| 1 |
/* Copyright (C) 2011-2015 Free Software Foundation, Inc. |
| 2 |
|
| 3 |
This file is part of GCC. |
| 4 |
|
| 5 |
GCC is free software; you can redistribute it and/or modify |
| 6 |
it under the terms of the GNU General Public License as published by |
| 7 |
the Free Software Foundation; either version 3, or (at your option) |
| 8 |
any later version. |
| 9 |
|
| 10 |
GCC is distributed in the hope that it will be useful, |
| 11 |
but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 13 |
GNU General Public License for more details. |
| 14 |
|
| 15 |
Under Section 7 of GPL version 3, you are granted additional |
| 16 |
permissions described in the GCC Runtime Library Exception, version |
| 17 |
3.1, as published by the Free Software Foundation. |
| 18 |
|
| 19 |
You should have received a copy of the GNU General Public License and |
| 20 |
a copy of the GCC Runtime Library Exception along with this program; |
| 21 |
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see |
| 22 |
<http://www.gnu.org/licenses/>. */ |
| 23 |
|
| 24 |
#ifndef _IMMINTRIN_H_INCLUDED |
| 25 |
# error "Never use <fmaintrin.h> directly; include <immintrin.h> instead." |
| 26 |
#endif |
| 27 |
|
| 28 |
#ifndef _FMAINTRIN_H_INCLUDED |
| 29 |
#define _FMAINTRIN_H_INCLUDED |
| 30 |
|
| 31 |
#ifndef __FMA__ |
| 32 |
#pragma GCC push_options |
| 33 |
#pragma GCC target("fma") |
| 34 |
#define __DISABLE_FMA__ |
| 35 |
#endif /* __FMA__ */ |
| 36 |
|
| 37 |
extern __inline __m128d |
| 38 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 39 |
_mm_fmadd_pd (__m128d __A, __m128d __B, __m128d __C) |
| 40 |
{ |
| 41 |
return (__m128d)__builtin_ia32_vfmaddpd ((__v2df)__A, (__v2df)__B, |
| 42 |
(__v2df)__C); |
| 43 |
} |
| 44 |
|
| 45 |
extern __inline __m256d |
| 46 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 47 |
_mm256_fmadd_pd (__m256d __A, __m256d __B, __m256d __C) |
| 48 |
{ |
| 49 |
return (__m256d)__builtin_ia32_vfmaddpd256 ((__v4df)__A, (__v4df)__B, |
| 50 |
(__v4df)__C); |
| 51 |
} |
| 52 |
|
| 53 |
extern __inline __m128 |
| 54 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 55 |
_mm_fmadd_ps (__m128 __A, __m128 __B, __m128 __C) |
| 56 |
{ |
| 57 |
return (__m128)__builtin_ia32_vfmaddps ((__v4sf)__A, (__v4sf)__B, |
| 58 |
(__v4sf)__C); |
| 59 |
} |
| 60 |
|
| 61 |
extern __inline __m256 |
| 62 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 63 |
_mm256_fmadd_ps (__m256 __A, __m256 __B, __m256 __C) |
| 64 |
{ |
| 65 |
return (__m256)__builtin_ia32_vfmaddps256 ((__v8sf)__A, (__v8sf)__B, |
| 66 |
(__v8sf)__C); |
| 67 |
} |
| 68 |
|
| 69 |
extern __inline __m128d |
| 70 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 71 |
_mm_fmadd_sd (__m128d __A, __m128d __B, __m128d __C) |
| 72 |
{ |
| 73 |
return (__m128d) __builtin_ia32_vfmaddsd3 ((__v2df)__A, (__v2df)__B, |
| 74 |
(__v2df)__C); |
| 75 |
} |
| 76 |
|
| 77 |
extern __inline __m128 |
| 78 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 79 |
_mm_fmadd_ss (__m128 __A, __m128 __B, __m128 __C) |
| 80 |
{ |
| 81 |
return (__m128) __builtin_ia32_vfmaddss3 ((__v4sf)__A, (__v4sf)__B, |
| 82 |
(__v4sf)__C); |
| 83 |
} |
| 84 |
|
| 85 |
extern __inline __m128d |
| 86 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 87 |
_mm_fmsub_pd (__m128d __A, __m128d __B, __m128d __C) |
| 88 |
{ |
| 89 |
return (__m128d)__builtin_ia32_vfmaddpd ((__v2df)__A, (__v2df)__B, |
| 90 |
-(__v2df)__C); |
| 91 |
} |
| 92 |
|
| 93 |
extern __inline __m256d |
| 94 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 95 |
_mm256_fmsub_pd (__m256d __A, __m256d __B, __m256d __C) |
| 96 |
{ |
| 97 |
return (__m256d)__builtin_ia32_vfmaddpd256 ((__v4df)__A, (__v4df)__B, |
