added perf. test for GFF

diff --git a/modules/gpu/src/cuda/ccomponetns.cu b/modules/gpu/src/cuda/ccomponetns.cu
index 3a54c49..11d4742 100644 (file)
--- a/modules/gpu/src/cuda/ccomponetns.cu
+++ b/modules/gpu/src/cuda/ccomponetns.cu
@@ -485,7 +485,9 @@ namespace cv { namespace gpu { namespace device
                 comps(y, x) = root(comps, comps(y, x));
         }
 
-        void labelComponents(const DevMem2D& edges, DevMem2Di comps, cudaStream_t stream)
+        enum {CC_NO_COMPACT = 0, CC_COMPACT_LABELS = 1};
+
+        void labelComponents(const DevMem2D& edges, DevMem2Di comps, int flags, cudaStream_t stream)
         {
             dim3 block(CTA_SIZE_X, CTA_SIZE_Y);
             dim3 grid(divUp(edges.cols, TILE_COLS), divUp(edges.rows, TILE_ROWS));
@@ -494,15 +496,12 @@ namespace cv { namespace gpu { namespace device
             cudaSafeCall( cudaGetLastError() );
 
             int tileSizeX = TILE_COLS, tileSizeY = TILE_ROWS;
-
-            cudaSafeCall( cudaGetLastError() );
-            // cudaSafeCall( cudaDeviceSynchronize() );
-
             while (grid.x > 1 || grid.y > 1)
             {
                 dim3 mergeGrid(ceilf(grid.x / 2.0), ceilf(grid.y / 2.0));
                 dim3 mergeBlock(STA_SIZE_MARGE_X, STA_SIZE_MARGE_Y);
-                std::cout << "merging: " << grid.y  << " x " << grid.x << " ---> " << mergeGrid.y <<  " x " << mergeGrid.x << " for tiles: " << tileSizeY << " x " << tileSizeX << std::endl;
+                // debug log
+                // std::cout << "merging: " << grid.y  << " x " << grid.x << " ---> " << mergeGrid.y <<  " x " << mergeGrid.x << " for tiles: " << tileSizeY << " x " << tileSizeX << std::endl;
                 crossMerge<<<mergeGrid, mergeBlock, 0, stream>>>(2, 2, tileSizeY, tileSizeX, edges, comps, ceilf(grid.y / 2.0) - grid.y / 2, ceilf(grid.x / 2.0) - grid.x / 2);
                 tileSizeX <<= 1;
                 tileSizeY <<= 1;
@@ -515,6 +514,7 @@ namespace cv { namespace gpu { namespace device
             grid.y = divUp(edges.rows, block.y);
             flatten<<<grid, block, 0, stream>>>(edges, comps);
             cudaSafeCall( cudaGetLastError() );
+
             if (stream == 0)
                 cudaSafeCall( cudaDeviceSynchronize() );
         }

diff --git a/modules/gpu/src/graphcuts.cpp b/modules/gpu/src/graphcuts.cpp
index dd8cf8e..73ae4c8 100644 (file)
--- a/modules/gpu/src/graphcuts.cpp
+++ b/modules/gpu/src/graphcuts.cpp
@@ -56,7 +56,7 @@ namespace cv { namespace gpu { namespace device
 {\r
     namespace ccl\r
     {\r
-        void labelComponents(const DevMem2D& edges, DevMem2Di comps, cudaStream_t stream);\r
+        void labelComponents(const DevMem2D& edges, DevMem2Di comps, int flags, cudaStream_t stream);\r
 \r
         template<typename T>\r
         void computeEdges(const DevMem2D& image, DevMem2D edges, const float4& lo, const float4& hi, cudaStream_t stream);\r
@@ -114,7 +114,7 @@ void cv::gpu::labelComponents(const GpuMat& mask, GpuMat& components, Stream& s)
         components.create(mask.size(), CV_32SC1);\r
 \r
     cudaStream_t stream = StreamAccessor::getStream(s);\r
-    device::ccl::labelComponents(mask, components, stream);\r
+    device::ccl::labelComponents(mask, components, 0, stream);\r
 }\r
 \r
 namespace\r