Package com.nativelibs4java.opencl

Examples of com.nativelibs4java.opencl.CLContext

16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
import com.nativelibs4java.util.NIOUtils;

public class SimpleTest {
  public static void main(String[] args) {
    System.out.println("Start");
    CLContext context = JavaCL.createBestContext();
    CLQueue queue = context.createDefaultQueue();

    float[] A = { 1.0f, 2.0f, 3.0f, 4.0f };
    float[] B = { 5.0f, 6.0f, 7.0f, 8.0f };
       
    String myKernelSource = ResourceInputStream.readResourceAsString("org/encog/plugins/opencl/kernels/simple.cl");
   
    FloatBuffer array1 = NIOUtils.directFloats(10, context.getByteOrder());
    FloatBuffer array2 = NIOUtils.directFloats(10, context.getByteOrder());
    FloatBuffer resultArray = NIOUtils.directFloats(10, context.getByteOrder());
   
    array1.put(A);
    array2.put(B);
   
    CLFloatBuffer b1 = context.createFloatBuffer(Usage.Input, array1, true);
    CLFloatBuffer b2 = context.createFloatBuffer(Usage.Input, array2, true);
    CLFloatBuffer b3 = context.createFloatBuffer(Usage.Output, resultArray, false);

    CLProgram program;
    try {
      program = context.createProgram(myKernelSource).build();
      CLKernel kernel = program.createKernel(
              "simpleKernel",
              b1,
              b2,
              b3
View Full Code Here
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
    }

    public static void main(String[] args) throws IOException, CLBuildException {
     // Create a context with the best double numbers support possible :
     // (try using DeviceFeature.GPU, DeviceFeature.CPU...)
        CLContext context = JavaCL.createBestContext(DeviceFeature.DoubleSupport);
       
        // Create a command queue, if possible able to execute multiple jobs in parallel
        // (out-of-order queues will still respect the CLEvent chaining)
        CLQueue queue = context.createDefaultOutOfOrderQueueIfPossible();

        DFT dft = new DFT(queue);
        //DFT2 dft = new DFT2(queue);

        // Create some fake test data :
View Full Code Here
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
        System.out.println("");
        System.out.println("");
       
        int dataSize = v1.length;
       
        CLContext context = JavaCL.createBestContext();
        CLQueue queue = context.createDefaultQueue();
        ByteOrder byteOrder = context.getByteOrder();
       
       
        DoubleBuffer v1_b = NIOUtils.directDoubles(dataSize, byteOrder);
        DoubleBuffer v2_b = NIOUtils.directDoubles(v2.length, byteOrder);
       
        FuncoesGPU.preencheBuffer(v1_b, v1);
        FuncoesGPU.preencheBuffer(v2_b, v2);       
       
        CLBuffer<Double> buffer_v1 = context.createDoubleBuffer(CLMem.Usage.Input,v1_b, true);
        CLBuffer<Double> buffer_v2 = context.createDoubleBuffer(CLMem.Usage.Input,v2_b, true);
        CLBuffer<Double> buffer_v3 = context.createDoubleBuffer(CLMem.Usage.Output, dataSize);
       
        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 });
View Full Code Here
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
        //leitura dos dados de entrada
        double[][] entradas = FuncoesCPU.lerArquivoEntradas(Param.nomeArquivo);

        //configuracao do opencl via JavaCL
        CLContext context = JavaCL.createBestContext();
        CLQueue queue = context.createDefaultQueue();
        ByteOrder byteOrder = context.getByteOrder();

        ArrayList<CLBuffer<Double>> clBufferEntradas = new ArrayList<CLBuffer<Double>>(entradas.length);
        for (int i = 0; i < entradas.length; i++) {
            DoubleBuffer dBufferEntrada = NIOUtils.directDoubles(entradas[i].length, byteOrder);
            FuncoesGPU.preencheBuffer(dBufferEntrada, entradas[i]);
            clBufferEntradas.add(context.createDoubleBuffer(CLMem.Usage.Input, dBufferEntrada, true));
        }

        CLBuffer<Double> clBufferEntrada = null;

        //leitura do arquivo cl e compilacao do programa
        String src = IOUtils.readText(new File("matvec.cl"));
        CLProgram program = context.createProgram(src);
        //CLKernel kernel = null;
        //CLEvent prodEvt = null;
        CLKernel kernelProdEscalar = program.createKernel("prod_escalar");
        CLKernel kernelS2 = program.createKernel("s2");
        CLKernel kernelS1 = program.createKernel("s1");
        CLKernel kernelAtualizaPesos3 = program.createKernel("atualiza_pesos_3");
        CLKernel kernelAtualizaPesos2 = program.createKernel("atualiza_pesos_2");
        CLKernel kernelAtualizaPesos1 = program.createKernel("atualiza_pesos_1");

        //----------------------------VARIAVEIS DA 1a CAMADA
        int qtdNeuronios_1 = 12;
        //gerado como vetor para facilitar o uso no kernel
        double[] pesos_1 = FuncoesCPU.gerarVetorAleatorio(qtdNeuronios_1 * qtdNeuronios_1, Param.min, Param.max);
        double[] pesos_1_bias = FuncoesCPU.gerarVetorAleatorio(qtdNeuronios_1, Param.min, Param.max);
        //double[] saida_camada_1 = new double[qtdNeuronios_1];

        DoubleBuffer dBufferPesos1 = NIOUtils.directDoubles(pesos_1.length * pesos_1.length, byteOrder);
        DoubleBuffer dBufferPesosBias1 = NIOUtils.directDoubles(pesos_1_bias.length, byteOrder);

