MKLDNNConv OpContext

deeplearning4j · Feb 25, 2019 · 8399e87 · 8399e87
1 parent 72a0251
commit 8399e87
Showing 1 changed file with 45 additions and 29 deletions.
diff --git a/...deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/mkldnn/MKLDNNConvHelper.java b/...deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/mkldnn/MKLDNNConvHelper.java
@@ -28,11 +28,13 @@
 import org.nd4j.linalg.activations.IActivation;
 import org.nd4j.linalg.api.ndarray.INDArray;
 import org.nd4j.linalg.api.ops.DynamicCustomOp;
+import org.nd4j.linalg.api.ops.OpContext;
 import org.nd4j.linalg.api.ops.impl.layers.convolution.Conv2D;
 import org.nd4j.linalg.api.ops.impl.layers.convolution.Conv2DDerivative;
 import org.nd4j.linalg.api.ops.impl.layers.convolution.config.Conv2DConfig;
 import org.nd4j.linalg.factory.Nd4j;
 import org.nd4j.linalg.primitives.Pair;
+import org.nd4j.linalg.util.ArrayUtil;
 
 import java.util.Collections;
 import java.util.Map;
@@ -43,6 +45,9 @@
  * @author Alex Black
  */
 public class MKLDNNConvHelper implements ConvolutionHelper {
+
+    protected OpContext context;
+
     @Override
     public boolean checkSupported() {
         return BaseMKLDNNHelper.mklDnnEnabled();
@@ -62,25 +67,32 @@ public Pair<Gradient, INDArray> backpropGradient(INDArray input, INDArray weight
                     kernel, strides, dilation);
         }
 
-        Conv2DConfig conf = Conv2DConfig.builder()
-                .dataFormat(Conv2DConfig.NCHW)
-                .kH(kernel[0]).kW(kernel[1])
-                .sH(strides[0]).sW(strides[1])
-                .pH(pad[0]).pW(pad[1])
-                .dH(dilation[0]).dH(dilation[1])
-                .isSameMode(convolutionMode == ConvolutionMode.Same)
-                .build();
+        if(context == null){
+            context = Nd4j.getExecutioner().buildContext();
+            context.setIArguments(kernel[0], kernel[1],
+                    strides[0], strides[1],
+                    pad[0], pad[1],
+                    dilation[0], dilation[1],
+                    ArrayUtil.fromBoolean(convolutionMode == ConvolutionMode.Same),
+                    0   //0=NCHW
+            );
+        };
 
         INDArray gradAtInput = workspaceMgr.createUninitialized(ArrayType.ACTIVATION_GRAD, input.dataType(), input.shape());
 
         INDArray[] inputsArr = biasGradView == null ? new INDArray[]{input, weightsPermute, delta} : new INDArray[]{input, weightsPermute, bias, delta};
         INDArray[] outputArr = biasGradView == null ? new INDArray[]{gradAtInput, weightGradViewPermute} : new INDArray[]{gradAtInput, weightGradViewPermute, biasGradView};
-        DynamicCustomOp op = Conv2DDerivative.derivativeBuilder()
-                .config(conf)
-                .inputArrays(inputsArr)
-                .outputs(outputArr)
-                .build();
-        Nd4j.exec(op);
+        context.getInputArrays().clear();
+        context.getOutputArrays().clear();
+        for( int i=0; i<inputsArr.length; i++ ){
+            context.setInputArray(i, inputsArr[i]);
+        }
+        for( int i=0; i<outputArr.length; i++ ){
+            context.setOutputArray(i, outputArr[i]);
+        }
+
+        Conv2DDerivative op = new Conv2DDerivative();
+        Nd4j.exec(op, context);
 
         Gradient g = new DefaultGradient();
         if(biasGradView != null) {
@@ -104,28 +116,32 @@ public INDArray preOutput(INDArray input, INDArray weights, INDArray bias, int[]
             outSize = ConvolutionUtils.getOutputSize(input, kernel, strides, pad, convolutionMode, dilation); //Also performs validation
         }
 
+        if(context == null){
+            context = Nd4j.getExecutioner().buildContext();
+            context.setIArguments(kernel[0], kernel[1],
+                    strides[0], strides[1],
+                    pad[0], pad[1],
+                    dilation[0], dilation[1],
+                    ArrayUtil.fromBoolean(convolutionMode == ConvolutionMode.Same),
+                    0   //0=NCHW
+            );
+        };
+
         int outDepth = (int) weights.size(0);
         INDArray out = workspaceMgr.createUninitialized(ArrayType.ACTIVATIONS, input.dataType(), input.size(0), outDepth, outSize[0], outSize[1]);
 
         //Note: conv2d op expects [kH, kW, iC, oC] weights... DL4J conv uses [oC, iC, kH, kW]
         weights = weights.permute(2,3,1,0);
 
-        Conv2DConfig conf = Conv2DConfig.builder()
-                .dataFormat(Conv2DConfig.NCHW)
-                .kH(kernel[0]).kW(kernel[1])
-                .sH(strides[0]).sW(strides[1])
-                .pH(pad[0]).pW(pad[1])
-                .dH(dilation[0]).dH(dilation[1])
-                .isSameMode(convolutionMode == ConvolutionMode.Same)
-                .build();
-
         INDArray[] inputsArr = bias == null ? new INDArray[]{input, weights} : new INDArray[]{input, weights, bias};
-        DynamicCustomOp op = Conv2D.builder()
-                .config(conf)
-                .inputArrays(inputsArr)
-                .outputs(new INDArray[]{out})
-                .build();
-        Nd4j.exec(op);
+        context.getInputArrays().clear();
+        for( int i=0; i<inputsArr.length; i++ ){
+            context.setInputArray(i, inputsArr[i]);
+        }
+        context.getOutputArrays().clear();
+        context.setOutputArray(0, out);
+        Conv2D op = new Conv2D();
+        Nd4j.exec(op, context);
 
         return out;
     }