avx512pfintrin.h revision 1.7 1 1.7 mrg /* Copyright (C) 2013-2022 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 <avx512pfintrin.h> directly; include <immintrin.h> instead."
26 1.1 mrg #endif
27 1.1 mrg
28 1.1 mrg #ifndef _AVX512PFINTRIN_H_INCLUDED
29 1.1 mrg #define _AVX512PFINTRIN_H_INCLUDED
30 1.1 mrg
31 1.1 mrg #ifndef __AVX512PF__
32 1.1 mrg #pragma GCC push_options
33 1.1 mrg #pragma GCC target("avx512pf")
34 1.1 mrg #define __DISABLE_AVX512PF__
35 1.1 mrg #endif /* __AVX512PF__ */
36 1.1 mrg
37 1.1 mrg /* Internal data types for implementing the intrinsics. */
38 1.1 mrg typedef long long __v8di __attribute__ ((__vector_size__ (64)));
39 1.1 mrg typedef int __v16si __attribute__ ((__vector_size__ (64)));
40 1.1 mrg
41 1.1 mrg /* The Intel API is flexible enough that we must allow aliasing with other
42 1.1 mrg vector types, and their scalar components. */
43 1.1 mrg typedef long long __m512i __attribute__ ((__vector_size__ (64), __may_alias__));
44 1.1 mrg
45 1.1 mrg typedef unsigned char __mmask8;
46 1.1 mrg typedef unsigned short __mmask16;
47 1.1 mrg
48 1.1 mrg #ifdef __OPTIMIZE__
49 1.1 mrg extern __inline void
50 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
51 1.3 mrg _mm512_prefetch_i32gather_pd (__m256i __index, void const *__addr,
52 1.3 mrg int __scale, int __hint)
53 1.1 mrg {
54 1.3 mrg __builtin_ia32_gatherpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
55 1.3 mrg __scale, __hint);
56 1.1 mrg }
57 1.1 mrg
58 1.1 mrg extern __inline void
59 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
60 1.3 mrg _mm512_prefetch_i32gather_ps (__m512i __index, void const *__addr,
61 1.3 mrg int __scale, int __hint)
62 1.1 mrg {
63 1.3 mrg __builtin_ia32_gatherpfdps ((__mmask16) 0xFFFF, (__v16si) __index, __addr,
64 1.3 mrg __scale, __hint);
65 1.1 mrg }
66 1.1 mrg
67 1.1 mrg extern __inline void
68 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
69 1.3 mrg _mm512_mask_prefetch_i32gather_pd (__m256i __index, __mmask8 __mask,
70 1.3 mrg void const *__addr, int __scale, int __hint)
71 1.1 mrg {
72 1.3 mrg __builtin_ia32_gatherpfdpd (__mask, (__v8si) __index, __addr, __scale,
73 1.3 mrg __hint);
74 1.1 mrg }
75 1.1 mrg
76 1.1 mrg extern __inline void
77 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
78 1.3 mrg _mm512_mask_prefetch_i32gather_ps (__m512i __index, __mmask16 __mask,
79 1.3 mrg void const *__addr, int __scale, int __hint)
80 1.1 mrg {
81 1.3 mrg __builtin_ia32_gatherpfdps (__mask, (__v16si) __index, __addr, __scale,
82 1.3 mrg __hint);
83 1.1 mrg }
84 1.1 mrg
85 1.1 mrg extern __inline void
86 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
87 1.3 mrg _mm512_prefetch_i64gather_pd (__m512i __index, void const *__addr,
88 1.3 mrg int __scale, int __hint)
89 1.1 mrg {
90 1.3 mrg __builtin_ia32_gatherpfqpd ((__mmask8) 0xFF, (__v8di) __index, __addr,
91 1.3 mrg __scale, __hint);
92 1.1 mrg }
93 1.1 mrg
94 1.1 mrg extern __inline void
95 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
96 1.3 mrg _mm512_prefetch_i64gather_ps (__m512i __index, void const *__addr,
97 1.3 mrg int __scale, int __hint)
98 1.1 mrg {
99 1.3 mrg __builtin_ia32_gatherpfqps ((__mmask8) 0xFF, (__v8di) __index, __addr,
100 1.3 mrg __scale, __hint);
101 1.1 mrg }
102 1.1 mrg
103 1.1 mrg extern __inline void
104 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
105 1.3 mrg _mm512_mask_prefetch_i64gather_pd (__m512i __index, __mmask8 __mask,
106 1.3 mrg void const *__addr, int __scale, int __hint)
107 1.1 mrg {
108 1.3 mrg __builtin_ia32_gatherpfqpd (__mask, (__v8di) __index, __addr, __scale,
109 1.3 mrg __hint);
110 1.1 mrg }
111 1.1 mrg
112 1.1 mrg extern __inline void
113 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
