[opencl] Fix small OpenCL alg issues

From comparison tests with CPU results

+ some minor speed improvements

Author: elpaso

python/analysis/auto_generated/raster/qgsaspectfilter.sip.in

@@ -30,11 +30,14 @@ nodata value if not present or outside of the border. Must be implemented by sub
 %End
 
 
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
 <<<<<<< d6e747ca57807318c0b3f6be8266f0b43c0f7747
 =======
 
 
 >>>>>>> More updates for opencl
+=======
+>>>>>>> [opencl] Fix small OpenCL alg issues
 };
 
 /************************************************************************

python/analysis/auto_generated/raster/qgsslopefilter.sip.in

@@ -29,10 +29,14 @@ Calculates output value from nine input values. The input values and the output
 nodata value if not present or outside of the border. Must be implemented by subclasses*
 %End
 
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
 <<<<<<< d6e747ca57807318c0b3f6be8266f0b43c0f7747
 
 =======
 >>>>>>> More updates for opencl
+=======
+
+>>>>>>> [opencl] Fix small OpenCL alg issues
 };
 
 /************************************************************************

src/analysis/raster/qgshillshadefilter.cpp

@@ -21,11 +21,19 @@
 QgsHillshadeFilter::QgsHillshadeFilter( const QString &inputFile, const QString &outputFile, const QString &outputFormat, double lightAzimuth,
                                         double lightAngle )
   : QgsDerivativeFilter( inputFile, outputFile, outputFormat )
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
   , mLightAzimuth( static_cast<float>( lightAzimuth ) )
   , mLightAngle( static_cast<float>( lightAngle ) )
   , mCosZenithRad( std::cos( static_cast<float>( lightAngle * M_PI ) / 180.0f ) )
   , mSinZenithRad( std::sin( static_cast<float>( lightAngle * M_PI ) / 180.0f ) )
   , mAzimuthRad( static_cast<float>( lightAzimuth * M_PI ) / 180.0f )
+=======
+  , mLightAzimuth( lightAzimuth )
+  , mLightAngle( lightAngle )
+  , mCosZenithRad( std::cos( mLightAngle * M_PI / 180.0 ) )
+  , mSinZenithRad( std::sin( mLightAngle * M_PI / 180.0 ) )
+  , mAzimuthRad( mLightAzimuth * M_PI / 180.0 )
+>>>>>>> [opencl] Fix small OpenCL alg issues
 {
 }
 
@@ -46,12 +54,17 @@ float QgsHillshadeFilter::processNineCellWindow( float *x11, float *x21, float *
   float aspect_rad = 0;
   if ( derX == 0 && derY == 0 ) //aspect undefined, take a neutral value. Better solutions?
   {
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
     aspect_rad = mAzimuthRad / 2.0f;
+=======
+    aspect_rad = mAzimuthRad / 2.0;
+>>>>>>> [opencl] Fix small OpenCL alg issues
   }
   else
   {
     aspect_rad = M_PI + std::atan2( derX, derY );
   }
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
   return std::max( 0.0f, 255.0f * ( ( mCosZenithRad * std::cos( slope_rad ) ) +
                                     ( mSinZenithRad * std::sin( slope_rad ) *
                                       std::cos( mAzimuthRad - aspect_rad ) ) ) );
@@ -84,23 +97,34 @@ void QgsHillshadeFilter::addExtraRasterParams( std::vector<float> &params )
 
