mirror of
https://github.com/VectorCamp/vectorscan.git
synced 2025-09-29 19:24:25 +03:00
SuperVector shuffle implementation and test function optimized
This commit is contained in:
@@ -45,10 +45,7 @@
|
||||
static really_inline
|
||||
u32 packedExtract128(m128 s, const m128 permute, const m128 compare) {
|
||||
m128 shuffled = pshufb_m128(s, permute);
|
||||
//int8x16_t res = (int8x16_t) pshufb_m128(s, permute);
|
||||
//printf("shufled:");
|
||||
//for(int i=15; i>=0; i--) {printf("%02x ", res[i]);}
|
||||
//printf("\n");
|
||||
print_m128_16x8("shufled", shuffled);
|
||||
m128 compared = and128(shuffled, compare);
|
||||
u16 rv = ~movemask128(eq128(compared, shuffled));
|
||||
return (u32)rv;
|
||||
|
@@ -463,14 +463,6 @@ char testbit128(m128 val, unsigned int n) {
|
||||
static really_inline
|
||||
m128 pshufb_m128(m128 a, m128 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
|
||||
|
@@ -603,7 +603,7 @@ template<>
|
||||
template<>
|
||||
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<>
|
||||
|
Reference in New Issue
Block a user