Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 1 | /*===---- avxintrin.h - AVX intrinsics -------------------------------------=== |
| 2 | * |
| 3 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 4 | * of this software and associated documentation files (the "Software"), to deal |
| 5 | * in the Software without restriction, including without limitation the rights |
| 6 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| 7 | * copies of the Software, and to permit persons to whom the Software is |
| 8 | * furnished to do so, subject to the following conditions: |
| 9 | * |
| 10 | * The above copyright notice and this permission notice shall be included in |
| 11 | * all copies or substantial portions of the Software. |
| 12 | * |
| 13 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 14 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 15 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 16 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 17 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 18 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
| 19 | * THE SOFTWARE. |
| 20 | * |
| 21 | *===-----------------------------------------------------------------------=== |
| 22 | */ |
| 23 | |
| 24 | #ifndef __IMMINTRIN_H |
| 25 | #error "Never use <avxintrin.h> directly; include <immintrin.h> instead." |
| 26 | #endif |
| 27 | |
| 28 | #ifndef __AVXINTRIN_H |
| 29 | #define __AVXINTRIN_H |
| 30 | |
| 31 | typedef double __v4df __attribute__ ((__vector_size__ (32))); |
| 32 | typedef float __v8sf __attribute__ ((__vector_size__ (32))); |
| 33 | typedef long long __v4di __attribute__ ((__vector_size__ (32))); |
| 34 | typedef int __v8si __attribute__ ((__vector_size__ (32))); |
| 35 | typedef short __v16hi __attribute__ ((__vector_size__ (32))); |
| 36 | typedef char __v32qi __attribute__ ((__vector_size__ (32))); |
| 37 | |
| 38 | /* We need an explicitly signed variant for char. Note that this shouldn't |
| 39 | * appear in the interface though. */ |
| 40 | typedef signed char __v32qs __attribute__((__vector_size__(32))); |
| 41 | |
| 42 | typedef float __m256 __attribute__ ((__vector_size__ (32))); |
| 43 | typedef double __m256d __attribute__((__vector_size__(32))); |
| 44 | typedef long long __m256i __attribute__((__vector_size__(32))); |
| 45 | |
| 46 | /* Define the default attributes for the functions in this file. */ |
| 47 | #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx"))) |
| 48 | |
| 49 | /* Arithmetic */ |
| 50 | /// \brief Adds two 256-bit vectors of [4 x double]. |
| 51 | /// |
| 52 | /// \headerfile <x86intrin.h> |
| 53 | /// |
| 54 | /// This intrinsic corresponds to the \c VADDPD / ADDPD instruction. |
| 55 | /// |
| 56 | /// \param __a |
| 57 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 58 | /// \param __b |
| 59 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 60 | /// \returns A 256-bit vector of [4 x double] containing the sums of both |
| 61 | /// operands. |
| 62 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 63 | _mm256_add_pd(__m256d __a, __m256d __b) |
| 64 | { |
| 65 | return __a+__b; |
| 66 | } |
| 67 | |
| 68 | /// \brief Adds two 256-bit vectors of [8 x float]. |
| 69 | /// |
| 70 | /// \headerfile <x86intrin.h> |
| 71 | /// |
| 72 | /// This intrinsic corresponds to the \c VADDPS / ADDPS instruction. |
| 73 | /// |
| 74 | /// \param __a |
| 75 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 76 | /// \param __b |
| 77 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 78 | /// \returns A 256-bit vector of [8 x float] containing the sums of both |
| 79 | /// operands. |
| 80 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 81 | _mm256_add_ps(__m256 __a, __m256 __b) |
| 82 | { |
| 83 | return __a+__b; |
| 84 | } |
| 85 | |
| 86 | /// \brief Subtracts two 256-bit vectors of [4 x double]. |
| 87 | /// |
| 88 | /// \headerfile <x86intrin.h> |
| 89 | /// |
| 90 | /// This intrinsic corresponds to the \c VSUBPD / SUBPD instruction. |
| 91 | /// |
| 92 | /// \param __a |
| 93 | /// A 256-bit vector of [4 x double] containing the minuend. |
| 94 | /// \param __b |
| 95 | /// A 256-bit vector of [4 x double] containing the subtrahend. |
| 96 | /// \returns A 256-bit vector of [4 x double] containing the differences between |
| 97 | /// both operands. |
| 98 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 99 | _mm256_sub_pd(__m256d __a, __m256d __b) |
| 100 | { |
| 101 | return __a-__b; |
| 102 | } |
| 103 | |
| 104 | /// \brief Subtracts two 256-bit vectors of [8 x float]. |
| 105 | /// |
| 106 | /// \headerfile <x86intrin.h> |
| 107 | /// |
| 108 | /// This intrinsic corresponds to the \c VSUBPS / SUBPS instruction. |
| 109 | /// |
| 110 | /// \param __a |
| 111 | /// A 256-bit vector of [8 x float] containing the minuend. |
| 112 | /// \param __b |
| 113 | /// A 256-bit vector of [8 x float] containing the subtrahend. |
| 114 | /// \returns A 256-bit vector of [8 x float] containing the differences between |
| 115 | /// both operands. |
| 116 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 117 | _mm256_sub_ps(__m256 __a, __m256 __b) |
| 118 | { |
| 119 | return __a-__b; |
| 120 | } |
| 121 | |
| 122 | /// \brief Adds the even-indexed values and subtracts the odd-indexed values of |
| 123 | /// two 256-bit vectors of [4 x double]. |
| 124 | /// |
| 125 | /// \headerfile <x86intrin.h> |
| 126 | /// |
| 127 | /// This intrinsic corresponds to the \c VADDSUBPD / ADDSUBPD instruction. |
| 128 | /// |
| 129 | /// \param __a |
| 130 | /// A 256-bit vector of [4 x double] containing the left source operand. |
| 131 | /// \param __b |
| 132 | /// A 256-bit vector of [4 x double] containing the right source operand. |
| 133 | /// \returns A 256-bit vector of [4 x double] containing the alternating sums |
| 134 | /// and differences between both operands. |
| 135 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 136 | _mm256_addsub_pd(__m256d __a, __m256d __b) |
| 137 | { |
| 138 | return (__m256d)__builtin_ia32_addsubpd256((__v4df)__a, (__v4df)__b); |
| 139 | } |
| 140 | |
| 141 | /// \brief Adds the even-indexed values and subtracts the odd-indexed values of |
| 142 | /// two 256-bit vectors of [8 x float]. |
| 143 | /// |
| 144 | /// \headerfile <x86intrin.h> |
| 145 | /// |
| 146 | /// This intrinsic corresponds to the \c VADDSUBPS / ADDSUBPS instruction. |
| 147 | /// |
| 148 | /// \param __a |
| 149 | /// A 256-bit vector of [8 x float] containing the left source operand. |
| 150 | /// \param __b |
| 151 | /// A 256-bit vector of [8 x float] containing the right source operand. |
| 152 | /// \returns A 256-bit vector of [8 x float] containing the alternating sums and |
| 153 | /// differences between both operands. |
| 154 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 155 | _mm256_addsub_ps(__m256 __a, __m256 __b) |
| 156 | { |
| 157 | return (__m256)__builtin_ia32_addsubps256((__v8sf)__a, (__v8sf)__b); |
| 158 | } |
| 159 | |
| 160 | /// \brief Divides two 256-bit vectors of [4 x double]. |
| 161 | /// |
| 162 | /// \headerfile <x86intrin.h> |
| 163 | /// |
| 164 | /// This intrinsic corresponds to the \c VDIVPD / DIVPD instruction. |
| 165 | /// |
| 166 | /// \param __a |
| 167 | /// A 256-bit vector of [4 x double] containing the dividend. |
| 168 | /// \param __b |
| 169 | /// A 256-bit vector of [4 x double] containing the divisor. |
| 170 | /// \returns A 256-bit vector of [4 x double] containing the quotients between |
| 171 | /// both operands. |
| 172 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 173 | _mm256_div_pd(__m256d __a, __m256d __b) |
| 174 | { |
| 175 | return __a / __b; |
| 176 | } |
| 177 | |
| 178 | /// \brief Divides two 256-bit vectors of [8 x float]. |
| 179 | /// |
| 180 | /// \headerfile <x86intrin.h> |
| 181 | /// |
| 182 | /// This intrinsic corresponds to the \c VDIVPS / DIVPS instruction. |
| 183 | /// |
| 184 | /// \param __a |
| 185 | /// A 256-bit vector of [8 x float] containing the dividend. |
| 186 | /// \param __b |
| 187 | /// A 256-bit vector of [8 x float] containing the divisor. |
| 188 | /// \returns A 256-bit vector of [8 x float] containing the quotients between |
| 189 | /// both operands. |
| 190 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 191 | _mm256_div_ps(__m256 __a, __m256 __b) |
| 192 | { |
| 193 | return __a / __b; |
| 194 | } |
| 195 | |
| 196 | /// \brief Compares two 256-bit vectors of [4 x double] and returns the greater |
| 197 | /// of each pair of values. |
| 198 | /// |
| 199 | /// \headerfile <x86intrin.h> |
| 200 | /// |
| 201 | /// This intrinsic corresponds to the \c VMAXPD / MAXPD instruction. |
| 202 | /// |
| 203 | /// \param __a |
| 204 | /// A 256-bit vector of [4 x double] containing one of the operands. |
| 205 | /// \param __b |
| 206 | /// A 256-bit vector of [4 x double] containing one of the operands. |
| 207 | /// \returns A 256-bit vector of [4 x double] containing the maximum values |
| 208 | /// between both operands. |
| 209 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 210 | _mm256_max_pd(__m256d __a, __m256d __b) |
| 211 | { |
| 212 | return (__m256d)__builtin_ia32_maxpd256((__v4df)__a, (__v4df)__b); |
| 213 | } |
| 214 | |
| 215 | /// \brief Compares two 256-bit vectors of [8 x float] and returns the greater |
| 216 | /// of each pair of values. |
| 217 | /// |
| 218 | /// \headerfile <x86intrin.h> |
| 219 | /// |
| 220 | /// This intrinsic corresponds to the \c VMAXPS / MAXPS instruction. |
| 221 | /// |
| 222 | /// \param __a |
| 223 | /// A 256-bit vector of [8 x float] containing one of the operands. |
| 224 | /// \param __b |
| 225 | /// A 256-bit vector of [8 x float] containing one of the operands. |
| 226 | /// \returns A 256-bit vector of [8 x float] containing the maximum values |
| 227 | /// between both operands. |
| 228 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 229 | _mm256_max_ps(__m256 __a, __m256 __b) |
| 230 | { |
| 231 | return (__m256)__builtin_ia32_maxps256((__v8sf)__a, (__v8sf)__b); |
| 232 | } |
| 233 | |
| 234 | /// \brief Compares two 256-bit vectors of [4 x double] and returns the lesser |
| 235 | /// of each pair of values. |
| 236 | /// |
| 237 | /// \headerfile <x86intrin.h> |
| 238 | /// |
| 239 | /// This intrinsic corresponds to the \c VMINPD / MINPD instruction. |
| 240 | /// |
| 241 | /// \param __a |
| 242 | /// A 256-bit vector of [4 x double] containing one of the operands. |
| 243 | /// \param __b |
| 244 | /// A 256-bit vector of [4 x double] containing one of the operands. |
| 245 | /// \returns A 256-bit vector of [4 x double] containing the minimum values |
| 246 | /// between both operands. |
| 247 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 248 | _mm256_min_pd(__m256d __a, __m256d __b) |
| 249 | { |
| 250 | return (__m256d)__builtin_ia32_minpd256((__v4df)__a, (__v4df)__b); |
| 251 | } |
| 252 | |
| 253 | /// \brief Compares two 256-bit vectors of [8 x float] and returns the lesser |
| 254 | /// of each pair of values. |
| 255 | /// |
| 256 | /// \headerfile <x86intrin.h> |
| 257 | /// |
| 258 | /// This intrinsic corresponds to the \c VMINPS / MINPS instruction. |
| 259 | /// |
| 260 | /// \param __a |
| 261 | /// A 256-bit vector of [8 x float] containing one of the operands. |
| 262 | /// \param __b |
| 263 | /// A 256-bit vector of [8 x float] containing one of the operands. |
| 264 | /// \returns A 256-bit vector of [8 x float] containing the minimum values |
| 265 | /// between both operands. |
| 266 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 267 | _mm256_min_ps(__m256 __a, __m256 __b) |
| 268 | { |
| 269 | return (__m256)__builtin_ia32_minps256((__v8sf)__a, (__v8sf)__b); |
