void NoneLocalMeans::recompileOpenCLCode(Image::pointer input) {
// Check if there is a need to recompile OpenCL code
if(input->getDimensions() == mDimensionCLCodeCompiledFor &&
input->getDataType() == mTypeCLCodeCompiledFor && !recompile)
return;
recompile = false;
OpenCLDevice::pointer device = getMainDevice();
std::string buildOptions = "";
if(!device->isWritingTo3DTexturesSupported()) {
buildOptions = "-DTYPE=" + getCTypeAsString(mOutputType);
}
buildOptions += " -D WINDOW=";
buildOptions += std::to_string((windowSize-1)/2);
buildOptions += " -D GROUP=";
buildOptions += std::to_string((groupSize-1)/2);
buildOptions += " -D KVERSION=";
buildOptions += std::to_string(k);
buildOptions += " -D EUCLID=";
buildOptions += std::to_string(k);
cl::Program program;
if(input->getDimensions() == 2) {
program = getOpenCLProgram(device, "2D", buildOptions);
} else {
//createOpenCLProgram(std::string(FAST_SOURCE_DIR) + "Algorithms/NoneLocalMeans/NoneLocalMeans3Dgs.cl", "3D");
program = getOpenCLProgram(device, "3D", buildOptions);
}
mKernel = cl::Kernel(program, "noneLocalMeans");
mDimensionCLCodeCompiledFor = input->getDimensions();
mTypeCLCodeCompiledFor = input->getDataType();
}
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
);
}
}
/*
void NoneLocalMeans::recompileOpenCLCode(Image::pointer input) {
// Check if there is a need to recompile OpenCL code
if (input->getDimensions() == mDimensionCLCodeCompiledFor &&
input->getDataType() == mTypeCLCodeCompiledFor && !recompile)
return;
OpenCLDevice::pointer device = getMainDevice();
recompile = false;
std::string buildOptions = "";
const bool writingTo3DTextures = device->getDevice().getInfo<CL_DEVICE_EXTENSIONS>().find("cl_khr_3d_image_writes") != std::string::npos;
if (!writingTo3DTextures) {
switch (mOutputType) {
case TYPE_FLOAT:
buildOptions += " -DTYPE=float";
break;
case TYPE_INT8:
buildOptions += " -DTYPE=char";
break;
case TYPE_UINT8:
buildOptions += " -DTYPE=uchar";
break;
case TYPE_INT16:
buildOptions += " -DTYPE=short";
break;
case TYPE_UINT16:
buildOptions += " -DTYPE=ushort";
break;
}
}
buildOptions += " -D WINDOW=";
buildOptions += std::to_string((windowSize-1)/2);
buildOptions += " -D GROUP=";
buildOptions += std::to_string((groupSize-1)/2);
std::string filename;
//might have to seperate color vs gray here, for better runtime
if (input->getDimensions() == 2) {
if(k == 0){
filename = "Algorithms/NoneLocalMeans/NoneLocalMeans2Dconstant.cl";
}else if(k == 1){
filename = "Algorithms/NoneLocalMeans/NoneLocalMeans2Dgaussian.cl";
}else{
filename = "Algorithms/NoneLocalMeans/NoneLocalMeans2Dconstant.cl";
}
//filename = "Algorithms/NoneLocalMeans/NoneLocalMeans2DgsPixelWise.cl";
//filename = "Algorithms/NoneLocalMeans/NoneLocalMeans2Dgs.cl";
//filename = "Algorithms/NoneLocalMeans/NoneLocalMeans2Dc.cl";
}
else {
filename = "Algorithms/NoneLocalMeans/NoneLocalMeans3Dgs.cl";
}
int programNr = device->createProgramFromSource(std::string(FAST_SOURCE_DIR) + filename, buildOptions);
mKernel = cl::Kernel(device->getProgram(programNr), "noneLocalMeans");
mDimensionCLCodeCompiledFor = input->getDimensions();
mTypeCLCodeCompiledFor = input->getDataType();
}*/
void NoneLocalMeans::execute() {
Image::pointer input = getStaticInputData<Image>(0);
Image::pointer output = getStaticOutputData<Image>(0);
// Initialize output image
ExecutionDevice::pointer device = getMainDevice();
if(mOutputTypeSet) {
output->create(input->getSize(), mOutputType, input->getNrOfComponents());
output->setSpacing(input->getSpacing());
} else {
output->createFromImage(input);
}
mOutputType = output->getDataType();
SceneGraph::setParentNode(output, input);
if(device->isHost()) {
switch(input->getDataType()) {
fastSwitchTypeMacro(executeAlgorithmOnHost<FAST_TYPE>(input, output, groupSize, windowSize, denoiseStrength, sigma));
}
} else {
OpenCLDevice::pointer clDevice = device;
recompileOpenCLCode(input);
cl::NDRange globalSize;
OpenCLImageAccess::pointer inputAccess = input->getOpenCLImageAccess(ACCESS_READ, device);
if(input->getDimensions() == 2) {
OpenCLImageAccess::pointer outputAccess = output->getOpenCLImageAccess(ACCESS_READ_WRITE, device);
mKernel.setArg(2, (denoiseStrength*denoiseStrength));
mKernel.setArg(3, (sigma*sigma));
globalSize = cl::NDRange(input->getWidth(),input->getHeight());
mKernel.setArg(0, *inputAccess->get2DImage());
mKernel.setArg(1, *outputAccess->get2DImage());
clDevice->getCommandQueue().enqueueNDRangeKernel(
mKernel,
cl::NullRange,
globalSize,
cl::NullRange
);
} else {
// Create an auxilliary image
//Image::pointer output2 = Image::New();
//output2->createFromImage(output);
globalSize = cl::NDRange(input->getWidth(),input->getHeight(),input->getDepth());
if(clDevice->isWritingTo3DTexturesSupported()) {
mKernel.setArg(2, (denoiseStrength*denoiseStrength));
mKernel.setArg(3, (sigma*sigma));
OpenCLImageAccess::pointer outputAccess = output->getOpenCLImageAccess(ACCESS_READ_WRITE, device);
//OpenCLImageAccess::pointer outputAccess2 = output2->getOpenCLImageAccess(ACCESS_READ_WRITE, device);
//cl::Image3D* image2;
cl::Image3D* image;
image = outputAccess->get3DImage();
//image2 = outputAccess->get3DImage();
mKernel.setArg(0, *inputAccess->get3DImage());
mKernel.setArg(1, *image);
clDevice->getCommandQueue().enqueueNDRangeKernel(
mKernel,
cl::NullRange,
globalSize,
cl::NullRange
);
}else{
mKernel.setArg(2, (denoiseStrength*denoiseStrength));
mKernel.setArg(3, (sigma*sigma));
OpenCLBufferAccess::pointer outputAccess = output->getOpenCLBufferAccess(ACCESS_READ_WRITE, device);
mKernel.setArg(0, *inputAccess->get3DImage());
mKernel.setArg(1, *outputAccess->get());
clDevice->getCommandQueue().enqueueNDRangeKernel(
mKernel,
cl::NullRange,
globalSize,
cl::NullRange
);
}
}
}
}