[1166] | 1 | /* Copyright (C) 2017-2021 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 | #if !defined _IMMINTRIN_H_INCLUDED
|
---|
| 25 | # error "Never use <avx512bitalgintrin.h> directly; include <x86intrin.h> instead."
|
---|
| 26 | #endif
|
---|
| 27 |
|
---|
| 28 | #ifndef _AVX512BITALGINTRIN_H_INCLUDED
|
---|
| 29 | #define _AVX512BITALGINTRIN_H_INCLUDED
|
---|
| 30 |
|
---|
| 31 | #ifndef __AVX512BITALG__
|
---|
| 32 | #pragma GCC push_options
|
---|
| 33 | #pragma GCC target("avx512bitalg")
|
---|
| 34 | #define __DISABLE_AVX512BITALG__
|
---|
| 35 | #endif /* __AVX512BITALG__ */
|
---|
| 36 |
|
---|
| 37 | extern __inline __m512i
|
---|
| 38 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 39 | _mm512_popcnt_epi8 (__m512i __A)
|
---|
| 40 | {
|
---|
| 41 | return (__m512i) __builtin_ia32_vpopcountb_v64qi ((__v64qi) __A);
|
---|
| 42 | }
|
---|
| 43 |
|
---|
| 44 | extern __inline __m512i
|
---|
| 45 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 46 | _mm512_popcnt_epi16 (__m512i __A)
|
---|
| 47 | {
|
---|
| 48 | return (__m512i) __builtin_ia32_vpopcountw_v32hi ((__v32hi) __A);
|
---|
| 49 | }
|
---|
| 50 |
|
---|
| 51 | #ifdef __DISABLE_AVX512BITALG__
|
---|
| 52 | #undef __DISABLE_AVX512BITALG__
|
---|
| 53 | #pragma GCC pop_options
|
---|
| 54 | #endif /* __DISABLE_AVX512BITALG__ */
|
---|
| 55 |
|
---|
| 56 | #if !defined(__AVX512BITALG__) || !defined(__AVX512BW__)
|
---|
| 57 | #pragma GCC push_options
|
---|
| 58 | #pragma GCC target("avx512bitalg,avx512bw")
|
---|
| 59 | #define __DISABLE_AVX512BITALGBW__
|
---|
| 60 | #endif /* __AVX512VLBW__ */
|
---|
| 61 |
|
---|
| 62 | extern __inline __m512i
|
---|
| 63 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 64 | _mm512_mask_popcnt_epi8 (__m512i __W, __mmask64 __U, __m512i __A)
|
---|
| 65 | {
|
---|
| 66 | return (__m512i) __builtin_ia32_vpopcountb_v64qi_mask ((__v64qi) __A,
|
---|
| 67 | (__v64qi) __W,
|
---|
| 68 | (__mmask64) __U);
|
---|
| 69 | }
|
---|
| 70 |
|
---|
| 71 | extern __inline __m512i
|
---|
| 72 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 73 | _mm512_maskz_popcnt_epi8 (__mmask64 __U, __m512i __A)
|
---|
| 74 | {
|
---|
| 75 | return (__m512i) __builtin_ia32_vpopcountb_v64qi_mask ((__v64qi) __A,
|
---|
| 76 | (__v64qi)
|
---|
| 77 | _mm512_setzero_si512 (),
|
---|
| 78 | (__mmask64) __U);
|
---|
| 79 | }
|
---|
| 80 | extern __inline __m512i
|
---|
| 81 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 82 | _mm512_mask_popcnt_epi16 (__m512i __W, __mmask32 __U, __m512i __A)
|
---|
| 83 | {
|
---|
| 84 | return (__m512i) __builtin_ia32_vpopcountw_v32hi_mask ((__v32hi) __A,
|
---|
| 85 | (__v32hi) __W,
|
---|
| 86 | (__mmask32) __U);
|
---|
| 87 | }
|
---|
| 88 |
|
---|
| 89 | extern __inline __m512i
|
---|
| 90 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 91 | _mm512_maskz_popcnt_epi16 (__mmask32 __U, __m512i __A)
|
---|
| 92 | {
|
---|
| 93 | return (__m512i) __builtin_ia32_vpopcountw_v32hi_mask ((__v32hi) __A,
|
---|
| 94 | (__v32hi)
|
---|
| 95 | _mm512_setzero_si512 (),
|
---|
| 96 | (__mmask32) __U);
|
---|
| 97 | }
|
---|
| 98 |
|
---|
| 99 | extern __inline __mmask64
|
---|
| 100 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 101 | _mm512_bitshuffle_epi64_mask (__m512i __A, __m512i __B)
|
---|
| 102 | {
|
---|
| 103 | return (__mmask64) __builtin_ia32_vpshufbitqmb512_mask ((__v64qi) __A,
|
---|
| 104 | (__v64qi) __B,
|
---|
| 105 | (__mmask64) -1);
|
---|
| 106 | }
|
---|
| 107 |
|
---|
| 108 | extern __inline __mmask64
|
---|
| 109 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 110 | _mm512_mask_bitshuffle_epi64_mask (__mmask64 __M, __m512i __A, __m512i __B)
