github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-10-01 03:57:43 +03:00
Code Issues Packages Projects Releases Wiki Activity

use unaligned loads for short scans

Browse Source
This commit is contained in:
Konstantinos Margaritis
2021-02-11 14:21:57 +02:00
parent d3e03ed88a
commit 9fd94e0062
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/hwlm/noodle_engine_sse.c
View File

@@ -50,7 +50,7 @@ hwlm_error_t scanSingleShort(const struct noodTable *n, const u8 *buf,
return HWLM_SUCCESS; return HWLM_SUCCESS;
} }
m128 mask128 = noCase ? caseMask : ones128(); m128 mask128 = noCase ? caseMask : ones128();
m128 v = and128(load128(d), mask128); m128 v = and128(loadu128(d), mask128);
// mask out where we can't match // mask out where we can't match
u32 mask = (0xFFFF >> (16 - l)); u32 mask = (0xFFFF >> (16 - l));
@@ -103,7 +103,7 @@ hwlm_error_t scanDoubleShort(const struct noodTable *n, const u8 *buf,
DEBUG_PRINTF("d %zu\n", d - buf); DEBUG_PRINTF("d %zu\n", d - buf);
m128 mask128 = noCase ? caseMask : ones128(); m128 mask128 = noCase ? caseMask : ones128();
m128 v = and128(load128(d), mask128); m128 v = and128(loadu128(d), mask128);
u32 z = movemask128(and128(lshiftbyte_m128(eq128(mask1, v), 1), u32 z = movemask128(and128(lshiftbyte_m128(eq128(mask1, v), 1),
eq128(mask2, v))); eq128(mask2, v)));