Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 1 | /*===------------- avx512vbmiintrin.h - VBMI 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 | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 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. */ |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 18 | #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi"), __min_vector_width__(512))) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 19 | |
| 20 | |
| 21 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 22 | _mm512_permutex2var_epi8(__m512i __A, __m512i __I, __m512i __B) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 23 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 24 | return (__m512i)__builtin_ia32_vpermi2varqi512((__v64qi)__A, (__v64qi)__I, |
| 25 | (__v64qi) __B); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 26 | } |
| 27 | |
| 28 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 29 | _mm512_mask_permutex2var_epi8(__m512i __A, __mmask64 __U, __m512i __I, |
| 30 | __m512i __B) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 31 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 32 | return (__m512i)__builtin_ia32_selectb_512(__U, |
| 33 | (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B), |
| 34 | (__v64qi)__A); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 35 | } |
| 36 | |
| 37 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 38 | _mm512_mask2_permutex2var_epi8(__m512i __A, __m512i __I, __mmask64 __U, |
| 39 | __m512i __B) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 40 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 41 | return (__m512i)__builtin_ia32_selectb_512(__U, |
| 42 | (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B), |
| 43 | (__v64qi)__I); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 44 | } |
| 45 | |
| 46 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 47 | _mm512_maskz_permutex2var_epi8(__mmask64 __U, __m512i __A, __m512i __I, |
| 48 | __m512i __B) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 49 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 50 | return (__m512i)__builtin_ia32_selectb_512(__U, |
| 51 | (__v64qi)_mm512_permutex2var_epi8(__A, __I, __B), |
| 52 | (__v64qi)_mm512_setzero_si512()); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 53 | } |
| 54 | |
| 55 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 56 | _mm512_permutexvar_epi8 (__m512i __A, __m512i __B) |
| 57 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 58 | return (__m512i)__builtin_ia32_permvarqi512((__v64qi) __B, (__v64qi) __A); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 59 | } |
| 60 | |
| 61 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 62 | _mm512_maskz_permutexvar_epi8 (__mmask64 __M, __m512i __A, |
| 63 | __m512i __B) |
| 64 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 65 | return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M, |
| 66 | (__v64qi)_mm512_permutexvar_epi8(__A, __B), |
| 67 | (__v64qi)_mm512_setzero_si512()); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 68 | } |
| 69 | |
| 70 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 71 | _mm512_mask_permutexvar_epi8 (__m512i __W, __mmask64 __M, __m512i __A, |
| 72 | __m512i __B) |
| 73 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 74 | return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M, |
| 75 | (__v64qi)_mm512_permutexvar_epi8(__A, __B), |
| 76 | (__v64qi)__W); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 77 | } |
| 78 | |
| 79 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
Logan Chien | dbcf412 | 2019-03-21 10:50:25 +0800 | [diff] [blame] | 80 | _mm512_multishift_epi64_epi8(__m512i __X, __m512i __Y) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 81 | { |
Logan Chien | dbcf412 | 2019-03-21 10:50:25 +0800 | [diff] [blame] | 82 | return (__m512i)__builtin_ia32_vpmultishiftqb512((__v64qi)__X, (__v64qi) __Y); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 83 | } |
| 84 | |
| 85 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
Logan Chien | dbcf412 | 2019-03-21 10:50:25 +0800 | [diff] [blame] | 86 | _mm512_mask_multishift_epi64_epi8(__m512i __W, __mmask64 __M, __m512i __X, |
| 87 | __m512i __Y) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 88 | { |
Logan Chien | dbcf412 | 2019-03-21 10:50:25 +0800 | [diff] [blame] | 89 | return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M, |
| 90 | (__v64qi)_mm512_multishift_epi64_epi8(__X, __Y), |
| 91 | (__v64qi)__W); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 92 | } |
| 93 | |
| 94 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
Logan Chien | dbcf412 | 2019-03-21 10:50:25 +0800 | [diff] [blame] | 95 | _mm512_maskz_multishift_epi64_epi8(__mmask64 __M, __m512i __X, __m512i __Y) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 96 | { |
Logan Chien | dbcf412 | 2019-03-21 10:50:25 +0800 | [diff] [blame] | 97 | return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M, |
| 98 | (__v64qi)_mm512_multishift_epi64_epi8(__X, __Y), |
| 99 | (__v64qi)_mm512_setzero_si512()); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 100 | } |
| 101 | |
| 102 | |
| 103 | #undef __DEFAULT_FN_ATTRS |
| 104 | |
| 105 | #endif |