Examples of CLEvent

com.jogamp.opencl.CLEvent
Event objects can be used for synchronizing command queues, e.g you can wait until a event occurs or they can also be used to capture profiling information that measure execution time of a command. Profiling of OpenCL commands can be enabled by using a {@link com.jogamp.opencl.CLCommandQueue} created with{@link com.jogamp.opencl.CLCommandQueue.Mode#PROFILING_MODE}. @author Michael Bien
com.nativelibs4java.opencl.CLEvent

Examples of com.jogamp.opencl.CLEvent


            vectorAddKernel.setArgs(clBufferA, clBufferB, clBufferC); // C = A+B
            queue.put1DRangeKernel(vectorAddKernel, 0, elements, groupSize, events);


            assertEquals(1, events.size());
            CLEvent probe = events.getEvent(0);
            out.println(probe);


            queue.putWaitForEvents(events, true);
            assertEquals(CLEvent.ExecutionStatus.COMPLETE, probe.getStatus());


            out.println(probe);
            long time = probe.getProfilingInfo(CLEvent.ProfilingCommand.END)
                      - probe.getProfilingInfo(CLEvent.ProfilingCommand.START);
            out.println("time: "+time);
            assertTrue(time > 0);


            events.release();
        }finally{

View Full Code Here

Examples of com.jogamp.opencl.CLEvent


            queue.put1DRangeKernel(vectorAddKernel, 0, elements, groupSize, conditions, events);
            assertEquals(1, events.size());


            Thread.sleep(1000);
            final CLEvent status = events.getEvent(0);


            assertEquals(ExecutionStatus.QUEUED, status.getStatus());
            condition.setComplete();
            assertTrue(condition.isComplete());


            queue.finish();
            assertTrue(status.isComplete());


        }finally{
            context.release();
        }

View Full Code Here

Examples of com.jogamp.opencl.CLEvent

            textRenderer.draw("device/time/precision", 10, height-15);


            for (int i = 0; i < slices; i++) {
                CLDevice device = queues[i].getDevice();
                boolean doubleFP = doublePrecision && isDoubleFPAvailable(device);
                CLEvent event = probes.getEvent(i);
                long start = event.getProfilingInfo(START);
                long end = event.getProfilingInfo(END);
                textRenderer.draw(device.getType().toString()+i +" "
                               + (int)((end-start)/1000000.0f)+"ms @"
                               + (doubleFP?"64bit":"32bit"), 10, height-(20+16*(slices-i)));
            }

View Full Code Here

Examples of com.nativelibs4java.opencl.CLEvent

    
    this.weightsBuffer.set(weights);
    this.layerOutputBuffer.set(layerOutput);    
        
    // execute
    CLEvent kernelCompletion = null;
    // The same kernel can be safely used by different threads, as long as setArgs + enqueueNDRange are in a synchronized block
    synchronized (kernel) {
        kernel.setArgs(this.weightsBuffer.getCLBuffer(),paramCLBuffer,this.layerOutputBuffer.getCLBuffer(),this.outputBuffer.getCLBuffer());
        kernelCompletion = kernel.enqueueNDRange(queue, new int[] { outputSize }, new int[] { 1 } );
    }
    //kernelCompletion.waitFor(); // better not to wait for it but to pass it as a dependent event to some other queuable operation (CLBuffer.read, for instance)
    
    this.outputBuffer.getCLBuffer().map(queue, MapFlags.Write, kernelCompletion);
    this.outputBuffer.getBuffer().rewind();
    this.outputBuffer.getBuffer().get(layerOutputF, 0, outputSize);
    for(int i=0;i<outputSize;i++) {
      layerOutput[outputIndex+i] = layerOutputF[i];
    }
    
    if( kernelCompletion!=null )
      kernelCompletion.release();
    
    return startIndex+(inputSize*outputSize);
  }

View Full Code Here

Examples of com.nativelibs4java.opencl.CLEvent

              b3
      );
      
      
      
      CLEvent kernelCompletion;
      // The same kernel can be safely used by different threads, as long as setArgs + enqueueNDRange are in a synchronized block
      synchronized (kernel) {
          //kernel.setArgs(b1,b2,b3);
          kernelCompletion = kernel.enqueueNDRange(queue, new int[] { A.length }, new int[] { 1 } );
      }
      kernelCompletion.waitFor(); // better not to wait for it but to pass it as a dependent event to some other queuable operation (CLBuffer.read, for instance)
      
