1 /*===---- fmaintrin.h - FMA 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 <fmaintrin.h> directly; include <immintrin.h> instead."
31 /* Define the default attributes for the functions in this file. */
32 #define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("fma"), __min_vector_width__(128)))
33 #define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("fma"), __min_vector_width__(256)))
35 static __inline__ __m128 __DEFAULT_FN_ATTRS128
36 _mm_fmadd_ps(__m128 __A, __m128 __B, __m128 __C)
38 return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
41 static __inline__ __m128d __DEFAULT_FN_ATTRS128
42 _mm_fmadd_pd(__m128d __A, __m128d __B, __m128d __C)
44 return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
47 static __inline__ __m128 __DEFAULT_FN_ATTRS128
48 _mm_fmadd_ss(__m128 __A, __m128 __B, __m128 __C)
50 return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
53 static __inline__ __m128d __DEFAULT_FN_ATTRS128
54 _mm_fmadd_sd(__m128d __A, __m128d __B, __m128d __C)
56 return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, (__v2df)__B, (__v2df)__C);
59 static __inline__ __m128 __DEFAULT_FN_ATTRS128
60 _mm_fmsub_ps(__m128 __A, __m128 __B, __m128 __C)
62 return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
65 static __inline__ __m128d __DEFAULT_FN_ATTRS128
66 _mm_fmsub_pd(__m128d __A, __m128d __B, __m128d __C)
68 return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
71 static __inline__ __m128 __DEFAULT_FN_ATTRS128
72 _mm_fmsub_ss(__m128 __A, __m128 __B, __m128 __C)
74 return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
77 static __inline__ __m128d __DEFAULT_FN_ATTRS128
78 _mm_fmsub_sd(__m128d __A, __m128d __B, __m128d __C)
80 return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, (__v2df)__B, -(__v2df)__C);
83 static __inline__ __m128 __DEFAULT_FN_ATTRS128
84 _mm_fnmadd_ps(__m128 __A, __m128 __B, __m128 __C)
86 return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
89 static __inline__ __m128d __DEFAULT_FN_ATTRS128
90 _mm_fnmadd_pd(__m128d __A, __m128d __B, __m128d __C)
92 return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, (__v2df)__C);
95 static __inline__ __m128 __DEFAULT_FN_ATTRS128
96 _mm_fnmadd_ss(__m128 __A, __m128 __B, __m128 __C)
98 return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, -(__v4sf)__B, (__v4sf)__C);
101 static __inline__ __m128d __DEFAULT_FN_ATTRS128
102 _mm_fnmadd_sd(__m128d __A, __m128d __B, __m128d __C)
104 return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, -(__v2df)__B, (__v2df)__C);
107 static __inline__ __m128 __DEFAULT_FN_ATTRS128
108 _mm_fnmsub_ps(__m128 __A, __m128 __B, __m128 __C)
110 return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
113 static __inline__ __m128d __DEFAULT_FN_ATTRS128
114 _mm_fnmsub_pd(__m128d __A, __m128d __B, __m128d __C)
116 return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C);
119 static __inline__ __m128 __DEFAULT_FN_ATTRS128
120 _mm_fnmsub_ss(__m128 __A, __m128 __B, __m128 __C)
122 return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, -(__v4sf)__B, -(__v4sf)__C);
125 static __inline__ __m128d __DEFAULT_FN_ATTRS128
126 _mm_fnmsub_sd(__m128d __A, __m128d __B, __m128d __C)
128 return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, -(__v2df)__B, -(__v2df)__C);
131 static __inline__ __m128 __DEFAULT_FN_ATTRS128
132 _mm_fmaddsub_ps(__m128 __A, __m128 __B, __m128 __C)
134 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
137 static __inline__ __m128d __DEFAULT_FN_ATTRS128
138 _mm_fmaddsub_pd(__m128d __A, __m128d __B, __m128d __C)
140 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
143 static __inline__ __m128 __DEFAULT_FN_ATTRS128
144 _mm_fmsubadd_ps(__m128 __A, __m128 __B, __m128 __C)
146 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
149 static __inline__ __m128d __DEFAULT_FN_ATTRS128
150 _mm_fmsubadd_pd(__m128d __A, __m128d __B, __m128d __C)
152 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
155 static __inline__ __m256 __DEFAULT_FN_ATTRS256
156 _mm256_fmadd_ps(__m256 __A, __m256 __B, __m256 __C)
158 return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
161 static __inline__ __m256d __DEFAULT_FN_ATTRS256
162 _mm256_fmadd_pd(__m256d __A, __m256d __B, __m256d __C)
164 return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
167 static __inline__ __m256 __DEFAULT_FN_ATTRS256
168 _mm256_fmsub_ps(__m256 __A, __m256 __B, __m256 __C)
170 return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
173 static __inline__ __m256d __DEFAULT_FN_ATTRS256
174 _mm256_fmsub_pd(__m256d __A, __m256d __B, __m256d __C)
176 return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
179 static __inline__ __m256 __DEFAULT_FN_ATTRS256
180 _mm256_fnmadd_ps(__m256 __A, __m256 __B, __m256 __C)
182 return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
185 static __inline__ __m256d __DEFAULT_FN_ATTRS256
186 _mm256_fnmadd_pd(__m256d __A, __m256d __B, __m256d __C)
188 return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, (__v4df)__C);
191 static __inline__ __m256 __DEFAULT_FN_ATTRS256
192 _mm256_fnmsub_ps(__m256 __A, __m256 __B, __m256 __C)
194 return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
197 static __inline__ __m256d __DEFAULT_FN_ATTRS256
198 _mm256_fnmsub_pd(__m256d __A, __m256d __B, __m256d __C)
200 return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, -(__v4df)__C);
203 static __inline__ __m256 __DEFAULT_FN_ATTRS256
204 _mm256_fmaddsub_ps(__m256 __A, __m256 __B, __m256 __C)
206 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
209 static __inline__ __m256d __DEFAULT_FN_ATTRS256
210 _mm256_fmaddsub_pd(__m256d __A, __m256d __B, __m256d __C)
212 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
215 static __inline__ __m256 __DEFAULT_FN_ATTRS256
216 _mm256_fmsubadd_ps(__m256 __A, __m256 __B, __m256 __C)
218 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
221 static __inline__ __m256d __DEFAULT_FN_ATTRS256
222 _mm256_fmsubadd_pd(__m256d __A, __m256d __B, __m256d __C)
224 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
227 #undef __DEFAULT_FN_ATTRS128
228 #undef __DEFAULT_FN_ATTRS256
230 #endif /* __FMAINTRIN_H */