Home | History | Annotate | Line # | Download | only in i386
fmaintrin.h revision 1.8 
      1  1.8  mrg /* Copyright (C) 2011-2019 Free Software Foundation, Inc.
      2  1.1  mrg 
      3  1.1  mrg    This file is part of GCC.
      4  1.1  mrg 
      5  1.1  mrg    GCC is free software; you can redistribute it and/or modify
      6  1.1  mrg    it under the terms of the GNU General Public License as published by
      7  1.1  mrg    the Free Software Foundation; either version 3, or (at your option)
      8  1.1  mrg    any later version.
      9  1.1  mrg 
     10  1.1  mrg    GCC is distributed in the hope that it will be useful,
     11  1.1  mrg    but WITHOUT ANY WARRANTY; without even the implied warranty of
     12  1.1  mrg    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
     13  1.1  mrg    GNU General Public License for more details.
     14  1.1  mrg 
     15  1.1  mrg    Under Section 7 of GPL version 3, you are granted additional
     16  1.1  mrg    permissions described in the GCC Runtime Library Exception, version
     17  1.1  mrg    3.1, as published by the Free Software Foundation.
     18  1.1  mrg 
     19  1.1  mrg    You should have received a copy of the GNU General Public License and
     20  1.1  mrg    a copy of the GCC Runtime Library Exception along with this program;
     21  1.1  mrg    see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
     22  1.1  mrg    <http://www.gnu.org/licenses/>.  */
     23  1.1  mrg 
     24  1.1  mrg #ifndef _IMMINTRIN_H_INCLUDED
     25  1.1  mrg # error "Never use <fmaintrin.h> directly; include <immintrin.h> instead."
     26  1.1  mrg #endif
     27  1.1  mrg 
     28  1.1  mrg #ifndef _FMAINTRIN_H_INCLUDED
     29  1.1  mrg #define _FMAINTRIN_H_INCLUDED
     30  1.1  mrg 
     31  1.1  mrg #ifndef __FMA__
     32  1.3  mrg #pragma GCC push_options
     33  1.3  mrg #pragma GCC target("fma")
     34  1.3  mrg #define __DISABLE_FMA__
     35  1.3  mrg #endif /* __FMA__ */
     36  1.1  mrg 
     37  1.1  mrg extern __inline __m128d
     38  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
     39  1.1  mrg _mm_fmadd_pd (__m128d __A, __m128d __B, __m128d __C)
     40  1.1  mrg {
     41  1.1  mrg   return (__m128d)__builtin_ia32_vfmaddpd ((__v2df)__A, (__v2df)__B,
     42  1.1  mrg                                            (__v2df)__C);
     43  1.1  mrg }
     44  1.1  mrg 
     45  1.1  mrg extern __inline __m256d
     46  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
     47  1.1  mrg _mm256_fmadd_pd (__m256d __A, __m256d __B, __m256d __C)
     48  1.1  mrg {
     49  1.1  mrg   return (__m256d)__builtin_ia32_vfmaddpd256 ((__v4df)__A, (__v4df)__B,
     50  1.1  mrg                                               (__v4df)__C);
     51  1.1  mrg }
     52  1.1  mrg 
     53  1.1  mrg extern __inline __m128
     54  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
     55  1.1  mrg _mm_fmadd_ps (__m128 __A, __m128 __B, __m128 __C)
     56  1.1  mrg {
     57  1.1  mrg   return (__m128)__builtin_ia32_vfmaddps ((__v4sf)__A, (__v4sf)__B,
     58  1.1  mrg                                           (__v4sf)__C);
     59  1.1  mrg }
     60  1.1  mrg 
     61  1.1  mrg extern __inline __m256
     62  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
     63  1.1  mrg _mm256_fmadd_ps (__m256 __A, __m256 __B, __m256 __C)
     64  1.1  mrg {
     65  1.1  mrg   return (__m256)__builtin_ia32_vfmaddps256 ((__v8sf)__A, (__v8sf)__B,
     66  1.1  mrg                                              (__v8sf)__C);
     67  1.1  mrg }
     68  1.1  mrg 
     69  1.1  mrg extern __inline __m128d
     70  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
