| 1 |
/*===---- tmmintrin.h - SSSE3 intrinsics -----------------------------------=== |
| 2 |
* |
| 3 |
* Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 |
* See https://llvm.org/LICENSE.txt for license information. |
| 5 |
* SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 |
* |
| 7 |
*===-----------------------------------------------------------------------=== |
| 8 |
*/ |
| 9 |
|
| 10 |
#ifndef __TMMINTRIN_H |
| 11 |
#define __TMMINTRIN_H |
| 12 |
|
| 13 |
#if !defined(__i386__) && !defined(__x86_64__) |
| 14 |
#error "This header is only meant to be used on x86 and x64 architecture" |
| 15 |
#endif |
| 16 |
|
| 17 |
#include <pmmintrin.h> |
| 18 |
|
| 19 |
/* Define the default attributes for the functions in this file. */ |
| 20 |
#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("ssse3"), __min_vector_width__(64))) |
| 21 |
#define __DEFAULT_FN_ATTRS_MMX __attribute__((__always_inline__, __nodebug__, __target__("mmx,ssse3"), __min_vector_width__(64))) |
| 22 |
|
| 23 |
/// Computes the absolute value of each of the packed 8-bit signed |
| 24 |
/// integers in the source operand and stores the 8-bit unsigned integer |
| 25 |
/// results in the destination. |
| 26 |
/// |
| 27 |
/// \headerfile <x86intrin.h> |
| 28 |
/// |
| 29 |
/// This intrinsic corresponds to the \c PABSB instruction. |
| 30 |
/// |
| 31 |
/// \param __a |
| 32 |
/// A 64-bit vector of [8 x i8]. |
| 33 |
/// \returns A 64-bit integer vector containing the absolute values of the |
| 34 |
/// elements in the operand. |
| 35 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 36 |
_mm_abs_pi8(__m64 __a) |
| 37 |
{ |
| 38 |
return (__m64)__builtin_ia32_pabsb((__v8qi)__a); |
| 39 |
} |
| 40 |
|
| 41 |
/// Computes the absolute value of each of the packed 8-bit signed |
| 42 |
/// integers in the source operand and stores the 8-bit unsigned integer |
| 43 |
/// results in the destination. |
| 44 |
/// |
| 45 |
/// \headerfile <x86intrin.h> |
| 46 |
/// |
| 47 |
/// This intrinsic corresponds to the \c VPABSB instruction. |
| 48 |
/// |
| 49 |
/// \param __a |
| 50 |
/// A 128-bit vector of [16 x i8]. |
| 51 |
/// \returns A 128-bit integer vector containing the absolute values of the |
| 52 |
/// elements in the operand. |
| 53 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 54 |
_mm_abs_epi8(__m128i __a) |
| 55 |
{ |
| 56 |
return (__m128i)__builtin_elementwise_abs((__v16qs)__a); |
| 57 |
} |
| 58 |
|
| 59 |
/// Computes the absolute value of each of the packed 16-bit signed |
| 60 |
/// integers in the source operand and stores the 16-bit unsigned integer |
| 61 |
/// results in the destination. |
| 62 |
/// |
| 63 |
/// \headerfile <x86intrin.h> |
| 64 |
/// |
| 65 |
/// This intrinsic corresponds to the \c PABSW instruction. |
| 66 |
/// |
| 67 |
/// \param __a |
| 68 |
/// A 64-bit vector of [4 x i16]. |
| 69 |
/// \returns A 64-bit integer vector containing the absolute values of the |
| 70 |
/// elements in the operand. |
| 71 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 72 |
_mm_abs_pi16(__m64 __a) |
| 73 |
{ |
| 74 |
return (__m64)__builtin_ia32_pabsw((__v4hi)__a); |
| 75 |
} |
| 76 |
|
| 77 |
/// Computes the absolute value of each of the packed 16-bit signed |
| 78 |
/// integers in the source operand and stores the 16-bit unsigned integer |
| 79 |
/// results in the destination. |
| 80 |
/// |
| 81 |
/// \headerfile <x86intrin.h> |
| 82 |
/// |
| 83 |
/// This intrinsic corresponds to the \c VPABSW instruction. |
| 84 |
/// |
| 85 |
/// \param __a |
| 86 |
/// A 128-bit vector of [8 x i16]. |
| 87 |
/// \returns A 128-bit integer vector containing the absolute values of the |
| 88 |
/// elements in the operand. |
| 89 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 90 |
_mm_abs_epi16(__m128i __a) |
| 91 |
{ |
| 92 |
return (__m128i)__builtin_elementwise_abs((__v8hi)__a); |
| 93 |
} |
| 94 |
|
| 95 |
/// Computes the absolute value of each of the packed 32-bit signed |
| 96 |
/// integers in the source operand and stores the 32-bit unsigned integer |
| 97 |
/// results in the destination. |
| 98 |
/// |
| 99 |
/// \headerfile <x86intrin.h> |
| 100 |
/// |
| 101 |
/// This intrinsic corresponds to the \c PABSD instruction. |
| 102 |
/// |
| 103 |
/// \param __a |
| 104 |
/// A 64-bit vector of [2 x i32]. |
| 105 |
/// \returns A 64-bit integer vector containing the absolute values of the |
| 106 |
/// elements in the operand. |
| 107 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 108 |
_mm_abs_pi32(__m64 __a) |
| 109 |
{ |
| 110 |
return (__m64)__builtin_ia32_pabsd((__v2si)__a); |
| 111 |
} |
| 112 |
