|
Blink API
|
|
Blink API
|
A simple example plug-in that uses the Blink API to do a two-pass box blur and gain on its input.
// Copyright (c) 2013 The Foundry Visionmongers Ltd. All Rights Reserved. //DDImage includes #include "DDImage/PlanarIop.h" #include "DDImage/Knobs.h" #include "DDImage/NukeWrapper.h" #include "DDImage/Blink.h" //Blink API includes #include "Blink/Blink.h" //Header containing the source strings for the Blink kernels we will use for image processing #include "blurAndGainExampleKernels.h" //Some information about the plug-in static const char* const CLASS = "BlinkBlurAndGain"; static const char* const HELP = "Example of a simple box blur and gain operator that uses the Blink API."; /* * Simple Blink API example plug-in that does a two-pass box blur and applies a gain. */ // Blink plug-ins do their image processing on tiles rather than scanlines. They should always inherit from // DD::Image::PlanarIops to get access to ImagePlanes, which can then be wrapped in Blink::Images for // Blink-based image processing. class BlinkBlurAndGain : public DD::Image::PlanarIop { protected: // Reference to the GPU device to process on. Blink::ComputeDevice _gpuDevice; // Whether to process on the GPU, if available bool _useGPUIfAvailable; // The radius to use for the blur kernel. int _radius; // The amount of gain to apply. float _gain; // This holds the ProgramSource for the blur kernel. Blink::ProgramSource _blurProgram; // This holds the ProgramSource for the gain kernel. Blink::ProgramSource _gainProgram; public: // Constructor. Initialize user controls and local variables to their default values. BlinkBlurAndGain(Node* node) : PlanarIop(node) , _gpuDevice(Blink::ComputeDevice::CurrentGPUDevice()) , _useGPUIfAvailable(true) , _blurProgram(BlurKernel) , _gainProgram(GainKernel) { _radius = 3; _gain = 2.0f; } // This function creates the controls that will be exposed to the user. void knobs(DD::Image::Knob_Callback f); void _validate(bool); void getRequests(const DD::Image::Box& box, const DD::Image::ChannelSet& channels, int count, DD::Image::RequestOutput &reqData) const; // Whether to process in stripes or full-frame. virtual bool useStripes() const; // Set the stripe height to use for processing. virtual size_t stripeHeight() const; // This function does all the work in a PlanarIop. It is guaranteed that outputPlane here will never have more than // the 4 channels which can be stored in a Blink::Image. void renderStripe(DD::Image::ImagePlane& outputPlane); // Return the name of the class. const char* Class() const { return CLASS; } // Return a brief description of the node. const char* node_help() const { return HELP; } // Information to the plug-in manager of DDImage/Nuke. static const Iop::Description description; }; void BlinkBlurAndGain::knobs(DD::Image::Knob_Callback f) { // Add GPU knobs Newline(f, "Local GPU: "); const bool hasGPU = _gpuDevice.available(); std::string gpuName = hasGPU ? _gpuDevice.name() : "Not available"; Named_Text_knob(f, "gpuName", gpuName.c_str()); Newline(f); Bool_knob(f, &_useGPUIfAvailable, "use_gpu", "Use GPU if available"); Divider(f); // Add a parameter for the blur radius. Int_knob(f, &_radius, "radius"); Tooltip(f, "The radius of the blur."); // Add a parameter for the gain amount. Float_knob(f, &_gain, DD::Image::IRange(0, 10), "gain"); Tooltip(f, "The amount of gain to apply."); } void BlinkBlurAndGain::_validate(bool for_real) { // Copy bbox channels etc from input0, which will validate it. copy_info(); // Pad the output area this node can produce by the size of the blur radius. info_.pad(_radius); } void BlinkBlurAndGain::getRequests(const DD::Image::Box& box, const DD::Image::ChannelSet& channels, int count, DD::Image::RequestOutput &reqData) const { DD::Image::Box inputBox = box; // Request _radius extra pixels around the input. inputBox.expand(_radius); reqData.request(&input0(), inputBox, channels, count); } /* * Override of PlanarIop function renderStripe(...). This does the image processing to produce a stripe of the output. */ void BlinkBlurAndGain::renderStripe(DD::Image::ImagePlane& outputPlane) { // Work out the bounds of the area we need to fetch from the input. This will depend on the output bounds. DD::Image::Box inputBox = outputPlane.bounds(); inputBox.expand(_radius); inputBox.intersect(input0().info()); // Make an ImagePlaneDescriptor that describes how the inputs should be stored. DD::Image::ImagePlaneDescriptor inputDescriptor(inputBox, // the bounds of the input we want to fetch outputPlane.packed(), // input data should be packed in the same way as the output plane outputPlane.channels(), // input data should have the same channels as the output plane outputPlane.nComps()); // input data should