|
---|
| 111 | {
|
---|
| 112 | return (__mmask64) __builtin_ia32_vpshufbitqmb512_mask ((__v64qi) __A,
|
---|
| 113 | (__v64qi) __B,
|
---|
| 114 | (__mmask64) __M);
|
---|
| 115 | }
|
---|
| 116 |
|
---|
| 117 | #ifdef __DISABLE_AVX512BITALGBW__
|
---|
| 118 | #undef __DISABLE_AVX512BITALGBW__
|
---|
| 119 | #pragma GCC pop_options
|
---|
| 120 | #endif /* __DISABLE_AVX512BITALGBW__ */
|
---|
| 121 |
|
---|
| 122 | #if !defined(__AVX512BITALG__) || !defined(__AVX512VL__) || !defined(__AVX512BW__)
|
---|
| 123 | #pragma GCC push_options
|
---|
| 124 | #pragma GCC target("avx512bitalg,avx512vl,avx512bw")
|
---|
| 125 | #define __DISABLE_AVX512BITALGVLBW__
|
---|
| 126 | #endif /* __AVX512VLBW__ */
|
---|
| 127 |
|
---|
| 128 | extern __inline __m256i
|
---|
| 129 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 130 | _mm256_mask_popcnt_epi8 (__m256i __W, __mmask32 __U, __m256i __A)
|
---|
| 131 | {
|
---|
| 132 | return (__m256i) __builtin_ia32_vpopcountb_v32qi_mask ((__v32qi) __A,
|
---|
| 133 | (__v32qi) __W,
|
---|
| 134 | (__mmask32) __U);
|
---|
| 135 | }
|
---|
| 136 |
|
---|
| 137 | extern __inline __m256i
|
---|
| 138 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 139 | _mm256_maskz_popcnt_epi8 (__mmask32 __U, __m256i __A)
|
---|
| 140 | {
|
---|
| 141 | return (__m256i) __builtin_ia32_vpopcountb_v32qi_mask ((__v32qi) __A,
|
---|
| 142 | (__v32qi)
|
---|
| 143 | _mm256_setzero_si256 (),
|
---|
| 144 | (__mmask32) __U);
|
---|
| 145 | }
|
---|
| 146 |
|
---|
| 147 | extern __inline __mmask32
|
---|
| 148 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 149 | _mm256_bitshuffle_epi64_mask (__m256i __A, __m256i __B)
|
---|
| 150 | {
|
---|
| 151 | return (__mmask32) __builtin_ia32_vpshufbitqmb256_mask ((__v32qi) __A,
|
---|
| 152 | (__v32qi) __B,
|
---|
| 153 | (__mmask32) -1);
|
---|
| 154 | }
|
---|
| 155 |
|
---|
| 156 | extern __inline __mmask32
|
---|
| 157 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 158 | _mm256_mask_bitshuffle_epi64_mask (__mmask32 __M, __m256i __A, __m256i __B)
|
---|
| 159 | {
|
---|
| 160 | return (__mmask32) __builtin_ia32_vpshufbitqmb256_mask ((__v32qi) __A,
|
---|
| 161 | (__v32qi) __B,
|
---|
| 162 | (__mmask32) __M);
|
---|
| 163 | }
|
---|
| 164 |
|
---|
| 165 | #ifdef __DISABLE_AVX512BITALGVLBW__
|
---|
| 166 | #undef __DISABLE_AVX512BITALGVLBW__
|
---|
| 167 | #pragma GCC pop_options
|
---|
| 168 | #endif /* __DISABLE_AVX512BITALGVLBW__ */
|
---|
| 169 |
|
---|
| 170 |
|
---|
| 171 | #if !defined(__AVX512BITALG__) || !defined(__AVX512VL__)
|
---|
| 172 | #pragma GCC push_options
|
---|
| 173 | #pragma GCC target("avx512bitalg,avx512vl")
|
---|
| 174 | #define __DISABLE_AVX512BITALGVL__
|
---|
| 175 | #endif /* __AVX512VLBW__ */
|
---|
| 176 |
|
---|
| 177 | extern __inline __mmask16
|
---|
| 178 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 179 | _mm_bitshuffle_epi64_mask (__m128i __A, __m128i __B)
|
---|
| 180 | {
|
---|
| 181 | return (__mmask16) __builtin_ia32_vpshufbitqmb128_mask ((__v16qi) __A,
|
---|
| 182 | (__v16qi) __B,
|
---|
| 183 | (__mmask16) -1);
|
---|
| 184 | }
|
---|
| 185 |
|
---|
| 186 | extern __inline __mmask16
|
---|
| 187 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 188 | _mm_mask_bitshuffle_epi64_mask (__mmask16 __M, __m128i __A, __m128i __B)
|
---|
| 189 | {
|
---|
| 190 | return (__mmask16) __builtin_ia32_vpshufbitqmb128_mask ((__v16qi) __A,
|
---|
| 191 | (__v16qi) __B,
|
---|
| 192 | (__mmask16) __M);
|
---|
| 193 | }
|
---|
| 194 |
|
---|
| 195 | extern __inline __m256i
