gcc/config/i386/avx512ifmavlintrin.h

   1 /* Copyright (C) 2013-2024 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 <avx512ifmavlintrin.h> directly; include <immintrin.h> instead."
  26 #endif
  27
  28 #ifndef _AVX512IFMAVLINTRIN_H_INCLUDED
  29 #define _AVX512IFMAVLINTRIN_H_INCLUDED
  30
  31 #if !defined(__AVX512VL__) || !defined(__AVX512IFMA__) || defined (__EVEX512__)
  32 #pragma GCC push_options
  33 #pragma GCC target("avx512ifma,avx512vl,no-evex512")
  34 #define __DISABLE_AVX512IFMAVL__
  35 #endif /* __AVX512IFMAVL__ */
  36
  37 #define _mm_madd52lo_epu64(A, B, C)                       \
  38   ((__m128i) __builtin_ia32_vpmadd52luq128 ((__v2di) (A), \
  39                                             (__v2di) (B), \
  40                                             (__v2di) (C)))
  41
  42 #define _mm_madd52hi_epu64(A, B, C)                       \
  43   ((__m128i) __builtin_ia32_vpmadd52huq128 ((__v2di) (A), \
  44                                             (__v2di) (B), \
  45                                             (__v2di) (C)))
  46
  47 #define _mm256_madd52lo_epu64(A, B, C)                    \
  48   ((__m256i) __builtin_ia32_vpmadd52luq256 ((__v4di) (A), \
  49                                             (__v4di) (B), \
  50                                             (__v4di) (C)))
  51
  52
  53 #define _mm256_madd52hi_epu64(A, B, C)                    \
  54   ((__m256i) __builtin_ia32_vpmadd52huq256 ((__v4di) (A), \
  55                                             (__v4di) (B), \
  56                                             (__v4di) (C)))
  57
  58 extern __inline __m128i
  59 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  60 _mm_mask_madd52lo_epu64 (__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y)
  61 {
  62   return (__m128i) __builtin_ia32_vpmadd52luq128_mask ((__v2di) __W,
  63                                                        (__v2di) __X,
  64                                                        (__v2di) __Y,
  65                                                        (__mmask8) __M);
  66 }
  67
  68 extern __inline __m128i
  69 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  70 _mm_mask_madd52hi_epu64 (__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y)
  71 {
  72   return (__m128i) __builtin_ia32_vpmadd52huq128_mask ((__v2di) __W,
  73                                                        (__v2di) __X,
  74                                                        (__v2di) __Y,
  75                                                        (__mmask8) __M);
  76 }
  77
  78 extern __inline __m256i
  79 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  80 _mm256_mask_madd52lo_epu64 (__m256i __W, __mmask8 __M, __m256i __X,
  81                             __m256i __Y)
  82 {
  83   return (__m256i) __builtin_ia32_vpmadd52luq256_mask ((__v4di) __W,
  84                                                        (__v4di) __X,
  85                                                        (__v4di) __Y,
  86                                                        (__mmask8) __M);
  87 }
  88
  89 extern __inline __m256i
  90 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  91 _mm256_mask_madd52hi_epu64 (__m256i __W, __mmask8 __M, __m256i __X,
  92                             __m256i __Y)
  93 {
  94   return (__m256i) __builtin_ia32_vpmadd52huq256_mask ((__v4di) __W,
  95                                                        (__v4di) __X,
  96                                                        (__v4di) __Y,
  97                                                        (__mmask8) __M);
  98 }
  99
 100 extern __inline __m128i
 101 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 102 _mm_maskz_madd52lo_epu64 (__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z)
 103 {
 104   return (__m128i) __builtin_ia32_vpmadd52luq128_maskz ((__v2di) __X,
 105                                                         (__v2di) __Y,
 106                                                         (__v2di) __Z,
 107                                                         (__mmask8) __M);
 108 }
 109
 110 extern __inline __m128i
 111 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 112 _mm_maskz_madd52hi_epu64 (__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z)
 113 {
 114   return (__m128i) __builtin_ia32_vpmadd52huq128_maskz ((__v2di) __X,
 115                                                         (__v2di) __Y,
 116                                                         (__v2di) __Z,
 117                                                         (__mmask8) __M);
 118 }
 119
 120 extern __inline __m256i
 121 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 122 _mm256_maskz_madd52lo_epu64 (__mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z)
 123 {
 124   return (__m256i) __builtin_ia32_vpmadd52luq256_maskz ((__v4di) __X,
 125                                                         (__v4di) __Y,
 126                                                         (__v4di) __Z,
 127                                                         (__mmask8) __M);
 128 }
 129
 130 extern __inline __m256i
 131 __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
 132 _mm256_maskz_madd52hi_epu64 (__mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z)
 133 {
 134   return (__m256i) __builtin_ia32_vpmadd52huq256_maskz ((__v4di) __X,
 135                                                         (__v4di) __Y,
 136                                                         (__v4di) __Z,
 137                                                         (__mmask8) __M);
 138 }
 139
 140 #ifdef __DISABLE_AVX512IFMAVL__
 141 #undef __DISABLE_AVX512IFMAVL__
 142 #pragma GCC pop_options
 143 #endif /* __DISABLE_AVX512IFMAVL__ */
 144
 145 #endif /* _AVX512IFMAVLINTRIN_H_INCLUDED */