1/*===---- fma4intrin.h - FMA4 intrinsics -----------------------------------===
2 *
3 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 * See https://llvm.org/LICENSE.txt for license information.
5 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 *
7 *===-----------------------------------------------------------------------===
8 */
9
10#ifndef __X86INTRIN_H
11#error "Never use <fma4intrin.h> directly; include <x86intrin.h> instead."
12#endif
13
14#ifndef __FMA4INTRIN_H
15#define __FMA4INTRIN_H
16
17#include <pmmintrin.h>
18
19/* Define the default attributes for the functions in this file. */
20#define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("fma4"), __min_vector_width__(128)))
21#define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("fma4"), __min_vector_width__(256)))
22
23#if defined(__cplusplus) && (__cplusplus >= 201103L)
24#define __DEFAULT_FN_ATTRS128_CONSTEXPR __DEFAULT_FN_ATTRS128 constexpr
25#define __DEFAULT_FN_ATTRS256_CONSTEXPR __DEFAULT_FN_ATTRS256 constexpr
26#else
27#define __DEFAULT_FN_ATTRS128_CONSTEXPR __DEFAULT_FN_ATTRS128
28#define __DEFAULT_FN_ATTRS256_CONSTEXPR __DEFAULT_FN_ATTRS256
29#endif
30
31static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
32_mm_macc_ps(__m128 __A, __m128 __B, __m128 __C) {
33 return (__m128)__builtin_elementwise_fma((__v4sf)__A, (__v4sf)__B,
34 (__v4sf)__C);
35}
36
37static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
38_mm_macc_pd(__m128d __A, __m128d __B, __m128d __C) {
39 return (__m128d)__builtin_elementwise_fma((__v2df)__A, (__v2df)__B,
40 (__v2df)__C);
41}
42
43static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
44_mm_macc_ss(__m128 __A, __m128 __B, __m128 __C) {
45 return _mm_set_ss(__builtin_elementwise_fma(__A[0], __B[0], __C[0]));
46}
47
48static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
49_mm_macc_sd(__m128d __A, __m128d __B, __m128d __C) {
50 return _mm_set_sd(__builtin_elementwise_fma(__A[0], __B[0], __C[0]));
51}
52
53static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
54_mm_msub_ps(__m128 __A, __m128 __B, __m128 __C) {
55 return (__m128)__builtin_elementwise_fma((__v4sf)__A, (__v4sf)__B,
56 -(__v4sf)__C);
57}
58
59static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
60_mm_msub_pd(__m128d __A, __m128d __B, __m128d __C) {
61 return (__m128d)__builtin_elementwise_fma((__v2df)__A, (__v2df)__B,
62 -(__v2df)__C);
63}
64
65static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
66_mm_msub_ss(__m128 __A, __m128 __B, __m128 __C) {
67 return _mm_set_ss(__builtin_elementwise_fma(__A[0], __B[0], -__C[0]));
68}
69
70static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
71_mm_msub_sd(__m128d __A, __m128d __B, __m128d __C) {
72 return _mm_set_sd(__builtin_elementwise_fma(__A[0], __B[0], -__C[0]));
73}
74
75static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
76_mm_nmacc_ps(__m128 __A, __m128 __B, __m128 __C) {
77 return (__m128)__builtin_elementwise_fma(-(__v4sf)__A, (__v4sf)__B,
78 (__v4sf)__C);
79}
80
81static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
82_mm_nmacc_pd(__m128d __A, __m128d __B, __m128d __C) {
83 return (__m128d)__builtin_elementwise_fma(-(__v2df)__A, (__v2df)__B,
84 (__v2df)__C);
85}
86
87static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
88_mm_nmacc_ss(__m128 __A, __m128 __B, __m128 __C) {
89 return _mm_set_ss(__builtin_elementwise_fma(-__A[0], __B[0], __C[0]));
90}
91
92static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
93_mm_nmacc_sd(__m128d __A, __m128d __B, __m128d __C) {
94 return _mm_set_sd(__builtin_elementwise_fma(-__A[0], __B[0], __C[0]));
95}
96
97static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
98_mm_nmsub_ps(__m128 __A, __m128 __B, __m128 __C) {
99 return (__m128)__builtin_elementwise_fma(-(__v4sf)__A, (__v4sf)__B,
100 -(__v4sf)__C);
101}
102
103static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
104_mm_nmsub_pd(__m128d __A, __m128d __B, __m128d __C) {
105 return (__m128d)__builtin_elementwise_fma(-(__v2df)__A, (__v2df)__B,