      FloatBuffer f = b3.read(queue, kernelCompletion);
      
      for(int i=0;i<A.length;i++) {
        System.out.println( A[i] + " * " + B[i] + " = " + f.get(i));

View Full Code Here

Examples of com.nativelibs4java.opencl.CLEvent


        // Set the args of the kernel :
        kernel.setArgs(inBuf, outBuf, length, forward ? 1 : -1);
        
        // Ask for `length` parallel executions of the kernel in 1 dimension :
        CLEvent dftEvt = kernel.enqueueNDRange(queue, new int[]{ length });


        // Return an NIO buffer read from the output CLBuffer :
        return outBuf.read(queue, dftEvt);
    }

View Full Code Here

Examples of com.nativelibs4java.opencl.CLEvent

        String src = IOUtils.readText(new File("matvec.cl"));
        CLProgram program = context.createProgram(src);
        CLKernel prod_escalar = program.createKernel("prod_escalar");
        prod_escalar.setArgs(buffer_v1, buffer_v2, buffer_v3, dataSize, dataSize);
        
        CLEvent prodEvt = prod_escalar.enqueueNDRange(queue, new int[] { dataSize });


        DoubleBuffer buffer_v4 = buffer_v3.read(queue,prodEvt);


        for(int i = 0; i < dataSize;i++){
            System.out.print(buffer_v4.get(i));

View Full Code Here

Examples of com.nativelibs4java.opencl.CLEvent

                 */
                // PRIMEIRA CAMADA
                kernelProdEscalar.setArgs(clBufferEntrada, clBufferPesos1, clBufferPesosBias1,
                        clBufferSaida1, qtdNeuronios_1, qtdNeuronios_1);
                //aqui diz quantos work itens trabalharao para executar o kernel
                CLEvent prodEvt = kernelProdEscalar.enqueueNDRange(queue, new int[]{qtdNeuronios_1});
                //faz a leitura do 'result'
                //DoubleBuffer dBufferResSaida1 = clBufferSaida1.read(queue,prodEvt);


                // SEGUNDA CAMADA
                clBufferSaida1 = context.createDoubleBuffer(CLMem.Usage.Input,
                        (DoubleBuffer) clBufferSaida1.read(queue, prodEvt), true);


                //TODO esta passando mais neuronios do que tem. Verica depois
                //pois nos testes funcionou. Tanto nos argumentos quanto no kernel
                kernelProdEscalar.setArgs(clBufferSaida1, clBufferPesos2, clBufferPesosBias2,
                        clBufferSaida2, qtdNeuronios_1, qtdNeuronios_1);


                prodEvt = kernelProdEscalar.enqueueNDRange(queue, new int[]{qtdNeuronios_1});


                //DoubleBuffer dBufferResSaida2 = clBufferSaida2.read(queue, prodEvt);


                //TERCEIRA CAMADA
                clBufferSaida2 = context.createDoubleBuffer(CLMem.Usage.Input,
                        (DoubleBuffer) clBufferSaida2.read(queue, prodEvt), true);


                //TODO esta passando mais neuronios do que tem. Verica depois
                //pois nos testes funcionou. Tanto nos argumentos quanto no kernel
                kernelProdEscalar.setArgs(clBufferSaida2, clBufferPesos3, clBufferPesosBias3,
                        clBufferSaida3, qtdNeuronios_2, qtdNeuronios_2);


                prodEvt = kernelProdEscalar.enqueueNDRange(queue, new int[]{qtdNeuronios_2});


                //DoubleBuffer dBufferSaidaFinal = clBufferSaida3.read(queue,prodEvt);


                // BACKPROPAGATION
                dSaidaFinal = ((DoubleBuffer) clBufferSaida3.read(queue, prodEvt)).get(0);
                erro = Param.target - dSaidaFinal;
                percentualErro = Math.abs((erro / Param.target) * 100);
                s3 = -2 * FuncoesCPU.derivativeSigmoid(dSaidaFinal) * erro;