|
| 113 |
/// Computes the absolute value of each of the packed 32-bit signed |
| 114 |
/// integers in the source operand and stores the 32-bit unsigned integer |
| 115 |
/// results in the destination. |
| 116 |
/// |
| 117 |
/// \headerfile <x86intrin.h> |
| 118 |
/// |
| 119 |
/// This intrinsic corresponds to the \c VPABSD instruction. |
| 120 |
/// |
| 121 |
/// \param __a |
| 122 |
/// A 128-bit vector of [4 x i32]. |
| 123 |
/// \returns A 128-bit integer vector containing the absolute values of the |
| 124 |
/// elements in the operand. |
| 125 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 126 |
_mm_abs_epi32(__m128i __a) |
| 127 |
{ |
| 128 |
return (__m128i)__builtin_elementwise_abs((__v4si)__a); |
| 129 |
} |
| 130 |
|
| 131 |
/// Concatenates the two 128-bit integer vector operands, and |
| 132 |
/// right-shifts the result by the number of bytes specified in the immediate |
| 133 |
/// operand. |
| 134 |
/// |
| 135 |
/// \headerfile <x86intrin.h> |
| 136 |
/// |
| 137 |
/// \code |
| 138 |
/// __m128i _mm_alignr_epi8(__m128i a, __m128i b, const int n); |
| 139 |
/// \endcode |
| 140 |
/// |
| 141 |
/// This intrinsic corresponds to the \c PALIGNR instruction. |
| 142 |
/// |
| 143 |
/// \param a |
| 144 |
/// A 128-bit vector of [16 x i8] containing one of the source operands. |
| 145 |
/// \param b |
| 146 |
/// A 128-bit vector of [16 x i8] containing one of the source operands. |
| 147 |
/// \param n |
| 148 |
/// An immediate operand specifying how many bytes to right-shift the result. |
| 149 |
/// \returns A 128-bit integer vector containing the concatenated right-shifted |
| 150 |
/// value. |
| 151 |
#define _mm_alignr_epi8(a, b, n) \ |
| 152 |
((__m128i)__builtin_ia32_palignr128((__v16qi)(__m128i)(a), \ |
| 153 |
(__v16qi)(__m128i)(b), (n))) |
| 154 |
|
| 155 |
/// Concatenates the two 64-bit integer vector operands, and right-shifts |
| 156 |
/// the result by the number of bytes specified in the immediate operand. |
| 157 |
/// |
| 158 |
/// \headerfile <x86intrin.h> |
| 159 |
/// |
| 160 |
/// \code |
| 161 |
/// __m64 _mm_alignr_pi8(__m64 a, __m64 b, const int n); |
| 162 |
/// \endcode |
| 163 |
/// |
| 164 |
/// This intrinsic corresponds to the \c PALIGNR instruction. |
| 165 |
/// |
| 166 |
/// \param a |
| 167 |
/// A 64-bit vector of [8 x i8] containing one of the source operands. |
| 168 |
/// \param b |
| 169 |
/// A 64-bit vector of [8 x i8] containing one of the source operands. |
| 170 |
/// \param n |
| 171 |
/// An immediate operand specifying how many bytes to right-shift the result. |
| 172 |
/// \returns A 64-bit integer vector containing the concatenated right-shifted |
| 173 |
/// value. |
| 174 |
#define _mm_alignr_pi8(a, b, n) \ |
| 175 |
((__m64)__builtin_ia32_palignr((__v8qi)(__m64)(a), (__v8qi)(__m64)(b), (n))) |
| 176 |
|
| 177 |
/// Horizontally adds the adjacent pairs of values contained in 2 packed |
| 178 |
/// 128-bit vectors of [8 x i16]. |
| 179 |
/// |
| 180 |
/// \headerfile <x86intrin.h> |
| 181 |
/// |
| 182 |
/// This intrinsic corresponds to the \c VPHADDW instruction. |
| 183 |
/// |
| 184 |
/// \param __a |
| 185 |
/// A 128-bit vector of [8 x i16] containing one of the source operands. The |
| 186 |
/// horizontal sums of the values are stored in the lower bits of the |
| 187 |
/// destination. |
| 188 |
/// \param __b |
| 189 |
/// A 128-bit vector of [8 x i16] containing one of the source operands. The |
| 190 |
/// horizontal sums of the values are stored in the upper bits of the |
| 191 |
/// destination. |
| 192 |
/// \returns A 128-bit vector of [8 x i16] containing the horizontal sums of |
| 193 |
/// both operands. |
| 194 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 195 |
_mm_hadd_epi16(__m128i __a, __m128i __b) |
| 196 |
{ |
| 197 |
return (__m128i)__builtin_ia32_phaddw128((__v8hi)__a, (__v8hi)__b); |
| 198 |
} |
| 199 |
|
| 200 |
/// Horizontally adds the adjacent pairs of values contained in 2 packed |
| 201 |
/// 128-bit vectors of [4 x i32]. |
| 202 |
/// |
| 203 |
/// \headerfile <x86intrin.h> |
| 204 |
/// |
| 205 |
/// This intrinsic corresponds to the \c VPHADDD instruction. |
| 206 |
/// |
| 207 |
/// \param __a |
| 208 |
/// A 128-bit vector of [4 x i32] containing one of the source operands. The |
| 209 |
/// horizontal sums of the values are stored in the lower bits of the |
| 210 |
/// destination. |
| 211 |
/// \param __b |
| 212 |
/// A 128-bit vector of [4 x i32] containing one of the source operands. The |
| 213 |
/// horizontal sums of the values are stored in the upper bits of the |
| 214 |
/// destination. |
