[1046] | 1 | /* Copyright (C) 2013-2015 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_mask_prefetch_i32gather_pd (__m256i index, __mmask8 mask,
|
---|
| 52 | void *addr, int scale, int hint)
|
---|
| 53 | {
|
---|
| 54 | __builtin_ia32_gatherpfdpd (mask, (__v8si) index, (long long const *) addr,
|
---|
| 55 | scale, hint);
|
---|
| 56 | }
|
---|
| 57 |
|
---|
| 58 | extern __inline void
|
---|
| 59 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 60 | _mm512_mask_prefetch_i32gather_ps (__m512i index, __mmask16 mask,
|
---|
| 61 | void *addr, int scale, int hint)
|
---|
| 62 | {
|
---|
| 63 | __builtin_ia32_gatherpfdps (mask, (__v16si) index, (int const *) addr,
|
---|
| 64 | scale, hint);
|
---|
| 65 | }
|
---|
| 66 |
|
---|
| 67 | extern __inline void
|
---|
| 68 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 69 | _mm512_mask_prefetch_i64gather_pd (__m512i index, __mmask8 mask,
|
---|
| 70 | void *addr, int scale, int hint)
|
---|
| 71 | {
|
---|
| 72 | __builtin_ia32_gatherpfqpd (mask, (__v8di) index, (long long const *) addr,
|
---|
| 73 | scale, hint);
|
---|
| 74 | }
|
---|
| 75 |
|
---|
| 76 | extern __inline void
|
---|
| 77 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 78 | _mm512_mask_prefetch_i64gather_ps (__m512i index, __mmask8 mask,
|
---|
| 79 | void *addr, int scale, int hint)
|
---|
| 80 | {
|
---|
| 81 | __builtin_ia32_gatherpfqps (mask, (__v8di) index, (int const *) addr,
|
---|
| 82 | scale, hint);
|
---|
| 83 | }
|
---|
| 84 |
|
---|
| 85 | extern __inline void
|
---|
| 86 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 87 | _mm512_prefetch_i32scatter_pd (void *addr, __m256i index, int scale,
|
---|
| 88 | int hint)
|
---|
| 89 | {
|
---|
| 90 | __builtin_ia32_scatterpfdpd ((__mmask8) 0xFF, (__v8si) index,
|
---|
| 91 | (long long const *)addr, scale, hint);
|
---|
| 92 | }
|
---|
| 93 |
|
---|
| 94 | extern __inline void
|
---|
| 95 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 96 | _mm512_prefetch_i32scatter_ps (void *addr, __m512i index, int scale,
|
---|
| 97 | int hint)
|
---|
| 98 | {
|
---|
| 99 | __builtin_ia32_scatterpfdps ((__mmask16) 0xFFFF, (__v16si) index, (int const *) addr,
|
---|
| 100 | scale, hint);
|
---|
| 101 | }
|
---|
| 102 |
|
---|
| 103 | extern __inline void
|
---|
| 104 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 105 | _mm512_mask_prefetch_i32scatter_pd (void *addr, __mmask8 mask,
|
---|
| 106 | __m256i index, int scale, int hint)
|
---|
| 107 | {
|
---|
| 108 | __builtin_ia32_scatterpfdpd (mask, (__v8si) index, (long long const *) addr,
|
---|
| 109 | scale, hint);
|
---|
| 110 | }
|
---|
| 111 |
|
---|
| 112 | extern __inline void
|
---|
| 113 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 114 | _mm512_mask_prefetch_i32scatter_ps (void *addr, __mmask16 mask,
|
---|
| 115 | __m512i index, int scale, int hint)
|
---|
| 116 | {
|
---|
| 117 | __builtin_ia32_scatterpfdps (mask, (__v16si) index, (int const *) addr,
|
---|
| 118 | scale, hint);
|
---|
| 119 | }
|
---|
| 120 |
|
---|
| 121 | extern __inline void
|
---|
| 122 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 123 | _mm512_prefetch_i64scatter_pd (void *addr, __m512i index, int scale,
|
---|
| 124 | int hint)
|
---|
| 125 | {
|
---|
| 126 | __builtin_ia32_scatterpfqpd ((__mmask8) 0xFF, (__v8di) index, (long long const *) addr,
|
---|
| 127 | scale, hint);
|
---|
| 128 | }
|
---|
| 129 |
|
---|
| 130 | extern __inline void
|
---|
| 131 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 132 | _mm512_prefetch_i64scatter_ps (void *addr, __m512i index, int scale,
|
---|
| 133 | int hint)
|
---|
| 134 | {
|
---|
| 135 | __builtin_ia32_scatterpfqps ((__mmask8) 0xFF, (__v8di) index, (int const *) addr,
|
---|
| 136 | scale, hint);
|
---|
| 137 | }
|
---|
| 138 |
|
---|
| 139 | extern __inline void
|
---|
| 140 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 141 | _mm512_mask_prefetch_i64scatter_pd (void *addr, __mmask16 mask,
|
---|
| 142 | __m512i index, int scale, int hint)