 #endif
 =======
+=======
+  return std::max( 0.0, 255.0 * ( ( mCosZenithRad * std::cos( slope_rad ) ) +
+                                  ( mSinZenithRad * std::sin( slope_rad ) *
+                                    std::cos( mAzimuthRad - aspect_rad ) ) ) );
+}
+
+#ifdef HAVE_OPENCL
+
+>>>>>>> [opencl] Fix small OpenCL alg issues
 void QgsHillshadeFilter::addExtraRasterParams( std::vector<float> &params )
 {
-  float azimuthRad =  -1 * mLightAzimuth * M_PI / 180.0;
-  float zenithRad = std::max( 0.0f, 90.0f - mLightAngle ) * M_PI / 180.0;
-  float cosZenithRad = std::cos( zenithRad );
-  float cos_az_mul_cos_alt_mul_z = std::cos( azimuthRad ) * cosZenithRad * mZFactor;
-  float sin_az_mul_cos_alt_mul_z = std::sin( azimuthRad ) * cosZenithRad * mZFactor;
-  float cos_az_mul_cos_alt_mul_z_mul_254 = 254.0 * cos_az_mul_cos_alt_mul_z;
-  float sin_az_mul_cos_alt_mul_z_mul_254 = 254.0 * sin_az_mul_cos_alt_mul_z;
-  float square_z = mZFactor * mZFactor;
-  float sin_altRadians_mul_254 = 254.0 * std::sin( zenithRad );
-
-  // For fast formula from GDAL DEM
-  params.push_back( cos_az_mul_cos_alt_mul_z_mul_254 ); // 5
-  params.push_back( sin_az_mul_cos_alt_mul_z_mul_254 ); // 6
-  params.push_back( square_z ); // 7
-  params.push_back( sin_altRadians_mul_254 ); // 8
+
+  // Original CPU formula
+  float zenith_rad = mLightAngle * M_PI / 180.0;
+  float azimuth_rad = mLightAzimuth * M_PI / 180.0;
+  params.push_back( zenith_rad ); // 5
+  params.push_back( azimuth_rad ); // 6
+
+  /*
+  params.push_back( std::cos( mLightAngle * M_PI / 180.0 ) ); // cos_zenith_rad 5
+  params.push_back( mLightAzimuth * M_PI / 180.0 ); // azimuth_rad 6
+  params.push_back( std::sin( mLightAzimuth * M_PI / 180.0 ) ); // sin_zenith_rad 7
+  */
 
 }
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
 >>>>>>> [opencl] Use fast formula for hillshade
+=======
+
+#endif
+>>>>>>> [opencl] Fix small OpenCL alg issues

src/analysis/raster/qgshillshadefilter.h

@@ -44,40 +44,56 @@ class ANALYSIS_EXPORT QgsHillshadeFilter: public QgsDerivativeFilter
     void setLightAngle( float angle );
 
   private:
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
 <<<<<<< a73bbbad21629d81b9b1d4217a096a930473eb5c
 
 #ifdef HAVE_OPENCL
 
 =======
 >>>>>>> [opencl] Use fast formula for hillshade
+=======
+
+#ifdef HAVE_OPENCL
+
+>>>>>>> [opencl] Fix small OpenCL alg issues
     const QString openClProgramBaseName() const override
     {
       return QStringLiteral( "hillshade" );
     }
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
 <<<<<<< a73bbbad21629d81b9b1d4217a096a930473eb5c
 #endif
 
 =======
 >>>>>>> [opencl] Use fast formula for hillshade
+=======
+#endif
+
+>>>>>>> [opencl] Fix small OpenCL alg issues
     float mLightAzimuth;
     float mLightAngle;
     // Precalculate for speed:
     float mCosZenithRad;
     float mSinZenithRad;
     float mAzimuthRad;
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
 
 
 #ifdef HAVE_OPENCL
   private:
 
     void addExtraRasterParams( std::vector<float> &params ) override;
 #endif
+=======
+>>>>>>> [opencl] Fix small OpenCL alg issues
 
 
-    // QgsNineCellFilter interface
+#ifdef HAVE_OPENCL
   private:
 
     void addExtraRasterParams( std::vector<float> &params ) override;
+#endif
+
 };
 
