testCorrelate2DIntraTD

src/TpHostGpu.h

@@ -129,8 +129,11 @@ public:
 //	void testImclt               (int num_runs); // 682 // not implemented
 	void testImcltRbgAll         (int num_runs); // 701
 
-	void testCorrelate2DIntra    (int num_runs);
-	void testCorrelate2DInterSelf(int num_runs);
+	void testCorrelate2DIntra    (int num_runs); // 762 - 885
+	void testCorrelate2DInterSelf(int num_runs); // 1136 - 1411
+	void testCorrelate2DIntraTD  (int num_runs, int quad_combine); // 886 - 1123
+
+
 
 	void saveClt(const char ** paths,  const char * prompt, float **     gpu_clt_h);
 	void saveRgb(const char ** paths,  const char * prompt, float **     gpu_corr_images_h);

src/test_tp.cu

@@ -32,7 +32,7 @@
 
 // all of the next 5 were disabled
 //#define NOCORR
-#define NOCORR_TD
+//#define NOCORR_TD
 #define NOTEXTURES //
 #define NOTEXTURE_RGBA //
 //#define NOTEXTURE_RGBAXXX //
@@ -883,10 +883,8 @@ int main(int argc, char **argv)
     free (corr_img);
 #endif // ifndef NOCORR
 
-
-
-
 #ifndef NOCORR_TD
+//#define QUAD_COMBINE
 //    cudaProfilerStart();
 // testing corr
     StopWatchInterface *timerCORRTD = 0;
@@ -903,10 +901,10 @@ int main(int argc, char **argv)
     	// FIXME: provide sel_pairs
         correlate2D<<<1,1>>>( // output TD tiles, no normalization
         		tpParams.num_cams,                      // int               num_cams,
-				TpParams.sel_pairs[0], // int               sel_pairs0           // unused bits should be 0
-				TpParams.sel_pairs[1], // int               sel_pairs1,           // unused bits should be 0
-				TpParams.sel_pairs[2], // int               sel_pairs2,           // unused bits should be 0
-				TpParams.sel_pairs[3], // int               sel_pairs3,           // unused bits should be 0
+				tpParams.sel_pairs[0], // int               sel_pairs0           // unused bits should be 0
+				tpParams.sel_pairs[1], // int               sel_pairs1,           // unused bits should be 0
+				tpParams.sel_pairs[2], // int               sel_pairs2,           // unused bits should be 0
+				tpParams.sel_pairs[3], // int               sel_pairs3,           // unused bits should be 0
         		gpu_clt,                       // float          ** gpu_clt,            // [NUM_CAMS] ->[TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE]
 				tpParams.num_colors,                    // int               colors,             // number of colors (3/1)
 				tpParams.color_weights[0], // 0.25,     // float             scale0,             // scale for R
@@ -1019,20 +1017,18 @@ int main(int argc, char **argv)
 //    int rslt_corr_size =   num_corrs * corr_size * corr_size;
 //    float * cpu_corr = (float *)malloc(rslt_corr_size * sizeof(float));
 
-	rslt_corr_size =   num_corrs * corr_length * corr_length;
+	rslt_corr_size =   num_corrs * tpParams.corr_length; // corr_size * corr_size;
 	corr_img_size = num_corr_indices * 16*16; // NAN
 	corr_img = (float *)malloc(corr_img_size * sizeof(float));
 	cpu_corr = (float *)malloc(rslt_corr_size * sizeof(float));
 	cpu_corr_indices = (int *) malloc(num_corr_indices * sizeof(int));
 
-
-
     checkCudaErrors(cudaMemcpy2D(
     		cpu_corr,
-			(corr_length * corr_length) * sizeof(float),
+			tpParams.corr_length * sizeof(float),
 			gpu_corrs,
 			dstride_corr,
-			(corr_length * corr_length) * sizeof(float),
+			tpParams.corr_length * sizeof(float),
 			num_corrs,
 			cudaMemcpyDeviceToHost));
     //    checkCudaErrors (cudaMalloc((void **)&gpu_corr_indices,        num_pairs * TILESX * TILESY*sizeof(int)));
@@ -1056,13 +1052,13 @@ int main(int argc, char **argv)
     	int ty = ctt / TILESX;
     	int tx = ctt % TILESX;
     	//		int src_offs0 = ict * tpParams.num_pairs * corr_size * corr_size;
-    	int src_offs0 = ict * corr_length * corr_length;
+    	int src_offs0 = ict * tpParams.corr_length;
     	int dst_offs0 = cpair * (num_tiles * 16 * 16) +  (ty * 16 * TILESX * 16) + (tx * 16);
 
-    	for (int iy = 0; iy < corr_length; iy++){
-    		int src_offs = src_offs0 + iy * corr_length; // ict * tpParams.num_pairs * corr_size * corr_size;
+    	for (int iy = 0; iy < tpParams.corr_size; iy++){
+    		int src_offs = src_offs0 + iy * tpParams.corr_size; // ict * tpParams.num_pairs * corr_size * corr_size;
     		int dst_offs = dst_offs0 + iy * (TILESX * 16);
-    		for (int ix = 0; ix < corr_length; ix++){
+    		for (int ix = 0; ix < tpParams.corr_size; ix++){
     			corr_img[dst_offs++] = cpu_corr[src_offs++];
     		}
     	}
@@ -1116,15 +1112,9 @@ int main(int argc, char **argv)
 			"clt/aux_intrascene-TD.raw"); // 			   const char *  path) // file path
 #endif
 
-
-
     free (cpu_corr_td);
 #endif  // if 1
     // reuse image, export TD data
-
-
-
-
     free (cpu_corr);
     free (cpu_corr_indices);
     free (corr_img);
@@ -1135,6 +1125,9 @@ int main(int argc, char **argv)
 
 
 
+
+
+
 // Testing "interframe" correlation with itself, assuming direct convert already ran