| 215 |
/// \returns A 128-bit vector of [4 x i32] containing the horizontal sums of |
| 216 |
/// both operands. |
| 217 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 218 |
_mm_hadd_epi32(__m128i __a, __m128i __b) |
| 219 |
{ |
| 220 |
return (__m128i)__builtin_ia32_phaddd128((__v4si)__a, (__v4si)__b); |
| 221 |
} |
| 222 |
|
| 223 |
/// Horizontally adds the adjacent pairs of values contained in 2 packed |
| 224 |
/// 64-bit vectors of [4 x i16]. |
| 225 |
/// |
| 226 |
/// \headerfile <x86intrin.h> |
| 227 |
/// |
| 228 |
/// This intrinsic corresponds to the \c PHADDW instruction. |
| 229 |
/// |
| 230 |
/// \param __a |
| 231 |
/// A 64-bit vector of [4 x i16] containing one of the source operands. The |
| 232 |
/// horizontal sums of the values are stored in the lower bits of the |
| 233 |
/// destination. |
| 234 |
/// \param __b |
| 235 |
/// A 64-bit vector of [4 x i16] containing one of the source operands. The |
| 236 |
/// horizontal sums of the values are stored in the upper bits of the |
| 237 |
/// destination. |
| 238 |
/// \returns A 64-bit vector of [4 x i16] containing the horizontal sums of both |
| 239 |
/// operands. |
| 240 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 241 |
_mm_hadd_pi16(__m64 __a, __m64 __b) |
| 242 |
{ |
| 243 |
return (__m64)__builtin_ia32_phaddw((__v4hi)__a, (__v4hi)__b); |
| 244 |
} |
| 245 |
|
| 246 |
/// Horizontally adds the adjacent pairs of values contained in 2 packed |
| 247 |
/// 64-bit vectors of [2 x i32]. |
| 248 |
/// |
| 249 |
/// \headerfile <x86intrin.h> |
| 250 |
/// |
| 251 |
/// This intrinsic corresponds to the \c PHADDD instruction. |
| 252 |
/// |
| 253 |
/// \param __a |
| 254 |
/// A 64-bit vector of [2 x i32] containing one of the source operands. The |
| 255 |
/// horizontal sums of the values are stored in the lower bits of the |
| 256 |
/// destination. |
| 257 |
/// \param __b |
| 258 |
/// A 64-bit vector of [2 x i32] containing one of the source operands. The |
| 259 |
/// horizontal sums of the values are stored in the upper bits of the |
| 260 |
/// destination. |
| 261 |
/// \returns A 64-bit vector of [2 x i32] containing the horizontal sums of both |
| 262 |
/// operands. |
| 263 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 264 |
_mm_hadd_pi32(__m64 __a, __m64 __b) |
| 265 |
{ |
| 266 |
return (__m64)__builtin_ia32_phaddd((__v2si)__a, (__v2si)__b); |
| 267 |
} |
| 268 |
|
| 269 |
/// Horizontally adds the adjacent pairs of values contained in 2 packed |
| 270 |
/// 128-bit vectors of [8 x i16]. Positive sums greater than 0x7FFF are |
| 271 |
/// saturated to 0x7FFF. Negative sums less than 0x8000 are saturated to |
| 272 |
/// 0x8000. |
| 273 |
/// |
| 274 |
/// \headerfile <x86intrin.h> |
| 275 |
/// |
| 276 |
/// This intrinsic corresponds to the \c VPHADDSW instruction. |
| 277 |
/// |
| 278 |
/// \param __a |
| 279 |
/// A 128-bit vector of [8 x i16] containing one of the source operands. The |
| 280 |
/// horizontal sums of the values are stored in the lower bits of the |
| 281 |
/// destination. |
| 282 |
/// \param __b |
| 283 |
/// A 128-bit vector of [8 x i16] containing one of the source operands. The |
| 284 |
/// horizontal sums of the values are stored in the upper bits of the |
| 285 |
/// destination. |
| 286 |
/// \returns A 128-bit vector of [8 x i16] containing the horizontal saturated |
| 287 |
/// sums of both operands. |
| 288 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 289 |
_mm_hadds_epi16(__m128i __a, __m128i __b) |
| 290 |
{ |
| 291 |
return (__m128i)__builtin_ia32_phaddsw128((__v8hi)__a, (__v8hi)__b); |
| 292 |
} |
| 293 |
|
| 294 |
/// Horizontally adds the adjacent pairs of values contained in 2 packed |
| 295 |
/// 64-bit vectors of [4 x i16]. Positive sums greater than 0x7FFF are |
| 296 |
/// saturated to 0x7FFF. Negative sums less than 0x8000 are saturated to |
| 297 |
/// 0x8000. |
| 298 |
/// |
| 299 |
/// \headerfile <x86intrin.h> |
| 300 |
/// |
| 301 |
/// This intrinsic corresponds to the \c PHADDSW instruction. |
| 302 |
/// |
| 303 |
/// \param __a |
| 304 |
/// A 64-bit vector of [4 x i16] containing one of the source operands. The |
| 305 |
/// horizontal sums of the values are stored in the lower bits of the |
| 306 |
/// destination. |
| 307 |
/// \param __b |
| 308 |
/// A 64-bit vector of [4 x i16] containing one of the source operands. The |
| 309 |
/// horizontal sums of the values are stored in the upper bits of the |
| 310 |
/// destination. |
| 311 |
/// \returns A 64-bit vector of [4 x i16] containing the horizontal saturated |
| 312 |
/// sums of both operands. |
