99 lines · c
1/*===------------- avx512vbmiintrin.h - VBMI intrinsics ------------------===2 *3 *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-exception7 *8 *===-----------------------------------------------------------------------===9 */10#ifndef __IMMINTRIN_H11#error "Never use <avx512vbmiintrin.h> directly; include <immintrin.h> instead."12#endif13 14#ifndef __VBMIINTRIN_H15#define __VBMIINTRIN_H16 17/* Define the default attributes for the functions in this file. */18#if defined(__cplusplus) && (__cplusplus >= 201103L)19#define __DEFAULT_FN_ATTRS \20 __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi"), \21 __min_vector_width__(512))) constexpr22#else23#define __DEFAULT_FN_ATTRS \24 __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi"), \25 __min_vector_width__(512)))26#endif27 28static __inline__ __m512i __DEFAULT_FN_ATTRS29_mm512_permutex2var_epi8(__m512i __A, __m512i __I, __m512i __B) {30 return (__m512i)__builtin_ia32_vpermi2varqi512((__v64qi)__A, (__v64qi)__I,31 (__v64qi) __B);32}33 34static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_mask_permutex2var_epi8(35 __m512i __A, __mmask64 __U, __m512i __I, __m512i __B) {36 return (__m512i)__builtin_ia32_selectb_512(__U,37 (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B),38 (__v64qi)__A);39}40 41static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_mask2_permutex2var_epi8(42 __m512i __A, __m512i __I, __mmask64 __U, __m512i __B) {43 return (__m512i)__builtin_ia32_selectb_512(__U,44 (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B),45 (__v64qi)__I);46}47 48static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_permutex2var_epi8(49 __mmask64 __U, __m512i __A, __m512i __I, __m512i __B) {50 return (__m512i)__builtin_ia32_selectb_512(__U,51 (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B),52 (__v64qi)_mm512_setzero_si512());53}54 55static __inline__ __m512i __DEFAULT_FN_ATTRS56_mm512_permutexvar_epi8(__m512i __A, __m512i __B) {57 return (__m512i)__builtin_ia32_permvarqi512((__v64qi) __B, (__v64qi) __A);58}59 60static __inline__ __m512i __DEFAULT_FN_ATTRS61_mm512_maskz_permutexvar_epi8(__mmask64 __M, __m512i __A, __m512i __B) {62 return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,63 (__v64qi)_mm512_permutexvar_epi8(__A, __B),64 (__v64qi)_mm512_setzero_si512());65}66 67static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_mask_permutexvar_epi8(68 __m512i __W, __mmask64 __M, __m512i __A, __m512i __B) {69 return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,70 (__v64qi)_mm512_permutexvar_epi8(__A, __B),71 (__v64qi)__W);72}73 74static __inline__ __m512i __DEFAULT_FN_ATTRS75_mm512_multishift_epi64_epi8(__m512i __X, __m512i __Y)76{77 return (__m512i)__builtin_ia32_vpmultishiftqb512((__v64qi)__X, (__v64qi) __Y);78}79 80static __inline__ __m512i __DEFAULT_FN_ATTRS81_mm512_mask_multishift_epi64_epi8(__m512i __W, __mmask64 __M, __m512i __X,82 __m512i __Y)83{84 return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,85 (__v64qi)_mm512_multishift_epi64_epi8(__X, __Y),86 (__v64qi)__W);87}88 89static __inline__ __m512i __DEFAULT_FN_ATTRS90_mm512_maskz_multishift_epi64_epi8(__mmask64 __M, __m512i __X, __m512i __Y)91{92 return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,93 (__v64qi)_mm512_multishift_epi64_epi8(__X, __Y),94 (__v64qi)_mm512_setzero_si512());95}96 97#undef __DEFAULT_FN_ATTRS98#endif99