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

Fattal: reduced number of temporary buffers needed

Browse Source
This commit is contained in:
Alberto Griggio 2017-11-08 21:49:50 +01:00
parent cb735125a2
commit b0ebcc30eb
1 changed files with 18 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
rtengine/tmo_fattal02.cc
View File

@ -488,7 +488,7 @@ void findMaxMinPercentile(const Array2Df& I,
} }
void solve_pde_fft(Array2Df *F, Array2Df *U, bool multithread); void solve_pde_fft(Array2Df *F, Array2Df *U, Array2Df *buf, bool multithread);
void tmo_fattal02(size_t width, void tmo_fattal02(size_t width,
size_t height, size_t height,
@ -625,7 +625,9 @@ void tmo_fattal02(size_t width,
{ {
gradients[k] = new Array2Df(pyramids[k]->getCols(), pyramids[k]->getRows()); gradients[k] = new Array2Df(pyramids[k]->getCols(), pyramids[k]->getRows());
avgGrad[k] = calculateGradients(pyramids[k],gradients[k], k); avgGrad[k] = calculateGradients(pyramids[k],gradients[k], k);
delete pyramids[k];
} }
delete[] pyramids;
// ph.setValue(12); // ph.setValue(12);
// calculate fi matrix // calculate fi matrix
@ -634,10 +636,8 @@ void tmo_fattal02(size_t width,
// dumpPFS( "FI.pfs", FI, "Y" ); // dumpPFS( "FI.pfs", FI, "Y" );
for ( int i=0 ; i<nlevels ; i++ ) for ( int i=0 ; i<nlevels ; i++ )
{ {
delete pyramids[i];
delete gradients[i]; delete gradients[i];
} }
delete[] pyramids;
delete[] gradients; delete[] gradients;
delete[] avgGrad; delete[] avgGrad;
// ph.setValue(16); // ph.setValue(16);
@ -702,7 +702,7 @@ void tmo_fattal02(size_t width,
} }
} }
delete Gx; //delete Gx; // RT - reused as temp buffer in solve_pde_fft, deleted later
delete Gy; delete Gy;
// solve pde and exponentiate (ie recover compressed image) // solve pde and exponentiate (ie recover compressed image)
@ -710,8 +710,9 @@ void tmo_fattal02(size_t width,
// if (fftsolver) // if (fftsolver)
{ {
MyMutex::MyLock lock(*fftwMutex); MyMutex::MyLock lock(*fftwMutex);
solve_pde_fft(FI, &L, multithread);//, ph); solve_pde_fft(FI, &L, Gx, multithread);//, ph);
} }
delete Gx;
delete FI; delete FI;
// else // else
// { // {
@ -966,7 +967,7 @@ std::vector<double> get_lambda(int n)
// not modified and the equation might not have a solution but an // not modified and the equation might not have a solution but an
// approximate solution with a minimum error is then calculated // approximate solution with a minimum error is then calculated
// double precision version // double precision version
void solve_pde_fft(Array2Df *F, Array2Df *U, bool multithread)/*, pfs::Progress &ph, void solve_pde_fft(Array2Df *F, Array2Df *U, Array2Df *buf, bool multithread)/*, pfs::Progress &ph,
bool adjust_bound)*/ bool adjust_bound)*/
{ {
// ph.setValue(20); // ph.setValue(20);
@ -974,6 +975,7 @@ void solve_pde_fft(Array2Df *F, Array2Df *U, bool multithread)/*, pfs::Progress
int width = F->getCols(); int width = F->getCols();
int height = F->getRows(); int height = F->getRows();
assert((int)U->getCols()==width && (int)U->getRows()==height); assert((int)U->getCols()==width && (int)U->getRows()==height);
assert(buf->getCols()==width && buf->getRows()==height);
// activate parallel execution of fft routines // activate parallel execution of fft routines
#ifdef RT_FFTW3F_OMP #ifdef RT_FFTW3F_OMP
