1 /* Copyright (C) 2013-2021 Free Software Foundation, Inc.
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 #define _mm256_dpbusd_epi32(A, B, C) \
38 ((__m256i) __builtin_ia32_vpdpbusd_v8si ((__v8si) (A), \
39 (__v8si) (B), \
40 (__v8si) (C)))
41
42 extern __inline __m256i
43 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_dpbusd_epi32(__m256i __A,__mmask8 __B,__m256i __C,__m256i __D)44 _mm256_mask_dpbusd_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
45 {
46 return (__m256i)__builtin_ia32_vpdpbusd_v8si_mask ((__v8si)__A, (__v8si) __C,
47 (__v8si) __D, (__mmask8)__B);
48 }
49
50 extern __inline __m256i
51 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_dpbusd_epi32(__mmask8 __A,__m256i __B,__m256i __C,__m256i __D)52 _mm256_maskz_dpbusd_epi32 (__mmask8 __A, __m256i __B, __m256i __C, __m256i __D)
53 {
54 return (__m256i)__builtin_ia32_vpdpbusd_v8si_maskz ((__v8si)__B,
55 (__v8si) __C, (__v8si) __D, (__mmask8)__A);
56 }
57
58 #define _mm_dpbusd_epi32(A, B, C) \
59 ((__m128i) __builtin_ia32_vpdpbusd_v4si ((__v4si) (A), \
60 (__v4si) (B), \
61 (__v4si) (C)))
62
63 extern __inline __m128i
64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_dpbusd_epi32(__m128i __A,__mmask8 __B,__m128i __C,__m128i __D)65 _mm_mask_dpbusd_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
66 {
67 return (__m128i)__builtin_ia32_vpdpbusd_v4si_mask ((__v4si)__A, (__v4si) __C,
68 (__v4si) __D, (__mmask8)__B);
69 }
70
71 extern __inline __m128i
72 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_dpbusd_epi32(__mmask8 __A,__m128i __B,__m128i __C,__m128i __D)73 _mm_maskz_dpbusd_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
74 {
75 return (__m128i)__builtin_ia32_vpdpbusd_v4si_maskz ((__v4si)__B,
76 (__v4si) __C, (__v4si) __D, (__mmask8)__A);
77 }
78
79 #define _mm256_dpbusds_epi32(A, B, C) \
80 ((__m256i) __builtin_ia32_vpdpbusds_v8si ((__v8si) (A), \
81 (__v8si) (B), \
82 (__v8si) (C)))
83
84 extern __inline __m256i
85 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_dpbusds_epi32(__m256i __A,__mmask8 __B,__m256i __C,__m256i __D)86 _mm256_mask_dpbusds_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
87 {
88 return (__m256i)__builtin_ia32_vpdpbusds_v8si_mask ((__v8si)__A,
89 (__v8si) __C, (__v8si) __D, (__mmask8)__B);
90 }
91
92 extern __inline __m256i
93 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_dpbusds_epi32(__mmask8 __A,__m256i __B,__m256i __C,__m256i __D)94 _mm256_maskz_dpbusds_epi32 (__mmask8 __A, __m256i __B, __m256i __C,
95 __m256i __D)
96 {
97 return (__m256i)__builtin_ia32_vpdpbusds_v8si_maskz ((__v8si)__B,
98 (__v8si) __C, (__v8si) __D, (__mmask8)__A);
99 }
100
101 #define _mm_dpbusds_epi32(A, B, C) \
102 ((__m128i) __builtin_ia32_vpdpbusds_v4si ((__v4si) (A), \
103 (__v4si) (B), \
104 (__v4si) (C)))
105
106 extern __inline __m128i
107 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_dpbusds_epi32(__m128i __A,__mmask8 __B,__m128i __C,__m128i __D)108 _mm_mask_dpbusds_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
109 {
110 return (__m128i)__builtin_ia32_vpdpbusds_v4si_mask ((__v4si)__A,
111 (__v4si) __C, (__v4si) __D, (__mmask8)__B);
112 }
113
114 extern __inline __m128i
115 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_dpbusds_epi32(__mmask8 __A,__m128i __B,__m128i __C,__m128i __D)116 _mm_maskz_dpbusds_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
117 {
118 return (__m128i)__builtin_ia32_vpdpbusds_v4si_maskz ((__v4si)__B,
119 (__v4si) __C, (__v4si) __D, (__mmask8)__A);
120 }
121
122 #define _mm256_dpwssd_epi32(A, B, C) \
123 ((__m256i) __builtin_ia32_vpdpwssd_v8si ((__v8si) (A), \