| 98 |
-(__v4df)__C); |
| 99 |
} |
| 100 |
|
| 101 |
extern __inline __m128 |
| 102 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 103 |
_mm_fmsub_ps (__m128 __A, __m128 __B, __m128 __C) |
| 104 |
{ |
| 105 |
return (__m128)__builtin_ia32_vfmaddps ((__v4sf)__A, (__v4sf)__B, |
| 106 |
-(__v4sf)__C); |
| 107 |
} |
| 108 |
|
| 109 |
extern __inline __m256 |
| 110 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 111 |
_mm256_fmsub_ps (__m256 __A, __m256 __B, __m256 __C) |
| 112 |
{ |
| 113 |
return (__m256)__builtin_ia32_vfmaddps256 ((__v8sf)__A, (__v8sf)__B, |
| 114 |
-(__v8sf)__C); |
| 115 |
} |
| 116 |
|
| 117 |
extern __inline __m128d |
| 118 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 119 |
_mm_fmsub_sd (__m128d __A, __m128d __B, __m128d __C) |
| 120 |
{ |
| 121 |
return (__m128d)__builtin_ia32_vfmaddsd3 ((__v2df)__A, (__v2df)__B, |
| 122 |
-(__v2df)__C); |
| 123 |
} |
| 124 |
|
| 125 |
extern __inline __m128 |
| 126 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 127 |
_mm_fmsub_ss (__m128 __A, __m128 __B, __m128 __C) |
| 128 |
{ |
| 129 |
return (__m128)__builtin_ia32_vfmaddss3 ((__v4sf)__A, (__v4sf)__B, |
| 130 |
-(__v4sf)__C); |
| 131 |
} |
| 132 |
|
| 133 |
extern __inline __m128d |
| 134 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 135 |
_mm_fnmadd_pd (__m128d __A, __m128d __B, __m128d __C) |
| 136 |
{ |
| 137 |
return (__m128d)__builtin_ia32_vfmaddpd (-(__v2df)__A, (__v2df)__B, |
| 138 |
(__v2df)__C); |
| 139 |
} |
| 140 |
|
| 141 |
extern __inline __m256d |
| 142 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 143 |
_mm256_fnmadd_pd (__m256d __A, __m256d __B, __m256d __C) |
| 144 |
{ |
| 145 |
return (__m256d)__builtin_ia32_vfmaddpd256 (-(__v4df)__A, (__v4df)__B, |
| 146 |
(__v4df)__C); |
| 147 |
} |
| 148 |
|
| 149 |
extern __inline __m128 |
| 150 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 151 |
_mm_fnmadd_ps (__m128 __A, __m128 __B, __m128 __C) |
| 152 |
{ |
| 153 |
return (__m128)__builtin_ia32_vfmaddps (-(__v4sf)__A, (__v4sf)__B, |
| 154 |
(__v4sf)__C); |
| 155 |
} |
| 156 |
|
| 157 |
extern __inline __m256 |
| 158 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 159 |
_mm256_fnmadd_ps (__m256 __A, __m256 __B, __m256 __C) |
| 160 |
{ |
| 161 |
return (__m256)__builtin_ia32_vfmaddps256 (-(__v8sf)__A, (__v8sf)__B, |
| 162 |
(__v8sf)__C); |
| 163 |
} |
| 164 |
|
| 165 |
extern __inline __m128d |
| 166 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 167 |
_mm_fnmadd_sd (__m128d __A, __m128d __B, __m128d __C) |
| 168 |
{ |
| 169 |
return (__m128d)__builtin_ia32_vfmaddsd3 ((__v2df)__A, -(__v2df)__B, |
| 170 |
(__v2df)__C); |
| 171 |
} |
| 172 |
|
| 173 |
extern __inline __m128 |
| 174 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 175 |
_mm_fnmadd_ss (__m128 __A, __m128 __B, __m128 __C) |
| 176 |
{ |
| 177 |
return (__m128)__builtin_ia32_vfmaddss3 ((__v4sf)__A, -(__v4sf)__B, |
| 178 |
(__v4sf)__C); |
| 179 |
} |
| 180 |
|
| 181 |
extern __inline __m128d |
| 182 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 183 |
_mm_fnmsub_pd (__m128d __A, __m128d __B, __m128d __C) |
| 184 |
{ |
| 185 |
return (__m128d)__builtin_ia32_vfmaddpd (-(__v2df)__A, (__v2df)__B, |
| 186 |
-(__v2df)__C); |
| 187 |
} |
| 188 |
|
| 189 |
extern __inline __m256d |
| 190 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 191 |
_mm256_fnmsub_pd (__m256d __A, __m256d __B, __m256d __C) |
| 192 |
{ |
| 193 |
return (__m256d)__builtin_ia32_vfmaddpd256 (-(__v4df)__A, (__v4df)__B, |
| 194 |
-(__v4df)__C); |
| 195 |
} |
| 196 |
|
| 197 |
extern __inline __m128 |
| 198 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 199 |
