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

Wavelet levels - some small optimizations nopatch

Browse Source
This commit is contained in:
jdc
2015-04-13 11:44:05 +02:00
parent 6a8ae4c485
commit 6240014433
1 changed files with 28 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
rtengine/ipwavelet.cc
View File

@@ -930,7 +930,8 @@ omp_set_nested(oldNested);
float averaP = 0.f, averaN = 0.f; float averaP = 0.f, averaN = 0.f;
float thres = 5.f;//different fom zero to take into account only data large enough float thres = 5.f;//different fom zero to take into account only data large enough
max=0.f;
min=0.f;
#ifdef _OPENMP #ifdef _OPENMP
#pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
@@ -1068,6 +1069,9 @@ StopWatch Stop1("Evaluate2");
float multL=(float)contrast*(maxl-1.f)/100.f + 1.f; float multL=(float)contrast*(maxl-1.f)/100.f + 1.f;
float multH=(float) contrast*(maxh-1.f)/100.f + 1.f; float multH=(float) contrast*(maxh-1.f)/100.f + 1.f;
double avedbl=0.f; // use double precision for big summations double avedbl=0.f; // use double precision for big summations
float max0 = 0.f;
float min0 = FLT_MAX;
if(contrast != 0.f) { // contrast = 0.f means that all will be multiplied by 1.f, so we can skip this step if(contrast != 0.f) { // contrast = 0.f means that all will be multiplied by 1.f, so we can skip this step
#ifdef _OPENMP #ifdef _OPENMP
#pragma omp parallel for reduction(+:avedbl) num_threads(wavNestedLevels) if(wavNestedLevels>1) #pragma omp parallel for reduction(+:avedbl) num_threads(wavNestedLevels) if(wavNestedLevels>1)
@@ -1075,17 +1079,37 @@ StopWatch Stop1("Evaluate2");
for (int i=0; i<W_L*H_L; i++) { for (int i=0; i<W_L*H_L; i++) {
avedbl += WavCoeffs_L0[i]; avedbl += WavCoeffs_L0[i];
} }
#pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
{
float lminL = FLT_MAX;
float lmaxL = 0.f;
#pragma omp for
for(int i = 0; i < W_L*H_L; i++) {
if(WavCoeffs_L0[i] < lminL) lminL = WavCoeffs_L0[i];
if(WavCoeffs_L0[i] > lmaxL) lmaxL = WavCoeffs_L0[i];
}
#pragma omp critical
{ if(lminL < min0) min0 = lminL;
if(lmaxL > max0) max0 = lmaxL;
}
}
} }
float max0=0.f;
float min0=50000.f;
#ifdef _OPENMP #ifdef _OPENMP
//#pragma omp parallel for num_threads(wavNestedLevels) if(wavNestedLevels>1) //#pragma omp parallel for num_threads(wavNestedLevels) if(wavNestedLevels>1)
#endif #endif
for (int i=0; i<W_L*H_L; i++) { /* for (int i=0; i<W_L*H_L; i++) {
if(WavCoeffs_L0[i] > max0) max0=WavCoeffs_L0[i]; if(WavCoeffs_L0[i] > max0) max0=WavCoeffs_L0[i];
if(WavCoeffs_L0[i] < min0) min0=WavCoeffs_L0[i]; if(WavCoeffs_L0[i] < min0) min0=WavCoeffs_L0[i];
} }
*/
max0/=327.68f; max0/=327.68f;
min0/=327.68f; min0/=327.68f;
float ave = avedbl / (double)(W_L*H_L); float ave = avedbl / (double)(W_L*H_L);