blob: 7bae2f4a3155c04713d4a8db2e003c35784978f3 [file] [log] [blame]
Logan Chien2833ffb2018-10-09 10:03:24 +08001/*===---- fma4intrin.h - FMA4 intrinsics -----------------------------------===
2 *
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:
9 *
10 * The above copyright notice and this permission notice shall be included in
11 * all copies or substantial portions of the Software.
12 *
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
19 * THE SOFTWARE.
20 *
21 *===-----------------------------------------------------------------------===
22 */
23
24#ifndef __X86INTRIN_H
25#error "Never use <fma4intrin.h> directly; include <x86intrin.h> instead."
26#endif
27
28#ifndef __FMA4INTRIN_H
29#define __FMA4INTRIN_H
30
31#include <pmmintrin.h>
32
33/* Define the default attributes for the functions in this file. */
Logan Chien55afb0a2018-10-15 10:42:14 +080034#define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("fma4"), __min_vector_width__(128)))
35#define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("fma4"), __min_vector_width__(256)))
Logan Chien2833ffb2018-10-09 10:03:24 +080036
Logan Chien55afb0a2018-10-15 10:42:14 +080037static __inline__ __m128 __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +080038_mm_macc_ps(__m128 __A, __m128 __B, __m128 __C)
39{
40 return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
41}
42
Logan Chien55afb0a2018-10-15 10:42:14 +080043static __inline__ __m128d __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +080044_mm_macc_pd(__m128d __A, __m128d __B, __m128d __C)
45{
46 return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
47}
48
Logan Chien55afb0a2018-10-15 10:42:14 +080049static __inline__ __m128 __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +080050_mm_macc_ss(__m128 __A, __m128 __B, __m128 __C)
51{
52 return (__m128)__builtin_ia32_vfmaddss((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
53}
54
Logan Chien55afb0a2018-10-15 10:42:14 +080055static __inline__ __m128d __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +080056_mm_macc_sd(__m128d __A, __m128d __B, __m128d __C)
57{
58 return (__m128d)__builtin_ia32_vfmaddsd((__v2df)__A, (__v2df)__B, (__v2df)__C);
59}
60
Logan Chien55afb0a2018-10-15 10:42:14 +080061static __inline__ __m128 __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +080062_mm_msub_ps(__m128 __A, __m128 __B, __m128 __C)
63{
Logan Chien55afb0a2018-10-15 10:42:14 +080064 return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +080065}
66
Logan Chien55afb0a2018-10-15 10:42:14 +080067static __inline__ __m128d __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +080068_mm_msub_pd(__m128d __A, __m128d __B, __m128d __C)
69{
Logan Chien55afb0a2018-10-15 10:42:14 +080070 return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +080071}
72
Logan Chien55afb0a2018-10-15 10:42:14 +080073static __inline__ __m128 __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +080074_mm_msub_ss(__m128 __A, __m128 __B, __m128 __C)
75{
Logan Chien55afb0a2018-10-15 10:42:14 +080076 return (__m128)__builtin_ia32_vfmaddss((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +080077}
78
Logan Chien55afb0a2018-10-15 10:42:14 +080079static __inline__ __m128d __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +080080_mm_msub_sd(__m128d __A, __m128d __B, __m128d __C)
81{
Logan Chien55afb0a2018-10-15 10:42:14 +080082 return (__m128d)__builtin_ia32_vfmaddsd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +080083}
84
Logan Chien55afb0a2018-10-15 10:42:14 +080085static __inline__ __m128 __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +080086_mm_nmacc_ps(__m128 __A, __m128 __B, __m128 __C)
87{
Logan Chien55afb0a2018-10-15 10:42:14 +080088 return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +080089}
90
Logan Chien55afb0a2018-10-15 10:42:14 +080091static __inline__ __m128d __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +080092_mm_nmacc_pd(__m128d __A, __m128d __B, __m128d __C)
93{
Logan Chien55afb0a2018-10-15 10:42:14 +080094 return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, (__v2df)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +080095}
96
Logan Chien55afb0a2018-10-15 10:42:14 +080097static __inline__ __m128 __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +080098_mm_nmacc_ss(__m128 __A, __m128 __B, __m128 __C)
99{
Logan Chien55afb0a2018-10-15 10:42:14 +0800100 return (__m128)__builtin_ia32_vfmaddss(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800101}
102
Logan Chien55afb0a2018-10-15 10:42:14 +0800103static __inline__ __m128d __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +0800104_mm_nmacc_sd(__m128d __A, __m128d __B, __m128d __C)
105{
Logan Chien55afb0a2018-10-15 10:42:14 +0800106 return (__m128d)__builtin_ia32_vfmaddsd(-(__v2df)__A, (__v2df)__B, (__v2df)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800107}
108
Logan Chien55afb0a2018-10-15 10:42:14 +0800109static __inline__ __m128 __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +0800110_mm_nmsub_ps(__m128 __A, __m128 __B, __m128 __C)
111{
Logan Chien55afb0a2018-10-15 10:42:14 +0800112 return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800113}
114
Logan Chien55afb0a2018-10-15 10:42:14 +0800115static __inline__ __m128d __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +0800116_mm_nmsub_pd(__m128d __A, __m128d __B, __m128d __C)
117{
Logan Chien55afb0a2018-10-15 10:42:14 +0800118 return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800119}
120
Logan Chien55afb0a2018-10-15 10:42:14 +0800121static __inline__ __m128 __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +0800122_mm_nmsub_ss(__m128 __A, __m128 __B, __m128 __C)
