = __builtin_ia32_vec_ext_v4sf ((__v4sf)(S), (N)); }
 388
 389 /* Extract specified single precision float element into the lower
 390    part of __m128.  */
 391 #define _MM_PICK_OUT_PS(X, N)                           \
 392   _mm_insert_ps (_mm_setzero_ps (), (X),                \
 393                  _MM_MK_INSERTPS_NDX ((N), 0, 0x0e))
 394
 395 /* Insert integer, S, into packed integer array element of D
 396    selected by index N.  */
 397
 398 #ifdef __OPTIMIZE__
 399 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 400 _mm_insert_epi8 (__m128i __D, int __S, const int __N)
 401 {
 402   return (__m128i) __builtin_ia32_vec_set_v16qi ((__v16qi)__D,
 403                                                  __S, __N);
 404 }
 405
 406 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 407 _mm_insert_epi32 (__m128i __D, int __S, const int __N)
 408 {
 409   return (__m128i) __builtin_ia32_vec_set_v4si ((__v4si)__D,
 410                                                  __S, __N);
 411 }
 412
 413 #ifdef __x86_64__
 414 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 415 _mm_insert_epi64 (__m128i __D, long long __S, const int __N)
 416 {
 417   return (__m128i) __builtin_ia32_vec_set_v2di ((__v2di)__D,
 418                                                  __S, __N);
 419 }
 420 #endif
 421 #else
 422 #define _mm_insert_epi8(D, S, N)                                        \
 423   ((__m128i) __builtin_ia32_vec_set_v16qi ((__v16qi)(__m128i)(D),       \
 424                                            (int)(S), (int)(N)))
 425
 426 #define _mm_insert_epi32(D, S, N)                               \
 427   ((__m128i) __builtin_ia32_vec_set_v4si ((__v4si)(__m128i)(D), \
 428                                           (int)(S), (int)(N)))
 429
 430 #ifdef __x86_64__
 431 #define _mm_insert_epi64(D, S, N)                                       \
 432   ((__m128i) __builtin_ia32_vec_set_v2di ((__v2di)(__m128i)(D),         \
 433                                           (long long)(S), (int)(N)))
 434 #endif
 435 #endif
 436
 437 /* Extract integer from packed integer array element of X selected by
 438    index N.  */
 439
 440 #ifdef __OPTIMIZE__
 441 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 442 _mm_extract_epi8 (__m128i __X, const int __N)
 443 {
 444    return (unsigned char) __builtin_ia32_vec_ext_v16qi ((__v16qi)__X, __N);
 445 }
 446
 447 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 448 _mm_extract_epi32 (__m128i __X, const int __N)
 449 {
 450    return __builtin_ia32_vec_ext_v4si ((__v4si)__X, __N);
 451 }
 452
 453 #ifdef __x86_64__
 454 extern __inline long long  __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 455 _mm_extract_epi64 (__m128i __X, const int __N)
 456 {
 457   return __builtin_ia32_vec_ext_v2di ((__v2di)__X, __N);
 458 }
 459 #endif
 460 #else
 461 #define _mm_extract_epi8(X, N) \
 462   ((int) (unsigned char) __builtin_ia32_vec_ext_v16qi ((__v16qi)(__m128i)(X), (int)(N)))
 463 #define _mm_extract_epi32(X, N) \
 464   ((int) __builtin_ia32_vec_ext_v4si ((__v4si)(__m128i)(X), (int)(N)))
 465
 466 #ifdef __x86_64__
 467 #define _mm_extract_epi64(X, N) \
 468   ((long long) __builtin_ia32_vec_ext_v2di ((__v2di)(__m128i)(X), (int)(N)))
 469 #endif
 470 #endif
 471
 472 /* Return horizontal packed word minimum and its index in bits [15:0]
 473    and bits [18:16] respectively.  */
 474 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 475 _mm_minpos_epu16 (__m128i __X)
 476 {
 477   return (__m128i) __builtin_ia32_phminposuw128 ((__v8hi)__X);
 478 }
 479
 480 /* Packed integer sign-extension.  */
 481
 482 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 483 _mm_cvtepi8_epi32 (__m128i __X)
 484 {