106 -(__v2df)__C);
107}
108
109static __inline__ __m128 __DEFAULT_FN_ATTRS128_CONSTEXPR
110_mm_nmsub_ss(__m128 __A, __m128 __B, __m128 __C) {
111 return _mm_set_ss(__builtin_elementwise_fma(-__A[0], __B[0], -__C[0]));
112}
113
114static __inline__ __m128d __DEFAULT_FN_ATTRS128_CONSTEXPR
115_mm_nmsub_sd(__m128d __A, __m128d __B, __m128d __C) {
116 return _mm_set_sd(__builtin_elementwise_fma(-__A[0], __B[0], -__C[0]));
117}
118
119static __inline__ __m128 __DEFAULT_FN_ATTRS128
120_mm_maddsub_ps(__m128 __A, __m128 __B, __m128 __C)
121{
122 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
123}
124
125static __inline__ __m128d __DEFAULT_FN_ATTRS128
126_mm_maddsub_pd(__m128d __A, __m128d __B, __m128d __C)
127{
128 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, (__v2df)__C);
129}
130
131static __inline__ __m128 __DEFAULT_FN_ATTRS128
132_mm_msubadd_ps(__m128 __A, __m128 __B, __m128 __C)
133{
134 return (__m128)__builtin_ia32_vfmaddsubps((__v4sf)__A, (__v4sf)__B, -(__v4sf)__C);
135}
136
137static __inline__ __m128d __DEFAULT_FN_ATTRS128
138_mm_msubadd_pd(__m128d __A, __m128d __B, __m128d __C)
139{
140 return (__m128d)__builtin_ia32_vfmaddsubpd((__v2df)__A, (__v2df)__B, -(__v2df)__C);
141}
142
143static __inline__ __m256 __DEFAULT_FN_ATTRS256_CONSTEXPR
144_mm256_macc_ps(__m256 __A, __m256 __B, __m256 __C) {
145 return (__m256)__builtin_elementwise_fma((__v8sf)__A, (__v8sf)__B,
146 (__v8sf)__C);
147}
148
149static __inline__ __m256d __DEFAULT_FN_ATTRS256_CONSTEXPR
150_mm256_macc_pd(__m256d __A, __m256d __B, __m256d __C) {
151 return (__m256d)__builtin_elementwise_fma((__v4df)__A, (__v4df)__B,
152 (__v4df)__C);
153}
154
155static __inline__ __m256 __DEFAULT_FN_ATTRS256_CONSTEXPR
156_mm256_msub_ps(__m256 __A, __m256 __B, __m256 __C) {
157 return (__m256)__builtin_elementwise_fma((__v8sf)__A, (__v8sf)__B,
158 -(__v8sf)__C);
159}
160
161static __inline__ __m256d __DEFAULT_FN_ATTRS256_CONSTEXPR
162_mm256_msub_pd(__m256d __A, __m256d __B, __m256d __C) {
163 return (__m256d)__builtin_elementwise_fma((__v4df)__A, (__v4df)__B,
164 -(__v4df)__C);
165}
166
167static __inline__ __m256 __DEFAULT_FN_ATTRS256_CONSTEXPR
168_mm256_nmacc_ps(__m256 __A, __m256 __B, __m256 __C) {
169 return (__m256)__builtin_elementwise_fma(-(__v8sf)__A, (__v8sf)__B,
170 (__v8sf)__C);
171}
172
173static __inline__ __m256d __DEFAULT_FN_ATTRS256_CONSTEXPR
174_mm256_nmacc_pd(__m256d __A, __m256d __B, __m256d __C) {
175 return (__m256d)__builtin_elementwise_fma(-(__v4df)__A, (__v4df)__B,
176 (__v4df)__C);
177}
178
179static __inline__ __m256 __DEFAULT_FN_ATTRS256_CONSTEXPR
180_mm256_nmsub_ps(__m256 __A, __m256 __B, __m256 __C) {
181 return (__m256)__builtin_elementwise_fma(-(__v8sf)__A, (__v8sf)__B,
182 -(__v8sf)__C);
183}
184
185static __inline__ __m256d __DEFAULT_FN_ATTRS256_CONSTEXPR
186_mm256_nmsub_pd(__m256d __A, __m256d __B, __m256d __C) {
187 return (__m256d)__builtin_elementwise_fma(-(__v4df)__A, (__v4df)__B,
188 -(__v4df)__C);
189}
190
191static __inline__ __m256 __DEFAULT_FN_ATTRS256
192_mm256_maddsub_ps(__m256 __A, __m256 __B, __m256 __C)
193{
194 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
195}
196
197static __inline__ __m256d __DEFAULT_FN_ATTRS256
198_mm256_maddsub_pd(__m256d __A, __m256d __B, __m256d __C)
199{
200 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, (__v4df)__C);
201}
202
203static __inline__ __m256 __DEFAULT_FN_ATTRS256
204_mm256_msubadd_ps(__m256 __A, __m256 __B, __m256 __C)
205{
206 return (__m256)__builtin_ia32_vfmaddsubps256((__v8sf)__A, (__v8sf)__B, -(__v8sf)__C);
207}
208
209static __inline__ __m256d __DEFAULT_FN_ATTRS256
210_mm256_msubadd_pd(__m256d __A, __m256d __B, __m256d __C)
211{
212 return (__m256d)__builtin_ia32_vfmaddsubpd256((__v4df)__A, (__v4df)__B, -(__v4df)__C);
213}
214
215#undef __DEFAULT_FN_ATTRS128
216#undef __DEFAULT_FN_ATTRS256
217#undef __DEFAULT_FN_ATTRS128_CONSTEXPR
218#undef __DEFAULT_FN_ATTRS256_CONSTEXPR
219
220#endif /* __FMA4INTRIN_H */