| 313 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 314 |
_mm_hadds_pi16(__m64 __a, __m64 __b) |
| 315 |
{ |
| 316 |
return (__m64)__builtin_ia32_phaddsw((__v4hi)__a, (__v4hi)__b); |
| 317 |
} |
| 318 |
|
| 319 |
/// Horizontally subtracts the adjacent pairs of values contained in 2 |
| 320 |
/// packed 128-bit vectors of [8 x i16]. |
| 321 |
/// |
| 322 |
/// \headerfile <x86intrin.h> |
| 323 |
/// |
| 324 |
/// This intrinsic corresponds to the \c VPHSUBW instruction. |
| 325 |
/// |
| 326 |
/// \param __a |
| 327 |
/// A 128-bit vector of [8 x i16] containing one of the source operands. The |
| 328 |
/// horizontal differences between the values are stored in the lower bits of |
| 329 |
/// the destination. |
| 330 |
/// \param __b |
| 331 |
/// A 128-bit vector of [8 x i16] containing one of the source operands. The |
| 332 |
/// horizontal differences between the values are stored in the upper bits of |
| 333 |
/// the destination. |
| 334 |
/// \returns A 128-bit vector of [8 x i16] containing the horizontal differences |
| 335 |
/// of both operands. |
| 336 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 337 |
_mm_hsub_epi16(__m128i __a, __m128i __b) |
| 338 |
{ |
| 339 |
return (__m128i)__builtin_ia32_phsubw128((__v8hi)__a, (__v8hi)__b); |
| 340 |
} |
| 341 |
|
| 342 |
/// Horizontally subtracts the adjacent pairs of values contained in 2 |
| 343 |
/// packed 128-bit vectors of [4 x i32]. |
| 344 |
/// |
| 345 |
/// \headerfile <x86intrin.h> |
| 346 |
/// |
| 347 |
/// This intrinsic corresponds to the \c VPHSUBD instruction. |
| 348 |
/// |
| 349 |
/// \param __a |
| 350 |
/// A 128-bit vector of [4 x i32] containing one of the source operands. The |
| 351 |
/// horizontal differences between the values are stored in the lower bits of |
| 352 |
/// the destination. |
| 353 |
/// \param __b |
| 354 |
/// A 128-bit vector of [4 x i32] containing one of the source operands. The |
| 355 |
/// horizontal differences between the values are stored in the upper bits of |
| 356 |
/// the destination. |
| 357 |
/// \returns A 128-bit vector of [4 x i32] containing the horizontal differences |
| 358 |
/// of both operands. |
| 359 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 360 |
_mm_hsub_epi32(__m128i __a, __m128i __b) |
| 361 |
{ |
| 362 |
return (__m128i)__builtin_ia32_phsubd128((__v4si)__a, (__v4si)__b); |
| 363 |
} |
| 364 |
|
| 365 |
/// Horizontally subtracts the adjacent pairs of values contained in 2 |
| 366 |
/// packed 64-bit vectors of [4 x i16]. |
| 367 |
/// |
| 368 |
/// \headerfile <x86intrin.h> |
| 369 |
/// |
| 370 |
/// This intrinsic corresponds to the \c PHSUBW instruction. |
| 371 |
/// |
| 372 |
/// \param __a |
| 373 |
/// A 64-bit vector of [4 x i16] containing one of the source operands. The |
| 374 |
/// horizontal differences between the values are stored in the lower bits of |
| 375 |
/// the destination. |
| 376 |
/// \param __b |
| 377 |
/// A 64-bit vector of [4 x i16] containing one of the source operands. The |
| 378 |
/// horizontal differences between the values are stored in the upper bits of |
| 379 |
/// the destination. |
| 380 |
/// \returns A 64-bit vector of [4 x i16] containing the horizontal differences |
| 381 |
/// of both operands. |
| 382 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 383 |
_mm_hsub_pi16(__m64 __a, __m64 __b) |
| 384 |
{ |
| 385 |
return (__m64)__builtin_ia32_phsubw((__v4hi)__a, (__v4hi)__b); |
| 386 |
} |
| 387 |
|
| 388 |
/// Horizontally subtracts the adjacent pairs of values contained in 2 |
| 389 |
/// packed 64-bit vectors of [2 x i32]. |
| 390 |
/// |
| 391 |
/// \headerfile <x86intrin.h> |
| 392 |
/// |
| 393 |
/// This intrinsic corresponds to the \c PHSUBD instruction. |
| 394 |
/// |
| 395 |
/// \param __a |
| 396 |
/// A 64-bit vector of [2 x i32] containing one of the source operands. The |
| 397 |
/// horizontal differences between the values are stored in the lower bits of |
| 398 |
/// the destination. |
| 399 |
/// \param __b |
| 400 |
/// A 64-bit vector of [2 x i32] containing one of the source operands. The |
| 401 |
/// horizontal differences between the values are stored in the upper bits of |
| 402 |
/// the destination. |
| 403 |
/// \returns A 64-bit vector of [2 x i32] containing the horizontal differences |
| 404 |
/// of both operands. |
| 405 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 406 |
_mm_hsub_pi32(__m64 __a, __m64 __b) |
| 407 |
{ |
| 408 |
return (__m64)__builtin_ia32_phsubd((__v2si)__a, (__v2si)__b); |
| 409 |
} |
| 410 |
|
| 411 |
/// Horizontally subtracts the adjacent pairs of values contained in 2 |
