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

ocl: adjust worksize for filter2D and boxFilter

Browse Source
This commit is contained in:
Alexander Alekhin
2013-10-30 14:57:46 +04:00
parent 8a4f1bbbdf
commit 7b0f018a74
3 changed files with 227 additions and 171 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
modules/ocl/src/cl_operations.cpp
View File

@@ -336,8 +336,7 @@ static std::string removeDuplicatedWhiteSpaces(const char * buildOptions)
return opt;
}
void openCLExecuteKernel_(Context *ctx, const cv::ocl::ProgramEntry* source, string kernelName, size_t globalThreads[3],
size_t localThreads[3], vector< pair<size_t, const void *> > &args, int channels,
cl_kernel openCLGetKernelFromSource(Context *ctx, const cv::ocl::ProgramEntry* source, string kernelName, int channels,
int depth, const char *build_options)
{
//construct kernel name
@@ -350,10 +349,14 @@ void openCLExecuteKernel_(Context *ctx, const cv::ocl::ProgramEntry* source, str
idxStr << "_D" << depth;
kernelName += idxStr.str();
cl_kernel kernel;
std::string fixedOptions = removeDuplicatedWhiteSpaces(build_options);
kernel = openCLGetKernelFromSource(ctx, source, kernelName, fixedOptions.c_str());
cl_kernel kernel = openCLGetKernelFromSource(ctx, source, kernelName, fixedOptions.c_str());
return kernel;
}
void openCLExecuteKernel(Context *ctx, cl_kernel kernel, size_t globalThreads[3],
size_t localThreads[3], vector< pair<size_t, const void *> > &args)
{
if ( localThreads != NULL)
{
globalThreads[0] = roundUp(globalThreads[0], localThreads[0]);
@@ -399,6 +402,15 @@ void openCLExecuteKernel_(Context *ctx, const cv::ocl::ProgramEntry* source, str
openCLSafeCall(clReleaseKernel(kernel));
}
void openCLExecuteKernel_(Context *ctx, const cv::ocl::ProgramEntry* source, string kernelName, size_t globalThreads[3],
size_t localThreads[3], vector< pair<size_t, const void *> > &args, int channels,
int depth, const char *build_options)
{
cl_kernel kernel = openCLGetKernelFromSource(ctx, source, kernelName, channels, depth, build_options);
openCLExecuteKernel(ctx, kernel, globalThreads, localThreads, args);
}
void openCLExecuteKernel(Context *ctx, const cv::ocl::ProgramEntry* source, string kernelName,
size_t globalThreads[3], size_t localThreads[3],
vector< pair<size_t, const void *> > &args, int channels, int depth)