avx512vnnivlintrin.h revision 1.1.1.1 1 1.1 mrg /* Copyright (C) 2013-2018 Free Software Foundation, Inc.
2 1.1 mrg
3 1.1 mrg This file is part of GCC.
4 1.1 mrg
5 1.1 mrg GCC is free software; you can redistribute it and/or modify
6 1.1 mrg it under the terms of the GNU General Public License as published by
7 1.1 mrg the Free Software Foundation; either version 3, or (at your option)
8 1.1 mrg any later version.
9 1.1 mrg
10 1.1 mrg GCC is distributed in the hope that it will be useful,
11 1.1 mrg but WITHOUT ANY WARRANTY; without even the implied warranty of
12 1.1 mrg MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 1.1 mrg GNU General Public License for more details.
14 1.1 mrg
15 1.1 mrg Under Section 7 of GPL version 3, you are granted additional
16 1.1 mrg permissions described in the GCC Runtime Library Exception, version
17 1.1 mrg 3.1, as published by the Free Software Foundation.
18 1.1 mrg
19 1.1 mrg You should have received a copy of the GNU General Public License and
20 1.1 mrg a copy of the GCC Runtime Library Exception along with this program;
21 1.1 mrg see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
22 1.1 mrg <http://www.gnu.org/licenses/>. */
23 1.1 mrg
24 1.1 mrg #ifndef _IMMINTRIN_H_INCLUDED
25 1.1 mrg #error "Never use <avx512vnnivlintrin.h> directly; include <immintrin.h> instead."
26 1.1 mrg #endif
27 1.1 mrg
28 1.1 mrg #ifndef _AVX512VNNIVLINTRIN_H_INCLUDED
29 1.1 mrg #define _AVX512VNNIVLINTRIN_H_INCLUDED
30 1.1 mrg
31 1.1 mrg #if !defined(__AVX512VL__) || !defined(__AVX512VNNI__)
32 1.1 mrg #pragma GCC push_options
33 1.1 mrg #pragma GCC target("avx512vnni,avx512vl")
34 1.1 mrg #define __DISABLE_AVX512VNNIVL__
35 1.1 mrg #endif /* __AVX512VNNIVL__ */
36 1.1 mrg
37 1.1 mrg extern __inline __m256i
38 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
39 1.1 mrg _mm256_dpbusd_epi32 (__m256i __A, __m256i __B, __m256i __C)
40 1.1 mrg {
41 1.1 mrg return (__m256i) __builtin_ia32_vpdpbusd_v8si ((__v8si)__A, (__v8si) __B,
42 1.1 mrg (__v8si) __C);
43 1.1 mrg }
44 1.1 mrg
45 1.1 mrg extern __inline __m256i
46 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
47 1.1 mrg _mm256_mask_dpbusd_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
48 1.1 mrg {
49 1.1 mrg return (__m256i)__builtin_ia32_vpdpbusd_v8si_mask ((__v8si)__A, (__v8si) __C,
50 1.1 mrg (__v8si) __D, (__mmask8)__B);
51 1.1 mrg }
52 1.1 mrg
53 1.1 mrg extern __inline __m256i
54 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
55 1.1 mrg _mm256_maskz_dpbusd_epi32 (__mmask8 __A, __m256i __B, __m256i __C, __m256i __D)
56 1.1 mrg {
57 1.1 mrg return (__m256i)__builtin_ia32_vpdpbusd_v8si_maskz ((__v8si)__B,
58 1.1 mrg (__v8si) __C, (__v8si) __D, (__mmask8)__A);
59 1.1 mrg }
60 1.1 mrg
61 1.1 mrg extern __inline __m128i
62 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
63 1.1 mrg _mm_dpbusd_epi32 (__m128i __A, __m128i __B, __m128i __C)
64 1.1 mrg {
65 1.1 mrg return (__m128i) __builtin_ia32_vpdpbusd_v4si ((__v4si)__A, (__v4si) __B,
66 1.1 mrg (__v4si) __C);
67 1.1 mrg }
68 1.1 mrg
69 1.1 mrg extern __inline __m128i
70 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
71 1.1 mrg _mm_mask_dpbusd_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
72 1.1 mrg {
73 1.1 mrg return (__m128i)__builtin_ia32_vpdpbusd_v4si_mask ((__v4si)__A, (__v4si) __C,
74 1.1 mrg (__v4si) __D, (__mmask8)__B);