     71  1.1  mrg _mm_fmadd_sd (__m128d __A, __m128d __B, __m128d __C)
     72  1.1  mrg {
     73  1.1  mrg   return (__m128d) __builtin_ia32_vfmaddsd3 ((__v2df)__A, (__v2df)__B,
     74  1.1  mrg                                              (__v2df)__C);
     75  1.1  mrg }
     76  1.1  mrg 
     77  1.1  mrg extern __inline __m128
     78  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
     79  1.1  mrg _mm_fmadd_ss (__m128 __A, __m128 __B, __m128 __C)
     80  1.1  mrg {
     81  1.1  mrg   return (__m128) __builtin_ia32_vfmaddss3 ((__v4sf)__A, (__v4sf)__B,
     82  1.1  mrg                                             (__v4sf)__C);
     83  1.1  mrg }
     84  1.1  mrg 
     85  1.1  mrg extern __inline __m128d
     86  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
     87  1.1  mrg _mm_fmsub_pd (__m128d __A, __m128d __B, __m128d __C)
     88  1.1  mrg {
     89  1.8  mrg   return (__m128d)__builtin_ia32_vfmsubpd ((__v2df)__A, (__v2df)__B,
     90  1.8  mrg                                            (__v2df)__C);
     91  1.1  mrg }
     92  1.1  mrg 
     93  1.1  mrg extern __inline __m256d
     94  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
     95  1.1  mrg _mm256_fmsub_pd (__m256d __A, __m256d __B, __m256d __C)
     96  1.1  mrg {
     97  1.8  mrg   return (__m256d)__builtin_ia32_vfmsubpd256 ((__v4df)__A, (__v4df)__B,
     98  1.8  mrg                                               (__v4df)__C);
     99  1.1  mrg }
    100  1.1  mrg 
    101  1.1  mrg extern __inline __m128
    102  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    103  1.1  mrg _mm_fmsub_ps (__m128 __A, __m128 __B, __m128 __C)
    104  1.1  mrg {
    105  1.8  mrg   return (__m128)__builtin_ia32_vfmsubps ((__v4sf)__A, (__v4sf)__B,
    106  1.8  mrg                                           (__v4sf)__C);
    107  1.1  mrg }
    108  1.1  mrg 
    109  1.1  mrg extern __inline __m256
    110  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    111  1.1  mrg _mm256_fmsub_ps (__m256 __A, __m256 __B, __m256 __C)
    112  1.1  mrg {
    113  1.8  mrg   return (__m256)__builtin_ia32_vfmsubps256 ((__v8sf)__A, (__v8sf)__B,
    114  1.8  mrg                                              (__v8sf)__C);
    115  1.1  mrg }
    116  1.1  mrg 
    117  1.1  mrg extern __inline __m128d
    118  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    119  1.1  mrg _mm_fmsub_sd (__m128d __A, __m128d __B, __m128d __C)
    120  1.1  mrg {
    121  1.8  mrg   return (__m128d)__builtin_ia32_vfmsubsd3 ((__v2df)__A, (__v2df)__B,
    122  1.8  mrg                                             (__v2df)__C);
    123  1.1  mrg }
    124  1.1  mrg 
    125  1.1  mrg extern __inline __m128
    126  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    127  1.1  mrg _mm_fmsub_ss (__m128 __A, __m128 __B, __m128 __C)
    128  1.1  mrg {
    129  1.8  mrg   return (__m128)__builtin_ia32_vfmsubss3 ((__v4sf)__A, (__v4sf)__B,
    130  1.8  mrg                                            (__v4sf)__C);
    131  1.1  mrg }
    132  1.1  mrg 
    133  1.1  mrg extern __inline __m128d
    134  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    135  1.1  mrg _mm_fnmadd_pd (__m128d __A, __m128d __B, __m128d __C)
    136  1.1  mrg {
    137  1.8  mrg   return (__m128d)__builtin_ia32_vfnmaddpd ((__v2df)__A, (__v2df)__B,
    138  1.8  mrg 					    (__v2df)__C);
    139  1.1  mrg }
    140  1.1  mrg 
    141  1.1  mrg extern __inline __m256d
    142  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    143  1.1  mrg _mm256_fnmadd_pd (__m256d __A, __m256d __B, __m256d __C)
    144  1.1  mrg {
    145  1.8  mrg   return (__m256d)__builtin_ia32_vfnmaddpd256 ((__v4df)__A, (__v4df)__B,