| 412 |
/// packed 128-bit vectors of [8 x i16]. Positive differences greater than |
| 413 |
/// 0x7FFF are saturated to 0x7FFF. Negative differences less than 0x8000 are |
| 414 |
/// saturated to 0x8000. |
| 415 |
/// |
| 416 |
/// \headerfile <x86intrin.h> |
| 417 |
/// |
| 418 |
/// This intrinsic corresponds to the \c VPHSUBSW instruction. |
| 419 |
/// |
| 420 |
/// \param __a |
| 421 |
/// A 128-bit vector of [8 x i16] containing one of the source operands. The |
| 422 |
/// horizontal differences between the values are stored in the lower bits of |
| 423 |
/// the destination. |
| 424 |
/// \param __b |
| 425 |
/// A 128-bit vector of [8 x i16] containing one of the source operands. The |
| 426 |
/// horizontal differences between the values are stored in the upper bits of |
| 427 |
/// the destination. |
| 428 |
/// \returns A 128-bit vector of [8 x i16] containing the horizontal saturated |
| 429 |
/// differences of both operands. |
| 430 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 431 |
_mm_hsubs_epi16(__m128i __a, __m128i __b) |
| 432 |
{ |
| 433 |
return (__m128i)__builtin_ia32_phsubsw128((__v8hi)__a, (__v8hi)__b); |
| 434 |
} |
| 435 |
|
| 436 |
/// Horizontally subtracts the adjacent pairs of values contained in 2 |
| 437 |
/// packed 64-bit vectors of [4 x i16]. Positive differences greater than |
| 438 |
/// 0x7FFF are saturated to 0x7FFF. Negative differences less than 0x8000 are |
| 439 |
/// saturated to 0x8000. |
| 440 |
/// |
| 441 |
/// \headerfile <x86intrin.h> |
| 442 |
/// |
| 443 |
/// This intrinsic corresponds to the \c PHSUBSW instruction. |
| 444 |
/// |
| 445 |
/// \param __a |
| 446 |
/// A 64-bit vector of [4 x i16] containing one of the source operands. The |
| 447 |
/// horizontal differences between the values are stored in the lower bits of |
| 448 |
/// the destination. |
| 449 |
/// \param __b |
| 450 |
/// A 64-bit vector of [4 x i16] containing one of the source operands. The |
| 451 |
/// horizontal differences between the values are stored in the upper bits of |
| 452 |
/// the destination. |
| 453 |
/// \returns A 64-bit vector of [4 x i16] containing the horizontal saturated |
| 454 |
/// differences of both operands. |
| 455 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 456 |
_mm_hsubs_pi16(__m64 __a, __m64 __b) |
| 457 |
{ |
| 458 |
return (__m64)__builtin_ia32_phsubsw((__v4hi)__a, (__v4hi)__b); |
| 459 |
} |
| 460 |
|
| 461 |
/// Multiplies corresponding pairs of packed 8-bit unsigned integer |
| 462 |
/// values contained in the first source operand and packed 8-bit signed |
| 463 |
/// integer values contained in the second source operand, adds pairs of |
| 464 |
/// contiguous products with signed saturation, and writes the 16-bit sums to |
| 465 |
/// the corresponding bits in the destination. |
| 466 |
/// |
| 467 |
/// For example, bits [7:0] of both operands are multiplied, bits [15:8] of |
| 468 |
/// both operands are multiplied, and the sum of both results is written to |
| 469 |
/// bits [15:0] of the destination. |
| 470 |
/// |
| 471 |
/// \headerfile <x86intrin.h> |
| 472 |
/// |
| 473 |
/// This intrinsic corresponds to the \c VPMADDUBSW instruction. |
| 474 |
/// |
| 475 |
/// \param __a |
| 476 |
/// A 128-bit integer vector containing the first source operand. |
| 477 |
/// \param __b |
| 478 |
/// A 128-bit integer vector containing the second source operand. |
| 479 |
/// \returns A 128-bit integer vector containing the sums of products of both |
| 480 |
/// operands: \n |
| 481 |
/// \a R0 := (\a __a0 * \a __b0) + (\a __a1 * \a __b1) \n |
| 482 |
/// \a R1 := (\a __a2 * \a __b2) + (\a __a3 * \a __b3) \n |
| 483 |
/// \a R2 := (\a __a4 * \a __b4) + (\a __a5 * \a __b5) \n |
| 484 |
/// \a R3 := (\a __a6 * \a __b6) + (\a __a7 * \a __b7) \n |
| 485 |
/// \a R4 := (\a __a8 * \a __b8) + (\a __a9 * \a __b9) \n |
| 486 |
/// \a R5 := (\a __a10 * \a __b10) + (\a __a11 * \a __b11) \n |
| 487 |
/// \a R6 := (\a __a12 * \a __b12) + (\a __a13 * \a __b13) \n |
| 488 |
/// \a R7 := (\a __a14 * \a __b14) + (\a __a15 * \a __b15) |
| 489 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 490 |
_mm_maddubs_epi16(__m128i __a, __m128i __b) |
| 491 |
{ |
| 492 |
return (__m128i)__builtin_ia32_pmaddubsw128((__v16qi)__a, (__v16qi)__b); |
| 493 |
} |
| 494 |
|
| 495 |
/// Multiplies corresponding pairs of packed 8-bit unsigned integer |
| 496 |
/// values contained in the first source operand and packed 8-bit signed |
| 497 |
/// integer values contained in the second source operand, adds pairs of |
| 498 |
