1 /*===---- smmintrin.h - SSE4 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 _SMMINTRIN_H 25 #define _SMMINTRIN_H 26 27 #ifndef __SSE4_1__ 28 #error "SSE4.1 instruction set not enabled" 29 #else 30 31 #include <tmmintrin.h> 32 33 /* SSE4 Rounding macros. */ 34 #define _MM_FROUND_TO_NEAREST_INT 0x00 35 #define _MM_FROUND_TO_NEG_INF 0x01 36 #define _MM_FROUND_TO_POS_INF 0x02 37 #define _MM_FROUND_TO_ZERO 0x03 38 #define _MM_FROUND_CUR_DIRECTION 0x04 39 40 #define _MM_FROUND_RAISE_EXC 0x00 41 #define _MM_FROUND_NO_EXC 0x08 42 43 #define _MM_FROUND_NINT (_MM_FROUND_RAISE_EXC | _MM_FROUND_TO_NEAREST_INT) 44 #define _MM_FROUND_FLOOR (_MM_FROUND_RAISE_EXC | _MM_FROUND_TO_NEG_INF) 45 #define _MM_FROUND_CEIL (_MM_FROUND_RAISE_EXC | _MM_FROUND_TO_POS_INF) 46 #define _MM_FROUND_TRUNC (_MM_FROUND_RAISE_EXC | _MM_FROUND_TO_ZERO) 47 #define _MM_FROUND_RINT (_MM_FROUND_RAISE_EXC | _MM_FROUND_CUR_DIRECTION) 48 #define _MM_FROUND_NEARBYINT (_MM_FROUND_NO_EXC | _MM_FROUND_CUR_DIRECTION) 49 50 #define _mm_ceil_ps(X) _mm_round_ps((X), _MM_FROUND_CEIL) 51 #define _mm_ceil_pd(X) _mm_round_pd((X), _MM_FROUND_CEIL) 52 #define _mm_ceil_ss(X, Y) _mm_round_ss((X), (Y), _MM_FROUND_CEIL) 53 #define _mm_ceil_sd(X, Y) _mm_round_sd((X), (Y), _MM_FROUND_CEIL) 54 55 #define _mm_floor_ps(X) _mm_round_ps((X), _MM_FROUND_FLOOR) 56 #define _mm_floor_pd(X) _mm_round_pd((X), _MM_FROUND_FLOOR) 57 #define _mm_floor_ss(X, Y) _mm_round_ss((X), (Y), _MM_FROUND_FLOOR) 58 #define _mm_floor_sd(X, Y) _mm_round_sd((X), (Y), _MM_FROUND_FLOOR) 59 60 #define _mm_round_ps(X, Y) __builtin_ia32_roundps((X), (Y)) 61 #define _mm_round_ss(X, Y, M) __builtin_ia32_roundss((X), (Y), (M)) 62 #define _mm_round_pd(X, M) __builtin_ia32_roundpd((X), (M)) 63 #define _mm_round_sd(X, Y, M) __builtin_ia32_roundsd((X), (Y), (M)) 64 65 /* SSE4 Packed Blending Intrinsics. */ 66 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) 67 _mm_blend_pd (__m128d __V1, __m128d __V2, const int __M) 68 { 69 return (__m128d) __builtin_ia32_blendpd ((__v2df)__V1, (__v2df)__V2, __M); 70 } 71 72 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 73 _mm_blend_ps (__m128 __V1, __m128 __V2, const int __M) 74 { 75 return (__m128) __builtin_ia32_blendps ((__v4sf)__V1, (__v4sf)__V2, __M); 76 } 77 78 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) 79 _mm_blendv_pd (__m128d __V1, __m128d __V2, __m128d __M) 80 { 81 return (__m128d) __builtin_ia32_blendvpd ((__v2df)__V1, (__v2df)__V2, 82 (__v2df)__M); 83 } 84 85 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 86 _mm_blendv_ps (__m128 __V1, __m128 __V2, __m128 __M) 87 { 88 return (__m128) __builtin_ia32_blendvps ((__v4sf)__V1, (__v4sf)__V2, 89 (__v4sf)__M); 90 } 91 92 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 93 _mm_blendv_epi8 (__m128i __V1, __m128i __V2, __m128i __M) 94 { 95 return (__m128i) __builtin_ia32_pblendvb128 ((__v16qi)__V1, (__v16qi)__V2, 96 (__v16qi)__M); 97 } 98 99 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 100 _mm_blend_epi16 (__m128i __V1, __m128i __V2, const int __M) 101 { 102 return (__m128i) __builtin_ia32_pblendw128 ((__v8hi)__V1, (__v8hi)__V2, __M); 103 } 104 105 /* SSE4 Dword Multiply Instructions. */ 106 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 107 _mm_mullo_epi32 (__m128i __V1, __m128i __V2) 108 { 109 return (__m128i) ((__v4si)__V1 * (__v4si)__V2); 110 } 111 112 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 113 _mm_mul_epi32 (__m128i __V1, __m128i __V2) 114 { 115 return (__m128i) __builtin_ia32_pmuldq128 ((__v4si)__V1, (__v4si)__V2); 116 } 117 118 /* SSE4 Floating Point Dot Product Instructions. */ 119 #define _mm_dp_ps(X, Y, M) __builtin_ia32_dpps ((X), (Y), (M)) 120 #define _mm_dp_pd(X, Y, M) __builtin_ia32_dppd ((X), (Y), (M)) 121 122 /* SSE4 Streaming Load Hint Instruction. */ 123 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 124 _mm_stream_load_si128 (__m128i *__V) 125 { 126 return (__m128i) __builtin_ia32_movntdqa ((__v2di *) __V); 127 } 128 129 /* SSE4 Packed Integer Min/Max Instructions. */ 130 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 131 _mm_min_epi8 (__m128i __V1, __m128i __V2) 132 { 133 return (__m128i) __builtin_ia32_pminsb128 ((__v16qi) __V1, (__v16qi) __V2); 134 } 135 136 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 137 _mm_max_epi8 (__m128i __V1, __m128i __V2) 138 { 139 return (__m128i) __builtin_ia32_pmaxsb128 ((__v16qi) __V1, (__v16qi) __V2); 140 } 141 142 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 143 _mm_min_epu16 (__m128i __V1, __m128i __V2) 144 { 145 return (__m128i) __builtin_ia32_pminuw128 ((__v8hi) __V1, (__v8hi) __V2); 146 } 147 148 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 149 _mm_max_epu16 (__m128i __V1, __m128i __V2) 150 { 151 return (__m128i) __builtin_ia32_pmaxuw128 ((__v8hi) __V1, (__v8hi) __V2); 152 } 153 154 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 155 _mm_min_epi32 (__m128i __V1, __m128i __V2) 156 { 157 return (__m128i) __builtin_ia32_pminsd128 ((__v4si) __V1, (__v4si) __V2); 158 } 159 160 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 161 _mm_max_epi32 (__m128i __V1, __m128i __V2) 162 { 163 return (__m128i) __builtin_ia32_pmaxsd128 ((__v4si) __V1, (__v4si) __V2); 164 } 165 166 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 167 _mm_min_epu32 (__m128i __V1, __m128i __V2) 168 { 169 return (__m128i) __builtin_ia32_pminud128((__v4si) __V1, (__v4si) __V2); 170 } 171 172 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 173 _mm_max_epu32 (__m128i __V1, __m128i __V2) 174 { 175 return (__m128i) __builtin_ia32_pmaxud128((__v4si) __V1, (__v4si) __V2); 176 } 177 178 /* SSE4 Insertion and Extraction from XMM Register Instructions. */ 179 #define _mm_insert_ps(X, Y, N) __builtin_ia32_insertps128((X), (Y), (N)) 180 #define _mm_extract_ps(X, N) (__extension__ \ 181 ({ union { int i; float f; } __t; \ 182 __v4sf __a = (__v4sf)(X); \ 183 __t.f = __a[N]; \ 184 __t.i;})) 185 186 /* Miscellaneous insert and extract macros. */ 187 /* Extract a single-precision float from X at index N into D. */ 188 #define _MM_EXTRACT_FLOAT(D, X, N) (__extension__ ({ __v4sf __a = (__v4sf)(X); \ 189 (D) = __a[N]; })) 