Commit ee0cfc3b authored by Andrey Filippov's avatar Andrey Filippov
Browse files

converting to varaible num_cams

parent 18d8e56b

src/TileProcessor.cuh

+725 −362

File changed.

Preview size limit exceeded, changes collapsed.

src/TileProcessor.h

+11 −3
Original line number Diff line number Diff line
@@ -44,10 +44,13 @@ 


extern "C" __global__ void convert_direct( // called with a single block, single thread

		//		struct CltExtra ** gpu_kernel_offsets, // [NUM_CAMS], // changed for jcuda to avoid struct parameters

		int                num_cams,           // actual number of cameras

		int                num_colors,         // actual number of colors: 3 for RGB, 1 for LWIR/mono

		float           ** gpu_kernel_offsets, // [NUM_CAMS],

		float           ** gpu_kernels,        // [NUM_CAMS],

		float           ** gpu_images,         // [NUM_CAMS],

		struct tp_task   * gpu_tasks,

		float            * gpu_ftasks,         // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16

//		struct tp_task   * gpu_tasks,

		float           ** gpu_clt,            // [NUM_CAMS][TILES-Y][TILES-X][NUM_COLORS][DTT_SIZE*DTT_SIZE]

		size_t             dstride,            // in floats (pixels)

		int                num_tiles,          // number of tiles in task
@@ -60,15 +63,17 @@ extern "C" __global__ void convert_direct( // called with a single block, single 
		int *              pnum_active_tiles,  //  indices to gpu_tasks

		int                tilesx);





extern "C" __global__ void correlate2D(

		int               num_cams,

		int *             sel_pairs,

		float          ** gpu_clt,            // [NUM_CAMS] ->[TILES-Y][TILES-X][NUM_COLORS][DTT_SIZE*DTT_SIZE]

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

		float             scale0,             // scale for R

		float             scale1,             // scale for B

		float             scale2,             // scale for G

		float             fat_zero,           // here - absolute

		struct tp_task  * gpu_tasks,          // array of per-tile tasks (now bits 4..9 - correlation pairs)

		float            * gpu_ftasks,         // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16

//		struct tp_task  * gpu_tasks,          // array of per-tile tasks (now bits 4..9 - correlation pairs)

		int               num_tiles,          // number of tiles in task

		int               tilesx,             // number of tile rows

		int             * gpu_corr_indices,   // packed tile+pair
@@ -99,6 +104,7 @@ extern "C" __global__ void corr2D_combine( 
		float           * gpu_corrs_combo);   // combined correlation output (one per tile)



extern "C" __global__ void textures_nonoverlap(

		int                num_cams,           // number of cameras used

		struct tp_task  * gpu_tasks,

		int               num_tiles,          // number of tiles in task list

//		int               num_tilesx,         // number of tiles in a row
@@ -121,6 +127,7 @@ extern "C" __global__ void textures_nonoverlap( 


extern "C"

__global__ void imclt_rbg_all(

		int                num_cams,

		float           ** gpu_clt,            // [NUM_CAMS][TILES-Y][TILES-X][NUM_COLORS][DTT_SIZE*DTT_SIZE]

		float           ** gpu_corr_images,    // [NUM_CAMS][WIDTH, 3 * HEIGHT]

		int                apply_lpf,
@@ -142,6 +149,7 @@ extern "C" __global__ void imclt_rbg( 
		const size_t      dstride);            // in floats (pixels)



extern "C" __global__ void generate_RBGA(

		int                num_cams,           // number of cameras used

		// Parameters to generate texture tasks

		struct tp_task   * gpu_tasks,

		int                num_tiles,          // number of tiles in task list

src/test_tp.cu

+194 −68

File changed.

Preview size limit exceeded, changes collapsed.

src/tp_defines.h

+5 −2
Original line number Diff line number Diff line
@@ -41,9 +41,10 @@ 
#ifndef JCUDA

#include <stdio.h>

#define THREADSX              (DTT_SIZE)

#define NUM_CAMS                       4

#define TEST_LWIR                      1

#define NUM_CAMS                      16 // now maximal number of cameras

#define NUM_PAIRS                      6

#define NUM_COLORS                     3

#define NUM_COLORS                     1 //3

#define IMG_WIDTH                   2592

#define IMG_HEIGHT                  1936

#define KERNELS_HOR                  164
@@ -55,11 +56,13 @@ 
#define CORR_TILES_PER_BLOCK           4

#define CORR_TILES_PER_BLOCK_NORMALIZE 4 // increase to 8?

#define CORR_TILES_PER_BLOCK_COMBINE   4 // increase to 16?

#define TEXTURE_THREADS               32 //

#define TEXTURE_THREADS_PER_TILE       8

#define TEXTURE_TILES_PER_BLOCK        1

#define IMCLT_THREADS_PER_TILE        16

#define IMCLT_TILES_PER_BLOCK          4

#define CORR_NTILE_SHIFT               8 // higher bits - number of a pair, other bits tile number

// only lower bit will be used to request correlations, correlation mask will be common for all the scene

#define CORR_PAIRS_MASK             0x3f// lower bits used to address correlation pair for the selected tile

#define CORR_TEXTURE_BIT               7 // bit 7 used to request texture for the tile

#define TASK_CORR_BITS                 4