/// contiguous products with signed saturation, and writes the 16-bit sums to |
| 499 |
/// the corresponding bits in the destination. |
| 500 |
/// |
| 501 |
/// For example, bits [7:0] of both operands are multiplied, bits [15:8] of |
| 502 |
/// both operands are multiplied, and the sum of both results is written to |
| 503 |
/// bits [15:0] of the destination. |
| 504 |
/// |
| 505 |
/// \headerfile <x86intrin.h> |
| 506 |
/// |
| 507 |
/// This intrinsic corresponds to the \c PMADDUBSW instruction. |
| 508 |
/// |
| 509 |
/// \param __a |
| 510 |
/// A 64-bit integer vector containing the first source operand. |
| 511 |
/// \param __b |
| 512 |
/// A 64-bit integer vector containing the second source operand. |
| 513 |
/// \returns A 64-bit integer vector containing the sums of products of both |
| 514 |
/// operands: \n |
| 515 |
/// \a R0 := (\a __a0 * \a __b0) + (\a __a1 * \a __b1) \n |
| 516 |
/// \a R1 := (\a __a2 * \a __b2) + (\a __a3 * \a __b3) \n |
| 517 |
/// \a R2 := (\a __a4 * \a __b4) + (\a __a5 * \a __b5) \n |
| 518 |
/// \a R3 := (\a __a6 * \a __b6) + (\a __a7 * \a __b7) |
| 519 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 520 |
_mm_maddubs_pi16(__m64 __a, __m64 __b) |
| 521 |
{ |
| 522 |
return (__m64)__builtin_ia32_pmaddubsw((__v8qi)__a, (__v8qi)__b); |
| 523 |
} |
| 524 |
|
| 525 |
/// Multiplies packed 16-bit signed integer values, truncates the 32-bit |
| 526 |
/// products to the 18 most significant bits by right-shifting, rounds the |
| 527 |
/// truncated value by adding 1, and writes bits [16:1] to the destination. |
| 528 |
/// |
| 529 |
/// \headerfile <x86intrin.h> |
| 530 |
/// |
| 531 |
/// This intrinsic corresponds to the \c VPMULHRSW instruction. |
| 532 |
/// |
| 533 |
/// \param __a |
| 534 |
/// A 128-bit vector of [8 x i16] containing one of the source operands. |
| 535 |
/// \param __b |
| 536 |
/// A 128-bit vector of [8 x i16] containing one of the source operands. |
| 537 |
/// \returns A 128-bit vector of [8 x i16] containing the rounded and scaled |
| 538 |
/// products of both operands. |
| 539 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 540 |
_mm_mulhrs_epi16(__m128i __a, __m128i __b) |
| 541 |
{ |
| 542 |
return (__m128i)__builtin_ia32_pmulhrsw128((__v8hi)__a, (__v8hi)__b); |
| 543 |
} |
| 544 |
|
| 545 |
/// Multiplies packed 16-bit signed integer values, truncates the 32-bit |
| 546 |
/// products to the 18 most significant bits by right-shifting, rounds the |
| 547 |
/// truncated value by adding 1, and writes bits [16:1] to the destination. |
| 548 |
/// |
| 549 |
/// \headerfile <x86intrin.h> |
| 550 |
/// |
| 551 |
/// This intrinsic corresponds to the \c PMULHRSW instruction. |
| 552 |
/// |
| 553 |
/// \param __a |
| 554 |
/// A 64-bit vector of [4 x i16] containing one of the source operands. |
| 555 |
/// \param __b |
| 556 |
/// A 64-bit vector of [4 x i16] containing one of the source operands. |
| 557 |
/// \returns A 64-bit vector of [4 x i16] containing the rounded and scaled |
| 558 |
/// products of both operands. |
| 559 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 560 |
_mm_mulhrs_pi16(__m64 __a, __m64 __b) |
| 561 |
{ |
| 562 |
return (__m64)__builtin_ia32_pmulhrsw((__v4hi)__a, (__v4hi)__b); |
| 563 |
} |
| 564 |
|
| 565 |
/// Copies the 8-bit integers from a 128-bit integer vector to the |
| 566 |
/// destination or clears 8-bit values in the destination, as specified by |
| 567 |
/// the second source operand. |
| 568 |
/// |
| 569 |
/// \headerfile <x86intrin.h> |
| 570 |
/// |
| 571 |
/// This intrinsic corresponds to the \c VPSHUFB instruction. |
| 572 |
/// |
| 573 |
/// \param __a |
| 574 |
/// A 128-bit integer vector containing the values to be copied. |
| 575 |
/// \param __b |
| 576 |
/// A 128-bit integer vector containing control bytes corresponding to |
| 577 |
/// positions in the destination: |
| 578 |
/// Bit 7: \n |
| 579 |
/// 1: Clear the corresponding byte in the destination. \n |
| 580 |
/// 0: Copy the selected source byte to the corresponding byte in the |
| 581 |
/// destination. \n |
| 582 |
/// Bits [6:4] Reserved. \n |
| 583 |
/// Bits [3:0] select the source byte to be copied. |
| 584 |
/// \returns A 128-bit integer vector containing the copied or cleared values. |
| 585 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 586 |
_mm_shuffle_epi8(__m128i __a, __m128i __b) |
| 587 |
{ |
| 588 |
return (__m128i)__builtin_ia32_pshufb128((__v16qi)__a, (__v16qi)__b); |
| 589 |
} |
| 590 |
|
| 591 |
/// Copies the 8-bit integers from a 64-bit integer vector to the |
| 592 |
/// destination or clears 8-bit values in the destination, as specified by |
