blob: 962b1a60a2585e5678c442707d3d5acffdc7e872 [file] [log] [blame]
Craig Topperb4ceb6f2011-12-30 09:15:03 +00001/*===---- 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
Craig Topperb4ceb6f2011-12-30 09:15:03 +000031#include <pmmintrin.h>
32
Eric Christopher4d1851682015-06-17 07:09:20 +000033/* Define the default attributes for the functions in this file. */
Michael Kupersteine45af542015-06-30 13:36:19 +000034#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("fma4")))
Eric Christopher4d1851682015-06-17 07:09:20 +000035
Michael Kupersteine45af542015-06-30 13:36:19 +000036static __inline__ __m128 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +000037_mm_macc_ps(__m128 __A, __m128 __B, __m128 __C)
38{
Craig Topper1aa231e2016-05-16 06:38:42 +000039 return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +000040}
41
Michael Kupersteine45af542015-06-30 13:36:19 +000042static __inline__ __m128d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +000043_mm_macc_pd(__m128d __A, __m128d __B, __m128d __C)
44{
Craig Topper1aa231e2016-05-16 06:38:42 +000045 return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +000046}
47
Michael Kupersteine45af542015-06-30 13:36:19 +000048static __inline__ __m128 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +000049_mm_macc_ss(__m128 __A, __m128 __B, __m128 __C)
50{
Craig Topper9e032ed2017-11-25 19:32:12 +000051 return (__m128)__builtin_ia32_vfmaddss((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +000052}
53
Michael Kupersteine45af542015-06-30 13:36:19 +000054static __inline__ __m128d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +000055_mm_macc_sd(__m128d __A, __m128d __B, __m128d __C)
56{
Craig Topper9e032ed2017-11-25 19:32:12 +000057 return (__m128d)__builtin_ia32_vfmaddsd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +000058}
59
Michael Kupersteine45af542015-06-30 13:36:19 +000060static __inline__ __m128 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +000061_mm_msub_ps(__m128 __A, __m128 __B, __m128 __C)
62{
Craig Topperb3d44732017-11-10 05:20:32 +000063 return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +000064}
65
Michael Kupersteine45af542015-06-30 13:36:19 +000066static __inline__ __m128d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +000067_mm_msub_pd(__m128d __A, __m128d __B, __m128d __C)
68{
Craig Topperb3d44732017-11-10 05:20:32 +000069 return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +000070}
71
Michael Kupersteine45af542015-06-30 13:36:19 +000072static __inline__ __m128 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +000073_mm_msub_ss(__m128 __A, __m128 __B, __m128 __C)
74{
Craig Topper9e032ed2017-11-25 19:32:12 +000075 return (__m128)__builtin_ia32_vfmaddss((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +000076}
77
Michael Kupersteine45af542015-06-30 13:36:19 +000078static __inline__ __m128d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +000079_mm_msub_sd(__m128d __A, __m128d __B, __m128d __C)
80{
Craig Topper9e032ed2017-11-25 19:32:12 +000081 return (__m128d)__builtin_ia32_vfmaddsd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +000082}
83
Michael Kupersteine45af542015-06-30 13:36:19 +000084static __inline__ __m128 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +000085_mm_nmacc_ps(__m128 __A, __m128 __B, __m128 __C)
86{
Craig Topperb3d44732017-11-10 05:20:32 +000087 return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +000088}
89
Michael Kupersteine45af542015-06-30 13:36:19 +000090static __inline__ __m128d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +000091_mm_nmacc_pd(__m128d __A, __m128d __B, __m128d __C)
92{
Craig Topperb3d44732017-11-10 05:20:32 +000093 return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, (__v2df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +000094}
95
Michael Kupersteine45af542015-06-30 13:36:19 +000096static __inline__ __m128 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +000097_mm_nmacc_ss(__m128 __A, __m128 __B, __m128 __C)
98{
Craig Topper9e032ed2017-11-25 19:32:12 +000099 return (__m128)__builtin_ia32_vfmaddss(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000100}
101
Michael Kupersteine45af542015-06-30 13:36:19 +0000102static __inline__ __m128d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000103_mm_nmacc_sd(__m128d __A, __m128d __B, __m128d __C)
104{
Craig Topper9e032ed2017-11-25 19:32:12 +0000105 return (__m128d)__builtin_ia32_vfmaddsd(-(__v2df)__A, (__v2df)__B, (__v2df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000106}
107
Michael Kupersteine45af542015-06-30 13:36:19 +0000108static __inline__ __m128 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000109_mm_nmsub_ps(__m128 __A, __m128 __B, __m128 __C)
110{
Craig Topperb3d44732017-11-10 05:20:32 +0000111 return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000112}
113
Michael Kupersteine45af542015-06-30 13:36:19 +0000114static __inline__ __m128d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000115_mm_nmsub_pd(__m128d __A, __m128d __B, __m128d __C)
116{
Craig Topperb3d44732017-11-10 05:20:32 +0000117 return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000118}
119
Michael Kupersteine45af542015-06-30 13:36:19 +0000120static __inline__ __m128 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000121_mm_nmsub_ss(__m128 __A, __m128 __B, __m128 __C)
122{
Craig Topper9e032ed2017-11-25 19:32:12 +0000123 return (__m128)__builtin_ia32_vfmaddss(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000124}
125