 485   return (__m128i) __builtin_ia32_pmovsxbd128 ((__v16qi)__X);
 486 }
 487
 488 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 489 _mm_cvtepi16_epi32 (__m128i __X)
 490 {
 491   return (__m128i) __builtin_ia32_pmovsxwd128 ((__v8hi)__X);
 492 }
 493
 494 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 495 _mm_cvtepi8_epi64 (__m128i __X)
 496 {
 497   return (__m128i) __builtin_ia32_pmovsxbq128 ((__v16qi)__X);
 498 }
 499
 500 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 501 _mm_cvtepi32_epi64 (__m128i __X)
 502 {
 503   return (__m128i) __builtin_ia32_pmovsxdq128 ((__v4si)__X);
 504 }
 505
 506 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 507 _mm_cvtepi16_epi64 (__m128i __X)
 508 {
 509   return (__m128i) __builtin_ia32_pmovsxwq128 ((__v8hi)__X);
 510 }
 511
 512 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 513 _mm_cvtepi8_epi16 (__m128i __X)
 514 {
 515   return (__m128i) __builtin_ia32_pmovsxbw128 ((__v16qi)__X);
 516 }
 517
 518 /* Packed integer zero-extension. */
 519
 520 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 521 _mm_cvtepu8_epi32 (__m128i __X)
 522 {
 523   return (__m128i) __builtin_ia32_pmovzxbd128 ((__v16qi)__X);
 524 }
 525
 526 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 527 _mm_cvtepu16_epi32 (__m128i __X)
 528 {
 529   return (__m128i) __builtin_ia32_pmovzxwd128 ((__v8hi)__X);
 530 }
 531
 532 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 533 _mm_cvtepu8_epi64 (__m128i __X)
 534 {
 535   return (__m128i) __builtin_ia32_pmovzxbq128 ((__v16qi)__X);
 536 }
 537
 538 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 539 _mm_cvtepu32_epi64 (__m128i __X)
 540 {
 541   return (__m128i) __builtin_ia32_pmovzxdq128 ((__v4si)__X);
 542 }
 543
 544 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 545 _mm_cvtepu16_epi64 (__m128i __X)
 546 {
 547   return (__m128i) __builtin_ia32_pmovzxwq128 ((__v8hi)__X);
 548 }
 549
 550 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 551 _mm_cvtepu8_epi16 (__m128i __X)
 552 {
 553   return (__m128i) __builtin_ia32_pmovzxbw128 ((__v16qi)__X);
 554 }
 555
 556 /* Pack 8 double words from 2 operands into 8 words of result with
 557    unsigned saturation. */
 558 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 559 _mm_packus_epi32 (__m128i __X, __m128i __Y)
 560 {
 561   return (__m128i) __builtin_ia32_packusdw128 ((__v4si)__X, (__v4si)__Y);
 562 }
 563
 564 /* Sum absolute 8-bit integer difference of adjacent groups of 4
 565    byte integers in the first 2 operands.  Starting offsets within
 566    operands are determined by the 3rd mask operand.  */
 567
 568 #ifdef __OPTIMIZE__
 569 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 570 _mm_mpsadbw_epu8 (__m128i __X, __m128i __Y, const int __M)
 571 {
 572   return (__m128i) __builtin_ia32_mpsadbw128 ((__v16qi)__X,
 573                                               (__v16qi)__Y, __M);
 574 }
 575 #else
 576 #define _mm_mpsadbw_epu8(X, Y, M)                                       \
 577   ((__m128i) __builtin_ia32_mpsadbw128 ((__v16qi)(__m128i)(X),          \
 578                                         (__v16qi)(__m128i)(Y), (int)(M)))
 579 #endif
 580
 581 /* Load double quadword using non-temporal aligned hint.  */
 582 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 583 _mm_stream_load_si128 (__m128i *__X)
 584 {
 585   return (__m128i) __builtin_ia32_movntdqa ((__v2di *) __X);
 586 }
 587
 588 #ifndef __SSE4_2__
 589 #pragma GCC push_options
 590 #pragma GCC target("sse4.2")
 591 #define __DISABLE_SSE4_2__
 592 #endif /* __SSE4_2__ */
 593
 594 /* These macros specify the source data format.  */
 595 #define _SIDD_UBYTE_OPS                 0x00
 596 #define _SIDD_UWORD_OPS                 0x01