| 270 | } |
| 271 | |
| 272 | /// \brief Multiplies two 256-bit vectors of [4 x double]. |
| 273 | /// |
| 274 | /// \headerfile <x86intrin.h> |
| 275 | /// |
| 276 | /// This intrinsic corresponds to the \c VMULPD / MULPD instruction. |
| 277 | /// |
| 278 | /// \param __a |
| 279 | /// A 256-bit vector of [4 x double] containing one of the operands. |
| 280 | /// \param __b |
| 281 | /// A 256-bit vector of [4 x double] containing one of the operands. |
| 282 | /// \returns A 256-bit vector of [4 x double] containing the products between |
| 283 | /// both operands. |
| 284 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 285 | _mm256_mul_pd(__m256d __a, __m256d __b) |
| 286 | { |
| 287 | return __a * __b; |
| 288 | } |
| 289 | |
| 290 | /// \brief Multiplies two 256-bit vectors of [8 x float]. |
| 291 | /// |
| 292 | /// \headerfile <x86intrin.h> |
| 293 | /// |
| 294 | /// This intrinsic corresponds to the \c VMULPS / MULPS instruction. |
| 295 | /// |
| 296 | /// \param __a |
| 297 | /// A 256-bit vector of [8 x float] containing one of the operands. |
| 298 | /// \param __b |
| 299 | /// A 256-bit vector of [8 x float] containing one of the operands. |
| 300 | /// \returns A 256-bit vector of [8 x float] containing the products between |
| 301 | /// both operands. |
| 302 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 303 | _mm256_mul_ps(__m256 __a, __m256 __b) |
| 304 | { |
| 305 | return __a * __b; |
| 306 | } |
| 307 | |
| 308 | /// \brief Calculates the square roots of the values stored in a 256-bit vector |
| 309 | /// of [4 x double]. |
| 310 | /// |
| 311 | /// \headerfile <x86intrin.h> |
| 312 | /// |
| 313 | /// This intrinsic corresponds to the \c VSQRTPD / SQRTPD instruction. |
| 314 | /// |
| 315 | /// \param __a |
| 316 | /// A 256-bit vector of [4 x double]. |
| 317 | /// \returns A 256-bit vector of [4 x double] containing the square roots of the |
| 318 | /// values in the operand. |
| 319 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 320 | _mm256_sqrt_pd(__m256d __a) |
| 321 | { |
| 322 | return (__m256d)__builtin_ia32_sqrtpd256((__v4df)__a); |
| 323 | } |
| 324 | |
| 325 | /// \brief Calculates the square roots of the values stored in a 256-bit vector |
| 326 | /// of [8 x float]. |
| 327 | /// |
| 328 | /// \headerfile <x86intrin.h> |
| 329 | /// |
| 330 | /// This intrinsic corresponds to the \c VSQRTPS / SQRTPS instruction. |
| 331 | /// |
| 332 | /// \param __a |
| 333 | /// A 256-bit vector of [8 x float]. |
| 334 | /// \returns A 256-bit vector of [8 x float] containing the square roots of the |
| 335 | /// values in the operand. |
| 336 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 337 | _mm256_sqrt_ps(__m256 __a) |
| 338 | { |
| 339 | return (__m256)__builtin_ia32_sqrtps256((__v8sf)__a); |
| 340 | } |
| 341 | |
| 342 | /// \brief Calculates the reciprocal square roots of the values stored in a |
| 343 | /// 256-bit vector of [8 x float]. |
| 344 | /// |
| 345 | /// \headerfile <x86intrin.h> |
| 346 | /// |
| 347 | /// This intrinsic corresponds to the \c VRSQRTPS / RSQRTPS instruction. |
| 348 | /// |
| 349 | /// \param __a |
| 350 | /// A 256-bit vector of [8 x float]. |
| 351 | /// \returns A 256-bit vector of [8 x float] containing the reciprocal square |
| 352 | /// roots of the values in the operand. |
| 353 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 354 | _mm256_rsqrt_ps(__m256 __a) |
| 355 | { |
| 356 | return (__m256)__builtin_ia32_rsqrtps256((__v8sf)__a); |
| 357 | } |
| 358 | |
| 359 | /// \brief Calculates the reciprocals of the values stored in a 256-bit vector |
| 360 | /// of [8 x float]. |
| 361 | /// |
| 362 | /// \headerfile <x86intrin.h> |
| 363 | /// |
| 364 | /// This intrinsic corresponds to the \c VRCPPS / RCPPS instruction. |
| 365 | /// |
| 366 | /// \param __a |
| 367 | /// A 256-bit vector of [8 x float]. |
| 368 | /// \returns A 256-bit vector of [8 x float] containing the reciprocals of the |
| 369 | /// values in the operand. |
| 370 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 371 | _mm256_rcp_ps(__m256 __a) |
| 372 | { |
| 373 | return (__m256)__builtin_ia32_rcpps256((__v8sf)__a); |
| 374 | } |
| 375 | |
| 376 | /// \brief Rounds the values stored in a 256-bit vector of [4 x double] as |
| 377 | /// specified by the byte operand. The source values are rounded to integer |
| 378 | /// values and returned as 64-bit double-precision floating-point values. |
| 379 | /// |
| 380 | /// \headerfile <x86intrin.h> |
| 381 | /// |
| 382 | /// \code |
| 383 | /// __m256d _mm256_round_pd(__m256d V, const int M); |
| 384 | /// \endcode |
| 385 | /// |
| 386 | /// This intrinsic corresponds to the \c VROUNDPD / ROUNDPD instruction. |
| 387 | /// |
| 388 | /// \param V |
| 389 | /// A 256-bit vector of [4 x double]. |
| 390 | /// \param M |
| 391 | /// An integer value that specifies the rounding operation. |
| 392 | /// Bits [7:4] are reserved. |
| 393 | /// Bit [3] is a precision exception value: |
| 394 | /// 0: A normal PE exception is used |
| 395 | /// 1: The PE field is not updated |
| 396 | /// Bit [2] is the rounding control source: |
| 397 | /// 0: Use bits [1:0] of M |
| 398 | /// 1: Use the current MXCSR setting |
| 399 | /// Bits [1:0] contain the rounding control definition: |
| 400 | /// 00: Nearest |
| 401 | /// 01: Downward (toward negative infinity) |
| 402 | /// 10: Upward (toward positive infinity) |
| 403 | /// 11: Truncated |
| 404 | /// \returns A 256-bit vector of [4 x double] containing the rounded values. |
| 405 | #define _mm256_round_pd(V, M) __extension__ ({ \ |
| 406 | (__m256d)__builtin_ia32_roundpd256((__v4df)(__m256d)(V), (M)); }) |
| 407 | |
| 408 | /// \brief Rounds the values stored in a 256-bit vector of [8 x float] as |
| 409 | /// specified by the byte operand. The source values are rounded to integer |
| 410 | /// values and returned as floating-point values. |
| 411 | /// |
| 412 | /// \headerfile <x86intrin.h> |
| 413 | /// |
| 414 | /// \code |
| 415 | /// __m256 _mm256_round_ps(__m256 V, const int M); |
| 416 | /// \endcode |
| 417 | /// |
| 418 | /// This intrinsic corresponds to the \c VROUNDPS / ROUNDPS instruction. |
| 419 | /// |
| 420 | /// \param V |
| 421 | /// A 256-bit vector of [8 x float]. |
| 422 | /// \param M |
| 423 | /// An integer value that specifies the rounding operation. |
| 424 | /// Bits [7:4] are reserved. |
| 425 | /// Bit [3] is a precision exception value: |
| 426 | /// 0: A normal PE exception is used |
| 427 | /// 1: The PE field is not updated |
| 428 | /// Bit [2] is the rounding control source: |
| 429 | /// 0: Use bits [1:0] of M |
| 430 | /// 1: Use the current MXCSR setting |
| 431 | /// Bits [1:0] contain the rounding control definition: |
| 432 | /// 00: Nearest |
| 433 | /// 01: Downward (toward negative infinity) |
| 434 | /// 10: Upward (toward positive infinity) |
| 435 | /// 11: Truncated |
| 436 | /// \returns A 256-bit vector of [8 x float] containing the rounded values. |
| 437 | #define _mm256_round_ps(V, M) __extension__ ({ \ |
| 438 | (__m256)__builtin_ia32_roundps256((__v8sf)(__m256)(V), (M)); }) |
| 439 | |
| 440 | /// \brief Round up the values stored in a 256-bit vector of [4 x double]. The |
| 441 | /// source values are rounded up to integer values and returned as 64-bit |
| 442 | /// double-precision floating-point values. |
| 443 | /// |
| 444 | /// \headerfile <x86intrin.h> |
| 445 | /// |
| 446 | /// \code |
| 447 | /// __m256d _mm256_ceil_pd(__m256d V); |
| 448 | /// \endcode |
| 449 | /// |
| 450 | /// This intrinsic corresponds to the \c VROUNDPD / ROUNDPD instruction. |
| 451 | /// |
| 452 | /// \param V |
| 453 | /// A 256-bit vector of [4 x double]. |
| 454 | /// \returns A 256-bit vector of [4 x double] containing the rounded up values. |
| 455 | #define _mm256_ceil_pd(V) _mm256_round_pd((V), _MM_FROUND_CEIL) |
| 456 | |
| 457 | /// \brief Round down the values stored in a 256-bit vector of [4 x double]. |
| 458 | /// The source values are rounded down to integer values and returned as |
| 459 | /// 64-bit double-precision floating-point values. |
| 460 | /// |
| 461 | /// \headerfile <x86intrin.h> |
| 462 | /// |
| 463 | /// \code |
| 464 | /// __m256d _mm256_floor_pd(__m256d V); |
| 465 | /// \endcode |
| 466 | /// |
| 467 | /// This intrinsic corresponds to the \c VROUNDPD / ROUNDPD instruction. |
| 468 | /// |
| 469 | /// \param V |
| 470 | /// A 256-bit vector of [4 x double]. |
| 471 | /// \returns A 256-bit vector of [4 x double] containing the rounded down |
| 472 | /// values. |
| 473 | #define _mm256_floor_pd(V) _mm256_round_pd((V), _MM_FROUND_FLOOR) |
| 474 | |
| 475 | /// \brief Round up the values stored in a 256-bit vector of [8 x float]. The |
| 476 | /// source values are rounded up to integer values and returned as |
| 477 | /// floating-point values. |
| 478 | /// |
| 479 | /// \headerfile <x86intrin.h> |
| 480 | /// |
| 481 | /// \code |
| 482 | /// __m256 _mm256_ceil_ps(__m256 V); |
| 483 | /// \endcode |
| 484 | /// |
| 485 | /// This intrinsic corresponds to the \c VROUNDPS / ROUNDPS instruction. |
| 486 | /// |
| 487 | /// \param V |
| 488 | /// A 256-bit vector of [8 x float]. |
| 489 | /// \returns A 256-bit vector of [8 x float] containing the rounded up values. |
| 490 | #define _mm256_ceil_ps(V) _mm256_round_ps((V), _MM_FROUND_CEIL) |
| 491 | |
| 492 | /// \brief Round down the values stored in a 256-bit vector of [8 x float]. The |
| 493 | /// source values are rounded down to integer values and returned as |
| 494 | /// floating-point values. |
| 495 | /// |
| 496 | /// \headerfile <x86intrin.h> |
| 497 | /// |
| 498 | /// \code |
| 499 | /// __m256 _mm256_floor_ps(__m256 V); |
| 500 | /// \endcode |
| 501 | /// |
| 502 | /// This intrinsic corresponds to the \c VROUNDPS / ROUNDPS instruction. |
| 503 | /// |
| 504 | /// \param V |
| 505 | /// A 256-bit vector of [8 x float]. |
| 506 | /// \returns A 256-bit vector of [8 x float] containing the rounded down values. |
| 507 | #define _mm256_floor_ps(V) _mm256_round_ps((V), _MM_FROUND_FLOOR) |
| 508 | |
| 509 | /* Logical */ |
| 510 | /// \brief Performs a bitwise AND of two 256-bit vectors of [4 x double]. |
| 511 | /// |
| 512 | /// \headerfile <x86intrin.h> |
| 513 | /// |
| 514 | /// This intrinsic corresponds to the \c VANDPD / ANDPD instruction. |
| 515 | /// |
| 516 | /// \param __a |
| 517 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 518 | /// \param __b |
| 519 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 520 | /// \returns A 256-bit vector of [4 x double] containing the bitwise AND of the |
| 521 | /// values between both operands. |
| 522 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 523 | _mm256_and_pd(__m256d __a, __m256d __b) |
| 524 | { |
| 525 | return (__m256d)((__v4di)__a & (__v4di)__b); |
| 526 | } |
| 527 | |
| 528 | /// \brief Performs a bitwise AND of two 256-bit vectors of [8 x float]. |
| 529 | /// |
| 530 | /// \headerfile <x86intrin.h> |
| 531 | /// |
| 532 | /// This intrinsic corresponds to the \c VANDPS / ANDPS instruction. |
| 533 | /// |
| 534 | /// \param __a |
| 535 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 536 | /// \param __b |
| 537 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 538 | /// \returns A 256-bit vector of [8 x float] containing the bitwise AND of the |
| 539 | /// values between both operands. |
| 540 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 541 | _mm256_and_ps(__m256 __a, __m256 __b) |
| 542 | { |
| 543 | return (__m256)((__v8si)__a & (__v8si)__b); |
| 544 | } |
| 545 | |
| 546 | /// \brief Performs a bitwise AND of two 256-bit vectors of [4 x double], using |