Michael Kupersteine45af542015-06-30 13:36:19 +0000126static __inline__ __m128d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000127_mm_nmsub_sd(__m128d __A, __m128d __B, __m128d __C)
128{
Craig Topper9e032ed2017-11-25 19:32:12 +0000129 return (__m128d)__builtin_ia32_vfmaddsd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000130}
131
Michael Kupersteine45af542015-06-30 13:36:19 +0000132static __inline__ __m128 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000133_mm_maddsub_ps(__m128 __A, __m128 __B, __m128 __C)
134{
Craig Topper1aa231e2016-05-16 06:38:42 +0000135 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000136}
137
Michael Kupersteine45af542015-06-30 13:36:19 +0000138static __inline__ __m128d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000139_mm_maddsub_pd(__m128d __A, __m128d __B, __m128d __C)
140{
Craig Topper1aa231e2016-05-16 06:38:42 +0000141 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000142}
143
Michael Kupersteine45af542015-06-30 13:36:19 +0000144static __inline__ __m128 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000145_mm_msubadd_ps(__m128 __A, __m128 __B, __m128 __C)
146{
Craig Topperb3d44732017-11-10 05:20:32 +0000147 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000148}
149
Michael Kupersteine45af542015-06-30 13:36:19 +0000150static __inline__ __m128d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000151_mm_msubadd_pd(__m128d __A, __m128d __B, __m128d __C)
152{
Craig Topperb3d44732017-11-10 05:20:32 +0000153 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000154}
155
Michael Kupersteine45af542015-06-30 13:36:19 +0000156static __inline__ __m256 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000157_mm256_macc_ps(__m256 __A, __m256 __B, __m256 __C)
158{
Craig Topper1aa231e2016-05-16 06:38:42 +0000159 return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000160}
161
Michael Kupersteine45af542015-06-30 13:36:19 +0000162static __inline__ __m256d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000163_mm256_macc_pd(__m256d __A, __m256d __B, __m256d __C)
164{
Craig Topper1aa231e2016-05-16 06:38:42 +0000165 return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000166}
167
Michael Kupersteine45af542015-06-30 13:36:19 +0000168static __inline__ __m256 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000169_mm256_msub_ps(__m256 __A, __m256 __B, __m256 __C)
170{
Craig Topperb3d44732017-11-10 05:20:32 +0000171 return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000172}
173
Michael Kupersteine45af542015-06-30 13:36:19 +0000174static __inline__ __m256d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000175_mm256_msub_pd(__m256d __A, __m256d __B, __m256d __C)
176{
Craig Topperb3d44732017-11-10 05:20:32 +0000177 return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000178}
179
Michael Kupersteine45af542015-06-30 13:36:19 +0000180static __inline__ __m256 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000181_mm256_nmacc_ps(__m256 __A, __m256 __B, __m256 __C)
182{
Craig Topperb3d44732017-11-10 05:20:32 +0000183 return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000184}
185
Michael Kupersteine45af542015-06-30 13:36:19 +0000186static __inline__ __m256d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000187_mm256_nmacc_pd(__m256d __A, __m256d __B, __m256d __C)
188{
Craig Topperb3d44732017-11-10 05:20:32 +0000189 return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, (__v4df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000190}
191
Michael Kupersteine45af542015-06-30 13:36:19 +0000192static __inline__ __m256 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000193_mm256_nmsub_ps(__m256 __A, __m256 __B, __m256 __C)
194{
Craig Topperb3d44732017-11-10 05:20:32 +0000195 return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000196}
197
Michael Kupersteine45af542015-06-30 13:36:19 +0000198static __inline__ __m256d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000199_mm256_nmsub_pd(__m256d __A, __m256d __B, __m256d __C)
200{
Craig Topperb3d44732017-11-10 05:20:32 +0000201 return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, -(__v4df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000202}
203
Michael Kupersteine45af542015-06-30 13:36:19 +0000204static __inline__ __m256 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000205_mm256_maddsub_ps(__m256 __A, __m256 __B, __m256 __C)
206{
Craig Topper1aa231e2016-05-16 06:38:42 +0000207 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000208}
209
Michael Kupersteine45af542015-06-30 13:36:19 +0000210static __inline__ __m256d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000211_mm256_maddsub_pd(__m256d __A, __m256d __B, __m256d __C)
212{
Craig Topper1aa231e2016-05-16 06:38:42 +0000213 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000214}
215
Michael Kupersteine45af542015-06-30 13:36:19 +0000216static __inline__ __m256 __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000217_mm256_msubadd_ps(__m256 __A, __m256 __B, __m256 __C)
218{
Craig Topperb3d44732017-11-10 05:20:32 +0000219 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000220}
221
Michael Kupersteine45af542015-06-30 13:36:19 +0000222static __inline__ __m256d __DEFAULT_FN_ATTRS
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000223_mm256_msubadd_pd(__m256d __A, __m256d __B, __m256d __C)
224{
Craig Topperb3d44732017-11-10 05:20:32 +0000225 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000226}
227
Michael Kupersteine45af542015-06-30 13:36:19 +0000228#undef __DEFAULT_FN_ATTRS
Eric Christopher4d1851682015-06-17 07:09:20 +0000229
Craig Topperb4ceb6f2011-12-30 09:15:03 +0000230#endif /* __FMA4INTRIN_H */