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

SuperVector shuffle implementation and test function optimized

Browse Source
This commit is contained in:
Apostolos Tapsas 2021-10-22 11:55:39 +00:00
parent 57301721f1
commit d43d6733b6
5 changed files with 7 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/nfa/limex_shuffle.h
View File

@ -45,10 +45,7 @@
static really_inline static really_inline
u32 packedExtract128(m128 s, const m128 permute, const m128 compare) { u32 packedExtract128(m128 s, const m128 permute, const m128 compare) {
m128 shuffled = pshufb_m128(s, permute); m128 shuffled = pshufb_m128(s, permute);
//int8x16_t res = (int8x16_t) pshufb_m128(s, permute); print_m128_16x8("shufled", shuffled);
//printf("shufled:");
//for(int i=15; i>=0; i--) {printf("%02x ", res[i]);}
//printf("\n");
m128 compared = and128(shuffled, compare); m128 compared = and128(shuffled, compare);
u16 rv = ~movemask128(eq128(compared, shuffled)); u16 rv = ~movemask128(eq128(compared, shuffled));
return (u32)rv; return (u32)rv;

8
src/util/arch/ppc64el/simd_utils.h
View File

@ -463,14 +463,6 @@ char testbit128(m128 val, unsigned int n) {
static really_inline static really_inline
m128 pshufb_m128(m128 a, m128 b) { m128 pshufb_m128(m128 a, m128 b) {
return (m128) vec_perm((uint8x16_t)a, (uint8x16_t)a, (uint8x16_t)b); return (m128) vec_perm((uint8x16_t)a, (uint8x16_t)a, (uint8x16_t)b);
//return (m128) vec_perm((int8x16_t)vec_splat_s8(0), (int8x16_t)a, (uint8x16_t)b);;
//uint8x16_t btransparent = vec_and((uint8x16_t)b, (uint8x16_t)vec_splats(0x8f));
//return (m128) vec_perm(a, a, btransparent);
//return (m128) vec_perm((int8x16_t)vec_splat_s8(0), (int8x16_t)b, (uint8x16_t)a);
//return (m128) vec_perm((int8x16_t)a, (int8x16_t)b, (uint8x16_t)vec_splat_s8(0));
//return (m128) vec_perm((int8x16_t)b, (int8x16_t)a, (uint8x16_t)vec_splat_s8(0));
} }
static really_inline static really_inline

2
src/util/supervector/arch/ppc64el/impl.cpp
View File

@ -603,7 +603,7 @@ template<>
template<> template<>
really_inline SuperVector<16> SuperVector<16>::pshufb<false>(SuperVector<16> b) really_inline SuperVector<16> SuperVector<16>::pshufb<false>(SuperVector<16> b)
{ {
return (m128) vec_permxor((int8x16_t)vec_splat_s8(0), (int8x16_t)u.v128[0], (int8x16_t) b.u.v128[0]); return (m128) vec_perm((uint8x16_t)u.v128[0], (uint8x16_t)u.v128[0], (uint8x16_t)b.u.v128[0]);
} }
template<> template<>

4
unit/internal/simd_utils.cpp
View File

@ -849,7 +849,7 @@ TEST(SimdUtilsTest, pshufb_m128) {
} }
u8 vec2[16]; u8 vec2[16];
for (int i=0; i<16; i++) { for (int i=0; i<16; i++) {
vec2[i]=i; vec2[i]=i + (rand() % 16 + 0);
} }
m128 v1 = loadu128(vec); m128 v1 = loadu128(vec);
m128 v2 = loadu128(vec2); m128 v2 = loadu128(vec2);
@ -857,7 +857,7 @@ TEST(SimdUtilsTest, pshufb_m128) {
u8 res[16]; u8 res[16];
store128(res, vres); store128(res, vres);
for (int i=0; i<16; i++) { for (int i=0; i<16; i++) {
ASSERT_EQ(vec[vec2[i]], res[i]); ASSERT_EQ(vec[vec2[i] % 16 ], res[i]);
} }
} }

4
unit/internal/supervector.cpp
View File

@ -280,13 +280,13 @@ TEST(SuperVectorUtilsTest,pshufb128c) {
} }
u8 vec2[16]; u8 vec2[16];
for (int i=0; i<16; i++) { for (int i=0; i<16; i++) {
vec2[i]=i; vec2[i]=i + (rand() % 15 + 0);
} }
auto SP1 = SuperVector<16>::loadu(vec); auto SP1 = SuperVector<16>::loadu(vec);
auto SP2 = SuperVector<16>::loadu(vec2); auto SP2 = SuperVector<16>::loadu(vec2);
auto SResult = SP1.template pshufb<true>(SP2); auto SResult = SP1.template pshufb<true>(SP2);
for (int i=0; i<16; i++) { for (int i=0; i<16; i++) {
ASSERT_EQ(vec[vec2[i]],SResult.u.u8[i]); ASSERT_EQ(vec[vec2[i] % 16 ],SResult.u.u8[i]);
} }
} }