123{
Logan Chien55afb0a2018-10-15 10:42:14 +0800124 return (__m128)__builtin_ia32_vfmaddss(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800125}
126
Logan Chien55afb0a2018-10-15 10:42:14 +0800127static __inline__ __m128d __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +0800128_mm_nmsub_sd(__m128d __A, __m128d __B, __m128d __C)
129{
Logan Chien55afb0a2018-10-15 10:42:14 +0800130 return (__m128d)__builtin_ia32_vfmaddsd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800131}
132
Logan Chien55afb0a2018-10-15 10:42:14 +0800133static __inline__ __m128 __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +0800134_mm_maddsub_ps(__m128 __A, __m128 __B, __m128 __C)
135{
136 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
137}
138
Logan Chien55afb0a2018-10-15 10:42:14 +0800139static __inline__ __m128d __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +0800140_mm_maddsub_pd(__m128d __A, __m128d __B, __m128d __C)
141{
142 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
143}
144
Logan Chien55afb0a2018-10-15 10:42:14 +0800145static __inline__ __m128 __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +0800146_mm_msubadd_ps(__m128 __A, __m128 __B, __m128 __C)
147{
Logan Chien55afb0a2018-10-15 10:42:14 +0800148 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800149}
150
Logan Chien55afb0a2018-10-15 10:42:14 +0800151static __inline__ __m128d __DEFAULT_FN_ATTRS128
Logan Chien2833ffb2018-10-09 10:03:24 +0800152_mm_msubadd_pd(__m128d __A, __m128d __B, __m128d __C)
153{
Logan Chien55afb0a2018-10-15 10:42:14 +0800154 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800155}
156
Logan Chien55afb0a2018-10-15 10:42:14 +0800157static __inline__ __m256 __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800158_mm256_macc_ps(__m256 __A, __m256 __B, __m256 __C)
159{
160 return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
161}
162
Logan Chien55afb0a2018-10-15 10:42:14 +0800163static __inline__ __m256d __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800164_mm256_macc_pd(__m256d __A, __m256d __B, __m256d __C)
165{
166 return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
167}
168
Logan Chien55afb0a2018-10-15 10:42:14 +0800169static __inline__ __m256 __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800170_mm256_msub_ps(__m256 __A, __m256 __B, __m256 __C)
171{
Logan Chien55afb0a2018-10-15 10:42:14 +0800172 return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800173}
174
Logan Chien55afb0a2018-10-15 10:42:14 +0800175static __inline__ __m256d __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800176_mm256_msub_pd(__m256d __A, __m256d __B, __m256d __C)
177{
Logan Chien55afb0a2018-10-15 10:42:14 +0800178 return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800179}
180
Logan Chien55afb0a2018-10-15 10:42:14 +0800181static __inline__ __m256 __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800182_mm256_nmacc_ps(__m256 __A, __m256 __B, __m256 __C)
183{
Logan Chien55afb0a2018-10-15 10:42:14 +0800184 return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800185}
186
Logan Chien55afb0a2018-10-15 10:42:14 +0800187static __inline__ __m256d __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800188_mm256_nmacc_pd(__m256d __A, __m256d __B, __m256d __C)
189{
Logan Chien55afb0a2018-10-15 10:42:14 +0800190 return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, (__v4df)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800191}
192
Logan Chien55afb0a2018-10-15 10:42:14 +0800193static __inline__ __m256 __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800194_mm256_nmsub_ps(__m256 __A, __m256 __B, __m256 __C)
195{
Logan Chien55afb0a2018-10-15 10:42:14 +0800196 return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800197}
198
Logan Chien55afb0a2018-10-15 10:42:14 +0800199static __inline__ __m256d __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800200_mm256_nmsub_pd(__m256d __A, __m256d __B, __m256d __C)
201{
Logan Chien55afb0a2018-10-15 10:42:14 +0800202 return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, -(__v4df)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800203}
204
Logan Chien55afb0a2018-10-15 10:42:14 +0800205static __inline__ __m256 __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800206_mm256_maddsub_ps(__m256 __A, __m256 __B, __m256 __C)
207{
208 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
209}
210
Logan Chien55afb0a2018-10-15 10:42:14 +0800211static __inline__ __m256d __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800212_mm256_maddsub_pd(__m256d __A, __m256d __B, __m256d __C)
213{
214 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
215}
216
Logan Chien55afb0a2018-10-15 10:42:14 +0800217static __inline__ __m256 __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800218_mm256_msubadd_ps(__m256 __A, __m256 __B, __m256 __C)
219{
Logan Chien55afb0a2018-10-15 10:42:14 +0800220 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800221}
222
Logan Chien55afb0a2018-10-15 10:42:14 +0800223static __inline__ __m256d __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800224_mm256_msubadd_pd(__m256d __A, __m256d __B, __m256d __C)
225{
Logan Chien55afb0a2018-10-15 10:42:14 +0800226 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
Logan Chien2833ffb2018-10-09 10:03:24 +0800227}
228
Logan Chien55afb0a2018-10-15 10:42:14 +0800229#undef __DEFAULT_FN_ATTRS128
230#undef __DEFAULT_FN_ATTRS256
Logan Chien2833ffb2018-10-09 10:03:24 +0800231
232#endif /* __FMA4INTRIN_H */