| 547 | /// the one's complement of the values contained in the first source operand. |
| 548 | /// |
| 549 | /// \headerfile <x86intrin.h> |
| 550 | /// |
| 551 | /// This intrinsic corresponds to the \c VANDNPD / ANDNPD instruction. |
| 552 | /// |
| 553 | /// \param __a |
| 554 | /// A 256-bit vector of [4 x double] containing the left source operand. The |
| 555 | /// one's complement of this value is used in the bitwise AND. |
| 556 | /// \param __b |
| 557 | /// A 256-bit vector of [4 x double] containing the right source operand. |
| 558 | /// \returns A 256-bit vector of [4 x double] containing the bitwise AND of the |
| 559 | /// values of the second operand and the one's complement of the first |
| 560 | /// operand. |
| 561 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 562 | _mm256_andnot_pd(__m256d __a, __m256d __b) |
| 563 | { |
| 564 | return (__m256d)(~(__v4di)__a & (__v4di)__b); |
| 565 | } |
| 566 | |
| 567 | /// \brief Performs a bitwise AND of two 256-bit vectors of [8 x float], using |
| 568 | /// the one's complement of the values contained in the first source operand. |
| 569 | /// |
| 570 | /// \headerfile <x86intrin.h> |
| 571 | /// |
| 572 | /// This intrinsic corresponds to the \c VANDNPS / ANDNPS instruction. |
| 573 | /// |
| 574 | /// \param __a |
| 575 | /// A 256-bit vector of [8 x float] containing the left source operand. The |
| 576 | /// one's complement of this value is used in the bitwise AND. |
| 577 | /// \param __b |
| 578 | /// A 256-bit vector of [8 x float] containing the right source operand. |
| 579 | /// \returns A 256-bit vector of [8 x float] containing the bitwise AND of the |
| 580 | /// values of the second operand and the one's complement of the first |
| 581 | /// operand. |
| 582 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 583 | _mm256_andnot_ps(__m256 __a, __m256 __b) |
| 584 | { |
| 585 | return (__m256)(~(__v8si)__a & (__v8si)__b); |
| 586 | } |
| 587 | |
| 588 | /// \brief Performs a bitwise OR of two 256-bit vectors of [4 x double]. |
| 589 | /// |
| 590 | /// \headerfile <x86intrin.h> |
| 591 | /// |
| 592 | /// This intrinsic corresponds to the \c VORPD / ORPD instruction. |
| 593 | /// |
| 594 | /// \param __a |
| 595 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 596 | /// \param __b |
| 597 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 598 | /// \returns A 256-bit vector of [4 x double] containing the bitwise OR of the |
| 599 | /// values between both operands. |
| 600 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 601 | _mm256_or_pd(__m256d __a, __m256d __b) |
| 602 | { |
| 603 | return (__m256d)((__v4di)__a | (__v4di)__b); |
| 604 | } |
| 605 | |
| 606 | /// \brief Performs a bitwise OR of two 256-bit vectors of [8 x float]. |
| 607 | /// |
| 608 | /// \headerfile <x86intrin.h> |
| 609 | /// |
| 610 | /// This intrinsic corresponds to the \c VORPS / ORPS instruction. |
| 611 | /// |
| 612 | /// \param __a |
| 613 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 614 | /// \param __b |
| 615 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 616 | /// \returns A 256-bit vector of [8 x float] containing the bitwise OR of the |
| 617 | /// values between both operands. |
| 618 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 619 | _mm256_or_ps(__m256 __a, __m256 __b) |
| 620 | { |
| 621 | return (__m256)((__v8si)__a | (__v8si)__b); |
| 622 | } |
| 623 | |
| 624 | /// \brief Performs a bitwise XOR of two 256-bit vectors of [4 x double]. |
| 625 | /// |
| 626 | /// \headerfile <x86intrin.h> |
| 627 | /// |
| 628 | /// This intrinsic corresponds to the \c VXORPD / XORPD instruction. |
| 629 | /// |
| 630 | /// \param __a |
| 631 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 632 | /// \param __b |
| 633 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 634 | /// \returns A 256-bit vector of [4 x double] containing the bitwise XOR of the |
| 635 | /// values between both operands. |
| 636 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 637 | _mm256_xor_pd(__m256d __a, __m256d __b) |
| 638 | { |
| 639 | return (__m256d)((__v4di)__a ^ (__v4di)__b); |
| 640 | } |
| 641 | |
| 642 | /// \brief Performs a bitwise XOR of two 256-bit vectors of [8 x float]. |
| 643 | /// |
| 644 | /// \headerfile <x86intrin.h> |
| 645 | /// |
| 646 | /// This intrinsic corresponds to the \c VXORPS / XORPS instruction. |
| 647 | /// |
| 648 | /// \param __a |
| 649 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 650 | /// \param __b |
| 651 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 652 | /// \returns A 256-bit vector of [8 x float] containing the bitwise XOR of the |
| 653 | /// values between both operands. |
| 654 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 655 | _mm256_xor_ps(__m256 __a, __m256 __b) |
| 656 | { |
| 657 | return (__m256)((__v8si)__a ^ (__v8si)__b); |
| 658 | } |
| 659 | |
| 660 | /* Horizontal arithmetic */ |
| 661 | /// \brief Horizontally adds the adjacent pairs of values contained in two |
| 662 | /// 256-bit vectors of [4 x double]. |
| 663 | /// |
| 664 | /// \headerfile <x86intrin.h> |
| 665 | /// |
| 666 | /// This intrinsic corresponds to the \c VHADDPD / HADDPD instruction. |
| 667 | /// |
| 668 | /// \param __a |
| 669 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 670 | /// The horizontal sums of the values are returned in the even-indexed |
| 671 | /// elements of a vector of [4 x double]. |
| 672 | /// \param __b |
| 673 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 674 | /// The horizontal sums of the values are returned in the odd-indexed |
| 675 | /// elements of a vector of [4 x double]. |
| 676 | /// \returns A 256-bit vector of [4 x double] containing the horizontal sums of |
| 677 | /// both operands. |
| 678 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 679 | _mm256_hadd_pd(__m256d __a, __m256d __b) |
| 680 | { |
| 681 | return (__m256d)__builtin_ia32_haddpd256((__v4df)__a, (__v4df)__b); |
| 682 | } |
| 683 | |
| 684 | /// \brief Horizontally adds the adjacent pairs of values contained in two |
| 685 | /// 256-bit vectors of [8 x float]. |
| 686 | /// |
| 687 | /// \headerfile <x86intrin.h> |
| 688 | /// |
| 689 | /// This intrinsic corresponds to the \c VHADDPS / HADDPS instruction. |
| 690 | /// |
| 691 | /// \param __a |
| 692 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 693 | /// The horizontal sums of the values are returned in the elements with |
| 694 | /// index 0, 1, 4, 5 of a vector of [8 x float]. |
| 695 | /// \param __b |
| 696 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 697 | /// The horizontal sums of the values are returned in the elements with |
| 698 | /// index 2, 3, 6, 7 of a vector of [8 x float]. |
| 699 | /// \returns A 256-bit vector of [8 x float] containing the horizontal sums of |
| 700 | /// both operands. |
| 701 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 702 | _mm256_hadd_ps(__m256 __a, __m256 __b) |
| 703 | { |
| 704 | return (__m256)__builtin_ia32_haddps256((__v8sf)__a, (__v8sf)__b); |
| 705 | } |
| 706 | |
| 707 | /// \brief Horizontally subtracts the adjacent pairs of values contained in two |
| 708 | /// 256-bit vectors of [4 x double]. |
| 709 | /// |
| 710 | /// \headerfile <x86intrin.h> |
| 711 | /// |
| 712 | /// This intrinsic corresponds to the \c VHSUBPD / HSUBPD instruction. |
| 713 | /// |
| 714 | /// \param __a |
| 715 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 716 | /// The horizontal differences between the values are returned in the |
| 717 | /// even-indexed elements of a vector of [4 x double]. |
| 718 | /// \param __b |
| 719 | /// A 256-bit vector of [4 x double] containing one of the source operands. |
| 720 | /// The horizontal differences between the values are returned in the |
| 721 | /// odd-indexed elements of a vector of [4 x double]. |
| 722 | /// \returns A 256-bit vector of [4 x double] containing the horizontal |
| 723 | /// differences of both operands. |
| 724 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 725 | _mm256_hsub_pd(__m256d __a, __m256d __b) |
| 726 | { |
| 727 | return (__m256d)__builtin_ia32_hsubpd256((__v4df)__a, (__v4df)__b); |
| 728 | } |
| 729 | |
| 730 | /// \brief Horizontally subtracts the adjacent pairs of values contained in two |
| 731 | /// 256-bit vectors of [8 x float]. |
| 732 | /// |
| 733 | /// \headerfile <x86intrin.h> |
| 734 | /// |
| 735 | /// This intrinsic corresponds to the \c VHSUBPS / HSUBPS instruction. |
| 736 | /// |
| 737 | /// \param __a |
| 738 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 739 | /// The horizontal differences between the values are returned in the |
| 740 | /// elements with index 0, 1, 4, 5 of a vector of [8 x float]. |
| 741 | /// \param __b |
| 742 | /// A 256-bit vector of [8 x float] containing one of the source operands. |
| 743 | /// The horizontal differences between the values are returned in the |
| 744 | /// elements with index 2, 3, 6, 7 of a vector of [8 x float]. |
| 745 | /// \returns A 256-bit vector of [8 x float] containing the horizontal |
| 746 | /// differences of both operands. |
| 747 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 748 | _mm256_hsub_ps(__m256 __a, __m256 __b) |
| 749 | { |
| 750 | return (__m256)__builtin_ia32_hsubps256((__v8sf)__a, (__v8sf)__b); |
| 751 | } |
| 752 | |
| 753 | /* Vector permutations */ |
| 754 | /// \brief Copies the values stored in a 128-bit vector of [2 x double] as |
| 755 | /// specified by the 128-bit integer vector operand. |
| 756 | /// |
| 757 | /// \headerfile <x86intrin.h> |
| 758 | /// |
| 759 | /// This intrinsic corresponds to the \c VPERMILPD / PERMILPD instruction. |
| 760 | /// |
| 761 | /// \param __a |
| 762 | /// A 128-bit vector of [2 x double]. |
| 763 | /// \param __c |
| 764 | /// A 128-bit integer vector operand specifying how the values are to be |
| 765 | /// copied. |
| 766 | /// Bit [1]: |
| 767 | /// 0: Bits [63:0] of the source are copied to bits [63:0] of the |
| 768 | /// returned vector |
| 769 | /// 1: Bits [127:64] of the source are copied to bits [63:0] of the |
| 770 | /// returned vector |
| 771 | /// Bit [65]: |
| 772 | /// 0: Bits [63:0] of the source are copied to bits [127:64] of the |
| 773 | /// returned vector |
| 774 | /// 1: Bits [127:64] of the source are copied to bits [127:64] of the |
| 775 | /// returned vector |
| 776 | /// \returns A 128-bit vector of [2 x double] containing the copied values. |
| 777 | static __inline __m128d __DEFAULT_FN_ATTRS |
| 778 | _mm_permutevar_pd(__m128d __a, __m128i __c) |
| 779 | { |
| 780 | return (__m128d)__builtin_ia32_vpermilvarpd((__v2df)__a, (__v2di)__c); |
| 781 | } |
| 782 | |
| 783 | /// \brief Copies the values stored in a 256-bit vector of [4 x double] as |
| 784 | /// specified by the 256-bit integer vector operand. |
| 785 | /// |
| 786 | /// \headerfile <x86intrin.h> |
| 787 | /// |
| 788 | /// This intrinsic corresponds to the \c VPERMILPD / PERMILPD instruction. |
| 789 | /// |
| 790 | /// \param __a |
| 791 | /// A 256-bit vector of [4 x double]. |
| 792 | /// \param __c |
| 793 | /// A 256-bit integer vector operand specifying how the values are to be |
| 794 | /// copied. |
| 795 | /// Bit [1]: |
| 796 | /// 0: Bits [63:0] of the source are copied to bits [63:0] of the |
| 797 | /// returned vector |
| 798 | /// 1: Bits [127:64] of the source are copied to bits [63:0] of the |
| 799 | /// returned vector |
| 800 | /// Bit [65]: |
| 801 | /// 0: Bits [63:0] of the source are copied to bits [127:64] of the |
| 802 | /// returned vector |
| 803 | /// 1: Bits [127:64] of the source are copied to bits [127:64] of the |
| 804 | /// returned vector |
| 805 | /// Bit [129]: |
