145
|
1 /* Copyright (C) 2013-2020 Free Software Foundation, Inc.
|
131
|
2
|
|
3 This file is part of GCC.
|
|
4
|
|
5 GCC is free software; you can redistribute it and/or modify
|
|
6 it under the terms of the GNU General Public License as published by
|
|
7 the Free Software Foundation; either version 3, or (at your option)
|
|
8 any later version.
|
|
9
|
|
10 GCC is distributed in the hope that it will be useful,
|
|
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
13 GNU General Public License for more details.
|
|
14
|
|
15 Under Section 7 of GPL version 3, you are granted additional
|
|
16 permissions described in the GCC Runtime Library Exception, version
|
|
17 3.1, as published by the Free Software Foundation.
|
|
18
|
|
19 You should have received a copy of the GNU General Public License and
|
|
20 a copy of the GCC Runtime Library Exception along with this program;
|
|
21 see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
|
22 <http://www.gnu.org/licenses/>. */
|
|
23
|
|
24 #ifndef _IMMINTRIN_H_INCLUDED
|
|
25 #error "Never use <avx512vnnivlintrin.h> directly; include <immintrin.h> instead."
|
|
26 #endif
|
|
27
|
|
28 #ifndef _AVX512VNNIVLINTRIN_H_INCLUDED
|
|
29 #define _AVX512VNNIVLINTRIN_H_INCLUDED
|
|
30
|
|
31 #if !defined(__AVX512VL__) || !defined(__AVX512VNNI__)
|
|
32 #pragma GCC push_options
|
|
33 #pragma GCC target("avx512vnni,avx512vl")
|
|
34 #define __DISABLE_AVX512VNNIVL__
|
|
35 #endif /* __AVX512VNNIVL__ */
|
|
36
|
|
37 extern __inline __m256i
|
|
38 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
39 _mm256_dpbusd_epi32 (__m256i __A, __m256i __B, __m256i __C)
|
|
40 {
|
|
41 return (__m256i) __builtin_ia32_vpdpbusd_v8si ((__v8si)__A, (__v8si) __B,
|
|
42 (__v8si) __C);
|
|
43 }
|
|
44
|
|
45 extern __inline __m256i
|
|
46 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
47 _mm256_mask_dpbusd_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
|
|
48 {
|
|
49 return (__m256i)__builtin_ia32_vpdpbusd_v8si_mask ((__v8si)__A, (__v8si) __C,
|
|
50 (__v8si) __D, (__mmask8)__B);
|
|
51 }
|
|
52
|
|
53 extern __inline __m256i
|
|
54 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
55 _mm256_maskz_dpbusd_epi32 (__mmask8 __A, __m256i __B, __m256i __C, __m256i __D)
|
|
56 {
|
|
57 return (__m256i)__builtin_ia32_vpdpbusd_v8si_maskz ((__v8si)__B,
|
|
58 (__v8si) __C, (__v8si) __D, (__mmask8)__A);
|
|
59 }
|
|
60
|
|
61 extern __inline __m128i
|
|
62 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
63 _mm_dpbusd_epi32 (__m128i __A, __m128i __B, __m128i __C)
|
|
64 {
|
|
65 return (__m128i) __builtin_ia32_vpdpbusd_v4si ((__v4si)__A, (__v4si) __B,
|
|
66 (__v4si) __C);
|
|
67 }
|
|
68
|
|
69 extern __inline __m128i
|
|
70 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
71 _mm_mask_dpbusd_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
|
|
72 {
|
|
73 return (__m128i)__builtin_ia32_vpdpbusd_v4si_mask ((__v4si)__A, (__v4si) __C,
|
|
74 (__v4si) __D, (__mmask8)__B);
|
|
75 }
|
|
76
|
|
77 extern __inline __m128i
|
|
78 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
79 _mm_maskz_dpbusd_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
|
|
80 {
|
|
81 return (__m128i)__builtin_ia32_vpdpbusd_v4si_maskz ((__v4si)__B,
|
|
82 (__v4si) __C, (__v4si) __D, (__mmask8)__A);
|
|
83 }
|
|
84
|
|
85 extern __inline __m256i
|
|
86 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
87 _mm256_dpbusds_epi32 (__m256i __A, __m256i __B, __m256i __C)
|
|
88 {
|
|
89 return (__m256i) __builtin_ia32_vpdpbusds_v8si ((__v8si)__A, (__v8si) __B,
|
|
90 (__v8si) __C);
|
|
91 }
|
|
92
|
|
93 extern __inline __m256i
|
|
94 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
95 _mm256_mask_dpbusds_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
|
|
96 {
|
|
97 return (__m256i)__builtin_ia32_vpdpbusds_v8si_mask ((__v8si)__A,
|
|
98 (__v8si) __C, (__v8si) __D, (__mmask8)__B);
|
|
99 }
|
|
100
|
|
101 extern __inline __m256i
|
|
102 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
103 _mm256_maskz_dpbusds_epi32 (__mmask8 __A, __m256i __B, __m256i __C,
|
|
104 __m256i __D)
|
|
105 {
|
|
106 return (__m256i)__builtin_ia32_vpdpbusds_v8si_maskz ((__v8si)__B,
|
|
107 (__v8si) __C, (__v8si) __D, (__mmask8)__A);
|
|
108 }
|
|
109
|
|
110 extern __inline __m128i
|
|
111 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
112 _mm_dpbusds_epi32 (__m128i __A, __m128i __B, __m128i __C)
|
|
113 {
|
|
114 return (__m128i) __builtin_ia32_vpdpbusds_v4si ((__v4si)__A, (__v4si) __B,
|
|
115 (__v4si) __C);
|
|
116 }
|
|
117
|
|
118 extern __inline __m128i
|
|
