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

use double precision for large summations, #4408

Browse Source
This commit is contained in:
heckflosse 2018-02-21 14:28:54 +01:00
parent fc751a0cce
commit e69952d654
5 changed files with 13 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
rtengine/EdgePreservingDecomposition.cc
View File

@ -42,7 +42,8 @@ float *SparseConjugateGradient(void Ax(float *Product, float *x, void *Pass), fl
} }
//s is preconditionment of r. Without, direct to r. //s is preconditionment of r. Without, direct to r.
float *s = r, rs = 0.0f; float *s = r;
double rs = 0.0; // use double precision for large summations
if(Preconditioner != nullptr) { if(Preconditioner != nullptr) {
s = new float[n]; s = new float[n];
@ -77,7 +78,7 @@ float *SparseConjugateGradient(void Ax(float *Product, float *x, void *Pass), fl
for(iterate = 0; iterate < MaximumIterates; iterate++) { for(iterate = 0; iterate < MaximumIterates; iterate++) {
//Get step size alpha, store ax while at it. //Get step size alpha, store ax while at it.
float ab = 0.0f; double ab = 0.0; // use double precision for large summations
Ax(ax, d, Pass); Ax(ax, d, Pass);
#ifdef _OPENMP #ifdef _OPENMP
#pragma omp parallel for reduction(+:ab) #pragma omp parallel for reduction(+:ab)
@ -94,7 +95,7 @@ float *SparseConjugateGradient(void Ax(float *Product, float *x, void *Pass), fl
ab = rs / ab; ab = rs / ab;
//Update x and r with this step size. //Update x and r with this step size.
float rms = 0.0; double rms = 0.0; // use double precision for large summations
#ifdef _OPENMP #ifdef _OPENMP
#pragma omp parallel for reduction(+:rms) #pragma omp parallel for reduction(+:rms)
#endif #endif

8
rtengine/PF_correct_RT.cc
View File

@ -70,7 +70,7 @@ void ImProcFunctions::PF_correct_RT(LabImage * src, LabImage * dst, double radiu
gaussianBlur (src->b, tmp1->b, src->W, src->H, radius); gaussianBlur (src->b, tmp1->b, src->W, src->H, radius);
} }
float chromave = 0.0f; double chromave = 0.0; // use double precision for large summations
#ifdef _OPENMP #ifdef _OPENMP
#pragma omp parallel #pragma omp parallel
@ -382,7 +382,7 @@ void ImProcFunctions::PF_correct_RTcam(CieImage * src, CieImage * dst, double ra
gaussianBlur (srbb, tmbb, src->W, src->H, radius); gaussianBlur (srbb, tmbb, src->W, src->H, radius);
} }
float chromave = 0.0f; double chromave = 0.0; // use double precision for large summations
#ifdef __SSE2__ #ifdef __SSE2__
@ -1042,7 +1042,7 @@ void ImProcFunctions::Badpixelscam(CieImage * src, CieImage * dst, double radius
// begin chroma badpixels // begin chroma badpixels
float chrommed = 0.f; double chrommed = 0.0; // use double precision for large summations
#ifdef _OPENMP #ifdef _OPENMP
#pragma omp parallel for reduction(+:chrommed) #pragma omp parallel for reduction(+:chrommed)
#endif #endif
@ -1646,7 +1646,7 @@ void ImProcFunctions::BadpixelsLab(LabImage * src, LabImage * dst, double radius
if(mode == 3) { if(mode == 3) {
// begin chroma badpixels // begin chroma badpixels
float chrommed = 0.f; double chrommed = 0.0; // use double precision for large summations
#ifdef _OPENMP #ifdef _OPENMP
#pragma omp parallel for reduction(+:chrommed) #pragma omp parallel for reduction(+:chrommed)
#endif #endif

4
rtengine/improcfun.cc
View File

@ -2002,8 +2002,7 @@ void ImProcFunctions::ciecam_02float (CieImage* ncie, float adap, int pW, int pw
hist16Q.clear(); hist16Q.clear();
} }
float sum = 0.f; double sum = 0.0; // use double precision for large summations
// float sumQ = 0.f;
#ifdef _OPENMP #ifdef _OPENMP
const int numThreads = min (max (width * height / 65536, 1), omp_get_max_threads()); const int numThreads = min (max (width * height / 65536, 1), omp_get_max_threads());
@ -2023,7 +2022,6 @@ void ImProcFunctions::ciecam_02float (CieImage* ncie, float adap, int pW, int pw
hist16Qthr.clear(); hist16Qthr.clear();
} }
// #pragma omp for reduction(+:sum,sumQ)
#pragma omp for reduction(+:sum) #pragma omp for reduction(+:sum)

6
rtengine/ipwavelet.cc
View File

@ -1271,7 +1271,7 @@ void ImProcFunctions::Aver( float * RESTRICT DataList, int datalen, float &aver
//find absolute mean //find absolute mean
int countP = 0, countN = 0; int countP = 0, countN = 0;
float averaP = 0.f, averaN = 0.f; double averaP = 0.0, averaN = 0.0; // use double precision for large summations
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; max = 0.f;
@ -1332,7 +1332,7 @@ void ImProcFunctions::Aver( float * RESTRICT DataList, int datalen, float &aver
void ImProcFunctions::Sigma( float * RESTRICT DataList, int datalen, float averagePlus, float averageNeg, float &sigmaPlus, float &sigmaNeg) void ImProcFunctions::Sigma( float * RESTRICT DataList, int datalen, float averagePlus, float averageNeg, float &sigmaPlus, float &sigmaNeg)
{ {
int countP = 0, countN = 0; int countP = 0, countN = 0;
float variP = 0.f, variN = 0.f; double variP = 0.0, variN = 0.0; // use double precision for large summations
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
#ifdef _OPENMP #ifdef _OPENMP
@ -1687,7 +1687,7 @@ void ImProcFunctions::WaveletcontAllL(LabImage * labco, float ** varhue, float *
float contrast = cp.contrast; float contrast = cp.contrast;
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 = 0.0; // use double precision for large summations
float max0 = 0.f; float max0 = 0.f;
float min0 = FLT_MAX; float min0 = FLT_MAX;

2
rtengine/tmo_fattal02.cc
View File

@ -276,7 +276,7 @@ float calculateGradients (Array2Df* H, Array2Df* G, int k, bool multithread)
const int width = H->getCols(); const int width = H->getCols();
const int height = H->getRows(); const int height = H->getRows();
const float divider = pow ( 2.0f, k + 1 ); const float divider = pow ( 2.0f, k + 1 );
float avgGrad = 0.0f; double avgGrad = 0.0; // use double precision for large summations
#pragma omp parallel for reduction(+:avgGrad) if(multithread) #pragma omp parallel for reduction(+:avgGrad) if(multithread)