generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

ocl: Canny: replace unsafe buffer to oclMat

Browse Source
This commit is contained in:
Alexander Alekhin 2013-10-27 18:55:42 +04:00
parent c1223f8dac
commit 9e527fc9f4
2 changed files with 21 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
modules/ocl/include/opencv2/ocl/ocl.hpp
View File

@ -984,12 +984,12 @@ namespace cv
struct CV_EXPORTS CannyBuf struct CV_EXPORTS CannyBuf
{ {
CannyBuf() : counter(NULL) {} CannyBuf() : counter(1, 1, CV_32S) { }
~CannyBuf() ~CannyBuf()
{ {
release(); release();
} }
explicit CannyBuf(const Size &image_size, int apperture_size = 3) : counter(NULL) explicit CannyBuf(const Size &image_size, int apperture_size = 3) : counter(1, 1, CV_32S)
{ {
create(image_size, apperture_size); create(image_size, apperture_size);
} }
@ -1001,7 +1001,7 @@ namespace cv
oclMat dx_buf, dy_buf; oclMat dx_buf, dy_buf;
oclMat edgeBuf; oclMat edgeBuf;
oclMat trackBuf1, trackBuf2; oclMat trackBuf1, trackBuf2;
void *counter; oclMat counter;
Ptr<FilterEngine_GPU> filterDX, filterDY; Ptr<FilterEngine_GPU> filterDX, filterDY;
}; };

