Java CLProgram.createKernel方法代码示例

import com.nativelibs4java.opencl.CLProgram; //导入方法依赖的package包/类
/**
 * Creates a GPUMiner object. This constructor should not be used - you
 * should instead use {@link
 * me.apemanzilla.krist.turbokrist.miners.MinerFactory}.
 * 
 * @param dev @param options @throws MinerInitException
 */
GPUMiner(CLDevice dev, MinerOptions options) throws MinerInitException {
	this.deviceName = dev.getName().trim();
	this.context = dev.getPlatform().createContext(null, new CLDevice[] { dev });
	this.queue = context.createDefaultQueue();
	ProgramBuilder pb = new ProgramBuilder("sha256.cl", "krist_miner.cl");
	CLProgram program;
	try {
		program = pb.build(context);
	} catch (ProgramBuildException e) {
		e.printStackTrace();
		throw new MinerInitException("Failed to build OpenCL program");
	}
	this.kernel = program.createKernel("krist_miner_basic");
	Pointer<Byte> addressPtr = Pointer.allocateBytes(10).order(context.getByteOrder());
	byte[] addressBytes = MinerUtils.getBytes(options.getKristAddress().getName());
	addressPtr.setArray(addressBytes);
	this.addressBuffer = context.createByteBuffer(Usage.Input, addressPtr);
	this.workSize = new int[] { options.getWorkSize(MinerFactory.generateSignature(dev)) };
}
 

import com.nativelibs4java.opencl.CLProgram; //导入方法依赖的package包/类
public synchronized CLKernel getKernel(Fun2 op, Primitive prim1, Primitive prim2, Primitive primOut, boolean secondOperandIsScalar) throws CLBuildException {
    Map<PrimitiveTrio, CLKernel> m = fun2Kernels.get(op);
    if (m == null)
        fun2Kernels.put(op, m = new HashMap<PrimitiveTrio, CLKernel>());

    PrimitiveTrio key = new PrimitiveTrio(prim1, prim2, primOut, secondOperandIsScalar);
    CLKernel ker = m.get(key);
    if (ker == null) {
        StringBuilder out = new StringBuilder(300);
        String name = createVectFun2Source(op, prim1, prim2, primOut, out, secondOperandIsScalar);
        CLProgram prog = getContext().createProgram(out.toString()).build();
        ker = prog.createKernel(name);
        m.put(key, ker);
    }
    return ker;
}
 

import com.nativelibs4java.opencl.CLProgram; //导入方法依赖的package包/类
private void loadKernels(CLProgram program) {
	addImage = program.createKernel("addImage");
	addConstant = program.createKernel("addConstant");
	subtractImage = program.createKernel("subtractImage");
	subtractConstant = program.createKernel("subtractConstant");
	multiplyImage = program.createKernel("multiplyImage");
	multiplyConstant = program.createKernel("multiplyConstant");
	divideImage = program.createKernel("divideImage");
	divideConstant = program.createKernel("divideConstant");
}
 

import com.nativelibs4java.opencl.CLProgram; //导入方法依赖的package包/类
private static long buildAndExecuteKernel(CLQueue queue, float realMin,
			float imaginaryMin, int realResolution, int imaginaryResolution,
			int maxIter, int magicNumber, float deltaReal,
			float deltaImaginary, Pointer<Integer> results, String src)
			throws CLBuildException, IOException {

		CLContext context = queue.getContext();
		long startTime = System.nanoTime();
//		if (useAutoGenWrapper) {
//			Mandelbrot mandelbrot = new Mandelbrot(context);
//			mandelbrot.mandelbrot(queue, new float[] { deltaReal,
//					deltaImaginary }, new float[] { realMin, imaginaryMin },
//
//			maxIter, magicNumber, realResolution,
//					context.createBuffer(CLMem.Usage.Output, results, false),
//
//					new int[] { realResolution, imaginaryResolution },
//					new int[] { 1, 1 });
//		} else {
			CLProgram program = context.createProgram(src).build();

			// Create a kernel instance from the mandelbrot kernel, passing in
			// parameters.
			CLKernel kernel = program.createKernel("mandelbrot", new float[] {
					deltaReal, deltaImaginary }, new float[] { realMin,
					imaginaryMin },

			maxIter, magicNumber, realResolution,
					context.createBuffer(CLMem.Usage.Output, results, false));

			// Enqueue and complete work using a 2D range of work groups
			// corrsponding to individual pizels in the set.
			// The work groups are 1x1 in size and their range is defined by the
			// desired resolution. This corresponds
			// to one device thread per pixel.

			kernel.enqueueNDRange(queue, new int[] { realResolution,
					imaginaryResolution }, new int[] { 1, 1 });
//		}
		queue.finish();
		long time = System.nanoTime() - startTime;
		return time;

	}
 

import com.nativelibs4java.opencl.CLProgram; //导入方法依赖的package包/类
public static void main(String[] args) throws IOException {

        CLContext context = JavaCL.createBestContext();
        CLQueue clQueue = context.createDefaultQueue();
        ByteOrder byteOrder = context.getByteOrder();

        int n = 1024;
        Pointer<Float> aPtr = allocateFloats(n).order(byteOrder);

        for (int i = 0; i < n; i++) {
            aPtr.set(i, (float)cos(i));
        }

        // Create OpenCL input/output buffers (using the native memory pointers aPtr and bPtr) :
        CLBuffer<Float> a = context.createBuffer(Usage.InputOutput, aPtr);