                //kernel = program.createKernel("s2");
                kernelS2.setArgs(clBufferPesos3, dSaidaFinal, erro,
                        clBufferSaida2, clBufferS2);


                prodEvt = kernelS2.enqueueNDRange(queue, new int[]{qtdNeuronios_2});
                //DoubleBuffer dBufferResS2 = clBufferS2.read(queue,prodEvt);




                clBufferS2 = context.createDoubleBuffer(CLMem.Usage.InputOutput,
                        (DoubleBuffer) clBufferS2.read(queue, prodEvt), true);
                //kernel = program.createKernel("s1");
                kernelS1.setArgs(clBufferPesos2,
                        clBufferS2,
                        clBufferSaida1,
                        clBufferS1,
                        qtdNeuronios_2);


                prodEvt = kernelS1.enqueueNDRange(queue, new int[]{qtdNeuronios_1});
                //DoubleBuffer dBufferResS1 = clBufferS1.read(queue,prodEvt);


                clBufferS1 = context.createDoubleBuffer(CLMem.Usage.InputOutput,
                        (DoubleBuffer) clBufferS1.read(queue, prodEvt), true);




                //ATUALIZANDO OS PESOS
                //CAMADA 3
                //kernel = program.createKernel("atualiza_pesos_3");
                kernelAtualizaPesos3.setArgs(clBufferPesos3, clBufferPesos2, dSaidaFinal,
                        erro, Param.taxaAprendizado);
                prodEvt = kernelAtualizaPesos3.enqueueNDRange(queue, new int[]{qtdNeuronios_2});
                //DoubleBuffer dBufferResPesos3 = clBufferPesos3.read(queue,prodEvt);


                clBufferPesos3 = context.createDoubleBuffer(CLMem.Usage.InputOutput,
                        (DoubleBuffer) clBufferPesos3.read(queue, prodEvt), true);


                dBufferPesosBias3.put(0, dBufferPesosBias3.get(0) - (Param.taxaAprendizado * s3 * 1));
                clBufferPesosBias3 = context.createDoubleBuffer(CLMem.Usage.InputOutput, dBufferPesosBias3, true);


                //CAMADA 2
                //kernel = program.createKernel("atualiza_pesos_2");
                kernelAtualizaPesos2.setArgs(clBufferPesos2, clBufferS2, clBufferSaida1, qtdNeuronios_1,
                        Param.taxaAprendizado, clBufferPesosBias2);


                prodEvt = kernelAtualizaPesos2.enqueueNDRange(queue, new int[]{qtdNeuronios_2});
                //DoubleBuffer dBufferResPesos2 = clBufferPesos2.read(queue,prodEvt);
                //DoubleBuffer dBufferResPesosBias2 = clBufferPesosBias2.read(queue,prodEvt);


                clBufferPesos2 = context.createDoubleBuffer(CLMem.Usage.InputOutput,
                        (DoubleBuffer) clBufferPesos2.read(queue, prodEvt), true);
                clBufferPesosBias2 = context.createDoubleBuffer(CLMem.Usage.InputOutput,
                        (DoubleBuffer) clBufferPesosBias2.read(queue, prodEvt), true);


                //CAMADA 3
                //kernel = program.createKernel("atualiza_pesos_1");
                kernelAtualizaPesos1.setArgs(clBufferPesos1, clBufferS1, clBufferEntrada,
                        qtdNeuronios_1, Param.taxaAprendizado, clBufferPesosBias1);
                prodEvt = kernelAtualizaPesos1.enqueueNDRange(queue, new int[]{qtdNeuronios_1});
                //DoubleBuffer dBufferResPesos1 = clBufferPesos1.read(queue,prodEvt);
                //DoubleBuffer dBufferResPesosBias1 = clBufferPesosBias1.read(queue,prodEvt);


                clBufferPesos1 = context.createDoubleBuffer(CLMem.Usage.InputOutput,
                        (DoubleBuffer) clBufferPesos1.read(queue, prodEvt), true);
                clBufferPesosBias1 = context.createDoubleBuffer(CLMem.Usage.InputOutput,
                        (DoubleBuffer) clBufferPesosBias1.read(queue, prodEvt), true);


            }
        }


        System.out.println("TESTE");
        for (int e = indiceTeste; e < entradas.length; e++) {
            // PRIMEIRA CAMADA
            kernelProdEscalar.setArgs(clBufferEntrada, clBufferPesos1, clBufferPesosBias1,
                    clBufferSaida1, qtdNeuronios_1, qtdNeuronios_1);
            //aqui diz quantos work itens trabalharao para executar o kernel
            CLEvent prodEvt = kernelProdEscalar.enqueueNDRange(queue, new int[]{qtdNeuronios_1});
            //faz a leitura do 'result'
            //DoubleBuffer dBufferResSaida1 = clBufferSaida1.read(queue,prodEvt);


            // SEGUNDA CAMADA
            clBufferSaida1 = context.createDoubleBuffer(CLMem.Usage.Input,

View Full Code Here

TOP

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.