| 806 | /// 0: Bits [191:128] of the source are copied to bits [191:128] of the |
| 807 | /// returned vector |
| 808 | /// 1: Bits [255:192] of the source are copied to bits [191:128] of the |
| 809 | /// returned vector |
| 810 | /// Bit [193]: |
| 811 | /// 0: Bits [191:128] of the source are copied to bits [255:192] of the |
| 812 | /// returned vector |
| 813 | /// 1: Bits [255:192] of the source are copied to bits [255:192] of the |
| 814 | /// returned vector |
| 815 | /// \returns A 256-bit vector of [4 x double] containing the copied values. |
| 816 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 817 | _mm256_permutevar_pd(__m256d __a, __m256i __c) |
| 818 | { |
| 819 | return (__m256d)__builtin_ia32_vpermilvarpd256((__v4df)__a, (__v4di)__c); |
| 820 | } |
| 821 | |
| 822 | /// \brief Copies the values stored in a 128-bit vector of [4 x float] as |
| 823 | /// specified by the 128-bit integer vector operand. |
| 824 | /// |
| 825 | /// \headerfile <x86intrin.h> |
| 826 | /// |
| 827 | /// This intrinsic corresponds to the \c VPERMILPS / PERMILPS instruction. |
| 828 | /// |
| 829 | /// \param __a |
| 830 | /// A 128-bit vector of [4 x float]. |
| 831 | /// \param __c |
| 832 | /// A 128-bit integer vector operand specifying how the values are to be |
| 833 | /// copied. |
| 834 | /// Bits [1:0]: |
| 835 | /// 00: Bits [31:0] of the source are copied to bits [31:0] of the |
| 836 | /// returned vector |
| 837 | /// 01: Bits [63:32] of the source are copied to bits [31:0] of the |
| 838 | /// returned vector |
| 839 | /// 10: Bits [95:64] of the source are copied to bits [31:0] of the |
| 840 | /// returned vector |
| 841 | /// 11: Bits [127:96] of the source are copied to bits [31:0] of the |
| 842 | /// returned vector |
| 843 | /// Bits [33:32]: |
| 844 | /// 00: Bits [31:0] of the source are copied to bits [63:32] of the |
| 845 | /// returned vector |
| 846 | /// 01: Bits [63:32] of the source are copied to bits [63:32] of the |
| 847 | /// returned vector |
| 848 | /// 10: Bits [95:64] of the source are copied to bits [63:32] of the |
| 849 | /// returned vector |
| 850 | /// 11: Bits [127:96] of the source are copied to bits [63:32] of the |
| 851 | /// returned vector |
| 852 | /// Bits [65:64]: |
| 853 | /// 00: Bits [31:0] of the source are copied to bits [95:64] of the |
| 854 | /// returned vector |
| 855 | /// 01: Bits [63:32] of the source are copied to bits [95:64] of the |
| 856 | /// returned vector |
| 857 | /// 10: Bits [95:64] of the source are copied to bits [95:64] of the |
| 858 | /// returned vector |
| 859 | /// 11: Bits [127:96] of the source are copied to bits [95:64] of the |
| 860 | /// returned vector |
| 861 | /// Bits [97:96]: |
| 862 | /// 00: Bits [31:0] of the source are copied to bits [127:96] of the |
| 863 | /// returned vector |
| 864 | /// 01: Bits [63:32] of the source are copied to bits [127:96] of the |
| 865 | /// returned vector |
| 866 | /// 10: Bits [95:64] of the source are copied to bits [127:96] of the |
| 867 | /// returned vector |
| 868 | /// 11: Bits [127:96] of the source are copied to bits [127:96] of the |
| 869 | /// returned vector |
| 870 | /// \returns A 128-bit vector of [4 x float] containing the copied values. |
| 871 | static __inline __m128 __DEFAULT_FN_ATTRS |
| 872 | _mm_permutevar_ps(__m128 __a, __m128i __c) |
| 873 | { |
| 874 | return (__m128)__builtin_ia32_vpermilvarps((__v4sf)__a, (__v4si)__c); |
| 875 | } |
| 876 | |
| 877 | /// \brief Copies the values stored in a 256-bit vector of [8 x float] as |
| 878 | /// specified by the 256-bit integer vector operand. |
| 879 | /// |
| 880 | /// \headerfile <x86intrin.h> |
| 881 | /// |
| 882 | /// This intrinsic corresponds to the \c VPERMILPS / PERMILPS instruction. |
| 883 | /// |
| 884 | /// \param __a |
| 885 | /// A 256-bit vector of [8 x float]. |
| 886 | /// \param __c |
| 887 | /// A 256-bit integer vector operand specifying how the values are to be |
| 888 | /// copied. |
| 889 | /// Bits [1:0]: |
| 890 | /// 00: Bits [31:0] of the source are copied to bits [31:0] of the |
| 891 | /// returned vector |
| 892 | /// 01: Bits [63:32] of the source are copied to bits [31:0] of the |
| 893 | /// returned vector |
| 894 | /// 10: Bits [95:64] of the source are copied to bits [31:0] of the |
| 895 | /// returned vector |
| 896 | /// 11: Bits [127:96] of the source are copied to bits [31:0] of the |
| 897 | /// returned vector |
| 898 | /// Bits [33:32]: |
| 899 | /// 00: Bits [31:0] of the source are copied to bits [63:32] of the |
| 900 | /// returned vector |
| 901 | /// 01: Bits [63:32] of the source are copied to bits [63:32] of the |
| 902 | /// returned vector |
| 903 | /// 10: Bits [95:64] of the source are copied to bits [63:32] of the |
| 904 | /// returned vector |
| 905 | /// 11: Bits [127:96] of the source are copied to bits [63:32] of the |
| 906 | /// returned vector |
| 907 | /// Bits [65:64]: |
| 908 | /// 00: Bits [31:0] of the source are copied to bits [95:64] of the |
| 909 | /// returned vector |
| 910 | /// 01: Bits [63:32] of the source are copied to bits [95:64] of the |
| 911 | /// returned vector |
| 912 | /// 10: Bits [95:64] of the source are copied to bits [95:64] of the |
| 913 | /// returned vector |
| 914 | /// 11: Bits [127:96] of the source are copied to bits [95:64] of the |
| 915 | /// returned vector |
| 916 | /// Bits [97:96]: |
| 917 | /// 00: Bits [31:0] of the source are copied to bits [127:96] of the |
| 918 | /// returned vector |
| 919 | /// 01: Bits [63:32] of the source are copied to bits [127:96] of the |
| 920 | /// returned vector |
| 921 | /// 10: Bits [95:64] of the source are copied to bits [127:96] of the |
| 922 | /// returned vector |
| 923 | /// 11: Bits [127:96] of the source are copied to bits [127:96] of the |
| 924 | /// returned vector |
| 925 | /// Bits [129:128]: |
| 926 | /// 00: Bits [159:128] of the source are copied to bits [159:128] of the |
| 927 | /// returned vector |
| 928 | /// 01: Bits [191:160] of the source are copied to bits [159:128] of the |
| 929 | /// returned vector |
| 930 | /// 10: Bits [223:192] of the source are copied to bits [159:128] of the |
| 931 | /// returned vector |
| 932 | /// 11: Bits [255:224] of the source are copied to bits [159:128] of the |
| 933 | /// returned vector |
| 934 | /// Bits [161:160]: |
| 935 | /// 00: Bits [159:128] of the source are copied to bits [191:160] of the |
| 936 | /// returned vector |
| 937 | /// 01: Bits [191:160] of the source are copied to bits [191:160] of the |
| 938 | /// returned vector |
| 939 | /// 10: Bits [223:192] of the source are copied to bits [191:160] of the |
| 940 | /// returned vector |
| 941 | /// 11: Bits [255:224] of the source are copied to bits [191:160] of the |
| 942 | /// returned vector |
| 943 | /// Bits [193:192]: |
| 944 | /// 00: Bits [159:128] of the source are copied to bits [223:192] of the |
| 945 | /// returned vector |
| 946 | /// 01: Bits [191:160] of the source are copied to bits [223:192] of the |
| 947 | /// returned vector |
| 948 | /// 10: Bits [223:192] of the source are copied to bits [223:192] of the |
| 949 | /// returned vector |
| 950 | /// 11: Bits [255:224] of the source are copied to bits [223:192] of the |
| 951 | /// returned vector |
| 952 | /// Bits [225:224]: |
| 953 | /// 00: Bits [159:128] of the source are copied to bits [255:224] of the |
| 954 | /// returned vector |
| 955 | /// 01: Bits [191:160] of the source are copied to bits [255:224] of the |
| 956 | /// returned vector |
| 957 | /// 10: Bits [223:192] of the source are copied to bits [255:224] of the |
| 958 | /// returned vector |
| 959 | /// 11: Bits [255:224] of the source are copied to bits [255:224] of the |
| 960 | /// returned vector |
| 961 | /// \returns A 256-bit vector of [8 x float] containing the copied values. |
| 962 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 963 | _mm256_permutevar_ps(__m256 __a, __m256i __c) |
| 964 | { |
| 965 | return (__m256)__builtin_ia32_vpermilvarps256((__v8sf)__a, (__v8si)__c); |
| 966 | } |
| 967 | |
| 968 | /// \brief Copies the values stored in a 128-bit vector of [2 x double] as |
| 969 | /// specified by the immediate integer operand. |
| 970 | /// |
| 971 | /// \headerfile <x86intrin.h> |
| 972 | /// |
| 973 | /// \code |
| 974 | /// __m128d _mm_permute_pd(__m128d A, const int C); |
| 975 | /// \endcode |
| 976 | /// |
| 977 | /// This intrinsic corresponds to the \c VPERMILPD / PERMILPD instruction. |
| 978 | /// |
| 979 | /// \param A |
| 980 | /// A 128-bit vector of [2 x double]. |
| 981 | /// \param C |
| 982 | /// An immediate integer operand specifying how the values are to be copied. |
| 983 | /// Bit [0]: |
| 984 | /// 0: Bits [63:0] of the source are copied to bits [63:0] of the |
| 985 | /// returned vector |
| 986 | /// 1: Bits [127:64] of the source are copied to bits [63:0] of the |
| 987 | /// returned vector |
| 988 | /// Bit [1]: |
| 989 | /// 0: Bits [63:0] of the source are copied to bits [127:64] of the |
| 990 | /// returned vector |
| 991 | /// 1: Bits [127:64] of the source are copied to bits [127:64] of the |
| 992 | /// returned vector |
| 993 | /// \returns A 128-bit vector of [2 x double] containing the copied values. |
| 994 | #define _mm_permute_pd(A, C) __extension__ ({ \ |
| 995 | (__m128d)__builtin_shufflevector((__v2df)(__m128d)(A), \ |
| 996 | (__v2df)_mm_setzero_pd(), \ |
| 997 | (C) & 0x1, ((C) & 0x2) >> 1); }) |
| 998 | |
| 999 | /// \brief Copies the values stored in a 256-bit vector of [4 x double] as |
| 1000 | /// specified by the immediate integer operand. |
| 1001 | /// |
| 1002 | /// \headerfile <x86intrin.h> |
| 1003 | /// |
| 1004 | /// \code |
| 1005 | /// __m256d _mm256_permute_pd(__m256d A, const int C); |
| 1006 | /// \endcode |
| 1007 | /// |
| 1008 | /// This intrinsic corresponds to the \c VPERMILPD / PERMILPD instruction. |
| 1009 | /// |
| 1010 | /// \param A |
| 1011 | /// A 256-bit vector of [4 x double]. |
| 1012 | /// \param C |
| 1013 | /// An immediate integer operand specifying how the values are to be copied. |
| 1014 | /// Bit [0]: |
| 1015 | /// 0: Bits [63:0] of the source are copied to bits [63:0] of the |
| 1016 | /// returned vector |
| 1017 | /// 1: Bits [127:64] of the source are copied to bits [63:0] of the |
| 1018 | /// returned vector |
| 1019 | /// Bit [1]: |
| 1020 | /// 0: Bits [63:0] of the source are copied to bits [127:64] of the |
| 1021 | /// returned vector |
| 1022 | /// 1: Bits [127:64] of the source are copied to bits [127:64] of the |
| 1023 | /// returned vector |
| 1024 | /// Bit [2]: |
| 1025 | /// 0: Bits [191:128] of the source are copied to bits [191:128] of the |
| 1026 | /// returned vector |
| 1027 | /// 1: Bits [255:192] of the source are copied to bits [191:128] of the |
| 1028 | /// returned vector |
| 1029 | /// Bit [3]: |
| 1030 | /// 0: Bits [191:128] of the source are copied to bits [255:192] of the |
| 1031 | /// returned vector |
| 1032 | /// 1: Bits [255:192] of the source are copied to bits [255:192] of the |
| 1033 | /// returned vector |
| 1034 | /// \returns A 256-bit vector of [4 x double] containing the copied values. |
| 1035 | #define _mm256_permute_pd(A, C) __extension__ ({ \ |
| 1036 | (__m256d)__builtin_shufflevector((__v4df)(__m256d)(A), \ |
| 1037 | (__v4df)_mm256_setzero_pd(), \ |
| 1038 | (C) & 0x1, ((C) & 0x2) >> 1, \ |
| 1039 | 2 + (((C) & 0x4) >> 2), \ |
| 1040 | 2 + (((C) & 0x8) >> 3)); }) |
| 1041 | |
| 1042 | /// \brief Copies the values stored in a 128-bit vector of [4 x float] as |
| 1043 | /// specified by the immediate integer operand. |
| 1044 | /// |
| 1045 | /// \headerfile <x86intrin.h> |
| 1046 | /// |
| 1047 | /// \code |
