Overview (JavaCL Core / BridJ 1.0.0-RC4 API)

Packages
Package	Description
com.nativelibs4java.opencl	Object-oriented wrappers around the OpenCL API (`JavaCL`, `CLPlatform`, `CLContext`, `CLProgram`, `CLKernel`, `CLBuffer`...)
com.ochafik.util.string

JavaCL is an Object-Oriented API that wraps the OpenCL API.
It makes it easy to do cross-platform GPGPU with Java.

Core classes :

JavaCL: entry point of JavaCL, useful to list platforms, create contexts.
CLContext: all the JavaCL data and programs must be created from a context instance.
CLQueue: OpenCL operations can be executed immediately or queued. A queue instance is however needed even for immediate operations.
CLMem: base class for buffers and 2D and 3D images.

JavaCL relies on OpenCL4Java, low-level bindings that closely match the OpenCL C API and use JNA+JNAerator for the native access interop.

Here's a very simple vector addition JavaCL program :

package com.nativelibs4java.opencl.demos.vectoradd;
import com.nativelibs4java.opencl.*;
import java.nio.*;

public class VectorAdd {

    public static void main(String[] args) {
        try {
            FloatBuffer a = FloatBuffer.wrap(new float[] {  1,  2,  3,  4 });
            FloatBuffer b = FloatBuffer.wrap(new float[] { 10, 20, 30, 40 });

            FloatBuffer sum = add(a, b);
            for (int i = 0, n = sum.capacity(); i < n; i++)
                System.out.println(sum.get(i));
        } catch (Exception ex) {
            ex.printStackTrace();
        }
    }

    public static FloatBuffer add(FloatBuffer a, FloatBuffer b) throws CLBuildException {
        int n = a.capacity();
        
        CLContext context = JavaCL.createBestContext();
        CLQueue queue = context.createDefaultQueue();
        
        String source = 
            "__kernel void addFloats(__global const float* a, __global const float* b, __global float* output)     " +
            "{                                                                                                     " +
            "   int i = get_global_id(0);                                                                          " +
            "   output[i] = a[i] + b[i];                                                                           " +
            "}                                                                                                     ";
        
        CLKernel kernel = context.createProgram(source).createKernel("addFloats");
        CLFloatBuffer aBuf = context.createFloatBuffer(CLMem.Usage.Input, a, true);
        CLFloatBuffer bBuf = context.createFloatBuffer(CLMem.Usage.Input, b, true);
        CLFloatBuffer outBuf = context.createFloatBuffer(CLMem.Usage.Output, n);
        kernel.setArgs(aBuf, bBuf, outBuf);
        
        kernel.enqueueNDRange(queue, new int[]{n}, new int[] { 1 });
        queue.finish();
    
        return outBuf.read(queue);
    }
}