void Dilation::execute() {
Image::pointer input = getInputData<Image>();
if(input->getDataType() != TYPE_UINT8) {
throw Exception("Data type of image given to Dilation must be UINT8");
}
Image::pointer output = getOutputData<Image>();
output->createFromImage(input);
SceneGraph::setParentNode(output, input);
output->fill(0);
OpenCLDevice::pointer device = std::dynamic_pointer_cast<OpenCLDevice>(getMainDevice());
cl::CommandQueue queue = device->getCommandQueue();
cl::Program program = getOpenCLProgram(device);
cl::Kernel dilateKernel(program, "dilate");
Vector3ui size = input->getSize();
OpenCLImageAccess::pointer access = input->getOpenCLImageAccess(ACCESS_READ, device);
dilateKernel.setArg(0, *access->get3DImage());
dilateKernel.setArg(2, mSize/2);
if(!device->isWritingTo3DTexturesSupported()) {
OpenCLBufferAccess::pointer access2 = output->getOpenCLBufferAccess(ACCESS_READ_WRITE, device);
dilateKernel.setArg(1, *access2->get());
queue.enqueueNDRangeKernel(
dilateKernel,
cl::NullRange,
cl::NDRange(size.x(), size.y(), size.z()),
cl::NullRange
);
} else {
OpenCLImageAccess::pointer access2 = output->getOpenCLImageAccess(ACCESS_READ_WRITE, device);
dilateKernel.setArg(1, *access2->get3DImage());
queue.enqueueNDRangeKernel(
dilateKernel,
cl::NullRange,
cl::NDRange(size.x(), size.y(), size.z()),
cl::NullRange
);
}
}
std::tr1::tuple<cl::Kernel,cl::Kernel,std::vector<cl::Buffer*>,cl::CommandQueue,cl::NDRange,cl::NDRange,cl::NDRange> init_cl(int levels, unsigned w, unsigned h, unsigned bits, std::string source, int deviceNumber)
{
std::vector<cl::Platform> platforms;
cl::Platform::get(&platforms);
if(platforms.size()==0) throw std::runtime_error("No OpenCL platforms found.");
std::cerr<<"Found "<<platforms.size()<<" platforms\n";
for(unsigned i=0;i<platforms.size();i++){
std::string vendor=platforms[0].getInfo<CL_PLATFORM_VENDOR>();
std::cerr<<" Platform "<<i<<" : "<<vendor<<"\n";
}
int selectedPlatform=0;
if(getenv("HPCE_SELECT_PLATFORM")){
selectedPlatform=atoi(getenv("HPCE_SELECT_PLATFORM"));
}
std::cerr<<"Choosing platform "<<selectedPlatform<<"\n";
cl::Platform platform=platforms.at(selectedPlatform);
std::vector<cl::Device> devices;
platform.getDevices(CL_DEVICE_TYPE_ALL, &devices);
if(devices.size()==0){
throw std::runtime_error("No opencl devices found.\n");
}
std::cerr<<"Found "<<devices.size()<<" devices\n";
for(unsigned i=0;i<devices.size();i++){
std::string name=devices[i].getInfo<CL_DEVICE_NAME>();
std::cerr<<" Device "<<i<<" : "<<name<<"\n";
}
int selectedDevice=0;
if (deviceNumber != -1) selectedDevice = deviceNumber;
std::cerr<<"Choosing device "<<selectedDevice<<"\n";
cl::Device device=devices.at(selectedDevice);
cl::Context context(devices);
std::string kernelSource=LoadSource(source.c_str());
cl::Program::Sources sources; // A vector of (data,length) pairs
sources.push_back(std::make_pair(kernelSource.c_str(), kernelSource.size()+1)); // push on our single string
cl::Program program(context, sources);
try{
program.build(devices);
}catch(...){
for(unsigned i=0;i<devices.size();i++){
std::cerr<<"Log for device "<<devices[i].getInfo<CL_DEVICE_NAME>()<<":\n\n";
std::cerr<<program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(devices[i])<<"\n\n";
}
throw;
}
size_t cbBuffer= (w*bits)/2;
std::vector<cl::Buffer*> gpuBuffers;
for (int i=0; i<abs(levels); i++)
{
gpuBuffers.push_back(new cl::Buffer(context, CL_MEM_READ_WRITE, cbBuffer));
gpuBuffers.push_back(new cl::Buffer(context, CL_MEM_READ_WRITE, cbBuffer));
}
gpuBuffers.push_back(new cl::Buffer(context, CL_MEM_READ_WRITE, cbBuffer)); // ... and one for luck.
std::string erodeKernelName;
std::string dilateKernelName;
switch (bits) {
case 1:
erodeKernelName = "erode_kernel_1";
dilateKernelName = "dilate_kernel_1";
break;
case 2:
erodeKernelName = "erode_kernel_2";
dilateKernelName = "dilate_kernel_2";
break;
case 4:
erodeKernelName = "erode_kernel_4";
dilateKernelName = "dilate_kernel_4";
break;
case 8:
erodeKernelName = "erode_kernel_8";
dilateKernelName = "dilate_kernel_8";
break;
case 16:
erodeKernelName = "erode_kernel_16";
dilateKernelName = "dilate_kernel_16";
break;
case 32:
erodeKernelName = "erode_kernel_32";
dilateKernelName = "dilate_kernel_32";
break;
default:
break;
}
cl::Kernel erodeKernel(program, erodeKernelName.c_str());
cl::Kernel dilateKernel(program, dilateKernelName.c_str());
cl::CommandQueue queue(context, device);
cl::NDRange offset(0, 0); // Always start iterations at x=0, y=0
cl::NDRange globalSize((w*bits)/32, 1); // Global size must match the original loops
cl::NDRange localSize=cl::NullRange; // We don't care about local size
return std::tr1::make_tuple(erodeKernel,dilateKernel,gpuBuffers,queue,offset,globalSize,localSize);
}
