annotate clang/lib/Headers/fma4intrin.h @ 176:de4ac79aef9d

...
author Shinji KONO <kono@ie.u-ryukyu.ac.jp>
date Mon, 25 May 2020 17:13:11 +0900
parents 1d019706d866
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
150
anatofuz
parents:
diff changeset
1 /*===---- fma4intrin.h - FMA4 intrinsics -----------------------------------===
anatofuz
parents:
diff changeset
2 *
anatofuz
parents:
diff changeset
3 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
anatofuz
parents:
diff changeset
4 * See https://llvm.org/LICENSE.txt for license information.
anatofuz
parents:
diff changeset
5 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
anatofuz
parents:
diff changeset
6 *
anatofuz
parents:
diff changeset
7 *===-----------------------------------------------------------------------===
anatofuz
parents:
diff changeset
8 */
anatofuz
parents:
diff changeset
9
anatofuz
parents:
diff changeset
10 #ifndef __X86INTRIN_H
anatofuz
parents:
diff changeset
11 #error "Never use <fma4intrin.h> directly; include <x86intrin.h> instead."
anatofuz
parents:
diff changeset
12 #endif
anatofuz
parents:
diff changeset
13
anatofuz
parents:
diff changeset
14 #ifndef __FMA4INTRIN_H
anatofuz
parents:
diff changeset
15 #define __FMA4INTRIN_H
anatofuz
parents:
diff changeset
16
anatofuz
parents:
diff changeset
17 #include <pmmintrin.h>
anatofuz
parents:
diff changeset
18
anatofuz
parents:
diff changeset
19 /* Define the default attributes for the functions in this file. */
anatofuz
parents:
diff changeset
20 #define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("fma4"), __min_vector_width__(128)))
anatofuz
parents:
diff changeset
21 #define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("fma4"), __min_vector_width__(256)))
anatofuz
parents:
diff changeset
22
anatofuz
parents:
diff changeset
23 static __inline__ __m128 __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
24 _mm_macc_ps(__m128 __A, __m128 __B, __m128 __C)
anatofuz
parents:
diff changeset
25 {
anatofuz
parents:
diff changeset
26 return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
anatofuz
parents:
diff changeset
27 }
anatofuz
parents:
diff changeset
28
anatofuz
parents:
diff changeset
29 static __inline__ __m128d __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
30 _mm_macc_pd(__m128d __A, __m128d __B, __m128d __C)
anatofuz
parents:
diff changeset
31 {
anatofuz
parents:
diff changeset
32 return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
anatofuz
parents:
diff changeset
33 }
anatofuz
parents:
diff changeset
34
anatofuz
parents:
diff changeset
35 static __inline__ __m128 __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
36 _mm_macc_ss(__m128 __A, __m128 __B, __m128 __C)
anatofuz
parents:
diff changeset
37 {
anatofuz
parents:
diff changeset
38 return (__m128)__builtin_ia32_vfmaddss((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
anatofuz
parents:
diff changeset
39 }
anatofuz
parents:
diff changeset
40
anatofuz
parents:
diff changeset
41 static __inline__ __m128d __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
42 _mm_macc_sd(__m128d __A, __m128d __B, __m128d __C)
anatofuz
parents:
diff changeset
43 {
anatofuz
parents:
diff changeset
44 return (__m128d)__builtin_ia32_vfmaddsd((__v2df)__A, (__v2df)__B, (__v2df)__C);
anatofuz
parents:
diff changeset
45 }
anatofuz
parents:
diff changeset
46
anatofuz
parents:
diff changeset
47 static __inline__ __m128 __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
48 _mm_msub_ps(__m128 __A, __m128 __B, __m128 __C)
anatofuz
parents:
diff changeset
49 {
anatofuz
parents:
diff changeset
50 return (__m128)__builtin_ia32_vfmaddps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
anatofuz
parents:
diff changeset
51 }
anatofuz
parents:
diff changeset
52
anatofuz
parents:
diff changeset
53 static __inline__ __m128d __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
54 _mm_msub_pd(__m128d __A, __m128d __B, __m128d __C)
anatofuz
parents:
diff changeset
55 {
anatofuz
parents:
diff changeset
56 return (__m128d)__builtin_ia32_vfmaddpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
anatofuz
parents:
diff changeset
57 }
anatofuz
parents:
diff changeset
58
anatofuz
parents:
diff changeset
59 static __inline__ __m128 __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
60 _mm_msub_ss(__m128 __A, __m128 __B, __m128 __C)
anatofuz
parents:
diff changeset
61 {
anatofuz
parents:
diff changeset
62 return (__m128)__builtin_ia32_vfmaddss((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
anatofuz
parents:
diff changeset
63 }
anatofuz
parents:
diff changeset
64
anatofuz
parents:
diff changeset
65 static __inline__ __m128d __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
66 _mm_msub_sd(__m128d __A, __m128d __B, __m128d __C)
anatofuz
parents:
diff changeset
67 {
anatofuz
parents:
diff changeset
68 return (__m128d)__builtin_ia32_vfmaddsd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
anatofuz