| 593 |
/// the second source operand. |
| 594 |
/// |
| 595 |
/// \headerfile <x86intrin.h> |
| 596 |
/// |
| 597 |
/// This intrinsic corresponds to the \c PSHUFB instruction. |
| 598 |
/// |
| 599 |
/// \param __a |
| 600 |
/// A 64-bit integer vector containing the values to be copied. |
| 601 |
/// \param __b |
| 602 |
/// A 64-bit integer vector containing control bytes corresponding to |
| 603 |
/// positions in the destination: |
| 604 |
/// Bit 7: \n |
| 605 |
/// 1: Clear the corresponding byte in the destination. \n |
| 606 |
/// 0: Copy the selected source byte to the corresponding byte in the |
| 607 |
/// destination. \n |
| 608 |
/// Bits [3:0] select the source byte to be copied. |
| 609 |
/// \returns A 64-bit integer vector containing the copied or cleared values. |
| 610 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 611 |
_mm_shuffle_pi8(__m64 __a, __m64 __b) |
| 612 |
{ |
| 613 |
return (__m64)__builtin_ia32_pshufb((__v8qi)__a, (__v8qi)__b); |
| 614 |
} |
| 615 |
|
| 616 |
/// For each 8-bit integer in the first source operand, perform one of |
| 617 |
/// the following actions as specified by the second source operand. |
| 618 |
/// |
| 619 |
/// If the byte in the second source is negative, calculate the two's |
| 620 |
/// complement of the corresponding byte in the first source, and write that |
| 621 |
/// value to the destination. If the byte in the second source is positive, |
| 622 |
/// copy the corresponding byte from the first source to the destination. If |
| 623 |
/// the byte in the second source is zero, clear the corresponding byte in |
| 624 |
/// the destination. |
| 625 |
/// |
| 626 |
/// \headerfile <x86intrin.h> |
| 627 |
/// |
| 628 |
/// This intrinsic corresponds to the \c VPSIGNB instruction. |
| 629 |
/// |
| 630 |
/// \param __a |
| 631 |
/// A 128-bit integer vector containing the values to be copied. |
| 632 |
/// \param __b |
| 633 |
/// A 128-bit integer vector containing control bytes corresponding to |
| 634 |
/// positions in the destination. |
| 635 |
/// \returns A 128-bit integer vector containing the resultant values. |
| 636 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 637 |
_mm_sign_epi8(__m128i __a, __m128i __b) |
| 638 |
{ |
| 639 |
return (__m128i)__builtin_ia32_psignb128((__v16qi)__a, (__v16qi)__b); |
| 640 |
} |
| 641 |
|
| 642 |
/// For each 16-bit integer in the first source operand, perform one of |
| 643 |
/// the following actions as specified by the second source operand. |
| 644 |
/// |
| 645 |
/// If the word in the second source is negative, calculate the two's |
| 646 |
/// complement of the corresponding word in the first source, and write that |
| 647 |
/// value to the destination. If the word in the second source is positive, |
| 648 |
/// copy the corresponding word from the first source to the destination. If |
| 649 |
/// the word in the second source is zero, clear the corresponding word in |
| 650 |
/// the destination. |
| 651 |
/// |
| 652 |
/// \headerfile <x86intrin.h> |
| 653 |
/// |
| 654 |
/// This intrinsic corresponds to the \c VPSIGNW instruction. |
| 655 |
/// |
| 656 |
/// \param __a |
| 657 |
/// A 128-bit integer vector containing the values to be copied. |
| 658 |
/// \param __b |
| 659 |
/// A 128-bit integer vector containing control words corresponding to |
| 660 |
/// positions in the destination. |
| 661 |
/// \returns A 128-bit integer vector containing the resultant values. |
| 662 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 663 |
_mm_sign_epi16(__m128i __a, __m128i __b) |
| 664 |
{ |
| 665 |
return (__m128i)__builtin_ia32_psignw128((__v8hi)__a, (__v8hi)__b); |
| 666 |
} |
| 667 |
|
| 668 |
/// For each 32-bit integer in the first source operand, perform one of |
| 669 |
/// the following actions as specified by the second source operand. |
| 670 |
/// |
| 671 |
/// If the doubleword in the second source is negative, calculate the two's |
| 672 |
/// complement of the corresponding word in the first source, and write that |
| 673 |
/// value to the destination. If the doubleword in the second source is |
| 674 |
/// positive, copy the corresponding word from the first source to the |
| 675 |
/// destination. If the doubleword in the second source is zero, clear the |
| 676 |
/// corresponding word in the destination. |
| 677 |
/// |
| 678 |
/// \headerfile <x86intrin.h> |
| 679 |
/// |
| 680 |
/// This intrinsic corresponds to the \c VPSIGND instruction. |
| 681 |
/// |
| 682 |
/// \param __a |
| 683 |
/// A 128-bit integer vector containing the values to be copied. |