114 1.3 mrg _mm512_mask_prefetch_i64gather_ps (__m512i __index, __mmask8 __mask,
115 1.3 mrg void const *__addr, int __scale, int __hint)
116 1.1 mrg {
117 1.3 mrg __builtin_ia32_gatherpfqps (__mask, (__v8di) __index, __addr, __scale,
118 1.3 mrg __hint);
119 1.1 mrg }
120 1.1 mrg
121 1.1 mrg extern __inline void
122 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
123 1.3 mrg _mm512_prefetch_i32scatter_pd (void *__addr, __m256i __index, int __scale,
124 1.3 mrg int __hint)
125 1.1 mrg {
126 1.3 mrg __builtin_ia32_scatterpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
127 1.3 mrg __scale, __hint);
128 1.1 mrg }
129 1.1 mrg
130 1.1 mrg extern __inline void
131 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
132 1.3 mrg _mm512_prefetch_i32scatter_ps (void *__addr, __m512i __index, int __scale,
133 1.3 mrg int __hint)
134 1.1 mrg {
135 1.3 mrg __builtin_ia32_scatterpfdps ((__mmask16) 0xFFFF, (__v16si) __index, __addr,
136 1.3 mrg __scale, __hint);
137 1.1 mrg }
138 1.1 mrg
139 1.1 mrg extern __inline void
140 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
141 1.3 mrg _mm512_mask_prefetch_i32scatter_pd (void *__addr, __mmask8 __mask,
142 1.3 mrg __m256i __index, int __scale, int __hint)
143 1.1 mrg {
144 1.3 mrg __builtin_ia32_scatterpfdpd (__mask, (__v8si) __index, __addr, __scale,
145 1.3 mrg __hint);
146 1.1 mrg }
147 1.1 mrg
148 1.1 mrg extern __inline void
149 1.1 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
150 1.3 mrg _mm512_mask_prefetch_i32scatter_ps (void *__addr, __mmask16 __mask,
151 1.3 mrg __m512i __index, int __scale, int __hint)
152 1.1 mrg {
153 1.3 mrg __builtin_ia32_scatterpfdps (__mask, (__v16si) __index, __addr, __scale,
154 1.3 mrg __hint);
155 1.3 mrg }
156 1.3 mrg
157 1.3 mrg extern __inline void
158 1.3 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
159 1.3 mrg _mm512_prefetch_i64scatter_pd (void *__addr, __m512i __index, int __scale,
160 1.3 mrg int __hint)
161 1.3 mrg {
162 1.3 mrg __builtin_ia32_scatterpfqpd ((__mmask8) 0xFF, (__v8di) __index,__addr,
163 1.3 mrg __scale, __hint);
164 1.3 mrg }
165 1.3 mrg
166 1.3 mrg extern __inline void
167 1.3 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
168 1.3 mrg _mm512_prefetch_i64scatter_ps (void *__addr, __m512i __index, int __scale,
169 1.3 mrg int __hint)
170 1.3 mrg {
171 1.3 mrg __builtin_ia32_scatterpfqps ((__mmask8) 0xFF, (__v8di) __index, __addr,
172 1.3 mrg __scale, __hint);
173 1.3 mrg }
174 1.3 mrg
175 1.3 mrg extern __inline void
176 1.3 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
177 1.5 mrg _mm512_mask_prefetch_i64scatter_pd (void *__addr, __mmask8 __mask,
178 1.3 mrg __m512i __index, int __scale, int __hint)
179 1.3 mrg {
180 1.3 mrg __builtin_ia32_scatterpfqpd (__mask, (__v8di) __index, __addr, __scale,
181 1.5 mrg __hint);
182 1.3 mrg }
183 1.3 mrg
184 1.3 mrg extern __inline void
185 1.3 mrg __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
186 1.5 mrg _mm512_mask_prefetch_i64scatter_ps (void *__addr, __mmask8 __mask,
187 1.3 mrg __m512i __index, int __scale, int __hint)
188 1.3 mrg {
189 1.3 mrg __builtin_ia32_scatterpfqps (__mask, (__v8di) __index, __addr, __scale,
190 1.3 mrg __hint);
191 1.1 mrg }
192 1.1 mrg
193 1.1 mrg #else
194 1.3 mrg #define _mm512_prefetch_i32gather_pd(INDEX, ADDR, SCALE, HINT) \
195 1.6 mrg __builtin_ia32_gatherpfdpd ((__mmask8)0xFF, (__v8si)(__m256i) (INDEX), \
196 1.6 mrg (void const *) (ADDR), (int) (SCALE), \
197 1.6 mrg (int) (HINT))
198 1.3 mrg
199 1.3 mrg #define _mm512_prefetch_i32gather_ps(INDEX, ADDR, SCALE, HINT) \
200 1.6 mrg __builtin_ia32_gatherpfdps ((__mmask16)0xFFFF, (__v16si)(__m512i) (INDEX), \
201 1.6 mrg (void const *) (ADDR), (int) (SCALE), \