 #endif // QGSHILLSHADEFILTER_H

src/analysis/raster/qgsninecellfilter.cpp

@@ -422,8 +422,13 @@ int QgsNineCellFilter::processRasterGPU( const QString &source, QgsFeedback *fee
         queue.enqueueWriteBuffer( *scanLineBuffer[rowIndex[2]], CL_TRUE, 0, bufferSize, scanLine.get() ); // row 0
 >>>>>>> [opencl] Use fast formula for hillshade
       }
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
       else // Overwrite from input, skip first and last
 >>>>>>> [opencl] Reduce memory footprint and optimize
+=======
+      else // Read line i + 1 and put it into scanline 3
+        // Overwrite from input, skip first and last
+>>>>>>> [opencl] Fix small OpenCL alg issues
       {
         if ( GDALRasterIO( rasterBand, GF_Read, 0, i + 1, xSize, 1, &scanLine[1], xSize, 1, GDT_Float32, 0, 0 ) != CE_None )
         {
@@ -497,7 +502,10 @@ int QgsNineCellFilter::processRasterGPU( const QString &source, QgsFeedback *fee
   return 0;
 }
 #endif
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
 
+=======
+>>>>>>> [opencl] Fix small OpenCL alg issues
 
 // TODO: return an anum instead of an int
 int QgsNineCellFilter::processRasterCPU( QgsFeedback *feedback )

src/analysis/raster/qgsruggednessfilter.h

@@ -47,10 +47,14 @@ class ANALYSIS_EXPORT QgsRuggednessFilter: public QgsNineCellFilter
     {
       return QStringLiteral( "ruggedness" );
     }
+<<<<<<< 573283f0dcf022e84bd615e84fd2656043a9722b
 <<<<<<< 8b81f1bb0993c3755019921eaa064d95f430c9db
 #endif
 =======
 >>>>>>> [opencl] Ruggedness index OpenCL program
+=======
+#endif
+>>>>>>> [opencl] Fix small OpenCL alg issues
 
 };
 

src/core/raster/qgshillshaderenderer.cpp

@@ -159,32 +159,6 @@ QgsRasterBlock *QgsHillshadeRenderer::block( int bandNo, const QgsRectangle &ext
   QRgb defaultNodataColor = NODATA_COLOR;
 
 
-  // Common pre-calculated values
-  float cellXSize = static_cast<float>( extent.width() ) / width;
-  float cellYSize = static_cast<float>( extent.height() ) / height;
-  float zenithRad = static_cast<float>( std::max( 0.0, 90 - mLightAngle ) * M_PI / 180.0 );
-  float azimuthRad = static_cast<float>( -1 * mLightAzimuth * M_PI / 180.0 );
-  float cosZenithRad = std::cos( zenithRad );
-  float sinZenithRad = std::sin( zenithRad );
-
-  // For fast formula from GDAL DEM
-  float cos_alt_mul_z = cosZenithRad * static_cast<float>( mZFactor );
-  float cos_az_mul_cos_alt_mul_z = std::cos( azimuthRad ) * cos_alt_mul_z;
-  float sin_az_mul_cos_alt_mul_z = std::sin( azimuthRad ) * cos_alt_mul_z;
-  float cos_az_mul_cos_alt_mul_z_mul_254 = 254.0f * cos_az_mul_cos_alt_mul_z;
-  float sin_az_mul_cos_alt_mul_z_mul_254 = 254.0f * sin_az_mul_cos_alt_mul_z;
-  float square_z = static_cast<float>( mZFactor * mZFactor );
-  float sin_altRadians_mul_254 = 254.0f * sinZenithRad;
-
-  // For multi directional
-  float sin_altRadians_mul_127 = 127.0f * sinZenithRad;
-  // 127.0 * std::cos(225.0 *  M_PI / 180.0) = -32.87001872802012
-  float cos225_az_mul_cos_alt_mul_z_mul_127 = -32.87001872802012f * cos_alt_mul_z;
-  float cos_alt_mul_z_mul_127 = 127.0f * cos_alt_mul_z;
-
-  QRgb defaultNodataColor = NODATA_COLOR;
-
-
 #ifdef HAVE_OPENCL
 
   // Use OpenCL? For now OpenCL is enabled in the default configuration only