| 684 |
/// \param __b |
| 685 |
/// A 128-bit integer vector containing control doublewords corresponding to |
| 686 |
/// positions in the destination. |
| 687 |
/// \returns A 128-bit integer vector containing the resultant values. |
| 688 |
static __inline__ __m128i __DEFAULT_FN_ATTRS |
| 689 |
_mm_sign_epi32(__m128i __a, __m128i __b) |
| 690 |
{ |
| 691 |
return (__m128i)__builtin_ia32_psignd128((__v4si)__a, (__v4si)__b); |
| 692 |
} |
| 693 |
|
| 694 |
/// For each 8-bit integer in the first source operand, perform one of |
| 695 |
/// the following actions as specified by the second source operand. |
| 696 |
/// |
| 697 |
/// If the byte in the second source is negative, calculate the two's |
| 698 |
/// complement of the corresponding byte in the first source, and write that |
| 699 |
/// value to the destination. If the byte in the second source is positive, |
| 700 |
/// copy the corresponding byte from the first source to the destination. If |
| 701 |
/// the byte in the second source is zero, clear the corresponding byte in |
| 702 |
/// the destination. |
| 703 |
/// |
| 704 |
/// \headerfile <x86intrin.h> |
| 705 |
/// |
| 706 |
/// This intrinsic corresponds to the \c PSIGNB instruction. |
| 707 |
/// |
| 708 |
/// \param __a |
| 709 |
/// A 64-bit integer vector containing the values to be copied. |
| 710 |
/// \param __b |
| 711 |
/// A 64-bit integer vector containing control bytes corresponding to |
| 712 |
/// positions in the destination. |
| 713 |
/// \returns A 64-bit integer vector containing the resultant values. |
| 714 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 715 |
_mm_sign_pi8(__m64 __a, __m64 __b) |
| 716 |
{ |
| 717 |
return (__m64)__builtin_ia32_psignb((__v8qi)__a, (__v8qi)__b); |
| 718 |
} |
| 719 |
|
| 720 |
/// For each 16-bit integer in the first source operand, perform one of |
| 721 |
/// the following actions as specified by the second source operand. |
| 722 |
/// |
| 723 |
/// If the word in the second source is negative, calculate the two's |
| 724 |
/// complement of the corresponding word in the first source, and write that |
| 725 |
/// value to the destination. If the word in the second source is positive, |
| 726 |
/// copy the corresponding word from the first source to the destination. If |
| 727 |
/// the word in the second source is zero, clear the corresponding word in |
| 728 |
/// the destination. |
| 729 |
/// |
| 730 |
/// \headerfile <x86intrin.h> |
| 731 |
/// |
| 732 |
/// This intrinsic corresponds to the \c PSIGNW instruction. |
| 733 |
/// |
| 734 |
/// \param __a |
| 735 |
/// A 64-bit integer vector containing the values to be copied. |
| 736 |
/// \param __b |
| 737 |
/// A 64-bit integer vector containing control words corresponding to |
| 738 |
/// positions in the destination. |
| 739 |
/// \returns A 64-bit integer vector containing the resultant values. |
| 740 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 741 |
_mm_sign_pi16(__m64 __a, __m64 __b) |
| 742 |
{ |
| 743 |
return (__m64)__builtin_ia32_psignw((__v4hi)__a, (__v4hi)__b); |
| 744 |
} |
| 745 |
|
| 746 |
/// For each 32-bit integer in the first source operand, perform one of |
| 747 |
/// the following actions as specified by the second source operand. |
| 748 |
/// |
| 749 |
/// If the doubleword in the second source is negative, calculate the two's |
| 750 |
/// complement of the corresponding doubleword in the first source, and |
| 751 |
/// write that value to the destination. If the doubleword in the second |
| 752 |
/// source is positive, copy the corresponding doubleword from the first |
| 753 |
/// source to the destination. If the doubleword in the second source is |
| 754 |
/// zero, clear the corresponding doubleword in the destination. |
| 755 |
/// |
| 756 |
/// \headerfile <x86intrin.h> |
| 757 |
/// |
| 758 |
/// This intrinsic corresponds to the \c PSIGND instruction. |
| 759 |
/// |
| 760 |
/// \param __a |
| 761 |
/// A 64-bit integer vector containing the values to be copied. |
| 762 |
/// \param __b |
| 763 |
/// A 64-bit integer vector containing two control doublewords corresponding |
| 764 |
/// to positions in the destination. |
| 765 |
/// \returns A 64-bit integer vector containing the resultant values. |
| 766 |
static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
| 767 |
_mm_sign_pi32(__m64 __a, __m64 __b) |
| 768 |
{ |
| 769 |
return (__m64)__builtin_ia32_psignd((__v2si)__a, (__v2si)__b); |
| 770 |
} |
| 771 |
|
| 772 |
#undef __DEFAULT_FN_ATTRS |
| 773 |
#undef __DEFAULT_FN_ATTRS_MMX |
| 774 |
|
| 775 |
#endif /* __TMMINTRIN_H */ |