 597 #define _SIDD_SBYTE_OPS                 0x02
 598 #define _SIDD_SWORD_OPS                 0x03
 599
 600 /* These macros specify the comparison operation.  */
 601 #define _SIDD_CMP_EQUAL_ANY             0x00
 602 #define _SIDD_CMP_RANGES                0x04
 603 #define _SIDD_CMP_EQUAL_EACH            0x08
 604 #define _SIDD_CMP_EQUAL_ORDERED         0x0c
 605
 606 /* These macros specify the polarity.  */
 607 #define _SIDD_POSITIVE_POLARITY         0x00
 608 #define _SIDD_NEGATIVE_POLARITY         0x10
 609 #define _SIDD_MASKED_POSITIVE_POLARITY  0x20
 610 #define _SIDD_MASKED_NEGATIVE_POLARITY  0x30
 611
 612 /* These macros specify the output selection in _mm_cmpXstri ().  */
 613 #define _SIDD_LEAST_SIGNIFICANT         0x00
 614 #define _SIDD_MOST_SIGNIFICANT          0x40
 615
 616 /* These macros specify the output selection in _mm_cmpXstrm ().  */
 617 #define _SIDD_BIT_MASK                  0x00
 618 #define _SIDD_UNIT_MASK                 0x40
 619
 620 /* Intrinsics for text/string processing.  */
 621
 622 #ifdef __OPTIMIZE__
 623 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 624 _mm_cmpistrm (__m128i __X, __m128i __Y, const int __M)
 625 {
 626   return (__m128i) __builtin_ia32_pcmpistrm128 ((__v16qi)__X,
 627                                                 (__v16qi)__Y,
 628                                                 __M);
 629 }
 630
 631 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 632 _mm_cmpistri (__m128i __X, __m128i __Y, const int __M)
 633 {
 634   return __builtin_ia32_pcmpistri128 ((__v16qi)__X,
 635                                       (__v16qi)__Y,
 636                                       __M);
 637 }
 638
 639 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 640 _mm_cmpestrm (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 641 {
 642   return (__m128i) __builtin_ia32_pcmpestrm128 ((__v16qi)__X, __LX,
 643                                                 (__v16qi)__Y, __LY,
 644                                                 __M);
 645 }
 646
 647 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 648 _mm_cmpestri (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 649 {
 650   return __builtin_ia32_pcmpestri128 ((__v16qi)__X, __LX,
 651                                       (__v16qi)__Y, __LY,
 652                                       __M);
 653 }
 654 #else
 655 #define _mm_cmpistrm(X, Y, M)                                           \
 656   ((__m128i) __builtin_ia32_pcmpistrm128 ((__v16qi)(__m128i)(X),        \
 657                                           (__v16qi)(__m128i)(Y), (int)(M)))
 658 #define _mm_cmpistri(X, Y, M)                                           \
 659   ((int) __builtin_ia32_pcmpistri128 ((__v16qi)(__m128i)(X),            \
 660                                       (__v16qi)(__m128i)(Y), (int)(M)))
 661
 662 #define _mm_cmpestrm(X, LX, Y, LY, M)                                   \
 663   ((__m128i) __builtin_ia32_pcmpestrm128 ((__v16qi)(__m128i)(X),        \
 664                                           (int)(LX), (__v16qi)(__m128i)(Y), \
 665                                           (int)(LY), (int)(M)))
 666 #define _mm_cmpestri(X, LX, Y, LY, M)                                   \
 667   ((int) __builtin_ia32_pcmpestri128 ((__v16qi)(__m128i)(X), (int)(LX), \
 668                                       (__v16qi)(__m128i)(Y), (int)(LY), \
 669                                       (int)(M)))
 670 #endif
 671
 672 /* Intrinsics for text/string processing and reading values of
 673    EFlags.  */
 674
 675 #ifdef __OPTIMIZE__
 676 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 677 _mm_cmpistra (__m128i __X, __m128i __Y, const int __M)
 678 {
