gvsafronov/rapidjson
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Use aligned SSE register load intrinsic.

Browse Source 
The code goes to the trouble of ensuring that data is aligned at a
16-byte boundary, then goes ahead and uses the unaligned form of the
load intrinsic _mm_loadu_si128.

Either the code shouldn't bother aligning the data to the start of the
whitespace, or it should use the aligned form of the intrinsic.
This commit is contained in:
Drew Noakes 2015-01-24 23:41:09 +00:00
parent 24dd7ef839
commit 41f3c60b22
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
include/rapidjson/reader.h
View File

@ -277,7 +277,7 @@ inline const char *SkipWhitespace_SIMD(const char* p) {
// The rest of string using SIMD // The rest of string using SIMD
static const char whitespace[16] = " \n\r\t"; static const char whitespace[16] = " \n\r\t";
const __m128i w = _mm_loadu_si128((const __m128i *)&whitespace[0]); const __m128i w = _mm_load_si128((const __m128i *)&whitespace[0]);
for (;; p += 16) { for (;; p += 16) {
const __m128i s = _mm_load_si128((const __m128i *)p); const __m128i s = _mm_load_si128((const __m128i *)p);