Cleanup fftw when closing RT, instead of after each time fftw is used. Speeds up fattal and luminance denoise

rtengine/FTblockDN.cc

@@ -41,6 +41,7 @@
 #ifdef _OPENMP
 #include <omp.h>
 #endif
+#include "StopWatch.h"
 
 #define TS 64       // Tile size
 #define offset 25   // shift between tiles
@@ -475,6 +476,7 @@ enum nrquality {QUALITY_STANDARD, QUALITY_HIGH};
 
 SSEFUNCTION void ImProcFunctions::RGB_denoise(int kall, Imagefloat * src, Imagefloat * dst, Imagefloat * calclum, float * ch_M, float *max_r, float *max_b, bool isRAW, const procparams::DirPyrDenoiseParams & dnparams, const double expcomp, const NoiseCurve & noiseLCurve, const NoiseCurve & noiseCCurve, float &nresi, float &highresi)
 {
+BENCHFUN
 //#ifdef _DEBUG
     MyTime t1e, t2e;
     t1e.set();
@@ -1781,7 +1783,6 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise(int kall, Imagefloat * src, Imagef
                 fftwf_destroy_plan(plan_backward_blox[0]);
                 fftwf_destroy_plan(plan_forward_blox[1]);
                 fftwf_destroy_plan(plan_backward_blox[1]);
-                fftwf_cleanup();
             }
         } while (memoryAllocationFailed && numTries < 2 && (options.rgbDenoiseThreadLimit == 0) && !ponder);
 

rtengine/init.cc

@@ -16,6 +16,7 @@
  *  You should have received a copy of the GNU General Public License
  *  along with RawTherapee.  If not, see <http://www.gnu.org/licenses/>.
  */
+#include <fftw3.h>
 #include "../rtgui/profilestorecombobox.h"
 #include "rtengine.h"
 #include "iccstore.h"
@@ -108,10 +109,15 @@ int init (const Settings* s, Glib::ustring baseDir, Glib::ustring userSettingsDi
 
 void cleanup ()
 {
-
     ProcParams::cleanup ();
     Color::cleanup ();
     RawImageSource::cleanup ();
+#ifdef RT_FFTW3F_OMP
+    fftwf_cleanup_threads();
+#else
+    fftwf_cleanup();
+#endif
+
 }
 
 StagedImageProcessor* StagedImageProcessor::create (InitialImage* initialImage)

rtengine/tmo_fattal02.cc

@@ -907,18 +907,6 @@ void solve_pde_fft (Array2Df *F, Array2Df *U, Array2Df *buf, bool multithread)/*
     for (int i = 0; i < width * height; i++) {
         (*U) (i) -= max;
     }
-
-    // fft parallel threads cleanup, better handled outside this function?
-#ifdef RT_FFTW3F_OMP
-
-    if (multithread) {
-        fftwf_cleanup_threads();
-    }
-
-#endif
-
-    // ph.setValue(90);
-    //DEBUG_STR << "solve_pde_fft: done" << std::endl;
 }