        FuncoesGPU.preencheBuffer(dBufferPesos1, pesos_1);
        FuncoesGPU.preencheBuffer(dBufferPesosBias1, pesos_1_bias);

        CLBuffer<Double> clBufferPesos1 = context.createDoubleBuffer(CLMem.Usage.InputOutput, dBufferPesos1, true);
        CLBuffer<Double> clBufferPesosBias1 = context.createDoubleBuffer(CLMem.Usage.InputOutput, dBufferPesosBias1, true);
        CLBuffer<Double> clBufferSaida1 = context.createDoubleBuffer(CLMem.Usage.InputOutput, qtdNeuronios_1);

        //----------------------------VARIAVEIS DA 2a CAMADA
        int qtdNeuronios_2 = 6;
        double[] pesos_2 = FuncoesCPU.gerarVetorAleatorio(qtdNeuronios_2 * qtdNeuronios_1, Param.min, Param.max);
        double[] pesos_2_bias = FuncoesCPU.gerarVetorAleatorio(qtdNeuronios_2, Param.min, Param.max);
        //double[] saida_camada_2 = new double[qtdNeuronios_2];

        DoubleBuffer dBufferPesos2 = NIOUtils.directDoubles(qtdNeuronios_2 * qtdNeuronios_1, byteOrder);
        DoubleBuffer dBufferPesosBias2 = NIOUtils.directDoubles(pesos_2_bias.length, byteOrder);

        FuncoesGPU.preencheBuffer(dBufferPesos2, pesos_2);
        FuncoesGPU.preencheBuffer(dBufferPesosBias2, pesos_2_bias);

        CLBuffer<Double> clBufferPesos2 = context.createDoubleBuffer(CLMem.Usage.InputOutput, dBufferPesos2, true);
        CLBuffer<Double> clBufferPesosBias2 = context.createDoubleBuffer(CLMem.Usage.InputOutput, dBufferPesosBias2, true);
        CLBuffer<Double> clBufferSaida2 = context.createDoubleBuffer(CLMem.Usage.InputOutput, qtdNeuronios_2);

        //----------------------------VARIAVEIS DA 3a CAMADA
        int qtdNeuronios_3 = 1;
        double[] dvPesos3 = FuncoesCPU.gerarVetorAleatorio(qtdNeuronios_3 * qtdNeuronios_2, Param.min, Param.max);
        double[] dvPesosBias3 = FuncoesCPU.gerarVetorAleatorio(qtdNeuronios_3, Param.min, Param.max);

        DoubleBuffer dBufferPesos3 = NIOUtils.directDoubles(qtdNeuronios_3 * qtdNeuronios_2, byteOrder);
        DoubleBuffer dBufferPesosBias3 = NIOUtils.directDoubles(dvPesosBias3.length, byteOrder);

        FuncoesGPU.preencheBuffer(dBufferPesos3, dvPesos3);
        FuncoesGPU.preencheBuffer(dBufferPesosBias3, dvPesosBias3);

        CLBuffer<Double> clBufferPesos3 = context.createDoubleBuffer(CLMem.Usage.InputOutput, dBufferPesos3, true);
        CLBuffer<Double> clBufferPesosBias3 = context.createDoubleBuffer(CLMem.Usage.InputOutput, dBufferPesosBias3, true);
        CLBuffer<Double> clBufferSaida3 = context.createDoubleBuffer(CLMem.Usage.Output, qtdNeuronios_3);

        // VARIAVEIS DO BACKPROPAGATION
        CLBuffer<Double> clBufferS2 = context.createDoubleBuffer(CLMem.Usage.InputOutput, qtdNeuronios_2);
        CLBuffer<Double> clBufferS1 = context.createDoubleBuffer(CLMem.Usage.InputOutput, qtdNeuronios_1);
        double dSaidaFinal = 0.0;
        double erro = 0.0;
        double s3 = 0.0;

        int epocas = 1000;
        int tamanhoTreinamento = (int) (entradas.length * 0.85);
        int indiceTeste = tamanhoTreinamento;

        long init = System.currentTimeMillis();
        double percentualErro = 0.0;
        for (int epoca = 0; epoca < epocas; epoca++) {

            for (int e = 0; e < tamanhoTreinamento; e++) {

                //TODO possivel ponto de latencia. Para toda entrada vai ter outro
                // 'for' somente para preencher o buffer.
                //DoubleBuffer dBufferEntrada = NIOUtils.directDoubles(entradas[e].length, byteOrder);
                //FuncoesGPU.preencheBuffer(dBufferEntrada, entradas[e]);
                //CLBuffer<Double> clBufferEntrada = context.createDoubleBuffer(CLMem.Usage.Input, dBufferEntrada, true);

                clBufferEntrada = clBufferEntradas.get(0);

                //kernel = program.createKernel("prod_escalar");

                /*args
                 * input
                 * pesos
                 * pesos bias
                 * result
                 * quantidade de neuronios
                 * quantidade pesos por neuronio
                 */
                // 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,
                    (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,
View Full Code Here
TOP

Related Classes of com.nativelibs4java.opencl.CLContext

  • org.encog.plugins.opencl.example.DFT
  • org.encog.plugins.opencl.example.SimpleTest
  • rnaopencl.RNAOpencl2
  • rnaopencl.Testes
    • Copyright © 2018 www.massapicom. All rights reserved.
    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.