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 /* Define the default attributes for the functions in this file. */
38 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__))
40 static __inline__ __m128 __DEFAULT_FN_ATTRS
41 _mm_macc_ps(__m128 __A, __m128 __B, __m128 __C)
43 return (__m128)__builtin_ia32_vfmaddps(__A, __B, __C);
46 static __inline__ __m128d __DEFAULT_FN_ATTRS
47 _mm_macc_pd(__m128d __A, __m128d __B, __m128d __C)
49 return (__m128d)__builtin_ia32_vfmaddpd(__A, __B, __C);
52 static __inline__ __m128 __DEFAULT_FN_ATTRS
53 _mm_macc_ss(__m128 __A, __m128 __B, __m128 __C)
55 return (__m128)__builtin_ia32_vfmaddss(__A, __B, __C);
58 static __inline__ __m128d __DEFAULT_FN_ATTRS
59 _mm_macc_sd(__m128d __A, __m128d __B, __m128d __C)
61 return (__m128d)__builtin_ia32_vfmaddsd(__A, __B, __C);
64 static __inline__ __m128 __DEFAULT_FN_ATTRS
65 _mm_msub_ps(__m128 __A, __m128 __B, __m128 __C)
67 return (__m128)__builtin_ia32_vfmsubps(__A, __B, __C);
70 static __inline__ __m128d __DEFAULT_FN_ATTRS
71 _mm_msub_pd(__m128d __A, __m128d __B, __m128d __C)
73 return (__m128d)__builtin_ia32_vfmsubpd(__A, __B, __C);
76 static __inline__ __m128 __DEFAULT_FN_ATTRS
77 _mm_msub_ss(__m128 __A, __m128 __B, __m128 __C)
79 return (__m128)__builtin_ia32_vfmsubss(__A, __B, __C);
82 static __inline__ __m128d __DEFAULT_FN_ATTRS
83 _mm_msub_sd(__m128d __A, __m128d __B, __m128d __C)
85 return (__m128d)__builtin_ia32_vfmsubsd(__A, __B, __C);
88 static __inline__ __m128 __DEFAULT_FN_ATTRS
89 _mm_nmacc_ps(__m128 __A, __m128 __B, __m128 __C)
91 return (__m128)__builtin_ia32_vfnmaddps(__A, __B, __C);
94 static __inline__ __m128d __DEFAULT_FN_ATTRS
95 _mm_nmacc_pd(__m128d __A, __m128d __B, __m128d __C)
97 return (__m128d)__builtin_ia32_vfnmaddpd(__A, __B, __C);
100 static __inline__ __m128 __DEFAULT_FN_ATTRS
101 _mm_nmacc_ss(__m128 __A, __m128 __B, __m128 __C)
103 return (__m128)__builtin_ia32_vfnmaddss(__A, __B, __C);
106 static __inline__ __m128d __DEFAULT_FN_ATTRS
107 _mm_nmacc_sd(__m128d __A, __m128d __B, __m128d __C)
109 return (__m128d)__builtin_ia32_vfnmaddsd(__A, __B, __C);
112 static __inline__ __m128 __DEFAULT_FN_ATTRS
113 _mm_nmsub_ps(__m128 __A, __m128 __B, __m128 __C)
115 return (__m128)__builtin_ia32_vfnmsubps(__A, __B, __C);
118 static __inline__ __m128d __DEFAULT_FN_ATTRS
119 _mm_nmsub_pd(__m128d __A, __m128d __B, __m128d __C)
121 return (__m128d)__builtin_ia32_vfnmsubpd(__A, __B, __C);
124 static __inline__ __m128 __DEFAULT_FN_ATTRS
125 _mm_nmsub_ss(__m128 __A, __m128 __B, __m128 __C)
127 return (__m128)__builtin_ia32_vfnmsubss(__A, __B, __C);
130 static __inline__ __m128d __DEFAULT_FN_ATTRS
131 _mm_nmsub_sd(__m128d __A, __m128d __B, __m128d __C)
133 return (__m128d)__builtin_ia32_vfnmsubsd(__A, __B, __C);