parents:
diff changeset
69 }
anatofuz
parents:
diff changeset
70
anatofuz
parents:
diff changeset
71 static __inline__ __m128 __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
72 _mm_nmacc_ps(__m128 __A, __m128 __B, __m128 __C)
anatofuz
parents:
diff changeset
73 {
anatofuz
parents:
diff changeset
74 return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
anatofuz
parents:
diff changeset
75 }
anatofuz
parents:
diff changeset
76
anatofuz
parents:
diff changeset
77 static __inline__ __m128d __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
78 _mm_nmacc_pd(__m128d __A, __m128d __B, __m128d __C)
anatofuz
parents:
diff changeset
79 {
anatofuz
parents:
diff changeset
80 return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, (__v2df)__C);
anatofuz
parents:
diff changeset
81 }
anatofuz
parents:
diff changeset
82
anatofuz
parents:
diff changeset
83 static __inline__ __m128 __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
84 _mm_nmacc_ss(__m128 __A, __m128 __B, __m128 __C)
anatofuz
parents:
diff changeset
85 {
anatofuz
parents:
diff changeset
86 return (__m128)__builtin_ia32_vfmaddss(-(__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
anatofuz
parents:
diff changeset
87 }
anatofuz
parents:
diff changeset
88
anatofuz
parents:
diff changeset
89 static __inline__ __m128d __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
90 _mm_nmacc_sd(__m128d __A, __m128d __B, __m128d __C)
anatofuz
parents:
diff changeset
91 {
anatofuz
parents:
diff changeset
92 return (__m128d)__builtin_ia32_vfmaddsd(-(__v2df)__A, (__v2df)__B, (__v2df)__C);
anatofuz
parents:
diff changeset
93 }
anatofuz
parents:
diff changeset
94
anatofuz
parents:
diff changeset
95 static __inline__ __m128 __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
96 _mm_nmsub_ps(__m128 __A, __m128 __B, __m128 __C)
anatofuz
parents:
diff changeset
97 {
anatofuz
parents:
diff changeset
98 return (__m128)__builtin_ia32_vfmaddps(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
anatofuz
parents:
diff changeset
99 }
anatofuz
parents:
diff changeset
100
anatofuz
parents:
diff changeset
101 static __inline__ __m128d __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
102 _mm_nmsub_pd(__m128d __A, __m128d __B, __m128d __C)
anatofuz
parents:
diff changeset
103 {
anatofuz
parents:
diff changeset
104 return (__m128d)__builtin_ia32_vfmaddpd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C);
anatofuz
parents:
diff changeset
105 }
anatofuz
parents:
diff changeset
106
anatofuz
parents:
diff changeset
107 static __inline__ __m128 __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
108 _mm_nmsub_ss(__m128 __A, __m128 __B, __m128 __C)
anatofuz
parents:
diff changeset
109 {
anatofuz
parents:
diff changeset
110 return (__m128)__builtin_ia32_vfmaddss(-(__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
anatofuz
parents:
diff changeset
111 }
anatofuz
parents:
diff changeset
112
anatofuz
parents:
diff changeset
113 static __inline__ __m128d __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
114 _mm_nmsub_sd(__m128d __A, __m128d __B, __m128d __C)
anatofuz
parents:
diff changeset
115 {
anatofuz
parents:
diff changeset
116 return (__m128d)__builtin_ia32_vfmaddsd(-(__v2df)__A, (__v2df)__B, -(__v2df)__C);
anatofuz
parents:
diff changeset
117 }
anatofuz
parents:
diff changeset
118
anatofuz
parents:
diff changeset
119 static __inline__ __m128 __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
120 _mm_maddsub_ps(__m128 __A, __m128 __B, __m128 __C)
anatofuz
parents:
diff changeset
121 {
anatofuz
parents:
diff changeset
122 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
anatofuz
parents:
diff changeset
123 }
anatofuz
parents:
diff changeset
124
anatofuz
parents:
diff changeset
125 static __inline__ __m128d __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
126 _mm_maddsub_pd(__m128d __A, __m128d __B, __m128d __C)
anatofuz
parents:
diff changeset
127 {
anatofuz
parents:
diff changeset
128 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
anatofuz
parents:
diff changeset
129 }
anatofuz
parents:
diff changeset
130
anatofuz
parents:
diff changeset
131 static __inline__ __m128 __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
132 _mm_msubadd_ps(__m128 __A, __m128 __B, __m128 __C)
anatofuz
parents:
diff changeset
133 {
anatofuz
parents:
diff changeset
134 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
anatofuz
parents:
diff changeset
135 }
anatofuz
parents:
diff changeset
136
anatofuz
parents:
diff changeset
137 static __inline__ __m128d __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
138 _mm_msubadd_pd(__m128d __A, __m128d __B, __m128d __C)
anatofuz
parents:
diff changeset
139 {
anatofuz
parents:
diff changeset
140 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
anatofuz
parents:
diff changeset
141 }
anatofuz
parents:
diff changeset
142
anatofuz
parents:
diff changeset
143 static __inline__ __m256 __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