42
modules/ocl/src/canny.cpp
View File

@ -49,7 +49,7 @@
using namespace cv; using namespace cv;
using namespace cv::ocl; using namespace cv::ocl;
cv::ocl::CannyBuf::CannyBuf(const oclMat &dx_, const oclMat &dy_) : dx(dx_), dy(dy_), counter(NULL) cv::ocl::CannyBuf::CannyBuf(const oclMat &dx_, const oclMat &dy_) : dx(dx_), dy(dy_), counter(1, 1, CV_32SC1)
{ {
CV_Assert(dx_.type() == CV_32SC1 && dy_.type() == CV_32SC1 && dx_.size() == dy_.size()); CV_Assert(dx_.type() == CV_32SC1 && dy_.type() == CV_32SC1 && dx_.size() == dy_.size());
@ -82,15 +82,6 @@ void cv::ocl::CannyBuf::create(const Size &image_size, int apperture_size)
ensureSizeIsEnough(1, image_size.width * image_size.height, CV_16UC2, trackBuf1); ensureSizeIsEnough(1, image_size.width * image_size.height, CV_16UC2, trackBuf1);
ensureSizeIsEnough(1, image_size.width * image_size.height, CV_16UC2, trackBuf2); ensureSizeIsEnough(1, image_size.width * image_size.height, CV_16UC2, trackBuf2);
int counter_i [1] = { 0 };
int err = 0;
if(counter)
{
openCLFree(counter);
}
counter = clCreateBuffer( *((cl_context*)getClContextPtr()), CL_MEM_COPY_HOST_PTR, sizeof(int), counter_i, &err );
openCLSafeCall(err);
} }
void cv::ocl::CannyBuf::release() void cv::ocl::CannyBuf::release()
@ -102,7 +93,6 @@ void cv::ocl::CannyBuf::release()
edgeBuf.release(); edgeBuf.release();
trackBuf1.release(); trackBuf1.release();
trackBuf2.release(); trackBuf2.release();
openCLFree(counter);
} }
namespace cv namespace cv
@ -118,9 +108,9 @@ namespace cv
void calcMap_gpu(oclMat &dx, oclMat &dy, oclMat &mag, oclMat &map, int rows, int cols, float low_thresh, float high_thresh); void calcMap_gpu(oclMat &dx, oclMat &dy, oclMat &mag, oclMat &map, int rows, int cols, float low_thresh, float high_thresh);
void edgesHysteresisLocal_gpu(oclMat &map, oclMat &st1, void *counter, int rows, int cols); void edgesHysteresisLocal_gpu(oclMat &map, oclMat &st1, oclMat& counter, int rows, int cols);
void edgesHysteresisGlobal_gpu(oclMat &map, oclMat &st1, oclMat &st2, void *counter, int rows, int cols); void edgesHysteresisGlobal_gpu(oclMat &map, oclMat &st1, oclMat &st2, oclMat& counter, int rows, int cols);
void getEdges_gpu(oclMat &map, oclMat &dst, int rows, int cols); void getEdges_gpu(oclMat &map, oclMat &dst, int rows, int cols);
} }
@ -322,7 +312,7 @@ void canny::calcMap_gpu(oclMat &dx, oclMat &dy, oclMat &mag, oclMat &map, int ro
openCLExecuteKernel(clCxt, &imgproc_canny, kernelName, globalThreads, localThreads, args, -1, -1); openCLExecuteKernel(clCxt, &imgproc_canny, kernelName, globalThreads, localThreads, args, -1, -1);
} }
void canny::edgesHysteresisLocal_gpu(oclMat &map, oclMat &st1, void *counter, int rows, int cols) void canny::edgesHysteresisLocal_gpu(oclMat &map, oclMat &st1, oclMat& counter, int rows, int cols)
{ {
Context *clCxt = map.clCxt; Context *clCxt = map.clCxt;
string kernelName = "edgesHysteresisLocal"; string kernelName = "edgesHysteresisLocal";
@ -330,7 +320,7 @@ void canny::edgesHysteresisLocal_gpu(oclMat &map, oclMat &st1, void *counter, in
args.push_back( make_pair( sizeof(cl_mem), (void *)&map.data)); args.push_back( make_pair( sizeof(cl_mem), (void *)&map.data));
args.push_back( make_pair( sizeof(cl_mem), (void *)&st1.data)); args.push_back( make_pair( sizeof(cl_mem), (void *)&st1.data));
args.push_back( make_pair( sizeof(cl_mem), (void *)&counter)); args.push_back( make_pair( sizeof(cl_mem), (void *)&counter.data));
args.push_back( make_pair( sizeof(cl_int), (void *)&rows)); args.push_back( make_pair( sizeof(cl_int), (void *)&rows));
args.push_back( make_pair( sizeof(cl_int), (void *)&cols)); args.push_back( make_pair( sizeof(cl_int), (void *)&cols));
args.push_back( make_pair( sizeof(cl_int), (void *)&map.step)); args.push_back( make_pair( sizeof(cl_int), (void *)&map.step));
@ -342,26 +332,30 @@ void canny::edgesHysteresisLocal_gpu(oclMat &map, oclMat &st1, void *counter, in
openCLExecuteKernel(clCxt, &imgproc_canny, kernelName, globalThreads, localThreads, args, -1, -1); openCLExecuteKernel(clCxt, &imgproc_canny, kernelName, globalThreads, localThreads, args, -1, -1);
} }
void canny::edgesHysteresisGlobal_gpu(oclMat &map, oclMat &st1, oclMat &st2, void *counter, int rows, int cols) void canny::edgesHysteresisGlobal_gpu(oclMat &map, oclMat &st1, oclMat &st2, oclMat& counter, int rows, int cols)
{ {
unsigned int count; Mat counterMat; counter.download(counterMat);
openCLSafeCall(clEnqueueReadBuffer(*(cl_command_queue*)getClCommandQueuePtr(), (cl_mem)counter, 1, 0, sizeof(float), &count, 0, NULL, NULL));
Context *clCxt = map.clCxt; Context *clCxt = map.clCxt;
string kernelName = "edgesHysteresisGlobal"; string kernelName = "edgesHysteresisGlobal";
vector< pair<size_t, const void *> > args; vector< pair<size_t, const void *> > args;
size_t localThreads[3] = {128, 1, 1}; size_t localThreads[3] = {128, 1, 1};
int count_i[1] = {0}; while(1 > 0)
while(count > 0)
{ {
openCLSafeCall(clEnqueueWriteBuffer(*(cl_command_queue*)getClCommandQueuePtr(), (cl_mem)counter, 1, 0, sizeof(int), &count_i, 0, NULL, NULL)); int count = counterMat.at<int>(0, 0);
CV_Assert(count >= 0);
if (count == 0)
break;
counterMat.at<int>(0, 0) = 0;
counter.upload(counterMat);
args.clear(); args.clear();
size_t globalThreads[3] = {std::min(count, 65535u) * 128, divUp(count, 65535), 1}; size_t globalThreads[3] = {std::min((unsigned)count, 65535u) * 128, divUp(count, 65535), 1};
args.push_back( make_pair( sizeof(cl_mem), (void *)&map.data)); args.push_back( make_pair( sizeof(cl_mem), (void *)&map.data));
args.push_back( make_pair( sizeof(cl_mem), (void *)&st1.data)); args.push_back( make_pair( sizeof(cl_mem), (void *)&st1.data));
args.push_back( make_pair( sizeof(cl_mem), (void *)&st2.data)); args.push_back( make_pair( sizeof(cl_mem), (void *)&st2.data));
args.push_back( make_pair( sizeof(cl_mem), (void *)&counter)); args.push_back( make_pair( sizeof(cl_mem), (void *)&counter.data));
args.push_back( make_pair( sizeof(cl_int), (void *)&rows)); args.push_back( make_pair( sizeof(cl_int), (void *)&rows));
args.push_back( make_pair( sizeof(cl_int), (void *)&cols)); args.push_back( make_pair( sizeof(cl_int), (void *)&cols));
args.push_back( make_pair( sizeof(cl_int), (void *)&count)); args.push_back( make_pair( sizeof(cl_int), (void *)&count));
@ -369,7 +363,7 @@ void canny::edgesHysteresisGlobal_gpu(oclMat &map, oclMat &st1, oclMat &st2, voi
args.push_back( make_pair( sizeof(cl_int), (void *)&map.offset)); args.push_back( make_pair( sizeof(cl_int), (void *)&map.offset));
openCLExecuteKernel(clCxt, &imgproc_canny, kernelName, globalThreads, localThreads, args, -1, -1); openCLExecuteKernel(clCxt, &imgproc_canny, kernelName, globalThreads, localThreads, args, -1, -1);
openCLSafeCall(clEnqueueReadBuffer(*(cl_command_queue*)getClCommandQueuePtr(), (cl_mem)counter, 1, 0, sizeof(int), &count, 0, NULL, NULL)); counter.download(counterMat);
std::swap(st1, st2); std::swap(st1, st2);
} }
} }