Commit 8caaa2db authored by Andrey Filippov's avatar Andrey Filippov
Browse files

debugging code to generate data for macroblocks correlation

parent fdc9840a

src/TileProcessor.cuh

+294 −68

File changed.

Preview size limit exceeded, changes collapsed.

src/TileProcessor.h

+10 −11
Original line number Original line Diff line number Diff line
@@ -75,11 +75,13 @@ extern "C" __global__ void correlate2D( 
		float           * gpu_corrs);          // correlation output data

		float           * gpu_corrs);          // correlation output data









extern "C" __global__ void textures_accumulate(

extern "C" __global__ void textures_nonoverlap(

		int             * woi,                // x, y, width,height

		struct tp_task  * gpu_tasks,

		float          ** gpu_clt,            // [NUM_CAMS] ->[TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE]

		int               num_tiles,          // number of tiles in task list

		size_t            num_texture_tiles,  // number of texture tiles to process

// declare arrays in device code?

		int             * gpu_texture_indices,// packed tile + bits (now only (1 << 7)

		int             * gpu_texture_indices,// packed tile + bits (now only (1 << 7)

		int             * pnum_texture_tiles,  // returns total number of elements in gpu_texture_indices array

		float          ** gpu_clt,            // [NUM_CAMS] ->[TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE]

		// TODO: use geometry_correction rXY !

		// TODO: use geometry_correction rXY !

		struct gc       * gpu_geometry_correction,

		struct gc       * gpu_geometry_correction,

		int               colors,             // number of colors (3/1)

		int               colors,             // number of colors (3/1)
@@ -91,14 +93,11 @@ extern "C" __global__ void textures_accumulate( 
		float             min_agree,          // minimal number of channels to agree on a point (real number to work with fuzzy averages)

		float             min_agree,          // minimal number of channels to agree on a point (real number to work with fuzzy averages)

		float             weights[3],         // scale for R,B,G

		float             weights[3],         // scale for R,B,G

		int               dust_remove,        // Do not reduce average weight when only one image differs much from the average

		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)?

//		int               keep_weights,       // return channel weights after A in RGBA (was removed) (should be 0 if gpu_texture_rbg)?

// combining both non-overlap and overlap (each calculated if pointer is not null )

// combining both non-overlap and overlap (each calculated if pointer is not null )

		size_t            texture_rbg_stride, // in floats

		size_t            texture_stride,     // in floats (now 256*4 = 1024)  // may be 0 if not needed

		float           * gpu_texture_rbg,    // (number of colors +1 + ?)*16*16 rgba texture tiles

		float           * gpu_texture_tiles,  // (number of colors +1 + ?)*16*16 rgba texture tiles    // may be 0 if not needed

		size_t            texture_stride,     // in floats (now 256*4 = 1024)

		float           * gpu_diff_rgb_combo); // diff[NUM_CAMS], R[NUM_CAMS], B[NUM_CAMS],G[NUM_CAMS] // may be 0 if not needed

		float           * gpu_texture_tiles,  // (number of colors +1 + ?)*16*16 rgba texture tiles

		float           * gpu_diff_rgb_combo); // diff[NUM_CAMS], R[NUM_CAMS], B[NUM_CAMS],G[NUM_CAMS]







extern "C"

extern "C"

__global__ void imclt_rbg_all(

__global__ void imclt_rbg_all(

src/test_tp.cu

+13 −14
Original line number Original line Diff line number Diff line
@@ -1075,12 +1075,14 @@ int main(int argc, char **argv) 
		// Channel0 weight = 0.294118

		// Channel0 weight = 0.294118

		// Channel1 weight = 0.117647

		// Channel1 weight = 0.117647

		// Channel2 weight = 0.588235

		// Channel2 weight = 0.588235

    	textures_accumulate<<<grid_texture,threads_texture>>> (

    	textures_nonoverlap<<<1,1>>> (

    			(int *) 0,             // int             * woi,                // x, y, width,height

                gpu_tasks,             // struct tp_task   * gpu_tasks,

		        gpu_clt ,              // float          ** gpu_clt,            // [NUM_CAMS] ->[TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE]

                tp_task_size,          // int                num_tiles,          // number of tiles in task list

				num_textures,          // size_t            num_texture_tiles,  // number of texture tiles to process

    	// declare arrays in device code?

				// requires initialized gpu_texture_indices

				gpu_texture_indices,   // int             * gpu_texture_indices,// packed tile + bits (now only (1 << 7)

				gpu_texture_indices,   // int             * gpu_texture_indices,// packed tile + bits (now only (1 << 7)

				gpu_num_texture_tiles, // int             * pnum_texture_tiles,  // returns total number of elements in gpu_texture_indices array

		        gpu_clt ,              // float          ** gpu_clt,            // [NUM_CAMS] ->[TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE]

    			// TODO: use geometry_correction rXY !

				gpu_geometry_correction, // struct gc     * gpu_geometry_correction,

				gpu_geometry_correction, // struct gc     * gpu_geometry_correction,

				texture_colors,        // int               colors,             // number of colors (3/1)

				texture_colors,        // int               colors,             // number of colors (3/1)

				(texture_colors == 1), // int               is_lwir,            // do not perform shot correction