144 _mm256_macc_ps(__m256 __A, __m256 __B, __m256 __C)
anatofuz
parents:
diff changeset
145 {
anatofuz
parents:
diff changeset
146 return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
anatofuz
parents:
diff changeset
147 }
anatofuz
parents:
diff changeset
148
anatofuz
parents:
diff changeset
149 static __inline__ __m256d __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
150 _mm256_macc_pd(__m256d __A, __m256d __B, __m256d __C)
anatofuz
parents:
diff changeset
151 {
anatofuz
parents:
diff changeset
152 return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
anatofuz
parents:
diff changeset
153 }
anatofuz
parents:
diff changeset
154
anatofuz
parents:
diff changeset
155 static __inline__ __m256 __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
156 _mm256_msub_ps(__m256 __A, __m256 __B, __m256 __C)
anatofuz
parents:
diff changeset
157 {
anatofuz
parents:
diff changeset
158 return (__m256)__builtin_ia32_vfmaddps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
anatofuz
parents:
diff changeset
159 }
anatofuz
parents:
diff changeset
160
anatofuz
parents:
diff changeset
161 static __inline__ __m256d __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
162 _mm256_msub_pd(__m256d __A, __m256d __B, __m256d __C)
anatofuz
parents:
diff changeset
163 {
anatofuz
parents:
diff changeset
164 return (__m256d)__builtin_ia32_vfmaddpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
anatofuz
parents:
diff changeset
165 }
anatofuz
parents:
diff changeset
166
anatofuz
parents:
diff changeset
167 static __inline__ __m256 __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
168 _mm256_nmacc_ps(__m256 __A, __m256 __B, __m256 __C)
anatofuz
parents:
diff changeset
169 {
anatofuz
parents:
diff changeset
170 return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
anatofuz
parents:
diff changeset
171 }
anatofuz
parents:
diff changeset
172
anatofuz
parents:
diff changeset
173 static __inline__ __m256d __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
174 _mm256_nmacc_pd(__m256d __A, __m256d __B, __m256d __C)
anatofuz
parents:
diff changeset
175 {
anatofuz
parents:
diff changeset
176 return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, (__v4df)__C);
anatofuz
parents:
diff changeset
177 }
anatofuz
parents:
diff changeset
178
anatofuz
parents:
diff changeset
179 static __inline__ __m256 __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
180 _mm256_nmsub_ps(__m256 __A, __m256 __B, __m256 __C)
anatofuz
parents:
diff changeset
181 {
anatofuz
parents:
diff changeset
182 return (__m256)__builtin_ia32_vfmaddps256(-(__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
anatofuz
parents:
diff changeset
183 }
anatofuz
parents:
diff changeset
184
anatofuz
parents:
diff changeset
185 static __inline__ __m256d __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
186 _mm256_nmsub_pd(__m256d __A, __m256d __B, __m256d __C)
anatofuz
parents:
diff changeset
187 {
anatofuz
parents:
diff changeset
188 return (__m256d)__builtin_ia32_vfmaddpd256(-(__v4df)__A, (__v4df)__B, -(__v4df)__C);
anatofuz
parents:
diff changeset
189 }
anatofuz
parents:
diff changeset
190
anatofuz
parents:
diff changeset
191 static __inline__ __m256 __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
192 _mm256_maddsub_ps(__m256 __A, __m256 __B, __m256 __C)
anatofuz
parents:
diff changeset
193 {
anatofuz
parents:
diff changeset
194 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
anatofuz
parents:
diff changeset
195 }
anatofuz
parents:
diff changeset
196
anatofuz
parents:
diff changeset
197 static __inline__ __m256d __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
198 _mm256_maddsub_pd(__m256d __A, __m256d __B, __m256d __C)
anatofuz
parents:
diff changeset
199 {
anatofuz
parents:
diff changeset
200 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
anatofuz
parents:
diff changeset
201 }
anatofuz
parents:
diff changeset
202
anatofuz
parents:
diff changeset
203 static __inline__ __m256 __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
204 _mm256_msubadd_ps(__m256 __A, __m256 __B, __m256 __C)
anatofuz
parents:
diff changeset
205 {
anatofuz
parents:
diff changeset
206 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
anatofuz
parents:
diff changeset
207 }
anatofuz
parents:
diff changeset
208
anatofuz
parents:
diff changeset
209 static __inline__ __m256d __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
210 _mm256_msubadd_pd(__m256d __A, __m256d __B, __m256d __C)
anatofuz
parents:
diff changeset
211 {
anatofuz
parents:
diff changeset
212 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
anatofuz
parents:
diff changeset
213 }
anatofuz
parents:
diff changeset
214
anatofuz
parents:
diff changeset
215 #undef __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
216 #undef __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
217
anatofuz
parents:
diff changeset
218 #endif /* __FMA4INTRIN_H */