Using Marco's help to make jcuda work with nvrtc with cdp

src/main/java/com/elphel/imagej/cameras/CLTParameters.java

@@ -333,7 +333,7 @@ public class CLTParameters {
 	public double     infinityDistance  = 10000;  // Distance to generate backdrop (0 - use regular backdrop)
 	public int        min_bgnd_tiles    = 10;     // Minimal number of background tiles to generate background
 	public boolean    shUseFlaps        = true;  // Split into shells with flaps
-	public boolean    shAggrFade        = true;  // Aggressive fade alpha (whole boundary)
+	public boolean    shAggrFade        = false; // true;  // Aggressive fade alpha (whole boundary)
 	public int        shMinArea         = 1;     // Minimal shell area (not counting flaps
 	public double     shMinStrength    = 0.2;   // Minimal value of the shell maximum strength
 

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

@@ -3867,7 +3867,9 @@ public class ImageDtt {
 
 		double [][] port_weights = new double[ports][tile_len];
 		double [][] color_avg =    new double[numcol][tile_len];
-		double [][] rgba = new double[numcol + 1 + (keep_weights?(ports + numcol + 1):0)][];
+//		double [][] rgba = new double[numcol + 1 + (keep_weights?(ports + numcol + 1):0)][];
+// need to pass keep_weights to the caller
+		double [][] rgba = new double[numcol + 1 + ports + (keep_weights?(numcol + 1):0)][];
 		int rms_start = numcol + 1 + ports;
 		if (keep_weights){
 			for (int ncol = 0; ncol <= numcol ; ncol++) if ((ncol == numcol) || (iclt_tile[0][ncol] != null)) {
@@ -4069,8 +4071,11 @@ public class ImageDtt {
 			rgba[ncol] = color_avg[ncol];
 		}
 		rgba[numcol] = alpha;
-
-		for (int i = 0; i < ports; i++)  rgba[numcol + 1 + i] = port_weights[i];
+//		if (keep_weights){
+			for (int i = 0; i < ports; i++) {
+				rgba[numcol + 1 + i] = port_weights[i];
+			}
+//		}
 		if (max_diff != null){
 			for (int ip = 0; ip < ports; ip++){
 				max_diff[ip] = 0;
@@ -5026,7 +5031,7 @@ public class ImageDtt {
 			System.out.println("iclt_2d():sharp_alpha=   "+sharp_alpha);
 		}
 		boolean has_weights = false;
-		boolean set_has_weight = false;
+		boolean set_has_weight = false; // not used
 		for (int i = 0; (i < tilesY) && !set_has_weight; i++){
 			for (int j = 0; (j < tilesX) && !set_has_weight; j++){
 				if (texture_tiles[i][j] != null) {
@@ -5064,15 +5069,15 @@ public class ImageDtt {
 						int n_half = transform_size / 2;
 						int lastY = tilesY-1;
 						int lastX = tilesX-1;
-						int offset = n_half * (transform_size * tilesX) + n_half;
+						int offset = n_half * (transform_size * tilesX) + n_half; // 4 pixels left and down (right/up when subtracted below)
 						for (int nTile = ai.getAndIncrement(); nTile < tiles_list[nser.get()].length; nTile = ai.getAndIncrement()) {
 							tileX = tiles_list[nser.get()][nTile][0];
 							tileY = tiles_list[nser.get()][nTile][1];
 							double [][] texture_tile =texture_tiles[tileY][tileX];
 							if (texture_tile != null) {
 								if (overlap) {
-									if ((tileY >0) && (tileX > 0) && (tileY < lastY) && (tileX < lastX)) { // fast, no extra checks
-										for (int i = 0; i < n2;i++){
+									if ((tileY >0) && (tileX > 0) && (tileY < lastY) && (tileX < lastX)) { // fast, no extra checks - ignore first/last rows and columns
+										for (int i = 0; i < n2; i++){
 											int start_line = ((tileY*transform_size + i) * tilesX + tileX)*transform_size - offset;
 											for (int chn = 0; chn < texture_tile.length; chn++) {
 												int schn = chn;
@@ -5082,6 +5087,7 @@ public class ImageDtt {
 												if (texture_tile[schn] == null) {
 													dpixels[chn] = null;
 												} else {
+													// should it be better to multiply each color by alpha before accumulating? No, it is already windowed!
 													if ((chn != 3) || !sharp_alpha) {
 														for (int j = 0; j<n2;j++) {
 															dpixels[chn][start_line + j] += texture_tile[schn][n2 * i + j];
@@ -9278,6 +9284,14 @@ public class ImageDtt {
 				}
 			}
 		}
+		// fix: removing extra slices
+		if (!clt_parameters.keep_weights && (texture_tiles[tileY][tileX]!=null)) {
+			if (numcol == 3 ) {
+				texture_tiles[tileY][tileX] = new double[][] {texture_tiles[tileY][tileX][0],texture_tiles[tileY][tileX][1],texture_tiles[tileY][tileX][2],texture_tiles[tileY][tileX][3]};
+			} else {
+				texture_tiles[tileY][tileX] = new double[][] {texture_tiles[tileY][tileX][0],texture_tiles[tileY][tileX][1]};
+			}
+		}
 	}
 