119 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
120 _mm_mask_dpbusds_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
|
|
121 {
|
|
122 return (__m128i)__builtin_ia32_vpdpbusds_v4si_mask ((__v4si)__A,
|
|
123 (__v4si) __C, (__v4si) __D, (__mmask8)__B);
|
|
124 }
|
|
125
|
|
126 extern __inline __m128i
|
|
127 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
128 _mm_maskz_dpbusds_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
|
|
129 {
|
|
130 return (__m128i)__builtin_ia32_vpdpbusds_v4si_maskz ((__v4si)__B,
|
|
131 (__v4si) __C, (__v4si) __D, (__mmask8)__A);
|
|
132 }
|
|
133
|
|
134 extern __inline __m256i
|
|
135 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
136 _mm256_dpwssd_epi32 (__m256i __A, __m256i __B, __m256i __C)
|
|
137 {
|
|
138 return (__m256i) __builtin_ia32_vpdpwssd_v8si ((__v8si)__A, (__v8si) __B,
|
|
139 (__v8si) __C);
|
|
140 }
|
|
141
|
|
142 extern __inline __m256i
|
|
143 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
144 _mm256_mask_dpwssd_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
|
|
145 {
|
|
146 return (__m256i)__builtin_ia32_vpdpwssd_v8si_mask ((__v8si)__A, (__v8si) __C,
|
|
147 (__v8si) __D, (__mmask8)__B);
|
|
148 }
|
|
149
|
|
150 extern __inline __m256i
|
|
151 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
152 _mm256_maskz_dpwssd_epi32 (__mmask8 __A, __m256i __B, __m256i __C, __m256i __D)
|
|
153 {
|
|
154 return (__m256i)__builtin_ia32_vpdpwssd_v8si_maskz ((__v8si)__B,
|
|
155 (__v8si) __C, (__v8si) __D, (__mmask8)__A);
|
|
156 }
|
|
157
|
|
158 extern __inline __m128i
|
|
159 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
160 _mm_dpwssd_epi32 (__m128i __A, __m128i __B, __m128i __C)
|
|
161 {
|
|
162 return (__m128i) __builtin_ia32_vpdpwssd_v4si ((__v4si)__A, (__v4si) __B,
|
|
163 (__v4si) __C);
|
|
164 }
|
|
165
|
|
166 extern __inline __m128i
|
|
167 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
168 _mm_mask_dpwssd_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
|
|
169 {
|
|
170 return (__m128i)__builtin_ia32_vpdpwssd_v4si_mask ((__v4si)__A, (__v4si) __C,
|
|
171 (__v4si) __D, (__mmask8)__B);
|
|
172 }
|
|
173
|
|
174 extern __inline __m128i
|
|
175 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
176 _mm_maskz_dpwssd_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
|
|
177 {
|
|
178 return (__m128i)__builtin_ia32_vpdpwssd_v4si_maskz ((__v4si)__B,
|
|
179 (__v4si) __C, (__v4si) __D, (__mmask8)__A);
|
|
180 }
|
|
181
|
|
182 extern __inline __m256i
|
|
183 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
184 _mm256_dpwssds_epi32 (__m256i __A, __m256i __B, __m256i __C)
|
|
185 {
|
|
186 return (__m256i) __builtin_ia32_vpdpwssds_v8si ((__v8si)__A, (__v8si) __B,
|
|
187 (__v8si) __C);
|
|
188 }
|
|
189
|
|
190 extern __inline __m256i
|
|
191 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
192 _mm256_mask_dpwssds_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
|
|
193 {
|
|
194 return (__m256i)__builtin_ia32_vpdpwssds_v8si_mask ((__v8si)__A,
|
|
195 (__v8si) __C, (__v8si) __D, (__mmask8)__B);
|
|
196 }
|
|
197
|
|
198 extern __inline __m256i
|
|
199 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
200 _mm256_maskz_dpwssds_epi32 (__mmask8 __A, __m256i __B, __m256i __C,
|
|
201 __m256i __D)
|
|
202 {
|
|
203 return (__m256i)__builtin_ia32_vpdpwssds_v8si_maskz ((__v8si)__B,
|
|
204 (__v8si) __C, (__v8si) __D, (__mmask8)__A);
|
|
205 }
|
|
206
|
|
207 extern __inline __m128i
|
|
208 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
209 _mm_dpwssds_epi32 (__m128i __A, __m128i __B, __m128i __C)
|
|
210 {
|
|
211 return (__m128i) __builtin_ia32_vpdpwssds_v4si ((__v4si)__A, (__v4si) __B,
|
|
212 (__v4si) __C);
|
|
213 }
|
|
214
|
|
215 extern __inline __m128i
|
|
216 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
217 _mm_mask_dpwssds_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
|
|
218 {
|
|
219 return (__m128i)__builtin_ia32_vpdpwssds_v4si_mask ((__v4si)__A,
|
|
220 (__v4si) __C, (__v4si) __D, (__mmask8)__B);
|
|
221 }
|
|
222
|
|
223 extern __inline __m128i
|
|
224 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
225 _mm_maskz_dpwssds_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
|
|
226 {
|
|
227 return (__m128i)__builtin_ia32_vpdpwssds_v4si_maskz ((__v4si)__B,
|
|
228 (__v4si) __C, (__v4si) __D, (__mmask8)__A);
|
|
229 }
|
|
230 #ifdef __DISABLE_AVX512VNNIVL__
|
|
231 #undef __DISABLE_AVX512VNNIVL__
|
|
232 #pragma GCC pop_options
|
|
233 #endif /* __DISABLE_AVX512VNNIVL__ */
|
|
234 #endif /* __DISABLE_AVX512VNNIVL__ */
|