75 1.1 mrg }
76 1.1 mrg
77 1.1 mrg extern __inline __m128i
78 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
79 1.1 mrg _mm_maskz_dpbusd_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
80 1.1 mrg {
81 1.1 mrg return (__m128i)__builtin_ia32_vpdpbusd_v4si_maskz ((__v4si)__B,
82 1.1 mrg (__v4si) __C, (__v4si) __D, (__mmask8)__A);
83 1.1 mrg }
84 1.1 mrg
85 1.1 mrg extern __inline __m256i
86 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
87 1.1 mrg _mm256_dpbusds_epi32 (__m256i __A, __m256i __B, __m256i __C)
88 1.1 mrg {
89 1.1 mrg return (__m256i) __builtin_ia32_vpdpbusds_v8si ((__v8si)__A, (__v8si) __B,
90 1.1 mrg (__v8si) __C);
91 1.1 mrg }
92 1.1 mrg
93 1.1 mrg extern __inline __m256i
94 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
95 1.1 mrg _mm256_mask_dpbusds_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
96 1.1 mrg {
97 1.1 mrg return (__m256i)__builtin_ia32_vpdpbusds_v8si_mask ((__v8si)__A,
98 1.1 mrg (__v8si) __C, (__v8si) __D, (__mmask8)__B);
99 1.1 mrg }
100 1.1 mrg
101 1.1 mrg extern __inline __m256i
102 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
103 1.1 mrg _mm256_maskz_dpbusds_epi32 (__mmask8 __A, __m256i __B, __m256i __C,
104 1.1 mrg __m256i __D)
105 1.1 mrg {
106 1.1 mrg return (__m256i)__builtin_ia32_vpdpbusds_v8si_maskz ((__v8si)__B,
107 1.1 mrg (__v8si) __C, (__v8si) __D, (__mmask8)__A);
108 1.1 mrg }
109 1.1 mrg
110 1.1 mrg extern __inline __m128i
111 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
112 1.1 mrg _mm_dpbusds_epi32 (__m128i __A, __m128i __B, __m128i __C)
113 1.1 mrg {
114 1.1 mrg return (__m128i) __builtin_ia32_vpdpbusds_v4si ((__v4si)__A, (__v4si) __B,
115 1.1 mrg (__v4si) __C);
116 1.1 mrg }
117 1.1 mrg
118 1.1 mrg extern __inline __m128i
119 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
120 1.1 mrg _mm_mask_dpbusds_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
121 1.1 mrg {
122 1.1 mrg return (__m128i)__builtin_ia32_vpdpbusds_v4si_mask ((__v4si)__A,
123 1.1 mrg (__v4si) __C, (__v4si) __D, (__mmask8)__B);
124 1.1 mrg }
125 1.1 mrg
126 1.1 mrg extern __inline __m128i
127 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
128 1.1 mrg _mm_maskz_dpbusds_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
129 1.1 mrg {
130 1.1 mrg return (__m128i)__builtin_ia32_vpdpbusds_v4si_maskz ((__v4si)__B,
131 1.1 mrg (__v4si) __C, (__v4si) __D, (__mmask8)__A);
132 1.1 mrg }
133 1.1 mrg
134 1.1 mrg extern __inline __m256i
135 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
136 1.1 mrg _mm256_dpwssd_epi32 (__m256i __A, __m256i __B, __m256i __C)
137 1.1 mrg {
138 1.1 mrg return (__m256i) __builtin_ia32_vpdpwssd_v8si ((__v8si)__A, (__v8si) __B,
139 1.1 mrg (__v8si) __C);
140 1.1 mrg }
141 1.1 mrg
142 1.1 mrg extern __inline __m256i
143 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
144 1.1 mrg _mm256_mask_dpwssd_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
145 1.1 mrg {
146 1.1 mrg return (__m256i)__builtin_ia32_vpdpwssd_v8si_mask ((__v8si)__A, (__v8si) __C,
147 1.1 mrg (__v8si) __D, (__mmask8)__B);
148 1.1 mrg }
149 1.1 mrg
150 1.1 mrg extern __inline __m256i
151 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
152 1.1 mrg _mm256_maskz_dpwssd_epi32 (__mmask8 __A, __m256i __B, __m256i __C, __m256i __D)
153 1.1 mrg {
154 1.1 mrg return (__m256i)__builtin_ia32_vpdpwssd_v8si_maskz ((__v8si)__B,
