gcc/config/i386/avx512bitalgvlintrin.h

   1 /* Copyright (C) 2023 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 <avx512bitalgvlintrin.h> directly; include <immintrin.h> instead."
  26 #endif
  27
  28 #ifndef _AVX512BITALGVLINTRIN_H_INCLUDED
  29 #define _AVX512BITALGVLINTRIN_H_INCLUDED
  30
  31 #if !defined(__AVX512BITALG__) || !defined(__AVX512VL__)
  32 #pragma GCC push_options
  33 #pragma GCC target("avx512bitalg,avx512vl")
  34 #define __DISABLE_AVX512BITALGVL__
  35 #endif /* __AVX512BITALGVL__ */
  36
  37 extern __inline __m256i
  38 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  39 _mm256_mask_popcnt_epi8 (__m256i __W, __mmask32 __U, __m256i __A)
  40 {
  41   return (__m256i) __builtin_ia32_vpopcountb_v32qi_mask ((__v32qi) __A,
  42                                                          (__v32qi) __W,
  43                                                          (__mmask32) __U);
  44 }
  45
  46 extern __inline __m256i
  47 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  48 _mm256_maskz_popcnt_epi8 (__mmask32 __U, __m256i __A)
  49 {
  50   return (__m256i) __builtin_ia32_vpopcountb_v32qi_mask ((__v32qi) __A,
  51                                                 (__v32qi)
  52                                                  _mm256_setzero_si256 (),
  53                                                 (__mmask32) __U);
  54 }
  55
  56 extern __inline __mmask32
  57 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  58 _mm256_bitshuffle_epi64_mask (__m256i __A, __m256i __B)
  59 {
  60   return (__mmask32) __builtin_ia32_vpshufbitqmb256_mask ((__v32qi) __A,
  61                                                  (__v32qi) __B,
  62                                                  (__mmask32) -1);
  63 }
  64
  65 extern __inline __mmask32
  66 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  67 _mm256_mask_bitshuffle_epi64_mask (__mmask32 __M, __m256i __A, __m256i __B)
  68 {
  69   return (__mmask32) __builtin_ia32_vpshufbitqmb256_mask ((__v32qi) __A,
  70                                                  (__v32qi) __B,
  71                                                  (__mmask32) __M);
  72 }
  73
  74 extern __inline __mmask16
  75 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  76 _mm_bitshuffle_epi64_mask (__m128i __A, __m128i __B)
  77 {
  78   return (__mmask16) __builtin_ia32_vpshufbitqmb128_mask ((__v16qi) __A,
  79                                                  (__v16qi) __B,
  80                                                  (__mmask16) -1);
  81 }
  82
  83 extern __inline __mmask16
  84 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  85 _mm_mask_bitshuffle_epi64_mask (__mmask16 __M, __m128i __A, __m128i __B)
  86 {
  87   return (__mmask16) __builtin_ia32_vpshufbitqmb128_mask ((__v16qi) __A,
  88                                                  (__v16qi) __B,
  89                                                  (__mmask16) __M);
  90 }
  91
  92 extern __inline __m256i
  93 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
  94 _mm256_popcnt_epi8 (__m256i __A)
  95 {
  96   return (__m256i) __builtin_ia32_vpopcountb_v32qi ((__v32qi) __A);
  97 }
  98
  99 extern __inline __m256i
 100 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 101 _mm256_popcnt_epi16 (__m256i __A)
 102 {
 103   return (__m256i) __builtin_ia32_vpopcountw_v16hi ((__v16hi) __A);
 104 }
 105
 106 extern __inline __m128i
 107 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 108 _mm_popcnt_epi8 (__m128i __A)
 109 {
 110   return (__m128i) __builtin_ia32_vpopcountb_v16qi ((__v16qi) __A);
 111 }
 112
 113 extern __inline __m128i
 114 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 115 _mm_popcnt_epi16 (__m128i __A)
 116 {
 117   return (__m128i) __builtin_ia32_vpopcountw_v8hi ((__v8hi) __A);
 118 }
 119
 120 extern __inline __m256i
 121 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 122 _mm256_mask_popcnt_epi16 (__m256i __W, __mmask16 __U, __m256i __A)
 123 {
 124   return (__m256i) __builtin_ia32_vpopcountw_v16hi_mask ((__v16hi) __A,
 125                                                         (__v16hi) __W,
 126                                                         (__mmask16) __U);
 127 }
 128
 129 extern __inline __m256i
 130 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 131 _mm256_maskz_popcnt_epi16 (__mmask16 __U, __m256i __A)
 132 {
 133   return (__m256i) __builtin_ia32_vpopcountw_v16hi_mask ((__v16hi) __A,
 134                                                 (__v16hi)
 135                                                 _mm256_setzero_si256 (),
 136                                                 (__mmask16) __U);
 137 }
 138
 139 extern __inline __m128i
 140 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 141 _mm_mask_popcnt_epi8 (__m128i __W, __mmask16 __U, __m128i __A)
 142 {
 143   return (__m128i) __builtin_ia32_vpopcountb_v16qi_mask ((__v16qi) __A,
 144                                                          (__v16qi) __W,
 145                                                          (__mmask16) __U);
 146 }
 147
 148 extern __inline __m128i
 149 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 150 _mm_maskz_popcnt_epi8 (__mmask16 __U, __m128i __A)
 151 {
 152   return (__m128i) __builtin_ia32_vpopcountb_v16qi_mask ((__v16qi) __A,
 153                                                          (__v16qi)
 154                                                          _mm_setzero_si128 (),
 155                                                          (__mmask16) __U);
 156 }
 157 extern __inline __m128i
 158 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 159 _mm_mask_popcnt_epi16 (__m128i __W, __mmask8 __U, __m128i __A)
 160 {
 161   return (__m128i) __builtin_ia32_vpopcountw_v8hi_mask ((__v8hi) __A,
 162                                                         (__v8hi) __W,
 163                                                         (__mmask8) __U);
 164 }
 165
 166 extern __inline __m128i
 167 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 168 _mm_maskz_popcnt_epi16 (__mmask8 __U, __m128i __A)
 169 {
 170   return (__m128i) __builtin_ia32_vpopcountw_v8hi_mask ((__v8hi) __A,
 171                                                         (__v8hi)
 172                                                         _mm_setzero_si128 (),
 173                                                         (__mmask8) __U);
 174 }
 175 #ifdef __DISABLE_AVX512BITALGVL__
 176 #undef __DISABLE_AVX512BITALGVL__
 177 #pragma GCC pop_options
 178 #endif /* __DISABLE_AVX512BITALGVL__ */
 179
 180 #endif /* _AVX512BITALGVLINTRIN_H_INCLUDED */