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

OMP mods in FTblockDN.cc (backing out those changes that lead to artifacts)

Browse Source 
More work is needed here to utilise AlignedBufferMP
This commit is contained in:
michael
2012-07-07 14:32:17 -04:00
parent bc46f0a871
commit 9735b3ff46
1 changed files with 35 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
rtengine/FTblockDN.cc
View File

@@ -205,6 +205,7 @@ namespace rtengine {
//now we have tile dimensions, overlaps
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//adding omp here slows it down
for (int tiletop=0; tiletop<imheight; tiletop+=tileHskip) {
for (int tileleft=0; tileleft<imwidth; tileleft+=tileWskip) {
@@ -222,10 +223,15 @@ namespace rtengine {
//pixel weight
array2D<float> totwt(width,height,ARRAY2D_CLEAR_DATA);//weight for combining DCT blocks
// OMP candidate?
//#ifdef _OPENMP
//#pragma omp parallel for
//#endif
//TODO: implement using AlignedBufferMP
//fill tile from image; convert RGB to "luma/chroma"
for (int i=tiletop, i1=0; i<tilebottom; i++, i1++)
for (int j=tileleft, j1=0; j<tileright; j++, j1++) {
for (int i=tiletop/*, i1=0*/; i<tilebottom; i++/*, i1++*/) {
int i1 = i - tiletop;
for (int j=tileleft/*, j1=0*/; j<tileright; j++/*, j1++*/) {
int j1 = j - tileleft;
float X = gain*src->r[i][j];//xyz_prophoto[0][0]*src->r[i][j] + xyz_prophoto[0][1]*src->g[i][j] + xyz_prophoto[0][2]*src->b[i][j];
float Y = gain*src->g[i][j];//xyz_prophoto[1][0]*src->r[i][j] + xyz_prophoto[1][1]*src->g[i][j] + xyz_prophoto[1][2]*src->b[i][j];
@@ -243,6 +249,7 @@ namespace rtengine {
Lin[i1][j1] = Y;
totwt[i1][j1] = 0;
}
}
//initial impulse denoise
if (dnparams.luma>0.01) {
@@ -318,7 +325,8 @@ namespace rtengine {
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
// Main detail recovery algorithm: Block loop
//OpenMP here
//OpenMP here
//adding omp here leads to artifacts
for (int vblk=0; vblk<numblox_H; vblk++) {
//printf("vblock=%d",vblk);
int vblkmod = vblk%8;
@@ -328,8 +336,12 @@ namespace rtengine {
float * buffer = new float [width + TS + 2*blkrad*offset];
float * datarow = buffer+blkrad*offset;
for (int i=0, row=top; i<TS; i++, row++) {
//#ifdef _OPENMP
//#pragma omp parallel for
//#endif
//TODO: implement using AlignedBufferMP
for (int i=0/*, row=top*/; i<TS; i++/*, row++*/) {
int row = top + i;
int rr = row;
if (row<0) {
rr = MIN(-row,height-1);
@@ -349,9 +361,7 @@ namespace rtengine {
}//now we have a padded data row
//now fill this row of the blocks with Lab high pass data
#ifdef _OPENMP
#pragma omp parallel for
#endif
//OMP here does not add speed, better handled on the outside loop
for (int hblk=0; hblk<numblox_W; hblk++) {
int left = (hblk-blkrad)*offset;
int indx = (hblk)*TS;//index of block in malloc
@@ -410,9 +420,10 @@ namespace rtengine {
fftwf_cleanup();
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
#ifdef _OPENMP
#pragma omp parallel for
#endif
//#ifdef _OPENMP
//#pragma omp parallel for
//#endif
//TODO: implement using AlignedBufferMP
for (int i=0; i<height; i++) {
for (int j=0; j<width; j++) {
//may want to include masking threshold for large hipass data to preserve edges/detail
@@ -444,11 +455,15 @@ namespace rtengine {
if (tileright<imwidth) Hmask[width-1-i] = mask;
}
//TODO: OMP candidate?
#ifdef _OPENMP
#pragma omp parallel for
#endif
//convert back to RGB and write to destination array
for (int i=tiletop, i1=0; i<tilebottom; i++, i1++) {
for (int i=tiletop/* i1=0*/; i<tilebottom; i++/*, i1++*/){
int i1 = i-tiletop;
float X,Y,Z;
for (int j=tileleft, j1=0; j<tileright; j++, j1++) {
for (int j=tileleft/*, j1=0*/; j<tileright; j++/*, j1++*/) {
int j1=j-tileleft;
Y = labdn->L[i1][j1];
X = (labdn->a[i1][j1]) + Y;
@@ -501,9 +516,10 @@ namespace rtengine {
boxabsblur(fLblox+blkstart, nbrwt, 3, 3, TS, TS);//blur neighbor weights for more robust estimation //for DCT
#ifdef _OPENMP
#pragma omp parallel for
#endif
//#ifdef _OPENMP
//#pragma omp parallel for
//#endif
//TODO: implement using AlignedBufferMP
for (int n=0; n<TS*TS; n++) { //for DCT
fLblox[blkstart+n] *= (1-expf(-SQR(nbrwt[n])/noisevar_Ldetail));
}//output neighbor averaged result
@@ -854,7 +870,7 @@ namespace rtengine {
boxblur(sfavea, sfavea, level+2, level+2, W_ab, H_ab);//increase smoothness by locally averaging shrinkage
boxblur(sfaveb, sfaveb, level+2, level+2, W_ab, H_ab);//increase smoothness by locally averaging shrinkage
//MK
#ifdef _OPENMP
#pragma omp parallel for
#endif