More editing to make dynamic number of cameras

src/TileProcessor.h

@@ -104,8 +104,9 @@ 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_cams,           // number of cameras
+		float            * gpu_ftasks,         // flattened tasks, 27 floats for quad EO, 99 floats
+		//		struct tp_task  * gpu_tasks,
 		int               num_tiles,          // number of tiles in task list
 //		int               num_tilesx,         // number of tiles in a row
 // declare arrays in device code?
@@ -151,7 +152,8 @@ extern "C" __global__ void imclt_rbg(
 extern "C" __global__ void generate_RBGA(
 		int                num_cams,           // number of cameras used
 		// Parameters to generate texture tasks
-		struct tp_task   * gpu_tasks,
+		float            * gpu_ftasks,         // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
+//		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)

src/geometry_correction.h

@@ -41,6 +41,8 @@
 #include "tp_defines.h"
 #endif
 
+#define get_task_size(x) (sizeof(struct tp_task)/sizeof(float) - 6 * (NUM_CAMS - x))
+
 #define NVRTC_BUG 1
 #ifndef M_PI
 #define M_PI  3.14159265358979323846 /* pi */
@@ -60,8 +62,9 @@ struct tp_task {
 		int      txy;
 		unsigned short sxy[2];
 	};
-	float xy[NUM_CAMS][2];
 	float target_disparity;
+	float xy[NUM_CAMS][2];
+//	float target_disparity;
 	float disp_dist[NUM_CAMS][4]; // calculated with getPortsCoordinates()
 };
 
@@ -142,7 +145,9 @@ struct gc {
 };
 #define RAD_COEFF_LEN 7
 extern "C" __global__ void get_tiles_offsets(
-		struct tp_task     * gpu_tasks,
+		int                  num_cams,
+//		struct tp_task     * gpu_tasks,
+		float              * gpu_ftasks,         // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
 		int                  num_tiles,          // number of tiles in task
 		struct gc          * gpu_geometry_correction,
 		struct corr_vector * gpu_correction_vector,
@@ -150,7 +155,9 @@ extern "C" __global__ void get_tiles_offsets(
 		trot_deriv   * gpu_rot_deriv);
 
 extern "C" __global__ void calculate_tiles_offsets(
-		struct tp_task     * gpu_tasks,
+		int                  num_cams,
+		float              * gpu_ftasks,         // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
+//		struct tp_task     * gpu_tasks,
 		int                  num_tiles,          // number of tiles in task
 		struct gc          * gpu_geometry_correction,
 		struct corr_vector * gpu_correction_vector,
@@ -160,6 +167,7 @@ extern "C" __global__ void calculate_tiles_offsets(
 
 // uses NUM_CAMS blocks, (3,3,3) threads
 extern "C" __global__ void calc_rot_deriv(
+		int                  num_cams,
 		struct corr_vector * gpu_correction_vector,
 		trot_deriv   * gpu_rot_deriv);
 
@@ -170,3 +178,4 @@ extern "C" __global__ void calcReverseDistortionTable(
 		float * rByRDist);
 
 
+

src/test_tp.cu

@@ -715,6 +715,7 @@ int main(int argc, char **argv)
     	}
 
     	calc_rot_deriv<<<grid_rot,threads_rot>>> (
+    			num_cams,                // int                  num_cams,
     			gpu_correction_vector ,           // 		struct corr_vector * gpu_correction_vector,
     			gpu_rot_deriv);                  // union trot_deriv   * gpu_rot_deriv);
 
@@ -821,6 +822,7 @@ int main(int argc, char **argv)
     	}
 /*
     	get_tiles_offsets<<<grid_geom,threads_geom>>> (
+    			num_cams,                // int                  num_cams,
     			gpu_tasks,               // struct tp_task     * gpu_tasks,
 				tp_task_size,            // int                  num_tiles,          // number of tiles in task list
 				gpu_geometry_correction, //	struct gc          * gpu_geometry_correction,
@@ -829,7 +831,9 @@ int main(int argc, char **argv)
 				gpu_rot_deriv);          // union trot_deriv   * gpu_rot_deriv);
 				*/
     	calculate_tiles_offsets<<<1,1>>> (
-    			gpu_tasks,               // struct tp_task     * gpu_tasks,
+    			num_cams,                // int                  num_cams,
+				gpu_ftasks,              // float              * gpu_ftasks,         // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
+//    			gpu_tasks,               // struct tp_task     * gpu_tasks,
 				tp_task_size,            // int                  num_tiles,          // number of tiles in task list
 				gpu_geometry_correction, //	struct gc          * gpu_geometry_correction,
 				gpu_correction_vector,   //	struct corr_vector * gpu_correction_vector,
@@ -1273,7 +1277,8 @@ int main(int argc, char **argv)
 		cudaFuncSetAttribute(textures_nonoverlap, cudaFuncAttributeMaxDynamicSharedMemorySize, 65536); // for CC 7.5
     	textures_nonoverlap<<<1,1>>> (
     			num_cams,              // int                num_cams,           // number of cameras used
-                gpu_tasks,             // struct tp_task   * gpu_tasks,
+				gpu_ftasks,          // float            * gpu_ftasks,         // flattened tasks, 27 floats for quad EO, 99 floats
+//                gpu_tasks,             // struct tp_task   * gpu_tasks,
                 tp_task_size,          // int                num_tiles,          // number of tiles in task list
     	// declare arrays in device code?
 				gpu_texture_indices,   // int             * gpu_texture_indices,// packed tile + bits (now only (1 << 7)
@@ -1365,7 +1370,8 @@ int main(int argc, char **argv)
     	generate_RBGA<<<1,1>>> (
     			num_cams,              // int                num_cams,           // number of cameras used
     	// Parameters to generate texture tasks
-                gpu_tasks,             // struct tp_task   * gpu_tasks,
+				gpu_ftasks,         // float            * gpu_ftasks,         // flattened tasks, 27 floats for quad EO, 99 floats for LWIR16
+//                gpu_tasks,             // struct tp_task   * gpu_tasks,
                 tp_task_size,          // int                num_tiles,          // number of tiles in task list
 		// Does not require initialized gpu_texture_indices to be initialized - just allocated, will generate.
 	            gpu_texture_indices,   // int              * gpu_texture_indices,// packed tile + bits (now only (1 << 7)

src/tp_defines.h

@@ -56,7 +56,8 @@
 #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               32 //
+#define NUM_THREADS                   32
 #define TEXTURE_THREADS_PER_TILE       8
 #define TEXTURE_TILES_PER_BLOCK        1
 #define IMCLT_THREADS_PER_TILE        16