1 /*===---- fma4intrin.h - FMA4 intrinsics -----------------------------------===
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:
10 * The above copyright notice and this permission notice shall be included in
11 * all copies or substantial portions of the Software.
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
21 *===-----------------------------------------------------------------------===
25 #error "Never use <fma4intrin.h> directly; include <x86intrin.h> instead."
28 #ifndef __FMA4INTRIN_H
29 #define __FMA4INTRIN_H
32 # error "FMA4 instruction set is not enabled"
35 #include <pmmintrin.h>
37 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
38 _mm_macc_ps(__m128 __A, __m128 __B, __m128 __C)
40 return (__m128)__builtin_ia32_vfmaddps(__A, __B, __C);
43 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
44 _mm_macc_pd(__m128d __A, __m128d __B, __m128d __C)
46 return (__m128d)__builtin_ia32_vfmaddpd(__A, __B, __C);
49 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
50 _mm_macc_ss(__m128 __A, __m128 __B, __m128 __C)
52 return (__m128)__builtin_ia32_vfmaddss(__A, __B, __C);
55 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
56 _mm_macc_sd(__m128d __A, __m128d __B, __m128d __C)
58 return (__m128d)__builtin_ia32_vfmaddsd(__A, __B, __C);
61 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
62 _mm_msub_ps(__m128 __A, __m128 __B, __m128 __C)
64 return (__m128)__builtin_ia32_vfmsubps(__A, __B, __C);
67 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
68 _mm_msub_pd(__m128d __A, __m128d __B, __m128d __C)
70 return (__m128d)__builtin_ia32_vfmsubpd(__A, __B, __C);
73 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
74 _mm_msub_ss(__m128 __A, __m128 __B, __m128 __C)
76 return (__m128)__builtin_ia32_vfmsubss(__A, __B, __C);
79 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
80 _mm_msub_sd(__m128d __A, __m128d __B, __m128d __C)
82 return (__m128d)__builtin_ia32_vfmsubsd(__A, __B, __C);
85 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
86 _mm_nmacc_ps(__m128 __A, __m128 __B, __m128 __C)
88 return (__m128)__builtin_ia32_vfnmaddps(__A, __B, __C);
91 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
92 _mm_nmacc_pd(__m128d __A, __m128d __B, __m128d __C)
94 return (__m128d)__builtin_ia32_vfnmaddpd(__A, __B, __C);
97 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
98 _mm_nmacc_ss(__m128 __A, __m128 __B, __m128 __C)
100 return (__m128)__builtin_ia32_vfnmaddss(__A, __B, __C);
103 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
104 _mm_nmacc_sd(__m128d __A, __m128d __B, __m128d __C)
106 return (__m128d)__builtin_ia32_vfnmaddsd(__A, __B, __C);
109 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
110 _mm_nmsub_ps(__m128 __A, __m128 __B, __m128 __C)
112 return (__m128)__builtin_ia32_vfnmsubps(__A, __B, __C);
115 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
116 _mm_nmsub_pd(__m128d __A, __m128d __B, __m128d __C)
118 return (__m128d)__builtin_ia32_vfnmsubpd(__A, __B, __C);
121 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
122 _mm_nmsub_ss(__m128 __A, __m128 __B, __m128 __C)
124 return (__m128)__builtin_ia32_vfnmsubss(__A, __B, __C);
127 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
128 _mm_nmsub_sd(__m128d __A, __m128d __B, __m128d __C)
130 return (__m128d)__builtin_ia32_vfnmsubsd(__A, __B, __C);
133 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
134 _mm_maddsub_ps(__m128 __A, __m128 __B, __m128 __C)
136 return (__m128)__builtin_ia32_vfmaddsubps(__A, __B, __C);
139 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
140 _mm_maddsub_pd(__m128d __A, __m128d __B, __m128d __C)
142 return (__m128d)__builtin_ia32_vfmaddsubpd(__A, __B, __C);
145 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
146 _mm_msubadd_ps(__m128 __A, __m128 __B, __m128 __C)
148 return (__m128)__builtin_ia32_vfmsubaddps(__A, __B, __C);
151 static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
152 _mm_msubadd_pd(__m128d __A, __m128d __B, __m128d __C)
154 return (__m128d)__builtin_ia32_vfmsubaddpd(__A, __B, __C);
157 static __inline__ __m256 __attribute__((__always_inline__, __nodebug__))
158 _mm256_macc_ps(__m256 __A, __m256 __B, __m256 __C)
160 return (__m256)__builtin_ia32_vfmaddps256(__A, __B, __C);
163 static __inline__ __m256d __attribute__((__always_inline__, __nodebug__))
164 _mm256_macc_pd(__m256d __A, __m256d __B, __m256d __C)
166 return (__m256d)__builtin_ia32_vfmaddpd256(__A, __B, __C);
169 static __inline__ __m256 __attribute__((__always_inline__, __nodebug__))
170 _mm256_msub_ps(__m256 __A, __m256 __B, __m256 __C)
172 return (__m256)__builtin_ia32_vfmsubps256(__A, __B, __C);
175 static __inline__ __m256d __attribute__((__always_inline__, __nodebug__))
176 _mm256_msub_pd(__m256d __A, __m256d __B, __m256d __C)
178 return (__m256d)__builtin_ia32_vfmsubpd256(__A, __B, __C);
181 static __inline__ __m256 __attribute__((__always_inline__, __nodebug__))
182 _mm256_nmacc_ps(__m256 __A, __m256 __B, __m256 __C)
184 return (__m256)__builtin_ia32_vfnmaddps256(__A, __B, __C);
187 static __inline__ __m256d __attribute__((__always_inline__, __nodebug__))
188 _mm256_nmacc_pd(__m256d __A, __m256d __B, __m256d __C)
190 return (__m256d)__builtin_ia32_vfnmaddpd256(__A, __B, __C);
193 static __inline__ __m256 __attribute__((__always_inline__, __nodebug__))
194 _mm256_nmsub_ps(__m256 __A, __m256 __B, __m256 __C)
196 return (__m256)__builtin_ia32_vfnmsubps256(__A, __B, __C);
199 static __inline__ __m256d __attribute__((__always_inline__, __nodebug__))
200 _mm256_nmsub_pd(__m256d __A, __m256d __B, __m256d __C)
202 return (__m256d)__builtin_ia32_vfnmsubpd256(__A, __B, __C);
205 static __inline__ __m256 __attribute__((__always_inline__, __nodebug__))
206 _mm256_maddsub_ps(__m256 __A, __m256 __B, __m256 __C)
208 return (__m256)__builtin_ia32_vfmaddsubps256(__A, __B, __C);
211 static __inline__ __m256d __attribute__((__always_inline__, __nodebug__))
212 _mm256_maddsub_pd(__m256d __A, __m256d __B, __m256d __C)
214 return (__m256d)__builtin_ia32_vfmaddsubpd256(__A, __B, __C);
217 static __inline__ __m256 __attribute__((__always_inline__, __nodebug__))
218 _mm256_msubadd_ps(__m256 __A, __m256 __B, __m256 __C)
220 return (__m256)__builtin_ia32_vfmsubaddps256(__A, __B, __C);
223 static __inline__ __m256d __attribute__((__always_inline__, __nodebug__))
224 _mm256_msubadd_pd(__m256d __A, __m256d __B, __m256d __C)
226 return (__m256d)__builtin_ia32_vfmsubaddpd256(__A, __B, __C);
229 #endif /* __FMA4__ */
231 #endif /* __FMA4INTRIN_H */