Debugged monochrome direct/inverse mclt conversions

src/dtt8x8.cu

@@ -951,6 +951,7 @@ __device__ void dttiv_color_2d(
     dctiv_nodiverg( // all colors
 			clt_tile + (DTT_SIZE1 * threadIdx.x), // [0][threadIdx.x], // pointer to start of row
 			1); //int inc);
+//	__syncthreads();// worsened
     if (color == BAYER_GREEN){
         dstiv_nodiverg( // all colors
 				clt_tile + DTT_SIZE1 * threadIdx.x + DTT_SIZE1 * DTT_SIZE, // clt_tile[1][threadIdx.x], // pointer to start of row
@@ -969,6 +970,7 @@ __device__ void dttiv_color_2d(
     dctiv_nodiverg( // all colors
     		clt_tile + threadIdx.x, //  &clt_tile[0][0][threadIdx.x], // pointer to start of column
 			DTT_SIZE1); // int inc,
+//	__syncthreads();// worsened
     if (color == BAYER_GREEN){
           dctiv_nodiverg( // all colors
         		clt_tile + threadIdx.x + (DTT_SIZE1 * DTT_SIZE), // &clt_tile[1][0][threadIdx.x], // pointer to start of column
@@ -977,6 +979,50 @@ __device__ void dttiv_color_2d(
   	 __syncthreads();// __syncwarp();
 }
 
+__device__ void dttiv_mono_2d(
+		float * clt_tile)
+{
+	// Copy 0-> 1
+
+    dctiv_nodiverg(
+			clt_tile + (DTT_SIZE1 * threadIdx.x) + (0 * DTT_SIZE1 * DTT_SIZE),
+			1); //int inc);
+    dstiv_nodiverg(
+    		clt_tile + (DTT_SIZE1 * threadIdx.x) + (1 * DTT_SIZE1 * DTT_SIZE),
+			1); //int inc);
+    dctiv_nodiverg(
+			clt_tile + (DTT_SIZE1 * threadIdx.x) + (2 * DTT_SIZE1 * DTT_SIZE),
+			1); //int inc);
+    dstiv_nodiverg(
+    		clt_tile + (DTT_SIZE1 * threadIdx.x) + (3 * DTT_SIZE1 * DTT_SIZE),
+			1); //int inc);
+	__syncthreads();// __syncwarp();
+
+#ifdef DEBUG222
+    if ((threadIdx.x) == 0){
+        printf("\nDTT Tiles after horizontal pass, color=%d\n",color);
+    	debug_print_clt1(clt_tile, color, (color== BAYER_GREEN)?3:1); // only 1 quadrant for R,B and 2 - for G
+    }
+     __syncthreads();// __syncwarp();
+#endif
+
+ 	dctiv_nodiverg( // CC
+ 			clt_tile + threadIdx.x,
+			DTT_SIZE1); // int inc,
+ 	dctiv_nodiverg( // SC
+ 			clt_tile + threadIdx.x + 1 * (DTT_SIZE1 * DTT_SIZE),
+			DTT_SIZE1); // int inc,
+ 	dstiv_nodiverg( // CS
+ 			clt_tile + threadIdx.x + 2 * (DTT_SIZE1 * DTT_SIZE), // &clt_tile[1][0][threadIdx.x], // pointer to start of column
+			DTT_SIZE1); // int inc,
+ 	dstiv_nodiverg( // SS
+ 			clt_tile + threadIdx.x + 3 * (DTT_SIZE1 * DTT_SIZE), // &clt_tile[1][0][threadIdx.x], // pointer to start of column
+			DTT_SIZE1); // int inc,
+  	 __syncthreads();// __syncwarp();
+}
+
+
+
 //
 // Uses 16 threads, gets 4*8*8 clt tiles, performs idtt-iv (swapping 1 and 2 quadrants) and then unfolds with window,
 // adding to the output 16x16 tile (to use Read-modify-write with 4 passes over the frame. Should be zeroed before the

