blob: 11aa8ceacf375c7c3815520f5fe8ed2027e2bf6c [file] [log] [blame]
Ying Wang60999142013-01-07 13:59:36 -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
Ying Wang60999142013-01-07 13:59:36 -080031#include <pmmintrin.h>
32
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -070033/* Define the default attributes for the functions in this file. */
34#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("fma4")))
35
36static __inline__ __m128 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -080037_mm_macc_ps(__m128 __A, __m128 __B, __m128 __C)
38{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -070039 return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Ying Wang60999142013-01-07 13:59:36 -080040}
41
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -070042static __inline__ __m128d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -080043_mm_macc_pd(__m128d __A, __m128d __B, __m128d __C)
44{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -070045 return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Ying Wang60999142013-01-07 13:59:36 -080046}
47
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -070048static __inline__ __m128 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -080049_mm_macc_ss(__m128 __A, __m128 __B, __m128 __C)
50{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -070051 return (__m128)__builtin_ia32_vfmaddss((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Ying Wang60999142013-01-07 13:59:36 -080052}
53
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -070054static __inline__ __m128d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -080055_mm_macc_sd(__m128d __A, __m128d __B, __m128d __C)
56{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -070057 return (__m128d)__builtin_ia32_vfmaddsd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Ying Wang60999142013-01-07 13:59:36 -080058}
59
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -070060static __inline__ __m128 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -080061_mm_msub_ps(__m128 __A, __m128 __B, __m128 __C)
62{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -070063 return (__m128)__builtin_ia32_vfmsubps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Ying Wang60999142013-01-07 13:59:36 -080064}
65
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -070066static __inline__ __m128d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -080067_mm_msub_pd(__m128d __A, __m128d __B, __m128d __C)
68{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -070069 return (__m128d)__builtin_ia32_vfmsubpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Ying Wang60999142013-01-07 13:59:36 -080070}
71
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -070072static __inline__ __m128 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -080073_mm_msub_ss(__m128 __A, __m128 __B, __m128 __C)
74{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -070075 return (__m128)__builtin_ia32_vfmsubss((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Ying Wang60999142013-01-07 13:59:36 -080076}
77
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -070078static __inline__ __m128d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -080079_mm_msub_sd(__m128d __A, __m128d __B, __m128d __C)
80{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -070081 return (__m128d)__builtin_ia32_vfmsubsd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Ying Wang60999142013-01-07 13:59:36 -080082}
83
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -070084static __inline__ __m128 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -080085_mm_nmacc_ps(__m128 __A, __m128 __B, __m128 __C)
86{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -070087 return (__m128)__builtin_ia32_vfnmaddps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Ying Wang60999142013-01-07 13:59:36 -080088}
89
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -070090static __inline__ __m128d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -080091_mm_nmacc_pd(__m128d __A, __m128d __B, __m128d __C)
92{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -070093 return (__m128d)__builtin_ia32_vfnmaddpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Ying Wang60999142013-01-07 13:59:36 -080094}
95
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -070096static __inline__ __m128 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -080097_mm_nmacc_ss(__m128 __A, __m128 __B, __m128 __C)
98{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -070099 return (__m128)__builtin_ia32_vfnmaddss((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Ying Wang60999142013-01-07 13:59:36 -0800100}
101
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700102static __inline__ __m128d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800103_mm_nmacc_sd(__m128d __A, __m128d __B, __m128d __C)
104{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700105 return (__m128d)__builtin_ia32_vfnmaddsd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Ying Wang60999142013-01-07 13:59:36 -0800106}
107
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700108static __inline__ __m128 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800109_mm_nmsub_ps(__m128 __A, __m128 __B, __m128 __C)
110{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700111 return (__m128)__builtin_ia32_vfnmsubps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Ying Wang60999142013-01-07 13:59:36 -0800112}
113
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700114static __inline__ __m128d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800115_mm_nmsub_pd(__m128d __A, __m128d __B, __m128d __C)
116{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700117 return (__m128d)__builtin_ia32_vfnmsubpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Ying Wang60999142013-01-07 13:59:36 -0800118}
119
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700120static __inline__ __m128 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800121_mm_nmsub_ss(__m128 __A, __m128 __B, __m128 __C)
122{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700123 return (__m128)__builtin_ia32_vfnmsubss((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Ying Wang60999142013-01-07 13:59:36 -0800124}
125
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700126static __inline__ __m128d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800127_mm_nmsub_sd(__m128d __A, __m128d __B, __m128d __C)