    146  1.8  mrg 					       (__v4df)__C);
    147  1.1  mrg }
    148  1.1  mrg 
    149  1.1  mrg extern __inline __m128
    150  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    151  1.1  mrg _mm_fnmadd_ps (__m128 __A, __m128 __B, __m128 __C)
    152  1.1  mrg {
    153  1.8  mrg   return (__m128)__builtin_ia32_vfnmaddps ((__v4sf)__A, (__v4sf)__B,
    154  1.8  mrg 					   (__v4sf)__C);
    155  1.1  mrg }
    156  1.1  mrg 
    157  1.1  mrg extern __inline __m256
    158  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    159  1.1  mrg _mm256_fnmadd_ps (__m256 __A, __m256 __B, __m256 __C)
    160  1.1  mrg {
    161  1.8  mrg   return (__m256)__builtin_ia32_vfnmaddps256 ((__v8sf)__A, (__v8sf)__B,
    162  1.8  mrg 					      (__v8sf)__C);
    163  1.1  mrg }
    164  1.1  mrg 
    165  1.1  mrg extern __inline __m128d
    166  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    167  1.1  mrg _mm_fnmadd_sd (__m128d __A, __m128d __B, __m128d __C)
    168  1.1  mrg {
    169  1.8  mrg   return (__m128d)__builtin_ia32_vfnmaddsd3 ((__v2df)__A, (__v2df)__B,
    170  1.8  mrg 					     (__v2df)__C);
    171  1.1  mrg }
    172  1.1  mrg 
    173  1.1  mrg extern __inline __m128
    174  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    175  1.1  mrg _mm_fnmadd_ss (__m128 __A, __m128 __B, __m128 __C)
    176  1.1  mrg {
    177  1.8  mrg   return (__m128)__builtin_ia32_vfnmaddss3 ((__v4sf)__A, (__v4sf)__B,
    178  1.8  mrg 					    (__v4sf)__C);
    179  1.1  mrg }
    180  1.1  mrg 
    181  1.1  mrg extern __inline __m128d
    182  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    183  1.1  mrg _mm_fnmsub_pd (__m128d __A, __m128d __B, __m128d __C)
    184  1.1  mrg {
    185  1.8  mrg   return (__m128d)__builtin_ia32_vfnmsubpd ((__v2df)__A, (__v2df)__B,
    186  1.8  mrg 					    (__v2df)__C);
    187  1.1  mrg }
    188  1.1  mrg 
    189  1.1  mrg extern __inline __m256d
    190  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    191  1.1  mrg _mm256_fnmsub_pd (__m256d __A, __m256d __B, __m256d __C)
    192  1.1  mrg {
    193  1.8  mrg   return (__m256d)__builtin_ia32_vfnmsubpd256 ((__v4df)__A, (__v4df)__B,
    194  1.8  mrg 					       (__v4df)__C);
    195  1.1  mrg }
    196  1.1  mrg 
    197  1.1  mrg extern __inline __m128
    198  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    199  1.1  mrg _mm_fnmsub_ps (__m128 __A, __m128 __B, __m128 __C)
    200  1.1  mrg {
    201  1.8  mrg   return (__m128)__builtin_ia32_vfnmsubps ((__v4sf)__A, (__v4sf)__B,
    202  1.8  mrg 					   (__v4sf)__C);
    203  1.1  mrg }
    204  1.1  mrg 
    205  1.1  mrg extern __inline __m256
    206  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    207  1.1  mrg _mm256_fnmsub_ps (__m256 __A, __m256 __B, __m256 __C)
    208  1.1  mrg {
    209  1.8  mrg   return (__m256)__builtin_ia32_vfnmsubps256 ((__v8sf)__A, (__v8sf)__B,
    210  1.8  mrg 					      (__v8sf)__C);
    211  1.1  mrg }
    212  1.1  mrg 
    213  1.1  mrg extern __inline __m128d
    214  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    215  1.1  mrg _mm_fnmsub_sd (__m128d __A, __m128d __B, __m128d __C)
    216  1.1  mrg {
    217  1.8  mrg   return (__m128d)__builtin_ia32_vfnmsubsd3 ((__v2df)__A, (__v2df)__B,
    218  1.8  mrg 					     (__v2df)__C);
    219  1.1  mrg }
    220  1.1  mrg 
    221  1.1  mrg extern __inline __m128
    222  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    223  1.1  mrg _mm_fnmsub_ss (__m128 __A, __m128 __B, __m128 __C)
    224  1.1  mrg {
    225  1.8  mrg   return (__m128)__builtin_ia32_vfnmsubss3 ((__v4sf)__A, (__v4sf)__B,
    226  1.8  mrg 					    (__v4sf)__C);
    227  1.1  mrg }
    228  1.1  mrg 
    229  1.1  mrg extern __inline __m128d