        // Read the program sources and compile them :
        String src = 
                    "__kernel void add_floats(global float* a, int n) {\n" +
                    "      int i = get_global_id(0);\n" +
                    "      if(i < n){\n" +
                    "      	a[i] = 2.f*a[i];\n" +
                    "       }\n" +
                    "}";

    //IOUtils.readText(new File("TutorialKernels.cl"));
        CLProgram program = context.createProgram(src).build();
        
        // Get and call the kernel :
        CLKernel addFloatsKernel = program.createKernel("add_floats");

        addFloatsKernel.setArgs(a, n);
        CLEvent evt = addFloatsKernel.enqueueNDRange(clQueue, new int[] { n });

        aPtr =  a.read(clQueue, evt); // blocks until add_floats finished
      
        // Print the first 10 output values :
        for (int i = 0; i < 10 && i < n; i++) {
            System.out.println("out[" + i + "] = " + aPtr.get(i));
        }

    }
 

import com.nativelibs4java.opencl.CLProgram; //导入方法依赖的package包/类
private static long buildAndExecuteKernel(CLQueue queue, float realMin, float imaginaryMin, int realResolution,
                                            int imaginaryResolution, int maxIter, int magicNumber, float deltaReal,
                                            float deltaImaginary, Pointer<Integer> results, String src) throws CLBuildException, IOException {

      CLContext context = queue.getContext();
      long startTime = System.nanoTime();
      if (useAutoGenWrapper) {
          Mandelbrot mandelbrot = new Mandelbrot(context);
          mandelbrot.mandelbrot(
              queue,
              new float[] { deltaReal, deltaImaginary },
              new float[] { realMin, imaginaryMin },

              maxIter,
              magicNumber,
              realResolution,
              context.createBuffer(CLMem.Usage.Output, results, false),

              new int[]{realResolution, imaginaryResolution},
              new int[]{1,1}
          );
      } else {
          CLProgram program = context.createProgram(src).build();

          //Create a kernel instance from the mandelbrot kernel, passing in parameters.
          CLKernel kernel = program.createKernel(
                  "mandelbrot",
                  new float[] { deltaReal, deltaImaginary },
                  new float[] { realMin, imaginaryMin },

                  maxIter,
                  magicNumber,
                  realResolution,
                  context.createBuffer(CLMem.Usage.Output, results, false)
          );

          //Enqueue and complete work using a 2D range of work groups corrsponding to individual pizels in the set.
          //The work groups are 1x1 in size and their range is defined by the desired resolution. This corresponds
          //to one device thread per pixel.

          kernel.enqueueNDRange(queue, new int[]{realResolution, imaginaryResolution}, new int[]{1,1});
      }
      queue.finish();
long time = System.nanoTime() - startTime;
return time;

  }
 

import com.nativelibs4java.opencl.CLProgram; //导入方法依赖的package包/类
public <T> CLEvent matrixTranspose(Primitive prim, CLBuffer<T> a, long aRows, long aColumns, long aStride, CLBuffer<T> out, CLEvent... eventsToWaitFor) throws CLBuildException {
    if (out == null)
        throw new IllegalArgumentException("Null output matrix !");
    //if (out != null)
    //    out = (CLBuffer<T>)context.createBuffer(Usage.Output, prim.primitiveType, aRows * aColumns);

    CLKernel[] kernels;
    synchronized (matrixTransposeKernels) {
        kernels = matrixTransposeKernels.get(prim);
        if (kernels == null) {
            String src =
            	prim.getRequiredPragmas() +
                "__kernel void transposeSelf(                                   \n" +
                "   __global double* a, int aRows, int aColumns, int aStride    \n" +
                ") {                                                            \n" +
                "    int i = get_global_id(0);                                  \n" +
                "    int j = get_global_id(1);                                  \n" +
                "                                                               \n" +
                "    if (i >= aRows || j >= aColumns || j >= i) return;         \n" +
                "                                                               \n" +
                "    size_t aIndex = i * aStride + j;                           \n" +
                "    size_t outIndex = j * aRows + i;                           \n" +
                "    double temp = a[outIndex];                                 \n" +
                "    a[outIndex] = a[aIndex];                                   \n" +
                "    a[aIndex] = temp;                                          \n" +
                "}                                                              \n" +
                "__kernel void transposeOther(                                  \n" +
                "   __global const double* a, int aRows, int aColumns, int aStride, \n" +
                "   __global double* out                                        \n" +
                ") {                                                            \n" +
                "    int i = get_global_id(0);                                  \n" +
                "    int j = get_global_id(1);                                  \n" +
                "                                                               \n" +
                "    if (i >= aRows || j >= aColumns) return;                   \n" +
                "                                                               \n" +
                "    size_t aIndex = i * aStride + j;                           \n" +
                "    size_t outIndex = j * aRows + i;                           \n" +
                "    out[outIndex] = a[aIndex];                                 \n" +
                "}                                                              \n"
            ;
            String clTypeName = prim.clTypeName();
            src = src.replaceAll("double", clTypeName);
            CLProgram program = context.createProgram(src);
            kernels = new CLKernel[] { program.createKernel("transposeSelf"), program.createKernel("transposeOther") };
            matrixTransposeKernels.put(prim, kernels);
        }
    }
    boolean self = a.equals(out);
    CLKernel kernel = kernels[self ? 0 : 1];
    synchronized (kernel) {
        if (self)
            kernel.setArgs(a, (int)aRows, (int)aColumns, (int)aStride);
        else
            kernel.setArgs(a, (int)aRows, (int)aColumns, (int)aStride, out);
        
        CLEvent evt = kernel.enqueueNDRange(queue, new int [] { (int)aRows, (int)aColumns }, eventsToWaitFor);
        return evt;
    }
}