128{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700129 return (__m128d)__builtin_ia32_vfnmsubsd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Ying Wang60999142013-01-07 13:59:36 -0800130}
131
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700132static __inline__ __m128 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800133_mm_maddsub_ps(__m128 __A, __m128 __B, __m128 __C)
134{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700135 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Ying Wang60999142013-01-07 13:59:36 -0800136}
137
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700138static __inline__ __m128d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800139_mm_maddsub_pd(__m128d __A, __m128d __B, __m128d __C)
140{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700141 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Ying Wang60999142013-01-07 13:59:36 -0800142}
143
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700144static __inline__ __m128 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800145_mm_msubadd_ps(__m128 __A, __m128 __B, __m128 __C)
146{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700147 return (__m128)__builtin_ia32_vfmsubaddps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
Ying Wang60999142013-01-07 13:59:36 -0800148}
149
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700150static __inline__ __m128d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800151_mm_msubadd_pd(__m128d __A, __m128d __B, __m128d __C)
152{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700153 return (__m128d)__builtin_ia32_vfmsubaddpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
Ying Wang60999142013-01-07 13:59:36 -0800154}
155
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700156static __inline__ __m256 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800157_mm256_macc_ps(__m256 __A, __m256 __B, __m256 __C)
158{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700159 return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Ying Wang60999142013-01-07 13:59:36 -0800160}
161
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700162static __inline__ __m256d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800163_mm256_macc_pd(__m256d __A, __m256d __B, __m256d __C)
164{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700165 return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
Ying Wang60999142013-01-07 13:59:36 -0800166}
167
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700168static __inline__ __m256 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800169_mm256_msub_ps(__m256 __A, __m256 __B, __m256 __C)
170{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700171 return (__m256)__builtin_ia32_vfmsubps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Ying Wang60999142013-01-07 13:59:36 -0800172}
173
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700174static __inline__ __m256d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800175_mm256_msub_pd(__m256d __A, __m256d __B, __m256d __C)
176{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700177 return (__m256d)__builtin_ia32_vfmsubpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
Ying Wang60999142013-01-07 13:59:36 -0800178}
179
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700180static __inline__ __m256 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800181_mm256_nmacc_ps(__m256 __A, __m256 __B, __m256 __C)
182{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700183 return (__m256)__builtin_ia32_vfnmaddps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Ying Wang60999142013-01-07 13:59:36 -0800184}
185
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700186static __inline__ __m256d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800187_mm256_nmacc_pd(__m256d __A, __m256d __B, __m256d __C)
188{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700189 return (__m256d)__builtin_ia32_vfnmaddpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
Ying Wang60999142013-01-07 13:59:36 -0800190}
191
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700192static __inline__ __m256 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800193_mm256_nmsub_ps(__m256 __A, __m256 __B, __m256 __C)
194{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700195 return (__m256)__builtin_ia32_vfnmsubps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Ying Wang60999142013-01-07 13:59:36 -0800196}
197
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700198static __inline__ __m256d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800199_mm256_nmsub_pd(__m256d __A, __m256d __B, __m256d __C)
200{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700201 return (__m256d)__builtin_ia32_vfnmsubpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
Ying Wang60999142013-01-07 13:59:36 -0800202}
203
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700204static __inline__ __m256 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800205_mm256_maddsub_ps(__m256 __A, __m256 __B, __m256 __C)
206{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700207 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Ying Wang60999142013-01-07 13:59:36 -0800208}
209
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700210static __inline__ __m256d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800211_mm256_maddsub_pd(__m256d __A, __m256d __B, __m256d __C)
212{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700213 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
Ying Wang60999142013-01-07 13:59:36 -0800214}
215
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700216static __inline__ __m256 __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800217_mm256_msubadd_ps(__m256 __A, __m256 __B, __m256 __C)
218{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700219 return (__m256)__builtin_ia32_vfmsubaddps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
Ying Wang60999142013-01-07 13:59:36 -0800220}
221
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700222static __inline__ __m256d __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800223_mm256_msubadd_pd(__m256d __A, __m256d __B, __m256d __C)
224{
Pirama Arumuga Nainarbb4374f2016-10-20 16:43:03 -0700225 return (__m256d)__builtin_ia32_vfmsubaddpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
Ying Wang60999142013-01-07 13:59:36 -0800226}
227
Pirama Arumuga Nainar4e74a022016-03-17 18:03:02 -0700228#undef __DEFAULT_FN_ATTRS
Ying Wang60999142013-01-07 13:59:36 -0800229
230#endif /* __FMA4INTRIN_H */