| 1048 | /// __m128 _mm_permute_ps(__m128 A, const int C); |
| 1049 | /// \endcode |
| 1050 | /// |
| 1051 | /// This intrinsic corresponds to the \c VPERMILPS / PERMILPS instruction. |
| 1052 | /// |
| 1053 | /// \param A |
| 1054 | /// A 128-bit vector of [4 x float]. |
| 1055 | /// \param C |
| 1056 | /// An immediate integer operand specifying how the values are to be copied. |
| 1057 | /// Bits [1:0]: |
| 1058 | /// 00: Bits [31:0] of the source are copied to bits [31:0] of the |
| 1059 | /// returned vector |
| 1060 | /// 01: Bits [63:32] of the source are copied to bits [31:0] of the |
| 1061 | /// returned vector |
| 1062 | /// 10: Bits [95:64] of the source are copied to bits [31:0] of the |
| 1063 | /// returned vector |
| 1064 | /// 11: Bits [127:96] of the source are copied to bits [31:0] of the |
| 1065 | /// returned vector |
| 1066 | /// Bits [3:2]: |
| 1067 | /// 00: Bits [31:0] of the source are copied to bits [63:32] of the |
| 1068 | /// returned vector |
| 1069 | /// 01: Bits [63:32] of the source are copied to bits [63:32] of the |
| 1070 | /// returned vector |
| 1071 | /// 10: Bits [95:64] of the source are copied to bits [63:32] of the |
| 1072 | /// returned vector |
| 1073 | /// 11: Bits [127:96] of the source are copied to bits [63:32] of the |
| 1074 | /// returned vector |
| 1075 | /// Bits [5:4]: |
| 1076 | /// 00: Bits [31:0] of the source are copied to bits [95:64] of the |
| 1077 | /// returned vector |
| 1078 | /// 01: Bits [63:32] of the source are copied to bits [95:64] of the |
| 1079 | /// returned vector |
| 1080 | /// 10: Bits [95:64] of the source are copied to bits [95:64] of the |
| 1081 | /// returned vector |
| 1082 | /// 11: Bits [127:96] of the source are copied to bits [95:64] of the |
| 1083 | /// returned vector |
| 1084 | /// Bits [7:6]: |
| 1085 | /// 00: Bits [31:0] of the source are copied to bits [127:96] of the |
| 1086 | /// returned vector |
| 1087 | /// 01: Bits [63:32] of the source are copied to bits [127:96] of the |
| 1088 | /// returned vector |
| 1089 | /// 10: Bits [95:64] of the source are copied to bits [127:96] of the |
| 1090 | /// returned vector |
| 1091 | /// 11: Bits [127:96] of the source are copied to bits [127:96] of the |
| 1092 | /// returned vector |
| 1093 | /// \returns A 128-bit vector of [4 x float] containing the copied values. |
| 1094 | #define _mm_permute_ps(A, C) __extension__ ({ \ |
| 1095 | (__m128)__builtin_shufflevector((__v4sf)(__m128)(A), \ |
| 1096 | (__v4sf)_mm_setzero_ps(), \ |
| 1097 | (C) & 0x3, ((C) & 0xc) >> 2, \ |
| 1098 | ((C) & 0x30) >> 4, ((C) & 0xc0) >> 6); }) |
| 1099 | |
| 1100 | /// \brief Copies the values stored in a 256-bit vector of [8 x float] as |
| 1101 | /// specified by the immediate integer operand. |
| 1102 | /// |
| 1103 | /// \headerfile <x86intrin.h> |
| 1104 | /// |
| 1105 | /// \code |
| 1106 | /// __m256 _mm256_permute_ps(__m256 A, const int C); |
| 1107 | /// \endcode |
| 1108 | /// |
| 1109 | /// This intrinsic corresponds to the \c VPERMILPS / PERMILPS instruction. |
| 1110 | /// |
| 1111 | /// \param A |
| 1112 | /// A 256-bit vector of [8 x float]. |
| 1113 | /// \param C |
| 1114 | /// An immediate integer operand specifying how the values are to be copied. |
| 1115 | /// Bits [1:0]: |
| 1116 | /// 00: Bits [31:0] of the source are copied to bits [31:0] of the |
| 1117 | /// returned vector |
| 1118 | /// 01: Bits [63:32] of the source are copied to bits [31:0] of the |
| 1119 | /// returned vector |
| 1120 | /// 10: Bits [95:64] of the source are copied to bits [31:0] of the |
| 1121 | /// returned vector |
| 1122 | /// 11: Bits [127:96] of the source are copied to bits [31:0] of the |
| 1123 | /// returned vector |
| 1124 | /// Bits [3:2]: |
| 1125 | /// 00: Bits [31:0] of the source are copied to bits [63:32] of the |
| 1126 | /// returned vector |
| 1127 | /// 01: Bits [63:32] of the source are copied to bits [63:32] of the |
| 1128 | /// returned vector |
| 1129 | /// 10: Bits [95:64] of the source are copied to bits [63:32] of the |
| 1130 | /// returned vector |
| 1131 | /// 11: Bits [127:96] of the source are copied to bits [63:32] of the |
| 1132 | /// returned vector |
| 1133 | /// Bits [5:4]: |
| 1134 | /// 00: Bits [31:0] of the source are copied to bits [95:64] of the |
| 1135 | /// returned vector |
| 1136 | /// 01: Bits [63:32] of the source are copied to bits [95:64] of the |
| 1137 | /// returned vector |
| 1138 | /// 10: Bits [95:64] of the source are copied to bits [95:64] of the |
| 1139 | /// returned vector |
| 1140 | /// 11: Bits [127:96] of the source are copied to bits [95:64] of the |
| 1141 | /// returned vector |
| 1142 | /// Bits [7:6]: |
| 1143 | /// 00: Bits [31:0] of the source are copied to bits [127:96] of the |
| 1144 | /// returned vector |
| 1145 | /// 01: Bits [63:32] of the source are copied to bits [127:96] of the |
| 1146 | /// returned vector |
| 1147 | /// 10: Bits [95:64] of the source are copied to bits [127:96] of the |
| 1148 | /// returned vector |
| 1149 | /// 11: Bits [127:96] of the source are copied to bits [127:96] of the |
| 1150 | /// returned vector |
| 1151 | /// Bits [1:0]: |
| 1152 | /// 00: Bits [159:128] of the source are copied to bits [159:128] of the |
| 1153 | /// returned vector |
| 1154 | /// 01: Bits [191:160] of the source are copied to bits [159:128] of the |
| 1155 | /// returned vector |
| 1156 | /// 10: Bits [223:192] of the source are copied to bits [159:128] of the |
| 1157 | /// returned vector |
| 1158 | /// 11: Bits [255:224] of the source are copied to bits [159:128] of the |
| 1159 | /// returned vector |
| 1160 | /// Bits [3:2]: |
| 1161 | /// 00: Bits [159:128] of the source are copied to bits [191:160] of the |
| 1162 | /// returned vector |
| 1163 | /// 01: Bits [191:160] of the source are copied to bits [191:160] of the |
| 1164 | /// returned vector |
| 1165 | /// 10: Bits [223:192] of the source are copied to bits [191:160] of the |
| 1166 | /// returned vector |
| 1167 | /// 11: Bits [255:224] of the source are copied to bits [191:160] of the |
| 1168 | /// returned vector |
| 1169 | /// Bits [5:4]: |
| 1170 | /// 00: Bits [159:128] of the source are copied to bits [223:192] of the |
| 1171 | /// returned vector |
| 1172 | /// 01: Bits [191:160] of the source are copied to bits [223:192] of the |
| 1173 | /// returned vector |
| 1174 | /// 10: Bits [223:192] of the source are copied to bits [223:192] of the |
| 1175 | /// returned vector |
| 1176 | /// 11: Bits [255:224] of the source are copied to bits [223:192] of the |
| 1177 | /// returned vector |
| 1178 | /// Bits [7:6]: |
| 1179 | /// 00: Bits [159:128] of the source are copied to bits [255:224] of the |
| 1180 | /// returned vector |
| 1181 | /// 01: Bits [191:160] of the source are copied to bits [255:224] of the |
| 1182 | /// returned vector |
| 1183 | /// 10: Bits [223:192] of the source are copied to bits [255:224] of the |
| 1184 | /// returned vector |
| 1185 | /// 11: Bits [255:224] of the source are copied to bits [255:224] of the |
| 1186 | /// returned vector |
| 1187 | /// \returns A 256-bit vector of [8 x float] containing the copied values. |
| 1188 | #define _mm256_permute_ps(A, C) __extension__ ({ \ |
| 1189 | (__m256)__builtin_shufflevector((__v8sf)(__m256)(A), \ |
| 1190 | (__v8sf)_mm256_setzero_ps(), \ |
| 1191 | (C) & 0x3, ((C) & 0xc) >> 2, \ |
| 1192 | ((C) & 0x30) >> 4, ((C) & 0xc0) >> 6, \ |
| 1193 | 4 + (((C) & 0x03) >> 0), \ |
| 1194 | 4 + (((C) & 0x0c) >> 2), \ |
| 1195 | 4 + (((C) & 0x30) >> 4), \ |
| 1196 | 4 + (((C) & 0xc0) >> 6)); }) |
| 1197 | |
| 1198 | #define _mm256_permute2f128_pd(V1, V2, M) __extension__ ({ \ |
| 1199 | (__m256d)__builtin_ia32_vperm2f128_pd256((__v4df)(__m256d)(V1), \ |
| 1200 | (__v4df)(__m256d)(V2), (M)); }) |
| 1201 | |
| 1202 | #define _mm256_permute2f128_ps(V1, V2, M) __extension__ ({ \ |
| 1203 | (__m256)__builtin_ia32_vperm2f128_ps256((__v8sf)(__m256)(V1), \ |
| 1204 | (__v8sf)(__m256)(V2), (M)); }) |
| 1205 | |
| 1206 | #define _mm256_permute2f128_si256(V1, V2, M) __extension__ ({ \ |
| 1207 | (__m256i)__builtin_ia32_vperm2f128_si256((__v8si)(__m256i)(V1), \ |
| 1208 | (__v8si)(__m256i)(V2), (M)); }) |
| 1209 | |
| 1210 | /* Vector Blend */ |
| 1211 | #define _mm256_blend_pd(V1, V2, M) __extension__ ({ \ |
| 1212 | (__m256d)__builtin_shufflevector((__v4df)(__m256d)(V1), \ |
| 1213 | (__v4df)(__m256d)(V2), \ |
| 1214 | (((M) & 0x01) ? 4 : 0), \ |
| 1215 | (((M) & 0x02) ? 5 : 1), \ |
| 1216 | (((M) & 0x04) ? 6 : 2), \ |
| 1217 | (((M) & 0x08) ? 7 : 3)); }) |
| 1218 | |
| 1219 | #define _mm256_blend_ps(V1, V2, M) __extension__ ({ \ |
| 1220 | (__m256)__builtin_shufflevector((__v8sf)(__m256)(V1), \ |
| 1221 | (__v8sf)(__m256)(V2), \ |
| 1222 | (((M) & 0x01) ? 8 : 0), \ |
| 1223 | (((M) & 0x02) ? 9 : 1), \ |
| 1224 | (((M) & 0x04) ? 10 : 2), \ |
| 1225 | (((M) & 0x08) ? 11 : 3), \ |
| 1226 | (((M) & 0x10) ? 12 : 4), \ |
| 1227 | (((M) & 0x20) ? 13 : 5), \ |
| 1228 | (((M) & 0x40) ? 14 : 6), \ |
| 1229 | (((M) & 0x80) ? 15 : 7)); }) |
| 1230 | |
| 1231 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1232 | _mm256_blendv_pd(__m256d __a, __m256d __b, __m256d __c) |
| 1233 | { |
| 1234 | return (__m256d)__builtin_ia32_blendvpd256( |
| 1235 | (__v4df)__a, (__v4df)__b, (__v4df)__c); |
| 1236 | } |
| 1237 | |
| 1238 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1239 | _mm256_blendv_ps(__m256 __a, __m256 __b, __m256 __c) |
| 1240 | { |
| 1241 | return (__m256)__builtin_ia32_blendvps256( |
| 1242 | (__v8sf)__a, (__v8sf)__b, (__v8sf)__c); |
| 1243 | } |
| 1244 | |
| 1245 | /* Vector Dot Product */ |
| 1246 | #define _mm256_dp_ps(V1, V2, M) __extension__ ({ \ |
| 1247 | (__m256)__builtin_ia32_dpps256((__v8sf)(__m256)(V1), \ |
| 1248 | (__v8sf)(__m256)(V2), (M)); }) |
| 1249 | |
| 1250 | /* Vector shuffle */ |
| 1251 | #define _mm256_shuffle_ps(a, b, mask) __extension__ ({ \ |
| 1252 | (__m256)__builtin_shufflevector((__v8sf)(__m256)(a), \ |
| 1253 | (__v8sf)(__m256)(b), \ |
| 1254 | (mask) & 0x3, \ |
| 1255 | ((mask) & 0xc) >> 2, \ |
| 1256 | (((mask) & 0x30) >> 4) + 8, \ |
| 1257 | (((mask) & 0xc0) >> 6) + 8, \ |
| 1258 | ((mask) & 0x3) + 4, \ |
| 1259 | (((mask) & 0xc) >> 2) + 4, \ |
| 1260 | (((mask) & 0x30) >> 4) + 12, \ |
| 1261 | (((mask) & 0xc0) >> 6) + 12); }) |
| 1262 | |
| 1263 | #define _mm256_shuffle_pd(a, b, mask) __extension__ ({ \ |
| 1264 | (__m256d)__builtin_shufflevector((__v4df)(__m256d)(a), \ |
| 1265 | (__v4df)(__m256d)(b), \ |
| 1266 | (mask) & 0x1, \ |
| 1267 | (((mask) & 0x2) >> 1) + 4, \ |
| 1268 | (((mask) & 0x4) >> 2) + 2, \ |
| 1269 | (((mask) & 0x8) >> 3) + 6); }) |
| 1270 | |
| 1271 | /* Compare */ |
| 1272 | #define _CMP_EQ_OQ 0x00 /* Equal (ordered, non-signaling) */ |
| 1273 | #define _CMP_LT_OS 0x01 /* Less-than (ordered, signaling) */ |
| 1274 | #define _CMP_LE_OS 0x02 /* Less-than-or-equal (ordered, signaling) */ |
| 1275 | #define _CMP_UNORD_Q 0x03 /* Unordered (non-signaling) */ |
| 1276 | #define _CMP_NEQ_UQ 0x04 /* Not-equal (unordered, non-signaling) */ |
| 1277 | #define _CMP_NLT_US 0x05 /* Not-less-than (unordered, signaling) */ |
| 1278 | #define _CMP_NLE_US 0x06 /* Not-less-than-or-equal (unordered, signaling) */ |
| 1279 | #define _CMP_ORD_Q 0x07 /* Ordered (nonsignaling) */ |
| 1280 | #define _CMP_EQ_UQ 0x08 /* Equal (unordered, non-signaling) */ |
| 1281 | #define _CMP_NGE_US 0x09 /* Not-greater-than-or-equal (unord, signaling) */ |
| 1282 | #define _CMP_NGT_US 0x0a /* Not-greater-than (unordered, signaling) */ |
| 1283 | #define _CMP_FALSE_OQ 0x0b /* False (ordered, non-signaling) */ |
| 1284 | #define _CMP_NEQ_OQ 0x0c /* Not-equal (ordered, non-signaling) */ |