_mm_fnmsub_ps (__m128 __A, __m128 __B, __m128 __C) |
| 200 |
{ |
| 201 |
return (__m128)__builtin_ia32_vfmaddps (-(__v4sf)__A, (__v4sf)__B, |
| 202 |
-(__v4sf)__C); |
| 203 |
} |
| 204 |
|
| 205 |
extern __inline __m256 |
| 206 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 207 |
_mm256_fnmsub_ps (__m256 __A, __m256 __B, __m256 __C) |
| 208 |
{ |
| 209 |
return (__m256)__builtin_ia32_vfmaddps256 (-(__v8sf)__A, (__v8sf)__B, |
| 210 |
-(__v8sf)__C); |
| 211 |
} |
| 212 |
|
| 213 |
extern __inline __m128d |
| 214 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 215 |
_mm_fnmsub_sd (__m128d __A, __m128d __B, __m128d __C) |
| 216 |
{ |
| 217 |
return (__m128d)__builtin_ia32_vfmaddsd3 ((__v2df)__A, -(__v2df)__B, |
| 218 |
-(__v2df)__C); |
| 219 |
} |
| 220 |
|
| 221 |
extern __inline __m128 |
| 222 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 223 |
_mm_fnmsub_ss (__m128 __A, __m128 __B, __m128 __C) |
| 224 |
{ |
| 225 |
return (__m128)__builtin_ia32_vfmaddss3 ((__v4sf)__A, -(__v4sf)__B, |
| 226 |
-(__v4sf)__C); |
| 227 |
} |
| 228 |
|
| 229 |
extern __inline __m128d |
| 230 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 231 |
_mm_fmaddsub_pd (__m128d __A, __m128d __B, __m128d __C) |
| 232 |
{ |
| 233 |
return (__m128d)__builtin_ia32_vfmaddsubpd ((__v2df)__A, (__v2df)__B, |
| 234 |
(__v2df)__C); |
| 235 |
} |
| 236 |
|
| 237 |
extern __inline __m256d |
| 238 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 239 |
_mm256_fmaddsub_pd (__m256d __A, __m256d __B, __m256d __C) |
| 240 |
{ |
| 241 |
return (__m256d)__builtin_ia32_vfmaddsubpd256 ((__v4df)__A, |
| 242 |
(__v4df)__B, |
| 243 |
(__v4df)__C); |
| 244 |
} |
| 245 |
|
| 246 |
extern __inline __m128 |
| 247 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 248 |
_mm_fmaddsub_ps (__m128 __A, __m128 __B, __m128 __C) |
| 249 |
{ |
| 250 |
return (__m128)__builtin_ia32_vfmaddsubps ((__v4sf)__A, (__v4sf)__B, |
| 251 |
(__v4sf)__C); |
| 252 |
} |
| 253 |
|
| 254 |
extern __inline __m256 |
| 255 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 256 |
_mm256_fmaddsub_ps (__m256 __A, __m256 __B, __m256 __C) |
| 257 |
{ |
| 258 |
return (__m256)__builtin_ia32_vfmaddsubps256 ((__v8sf)__A, |
| 259 |
(__v8sf)__B, |
| 260 |
(__v8sf)__C); |
| 261 |
} |
| 262 |
|
| 263 |
extern __inline __m128d |
| 264 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 265 |
_mm_fmsubadd_pd (__m128d __A, __m128d __B, __m128d __C) |
| 266 |
{ |
| 267 |
return (__m128d)__builtin_ia32_vfmaddsubpd ((__v2df)__A, (__v2df)__B, |
| 268 |
-(__v2df)__C); |
| 269 |
} |
| 270 |
|
| 271 |
extern __inline __m256d |
| 272 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 273 |
_mm256_fmsubadd_pd (__m256d __A, __m256d __B, __m256d __C) |
| 274 |
{ |
| 275 |
return (__m256d)__builtin_ia32_vfmaddsubpd256 ((__v4df)__A, |
| 276 |
(__v4df)__B, |
| 277 |
-(__v4df)__C); |
| 278 |
} |
| 279 |
|
| 280 |
extern __inline __m128 |
| 281 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 282 |
_mm_fmsubadd_ps (__m128 __A, __m128 __B, __m128 __C) |
| 283 |
{ |
| 284 |
return (__m128)__builtin_ia32_vfmaddsubps ((__v4sf)__A, (__v4sf)__B, |
| 285 |
-(__v4sf)__C); |
| 286 |
} |
| 287 |
|
| 288 |
extern __inline __m256 |
| 289 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 290 |
_mm256_fmsubadd_ps (__m256 __A, __m256 __B, __m256 __C) |
| 291 |
{ |
| 292 |
return (__m256)__builtin_ia32_vfmaddsubps256 ((__v8sf)__A, |
| 293 |
(__v8sf)__B, |
| 294 |
-(__v8sf)__C); |
| 295 |
} |
| 296 |
|
| 297 |
#ifdef __DISABLE_FMA__ |
| 298 |
#undef __DISABLE_FMA__ |
| 299 |
#pragma GCC pop_options |
| 300 |
#endif /* __DISABLE_FMA__ */ |
| 301 |
|
| 302 |
#endif |
