| 1 | /* Copyright (C) 2015-2021 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 <avx5124vnniwintrin.h> directly; include <x86intrin.h> instead."
|
|---|
| 26 | #endif
|
|---|
| 27 |
|
|---|
| 28 | #ifndef _AVX5124VNNIWINTRIN_H_INCLUDED
|
|---|
| 29 | #define _AVX5124VNNIWINTRIN_H_INCLUDED
|
|---|
| 30 |
|
|---|
| 31 | #ifndef __AVX5124VNNIW__
|
|---|
| 32 | #pragma GCC push_options
|
|---|
| 33 | #pragma GCC target("avx5124vnniw")
|
|---|
| 34 | #define __DISABLE_AVX5124VNNIW__
|
|---|
| 35 | #endif /* __AVX5124VNNIW__ */
|
|---|
| 36 |
|
|---|
| 37 | extern __inline __m512i
|
|---|
| 38 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|---|
| 39 | _mm512_4dpwssd_epi32 (__m512i __A, __m512i __B, __m512i __C,
|
|---|
| 40 | __m512i __D, __m512i __E, __m128i *__F)
|
|---|
| 41 | {
|
|---|
| 42 | return (__m512i) __builtin_ia32_vp4dpwssd ((__v16si) __B,
|
|---|
| 43 | (__v16si) __C,
|
|---|
| 44 | (__v16si) __D,
|
|---|
| 45 | (__v16si) __E,
|
|---|
| 46 | (__v16si) __A,
|
|---|
| 47 | (const __v4si *) __F);
|
|---|
| 48 | }
|
|---|
| 49 |
|
|---|
| 50 | extern __inline __m512i
|
|---|
| 51 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|---|
| 52 | _mm512_mask_4dpwssd_epi32 (__m512i __A, __mmask16 __U, __m512i __B,
|
|---|
| 53 | __m512i __C, __m512i __D, __m512i __E,
|
|---|
| 54 | __m128i *__F)
|
|---|
| 55 | {
|
|---|
| 56 | return (__m512i) __builtin_ia32_vp4dpwssd_mask ((__v16si) __B,
|
|---|
| 57 | (__v16si) __C,
|
|---|
| 58 | (__v16si) __D,
|
|---|
| 59 | (__v16si) __E,
|
|---|
| 60 | (__v16si) __A,
|
|---|
| 61 | (const __v4si *) __F,
|
|---|
| 62 | (__v16si) __A,
|
|---|
| 63 | (__mmask16) __U);
|
|---|
| 64 | }
|
|---|
| 65 |
|
|---|
| 66 | extern __inline __m512i
|
|---|
| 67 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|---|
| 68 | _mm512_maskz_4dpwssd_epi32 (__mmask16 __U, __m512i __A, __m512i __B,
|
|---|
| 69 | __m512i __C, __m512i __D, __m512i __E,
|
|---|
| 70 | __m128i *__F)
|
|---|
| 71 | {
|
|---|
| 72 | return (__m512i) __builtin_ia32_vp4dpwssd_mask ((__v16si) __B,
|
|---|
| 73 | (__v16si) __C,
|
|---|
| 74 | (__v16si) __D,
|
|---|
| 75 | (__v16si) __E,
|
|---|
| 76 | (__v16si) __A,
|
|---|
| 77 | (const __v4si *) __F,
|
|---|
| 78 | (__v16si) _mm512_setzero_ps (),
|
|---|
| 79 | (__mmask16) __U);
|
|---|
| 80 | }
|
|---|
| 81 |
|
|---|
| 82 | extern __inline __m512i
|
|---|
| 83 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|---|
| 84 | _mm512_4dpwssds_epi32 (__m512i __A, __m512i __B, __m512i __C,
|
|---|
| 85 | __m512i __D, __m512i __E, __m128i *__F)
|
|---|
| 86 | {
|
|---|
| 87 | return (__m512i) __builtin_ia32_vp4dpwssds ((__v16si) __B,
|
|---|
| 88 | (__v16si) __C,
|
|---|
| 89 | (__v16si) __D,
|
|---|
| 90 | (__v16si) __E,
|
|---|
| 91 | (__v16si) __A,
|
|---|
| 92 | (const __v4si *) __F);
|
|---|
| 93 | }
|
|---|
| 94 |
|
|---|
| 95 | extern __inline __m512i
|
|---|
| 96 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|---|
| 97 | _mm512_mask_4dpwssds_epi32 (__m512i __A, __mmask16 __U, __m512i __B,
|
|---|
| 98 | __m512i __C, __m512i __D, __m512i __E,
|
|---|
| 99 | __m128i *__F)
|
|---|
| 100 | {
|
|---|
| 101 | return (__m512i) __builtin_ia32_vp4dpwssds_mask ((__v16si) __B,
|
|---|
| 102 | (__v16si) __C,
|
|---|
| 103 | (__v16si) __D,
|
|---|
| 104 | (__v16si) __E,
|
|---|
| 105 | (__v16si) __A,
|
|---|
| 106 | (const __v4si *) __F,
|
|---|
| 107 | (__v16si) __A,
|
|---|
| 108 | (__mmask16) __U);
|
|---|
| 109 | }
|
|---|
| 110 |
|
|---|
| 111 | extern __inline __m512i
|
|---|
| 112 | __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|---|
| 113 | _mm512_maskz_4dpwssds_epi32 (__mmask16 __U, __m512i __A, __m512i __B,
|
|---|
| 114 | __m512i __C, __m512i __D, __m512i __E,
|
|---|
| 115 | __m128i *__F)
|
|---|
| 116 | {
|
|---|
| 117 | return (__m512i) __builtin_ia32_vp4dpwssds_mask ((__v16si) __B,
|
|---|
| 118 | (__v16si) __C,
|
|---|
| 119 | (__v16si) __D,
|
|---|
| 120 | (__v16si) __E,
|
|---|
| 121 | (__v16si) __A,
|
|---|
| 122 | (const __v4si *) __F,
|
|---|
| 123 | (__v16si) _mm512_setzero_ps (),
|
|---|
| 124 | (__mmask16) __U);
|
|---|
| 125 | }
|
|---|
| 126 |
|
|---|
| 127 | #ifdef __DISABLE_AVX5124VNNIW__
|
|---|
| 128 | #undef __DISABLE_AVX5124VNNIW__
|
|---|
| 129 | #pragma GCC pop_options
|
|---|
| 130 | #endif /* __DISABLE_AVX5124VNNIW__ */
|
|---|
| 131 |
|
|---|
| 132 | #endif /* _AVX5124VNNIWINTRIN_H_INCLUDED */
|
|---|
