debugging code to generate data for macroblocks correlation

src/main/java/com/elphel/imagej/tileprocessor/TwoQuadCLT.java

@@ -2083,10 +2083,12 @@ public class TwoQuadCLT {
 		// corr_indices array of integers to be passed to GPU
 //		gPUTileProcessor.setCorrIndices(corr_indices);
 
+/*
 		int [] texture_indices = gPUTileProcessor.getTextureTasks(
 				tp_tasks);
 		gPUTileProcessor.setTextureIndices(
 				texture_indices);
+*/
 		gPUTileProcessor.setGeometryCorrection(
 				quadCLT_main.getGeometryCorrection(),
 				false); // boolean use_java_rByRDist) { // false - use newer GPU execCalcReverseDistortions); // once
@@ -2146,12 +2148,11 @@ public class TwoQuadCLT {
 				clt_parameters.diff_sigma,     // double    diff_sigma,         // pixel value/pixel change
 				clt_parameters.diff_threshold, // double    diff_threshold,     // pixel value/pixel change
 				clt_parameters.min_agree,      // double    min_agree,          // minimal number of channels to agree on a point (real number to work with fuzzy averages)
-				clt_parameters.dust_remove,    // boolean   dust_remove,
-				clt_parameters.keep_weights);  // boolean   keep_weights);
+				clt_parameters.dust_remove);     // boolean   dust_remove,
 		long endTextures = System.nanoTime();
-
 // run texturesRBGA
 		long startTexturesRBGA = System.nanoTime();   // System.nanoTime();
+
 		for (int i = 0; i < NREPEAT; i++ ) gPUTileProcessor.execRBGA(
 				col_weights,                   // double [] color_weights,
 				quadCLT_main.isLwir(),         // boolean   is_lwir,
@@ -2193,7 +2194,26 @@ public class TwoQuadCLT {
 
 		int out_width =  GPUTileProcessor.IMG_WIDTH +  GPUTileProcessor.DTT_SIZE;
 		int out_height = GPUTileProcessor.IMG_HEIGHT + GPUTileProcessor.DTT_SIZE;
-
+		int tilesX =  GPUTileProcessor.IMG_WIDTH / GPUTileProcessor.DTT_SIZE;
+		int tilesY =  GPUTileProcessor.IMG_HEIGHT / GPUTileProcessor.DTT_SIZE;
+		// show extra
+		/* */
+		String [] extra_group_titles = {"DIFF","Red","Blue","Green"};
+		String [] extra_titles = new String [extra_group_titles.length*GPUTileProcessor.NUM_CAMS];
+		for (int g = 0; g < extra_group_titles.length;g++) {
+			for (int ncam=0; ncam < GPUTileProcessor.NUM_CAMS;ncam++) {
+				extra_titles[g * GPUTileProcessor.NUM_CAMS+ncam]= extra_group_titles[g]+"-"+ncam;
+			}
+		}
+		float [][] extra = gPUTileProcessor.getExtra();
+		(new ShowDoubleFloatArrays()).showArrays(
+				extra,
+				tilesX,
+				tilesY,
+				true,
+				name+"-EXTRA-D"+clt_parameters.disparity,
+				extra_titles);
+		/* */
 		ImagePlus [] imps_RGB = new ImagePlus[iclt_fimg.length];
 		for (int ncam = 0; ncam < iclt_fimg.length; ncam++) {
 			String title=name+"-"+String.format("%02d", ncam);
@@ -2215,8 +2235,6 @@ public class TwoQuadCLT {
 
 		}
 		//show_corr
-		int tilesX =  GPUTileProcessor.IMG_WIDTH / GPUTileProcessor.DTT_SIZE;
-		int tilesY =  GPUTileProcessor.IMG_HEIGHT / GPUTileProcessor.DTT_SIZE;
 		int [] wh = new int[2];
 		if (clt_parameters.show_corr) {
 			int [] corr_indices = gPUTileProcessor.getCorrIndices();
@@ -2234,7 +2252,7 @@ public class TwoQuadCLT {
 					wh[0],
 					wh[1],
 					true,
-					"CORR2D",
+					name+"-CORR2D-D"+clt_parameters.disparity,
 					GPUTileProcessor.getCorrTitles());
 		}
 // convert to overlapping and show
@@ -2369,14 +2387,16 @@ public class TwoQuadCLT {
 		}
 
 		// convert textures to RGBA in Java
-		if (clt_parameters.show_rgba_color && (debugLevel > 0)) { // disabling
+		if (clt_parameters.show_rgba_color && (debugLevel > 100)) { // disabling
 			int numcol = quadCLT_main.isMonochrome()?1:3;
 			int ports = imp_quad_main.length;
+			int [] texture_indices = gPUTileProcessor.getTextureIndices();
 			int          num_src_slices = numcol + 1 + (clt_parameters.keep_weights?(ports + numcol + 1):0); // 12 ; // calculate
 //			float [][][] ftextures = gPUTileProcessor.getTextures(
 //		    		(is_mono?1:3), // int     num_colors,
 //		    		clt_parameters.keep_weights); // boolean keep_weights);
 			float [] flat_textures =  gPUTileProcessor.getFlatTextures(
+					texture_indices.length,
 		    		(is_mono?1:3), // int     num_colors,
 		    		clt_parameters.keep_weights); // boolean keep_weights);
 	    	int texture_slice_size = (2 * GPUTileProcessor.DTT_SIZE)* (2 * GPUTileProcessor.DTT_SIZE);

src/main/resources/kernels/TileProcessor.h

@@ -75,11 +75,13 @@ extern "C" __global__ void correlate2D(
 		float           * gpu_corrs);          // correlation output data
 
 
-extern "C" __global__ void textures_accumulate(
-		int             * woi,                // x, y, width,height
-		float          ** gpu_clt,            // [NUM_CAMS] ->[TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE]
-		size_t            num_texture_tiles,  // number of texture tiles to process
+extern "C" __global__ void textures_nonoverlap(
+		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             * 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 !
 		struct gc       * gpu_geometry_correction,
 		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             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               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 )
-		size_t            texture_rbg_stride, // in floats
-		float           * gpu_texture_rbg,    // (number of colors +1 + ?)*16*16 rgba texture tiles
-		size_t            texture_stride,     // in floats (now 256*4 = 1024)
-		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]
-
+		size_t            texture_stride,     // in floats (now 256*4 = 1024)  // may be 0 if not needed
+		float           * gpu_texture_tiles,  // (number of colors +1 + ?)*16*16 rgba texture tiles    // may be 0 if not needed
+		float           * gpu_diff_rgb_combo); // diff[NUM_CAMS], R[NUM_CAMS], B[NUM_CAMS],G[NUM_CAMS] // may be 0 if not needed
 
 extern "C"
 __global__ void imclt_rbg_all(