155 1.1 mrg (__v8si) __C, (__v8si) __D, (__mmask8)__A);
156 1.1 mrg }
157 1.1 mrg
158 1.1 mrg extern __inline __m128i
159 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
160 1.1 mrg _mm_dpwssd_epi32 (__m128i __A, __m128i __B, __m128i __C)
161 1.1 mrg {
162 1.1 mrg return (__m128i) __builtin_ia32_vpdpwssd_v4si ((__v4si)__A, (__v4si) __B,
163 1.1 mrg (__v4si) __C);
164 1.1 mrg }
165 1.1 mrg
166 1.1 mrg extern __inline __m128i
167 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
168 1.1 mrg _mm_mask_dpwssd_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
169 1.1 mrg {
170 1.1 mrg return (__m128i)__builtin_ia32_vpdpwssd_v4si_mask ((__v4si)__A, (__v4si) __C,
171 1.1 mrg (__v4si) __D, (__mmask8)__B);
172 1.1 mrg }
173 1.1 mrg
174 1.1 mrg extern __inline __m128i
175 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
176 1.1 mrg _mm_maskz_dpwssd_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
177 1.1 mrg {
178 1.1 mrg return (__m128i)__builtin_ia32_vpdpwssd_v4si_maskz ((__v4si)__B,
179 1.1 mrg (__v4si) __C, (__v4si) __D, (__mmask8)__A);
180 1.1 mrg }
181 1.1 mrg
182 1.1 mrg extern __inline __m256i
183 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
184 1.1 mrg _mm256_dpwssds_epi32 (__m256i __A, __m256i __B, __m256i __C)
185 1.1 mrg {
186 1.1 mrg return (__m256i) __builtin_ia32_vpdpwssds_v8si ((__v8si)__A, (__v8si) __B,
187 1.1 mrg (__v8si) __C);
188 1.1 mrg }
189 1.1 mrg
190 1.1 mrg extern __inline __m256i
191 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
192 1.1 mrg _mm256_mask_dpwssds_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
193 1.1 mrg {
194 1.1 mrg return (__m256i)__builtin_ia32_vpdpwssds_v8si_mask ((__v8si)__A,
195 1.1 mrg (__v8si) __C, (__v8si) __D, (__mmask8)__B);
196 1.1 mrg }
197 1.1 mrg
198 1.1 mrg extern __inline __m256i
199 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
200 1.1 mrg _mm256_maskz_dpwssds_epi32 (__mmask8 __A, __m256i __B, __m256i __C,
201 1.1 mrg __m256i __D)
202 1.1 mrg {
203 1.1 mrg return (__m256i)__builtin_ia32_vpdpwssds_v8si_maskz ((__v8si)__B,
204 1.1 mrg (__v8si) __C, (__v8si) __D, (__mmask8)__A);
205 1.1 mrg }
206 1.1 mrg
207 1.1 mrg extern __inline __m128i
208 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
209 1.1 mrg _mm_dpwssds_epi32 (__m128i __A, __m128i __B, __m128i __C)
210 1.1 mrg {
211 1.1 mrg return (__m128i) __builtin_ia32_vpdpwssds_v4si ((__v4si)__A, (__v4si) __B,
212 1.1 mrg (__v4si) __C);
213 1.1 mrg }
214 1.1 mrg
215 1.1 mrg extern __inline __m128i
216 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
217 1.1 mrg _mm_mask_dpwssds_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
218 1.1 mrg {
219 1.1 mrg return (__m128i)__builtin_ia32_vpdpwssds_v4si_mask ((__v4si)__A,
220 1.1 mrg (__v4si) __C, (__v4si) __D, (__mmask8)__B);
221 1.1 mrg }
222 1.1 mrg
223 1.1 mrg extern __inline __m128i
224 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
225 1.1 mrg _mm_maskz_dpwssds_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
226 1.1 mrg {
227 1.1 mrg return (__m128i)__builtin_ia32_vpdpwssds_v4si_maskz ((__v4si)__B,
228 1.1 mrg (__v4si) __C, (__v4si) __D, (__mmask8)__A);
229 1.1 mrg }
230 1.1 mrg #ifdef __DISABLE_AVX512VNNIVL__
231 1.1 mrg #undef __DISABLE_AVX512VNNIVL__
232 1.1 mrg #pragma GCC pop_options
233 1.1 mrg #endif /* __DISABLE_AVX512VNNIVL__ */
234 1.1 mrg #endif /* __DISABLE_AVX512VNNIVL__ */
235