 679   return __builtin_ia32_pcmpistria128 ((__v16qi)__X,
 680                                        (__v16qi)__Y,
 681                                        __M);
 682 }
 683
 684 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 685 _mm_cmpistrc (__m128i __X, __m128i __Y, const int __M)
 686 {
 687   return __builtin_ia32_pcmpistric128 ((__v16qi)__X,
 688                                        (__v16qi)__Y,
 689                                        __M);
 690 }
 691
 692 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 693 _mm_cmpistro (__m128i __X, __m128i __Y, const int __M)
 694 {
 695   return __builtin_ia32_pcmpistrio128 ((__v16qi)__X,
 696                                        (__v16qi)__Y,
 697                                        __M);
 698 }
 699
 700 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 701 _mm_cmpistrs (__m128i __X, __m128i __Y, const int __M)
 702 {
 703   return __builtin_ia32_pcmpistris128 ((__v16qi)__X,
 704                                        (__v16qi)__Y,
 705                                        __M);
 706 }
 707
 708 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 709 _mm_cmpistrz (__m128i __X, __m128i __Y, const int __M)
 710 {
 711   return __builtin_ia32_pcmpistriz128 ((__v16qi)__X,
 712                                        (__v16qi)__Y,
 713                                        __M);
 714 }
 715
 716 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 717 _mm_cmpestra (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 718 {
 719   return __builtin_ia32_pcmpestria128 ((__v16qi)__X, __LX,
 720                                        (__v16qi)__Y, __LY,
 721                                        __M);
 722 }
 723
 724 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 725 _mm_cmpestrc (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 726 {
 727   return __builtin_ia32_pcmpestric128 ((__v16qi)__X, __LX,
 728                                        (__v16qi)__Y, __LY,
 729                                        __M);
 730 }
 731
 732 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 733 _mm_cmpestro (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 734 {
 735   return __builtin_ia32_pcmpestrio128 ((__v16qi)__X, __LX,
 736                                        (__v16qi)__Y, __LY,
 737                                        __M);
 738 }
 739
 740 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 741 _mm_cmpestrs (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 742 {
 743   return __builtin_ia32_pcmpestris128 ((__v16qi)__X, __LX,
 744                                        (__v16qi)__Y, __LY,
 745                                        __M);
 746 }
 747
 748 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 749 _mm_cmpestrz (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 750 {
 751   return __builtin_ia32_pcmpestriz128 ((__v16qi)__X, __LX,
 752                                        (__v16qi)__Y, __LY,
 753                                        __M);
 754 }
 755 #else
 756 #define _mm_cmpistra(X, Y, M)                                           \
 757   ((int) __builtin_ia32_pcmpistria128 ((__v16qi)(__m128i)(X),           \
 758                                        (__v16qi)(__m128i)(Y), (int)(M)))
 759 #define _mm_cmpistrc(X, Y, M)                                           \
 760   ((int) __builtin_ia32_pcmpistric128 ((__v16qi)(__m128i)(X),           \
 761                                        (__v16qi)(__m128i)(Y), (int)(M)))
 762 #define _mm_cmpistro(X, Y, M)                                           \
 763   ((int) __builtin_ia32_pcmpistrio128 ((__v16qi)(__m128i)(X),           \
 764                                        (__v16qi)(__m128i)(Y), (int)(M)))