|
---|
| 196 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 197 | _mm256_popcnt_epi8 (__m256i __A)
|
---|
| 198 | {
|
---|
| 199 | return (__m256i) __builtin_ia32_vpopcountb_v32qi ((__v32qi) __A);
|
---|
| 200 | }
|
---|
| 201 |
|
---|
| 202 | extern __inline __m256i
|
---|
| 203 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 204 | _mm256_popcnt_epi16 (__m256i __A)
|
---|
| 205 | {
|
---|
| 206 | return (__m256i) __builtin_ia32_vpopcountw_v16hi ((__v16hi) __A);
|
---|
| 207 | }
|
---|
| 208 |
|
---|
| 209 | extern __inline __m128i
|
---|
| 210 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 211 | _mm_popcnt_epi8 (__m128i __A)
|
---|
| 212 | {
|
---|
| 213 | return (__m128i) __builtin_ia32_vpopcountb_v16qi ((__v16qi) __A);
|
---|
| 214 | }
|
---|
| 215 |
|
---|
| 216 | extern __inline __m128i
|
---|
| 217 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 218 | _mm_popcnt_epi16 (__m128i __A)
|
---|
| 219 | {
|
---|
| 220 | return (__m128i) __builtin_ia32_vpopcountw_v8hi ((__v8hi) __A);
|
---|
| 221 | }
|
---|
| 222 |
|
---|
| 223 | extern __inline __m256i
|
---|
| 224 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 225 | _mm256_mask_popcnt_epi16 (__m256i __W, __mmask16 __U, __m256i __A)
|
---|
| 226 | {
|
---|
| 227 | return (__m256i) __builtin_ia32_vpopcountw_v16hi_mask ((__v16hi) __A,
|
---|
| 228 | (__v16hi) __W,
|
---|
| 229 | (__mmask16) __U);
|
---|
| 230 | }
|
---|
| 231 |
|
---|
| 232 | extern __inline __m256i
|
---|
| 233 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 234 | _mm256_maskz_popcnt_epi16 (__mmask16 __U, __m256i __A)
|
---|
| 235 | {
|
---|
| 236 | return (__m256i) __builtin_ia32_vpopcountw_v16hi_mask ((__v16hi) __A,
|
---|
| 237 | (__v16hi)
|
---|
| 238 | _mm256_setzero_si256 (),
|
---|
| 239 | (__mmask16) __U);
|
---|
| 240 | }
|
---|
| 241 |
|
---|
| 242 | extern __inline __m128i
|
---|
| 243 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 244 | _mm_mask_popcnt_epi8 (__m128i __W, __mmask16 __U, __m128i __A)
|
---|
| 245 | {
|
---|
| 246 | return (__m128i) __builtin_ia32_vpopcountb_v16qi_mask ((__v16qi) __A,
|
---|
| 247 | (__v16qi) __W,
|
---|
| 248 | (__mmask16) __U);
|
---|
| 249 | }
|
---|
| 250 |
|
---|
| 251 | extern __inline __m128i
|
---|
| 252 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 253 | _mm_maskz_popcnt_epi8 (__mmask16 __U, __m128i __A)
|
---|
| 254 | {
|
---|
| 255 | return (__m128i) __builtin_ia32_vpopcountb_v16qi_mask ((__v16qi) __A,
|
---|
| 256 | (__v16qi)
|
---|
| 257 | _mm_setzero_si128 (),
|
---|
| 258 | (__mmask16) __U);
|
---|
| 259 | }
|
---|
| 260 | extern __inline __m128i
|
---|
| 261 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 262 | _mm_mask_popcnt_epi16 (__m128i __W, __mmask8 __U, __m128i __A)
|
---|
| 263 | {
|
---|
| 264 | return (__m128i) __builtin_ia32_vpopcountw_v8hi_mask ((__v8hi) __A,
|
---|
| 265 | (__v8hi) __W,
|
---|
| 266 | (__mmask8) __U);
|
---|
| 267 | }
|
---|
| 268 |
|
---|
| 269 | extern __inline __m128i
|
---|
| 270 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
---|
| 271 | _mm_maskz_popcnt_epi16 (__mmask8 __U, __m128i __A)
|
---|
| 272 | {
|
---|
| 273 | return (__m128i) __builtin_ia32_vpopcountw_v8hi_mask ((__v8hi) __A,
|
---|
| 274 | (__v8hi)
|
---|
| 275 | _mm_setzero_si128 (),
|
---|
| 276 | (__mmask8) __U);
|
---|
| 277 | }
|
---|
| 278 | #ifdef __DISABLE_AVX512BITALGVL__
|
---|
| 279 | #undef __DISABLE_AVX512BITALGVL__
|
---|
| 280 | #pragma GCC pop_options
|
---|
| 281 | #endif /* __DISABLE_AVX512BITALGBW__ */
|
---|
| 282 |
|
---|
| 283 | #endif /* _AVX512BITALGINTRIN_H_INCLUDED */
|
---|