whytofly
/
qt6-bb10
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Squeeze more vectorized rounds out of JSON Latin1String conversion. 

E.g. for a string of length 16, with i == 0, this loop can still run.
Same for the case where length is 8.

Change-Id: Ie95832b50ddeba2e0dfb0e3308e4c7a5376bb969
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
bb10
Erik Verbruggen 2016-05-25 10:34:29 +02:00 committed by Erik Verbruggen
parent 2cf63c71eb
commit 5904822b50
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/corelib/json/qjson_p.h
View File

@ -396,7 +396,7 @@ public:
const ushort *uc = (const ushort *)str.unicode();
int i = 0;
#ifdef __SSE2__
for ( ; i + 16 < len; i += 16) {
for ( ; i + 16 <= len; i += 16) {
__m128i chunk1 = _mm_loadu_si128((__m128i*)&uc[i]); // load
__m128i chunk2 = _mm_loadu_si128((__m128i*)&uc[i + 8]); // load
// pack the two vector to 16 x 8bits elements
@ -405,7 +405,7 @@ public:
}
# ifdef Q_PROCESSOR_X86_64
// we can do one more round, of 8 characters
if (i + 8 < len) {
if (i + 8 <= len) {
__m128i chunk = _mm_loadu_si128((__m128i*)&uc[i]); // load
// pack with itself, we'll discard the high part anyway
chunk = _mm_packus_epi16(chunk, chunk);