have the same number of components as the output plane // Make an ImagePlane that satisfies this description. DD::Image::ImagePlane inputPlane(inputDescriptor); // Fetch the data from input0 into our ImagePlane. input0().fetchPlane(inputPlane); // This function must be called on the output plane before writing to it. outputPlane.makeWritable(); // Wrap the input and output planes as Blink images. The underlying data stays the same. Blink::Image outputPlaneAsImage; Blink::Image inputPlaneAsImage; bool success = (DD::Image::Blink::ImagePlaneAsBlinkImage(outputPlane, outputPlaneAsImage) && DD::Image::Blink::ImagePlaneAsBlinkImage(inputPlane, inputPlaneAsImage)); // Check the fetch succeeded. if (!success) { error("Unable to fetch Blink image for image plane."); return; } // Has the user requested GPU processing, and is the GPU available for processing on? bool usingGPU = _useGPUIfAvailable && _gpuDevice.available(); // Make a temporary image to blur into, on the GPU or CPU as appropriate. This should have the same layout as the input image. Blink::Image intermediateImage = usingGPU ? inputPlaneAsImage.makeLike(_gpuDevice) : inputPlaneAsImage.makeLike(Blink::ComputeDevice::CurrentCPUDevice()); // Get a reference to the ComputeDevice to do our processing on. Blink::ComputeDevice computeDevice = usingGPU ? _gpuDevice : Blink::ComputeDevice::CurrentCPUDevice(); // Distribute our input image from the device used by NUKE to our ComputeDevice. Blink::Image inputImageOnComputeDevice = inputPlaneAsImage.distributeTo(computeDevice); // This will bind the compute device to the calling thread. This bind should always be performed before // beginning any image processing with Blink. Blink::ComputeDeviceBinder binder(computeDevice); // Make a vector containing the images we want to run the first kernel over. std::vector<Blink::Image> images; images.push_back(inputImageOnComputeDevice); images.push_back(intermediateImage); // Make a Blink::Kernel from the source in _blurProgram to do the horizontal blur. Blink::Kernel horizontalBlurKernel(_blurProgram, computeDevice, images, kBlinkCodegenDefault); horizontalBlurKernel.setParamValue("Radius", _radius); horizontalBlurKernel.setParamValue("Horizontal", true); // Run the horizontal blur kernel. Because we haven't specified the bounds to iterate over, these // will be the bounds of the last image passed in - in this case, the intermediateImage. horizontalBlurKernel.iterate(); // If we are on the GPU, we need to make an output image, otherwise we can just use the one from NUKE. Blink::Image outputImage = usingGPU ? outputPlaneAsImage.makeLike(_gpuDevice) : outputPlaneAsImage; // Reset the images vector to hold the images needed for the vertical blur. images.clear(); images.push_back(intermediateImage); images.push_back(outputImage); // Make a Blink::Kernel from the source in _blurProgram to do the vertical blur. Blink::Kernel verticalBlurKernel(_blurProgram, computeDevice, images, kBlinkCodegenDefault); verticalBlurKernel.setParamValue("Radius", _radius); verticalBlurKernel.setParamValue("Horizontal", false); // Run the vertical blur kernel. Because we haven't specified the bounds to iterate over, these // will be the bounds of the last image passed in - in this case, the outputImage. verticalBlurKernel.iterate(); // The gain kernel requires only the output image as input/output. images.clear(); images.push_back(outputImage); // Make a Blink::Kernel from the source in _gainProgram to do the gain. Blink::Kernel gainKernel(_gainProgram, computeDevice, images, kBlinkCodegenDefault); gainKernel.setParamValue("Gain", _gain); // Run the gain kernel over the output image. gainKernel.iterate(); // If we're using the GPU, copy the result back to NUKE's output plane. if (usingGPU) { outputPlaneAsImage.copyFrom(outputImage); } } /* * Override of PlanarIop function; returns true. * renderStripe() will be called on imageplanes of size info().w() x stripeHeight(). */ bool BlinkBlurAndGain::useStripes() const { return true; } /* * Override of PlanarIop function; * returns a fixed stripe height of 256. */ size_t BlinkBlurAndGain::stripeHeight() const { return 256; } /* This is a function that creates an instance of the operator, and is needed for the Iop::Description to work. */ static DD::Image::Iop* BlinkBlurAndGainCreate(Node* node) { return new BlinkBlurAndGain(node); } /* The Iop::Description is how NUKE knows what the name of the operator is, how to create one, and the menu item to show the user. The menu item may be 0 if you do not want the operator to be visible. */ const DD::Image::Iop::Description BlinkBlurAndGain::description(CLASS, "Filter/BlinkBlurAndGain", BlinkBlurAndGainCreate);
| ©2013 The Foundry Visionmongers, Ltd. All Rights Reserved. |