|
---|
| 143 | {
|
---|
| 144 | __builtin_ia32_scatterpfqpd (mask, (__v8di) index, (long long const *) addr,
|
---|
| 145 | scale, hint);
|
---|
| 146 | }
|
---|
| 147 |
|
---|
| 148 | extern __inline void
|
---|
| 149 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 150 | _mm512_mask_prefetch_i64scatter_ps (void *addr, __mmask16 mask,
|
---|
| 151 | __m512i index, int scale, int hint)
|
---|
| 152 | {
|
---|
| 153 | __builtin_ia32_scatterpfqps (mask, (__v8di) index, (int const *) addr,
|
---|
| 154 | scale, hint);
|
---|
| 155 | }
|
---|
| 156 |
|
---|
| 157 | #else
|
---|
| 158 | #define _mm512_mask_prefetch_i32gather_pd(INDEX, MASK, ADDR, SCALE, HINT) \
|
---|
| 159 | __builtin_ia32_gatherpfdpd ((__mmask8)MASK, (__v8si)(__m256i)INDEX, \
|
---|
| 160 | (long long const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 161 |
|
---|
| 162 | #define _mm512_mask_prefetch_i32gather_ps(INDEX, MASK, ADDR, SCALE, HINT) \
|
---|
| 163 | __builtin_ia32_gatherpfdps ((__mmask16)MASK, (__v16si)(__m512i)INDEX, \
|
---|
| 164 | (int const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 165 |
|
---|
| 166 | #define _mm512_mask_prefetch_i64gather_pd(INDEX, MASK, ADDR, SCALE, HINT) \
|
---|
| 167 | __builtin_ia32_gatherpfqpd ((__mmask8)MASK, (__v8di)(__m512i)INDEX, \
|
---|
| 168 | (long long const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 169 |
|
---|
| 170 | #define _mm512_mask_prefetch_i64gather_ps(INDEX, MASK, ADDR, SCALE, HINT) \
|
---|
| 171 | __builtin_ia32_gatherpfqps ((__mmask8)MASK, (__v8di)(__m512i)INDEX, \
|
---|
| 172 | (int const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 173 |
|
---|
| 174 | #define _mm512_prefetch_i32scatter_pd(ADDR, INDEX, SCALE, HINT) \
|
---|
| 175 | __builtin_ia32_scatterpfdpd ((__mmask8)0xFF, (__v8si)(__m256i)INDEX, \
|
---|
| 176 | (long long const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 177 |
|
---|
| 178 | #define _mm512_prefetch_i32scatter_ps(ADDR, INDEX, SCALE, HINT) \
|
---|
| 179 | __builtin_ia32_scatterpfdps ((__mmask16)0xFFFF, (__v16si)(__m512i)INDEX, \
|
---|
| 180 | (int const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 181 |
|
---|
| 182 | #define _mm512_mask_prefetch_i32scatter_pd(ADDR, MASK, INDEX, SCALE, HINT) \
|
---|
| 183 | __builtin_ia32_scatterpfdpd ((__mmask8)MASK, (__v8si)(__m256i)INDEX, \
|
---|
| 184 | (long long const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 185 |
|
---|
| 186 | #define _mm512_mask_prefetch_i32scatter_ps(ADDR, MASK, INDEX, SCALE, HINT) \
|
---|
| 187 | __builtin_ia32_scatterpfdps ((__mmask16)MASK, (__v16si)(__m512i)INDEX, \
|
---|
| 188 | (int const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 189 |
|
---|
| 190 | #define _mm512_prefetch_i64scatter_pd(ADDR, INDEX, SCALE, HINT) \
|
---|
| 191 | __builtin_ia32_scatterpfqpd ((__mmask8)0xFF, (__v8di)(__m512i)INDEX, \
|
---|
| 192 | (long long const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 193 |
|
---|
| 194 | #define _mm512_prefetch_i64scatter_ps(ADDR, INDEX, SCALE, HINT) \
|
---|
| 195 | __builtin_ia32_scatterpfqps ((__mmask8)0xFF, (__v8di)(__m512i)INDEX, \
|
---|
| 196 | (int const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 197 |
|
---|
| 198 | #define _mm512_mask_prefetch_i64scatter_pd(ADDR, MASK, INDEX, SCALE, HINT) \
|
---|
| 199 | __builtin_ia32_scatterpfqpd ((__mmask8)MASK, (__v8di)(__m512i)INDEX, \
|
---|
| 200 | (long long const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 201 |
|
---|
| 202 | #define _mm512_mask_prefetch_i64scatter_ps(ADDR, MASK, INDEX, SCALE, HINT) \
|
---|
| 203 | __builtin_ia32_scatterpfqps ((__mmask8)MASK, (__v8di)(__m512i)INDEX, \
|
---|
| 204 | (int const *)ADDR, (int)SCALE, (int)HINT)
|
---|
| 205 | #endif
|
---|
| 206 |
|
---|
| 207 | #ifdef __DISABLE_AVX512PF__
|
---|
| 208 | #undef __DISABLE_AVX512PF__
|
---|
| 209 | #pragma GCC pop_options
|
---|
| 210 | #endif /* __DISABLE_AVX512PF__ */
|
---|
| 211 |
|
---|
| 212 | #endif /* _AVX512PFINTRIN_H_INCLUDED */
|
---|