| 1285 | #define _CMP_GE_OS 0x0d /* Greater-than-or-equal (ordered, signaling) */ |
| 1286 | #define _CMP_GT_OS 0x0e /* Greater-than (ordered, signaling) */ |
| 1287 | #define _CMP_TRUE_UQ 0x0f /* True (unordered, non-signaling) */ |
| 1288 | #define _CMP_EQ_OS 0x10 /* Equal (ordered, signaling) */ |
| 1289 | #define _CMP_LT_OQ 0x11 /* Less-than (ordered, non-signaling) */ |
| 1290 | #define _CMP_LE_OQ 0x12 /* Less-than-or-equal (ordered, non-signaling) */ |
| 1291 | #define _CMP_UNORD_S 0x13 /* Unordered (signaling) */ |
| 1292 | #define _CMP_NEQ_US 0x14 /* Not-equal (unordered, signaling) */ |
| 1293 | #define _CMP_NLT_UQ 0x15 /* Not-less-than (unordered, non-signaling) */ |
| 1294 | #define _CMP_NLE_UQ 0x16 /* Not-less-than-or-equal (unord, non-signaling) */ |
| 1295 | #define _CMP_ORD_S 0x17 /* Ordered (signaling) */ |
| 1296 | #define _CMP_EQ_US 0x18 /* Equal (unordered, signaling) */ |
| 1297 | #define _CMP_NGE_UQ 0x19 /* Not-greater-than-or-equal (unord, non-sign) */ |
| 1298 | #define _CMP_NGT_UQ 0x1a /* Not-greater-than (unordered, non-signaling) */ |
| 1299 | #define _CMP_FALSE_OS 0x1b /* False (ordered, signaling) */ |
| 1300 | #define _CMP_NEQ_OS 0x1c /* Not-equal (ordered, signaling) */ |
| 1301 | #define _CMP_GE_OQ 0x1d /* Greater-than-or-equal (ordered, non-signaling) */ |
| 1302 | #define _CMP_GT_OQ 0x1e /* Greater-than (ordered, non-signaling) */ |
| 1303 | #define _CMP_TRUE_US 0x1f /* True (unordered, signaling) */ |
| 1304 | |
| 1305 | #define _mm_cmp_pd(a, b, c) __extension__ ({ \ |
| 1306 | (__m128d)__builtin_ia32_cmppd((__v2df)(__m128d)(a), \ |
| 1307 | (__v2df)(__m128d)(b), (c)); }) |
| 1308 | |
| 1309 | #define _mm_cmp_ps(a, b, c) __extension__ ({ \ |
| 1310 | (__m128)__builtin_ia32_cmpps((__v4sf)(__m128)(a), \ |
| 1311 | (__v4sf)(__m128)(b), (c)); }) |
| 1312 | |
| 1313 | #define _mm256_cmp_pd(a, b, c) __extension__ ({ \ |
| 1314 | (__m256d)__builtin_ia32_cmppd256((__v4df)(__m256d)(a), \ |
| 1315 | (__v4df)(__m256d)(b), (c)); }) |
| 1316 | |
| 1317 | #define _mm256_cmp_ps(a, b, c) __extension__ ({ \ |
| 1318 | (__m256)__builtin_ia32_cmpps256((__v8sf)(__m256)(a), \ |
| 1319 | (__v8sf)(__m256)(b), (c)); }) |
| 1320 | |
| 1321 | #define _mm_cmp_sd(a, b, c) __extension__ ({ \ |
| 1322 | (__m128d)__builtin_ia32_cmpsd((__v2df)(__m128d)(a), \ |
| 1323 | (__v2df)(__m128d)(b), (c)); }) |
| 1324 | |
| 1325 | #define _mm_cmp_ss(a, b, c) __extension__ ({ \ |
| 1326 | (__m128)__builtin_ia32_cmpss((__v4sf)(__m128)(a), \ |
| 1327 | (__v4sf)(__m128)(b), (c)); }) |
| 1328 | |
| 1329 | static __inline int __DEFAULT_FN_ATTRS |
| 1330 | _mm256_extract_epi32(__m256i __a, const int __imm) |
| 1331 | { |
| 1332 | __v8si __b = (__v8si)__a; |
| 1333 | return __b[__imm & 7]; |
| 1334 | } |
| 1335 | |
| 1336 | static __inline int __DEFAULT_FN_ATTRS |
| 1337 | _mm256_extract_epi16(__m256i __a, const int __imm) |
| 1338 | { |
| 1339 | __v16hi __b = (__v16hi)__a; |
| 1340 | return __b[__imm & 15]; |
| 1341 | } |
| 1342 | |
| 1343 | static __inline int __DEFAULT_FN_ATTRS |
| 1344 | _mm256_extract_epi8(__m256i __a, const int __imm) |
| 1345 | { |
| 1346 | __v32qi __b = (__v32qi)__a; |
| 1347 | return __b[__imm & 31]; |
| 1348 | } |
| 1349 | |
| 1350 | #ifdef __x86_64__ |
| 1351 | static __inline long long __DEFAULT_FN_ATTRS |
| 1352 | _mm256_extract_epi64(__m256i __a, const int __imm) |
| 1353 | { |
| 1354 | __v4di __b = (__v4di)__a; |
| 1355 | return __b[__imm & 3]; |
| 1356 | } |
| 1357 | #endif |
| 1358 | |
| 1359 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1360 | _mm256_insert_epi32(__m256i __a, int __b, int const __imm) |
| 1361 | { |
| 1362 | __v8si __c = (__v8si)__a; |
| 1363 | __c[__imm & 7] = __b; |
| 1364 | return (__m256i)__c; |
| 1365 | } |
| 1366 | |
| 1367 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1368 | _mm256_insert_epi16(__m256i __a, int __b, int const __imm) |
| 1369 | { |
| 1370 | __v16hi __c = (__v16hi)__a; |
| 1371 | __c[__imm & 15] = __b; |
| 1372 | return (__m256i)__c; |
| 1373 | } |
| 1374 | |
| 1375 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1376 | _mm256_insert_epi8(__m256i __a, int __b, int const __imm) |
| 1377 | { |
| 1378 | __v32qi __c = (__v32qi)__a; |
| 1379 | __c[__imm & 31] = __b; |
| 1380 | return (__m256i)__c; |
| 1381 | } |
| 1382 | |
| 1383 | #ifdef __x86_64__ |
| 1384 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1385 | _mm256_insert_epi64(__m256i __a, long long __b, int const __imm) |
| 1386 | { |
| 1387 | __v4di __c = (__v4di)__a; |
| 1388 | __c[__imm & 3] = __b; |
| 1389 | return (__m256i)__c; |
| 1390 | } |
| 1391 | #endif |
| 1392 | |
| 1393 | /* Conversion */ |
| 1394 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1395 | _mm256_cvtepi32_pd(__m128i __a) |
| 1396 | { |
| 1397 | return (__m256d)__builtin_ia32_cvtdq2pd256((__v4si) __a); |
| 1398 | } |
| 1399 | |
| 1400 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1401 | _mm256_cvtepi32_ps(__m256i __a) |
| 1402 | { |
| 1403 | return (__m256)__builtin_ia32_cvtdq2ps256((__v8si) __a); |
| 1404 | } |
| 1405 | |
| 1406 | static __inline __m128 __DEFAULT_FN_ATTRS |
| 1407 | _mm256_cvtpd_ps(__m256d __a) |
| 1408 | { |
| 1409 | return (__m128)__builtin_ia32_cvtpd2ps256((__v4df) __a); |
| 1410 | } |
| 1411 | |
| 1412 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1413 | _mm256_cvtps_epi32(__m256 __a) |
| 1414 | { |
| 1415 | return (__m256i)__builtin_ia32_cvtps2dq256((__v8sf) __a); |
| 1416 | } |
| 1417 | |
| 1418 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1419 | _mm256_cvtps_pd(__m128 __a) |
| 1420 | { |
| 1421 | return (__m256d)__builtin_ia32_cvtps2pd256((__v4sf) __a); |
| 1422 | } |
| 1423 | |
| 1424 | static __inline __m128i __DEFAULT_FN_ATTRS |
| 1425 | _mm256_cvttpd_epi32(__m256d __a) |
| 1426 | { |
| 1427 | return (__m128i)__builtin_ia32_cvttpd2dq256((__v4df) __a); |
| 1428 | } |
| 1429 | |
| 1430 | static __inline __m128i __DEFAULT_FN_ATTRS |
| 1431 | _mm256_cvtpd_epi32(__m256d __a) |
| 1432 | { |
| 1433 | return (__m128i)__builtin_ia32_cvtpd2dq256((__v4df) __a); |
| 1434 | } |
| 1435 | |
| 1436 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1437 | _mm256_cvttps_epi32(__m256 __a) |
| 1438 | { |
| 1439 | return (__m256i)__builtin_ia32_cvttps2dq256((__v8sf) __a); |
| 1440 | } |
| 1441 | |
| 1442 | /* Vector replicate */ |
| 1443 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1444 | _mm256_movehdup_ps(__m256 __a) |
| 1445 | { |
| 1446 | return __builtin_shufflevector(__a, __a, 1, 1, 3, 3, 5, 5, 7, 7); |
| 1447 | } |
| 1448 | |
| 1449 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1450 | _mm256_moveldup_ps(__m256 __a) |
| 1451 | { |
| 1452 | return __builtin_shufflevector(__a, __a, 0, 0, 2, 2, 4, 4, 6, 6); |
| 1453 | } |
| 1454 | |
| 1455 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1456 | _mm256_movedup_pd(__m256d __a) |
| 1457 | { |
| 1458 | return __builtin_shufflevector(__a, __a, 0, 0, 2, 2); |
| 1459 | } |
| 1460 | |
| 1461 | /* Unpack and Interleave */ |
| 1462 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1463 | _mm256_unpackhi_pd(__m256d __a, __m256d __b) |
| 1464 | { |
| 1465 | return __builtin_shufflevector(__a, __b, 1, 5, 1+2, 5+2); |
| 1466 | } |
| 1467 | |
| 1468 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1469 | _mm256_unpacklo_pd(__m256d __a, __m256d __b) |
| 1470 | { |
| 1471 | return __builtin_shufflevector(__a, __b, 0, 4, 0+2, 4+2); |
| 1472 | } |
| 1473 | |
| 1474 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1475 | _mm256_unpackhi_ps(__m256 __a, __m256 __b) |
| 1476 | { |
| 1477 | return __builtin_shufflevector(__a, __b, 2, 10, 2+1, 10+1, 6, 14, 6+1, 14+1); |
| 1478 | } |
| 1479 | |
| 1480 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1481 | _mm256_unpacklo_ps(__m256 __a, __m256 __b) |
| 1482 | { |
| 1483 | return __builtin_shufflevector(__a, __b, 0, 8, 0+1, 8+1, 4, 12, 4+1, 12+1); |
| 1484 | } |
| 1485 | |
| 1486 | /* Bit Test */ |
| 1487 | static __inline int __DEFAULT_FN_ATTRS |
| 1488 | _mm_testz_pd(__m128d __a, __m128d __b) |
| 1489 | { |
| 1490 | return __builtin_ia32_vtestzpd((__v2df)__a, (__v2df)__b); |
| 1491 | } |
| 1492 | |
| 1493 | static __inline int __DEFAULT_FN_ATTRS |
| 1494 | _mm_testc_pd(__m128d __a, __m128d __b) |
| 1495 | { |
| 1496 | return __builtin_ia32_vtestcpd((__v2df)__a, (__v2df)__b); |
| 1497 | } |
| 1498 | |
| 1499 | static __inline int __DEFAULT_FN_ATTRS |
| 1500 | _mm_testnzc_pd(__m128d __a, __m128d __b) |
| 1501 | { |
| 1502 | return __builtin_ia32_vtestnzcpd((__v2df)__a, (__v2df)__b); |
| 1503 | } |
| 1504 | |
| 1505 | static __inline int __DEFAULT_FN_ATTRS |
| 1506 | _mm_testz_ps(__m128 __a, __m128 __b) |
| 1507 | { |
| 1508 | return __builtin_ia32_vtestzps((__v4sf)__a, (__v4sf)__b); |
| 1509 | } |
| 1510 | |
| 1511 | static __inline int __DEFAULT_FN_ATTRS |
| 1512 | _mm_testc_ps(__m128 __a, __m128 __b) |
| 1513 | { |
| 1514 | return __builtin_ia32_vtestcps((__v4sf)__a, (__v4sf)__b); |
| 1515 | } |
| 1516 | |
| 1517 | static __inline int __DEFAULT_FN_ATTRS |
| 1518 | _mm_testnzc_ps(__m128 __a, __m128 __b) |
| 1519 | { |
| 1520 | return __builtin_ia32_vtestnzcps((__v4sf)__a, (__v4sf)__b); |
| 1521 | } |
| 1522 | |
| 1523 | static __inline int __DEFAULT_FN_ATTRS |
| 1524 | _mm256_testz_pd(__m256d __a, __m256d __b) |
| 1525 | { |
| 1526 | return __builtin_ia32_vtestzpd256((__v4df)__a, (__v4df)__b); |
| 1527 | } |
| 1528 | |
| 1529 | static __inline int __DEFAULT_FN_ATTRS |
| 1530 | _mm256_testc_pd(__m256d __a, __m256d __b) |
| 1531 | { |
| 1532 | return __builtin_ia32_vtestcpd256((__v4df)__a, (__v4df)__b); |
| 1533 | } |
| 1534 | |
| 1535 | static __inline int __DEFAULT_FN_ATTRS |
| 1536 | _mm256_testnzc_pd(__m256d __a, __m256d __b) |
| 1537 | { |
| 1538 | return __builtin_ia32_vtestnzcpd256((__v4df)__a, (__v4df)__b); |
| 1539 | } |
| 1540 | |
| 1541 | static __inline int __DEFAULT_FN_ATTRS |
| 1542 | _mm256_testz_ps(__m256 __a, __m256 __b) |
| 1543 | { |
| 1544 | return __builtin_ia32_vtestzps256((__v8sf)__a, (__v8sf)__b); |
| 1545 | } |
| 1546 | |
| 1547 | static __inline int __DEFAULT_FN_ATTRS |
| 1548 | _mm256_testc_ps(__m256 __a, __m256 __b) |
| 1549 | { |
| 1550 | return __builtin_ia32_vtestcps256((__v8sf)__a, (__v8sf)__b); |
| 1551 | } |
| 1552 | |
| 1553 | static __inline int __DEFAULT_FN_ATTRS |
| 1554 | _mm256_testnzc_ps(__m256 __a, __m256 __b) |
| 1555 | { |
| 1556 | return __builtin_ia32_vtestnzcps256((__v8sf)__a, (__v8sf)__b); |
| 1557 | } |
| 1558 | |
| 1559 | static __inline int __DEFAULT_FN_ATTRS |
| 1560 | _mm256_testz_si256(__m256i __a, __m256i __b) |
| 1561 | { |
| 1562 | return __builtin_ia32_ptestz256((__v4di)__a, (__v4di)__b); |
| 1563 | } |
| 1564 | |
| 1565 | static __inline int __DEFAULT_FN_ATTRS |
| 1566 | _mm256_testc_si256(__m256i __a, __m256i __b) |
| 1567 | { |
| 1568 | return __builtin_ia32_ptestc256((__v4di)__a, (__v4di)__b); |
| 1569 | } |
| 1570 | |
| 1571 | static __inline int __DEFAULT_FN_ATTRS |
| 1572 | _mm256_testnzc_si256(__m256i __a, __m256i __b) |
| 1573 | { |
| 1574 | return __builtin_ia32_ptestnzc256((__v4di)__a, (__v4di)__b); |
| 1575 | } |
| 1576 | |
| 1577 | /* Vector extract sign mask */ |
| 1578 | static __inline int __DEFAULT_FN_ATTRS |
| 1579 | _mm256_movemask_pd(__m256d __a) |
| 1580 | { |
| 1581 | return __builtin_ia32_movmskpd256((__v4df)__a); |
| 1582 | } |
| 1583 | |
| 1584 | static __inline int __DEFAULT_FN_ATTRS |
| 1585 | _mm256_movemask_ps(__m256 __a) |
| 1586 | { |
| 1587 | return __builtin_ia32_movmskps256((__v8sf)__a); |
| 1588 | } |
| 1589 | |
| 1590 | /* Vector __zero */ |
| 1591 | static __inline void __DEFAULT_FN_ATTRS |
| 1592 | _mm256_zeroall(void) |
| 1593 | { |
| 1594 | __builtin_ia32_vzeroall(); |
| 1595 | } |
| 1596 | |
| 1597 | static __inline void __DEFAULT_FN_ATTRS |
| 1598 | _mm256_zeroupper(void) |
