liz/rawTherapee
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

removed streaming code because after adding _mm_mfence() it wasn't faster than the non SSE memset

Browse Source
This commit is contained in:
heckflosse
2016-01-26 23:27:52 +01:00
parent ee665d6790
commit ded93005d9
1 changed files with 0 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
rtengine/amaze_demosaic_RT.cc
View File

@@ -181,17 +181,7 @@ SSEFUNCTION void RawImageSource::amaze_demosaic_RT(int winx, int winy, int winw,
for (int top = winy - 16; top < winy + height; top += ts - 32) {
for (int left = winx - 16; left < winx + width; left += ts - 32) {
#ifdef __SSE2__
// Using SSE2 we can zero the memory without cache pollution
vfloat zerov = ZEROV;
for(int i = 3 * tsh; i < (ts - 6)*tsh; i += 16) {
_mm_stream_ps((float*)&nyquist[i], zerov);
}
#else
memset(&nyquist[3 * tsh], 0, sizeof(unsigned char) * (ts - 6) * tsh);
#endif
//location of tile bottom edge
int bottom = min(top + ts, winy + height + 16);
//location of tile right edge
@@ -983,8 +973,6 @@ SSEFUNCTION void RawImageSource::amaze_demosaic_RT(int winx, int winy, int winw,
// refine Nyquist areas using G curvatures
if(doNyquist) {
for (int rr = nystartrow; rr < nyendrow; rr++)
// TODO_INGO: maybe this part is also worth vectorizing using _mm_movemask_ps
for (int indx = rr * ts + nystartcol + (FC(rr, 2) & 1); indx < rr * ts + nyendcol; indx += 2) {
if (nyquist2[indx >> 1]) {