202 1.6 mrg (int) (HINT))
203 1.3 mrg
204 1.1 mrg #define _mm512_mask_prefetch_i32gather_pd(INDEX, MASK, ADDR, SCALE, HINT) \
205 1.6 mrg __builtin_ia32_gatherpfdpd ((__mmask8) (MASK), (__v8si)(__m256i) (INDEX), \
206 1.6 mrg (void const *) (ADDR), (int) (SCALE), \
207 1.6 mrg (int) (HINT))
208 1.1 mrg
209 1.1 mrg #define _mm512_mask_prefetch_i32gather_ps(INDEX, MASK, ADDR, SCALE, HINT) \
210 1.6 mrg __builtin_ia32_gatherpfdps ((__mmask16) (MASK), (__v16si)(__m512i) (INDEX),\
211 1.6 mrg (void const *) (ADDR), (int) (SCALE), \
212 1.6 mrg (int) (HINT))
213 1.3 mrg
214 1.3 mrg #define _mm512_prefetch_i64gather_pd(INDEX, ADDR, SCALE, HINT) \
215 1.6 mrg __builtin_ia32_gatherpfqpd ((__mmask8)0xFF, (__v8di)(__m512i) (INDEX), \
216 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
217 1.3 mrg
218 1.3 mrg #define _mm512_prefetch_i64gather_ps(INDEX, ADDR, SCALE, HINT) \
219 1.6 mrg __builtin_ia32_gatherpfqps ((__mmask8)0xFF, (__v8di)(__m512i) (INDEX), \
220 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
221 1.1 mrg
222 1.1 mrg #define _mm512_mask_prefetch_i64gather_pd(INDEX, MASK, ADDR, SCALE, HINT) \
223 1.6 mrg __builtin_ia32_gatherpfqpd ((__mmask8) (MASK), (__v8di)(__m512i) (INDEX), \
224 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
225 1.1 mrg
226 1.1 mrg #define _mm512_mask_prefetch_i64gather_ps(INDEX, MASK, ADDR, SCALE, HINT) \
227 1.6 mrg __builtin_ia32_gatherpfqps ((__mmask8) (MASK), (__v8di)(__m512i) (INDEX), \
228 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
229 1.1 mrg
230 1.1 mrg #define _mm512_prefetch_i32scatter_pd(ADDR, INDEX, SCALE, HINT) \
231 1.6 mrg __builtin_ia32_scatterpfdpd ((__mmask8)0xFF, (__v8si)(__m256i) (INDEX), \
232 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
233 1.1 mrg
234 1.1 mrg #define _mm512_prefetch_i32scatter_ps(ADDR, INDEX, SCALE, HINT) \
235 1.6 mrg __builtin_ia32_scatterpfdps ((__mmask16)0xFFFF, (__v16si)(__m512i) (INDEX),\
236 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
237 1.1 mrg
238 1.1 mrg #define _mm512_mask_prefetch_i32scatter_pd(ADDR, MASK, INDEX, SCALE, HINT) \
239 1.6 mrg __builtin_ia32_scatterpfdpd ((__mmask8) (MASK), (__v8si)(__m256i) (INDEX), \
240 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
241 1.1 mrg
242 1.1 mrg #define _mm512_mask_prefetch_i32scatter_ps(ADDR, MASK, INDEX, SCALE, HINT) \
243 1.6 mrg __builtin_ia32_scatterpfdps ((__mmask16) (MASK), \
244 1.6 mrg (__v16si)(__m512i) (INDEX), \
245 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
246 1.1 mrg
247 1.1 mrg #define _mm512_prefetch_i64scatter_pd(ADDR, INDEX, SCALE, HINT) \
248 1.6 mrg __builtin_ia32_scatterpfqpd ((__mmask8)0xFF, (__v8di)(__m512i) (INDEX), \
249 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
250 1.1 mrg
251 1.1 mrg #define _mm512_prefetch_i64scatter_ps(ADDR, INDEX, SCALE, HINT) \
252 1.6 mrg __builtin_ia32_scatterpfqps ((__mmask8)0xFF, (__v8di)(__m512i) (INDEX), \
253 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
254 1.1 mrg
255 1.1 mrg #define _mm512_mask_prefetch_i64scatter_pd(ADDR, MASK, INDEX, SCALE, HINT) \
256 1.6 mrg __builtin_ia32_scatterpfqpd ((__mmask8) (MASK), (__v8di)(__m512i) (INDEX), \
257 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
258 1.1 mrg
259 1.1 mrg #define _mm512_mask_prefetch_i64scatter_ps(ADDR, MASK, INDEX, SCALE, HINT) \
260 1.6 mrg __builtin_ia32_scatterpfqps ((__mmask8) (MASK), (__v8di)(__m512i) (INDEX), \
261 1.6 mrg (void *) (ADDR), (int) (SCALE), (int) (HINT))
262 1.1 mrg #endif
263 1.1 mrg
264 1.1 mrg #ifdef __DISABLE_AVX512PF__
265 1.1 mrg #undef __DISABLE_AVX512PF__
266 1.1 mrg #pragma GCC pop_options
267 1.1 mrg #endif /* __DISABLE_AVX512PF__ */
268 1.1 mrg
269 1.1 mrg #endif /* _AVX512PFINTRIN_H_INCLUDED */
270