@ -997,7 +999,7 @@ void solve_pde_fft(Array2Df *F, Array2Df *U, bool multithread)/*, pfs::Progress
// transforms F into eigenvector space: Ftr = // transforms F into eigenvector space: Ftr =
//DEBUG_STR << "solve_pde_fft: transform F to ev space (fft)" << std::endl; //DEBUG_STR << "solve_pde_fft: transform F to ev space (fft)" << std::endl;
Array2Df* F_tr = new Array2Df(width,height); Array2Df* F_tr = buf; //new Array2Df(width,height);
transform_normal2ev(F, F_tr); transform_normal2ev(F, F_tr);
// TODO: F no longer needed so could release memory, but as it is an // TODO: F no longer needed so could release memory, but as it is an
// input parameter we won't do that // input parameter we won't do that
@ -1030,7 +1032,7 @@ void solve_pde_fft(Array2Df *F, Array2Df *U, bool multithread)/*, pfs::Progress
// transforms U_tr back to the normal space // transforms U_tr back to the normal space
//DEBUG_STR << "solve_pde_fft: transform U_tr to normal space (fft)" << std::endl; //DEBUG_STR << "solve_pde_fft: transform U_tr to normal space (fft)" << std::endl;
transform_ev2normal(F_tr, U); transform_ev2normal(F_tr, U);
delete F_tr; // no longer needed so release memory // delete F_tr; // no longer needed so release memory
// the solution U as calculated will satisfy something like int U = 0 // the solution U as calculated will satisfy something like int U = 0
// since for any constant c, U-c is also a solution and we are mainly // since for any constant c, U-c is also a solution and we are mainly
@ -1266,7 +1268,6 @@ void ImProcFunctions::ToneMapFattal02(Imagefloat *rgb)
Median_Denoise(Yr, Yr, luminance_noise_floor, w, h, med, 1, num_threads); Median_Denoise(Yr, Yr, luminance_noise_floor, w, h, med, 1, num_threads);
} }
float noise = alpha * 0.01f; float noise = alpha * 0.01f;
if (settings->verbose) { if (settings->verbose) {
@ -1276,11 +1277,9 @@ void ImProcFunctions::ToneMapFattal02(Imagefloat *rgb)
int w2 = find_fast_dim(w) + 1; int w2 = find_fast_dim(w) + 1;
int h2 = find_fast_dim(h) + 1; int h2 = find_fast_dim(h) + 1;
Array2Df buf(w2, h2); Array2Df L(w2, h2);
rescale_nearest(Yr, buf, multiThread); rescale_nearest(Yr, L, multiThread);
tmo_fattal02(w2, h2, buf, buf, alpha, beta, noise, detail_level, multiThread); tmo_fattal02(w2, h2, L, L, alpha, beta, noise, detail_level, multiThread);
Array2Df L(w, h);
rescale_nearest(buf, L, multiThread);
// tmo_fattal02(w, h, Yr, L, alpha, beta, noise, detail_level, multiThread); // tmo_fattal02(w, h, Yr, L, alpha, beta, noise, detail_level, multiThread);
@ -1288,9 +1287,11 @@ void ImProcFunctions::ToneMapFattal02(Imagefloat *rgb)
#pragma omp parallel for if(multiThread) #pragma omp parallel for if(multiThread)
#endif #endif
for (int y = 0; y < h; y++) { for (int y = 0; y < h; y++) {
int yy = y * h2 / h;
for (int x = 0; x < w; x++) { for (int x = 0; x < w; x++) {
int xx = x * w2 / w;
float Y = Yr(x, y); float Y = Yr(x, y);
float l = std::max(L(x, y), epsilon) * (65535.f / Y); float l = std::max(L(xx, yy), epsilon) * (65535.f / Y);
rgb->r(y, x) = std::max(rgb->r(y, x), 0.f) * l; rgb->r(y, x) = std::max(rgb->r(y, x), 0.f) * l;
rgb->g(y, x) = std::max(rgb->g(y, x), 0.f) * l; rgb->g(y, x) = std::max(rgb->g(y, x), 0.f) * l;
rgb->b(y, x) = std::max(rgb->b(y, x), 0.f) * l; rgb->b(y, x) = std::max(rgb->b(y, x), 0.f) * l;