github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-06-28 16:41:01 +03:00
Code Issues Packages Projects Releases Wiki Activity

Fix combine2x128

Browse Source
This commit is contained in:
Matthew Barr 2016-09-13 15:07:16 +10:00
parent 8ff7a3cdbb
commit 99e14df117
3 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
src/nfa/shufti.c
View File

@ -308,7 +308,7 @@ const u8 *fwdBlockShort(m256 mask, m128 chars, const u8 *buf,
const m256 low4bits) {
// do the hi and lo shuffles in the one avx register
m256 c = set2x128(chars);
c = _mm256_srlv_epi64(c, _mm256_set_epi64x(0, 0, 4, 4));
c = _mm256_srlv_epi64(c, _mm256_set_epi64x(4, 4, 0, 0));
c = and256(c, low4bits);
m256 c_shuf = vpshufb(mask, c);
m128 t = and128(movdq_hi(c_shuf), cast256to128(c_shuf));
@ -440,7 +440,7 @@ const u8 *revBlockShort(m256 mask, m128 chars, const u8 *buf,
const m256 low4bits) {
// do the hi and lo shuffles in the one avx register
m256 c = set2x128(chars);
c = _mm256_srlv_epi64(c, _mm256_set_epi64x(0, 0, 4, 4));
c = _mm256_srlv_epi64(c, _mm256_set_epi64x(4, 4, 0, 0));
c = and256(c, low4bits);
m256 c_shuf = vpshufb(mask, c);
m128 t = and128(movdq_hi(c_shuf), cast256to128(c_shuf));
@ -565,7 +565,7 @@ const u8 *fwdBlockShort2(m256 mask1, m256 mask2, m128 chars, const u8 *buf,
const m256 low4bits) {
// do the hi and lo shuffles in the one avx register
m256 c = set2x128(chars);
c = _mm256_srlv_epi64(c, _mm256_set_epi64x(0, 0, 4, 4));
c = _mm256_srlv_epi64(c, _mm256_set_epi64x(4, 4, 0, 0));
c = and256(c, low4bits);
m256 c_shuf1 = vpshufb(mask1, c);
m256 c_shuf2 = rshift128_m256(vpshufb(mask2, c), 1);

4
src/rose/rose_build_bytecode.cpp
View File

@ -3054,8 +3054,8 @@ bool makeRoleShufti(const vector<LookEntry> &look,
neg_mask &= 0xffff;
array<u8, 32> nib_mask;
array<u8, 16> bucket_select_mask_16;
copy(hi_mask.begin(), hi_mask.begin() + 16, nib_mask.begin());
copy(lo_mask.begin(), lo_mask.begin() + 16, nib_mask.begin() + 16);
copy(lo_mask.begin(), lo_mask.begin() + 16, nib_mask.begin());
copy(hi_mask.begin(), hi_mask.begin() + 16, nib_mask.begin() + 16);
copy(bucket_select_lo.begin(), bucket_select_lo.begin() + 16,
bucket_select_mask_16.begin());
auto ri = make_unique<RoseInstrCheckShufti16x8>

6
src/util/simd_utils.h
View File

@ -658,8 +658,8 @@ m128 movdq_lo(m256 x) {
}
static really_inline
m256 combine2x128(m128 a, m128 b) {
m256 rv = {a, b};
m256 combine2x128(m128 hi, m128 lo) {
m256 rv = {lo, hi};
return rv;
}
@ -712,7 +712,7 @@ m256 combine2x128(m128 hi, m128 lo) {
#if defined(_mm256_set_m128i)
return _mm256_set_m128i(hi, lo);
#else
return insert128to256(cast128to256(hi), lo, 1);
return insert128to256(cast128to256(lo), hi, 1);
#endif
}
#endif //AVX2