| 1599 | { |
| 1600 | __builtin_ia32_vzeroupper(); |
| 1601 | } |
| 1602 | |
| 1603 | /* Vector load with broadcast */ |
| 1604 | static __inline __m128 __DEFAULT_FN_ATTRS |
| 1605 | _mm_broadcast_ss(float const *__a) |
| 1606 | { |
| 1607 | float __f = *__a; |
| 1608 | return (__m128)(__v4sf){ __f, __f, __f, __f }; |
| 1609 | } |
| 1610 | |
| 1611 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1612 | _mm256_broadcast_sd(double const *__a) |
| 1613 | { |
| 1614 | double __d = *__a; |
| 1615 | return (__m256d)(__v4df){ __d, __d, __d, __d }; |
| 1616 | } |
| 1617 | |
| 1618 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1619 | _mm256_broadcast_ss(float const *__a) |
| 1620 | { |
| 1621 | float __f = *__a; |
| 1622 | return (__m256)(__v8sf){ __f, __f, __f, __f, __f, __f, __f, __f }; |
| 1623 | } |
| 1624 | |
| 1625 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1626 | _mm256_broadcast_pd(__m128d const *__a) |
| 1627 | { |
| 1628 | return (__m256d)__builtin_ia32_vbroadcastf128_pd256(__a); |
| 1629 | } |
| 1630 | |
| 1631 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1632 | _mm256_broadcast_ps(__m128 const *__a) |
| 1633 | { |
| 1634 | return (__m256)__builtin_ia32_vbroadcastf128_ps256(__a); |
| 1635 | } |
| 1636 | |
| 1637 | /* SIMD load ops */ |
| 1638 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1639 | _mm256_load_pd(double const *__p) |
| 1640 | { |
| 1641 | return *(__m256d *)__p; |
| 1642 | } |
| 1643 | |
| 1644 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1645 | _mm256_load_ps(float const *__p) |
| 1646 | { |
| 1647 | return *(__m256 *)__p; |
| 1648 | } |
| 1649 | |
| 1650 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1651 | _mm256_loadu_pd(double const *__p) |
| 1652 | { |
| 1653 | struct __loadu_pd { |
| 1654 | __m256d __v; |
| 1655 | } __attribute__((__packed__, __may_alias__)); |
| 1656 | return ((struct __loadu_pd*)__p)->__v; |
| 1657 | } |
| 1658 | |
| 1659 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1660 | _mm256_loadu_ps(float const *__p) |
| 1661 | { |
| 1662 | struct __loadu_ps { |
| 1663 | __m256 __v; |
| 1664 | } __attribute__((__packed__, __may_alias__)); |
| 1665 | return ((struct __loadu_ps*)__p)->__v; |
| 1666 | } |
| 1667 | |
| 1668 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1669 | _mm256_load_si256(__m256i const *__p) |
| 1670 | { |
| 1671 | return *__p; |
| 1672 | } |
| 1673 | |
| 1674 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1675 | _mm256_loadu_si256(__m256i const *__p) |
| 1676 | { |
| 1677 | struct __loadu_si256 { |
| 1678 | __m256i __v; |
| 1679 | } __attribute__((__packed__, __may_alias__)); |
| 1680 | return ((struct __loadu_si256*)__p)->__v; |
| 1681 | } |
| 1682 | |
| 1683 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1684 | _mm256_lddqu_si256(__m256i const *__p) |
| 1685 | { |
| 1686 | return (__m256i)__builtin_ia32_lddqu256((char const *)__p); |
| 1687 | } |
| 1688 | |
| 1689 | /* SIMD store ops */ |
| 1690 | static __inline void __DEFAULT_FN_ATTRS |
| 1691 | _mm256_store_pd(double *__p, __m256d __a) |
| 1692 | { |
| 1693 | *(__m256d *)__p = __a; |
| 1694 | } |
| 1695 | |
| 1696 | static __inline void __DEFAULT_FN_ATTRS |
| 1697 | _mm256_store_ps(float *__p, __m256 __a) |
| 1698 | { |
| 1699 | *(__m256 *)__p = __a; |
| 1700 | } |
| 1701 | |
| 1702 | static __inline void __DEFAULT_FN_ATTRS |
| 1703 | _mm256_storeu_pd(double *__p, __m256d __a) |
| 1704 | { |
| 1705 | __builtin_ia32_storeupd256(__p, (__v4df)__a); |
| 1706 | } |
| 1707 | |
| 1708 | static __inline void __DEFAULT_FN_ATTRS |
| 1709 | _mm256_storeu_ps(float *__p, __m256 __a) |
| 1710 | { |
| 1711 | __builtin_ia32_storeups256(__p, (__v8sf)__a); |
| 1712 | } |
| 1713 | |
| 1714 | static __inline void __DEFAULT_FN_ATTRS |
| 1715 | _mm256_store_si256(__m256i *__p, __m256i __a) |
| 1716 | { |
| 1717 | *__p = __a; |
| 1718 | } |
| 1719 | |
| 1720 | static __inline void __DEFAULT_FN_ATTRS |
| 1721 | _mm256_storeu_si256(__m256i *__p, __m256i __a) |
| 1722 | { |
| 1723 | __builtin_ia32_storedqu256((char *)__p, (__v32qi)__a); |
| 1724 | } |
| 1725 | |
| 1726 | /* Conditional load ops */ |
| 1727 | static __inline __m128d __DEFAULT_FN_ATTRS |
| 1728 | _mm_maskload_pd(double const *__p, __m128i __m) |
| 1729 | { |
| 1730 | return (__m128d)__builtin_ia32_maskloadpd((const __v2df *)__p, (__v2di)__m); |
| 1731 | } |
| 1732 | |
| 1733 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1734 | _mm256_maskload_pd(double const *__p, __m256i __m) |
| 1735 | { |
| 1736 | return (__m256d)__builtin_ia32_maskloadpd256((const __v4df *)__p, |
| 1737 | (__v4di)__m); |
| 1738 | } |
| 1739 | |
| 1740 | static __inline __m128 __DEFAULT_FN_ATTRS |
| 1741 | _mm_maskload_ps(float const *__p, __m128i __m) |
| 1742 | { |
| 1743 | return (__m128)__builtin_ia32_maskloadps((const __v4sf *)__p, (__v4si)__m); |
| 1744 | } |
| 1745 | |
| 1746 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1747 | _mm256_maskload_ps(float const *__p, __m256i __m) |
| 1748 | { |
| 1749 | return (__m256)__builtin_ia32_maskloadps256((const __v8sf *)__p, (__v8si)__m); |
| 1750 | } |
| 1751 | |
| 1752 | /* Conditional store ops */ |
| 1753 | static __inline void __DEFAULT_FN_ATTRS |
| 1754 | _mm256_maskstore_ps(float *__p, __m256i __m, __m256 __a) |
| 1755 | { |
| 1756 | __builtin_ia32_maskstoreps256((__v8sf *)__p, (__v8si)__m, (__v8sf)__a); |
| 1757 | } |
| 1758 | |
| 1759 | static __inline void __DEFAULT_FN_ATTRS |
| 1760 | _mm_maskstore_pd(double *__p, __m128i __m, __m128d __a) |
| 1761 | { |
| 1762 | __builtin_ia32_maskstorepd((__v2df *)__p, (__v2di)__m, (__v2df)__a); |
| 1763 | } |
| 1764 | |
| 1765 | static __inline void __DEFAULT_FN_ATTRS |
| 1766 | _mm256_maskstore_pd(double *__p, __m256i __m, __m256d __a) |
| 1767 | { |
| 1768 | __builtin_ia32_maskstorepd256((__v4df *)__p, (__v4di)__m, (__v4df)__a); |
| 1769 | } |
| 1770 | |
| 1771 | static __inline void __DEFAULT_FN_ATTRS |
| 1772 | _mm_maskstore_ps(float *__p, __m128i __m, __m128 __a) |
| 1773 | { |
| 1774 | __builtin_ia32_maskstoreps((__v4sf *)__p, (__v4si)__m, (__v4sf)__a); |
| 1775 | } |
| 1776 | |
| 1777 | /* Cacheability support ops */ |
| 1778 | static __inline void __DEFAULT_FN_ATTRS |
| 1779 | _mm256_stream_si256(__m256i *__a, __m256i __b) |
| 1780 | { |
| 1781 | __builtin_ia32_movntdq256((__v4di *)__a, (__v4di)__b); |
| 1782 | } |
| 1783 | |
| 1784 | static __inline void __DEFAULT_FN_ATTRS |
| 1785 | _mm256_stream_pd(double *__a, __m256d __b) |
| 1786 | { |
| 1787 | __builtin_ia32_movntpd256(__a, (__v4df)__b); |
| 1788 | } |
| 1789 | |
| 1790 | static __inline void __DEFAULT_FN_ATTRS |
| 1791 | _mm256_stream_ps(float *__p, __m256 __a) |
| 1792 | { |
| 1793 | __builtin_ia32_movntps256(__p, (__v8sf)__a); |
| 1794 | } |
| 1795 | |
| 1796 | /* Create vectors */ |
| 1797 | static __inline__ __m256d __DEFAULT_FN_ATTRS |
| 1798 | _mm256_undefined_pd() |
| 1799 | { |
| 1800 | return (__m256d)__builtin_ia32_undef256(); |
| 1801 | } |
| 1802 | |
| 1803 | static __inline__ __m256 __DEFAULT_FN_ATTRS |
| 1804 | _mm256_undefined_ps() |
| 1805 | { |
| 1806 | return (__m256)__builtin_ia32_undef256(); |
| 1807 | } |
| 1808 | |
| 1809 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
| 1810 | _mm256_undefined_si256() |
| 1811 | { |
| 1812 | return (__m256i)__builtin_ia32_undef256(); |
| 1813 | } |
| 1814 | |
| 1815 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1816 | _mm256_set_pd(double __a, double __b, double __c, double __d) |
| 1817 | { |
| 1818 | return (__m256d){ __d, __c, __b, __a }; |
| 1819 | } |
| 1820 | |
| 1821 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1822 | _mm256_set_ps(float __a, float __b, float __c, float __d, |
| 1823 | float __e, float __f, float __g, float __h) |
| 1824 | { |
| 1825 | return (__m256){ __h, __g, __f, __e, __d, __c, __b, __a }; |
| 1826 | } |
| 1827 | |
| 1828 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1829 | _mm256_set_epi32(int __i0, int __i1, int __i2, int __i3, |
| 1830 | int __i4, int __i5, int __i6, int __i7) |
| 1831 | { |
| 1832 | return (__m256i)(__v8si){ __i7, __i6, __i5, __i4, __i3, __i2, __i1, __i0 }; |
| 1833 | } |
| 1834 | |
| 1835 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1836 | _mm256_set_epi16(short __w15, short __w14, short __w13, short __w12, |
| 1837 | short __w11, short __w10, short __w09, short __w08, |
| 1838 | short __w07, short __w06, short __w05, short __w04, |
| 1839 | short __w03, short __w02, short __w01, short __w00) |
| 1840 | { |
| 1841 | return (__m256i)(__v16hi){ __w00, __w01, __w02, __w03, __w04, __w05, __w06, |
| 1842 | __w07, __w08, __w09, __w10, __w11, __w12, __w13, __w14, __w15 }; |
| 1843 | } |
| 1844 | |
| 1845 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1846 | _mm256_set_epi8(char __b31, char __b30, char __b29, char __b28, |
| 1847 | char __b27, char __b26, char __b25, char __b24, |
| 1848 | char __b23, char __b22, char __b21, char __b20, |
| 1849 | char __b19, char __b18, char __b17, char __b16, |
| 1850 | char __b15, char __b14, char __b13, char __b12, |
| 1851 | char __b11, char __b10, char __b09, char __b08, |
| 1852 | char __b07, char __b06, char __b05, char __b04, |
| 1853 | char __b03, char __b02, char __b01, char __b00) |
| 1854 | { |
| 1855 | return (__m256i)(__v32qi){ |
| 1856 | __b00, __b01, __b02, __b03, __b04, __b05, __b06, __b07, |
| 1857 | __b08, __b09, __b10, __b11, __b12, __b13, __b14, __b15, |
| 1858 | __b16, __b17, __b18, __b19, __b20, __b21, __b22, __b23, |
| 1859 | __b24, __b25, __b26, __b27, __b28, __b29, __b30, __b31 |
| 1860 | }; |
| 1861 | } |
| 1862 | |
| 1863 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1864 | _mm256_set_epi64x(long long __a, long long __b, long long __c, long long __d) |
| 1865 | { |
| 1866 | return (__m256i)(__v4di){ __d, __c, __b, __a }; |
| 1867 | } |
| 1868 | |
| 1869 | /* Create vectors with elements in reverse order */ |
| 1870 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1871 | _mm256_setr_pd(double __a, double __b, double __c, double __d) |
| 1872 | { |
| 1873 | return (__m256d){ __a, __b, __c, __d }; |
| 1874 | } |
| 1875 | |
| 1876 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1877 | _mm256_setr_ps(float __a, float __b, float __c, float __d, |
| 1878 | float __e, float __f, float __g, float __h) |
| 1879 | { |
| 1880 | return (__m256){ __a, __b, __c, __d, __e, __f, __g, __h }; |
| 1881 | } |
| 1882 | |
| 1883 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1884 | _mm256_setr_epi32(int __i0, int __i1, int __i2, int __i3, |
| 1885 | int __i4, int __i5, int __i6, int __i7) |
| 1886 | { |
| 1887 | return (__m256i)(__v8si){ __i0, __i1, __i2, __i3, __i4, __i5, __i6, __i7 }; |
| 1888 | } |
| 1889 | |
| 1890 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1891 | _mm256_setr_epi16(short __w15, short __w14, short __w13, short __w12, |
| 1892 | short __w11, short __w10, short __w09, short __w08, |
| 1893 | short __w07, short __w06, short __w05, short __w04, |
| 1894 | short __w03, short __w02, short __w01, short __w00) |
| 1895 | { |
| 1896 | return (__m256i)(__v16hi){ __w15, __w14, __w13, __w12, __w11, __w10, __w09, |
| 1897 | __w08, __w07, __w06, __w05, __w04, __w03, __w02, __w01, __w00 }; |
| 1898 | } |
| 1899 | |
| 1900 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1901 | _mm256_setr_epi8(char __b31, char __b30, char __b29, char __b28, |
| 1902 | char __b27, char __b26, char __b25, char __b24, |
| 1903 | char __b23, char __b22, char __b21, char __b20, |
| 1904 | char __b19, char __b18, char __b17, char __b16, |
| 1905 | char __b15, char __b14, char __b13, char __b12, |
| 1906 | char __b11, char __b10, char __b09, char __b08, |
| 1907 | char __b07, char __b06, char __b05, char __b04, |
