Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 1 | /*===---- 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 Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 34 | #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 Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 36 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 37 | static __inline__ __m128 __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 38 | _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 Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 43 | static __inline__ __m128d __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 44 | _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 Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 49 | static __inline__ __m128 __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 50 | _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 Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 55 | static __inline__ __m128d __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 56 | _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 Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 61 | static __inline__ __m128 __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 62 | _mm_msub_ps(__m128 __A, __m128 __B, __m128 __C) |
| 63 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 64 | return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 65 | } |
| 66 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 67 | static __inline__ __m128d __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 68 | _mm_msub_pd(__m128d __A, __m128d __B, __m128d __C) |
| 69 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 70 | return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, -(__v2df)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 71 | } |
| 72 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 73 | static __inline__ __m128 __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 74 | _mm_msub_ss(__m128 __A, __m128 __B, __m128 __C) |
| 75 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 76 | return (__m128)__builtin_ia32_vfmaddss((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 77 | } |
| 78 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 79 | static __inline__ __m128d __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 80 | _mm_msub_sd(__m128d __A, __m128d __B, __m128d __C) |
| 81 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 82 | return (__m128d)__builtin_ia32_vfmaddsd((__v2df)__A, (__v2df)__B, -(__v2df)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 83 | } |
| 84 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 85 | static __inline__ __m128 __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 86 | _mm_nmacc_ps(__m128 __A, __m128 __B, __m128 __C) |
| 87 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 88 | return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 89 | } |
| 90 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 91 | static __inline__ __m128d __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 92 | _mm_nmacc_pd(__m128d __A, __m128d __B, __m128d __C) |
| 93 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 94 | return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, (__v2df)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 95 | } |
| 96 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 97 | static __inline__ __m128 __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 98 | _mm_nmacc_ss(__m128 __A, __m128 __B, __m128 __C) |
| 99 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 100 | return (__m128)__builtin_ia32_vfmaddss(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 101 | } |
| 102 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 103 | static __inline__ __m128d __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 104 | _mm_nmacc_sd(__m128d __A, __m128d __B, __m128d __C) |
| 105 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 106 | return (__m128d)__builtin_ia32_vfmaddsd(-(__v2df)__A, (__v2df)__B, (__v2df)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 107 | } |
| 108 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 109 | static __inline__ __m128 __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 110 | _mm_nmsub_ps(__m128 __A, __m128 __B, __m128 __C) |
| 111 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 112 | return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 113 | } |
| 114 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 115 | static __inline__ __m128d __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 116 | _mm_nmsub_pd(__m128d __A, __m128d __B, __m128d __C) |
| 117 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 118 | return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 119 | } |
| 120 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 121 | static __inline__ __m128 __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 122 | _mm_nmsub_ss(__m128 __A, __m128 __B, __m128 __C) |
| 123 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 124 | return (__m128)__builtin_ia32_vfmaddss(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 125 | } |
| 126 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 127 | static __inline__ __m128d __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 128 | _mm_nmsub_sd(__m128d __A, __m128d __B, __m128d __C) |
| 129 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 130 | return (__m128d)__builtin_ia32_vfmaddsd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 131 | } |
| 132 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 133 | static __inline__ __m128 __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 134 | _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 Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 139 | static __inline__ __m128d __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 140 | _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 Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 145 | static __inline__ __m128 __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 146 | _mm_msubadd_ps(__m128 __A, __m128 __B, __m128 __C) |
| 147 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 148 | return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 149 | } |
| 150 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 151 | static __inline__ __m128d __DEFAULT_FN_ATTRS128 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 152 | _mm_msubadd_pd(__m128d __A, __m128d __B, __m128d __C) |
| 153 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 154 | return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, -(__v2df)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 155 | } |
| 156 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 157 | static __inline__ __m256 __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 158 | _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 Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 163 | static __inline__ __m256d __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 164 | _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 Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 169 | static __inline__ __m256 __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 170 | _mm256_msub_ps(__m256 __A, __m256 __B, __m256 __C) |
| 171 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 172 | return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 173 | } |
| 174 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 175 | static __inline__ __m256d __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 176 | _mm256_msub_pd(__m256d __A, __m256d __B, __m256d __C) |
| 177 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 178 | return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 179 | } |
| 180 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 181 | static __inline__ __m256 __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 182 | _mm256_nmacc_ps(__m256 __A, __m256 __B, __m256 __C) |
| 183 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 184 | return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, (__v8sf)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 185 | } |
| 186 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 187 | static __inline__ __m256d __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 188 | _mm256_nmacc_pd(__m256d __A, __m256d __B, __m256d __C) |
| 189 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 190 | return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, (__v4df)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 191 | } |
| 192 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 193 | static __inline__ __m256 __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 194 | _mm256_nmsub_ps(__m256 __A, __m256 __B, __m256 __C) |
| 195 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 196 | return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, -(__v8sf)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 197 | } |
| 198 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 199 | static __inline__ __m256d __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 200 | _mm256_nmsub_pd(__m256d __A, __m256d __B, __m256d __C) |
| 201 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 202 | return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, -(__v4df)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 203 | } |
| 204 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 205 | static __inline__ __m256 __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 206 | _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 Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 211 | static __inline__ __m256d __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 212 | _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 Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 217 | static __inline__ __m256 __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 218 | _mm256_msubadd_ps(__m256 __A, __m256 __B, __m256 __C) |
| 219 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 220 | return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 221 | } |
| 222 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 223 | static __inline__ __m256d __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 224 | _mm256_msubadd_pd(__m256d __A, __m256d __B, __m256d __C) |
| 225 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 226 | return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 227 | } |
| 228 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame^] | 229 | #undef __DEFAULT_FN_ATTRS128 |
| 230 | #undef __DEFAULT_FN_ATTRS256 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 231 | |
| 232 | #endif /* __FMA4INTRIN_H */ |