利用JOCL进行opencl开发之helloworld

2019独角兽企业重金招聘Python工程师标准>>>

JOCL(Java bindings for OpenCL)提供的API保持与OpenCL的原始API尽可能的相近。功能全部采用静态方法实现，语义和这些方法的签名与原来的库函数保持一致。除了一些Java语言的特殊限制。

之所以选JOCL 

1 本人java开发,不懂C++ 

2 本人笔记本上只有AMD渣显卡

之前选过aparapi但没成功一直报错,可能是显卡原因.

本地显卡是AMD Radeon HD 7600M系列 驱动程序版本15.200.1062.1004  本地CPU是i7 3740QM 

首先 安装好正确的驱动 然后安装AMD提供的AMD APP SDK 2.9 目前版本是2.9  这俩自行去官网下载.

http://support.amd.com/en-us/kb-articles/Pages/OpenCL2-Driver.aspx

成功后cmd下执行clinfo -v  能出来结果说明安装ok了

然后下载jocl

http://www.jocl.org/downloads/downloads.html

具体参考

https://jogamp.org/jocl/doc/HowToBuild.html

如果不想自己编译源码  直接用archive 下载地址

http://jogamp.org/deployment/jogamp-current/archive/

下载这个最大的文件后解压

取\jogamp-all-platforms\jar下面的gluegen-rt.jar  gluegen-rt-natives-windows-amd64.jar  jocl.jar  jocl-natives-windows-amd64.jar这四个jar到自己项目的lib中

然后写个demo 用下面的代码

import com.jogamp.opencl.CLBuffer;
import com.jogamp.opencl.CLCommandQueue;
import com.jogamp.opencl.CLContext;
import com.jogamp.opencl.CLDevice;
import com.jogamp.opencl.CLKernel;
import com.jogamp.opencl.CLProgram;

import java.io.IOException;
import java.nio.FloatBuffer;
import java.util.Random;

import static java.lang.System.*;
import static com.jogamp.opencl.CLMemory.Mem.*;
import static java.lang.Math.*;

/**
 * Hello Java OpenCL example. Adds all elements of buffer A to buffer B
 * and stores the result in buffer C.<br/>
 * Sample was inspired by the Nvidia VectorAdd example written in C/C++
 * which is bundled in the Nvidia OpenCL SDK.
 *
 * @author Michael Bien
 */
public class HelloJOCL {

    public static void main(String[] args) throws IOException {

        // set up (uses default CLPlatform and creates context for all devices)
        CLContext context = CLContext.create();
        out.println("created "+context);

        // always make sure to release the context under all circumstances
        // not needed for this particular sample but recommented
        try{

            // select fastest device
            CLDevice device = null;
            device = context.getMaxFlopsDevice(CLDevice.Type.GPU);
            //device = context.getMaxFlopsDevice(CLDevice.Type.CPU);
            out.println("using "+device);

            // create command queue on device.
            CLCommandQueue queue = device.createCommandQueue();

            int elementCount = 59449477;                                  // Length of arrays to process
            int localWorkSize = min(device.getMaxWorkGroupSize(), 256);  // Local work size dimensions
            int globalWorkSize = roundUp(localWorkSize, elementCount);   // rounded up to the nearest multiple of the localWorkSize

            // load sources, create and build program
            CLProgram program = context.createProgram(HelloJOCL.class.getResourceAsStream("VectorAdd.cl")).build();

            // A, B are input buffers, C is for the result
            CLBuffer<FloatBuffer> clBufferA = context.createFloatBuffer(globalWorkSize, READ_ONLY);
            CLBuffer<FloatBuffer> clBufferB = context.createFloatBuffer(globalWorkSize, READ_ONLY);
            CLBuffer<FloatBuffer> clBufferC = context.createFloatBuffer(globalWorkSize, WRITE_ONLY);

            out.println("used device memory: "
                    + (clBufferA.getCLSize()+clBufferB.getCLSize()+clBufferC.getCLSize())/1000000 +"MB");

            // fill input buffers with random numbers
            // (just to have test data; seed is fixed -> results will not change between runs).
            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)
                    .putWriteBuffer(clBufferB, false)
                    .put1DRangeKernel(kernel, 0, globalWorkSize, localWorkSize)
                    .putReadBuffer(clBufferC, true);
            time = nanoTime() - time;

            // print first few elements of the resulting buffer to the console.
            out.println("a+b=c results snapshot: ");
            for(int i = 0; i < 10; i++)
                out.print(clBufferC.getBuffer().get() + ", ");
            out.println("...; " + clBufferC.getBuffer().remaining() + " more");

            out.println("computation took: "+(time/1000000)+"ms");

        }finally{
            // cleanup all resources associated with this context.
            context.release();
        }

    }

    private static void fillBuffer(FloatBuffer buffer, int seed) {
        Random rnd = new Random(seed);
        while(buffer.remaining() != 0)
            buffer.put(rnd.nextFloat()*100);
        buffer.rewind();
    }

    private static int roundUp(int groupSize, int globalSize) {
        int r = globalSize % groupSize;
        if (r == 0) {
            return globalSize;
        } else {
            return globalSize + groupSize - r;
        }
    }

}

这个程序大概算5000万次加法运算吧 

device = context.getMaxFlopsDevice(CLDevice.Type.GPU);

这里可以指定用CPU还是GPU

测试后如图

,没优化代码情况下, 这个显卡还没我CPU算的快

还有其它的demo 自己看官网文档即可

 

 

 

 

转载于:https://my.oschina.net/qutterrtl/blog/1531327