190 191 /* Or together 2 sets of indexes (X and Y) with the zeroing bits (Z) to create 192 an index suitable for _mm_insert_ps. */ 193 #define _MM_MK_INSERTPS_NDX(X, Y, Z) (((X) << 6) | ((Y) << 4) | (Z)) 194 195 /* Extract a float from X at index N into the first index of the return. */ 196 #define _MM_PICK_OUT_PS(X, N) _mm_insert_ps (_mm_setzero_ps(), (X), \ 197 _MM_MK_INSERTPS_NDX((N), 0, 0x0e)) 198 199 /* Insert int into packed integer array at index. */ 200 #define _mm_insert_epi8(X, I, N) (__extension__ ({ __v16qi __a = (__v16qi)(X); \ 201 __a[N] = I; \ 202 __a;})) 203 #define _mm_insert_epi32(X, I, N) (__extension__ ({ __v4si __a = (__v4si)(X); \ 204 __a[N] = I; \ 205 __a;})) 206 #ifdef __x86_64__ 207 #define _mm_insert_epi64(X, I, N) (__extension__ ({ __v2di __a = (__v2di)(X); \ 208 __a[N] = I; \ 209 __a;})) 210 #endif /* __x86_64__ */ 211 212 /* Extract int from packed integer array at index. This returns the element 213 * as a zero extended value, so it is unsigned. 214 */ 215 #define _mm_extract_epi8(X, N) (__extension__ ({ __v16qi __a = (__v16qi)(X); \ 216 (unsigned char)__a[N];})) 217 #define _mm_extract_epi32(X, N) (__extension__ ({ __v4si __a = (__v4si)(X); \ 218 (unsigned)__a[N];})) 219 #ifdef __x86_64__ 220 #define _mm_extract_epi64(X, N) (__extension__ ({ __v2di __a = (__v2di)(X); \ 221 __a[N];})) 222 #endif /* __x86_64 */ 223 224 /* SSE4 128-bit Packed Integer Comparisons. */ 225 static __inline__ int __attribute__((__always_inline__, __nodebug__)) 226 _mm_testz_si128(__m128i __M, __m128i __V) 227 { 228 return __builtin_ia32_ptestz128((__v2di)__M, (__v2di)__V); 229 } 230 231 static __inline__ int __attribute__((__always_inline__, __nodebug__)) 232 _mm_testc_si128(__m128i __M, __m128i __V) 233 { 234 return __builtin_ia32_ptestc128((__v2di)__M, (__v2di)__V); 235 } 236 237 static __inline__ int __attribute__((__always_inline__, __nodebug__)) 238 _mm_testnzc_si128(__m128i __M, __m128i __V) 239 { 240 return __builtin_ia32_ptestnzc128((__v2di)__M, (__v2di)__V); 241 } 242 243 #define _mm_test_all_ones(V) _mm_testc_si128((V), _mm_cmpeq_epi32((V), (V))) 244 #define _mm_test_mix_ones_zeros(M, V) _mm_testnzc_si128((M), (V)) 245 #define _mm_test_all_zeros(M, V) _mm_testz_si128 ((V), (V)) 246 247 /* SSE4 64-bit Packed Integer Comparisons. */ 248 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 249 _mm_cmpeq_epi64(__m128i __V1, __m128i __V2) 250 { 251 return (__m128i) __builtin_ia32_pcmpeqq((__v2di)__V1, (__v2di)__V2); 252 } 253 254 /* SSE4 Packed Integer Sign-Extension. */ 255 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 256 _mm_cvtepi8_epi16(__m128i __V) 257 { 258 return (__m128i) __builtin_ia32_pmovsxbw128((__v16qi) __V); 259 } 260 261 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 262 _mm_cvtepi8_epi32(__m128i __V) 263 { 264 return (__m128i) __builtin_ia32_pmovsxbd128((__v16qi) __V); 265 } 266 267 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 268 _mm_cvtepi8_epi64(__m128i __V) 269 { 270 return (__m128i) __builtin_ia32_pmovsxbq128((__v16qi) __V); 271 } 272 273 