Speedup and bugfix for new Sh/Hl tool

rtengine/dcrop.cc

@@ -1165,8 +1165,6 @@ bool Crop::setCropSizes (int rcx, int rcy, int rcw, int rch, int skip, bool inte
     orw = bw;
     orh = bh;
 
-    ProcParams& params = parent->params;
-
     parent->ipf.transCoord (parent->fw, parent->fh, bx1, by1, bw, bh, orx, ory, orw, orh);
 
     if (check_need_larger_crop_for_lcp_distortion (parent->fw, parent->fh, orx, ory, orw, orh, parent->params)) {

rtengine/ipshadowshighlights.cc

@@ -18,12 +18,10 @@
  *  along with RawTherapee.  If not, see <http://www.gnu.org/licenses/>.
  */
 
-#ifdef _OPENMP
-#include <omp.h>
-#endif
-
 #include "improcfun.h"
 #include "gauss.h"
+#include "sleef.c"
+#include "opthelper.h"
 
 namespace rtengine {
 
@@ -38,36 +36,39 @@ void ImProcFunctions::shadowsHighlights(LabImage *lab)
 
     array2D<float> mask(width, height);
     const float sigma = params->sh.radius * 5.f / scale;
+    LUTf f(32768);
 
     const auto apply =
         [&](int amount, int tonalwidth, bool hl) -> void
         {
             const float thresh = tonalwidth * 327.68f;
             const float scale = hl ? (thresh > 0.f ? 0.9f / thresh : 1.f) : thresh * 0.9f;
+
 #ifdef _OPENMP
-            #pragma omp parallel for if (multiThread)
+            #pragma omp parallel if (multiThread)
+#endif
+            {
+
+#ifdef _OPENMP
+                #pragma omp for
 #endif
                 for (int y = 0; y < height; ++y) {
                     for (int x = 0; x < width; ++x) {
                         float l = lab->L[y][x];
                         if (hl) {
-                        mask[y][x] = (l > thresh) ? 1.f : std::pow(l * scale, 4);
+                            mask[y][x] = (l > thresh) ? 1.f : pow4(l * scale);
                         } else {
-                        mask[y][x] = l <= thresh ? 1.f : std::pow(scale / l, 4);
+                            mask[y][x] = l <= thresh ? 1.f : pow4(scale / l);
                         }
                     }
                 }
-#ifdef _OPENMP
+
-            #pragma omp parallel
-#endif
-            {
                 gaussianBlur(mask, mask, width, height, sigma);
             }
 
             const float base = std::pow(4.f, float(amount)/100.f);
             const float gamma = hl ? base : 1.f / base;
 
-            LUTf f(32768);
             const float contrast = std::pow(2.f, float(amount)/100.f);
             DiagonalCurve sh_contrast({
                     DCT_NURBS,
@@ -77,20 +78,41 @@ void ImProcFunctions::shadowsHighlights(LabImage *lab)
                     0.375, std::pow(0.375 / 0.25, contrast) * 0.25,
                     1, 1
                 });
+
+            if(!hl) {
 #ifdef _OPENMP
                 #pragma omp parallel for if (multiThread)
 #endif
                 for (int l = 0; l < 32768; ++l) {
-                auto base = std::pow(l / 32768.f, gamma);
+                    auto base = pow_F(l / 32768.f, gamma);
-                if (!hl) {
                     // get a bit more contrast in the shadows
                     base = sh_contrast.getVal(base);
-                }
                     f[l] = base * 32768.f;
                 }
+            } else {
+#ifdef __SSE2__
+                vfloat c32768v = F2V(32768.f);
+                vfloat lv = _mm_setr_ps(0,1,2,3);
+                vfloat fourv = F2V(4.f);
+                vfloat gammav = F2V(gamma);
+                for (int l = 0; l < 32768; l += 4) {
+                    vfloat basev = pow_F(lv / c32768v, gammav);
+                    STVFU(f[l], basev * c32768v);
+                    lv += fourv;
+                }
+#else
+#ifdef _OPENMP
+                #pragma omp parallel for if (multiThread)
+#endif
+                for (int l = 0; l < 32768; ++l) {
+                    auto base = pow_F(l / 32768.f, gamma);
+                    f[l] = base * 32768.f;
+                }
+#endif
+            }
 
 #ifdef _OPENMP
-            #pragma omp parallel for if (multiThread)
+            #pragma omp parallel for schedule(dynamic,16) if (multiThread)
 #endif
             for (int y = 0; y < height; ++y) {
                 for (int x = 0; x < width; ++x) {

rtengine/rt_math.h

@@ -38,6 +38,12 @@ constexpr T SQR(T x)
     return x * x;
 }
 
+template<typename T>
+constexpr T pow4(T x)
+{
+    return SQR(SQR(x));
+}
+
 template<typename T>
 constexpr const T& min(const T& a)
 {

rtgui/ppversion.h

@@ -1,16 +1,18 @@
 #pragma once
 
 // This number has to be incremented whenever the PP3 file format is modified or the behaviour of a tool changes
-#define PPVERSION 331
+#define PPVERSION 332
 #define PPVERSION_AEXP 301 //value of PPVERSION when auto exposure algorithm was modified
 
 /*
   Log of version changes
-   331  2018-14-02
+   332  2018-04-26
+        new Shadows/Highlights tool
+   331  2018-02-14
         changed wavelet.Lmethod to int
-   330  2018-20-01
+   330  2018-01-20
         Added 'Auto-matched Tone Curve' button, performing histogram matching
-   329  2017-12-09
+   329  2017-09-12
         Added 'Enabled' flag for Channel Mixer, RGB Curves, HSV Equalizer and L*a*b* Adjustments
    328  2017-11-22
         Fix wrong type of ff_clipControl

rtgui/shadowshighlights.cc

@@ -37,7 +37,7 @@ ShadowsHighlights::ShadowsHighlights () : FoldableToolPanel(this, "shadowshighli
 
     pack_start (*Gtk::manage (new  Gtk::HSeparator()));
 
-    radius = Gtk::manage (new Adjuster (M("TP_SHADOWSHLIGHTS_RADIUS"), 5, 100, 1, 30));
+    radius = Gtk::manage (new Adjuster (M("TP_SHADOWSHLIGHTS_RADIUS"), 5, 100, 1, 40));
     pack_start (*radius);
 
     radius->setAdjusterListener (this);