Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 1 | /*===------------- avx512vnniintrin.h - VNNI intrinsics ------------------=== |
| 2 | * |
| 3 | * |
Logan Chien | df4f766 | 2019-09-04 16:45:23 -0700 | [diff] [blame] | 4 | * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 5 | * See https://llvm.org/LICENSE.txt for license information. |
| 6 | * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 7 | * |
| 8 | *===-----------------------------------------------------------------------=== |
| 9 | */ |
| 10 | #ifndef __IMMINTRIN_H |
| 11 | #error "Never use <avx512vnniintrin.h> directly; include <immintrin.h> instead." |
| 12 | #endif |
| 13 | |
| 14 | #ifndef __AVX512VNNIINTRIN_H |
| 15 | #define __AVX512VNNIINTRIN_H |
| 16 | |
| 17 | /* Define the default attributes for the functions in this file. */ |
| 18 | #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512vnni"), __min_vector_width__(512))) |
| 19 | |
| 20 | |
| 21 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 22 | _mm512_dpbusd_epi32(__m512i __S, __m512i __A, __m512i __B) |
| 23 | { |
| 24 | return (__m512i)__builtin_ia32_vpdpbusd512((__v16si)__S, (__v16si)__A, |
| 25 | (__v16si)__B); |
| 26 | } |
| 27 | |
| 28 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 29 | _mm512_mask_dpbusd_epi32(__m512i __S, __mmask16 __U, __m512i __A, __m512i __B) |
| 30 | { |
| 31 | return (__m512i)__builtin_ia32_selectd_512(__U, |
| 32 | (__v16si)_mm512_dpbusd_epi32(__S, __A, __B), |
| 33 | (__v16si)__S); |
| 34 | } |
| 35 | |
| 36 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 37 | _mm512_maskz_dpbusd_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B) |
| 38 | { |
| 39 | return (__m512i)__builtin_ia32_selectd_512(__U, |
| 40 | (__v16si)_mm512_dpbusd_epi32(__S, __A, __B), |
| 41 | (__v16si)_mm512_setzero_si512()); |
| 42 | } |
| 43 | |
| 44 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 45 | _mm512_dpbusds_epi32(__m512i __S, __m512i __A, __m512i __B) |
| 46 | { |
| 47 | return (__m512i)__builtin_ia32_vpdpbusds512((__v16si)__S, (__v16si)__A, |
| 48 | (__v16si)__B); |
| 49 | } |
| 50 | |
| 51 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 52 | _mm512_mask_dpbusds_epi32(__m512i __S, __mmask16 __U, __m512i __A, __m512i __B) |
| 53 | { |
| 54 | return (__m512i)__builtin_ia32_selectd_512(__U, |
| 55 | (__v16si)_mm512_dpbusds_epi32(__S, __A, __B), |
| 56 | (__v16si)__S); |
| 57 | } |
| 58 | |
| 59 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 60 | _mm512_maskz_dpbusds_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B) |
| 61 | { |
| 62 | return (__m512i)__builtin_ia32_selectd_512(__U, |
| 63 | (__v16si)_mm512_dpbusds_epi32(__S, __A, __B), |
| 64 | (__v16si)_mm512_setzero_si512()); |
| 65 | } |
| 66 | |
| 67 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 68 | _mm512_dpwssd_epi32(__m512i __S, __m512i __A, __m512i __B) |
| 69 | { |
| 70 | return (__m512i)__builtin_ia32_vpdpwssd512((__v16si)__S, (__v16si)__A, |
| 71 | (__v16si)__B); |
| 72 | } |
| 73 | |
| 74 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 75 | _mm512_mask_dpwssd_epi32(__m512i __S, __mmask16 __U, __m512i __A, __m512i __B) |
| 76 | { |
| 77 | return (__m512i)__builtin_ia32_selectd_512(__U, |
| 78 | (__v16si)_mm512_dpwssd_epi32(__S, __A, __B), |
| 79 | (__v16si)__S); |
| 80 | } |
| 81 | |
| 82 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 83 | _mm512_maskz_dpwssd_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B) |
| 84 | { |
| 85 | return (__m512i)__builtin_ia32_selectd_512(__U, |
| 86 | (__v16si)_mm512_dpwssd_epi32(__S, __A, __B), |
| 87 | (__v16si)_mm512_setzero_si512()); |
| 88 | } |
| 89 | |
| 90 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 91 | _mm512_dpwssds_epi32(__m512i __S, __m512i __A, __m512i __B) |
| 92 | { |
| 93 | return (__m512i)__builtin_ia32_vpdpwssds512((__v16si)__S, (__v16si)__A, |
| 94 | (__v16si)__B); |
| 95 | } |
| 96 | |
| 97 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 98 | _mm512_mask_dpwssds_epi32(__m512i __S, __mmask16 __U, __m512i __A, __m512i __B) |
| 99 | { |
| 100 | return (__m512i)__builtin_ia32_selectd_512(__U, |
| 101 | (__v16si)_mm512_dpwssds_epi32(__S, __A, __B), |
| 102 | (__v16si)__S); |
| 103 | } |
| 104 | |
| 105 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 106 | _mm512_maskz_dpwssds_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B) |
| 107 | { |
| 108 | return (__m512i)__builtin_ia32_selectd_512(__U, |
| 109 | (__v16si)_mm512_dpwssds_epi32(__S, __A, __B), |
| 110 | (__v16si)_mm512_setzero_si512()); |
| 111 | } |
| 112 | |
| 113 | #undef __DEFAULT_FN_ATTRS |
| 114 | |
| 115 | #endif |