124 (__v8si) (B), \
125 (__v8si) (C)))
126
127 extern __inline __m256i
128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_dpwssd_epi32(__m256i __A,__mmask8 __B,__m256i __C,__m256i __D)129 _mm256_mask_dpwssd_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
130 {
131 return (__m256i)__builtin_ia32_vpdpwssd_v8si_mask ((__v8si)__A, (__v8si) __C,
132 (__v8si) __D, (__mmask8)__B);
133 }
134
135 extern __inline __m256i
136 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_dpwssd_epi32(__mmask8 __A,__m256i __B,__m256i __C,__m256i __D)137 _mm256_maskz_dpwssd_epi32 (__mmask8 __A, __m256i __B, __m256i __C, __m256i __D)
138 {
139 return (__m256i)__builtin_ia32_vpdpwssd_v8si_maskz ((__v8si)__B,
140 (__v8si) __C, (__v8si) __D, (__mmask8)__A);
141 }
142
143 #define _mm_dpwssd_epi32(A, B, C) \
144 ((__m128i) __builtin_ia32_vpdpwssd_v4si ((__v4si) (A), \
145 (__v4si) (B), \
146 (__v4si) (C)))
147
148 extern __inline __m128i
149 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_dpwssd_epi32(__m128i __A,__mmask8 __B,__m128i __C,__m128i __D)150 _mm_mask_dpwssd_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
151 {
152 return (__m128i)__builtin_ia32_vpdpwssd_v4si_mask ((__v4si)__A, (__v4si) __C,
153 (__v4si) __D, (__mmask8)__B);
154 }
155
156 extern __inline __m128i
157 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_dpwssd_epi32(__mmask8 __A,__m128i __B,__m128i __C,__m128i __D)158 _mm_maskz_dpwssd_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
159 {
160 return (__m128i)__builtin_ia32_vpdpwssd_v4si_maskz ((__v4si)__B,
161 (__v4si) __C, (__v4si) __D, (__mmask8)__A);
162 }
163
164 #define _mm256_dpwssds_epi32(A, B, C) \
165 ((__m256i) __builtin_ia32_vpdpwssds_v8si ((__v8si) (A), \
166 (__v8si) (B), \
167 (__v8si) (C)))
168
169 extern __inline __m256i
170 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_dpwssds_epi32(__m256i __A,__mmask8 __B,__m256i __C,__m256i __D)171 _mm256_mask_dpwssds_epi32 (__m256i __A, __mmask8 __B, __m256i __C, __m256i __D)
172 {
173 return (__m256i)__builtin_ia32_vpdpwssds_v8si_mask ((__v8si)__A,
174 (__v8si) __C, (__v8si) __D, (__mmask8)__B);
175 }
176
177 extern __inline __m256i
178 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_dpwssds_epi32(__mmask8 __A,__m256i __B,__m256i __C,__m256i __D)179 _mm256_maskz_dpwssds_epi32 (__mmask8 __A, __m256i __B, __m256i __C,
180 __m256i __D)
181 {
182 return (__m256i)__builtin_ia32_vpdpwssds_v8si_maskz ((__v8si)__B,
183 (__v8si) __C, (__v8si) __D, (__mmask8)__A);
184 }
185
186 #define _mm_dpwssds_epi32(A, B, C) \
187 ((__m128i) __builtin_ia32_vpdpwssds_v4si ((__v4si) (A), \
188 (__v4si) (B), \
189 (__v4si) (C)))
190
191 extern __inline __m128i
192 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_dpwssds_epi32(__m128i __A,__mmask8 __B,__m128i __C,__m128i __D)193 _mm_mask_dpwssds_epi32 (__m128i __A, __mmask8 __B, __m128i __C, __m128i __D)
194 {
195 return (__m128i)__builtin_ia32_vpdpwssds_v4si_mask ((__v4si)__A,
196 (__v4si) __C, (__v4si) __D, (__mmask8)__B);
197 }
198
199 extern __inline __m128i
200 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_dpwssds_epi32(__mmask8 __A,__m128i __B,__m128i __C,__m128i __D)201 _mm_maskz_dpwssds_epi32 (__mmask8 __A, __m128i __B, __m128i __C, __m128i __D)
202 {
203 return (__m128i)__builtin_ia32_vpdpwssds_v4si_maskz ((__v4si)__B,
204 (__v4si) __C, (__v4si) __D, (__mmask8)__A);
205 }
206 #ifdef __DISABLE_AVX512VNNIVL__
207 #undef __DISABLE_AVX512VNNIVL__
208 #pragma GCC pop_options
209 #endif /* __DISABLE_AVX512VNNIVL__ */
210 #endif /* __DISABLE_AVX512VNNIVL__ */
211