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 274 _mm_cvtepi16_epi32(__m128i __V) 275 { 276 return (__m128i) __builtin_ia32_pmovsxwd128((__v8hi) __V); 277 } 278 279 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 280 _mm_cvtepi16_epi64(__m128i __V) 281 { 282 return (__m128i) __builtin_ia32_pmovsxwq128((__v8hi)__V); 283 } 284 285 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 286 _mm_cvtepi32_epi64(__m128i __V) 287 { 288 return (__m128i) __builtin_ia32_pmovsxdq128((__v4si)__V); 289 } 290 291 /* SSE4 Packed Integer Zero-Extension. */ 292 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 293 _mm_cvtepu8_epi16(__m128i __V) 294 { 295 return (__m128i) __builtin_ia32_pmovzxbw128((__v16qi) __V); 296 } 297 298 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 299 _mm_cvtepu8_epi32(__m128i __V) 300 { 301 return (__m128i) __builtin_ia32_pmovzxbd128((__v16qi)__V); 302 } 303 304 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 305 _mm_cvtepu8_epi64(__m128i __V) 306 { 307 return (__m128i) __builtin_ia32_pmovzxbq128((__v16qi)__V); 308 } 309 310 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 311 _mm_cvtepu16_epi32(__m128i __V) 312 { 313 return (__m128i) __builtin_ia32_pmovzxwd128((__v8hi)__V); 314 } 315 316 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 317 _mm_cvtepu16_epi64(__m128i __V) 318 { 319 return (__m128i) __builtin_ia32_pmovzxwq128((__v8hi)__V); 320 } 321 322 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 323 _mm_cvtepu32_epi64(__m128i __V) 324 { 325 return (__m128i) __builtin_ia32_pmovzxdq128((__v4si)__V); 326 } 327 328 /* SSE4 Pack with Unsigned Saturation. */ 329 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 330 _mm_packus_epi32(__m128i __V1, __m128i __V2) 331 { 332 return (__m128i) __builtin_ia32_packusdw128((__v4si)__V1, (__v4si)__V2); 333 } 334 335 /* SSE4 Multiple Packed Sums of Absolute Difference. */ 336 #define _mm_mpsadbw_epu8(X, Y, M) __builtin_ia32_mpsadbw128((X), (Y), (M)) 337 338 /* These definitions are normally in nmmintrin.h, but gcc puts them in here 339 so we'll do the same. */ 340 #ifdef __SSE4_2__ 341 342 /* These specify the type of data that we're comparing. */ 343 #define _SIDD_UBYTE_OPS 0x00 344 #define _SIDD_UWORD_OPS 0x01 345 #define _SIDD_SBYTE_OPS 0x02 346 #define _SIDD_SWORD_OPS 0x03 347 348 /* These specify the type of comparison operation. */ 349 #define _SIDD_CMP_EQUAL_ANY 0x00 350 #define _SIDD_CMP_RANGES 0x04 351 #define _SIDD_CMP_EQUAL_EACH 0x08 352 #define _SIDD_CMP_EQUAL_ORDERED 0x0c 353 354 /* These macros specify the polarity of the operation. */ 355 #define _SIDD_POSITIVE_POLARITY 0x00 356 #define _SIDD_NEGATIVE_POLARITY 0x10 357 #define _SIDD_MASKED_POSITIVE_POLARITY 0x20 358 #define _SIDD_MASKED_NEGATIVE_POLARITY 0x30 359 360 /* These macros are used in _mm_cmpXstri() to specify the return. */ 361 #define _SIDD_LEAST_SIGNIFICANT 0x00 362 #define _SIDD_MOST_SIGNIFICANT 0x40 363 364 /* These macros are used in _mm_cmpXstri() to specify the return. */ 365 #define _SIDD_BIT_MASK 0x00 366 #define _SIDD_UNIT_MASK 0x40 367 368 /* SSE4.2 Packed Comparison