github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-11-20 10:56:40 +03:00
Code Issues Packages Projects Releases Wiki Activity

introduce Sheng-McClellan hybrid

Browse Source
This commit is contained in:
Alex Coyte
2016-12-01 14:32:47 +11:00
committed by Matthew Barr
parent f626276271
commit e51b6d23b9
35 changed files with 3804 additions and 206 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
src/nfa/limex_shuffle.h
View File

@@ -41,52 +41,6 @@
#include "util/bitutils.h"
#include "util/simd_utils.h"
#if defined(__BMI2__) || (defined(_WIN32) && defined(__AVX2__))
#define HAVE_PEXT
#endif
static really_inline
u32 packedExtract32(u32 x, u32 mask) {
#if defined(HAVE_PEXT)
// Intel BMI2 can do this operation in one instruction.
return _pext_u32(x, mask);
#else
u32 result = 0, num = 1;
while (mask != 0) {
u32 bit = findAndClearLSB_32(&mask);
if (x & (1U << bit)) {
assert(num != 0); // more than 32 bits!
result |= num;
}
num <<= 1;
}
return result;
#endif
}
static really_inline
u32 packedExtract64(u64a x, u64a mask) {
#if defined(HAVE_PEXT) && defined(ARCH_64_BIT)
// Intel BMI2 can do this operation in one instruction.
return _pext_u64(x, mask);
#else
u32 result = 0, num = 1;
while (mask != 0) {
u32 bit = findAndClearLSB_64(&mask);
if (x & (1ULL << bit)) {
assert(num != 0); // more than 32 bits!
result |= num;
}
num <<= 1;
}
return result;
#endif
}
#undef HAVE_PEXT
static really_inline
u32 packedExtract128(m128 s, const m128 permute, const m128 compare) {
m128 shuffled = pshufb(s, permute);