From 09c55ca6eccae42892139791cc560643f7368c25 Mon Sep 17 00:00:00 2001
From: heckflosse <heckflosse67@gmx.de>
Date: Tue, 4 Dec 2018 23:13:20 +0100
Subject: [PATCH] Small speedup and code cleanup for autocontrast calculation

---
 rtengine/rt_algo.cc | 28 ++++++++++++++++------------
 1 file changed, 16 insertions(+), 12 deletions(-)

diff --git a/rtengine/rt_algo.cc b/rtengine/rt_algo.cc
index 4552593cc..a258287bf 100644
--- a/rtengine/rt_algo.cc
+++ b/rtengine/rt_algo.cc
@@ -203,20 +203,19 @@ void buildBlendMask(float** luminance, float **blend, int W, int H, float &contr
             }
         }
     } else {
-        constexpr float scale = 0.0625f / 327.68f;
         if (autoContrast) {
             for (int pass = 0; pass < 2; ++pass) {
                 const int tilesize = 80 / (pass + 1);
                 const int skip = pass < 1 ? tilesize : tilesize / 4;
                 const int numTilesW = W / skip - 3 * pass;
                 const int numTilesH = H / skip - 3 * pass;
-                std::vector<std::vector<std::pair<float, float>>> variances(numTilesH, std::vector<std::pair<float, float>>(numTilesW));
+                std::vector<std::vector<float>> variances(numTilesH, std::vector<float>(numTilesW));
 
-                #pragma omp parallel for
+                #pragma omp parallel for schedule(dynamic)
                 for (int i = 0; i < numTilesH; ++i) {
-                    int tileY = i * skip;
+                    const int tileY = i * skip;
                     for (int j = 0; j < numTilesW; ++j) {
-                        int tileX = j * skip;
+                        const int tileX = j * skip;
 #ifdef __SSE2__
                         vfloat avgv = ZEROV;
                         for (int y = tileY; y < tileY + tilesize; ++y) {
@@ -226,7 +225,7 @@ void buildBlendMask(float** luminance, float **blend, int W, int H, float &contr
                         }
                         float avg = vhadd(avgv);
 #else
-                        float avg = 0.;
+                        float avg = 0.f;
                         for (int y = tileY; y < tileY + tilesize; ++y) {
                             for (int x = tileX; x < tileX + tilesize; ++x) {
                                 avg += luminance[y][x];
@@ -234,6 +233,11 @@ void buildBlendMask(float** luminance, float **blend, int W, int H, float &contr
                         }
 #endif
                         avg /= SQR(tilesize);
+                        if (avg < 2000.f || avg > 20000.f) {
+                            // too dark or too bright => skip the tile
+                            variances[i][j] = RT_INFINITY_F;
+                            continue;
+                        }
 #ifdef __SSE2__
                         vfloat varv = ZEROV;
                         avgv = F2V(avg);
@@ -244,16 +248,15 @@ void buildBlendMask(float** luminance, float **blend, int W, int H, float &contr
                         }
                         float var = vhadd(varv);
 #else
-                        float var = 0.0;
+                        float var = 0.f;
                         for (int y = tileY; y < tileY + tilesize; ++y) {
                             for (int x = tileX; x < tileX + tilesize; ++x) {
                                 var += SQR(luminance[y][x] - avg);
                             }
                         }
-    #endif
+#endif
                         var /= (SQR(tilesize) * avg);
-                        variances[i][j].first = var;
-                        variances[i][j].second = avg;
+                        variances[i][j] = var;
                     }
                 }
 
@@ -261,8 +264,8 @@ void buildBlendMask(float** luminance, float **blend, int W, int H, float &contr
                 int minI = 0, minJ = 0;
                 for (int i = 0; i < numTilesH; ++i) {
                     for (int j = 0; j < numTilesW; ++j) {
-                        if (variances[i][j].first < minvar && variances[i][j].second > 2000.f && variances[i][j].second < 20000.f) {
-                            minvar = variances[i][j].first;
+                        if (variances[i][j] < minvar) {
+                            minvar = variances[i][j];
                             minI = i;
                             minJ = j;
                         }
@@ -295,6 +298,7 @@ void buildBlendMask(float** luminance, float **blend, int W, int H, float &contr
                 }
             }
         } else {
+            constexpr float scale = 0.0625f / 327.68f;
 #ifdef _OPENMP
             #pragma omp parallel
 #endif