1 /*===---- fmaintrin.h - FMA intrinsics -------------------------------------===
3 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 * See https://llvm.org/LICENSE.txt for license information.
5 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 *===-----------------------------------------------------------------------===
11 #error "Never use <fmaintrin.h> directly; include <immintrin.h> instead."
17 /* Define the default attributes for the functions in this file. */
18 #define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("fma"), __min_vector_width__(128)))
19 #define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("fma"), __min_vector_width__(256)))
21 static __inline__ __m128 __DEFAULT_FN_ATTRS128
22 _mm_fmadd_ps(__m128 __A, __m128 __B, __m128 __C)
24 return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
27 static __inline__ __m128d __DEFAULT_FN_ATTRS128
28 _mm_fmadd_pd(__m128d __A, __m128d __B, __m128d __C)
30 return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
33 static __inline__ __m128 __DEFAULT_FN_ATTRS128
34 _mm_fmadd_ss(__m128 __A, __m128 __B, __m128 __C)
36 return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
39 static __inline__ __m128d __DEFAULT_FN_ATTRS128
40 _mm_fmadd_sd(__m128d __A, __m128d __B, __m128d __C)
42 return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, (__v2df)__B, (__v2df)__C);
45 static __inline__ __m128 __DEFAULT_FN_ATTRS128
46 _mm_fmsub_ps(__m128 __A, __m128 __B, __m128 __C)
48 return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
51 static __inline__ __m128d __DEFAULT_FN_ATTRS128
52 _mm_fmsub_pd(__m128d __A, __m128d __B, __m128d __C)
54 return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
57 static __inline__ __m128 __DEFAULT_FN_ATTRS128
58 _mm_fmsub_ss(__m128 __A, __m128 __B, __m128 __C)
60 return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
63 static __inline__ __m128d __DEFAULT_FN_ATTRS128
64 _mm_fmsub_sd(__m128d __A, __m128d __B, __m128d __C)
66 return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, (__v2df)__B, -(__v2df)__C);
69 static __inline__ __m128 __DEFAULT_FN_ATTRS128
70 _mm_fnmadd_ps(__m128 __A, __m128 __B, __m128 __C)
72 return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
75 static __inline__ __m128d __DEFAULT_FN_ATTRS128
76 _mm_fnmadd_pd(__m128d __A, __m128d __B, __m128d __C)
78 return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, (__v2df)__C);
81 static __inline__ __m128 __DEFAULT_FN_ATTRS128
82 _mm_fnmadd_ss(__m128 __A, __m128 __B, __m128 __C)
84 return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, -(__v4sf)__B, (__v4sf)__C);
87 static __inline__ __m128d __DEFAULT_FN_ATTRS128
88 _mm_fnmadd_sd(__m128d __A, __m128d __B, __m128d __C)
90 return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, -(__v2df)__B, (__v2df)__C);
93 static __inline__ __m128 __DEFAULT_FN_ATTRS128
94 _mm_fnmsub_ps(__m128 __A, __m128 __B, __m128 __C)
96 return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
99 static __inline__ __m128d __DEFAULT_FN_ATTRS128
100 _mm_fnmsub_pd(__m128d __A, __m128d __B, __m128d __C)
102 return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C);
105 static __inline__ __m128 __DEFAULT_FN_ATTRS128
106 _mm_fnmsub_ss(__m128 __A, __m128 __B, __m128 __C)
108 return (__m128)__builtin_ia32_vfmaddss3((__v4sf)__A, -(__v4sf)__B, -(__v4sf)__C);
111 static __inline__ __m128d __DEFAULT_FN_ATTRS128
112 _mm_fnmsub_sd(__m128d __A, __m128d __B, __m128d __C)
114 return (__m128d)__builtin_ia32_vfmaddsd3((__v2df)__A, -(__v2df)__B, -(__v2df)__C);
117 static __inline__ __m128 __DEFAULT_FN_ATTRS128
118 _mm_fmaddsub_ps(__m128 __A, __m128 __B, __m128 __C)
120 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
123 static __inline__ __m128d __DEFAULT_FN_ATTRS128
124 _mm_fmaddsub_pd(__m128d __A, __m128d __B, __m128d __C)
126 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
129 static __inline__ __m128 __DEFAULT_FN_ATTRS128
130 _mm_fmsubadd_ps(__m128 __A, __m128 __B, __m128 __C)
132 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
135 static __inline__ __m128d __DEFAULT_FN_ATTRS128
136 _mm_fmsubadd_pd(__m128d __A, __m128d __B, __m128d __C)
138 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
141 static __inline__ __m256 __DEFAULT_FN_ATTRS256
142 _mm256_fmadd_ps(__m256 __A, __m256 __B, __m256 __C)
144 return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
147 static __inline__ __m256d __DEFAULT_FN_ATTRS256
148 _mm256_fmadd_pd(__m256d __A, __m256d __B, __m256d __C)
150 return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
153 static __inline__ __m256 __DEFAULT_FN_ATTRS256
154 _mm256_fmsub_ps(__m256 __A, __m256 __B, __m256 __C)
156 return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
159 static __inline__ __m256d __DEFAULT_FN_ATTRS256
160 _mm256_fmsub_pd(__m256d __A, __m256d __B, __m256d __C)
162 return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
165 static __inline__ __m256 __DEFAULT_FN_ATTRS256
166 _mm256_fnmadd_ps(__m256 __A, __m256 __B, __m256 __C)
168 return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
171 static __inline__ __m256d __DEFAULT_FN_ATTRS256
172 _mm256_fnmadd_pd(__m256d __A, __m256d __B, __m256d __C)
174 return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, (__v4df)__C);
177 static __inline__ __m256 __DEFAULT_FN_ATTRS256
178 _mm256_fnmsub_ps(__m256 __A, __m256 __B, __m256 __C)
180 return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
183 static __inline__ __m256d __DEFAULT_FN_ATTRS256
184 _mm256_fnmsub_pd(__m256d __A, __m256d __B, __m256d __C)
186 return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, -(__v4df)__C);
189 static __inline__ __m256 __DEFAULT_FN_ATTRS256
190 _mm256_fmaddsub_ps(__m256 __A, __m256 __B, __m256 __C)
192 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
195 static __inline__ __m256d __DEFAULT_FN_ATTRS256
196 _mm256_fmaddsub_pd(__m256d __A, __m256d __B, __m256d __C)
198 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
201 static __inline__ __m256 __DEFAULT_FN_ATTRS256
202 _mm256_fmsubadd_ps(__m256 __A, __m256 __B, __m256 __C)
204 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
207 static __inline__ __m256d __DEFAULT_FN_ATTRS256
208 _mm256_fmsubadd_pd(__m256d __A, __m256d __B, __m256d __C)
210 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
213 #undef __DEFAULT_FN_ATTRS128
214 #undef __DEFAULT_FN_ATTRS256
216 #endif /* __FMAINTRIN_H */