 765 #define _mm_cmpistrs(X, Y, M)                                           \
 766   ((int) __builtin_ia32_pcmpistris128 ((__v16qi)(__m128i)(X),           \
 767                                        (__v16qi)(__m128i)(Y), (int)(M)))
 768 #define _mm_cmpistrz(X, Y, M)                                           \
 769   ((int) __builtin_ia32_pcmpistriz128 ((__v16qi)(__m128i)(X),           \
 770                                        (__v16qi)(__m128i)(Y), (int)(M)))
 771
 772 #define _mm_cmpestra(X, LX, Y, LY, M)                                   \
 773   ((int) __builtin_ia32_pcmpestria128 ((__v16qi)(__m128i)(X), (int)(LX), \
 774                                        (__v16qi)(__m128i)(Y), (int)(LY), \
 775                                        (int)(M)))
 776 #define _mm_cmpestrc(X, LX, Y, LY, M)                                   \
 777   ((int) __builtin_ia32_pcmpestric128 ((__v16qi)(__m128i)(X), (int)(LX), \
 778                                        (__v16qi)(__m128i)(Y), (int)(LY), \
 779                                        (int)(M)))
 780 #define _mm_cmpestro(X, LX, Y, LY, M)                                   \
 781   ((int) __builtin_ia32_pcmpestrio128 ((__v16qi)(__m128i)(X), (int)(LX), \
 782                                        (__v16qi)(__m128i)(Y), (int)(LY), \
 783                                        (int)(M)))
 784 #define _mm_cmpestrs(X, LX, Y, LY, M)                                   \
 785   ((int) __builtin_ia32_pcmpestris128 ((__v16qi)(__m128i)(X), (int)(LX), \
 786                                        (__v16qi)(__m128i)(Y), (int)(LY), \
 787                                        (int)(M)))
 788 #define _mm_cmpestrz(X, LX, Y, LY, M)                                   \
 789   ((int) __builtin_ia32_pcmpestriz128 ((__v16qi)(__m128i)(X), (int)(LX), \
 790                                        (__v16qi)(__m128i)(Y), (int)(LY), \
 791                                        (int)(M)))
 792 #endif
 793
 794 /* Packed integer 64-bit comparison, zeroing or filling with ones
 795    corresponding parts of result.  */
 796 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 797 _mm_cmpgt_epi64 (__m128i __X, __m128i __Y)
 798 {
 799   return (__m128i) ((__v2di)__X > (__v2di)__Y);
 800 }
 801
 802 #ifdef __DISABLE_SSE4_2__
 803 #undef __DISABLE_SSE4_2__
 804 #pragma GCC pop_options
 805 #endif /* __DISABLE_SSE4_2__ */
 806
 807 #ifdef __DISABLE_SSE4_1__
 808 #undef __DISABLE_SSE4_1__
 809 #pragma GCC pop_options
 810 #endif /* __DISABLE_SSE4_1__ */
 811
 812 #include <popcntintrin.h>
 813
 814 #ifndef __CRC32__
 815 #pragma GCC push_options
 816 #pragma GCC target("crc32")
 817 #define __DISABLE_CRC32__
 818 #endif /* __CRC32__ */
 819
 820 /* Accumulate CRC32 (polynomial 0x11EDC6F41) value.  */
 821 extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 822 _mm_crc32_u8 (unsigned int __C, unsigned char __V)
 823 {
 824   return __builtin_ia32_crc32qi (__C, __V);
 825 }
 826
 827 extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 828 _mm_crc32_u16 (unsigned int __C, unsigned short __V)
 829 {
 830   return __builtin_ia32_crc32hi (__C, __V);
 831 }
 832
 833 extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 834 _mm_crc32_u32 (unsigned int __C, unsigned int __V)
 835 {
 836   return __builtin_ia32_crc32si (__C, __V);
 837 }
 838
 839 #ifdef __x86_64__
 840 extern __inline unsigned long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 841 _mm_crc32_u64 (unsigned long long __C, unsigned long long __V)
 842 {
 843   return __builtin_ia32_crc32di (__C, __V);
 844 }
 845 #endif
 846
 847 #ifdef __DISABLE_CRC32__
 848 #undef __DISABLE_CRC32__
 849 #pragma GCC pop_options
 850 #endif /* __DISABLE_CRC32__ */
 851
 852 #endif /* _SMMINTRIN_H_INCLUDED */