github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-09-29 19:24:25 +03:00
Code Issues Packages Projects Releases Wiki Activity

initial SSE/AVX2 implementation

Browse Source
This commit is contained in:
Konstantinos Margaritis
2021-09-20 23:52:31 +03:00
committed by Konstantinos Margaritis
parent e5e2057ca9
commit e7161fdfec
5 changed files with 595 additions and 438 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
src/nfa/shufti_simd.hpp
View File

@@ -51,7 +51,7 @@ typename SuperVector<S>::movemask_type block(SuperVector<S> mask_lo, SuperVector
SuperVector<S> c_lo = chars & low4bits;
c_lo = mask_lo.pshufb(c_lo);
SuperVector<S> c_hi = mask_hi.pshufb(chars.rshift64(4) & low4bits);
SuperVector<S> c_hi = mask_hi.pshufb(chars.template vshr_64_imm<4>() & low4bits);
SuperVector<S> t = c_lo & c_hi;
return t.eqmask(SuperVector<S>::Zeroes());
@@ -212,7 +212,7 @@ const u8 *fwdBlockDouble(SuperVector<S> mask1_lo, SuperVector<S> mask1_hi, Super
const SuperVector<S> low4bits = SuperVector<S>::dup_u8(0xf);
SuperVector<S> chars_lo = chars & low4bits;
chars_lo.print8("chars_lo");
SuperVector<S> chars_hi = chars.rshift64(4) & low4bits;
SuperVector<S> chars_hi = chars.template vshr_64_imm<4>() & low4bits;
chars_hi.print8("chars_hi");
SuperVector<S> c1_lo = mask1_lo.pshufb(chars_lo);
c1_lo.print8("c1_lo");
@@ -227,8 +227,8 @@ const u8 *fwdBlockDouble(SuperVector<S> mask1_lo, SuperVector<S> mask1_hi, Super
c2_hi.print8("c2_hi");
SuperVector<S> t2 = c2_lo | c2_hi;
t2.print8("t2");
t2.rshift128(1).print8("t2.rshift128(1)");
SuperVector<S> t = t1 | (t2.rshift128(1));
t2.template vshr_128_imm<1>().print8("t2.rshift128(1)");
SuperVector<S> t = t1 | (t2.template vshr_128_imm<1>());
t.print8("t");
typename SuperVector<S>::movemask_type z = t.eqmask(SuperVector<S>::Ones());
@@ -250,7 +250,7 @@ static really_inline const u8 *shuftiDoubleMini(SuperVector<S> mask1_lo, SuperVe
SuperVector<S> chars_lo = chars & low4bits;
chars_lo.print8("chars_lo");
SuperVector<S> chars_hi = chars.rshift64(4) & low4bits;
SuperVector<S> chars_hi = chars.template vshr_64_imm<4>() & low4bits;
chars_hi.print8("chars_hi");
SuperVector<S> c1_lo = mask1_lo.pshufb_maskz(chars_lo, len);
c1_lo.print8("c1_lo");
@@ -265,8 +265,8 @@ static really_inline const u8 *shuftiDoubleMini(SuperVector<S> mask1_lo, SuperVe
c2_hi.print8("c2_hi");
SuperVector<S> t2 = c2_lo | c2_hi;
t2.print8("t2");
t2.rshift128(1).print8("t2.rshift128(1)");
SuperVector<S> t = t1 | (t2.rshift128(1));
t2.template vshr_128_imm<1>().print8("t2.rshift128(1)");
SuperVector<S> t = t1 | (t2.template vshr_128_imm<1>());
t.print8("t");
typename SuperVector<S>::movemask_type z = t.eqmask(SuperVector<S>::Ones());

2
src/nfa/truffle_simd.hpp
View File

@@ -53,7 +53,7 @@ typename SuperVector<S>::movemask_type block(SuperVector<S> shuf_mask_lo_highcle
SuperVector<S> shuf1 = shuf_mask_lo_highclear.pshufb(v);
SuperVector<S> t1 = v ^ highconst;
SuperVector<S> shuf2 = shuf_mask_lo_highset.pshufb(t1);
SuperVector<S> t2 = highconst.opandnot(v.rshift64(4));
SuperVector<S> t2 = highconst.opandnot(v.template vshr_64_imm<4>());
SuperVector<S> shuf3 = shuf_mask_hi.pshufb(t2);
SuperVector<S> tmp = (shuf1 | shuf2) & shuf3;