GenerateRgbaHost - from class instance

src/GenerateRgbaHost.h

+#ifndef GENERATE_RGBA_HOST_H_
+#define GENERATE_RGBA_HOST_H_
+
+class GenerateRgbaHost {
+public:
+	GenerateRgbaHost();
+	~GenerateRgbaHost();
+	void generate_RBGA_host(
+			int                num_cams,           // number of cameras used
+			// Parameters to generate texture tasks
+			float            * gpu_ftasks,         // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16p//		struct tp_task   * gpu_tasks,
+			int                num_tiles,          // number of tiles in task list
+			// declare arrays in device code?
+			int              * gpu_texture_indices,// packed tile + bits (now only (1 << 7)
+			int              * gpu_num_texture_tiles,  // number of texture tiles to process  (8 separate elements for accumulation)
+			int              * gpu_woi,                // x,y,width,height of the woi
+			int                width,  // <= TILES-X, use for faster processing of LWIR images (should be actual + 1)
+			int                height, // <= TILES-Y, use for faster processing of LWIR images
+			// Parameters for the texture generation
+			float          ** gpu_clt,            // [num_cams] ->[TILES-Y][TILES-X][colors][DTT_SIZE*DTT_SIZE]
+			// TODO: use geometry_correction rXY !
+			struct gc       * gpu_geometry_correction,
+			int               colors,             // number of colors (3/1)
+			int               is_lwir,            // do not perform shot correction
+			const float       cpu_params[5],      // mitigating CUDA_ERROR_INVALID_PTX
+			const float       weights[3],         // scale for R,B,G should be host_array, not gpu
+			int               dust_remove,        // Do not reduce average weight when only one image differs much from the average
+			int               keep_weights,       // return channel weights after A in RGBA (was removed) (should be 0 if gpu_texture_rbg)?
+			const int         texture_rbga_stride,     // in floats
+			float           * gpu_texture_tiles);  // (number of colors +1 + ?)*16*16 rgba texture tiles
+};
+#endif

src/test_tp.cu

@@ -65,7 +65,8 @@
 #include "tp_files.h"
 //#include "tp_paths.cuh"
 #include "tp_paths.h"
-
+#include "GenerateRgbaHost.h"
+/*
 #if TEST_LWIR
 	#define IMG_WIDTH                   640
 	#define IMG_HEIGHT                  512
@@ -83,9 +84,9 @@
 #define TILESX        (IMG_WIDTH / DTT_SIZE)
 #define TILESY        (IMG_HEIGHT / DTT_SIZE)
 #define TILESYA       ((TILESY +3) & (~3))
-
-#include "generate_RGBA_host.inc"
-
+*/
+//#include "generate_RGBA_host.inc"
+#if 0
 void generate_RBGA_host(
 		int                num_cams,           // number of cameras used
 		// Parameters to generate texture tasks
@@ -359,7 +360,7 @@ void generate_RBGA_host(
 //	 checkCudaErrors(cudaFree(gpu_num_texture_tiles));
 	 //	 __syncthreads();
 }
-
+#endif // 0
 
 
 /**
@@ -392,6 +393,7 @@ int main(int argc, char **argv)
 #endif
 
     TpPaths tpPaths(use_lwir);
+    GenerateRgbaHost generateRgbaHost{}; //  = new GenerateRgbaHost();
 
 #if TEST_LWIR
 
@@ -474,22 +476,13 @@ int main(int argc, char **argv)
 
     int KERN_TILES = KERNELS_HOR *  KERNELS_VERT * num_colors; // NUM_COLORS;
     int KERN_SIZE =  KERN_TILES * 4 * 64;
-
     int CORR_SIZE = (2 * CORR_OUT_RAD + 1) * (2 * CORR_OUT_RAD + 1);
-
-
-
-
     float            * host_kern_buf =  (float *)malloc(KERN_SIZE * sizeof(float));
-// static - see https://stackoverflow.com/questions/20253267/segmentation-fault-before-main
-///    static struct tp_task     task_data   [TILESX*TILESY]; // maximal length - each tile
-///    static struct tp_task     task_data1  [TILESX*TILESY]; // maximal length - each tile
-
     float * ftask_data  =  (float *) malloc(TILESX * TILESY * task_size * sizeof(float));
     float * ftask_data1  = (float *) malloc(TILESX * TILESY * task_size * sizeof(float));
 
     trot_deriv  rot_deriv;
-///    int                corr_indices         [NUM_PAIRS*TILESX*TILESY];
+
     int                texture_indices      [TILESX*TILESYA];
     int                cpu_woi              [4];
 
@@ -521,7 +514,6 @@ int main(int argc, char **argv)
     float            * gpu_generate_RBGA_params;
     int                num_corrs;
     int                num_textures;
-/// int                num_ports = NUM_CAMS;
     // GPU pointers to GPU pointers to memory
     float           ** gpu_kernels; //           [NUM_CAMS];
     struct CltExtra ** gpu_kernel_offsets; //    [NUM_CAMS];
@@ -532,7 +524,6 @@ int main(int argc, char **argv)
 
 
     // GPU pointers to GPU memory
-///    struct tp_task  * gpu_tasks;  // TODO: ***** remove ! **** DONE
     float *           gpu_ftasks; // TODO: ***** allocate ! **** DONE
     int *             gpu_active_tiles;
     int *             gpu_num_active;
@@ -554,6 +545,7 @@ int main(int argc, char **argv)
     size_t  dstride_textures_rbga; // in bytes ! for one rgba/ya 16x16 tile
 
     struct gc fgeometry_correction;
+
     float*  correction_vector;
     int     correction_vector_length;
     float * rByRDist;
@@ -2044,7 +2036,7 @@ int main(int argc, char **argv)
     	}
     	// FIXME: update to use new correlations and num_cams
 #ifdef NO_DP
-    	generate_RBGA_host (
+    	generateRgbaHost.generate_RBGA_host (
     			num_cams,              // int                num_cams,           // number of cameras used
     	// Parameters to generate texture tasks
 				gpu_ftasks,         // float            * gpu_ftasks,         // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
@@ -2222,6 +2214,7 @@ int main(int argc, char **argv)
 	free (correction_vector);
 	free (ftask_data);
 	free (ftask_data1);
+//	delete generateRgbaHost;
 
 	exit(0);
 }

src/tp_defines.h

@@ -155,6 +155,25 @@
 
 #endif //#ifdef 	DEBUG_ANY
 
+#if TEST_LWIR
+	#define IMG_WIDTH                   640
+	#define IMG_HEIGHT                  512
+	#define KERNELS_HOR                 82 // 80+2
+	#define KERNELS_VERT                66 // 64+2
+#else
+	#define IMG_WIDTH                  2592
+	#define IMG_HEIGHT                 1936
+	#define KERNELS_HOR                 164  // 2592 / 16 + 2
+	#define KERNELS_VERT                123  // 1936 / 16 + 2
+#endif
+#define CORR_OUT_RAD                   7 // full tile (15x15), was 4 (9x9)
+#define DBG_DISPARITY                  0.0 // 56.0//   0.0 // 56.0 // disparity for which to calculate offsets (not needed in Java)
+// only used in C++ test
+#define TILESX        (IMG_WIDTH / DTT_SIZE)
+#define TILESY        (IMG_HEIGHT / DTT_SIZE)
+#define TILESYA       ((TILESY +3) & (~3))
+
+
 
 #endif //#ifndef JCUDA