136 static __inline__ __m128 __DEFAULT_FN_ATTRS
137 _mm_maddsub_ps(__m128 __A, __m128 __B, __m128 __C)
139 return (__m128)__builtin_ia32_vfmaddsubps(__A, __B, __C);
142 static __inline__ __m128d __DEFAULT_FN_ATTRS
143 _mm_maddsub_pd(__m128d __A, __m128d __B, __m128d __C)
145 return (__m128d)__builtin_ia32_vfmaddsubpd(__A, __B, __C);
148 static __inline__ __m128 __DEFAULT_FN_ATTRS
149 _mm_msubadd_ps(__m128 __A, __m128 __B, __m128 __C)
151 return (__m128)__builtin_ia32_vfmsubaddps(__A, __B, __C);
154 static __inline__ __m128d __DEFAULT_FN_ATTRS
155 _mm_msubadd_pd(__m128d __A, __m128d __B, __m128d __C)
157 return (__m128d)__builtin_ia32_vfmsubaddpd(__A, __B, __C);
160 static __inline__ __m256 __DEFAULT_FN_ATTRS
161 _mm256_macc_ps(__m256 __A, __m256 __B, __m256 __C)
163 return (__m256)__builtin_ia32_vfmaddps256(__A, __B, __C);
166 static __inline__ __m256d __DEFAULT_FN_ATTRS
167 _mm256_macc_pd(__m256d __A, __m256d __B, __m256d __C)
169 return (__m256d)__builtin_ia32_vfmaddpd256(__A, __B, __C);
172 static __inline__ __m256 __DEFAULT_FN_ATTRS
173 _mm256_msub_ps(__m256 __A, __m256 __B, __m256 __C)
175 return (__m256)__builtin_ia32_vfmsubps256(__A, __B, __C);
178 static __inline__ __m256d __DEFAULT_FN_ATTRS
179 _mm256_msub_pd(__m256d __A, __m256d __B, __m256d __C)
181 return (__m256d)__builtin_ia32_vfmsubpd256(__A, __B, __C);
184 static __inline__ __m256 __DEFAULT_FN_ATTRS
185 _mm256_nmacc_ps(__m256 __A, __m256 __B, __m256 __C)
187 return (__m256)__builtin_ia32_vfnmaddps256(__A, __B, __C);
190 static __inline__ __m256d __DEFAULT_FN_ATTRS
191 _mm256_nmacc_pd(__m256d __A, __m256d __B, __m256d __C)
193 return (__m256d)__builtin_ia32_vfnmaddpd256(__A, __B, __C);
196 static __inline__ __m256 __DEFAULT_FN_ATTRS
197 _mm256_nmsub_ps(__m256 __A, __m256 __B, __m256 __C)
199 return (__m256)__builtin_ia32_vfnmsubps256(__A, __B, __C);
202 static __inline__ __m256d __DEFAULT_FN_ATTRS
203 _mm256_nmsub_pd(__m256d __A, __m256d __B, __m256d __C)
205 return (__m256d)__builtin_ia32_vfnmsubpd256(__A, __B, __C);
208 static __inline__ __m256 __DEFAULT_FN_ATTRS
209 _mm256_maddsub_ps(__m256 __A, __m256 __B, __m256 __C)
211 return (__m256)__builtin_ia32_vfmaddsubps256(__A, __B, __C);
214 static __inline__ __m256d __DEFAULT_FN_ATTRS
215 _mm256_maddsub_pd(__m256d __A, __m256d __B, __m256d __C)
217 return (__m256d)__builtin_ia32_vfmaddsubpd256(__A, __B, __C);
220 static __inline__ __m256 __DEFAULT_FN_ATTRS
221 _mm256_msubadd_ps(__m256 __A, __m256 __B, __m256 __C)
223 return (__m256)__builtin_ia32_vfmsubaddps256(__A, __B, __C);
226 static __inline__ __m256d __DEFAULT_FN_ATTRS
227 _mm256_msubadd_pd(__m256d __A, __m256d __B, __m256d __C)
229 return (__m256d)__builtin_ia32_vfmsubaddpd256(__A, __B, __C);
232 #undef __DEFAULT_FN_ATTRS
234 #endif /* __FMA4__ */
236 #endif /* __FMA4INTRIN_H */