Examples of com.jogamp.opencl.CLKernel

com.jogamp.opencl.CLKernel
High level abstraction for an OpenCL Kernel. A kernel is a function declared in a program. A kernel is identified by the kernel qualifier applied to any function in a program. A kernel object encapsulates the specific kernel function declared in a program and the argument values to be used when executing this kernel function.
Example:
```
 CLKernel addKernel = program.createCLKernel("add"); addKernel.setArgs(clBufferA, clBufferB); ... queue.putEnqueue1DKernel(addKernel, 0, clBufferA.getSize(), 0); 
```
CLKernel provides utility methods for setting vector types (float4, int2...) with up to 4 elements. Larger vectors like float16 can be set using {@link #setArg(int,java.nio.Buffer)}. Arguments pointing to {@link CLBuffer}s or {@link CLImage}s can be set using {@link #setArg(int,com.jogamp.opencl.CLMemory) }or its relative putArg(..) methods.

CLKernel is not threadsafe. However it is perfectly safe to create a new instance of a CLKernel for every involved Thread.
@see CLProgram#createCLKernel(java.lang.String) @see CLProgram#createCLKernels() @author Michael Bien

        @Override
        public IntBuffer execute(CLSingleProgramQueueContext qc) {
            
            CLCommandQueue queue = qc.getQueue();
            CLContext context = qc.getCLContext();
            CLKernel kernel = qc.getKernel("compute");


//            System.out.println(Thread.currentThread().getName()+" / "+queue);
            assertFalse(qc.isReleased());
            assertFalse(queue.isReleased());
            assertFalse(context.isReleased());
            assertFalse(kernel.isReleased());


            CLBuffer<IntBuffer> buffer = null;
            try{


                buffer = context.createBuffer(data);
                int gws = buffer.getCLCapacity();


                kernel.putArg(buffer).putArg(gws).rewind();


                queue.putWriteBuffer(buffer, false);
                queue.put1DRangeKernel(kernel, 0, gws, 0);
                queue.putReadBuffer(buffer, true);
            }finally{

View Full Code Here

            fillBuffer(clBufferA.getBuffer(), 12345);
            fillBuffer(clBufferB.getBuffer(), 67890);


            // get a reference to the kernel function with the name 'VectorAdd'
            // and map the buffers to its input parameters.
            CLKernel kernel = program.createCLKernel("VectorAdd");
            kernel.putArgs(clBufferA, clBufferB, clBufferC).putArg(elementCount);


            // asynchronous write of data to GPU device,
            // followed by blocking read to get the computed results back.
            long time = nanoTime();
            queue.putWriteBuffer(clBufferA, false)

View Full Code Here

            
            int localWorkSize = queue.getDevice().getMaxWorkGroupSize(); // Local work size dimensions
            int globalWorkSize = roundUp(localWorkSize, fb.capacity());  // rounded up to the nearest multiple of the localWorkSize
            
            // create kernel and set function parameters
            CLKernel kernel = program.createCLKernel("gamma");
            
            // original lenna
            show(image, 0, 50, "reference");
            
            // a few gamma corrected versions

View Full Code Here


        // TODO Only power-of-two array lengths are supported so far


        dir = (dir != 0) ? 1 : 0;


        CLKernel sortlocal1  = kernels.get(BITONIC_SORT_LOCAL1);
        CLKernel sortlocal   = kernels.get(BITONIC_SORT_LOCAL);
        CLKernel mergeGlobal = kernels.get(BITONIC_MERGE_GLOBAL);
        CLKernel mergeLocal  = kernels.get(BITONIC_MERGE_LOCAL);


        if (arrayLength <= LOCAL_SIZE_LIMIT) {


            //        oclCheckError( (batch * arrayLength) % LOCAL_SIZE_LIMIT == 0, shrTRUE );


            //Launch bitonicSortLocal
            sortlocal.putArgs(dstKey, srcKey)
                     .putArg(arrayLength).putArg(dir).rewind();


            int localWorkSize = LOCAL_SIZE_LIMIT / 2;
            int globalWorkSize = batch * arrayLength / 2;
            queue.put1DRangeKernel(sortlocal, 0, globalWorkSize, localWorkSize);


        } else {


            //Launch bitonicSortLocal1
            sortlocal1.setArgs(dstKey, srcKey);


            int localWorkSize = LOCAL_SIZE_LIMIT / 2;
            int globalWorkSize = batch * arrayLength / 2;


            queue.put1DRangeKernel(sortlocal1, 0, globalWorkSize, localWorkSize);


            for (int size = 2 * LOCAL_SIZE_LIMIT; size <= arrayLength; size <<= 1) {
                for (int stride = size / 2; stride > 0; stride >>= 1) {
                    if (stride >= LOCAL_SIZE_LIMIT) {
                        //Launch bitonicMergeGlobal
                        mergeGlobal.putArgs(dstKey, dstKey)
                                   .putArg(arrayLength).putArg(size).putArg(stride).putArg(dir).rewind();


                        globalWorkSize = batch * arrayLength / 2;
                        queue.put1DRangeKernel(mergeGlobal, 0, globalWorkSize, 0);
                    } else {
                        //Launch bitonicMergeLocal
                        mergeLocal.putArgs(dstKey, dstKey)
                                  .putArg(arrayLength).putArg(stride).putArg(size).putArg(dir).rewind();


                        localWorkSize = LOCAL_SIZE_LIMIT / 2;
                        globalWorkSize = batch * arrayLength / 2;

View Full Code Here

TOP

Related Classes of com.jogamp.opencl.CLKernel

com.jogamp.common.nio.NativeSizeBuffer

com.jogamp.opencl.demos.bitonicsort.BitonicSort

com.jogamp.opencl.demos.gamma.CLSimpleGammaCorrection

com.jogamp.opencl.demos.hellojocl.HelloJOCL

com.jogamp.opencl.util.concurrent.CLMultiContextTest$CLTestTask

java.nio.ByteBuffer

All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.