github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-09-29 11:16:29 +03:00
Code Issues Packages Projects Releases Wiki Activity

AVX512VBMI Teddy.

Browse Source
This commit is contained in:
Chang, Harry
2019-11-05 21:19:38 +08:00
committed by Hong, Yang A
parent 55f336751d
commit 43204dda48
5 changed files with 341 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/util/arch.h
View File

@@ -1,5 +1,5 @@
/*
* Copyright (c) 2017, Intel Corporation
* Copyright (c) 2017-2020, Intel Corporation
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
@@ -57,6 +57,10 @@
#define HAVE_AVX512
#endif
#if defined(__AVX512VBMI__)
#define HAVE_AVX512VBMI
#endif
/*
* ICC and MSVC don't break out POPCNT or BMI/2 as separate pre-def macros
*/

16
src/util/simd_utils.h
View File

@@ -1,5 +1,5 @@
/*
* Copyright (c) 2015-2017, Intel Corporation
* Copyright (c) 2015-2020, Intel Corporation
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
@@ -150,6 +150,14 @@ static really_inline u32 movd(const m128 in) {
return _mm_cvtsi128_si32(in);
}
#if defined(HAVE_AVX512)
static really_inline u32 movd512(const m512 in) {
// NOTE: seems gcc doesn't support _mm512_cvtsi512_si32(in),
// so we use 2-step convertions to work around.
return _mm_cvtsi128_si32(_mm512_castsi512_si128(in));
}
#endif
static really_inline u64a movq(const m128 in) {
#if defined(ARCH_X86_64)
return _mm_cvtsi128_si64(in);
@@ -318,6 +326,12 @@ static really_inline
m512 maskz_pshufb_m512(__mmask64 k, m512 a, m512 b) {
return _mm512_maskz_shuffle_epi8(k, a, b);
}
#if defined(HAVE_AVX512VBMI)
#define vpermb512(idx, a) _mm512_permutexvar_epi8(idx, a)
#define maskz_vpermb512(k, idx, a) _mm512_maskz_permutexvar_epi8(k, idx, a)
#endif
#endif
static really_inline