| 1908 | char __b03, char __b02, char __b01, char __b00) |
| 1909 | { |
| 1910 | return (__m256i)(__v32qi){ |
| 1911 | __b31, __b30, __b29, __b28, __b27, __b26, __b25, __b24, |
| 1912 | __b23, __b22, __b21, __b20, __b19, __b18, __b17, __b16, |
| 1913 | __b15, __b14, __b13, __b12, __b11, __b10, __b09, __b08, |
| 1914 | __b07, __b06, __b05, __b04, __b03, __b02, __b01, __b00 }; |
| 1915 | } |
| 1916 | |
| 1917 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1918 | _mm256_setr_epi64x(long long __a, long long __b, long long __c, long long __d) |
| 1919 | { |
| 1920 | return (__m256i)(__v4di){ __a, __b, __c, __d }; |
| 1921 | } |
| 1922 | |
| 1923 | /* Create vectors with repeated elements */ |
| 1924 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1925 | _mm256_set1_pd(double __w) |
| 1926 | { |
| 1927 | return (__m256d){ __w, __w, __w, __w }; |
| 1928 | } |
| 1929 | |
| 1930 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1931 | _mm256_set1_ps(float __w) |
| 1932 | { |
| 1933 | return (__m256){ __w, __w, __w, __w, __w, __w, __w, __w }; |
| 1934 | } |
| 1935 | |
| 1936 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1937 | _mm256_set1_epi32(int __i) |
| 1938 | { |
| 1939 | return (__m256i)(__v8si){ __i, __i, __i, __i, __i, __i, __i, __i }; |
| 1940 | } |
| 1941 | |
| 1942 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1943 | _mm256_set1_epi16(short __w) |
| 1944 | { |
| 1945 | return (__m256i)(__v16hi){ __w, __w, __w, __w, __w, __w, __w, __w, __w, __w, |
| 1946 | __w, __w, __w, __w, __w, __w }; |
| 1947 | } |
| 1948 | |
| 1949 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1950 | _mm256_set1_epi8(char __b) |
| 1951 | { |
| 1952 | return (__m256i)(__v32qi){ __b, __b, __b, __b, __b, __b, __b, __b, __b, __b, |
| 1953 | __b, __b, __b, __b, __b, __b, __b, __b, __b, __b, __b, __b, __b, __b, __b, |
| 1954 | __b, __b, __b, __b, __b, __b, __b }; |
| 1955 | } |
| 1956 | |
| 1957 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1958 | _mm256_set1_epi64x(long long __q) |
| 1959 | { |
| 1960 | return (__m256i)(__v4di){ __q, __q, __q, __q }; |
| 1961 | } |
| 1962 | |
| 1963 | /* Create __zeroed vectors */ |
| 1964 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1965 | _mm256_setzero_pd(void) |
| 1966 | { |
| 1967 | return (__m256d){ 0, 0, 0, 0 }; |
| 1968 | } |
| 1969 | |
| 1970 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1971 | _mm256_setzero_ps(void) |
| 1972 | { |
| 1973 | return (__m256){ 0, 0, 0, 0, 0, 0, 0, 0 }; |
| 1974 | } |
| 1975 | |
| 1976 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1977 | _mm256_setzero_si256(void) |
| 1978 | { |
| 1979 | return (__m256i){ 0LL, 0LL, 0LL, 0LL }; |
| 1980 | } |
| 1981 | |
| 1982 | /* Cast between vector types */ |
| 1983 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 1984 | _mm256_castpd_ps(__m256d __a) |
| 1985 | { |
| 1986 | return (__m256)__a; |
| 1987 | } |
| 1988 | |
| 1989 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 1990 | _mm256_castpd_si256(__m256d __a) |
| 1991 | { |
| 1992 | return (__m256i)__a; |
| 1993 | } |
| 1994 | |
| 1995 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 1996 | _mm256_castps_pd(__m256 __a) |
| 1997 | { |
| 1998 | return (__m256d)__a; |
| 1999 | } |
| 2000 | |
| 2001 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 2002 | _mm256_castps_si256(__m256 __a) |
| 2003 | { |
| 2004 | return (__m256i)__a; |
| 2005 | } |
| 2006 | |
| 2007 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 2008 | _mm256_castsi256_ps(__m256i __a) |
| 2009 | { |
| 2010 | return (__m256)__a; |
| 2011 | } |
| 2012 | |
| 2013 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 2014 | _mm256_castsi256_pd(__m256i __a) |
| 2015 | { |
| 2016 | return (__m256d)__a; |
| 2017 | } |
| 2018 | |
| 2019 | static __inline __m128d __DEFAULT_FN_ATTRS |
| 2020 | _mm256_castpd256_pd128(__m256d __a) |
| 2021 | { |
| 2022 | return __builtin_shufflevector(__a, __a, 0, 1); |
| 2023 | } |
| 2024 | |
| 2025 | static __inline __m128 __DEFAULT_FN_ATTRS |
| 2026 | _mm256_castps256_ps128(__m256 __a) |
| 2027 | { |
| 2028 | return __builtin_shufflevector(__a, __a, 0, 1, 2, 3); |
| 2029 | } |
| 2030 | |
| 2031 | static __inline __m128i __DEFAULT_FN_ATTRS |
| 2032 | _mm256_castsi256_si128(__m256i __a) |
| 2033 | { |
| 2034 | return __builtin_shufflevector(__a, __a, 0, 1); |
| 2035 | } |
| 2036 | |
| 2037 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 2038 | _mm256_castpd128_pd256(__m128d __a) |
| 2039 | { |
| 2040 | return __builtin_shufflevector(__a, __a, 0, 1, -1, -1); |
| 2041 | } |
| 2042 | |
| 2043 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 2044 | _mm256_castps128_ps256(__m128 __a) |
| 2045 | { |
| 2046 | return __builtin_shufflevector(__a, __a, 0, 1, 2, 3, -1, -1, -1, -1); |
| 2047 | } |
| 2048 | |
| 2049 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 2050 | _mm256_castsi128_si256(__m128i __a) |
| 2051 | { |
| 2052 | return __builtin_shufflevector(__a, __a, 0, 1, -1, -1); |
| 2053 | } |
| 2054 | |
| 2055 | /* |
| 2056 | Vector insert. |
| 2057 | We use macros rather than inlines because we only want to accept |
| 2058 | invocations where the immediate M is a constant expression. |
| 2059 | */ |
| 2060 | #define _mm256_insertf128_ps(V1, V2, M) __extension__ ({ \ |
| 2061 | (__m256)__builtin_shufflevector( \ |
| 2062 | (__v8sf)(__m256)(V1), \ |
| 2063 | (__v8sf)_mm256_castps128_ps256((__m128)(V2)), \ |
| 2064 | (((M) & 1) ? 0 : 8), \ |
| 2065 | (((M) & 1) ? 1 : 9), \ |
| 2066 | (((M) & 1) ? 2 : 10), \ |
| 2067 | (((M) & 1) ? 3 : 11), \ |
| 2068 | (((M) & 1) ? 8 : 4), \ |
| 2069 | (((M) & 1) ? 9 : 5), \ |
| 2070 | (((M) & 1) ? 10 : 6), \ |
| 2071 | (((M) & 1) ? 11 : 7) );}) |
| 2072 | |
| 2073 | #define _mm256_insertf128_pd(V1, V2, M) __extension__ ({ \ |
| 2074 | (__m256d)__builtin_shufflevector( \ |
| 2075 | (__v4df)(__m256d)(V1), \ |
| 2076 | (__v4df)_mm256_castpd128_pd256((__m128d)(V2)), \ |
| 2077 | (((M) & 1) ? 0 : 4), \ |
| 2078 | (((M) & 1) ? 1 : 5), \ |
| 2079 | (((M) & 1) ? 4 : 2), \ |
| 2080 | (((M) & 1) ? 5 : 3) );}) |
| 2081 | |
| 2082 | #define _mm256_insertf128_si256(V1, V2, M) __extension__ ({ \ |
| 2083 | (__m256i)__builtin_shufflevector( \ |
| 2084 | (__v4di)(__m256i)(V1), \ |
| 2085 | (__v4di)_mm256_castsi128_si256((__m128i)(V2)), \ |
| 2086 | (((M) & 1) ? 0 : 4), \ |
| 2087 | (((M) & 1) ? 1 : 5), \ |
| 2088 | (((M) & 1) ? 4 : 2), \ |
| 2089 | (((M) & 1) ? 5 : 3) );}) |
| 2090 | |
| 2091 | /* |
| 2092 | Vector extract. |
| 2093 | We use macros rather than inlines because we only want to accept |
| 2094 | invocations where the immediate M is a constant expression. |
| 2095 | */ |
| 2096 | #define _mm256_extractf128_ps(V, M) __extension__ ({ \ |
| 2097 | (__m128)__builtin_shufflevector( \ |
| 2098 | (__v8sf)(__m256)(V), \ |
| 2099 | (__v8sf)(_mm256_setzero_ps()), \ |
| 2100 | (((M) & 1) ? 4 : 0), \ |
| 2101 | (((M) & 1) ? 5 : 1), \ |
| 2102 | (((M) & 1) ? 6 : 2), \ |
| 2103 | (((M) & 1) ? 7 : 3) );}) |
| 2104 | |
| 2105 | #define _mm256_extractf128_pd(V, M) __extension__ ({ \ |
| 2106 | (__m128d)__builtin_shufflevector( \ |
| 2107 | (__v4df)(__m256d)(V), \ |
| 2108 | (__v4df)(_mm256_setzero_pd()), \ |
| 2109 | (((M) & 1) ? 2 : 0), \ |
| 2110 | (((M) & 1) ? 3 : 1) );}) |
| 2111 | |
| 2112 | #define _mm256_extractf128_si256(V, M) __extension__ ({ \ |
| 2113 | (__m128i)__builtin_shufflevector( \ |
| 2114 | (__v4di)(__m256i)(V), \ |
| 2115 | (__v4di)(_mm256_setzero_si256()), \ |
| 2116 | (((M) & 1) ? 2 : 0), \ |
| 2117 | (((M) & 1) ? 3 : 1) );}) |
| 2118 | |
| 2119 | /* SIMD load ops (unaligned) */ |
| 2120 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 2121 | _mm256_loadu2_m128(float const *__addr_hi, float const *__addr_lo) |
| 2122 | { |
| 2123 | struct __loadu_ps { |
| 2124 | __m128 __v; |
| 2125 | } __attribute__((__packed__, __may_alias__)); |
| 2126 | |
| 2127 | __m256 __v256 = _mm256_castps128_ps256(((struct __loadu_ps*)__addr_lo)->__v); |
| 2128 | return _mm256_insertf128_ps(__v256, ((struct __loadu_ps*)__addr_hi)->__v, 1); |
| 2129 | } |
| 2130 | |
| 2131 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 2132 | _mm256_loadu2_m128d(double const *__addr_hi, double const *__addr_lo) |
| 2133 | { |
| 2134 | struct __loadu_pd { |
| 2135 | __m128d __v; |
| 2136 | } __attribute__((__packed__, __may_alias__)); |
| 2137 | |
| 2138 | __m256d __v256 = _mm256_castpd128_pd256(((struct __loadu_pd*)__addr_lo)->__v); |
| 2139 | return _mm256_insertf128_pd(__v256, ((struct __loadu_pd*)__addr_hi)->__v, 1); |
| 2140 | } |
| 2141 | |
| 2142 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 2143 | _mm256_loadu2_m128i(__m128i const *__addr_hi, __m128i const *__addr_lo) |
| 2144 | { |
| 2145 | struct __loadu_si128 { |
| 2146 | __m128i __v; |
| 2147 | } __attribute__((__packed__, __may_alias__)); |
| 2148 | __m256i __v256 = _mm256_castsi128_si256( |
| 2149 | ((struct __loadu_si128*)__addr_lo)->__v); |
| 2150 | return _mm256_insertf128_si256(__v256, |
| 2151 | ((struct __loadu_si128*)__addr_hi)->__v, 1); |
| 2152 | } |
| 2153 | |
| 2154 | /* SIMD store ops (unaligned) */ |
| 2155 | static __inline void __DEFAULT_FN_ATTRS |
| 2156 | _mm256_storeu2_m128(float *__addr_hi, float *__addr_lo, __m256 __a) |
| 2157 | { |
| 2158 | __m128 __v128; |
| 2159 | |
| 2160 | __v128 = _mm256_castps256_ps128(__a); |
| 2161 | __builtin_ia32_storeups(__addr_lo, __v128); |
| 2162 | __v128 = _mm256_extractf128_ps(__a, 1); |
| 2163 | __builtin_ia32_storeups(__addr_hi, __v128); |
| 2164 | } |
| 2165 | |
| 2166 | static __inline void __DEFAULT_FN_ATTRS |
| 2167 | _mm256_storeu2_m128d(double *__addr_hi, double *__addr_lo, __m256d __a) |
| 2168 | { |
| 2169 | __m128d __v128; |
| 2170 | |
| 2171 | __v128 = _mm256_castpd256_pd128(__a); |
| 2172 | __builtin_ia32_storeupd(__addr_lo, __v128); |
| 2173 | __v128 = _mm256_extractf128_pd(__a, 1); |
| 2174 | __builtin_ia32_storeupd(__addr_hi, __v128); |
| 2175 | } |
| 2176 | |
| 2177 | static __inline void __DEFAULT_FN_ATTRS |
| 2178 | _mm256_storeu2_m128i(__m128i *__addr_hi, __m128i *__addr_lo, __m256i __a) |
| 2179 | { |
| 2180 | __m128i __v128; |
| 2181 | |
| 2182 | __v128 = _mm256_castsi256_si128(__a); |
| 2183 | __builtin_ia32_storedqu((char *)__addr_lo, (__v16qi)__v128); |
| 2184 | __v128 = _mm256_extractf128_si256(__a, 1); |
| 2185 | __builtin_ia32_storedqu((char *)__addr_hi, (__v16qi)__v128); |
| 2186 | } |
| 2187 | |
| 2188 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 2189 | _mm256_set_m128 (__m128 __hi, __m128 __lo) { |
| 2190 | return (__m256) __builtin_shufflevector(__lo, __hi, 0, 1, 2, 3, 4, 5, 6, 7); |
| 2191 | } |
| 2192 | |
| 2193 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 2194 | _mm256_set_m128d (__m128d __hi, __m128d __lo) { |
| 2195 | return (__m256d)_mm256_set_m128((__m128)__hi, (__m128)__lo); |
| 2196 | } |
| 2197 | |
| 2198 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 2199 | _mm256_set_m128i (__m128i __hi, __m128i __lo) { |
| 2200 | return (__m256i)_mm256_set_m128((__m128)__hi, (__m128)__lo); |
| 2201 | } |
| 2202 | |
| 2203 | static __inline __m256 __DEFAULT_FN_ATTRS |
| 2204 | _mm256_setr_m128 (__m128 __lo, __m128 __hi) { |
| 2205 | return _mm256_set_m128(__hi, __lo); |
| 2206 | } |
| 2207 | |
| 2208 | static __inline __m256d __DEFAULT_FN_ATTRS |
| 2209 | _mm256_setr_m128d (__m128d __lo, __m128d __hi) { |
| 2210 | return (__m256d)_mm256_set_m128((__m128)__hi, (__m128)__lo); |
| 2211 | } |
| 2212 | |
| 2213 | static __inline __m256i __DEFAULT_FN_ATTRS |
| 2214 | _mm256_setr_m128i (__m128i __lo, __m128i __hi) { |
| 2215 | return (__m256i)_mm256_set_m128((__m128)__hi, (__m128)__lo); |
| 2216 | } |
| 2217 | |
| 2218 | #undef __DEFAULT_FN_ATTRS |
| 2219 | |
| 2220 | #endif /* __AVXINTRIN_H */ |