				(texture_colors == 1), // int               is_lwir,            // do not perform shot correction
@@ -1091,14 +1093,11 @@ int main(int argc, char **argv) 
				3.0,                   // float             min_agree,          // minimal number of channels to agree on a point (real number to work with fuzzy averages)

				3.0,                   // float             min_agree,          // minimal number of channels to agree on a point (real number to work with fuzzy averages)

				gpu_color_weights,     // float             weights[3],         // scale for R

				gpu_color_weights,     // float             weights[3],         // scale for R

				1,                     // int               dust_remove,        // Do not reduce average weight when only one image differes much from the average

				1,                     // int               dust_remove,        // Do not reduce average weight when only one image differes much from the average

				keep_texture_weights,  // int               keep_weights,       // return channel weights after A in RGBA

    	// combining both non-overlap and overlap (each calculated if pointer is not null )

    	// combining both non-overlap and overlap (each calculated if pointer is not null )

    			0, // const size_t      texture_rbg_stride, // in floats

				0, // dstride_textures/sizeof(float), // size_t            texture_stride,     // in floats (now 256*4 = 1024)  // may be 0 if not needed

    			(float *) 0, // float           * gpu_texture_rbg,     // (number of colors +1 + ?)*16*16 rgba texture tiles

//				gpu_textures,         // float           * gpu_texture_tiles,  // (number of colors +1 + ?)*16*16 rgba texture tiles    // may be 0 if not needed

				dstride_textures/sizeof(float), // const size_t      texture_stride,     // in floats (now 256*4 = 1024)

				(float *) 0,          // gpu_textures,         // float           * gpu_texture_tiles,  // (number of colors +1 + ?)*16*16 rgba texture tiles    // may be 0 if not needed

				gpu_textures,    // float           * gpu_texture_tiles);  // 4*16*16 rgba texture tiles

				gpu_diff_rgb_combo);  // float           * gpu_diff_rgb_combo); // diff[NUM_CAMS], R[NUM_CAMS], B[NUM_CAMS],G[NUM_CAMS] // may be 0 if not needed

				gpu_diff_rgb_combo);             // float           * gpu_diff_rgb_combo) // diff[NUM_CAMS], R[NUM_CAMS], B[NUM_CAMS],G[NUM_CAMS]



    	getLastCudaError("Kernel failure");

    	getLastCudaError("Kernel failure");

    	checkCudaErrors(cudaDeviceSynchronize());

    	checkCudaErrors(cudaDeviceSynchronize());

    	printf("test pass: %d\n",i);

    	printf("test pass: %d\n",i);
@@ -1251,7 +1250,7 @@ int main(int argc, char **argv) 
    		sdkStartTimer(&timerRGBA);

    		sdkStartTimer(&timerRGBA);

    	}

    	}





    	generate_RBGA<<<grid_rgba,threads_rgba>>> (

    	generate_RBGA<<<1,1>>> (

    	// Parameters to generate texture tasks

    	// Parameters to generate texture tasks

                gpu_tasks,             // struct tp_task   * gpu_tasks,

                gpu_tasks,             // struct tp_task   * gpu_tasks,

                tp_task_size,          // int                num_tiles,          // number of tiles in task list

                tp_task_size,          // int                num_tiles,          // number of tiles in task list
@@ -1276,7 +1275,7 @@ int main(int argc, char **argv) 
	            0,                     // int               keep_weights,       // return channel weights after A in RGBA

	            0,                     // int               keep_weights,       // return channel weights after A in RGBA

				dstride_textures_rbga/sizeof(float), // 	const size_t      texture_rbga_stride,     // in floats

				dstride_textures_rbga/sizeof(float), // 	const size_t      texture_rbga_stride,     // in floats

				gpu_textures_rbga,     // 	float           * gpu_texture_tiles)    // (number of colors +1 + ?)*16*16 rgba texture tiles

				gpu_textures_rbga,     // 	float           * gpu_texture_tiles)    // (number of colors +1 + ?)*16*16 rgba texture tiles

				gpu_diff_rgb_combo);   // float           * gpu_diff_rgb_combo) // diff[NUM_CAMS], R[NUM_CAMS], B[NUM_CAMS],G[NUM_CAMS]

				(float *) 0 ); // gpu_diff_rgb_combo);   // float           * gpu_diff_rgb_combo) // diff[NUM_CAMS], R[NUM_CAMS], B[NUM_CAMS],G[NUM_CAMS]





    	getLastCudaError("Kernel failure");

    	getLastCudaError("Kernel failure");

    	checkCudaErrors(cudaDeviceSynchronize());

    	checkCudaErrors(cudaDeviceSynchronize());

src/tp_defines.h

+5 −1
Original line number Original line Diff line number Diff line
@@ -114,8 +114,12 @@ 
// geom

// geom

//#define DEBUG20 1

//#define DEBUG20 1





#define DEBUG21 1

// #define DEBUG21 1 // Geometry Correction

#if (DBG_TILE_X >= 0) && (DBG_TILE_Y >= 0)





#define DEBUG22 1



#endif //#if (DBG_TILE_X >= 0) && (DBG_TILE_Y >= 0)





#endif //#ifndef JCUDA

#endif //#ifndef JCUDA