Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line | 
|---|---|---|---|
| 14 | pmbaty | 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-exception | 
        ||
| 7 |  * | 
        ||
| 8 |  *===-----------------------------------------------------------------------=== | 
        ||
| 9 |  */ | 
        ||
| 10 | #ifndef __IMMINTRIN_H | 
        ||
| 11 | #error "Never use <avx512vbmiintrin.h> directly; include <immintrin.h> instead." | 
        ||
| 12 | #endif | 
        ||
| 13 | |||
| 14 | #ifndef __VBMIINTRIN_H | 
        ||
| 15 | #define __VBMIINTRIN_H | 
        ||
| 16 | |||
| 17 | /* Define the default attributes for the functions in this file. */ | 
        ||
| 18 | #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi"), __min_vector_width__(512))) | 
        ||
| 19 | |||
| 20 | |||
| 21 | static __inline__ __m512i __DEFAULT_FN_ATTRS | 
        ||
| 22 | _mm512_permutex2var_epi8(__m512i __A, __m512i __I, __m512i __B)  | 
        ||
| 23 | { | 
        ||
| 24 | return (__m512i)__builtin_ia32_vpermi2varqi512((__v64qi)__A, (__v64qi)__I,  | 
        ||
| 25 | (__v64qi) __B);  | 
        ||
| 26 | } | 
        ||
| 27 | |||
| 28 | static __inline__ __m512i __DEFAULT_FN_ATTRS | 
        ||
| 29 | _mm512_mask_permutex2var_epi8(__m512i __A, __mmask64 __U, __m512i __I, | 
        ||
| 30 |                               __m512i __B) | 
        ||
| 31 | { | 
        ||
| 32 | return (__m512i)__builtin_ia32_selectb_512(__U,  | 
        ||
| 33 | (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B),  | 
        ||
| 34 | (__v64qi)__A);  | 
        ||
| 35 | } | 
        ||
| 36 | |||
| 37 | static __inline__ __m512i __DEFAULT_FN_ATTRS | 
        ||
| 38 | _mm512_mask2_permutex2var_epi8(__m512i __A, __m512i __I, __mmask64 __U, | 
        ||
| 39 |                                __m512i __B) | 
        ||
| 40 | { | 
        ||
| 41 | return (__m512i)__builtin_ia32_selectb_512(__U,  | 
        ||
| 42 | (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B),  | 
        ||
| 43 | (__v64qi)__I);  | 
        ||
| 44 | } | 
        ||
| 45 | |||
| 46 | static __inline__ __m512i __DEFAULT_FN_ATTRS | 
        ||
| 47 | _mm512_maskz_permutex2var_epi8(__mmask64 __U, __m512i __A, __m512i __I, | 
        ||
| 48 |                                __m512i __B) | 
        ||
| 49 | { | 
        ||
| 50 | return (__m512i)__builtin_ia32_selectb_512(__U,  | 
        ||
| 51 | (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B),  | 
        ||
| 52 | (__v64qi)_mm512_setzero_si512());  | 
        ||
| 53 | } | 
        ||
| 54 | |||
| 55 | static __inline__ __m512i __DEFAULT_FN_ATTRS | 
        ||
| 56 | _mm512_permutexvar_epi8 (__m512i __A, __m512i __B)  | 
        ||
| 57 | { | 
        ||
| 58 | return (__m512i)__builtin_ia32_permvarqi512((__v64qi) __B, (__v64qi) __A);  | 
        ||
| 59 | } | 
        ||
| 60 | |||
| 61 | static __inline__ __m512i __DEFAULT_FN_ATTRS | 
        ||
| 62 | _mm512_maskz_permutexvar_epi8 (__mmask64 __M, __m512i __A, | 
        ||
| 63 |         __m512i __B) | 
        ||
| 64 | { | 
        ||
| 65 | return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,  | 
        ||
| 66 | (__v64qi)_mm512_permutexvar_epi8(__A, __B),  | 
        ||
| 67 | (__v64qi)_mm512_setzero_si512());  | 
        ||
| 68 | } | 
        ||
| 69 | |||
| 70 | static __inline__ __m512i __DEFAULT_FN_ATTRS | 
        ||
| 71 | _mm512_mask_permutexvar_epi8 (__m512i __W, __mmask64 __M, __m512i __A, | 
        ||
| 72 |              __m512i __B) | 
        ||
| 73 | { | 
        ||
| 74 | return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,  | 
        ||
| 75 | (__v64qi)_mm512_permutexvar_epi8(__A, __B),  | 
        ||
| 76 | (__v64qi)__W);  | 
        ||
| 77 | } | 
        ||
| 78 | |||
| 79 | static __inline__ __m512i __DEFAULT_FN_ATTRS | 
        ||
| 80 | _mm512_multishift_epi64_epi8(__m512i __X, __m512i __Y)  | 
        ||
| 81 | { | 
        ||
| 82 | return (__m512i)__builtin_ia32_vpmultishiftqb512((__v64qi)__X, (__v64qi) __Y);  | 
        ||
| 83 | } | 
        ||
| 84 | |||
| 85 | static __inline__ __m512i __DEFAULT_FN_ATTRS | 
        ||
| 86 | _mm512_mask_multishift_epi64_epi8(__m512i __W, __mmask64 __M, __m512i __X, | 
        ||
| 87 |                                   __m512i __Y) | 
        ||
| 88 | { | 
        ||
| 89 | return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,  | 
        ||
| 90 | (__v64qi)_mm512_multishift_epi64_epi8(__X, __Y),  | 
        ||
| 91 | (__v64qi)__W);  | 
        ||
| 92 | } | 
        ||
| 93 | |||
| 94 | static __inline__ __m512i __DEFAULT_FN_ATTRS | 
        ||
| 95 | _mm512_maskz_multishift_epi64_epi8(__mmask64 __M, __m512i __X, __m512i __Y)  | 
        ||
| 96 | { | 
        ||
| 97 | return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,  | 
        ||
| 98 | (__v64qi)_mm512_multishift_epi64_epi8(__X, __Y),  | 
        ||
| 99 | (__v64qi)_mm512_setzero_si512());  | 
        ||
| 100 | } | 
        ||
| 101 | |||
| 102 | |||
| 103 | #undef __DEFAULT_FN_ATTRS | 
        ||
| 104 | |||
| 105 | #endif |