Intrinsics. */ 369 #define _mm_cmpistrm(A, B, M) __builtin_ia32_pcmpistrm128((A), (B), (M)) 370 #define _mm_cmpistri(A, B, M) __builtin_ia32_pcmpistri128((A), (B), (M)) 371 372 #define _mm_cmpestrm(A, LA, B, LB, M) \ 373 __builtin_ia32_pcmpestrm128((A), (LA), (B), (LB), (M)) 374 #define _mm_cmpestri(X, LX, Y, LY, M) \ 375 __builtin_ia32_pcmpestri128((A), (LA), (B), (LB), (M)) 376 377 /* SSE4.2 Packed Comparison Intrinsics and EFlag Reading. */ 378 #define _mm_cmpistra(A, LA, B, LB, M) \ 379 __builtin_ia32_pcmpistria128((A), (LA), (B), (LB), (M)) 380 #define _mm_cmpistrc(A, LA, B, LB, M) \ 381 __builtin_ia32_pcmpistric128((A), (LA), (B), (LB), (M)) 382 #define _mm_cmpistro(A, LA, B, LB, M) \ 383 __builtin_ia32_pcmpistrio128((A), (LA), (B), (LB), (M)) 384 #define _mm_cmpistrs(A, LA, B, LB, M) \ 385 __builtin_ia32_pcmpistris128((A), (LA), (B), (LB), (M)) 386 #define _mm_cmpistrz(A, LA, B, LB, M) \ 387 __builtin_ia32_pcmpistriz128((A), (LA), (B), (LB), (M)) 388 389 #define _mm_cmpestra(A, LA, B, LB, M) \ 390 __builtin_ia32_pcmpestria128((A), (LA), (B), (LB), (M)) 391 #define _mm_cmpestrc(A, LA, B, LB, M) \ 392 __builtin_ia32_pcmpestric128((A), (LA), (B), (LB), (M)) 393 #define _mm_cmpestro(A, LA, B, LB, M) \ 394 __builtin_ia32_pcmpestrio128((A), (LA), (B), (LB), (M)) 395 #define _mm_cmpestrs(A, LA, B, LB, M) \ 396 __builtin_ia32_pcmpestris128((A), (LA), (B), (LB), (M)) 397 #define _mm_cmpestrz(A, LA, B, LB, M) \ 398 __builtin_ia32_pcmpestriz128((A), (LA), (B), (LB), (M)) 399 400 /* SSE4.2 Compare Packed Data -- Greater Than. */ 401 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 402 _mm_cmpgt_epi64(__m128i __V1, __m128i __V2) 403 { 404 return __builtin_ia32_pcmpgtq((__v2di)__V1, (__v2di)__V2); 405 } 406 407 /* SSE4.2 Accumulate CRC32. */ 408 static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) 409 _mm_crc32_u8(unsigned int __C, unsigned char __D) 410 { 411 return __builtin_ia32_crc32qi(__C, __D); 412 } 413 414 static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) 415 _mm_crc32_u16(unsigned int __C, unsigned short __D) 416 { 417 return __builtin_ia32_crc32hi(__C, __D); 418 } 419 420 static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) 421 _mm_crc32_u32(unsigned int __C, unsigned int __D) 422 { 423 return __builtin_ia32_crc32si(__C, __D); 424 } 425 426 #ifdef __x86_64__ 427 static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__)) 428 _mm_crc32_u64(unsigned long long __C, unsigned long long __D) 429 { 430 return __builtin_ia32_crc32di(__C, __D); 431 } 432 #endif /* __x86_64__ */ 433 434 /* SSE4.2 Population Count. */ 435 static __inline__ int __attribute__((__always_inline__, __nodebug__)) 436 _mm_popcnt_u32(unsigned int __A) 437 { 438 return __builtin_popcount(__A); 439 } 440 441 #ifdef __x86_64__ 442 static __inline__ long long __attribute__((__always_inline__, __nodebug__)) 443 _mm_popcnt_u64(unsigned long long __A) 444 { 445 return __builtin_popcountll(__A); 446 } 447 #endif /* __x86_64__ */ 448 449 #endif /* __SSE4_2__ */ 450 #endif /* __SSE4_1__ */ 451 452 #endif /* _SMMINTRIN_H */ 453
