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