 	public double [][] get2DCorrs(

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

@@ -8985,6 +8985,7 @@ public class QuadCLT {
 			  tp.clt_3d_passes.add(latest_scan); // put it back
 		  }
 		  int next_pass = tp.clt_3d_passes.size(); //
+		  // Create tasks to scan, have tasks, disparity and border tiles in tp.clt_3d_passes
 		  tp.thirdPassSetupSurf( // prepare tile tasks for the second pass based on the previous one(s) // needs last scan
 				  clt_parameters,
 				  //FIXME: make a special parameter?
@@ -9564,6 +9565,7 @@ public class QuadCLT {
 		  double [] alpha_zero = new double [4*image_dtt.transform_size*image_dtt.transform_size];
 		  int alpha_index = 3;
 		  for (int i = 0; i < alpha_zero.length; i++) alpha_zero[i]=0.0;
+		  // border tiles are copied, alpha from alphaFade (not multiplied?)
 		  for (int tileY = 0; tileY < tilesY; tileY++){
 			  for (int tileX = 0; tileX < tilesX; tileX++){
 				  texture_tiles_cluster[tileY][tileX]= null;
@@ -9607,7 +9609,7 @@ public class QuadCLT {
 				  texture_overlap[alpha_index][i] = d;
 			  }
 		  }
-		  // for now - use just RGB. Later add option for RGBA
+		  // for now - use just RGB. Later add option for RGBA (?)
 		  double [][] texture_rgb = {texture_overlap[0],texture_overlap[1],texture_overlap[2]};
 		  double [][] texture_rgba = {texture_overlap[0],texture_overlap[1],texture_overlap[2],texture_overlap[3]};
 		  double [][] texture_rgbx = ((clt_parameters.alpha1 > 0)? texture_rgba: texture_rgb);

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

@@ -2191,7 +2191,9 @@ public class TwoQuadCLT {
 
 		}
 		if (clt_parameters.show_rgba_color) {
-			int          num_src_slices = 12 ; // calculate
+			int numcol = quadCLT_main.isMonochrome()?1:3;
+			int ports = imp_quad_main.length;
+			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);
@@ -3533,7 +3535,7 @@ if (debugLevel > -100) return true; // temporarily !
 			}
 		}
 
-		quadCLT_main.writeKml(debugLevel ); // alos generated with x3d model
+		quadCLT_main.writeKml(debugLevel ); // also generated with x3d model
 
 		String jp4_copy_path= quadCLT_main.correctionsParameters.selectX3dDirectory(
 				set_name, // quad timestamp. Will be ignored if correctionsParameters.use_x3d_subdirs is false

src/main/java/com/elphel/imagej/x3d/export/X3dOutput.java

@@ -69,7 +69,7 @@ public class X3dOutput {
 			CLTParameters                                   clt_parameters,
 			EyesisCorrectionParameters.CorrectionParameters correctionsParameters,
 			GeometryCorrection                              geometry_correction,
-			ArrayList <CLTPass3d>                           clt_3d_passes){
+			ArrayList <CLTPass3d>                           clt_3d_passes){ // to scan for textures, contain disp, tasks, border tiles
 		this.clt_parameters =        clt_parameters;
 		this.correctionsParameters = correctionsParameters;
 		this.geometry_correction =    geometry_correction;