src/dtt8x8.h

@@ -88,6 +88,8 @@ extern __device__ void dttii_2d(
 extern __device__ void dttiv_color_2d(
 		float * clt_tile,
 		int color);
+extern __device__ void dttiv_mono_2d(
+		float * clt_tile);
 extern __device__ void imclt(
 		float * clt_tile,   //        [4][DTT_SIZE][DTT_SIZE1], // +1 to alternate column ports [4][8][9]
 		float * mclt_tile );

src/test_tp.cu

@@ -34,7 +34,7 @@
 #define NOCORR_TD
 #define NOTEXTURES
 #define NOTEXTURE_RGBA
-
+#define SAVE_CLT
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -330,7 +330,7 @@ int main(int argc, char **argv)
 			"/home/eyesis/git/tile_processor_gpu/clt/aux_chn15.portsxy"};
 
 //#ifndef DBG_TILE
-    /*
+#ifdef SAVE_CLT
     const char* ports_clt_file[] = { // never referenced
     		"/home/eyesis/git/tile_processor_gpu/clt/aux_chn0.clt",
 			"/home/eyesis/git/tile_processor_gpu/clt/aux_chn1.clt",
@@ -348,7 +348,8 @@ int main(int argc, char **argv)
 			"/home/eyesis/git/tile_processor_gpu/clt/aux_chn13.clt",
 			"/home/eyesis/git/tile_processor_gpu/clt/aux_chn14.clt",
 			"/home/eyesis/git/tile_processor_gpu/clt/aux_chn15.clt"};
-*/
+
+#endif
     const char* result_rbg_file[] = {
     		"/home/eyesis/git/tile_processor_gpu/clt/aux_chn0.rbg",
 			"/home/eyesis/git/tile_processor_gpu/clt/aux_chn1.rbg",
@@ -401,15 +402,13 @@ int main(int argc, char **argv)
 			"/home/eyesis/git/tile_processor_gpu/clt/main_chn1.portsxy",
 			"/home/eyesis/git/tile_processor_gpu/clt/main_chn2.portsxy",
 			"/home/eyesis/git/tile_processor_gpu/clt/main_chn3.portsxy"};
-
-//#ifndef DBG_TILE
-    /*
+#ifdef SAVE_CLT
     const char* ports_clt_file[] = { // never referenced
     		"/home/eyesis/git/tile_processor_gpu/clt/main_chn0.clt",
 			"/home/eyesis/git/tile_processor_gpu/clt/main_chn1.clt",
 			"/home/eyesis/git/tile_processor_gpu/clt/main_chn2.clt",
 			"/home/eyesis/git/tile_processor_gpu/clt/main_chn3.clt"};
-*/
+#endif
     const char* result_rbg_file[] = {
     		"/home/eyesis/git/tile_processor_gpu/clt/main_chn0.rbg",
 			"/home/eyesis/git/tile_processor_gpu/clt/main_chn1.rbg",
@@ -1133,12 +1132,12 @@ int main(int argc, char **argv)
 				gpu_clt_h[ncam],
 				rslt_size * sizeof(float),
     			cudaMemcpyDeviceToHost));
-#ifndef DBG_TILE
+//#ifndef DBG_TILE
         printf("Writing CLT data to %s\n",  ports_clt_file[ncam]);
     	writeFloatsToFile(cpu_clt, // float *       data, // allocated array
     			rslt_size, // int           size, // length in elements
 				ports_clt_file[ncam]); // 			   const char *  path) // file path
-#endif
+//#endif
     }
 #endif
 

src/tp_defines.h

@@ -140,11 +140,12 @@
 // geom
 //#define DEBUG20 1
 
+
 #if (DBG_TILE_X >= 0) && (DBG_TILE_Y >= 0)
 #define DEBUG20 1 // Geometry Correction
 #define DEBUG21 1 // Geometry Correction
 //#define DEBUG210 1
-
+#define DEBUG30 1
 //#define DEBUG22 1
 //#define DEBUG23 1