예제 #1
0
/*
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
                );
            }
            
            
        }
    }
}
예제 #2
0
void SeededRegionGrowing::execute() {
    if(mSeedPoints.size() == 0)
        throw Exception("No seed points supplied to SeededRegionGrowing");

    Image::pointer input = getStaticInputData<Image>();
    if(input->getNrOfComponents() != 1)
        throw Exception("Seeded region growing currently doesn't support images with several components.");

    Segmentation::pointer output = getStaticOutputData<Segmentation>();

    // Initialize output image
    output->createFromImage(input, getMainDevice());

    if(getMainDevice()->isHost()) {
        ImageAccess::pointer inputAccess = input->getImageAccess(ACCESS_READ);
        void* inputData = inputAccess->get();
        switch(input->getDataType()) {
            fastSwitchTypeMacro(executeOnHost<FAST_TYPE>((FAST_TYPE*)inputData, output));
        }
    } else {
        OpenCLDevice::pointer device = getMainDevice();

        recompileOpenCLCode(input);

        ImageAccess::pointer access = output->getImageAccess(ACCESS_READ_WRITE);
        uchar* outputData = (uchar*)access->get();
        // Initialize to all 0s
        memset(outputData,0,sizeof(uchar)*output->getWidth()*output->getHeight()*output->getDepth());

        // Add sedd points
        for(int i = 0; i < mSeedPoints.size(); i++) {
            Vector3ui pos = mSeedPoints[i];

            // Check if seed point is in bounds
            if(pos.x() < 0 || pos.y() < 0 || pos.z() < 0 ||
                pos.x() >= output->getWidth() || pos.y() >= output->getHeight() || pos.z() >= output->getDepth())
                throw Exception("One of the seed points given to SeededRegionGrowing was out of bounds.");

            outputData[pos.x() + pos.y()*output->getWidth() + pos.z()*output->getWidth()*output->getHeight()] = 2;
        }
        access->release();

        cl::NDRange globalSize;
        if(output->getDimensions() == 2) {
            globalSize = cl::NDRange(input->getWidth(),input->getHeight());
            OpenCLImageAccess2D::pointer inputAccess = input->getOpenCLImageAccess2D(ACCESS_READ, device);
            mKernel.setArg(0, *inputAccess->get());
        } else {
            globalSize = cl::NDRange(input->getWidth(),input->getHeight(), input->getDepth());
            OpenCLImageAccess3D::pointer inputAccess = input->getOpenCLImageAccess3D(ACCESS_READ, device);
            mKernel.setArg(0, *inputAccess->get());
        }

        OpenCLBufferAccess::pointer outputAccess = output->getOpenCLBufferAccess(ACCESS_READ_WRITE, device);
        cl::Buffer stopGrowingBuffer = cl::Buffer(
                device->getContext(),
                CL_MEM_READ_WRITE,
                sizeof(char));
        cl::CommandQueue queue = device->getCommandQueue();
        mKernel.setArg(1, *outputAccess->get());
        mKernel.setArg(2, stopGrowingBuffer);
        mKernel.setArg(3, mMinimumIntensity);
        mKernel.setArg(4, mMaximumIntensity);

        bool stopGrowing = false;
        char stopGrowingInit = 1;
        char * stopGrowingResult = new char;
        int iterations = 0;
        do {
            iterations++;
            queue.enqueueWriteBuffer(stopGrowingBuffer, CL_TRUE, 0, sizeof(char), &stopGrowingInit);

            queue.enqueueNDRangeKernel(
                    mKernel,
                    cl::NullRange,
                    globalSize,
                    cl::NullRange
            );

            queue.enqueueReadBuffer(stopGrowingBuffer, CL_TRUE, 0, sizeof(char), stopGrowingResult);
            if(*stopGrowingResult == 1)
                stopGrowing = true;
        } while(!stopGrowing);
    }

}