    230  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    231  1.1  mrg _mm_fmaddsub_pd (__m128d __A, __m128d __B, __m128d __C)
    232  1.1  mrg {
    233  1.1  mrg   return (__m128d)__builtin_ia32_vfmaddsubpd ((__v2df)__A, (__v2df)__B,
    234  1.1  mrg                                               (__v2df)__C);
    235  1.1  mrg }
    236  1.1  mrg 
    237  1.1  mrg extern __inline __m256d
    238  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    239  1.1  mrg _mm256_fmaddsub_pd (__m256d __A, __m256d __B, __m256d __C)
    240  1.1  mrg {
    241  1.1  mrg   return (__m256d)__builtin_ia32_vfmaddsubpd256 ((__v4df)__A,
    242  1.1  mrg                                                  (__v4df)__B,
    243  1.1  mrg                                                  (__v4df)__C);
    244  1.1  mrg }
    245  1.1  mrg 
    246  1.1  mrg extern __inline __m128
    247  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    248  1.1  mrg _mm_fmaddsub_ps (__m128 __A, __m128 __B, __m128 __C)
    249  1.1  mrg {
    250  1.1  mrg   return (__m128)__builtin_ia32_vfmaddsubps ((__v4sf)__A, (__v4sf)__B,
    251  1.1  mrg                                              (__v4sf)__C);
    252  1.1  mrg }
    253  1.1  mrg 
    254  1.1  mrg extern __inline __m256
    255  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    256  1.1  mrg _mm256_fmaddsub_ps (__m256 __A, __m256 __B, __m256 __C)
    257  1.1  mrg {
    258  1.1  mrg   return (__m256)__builtin_ia32_vfmaddsubps256 ((__v8sf)__A,
    259  1.1  mrg                                                 (__v8sf)__B,
    260  1.1  mrg                                                 (__v8sf)__C);
    261  1.1  mrg }
    262  1.1  mrg 
    263  1.1  mrg extern __inline __m128d
    264  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    265  1.1  mrg _mm_fmsubadd_pd (__m128d __A, __m128d __B, __m128d __C)
    266  1.1  mrg {
    267  1.1  mrg   return (__m128d)__builtin_ia32_vfmaddsubpd ((__v2df)__A, (__v2df)__B,
    268  1.1  mrg                                               -(__v2df)__C);
    269  1.1  mrg }
    270  1.1  mrg 
    271  1.1  mrg extern __inline __m256d
    272  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    273  1.1  mrg _mm256_fmsubadd_pd (__m256d __A, __m256d __B, __m256d __C)
    274  1.1  mrg {
    275  1.1  mrg   return (__m256d)__builtin_ia32_vfmaddsubpd256 ((__v4df)__A,
    276  1.1  mrg                                                  (__v4df)__B,
    277  1.1  mrg                                                  -(__v4df)__C);
    278  1.1  mrg }
    279  1.1  mrg 
    280  1.1  mrg extern __inline __m128
    281  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    282  1.1  mrg _mm_fmsubadd_ps (__m128 __A, __m128 __B, __m128 __C)
    283  1.1  mrg {
    284  1.1  mrg   return (__m128)__builtin_ia32_vfmaddsubps ((__v4sf)__A, (__v4sf)__B,
    285  1.1  mrg                                              -(__v4sf)__C);
    286  1.1  mrg }
    287  1.1  mrg 
    288  1.1  mrg extern __inline __m256
    289  1.1  mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
    290  1.1  mrg _mm256_fmsubadd_ps (__m256 __A, __m256 __B, __m256 __C)
    291  1.1  mrg {
    292  1.1  mrg   return (__m256)__builtin_ia32_vfmaddsubps256 ((__v8sf)__A,
    293  1.1  mrg                                                 (__v8sf)__B,
    294  1.1  mrg                                                 -(__v8sf)__C);
    295  1.1  mrg }
    296  1.1  mrg 
    297  1.3  mrg #ifdef __DISABLE_FMA__
    298  1.3  mrg #undef __DISABLE_FMA__
    299  1.3  mrg #pragma GCC pop_options
    300  1.3  mrg #endif /* __DISABLE_FMA__ */
    301  1.1  mrg 
    302  1.1  mrg #endif
    303