nd4j/nd4j-backends/nd4j-api-parent/nd4j-api/src/main/java/org/nd4j/linalg/api/ops/impl/shape/Reshape.java
/*
* ******************************************************************************
* *
* *
* * This program and the accompanying materials are made available under the
* * terms of the Apache License, Version 2.0 which is available at
* * https://www.apache.org/licenses/LICENSE-2.0.
* *
* * See the NOTICE file distributed with this work for additional
* * information regarding copyright ownership.
* * Unless required by applicable law or agreed to in writing, software
* * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
* * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
* * License for the specific language governing permissions and limitations
* * under the License.
* *
* * SPDX-License-Identifier: Apache-2.0
* *****************************************************************************
*/
package org.nd4j.linalg.api.ops.impl.shape;
import lombok.NoArgsConstructor;
import lombok.NonNull;
import lombok.extern.slf4j.Slf4j;
import lombok.val;
import onnx.Onnx;
import org.nd4j.autodiff.samediff.SDVariable;
import org.nd4j.autodiff.samediff.SameDiff;
import org.nd4j.common.base.Preconditions;
import org.nd4j.imports.descriptors.properties.PropertyMapping;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.ops.CustomOp;
import org.nd4j.linalg.api.ops.DynamicCustomOp;
import org.nd4j.linalg.api.ops.OpContext;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.shade.guava.primitives.Longs;
import org.tensorflow.framework.AttrValue;
import org.tensorflow.framework.GraphDef;
import org.tensorflow.framework.NodeDef;
import java.util.Collections;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
@Slf4j
@NoArgsConstructor
public class Reshape extends DynamicCustomOp {
private long[] shape;
public final static int C_ORDER = -99;
public final static int F_ORDER = -102;
private transient boolean reshapeWithViewPossible = false;
public Reshape(SameDiff sameDiff, SDVariable i_v, long[] shape) {
super(null, sameDiff, new SDVariable[]{i_v});
this.shape = shape;
//c ordering: see (char) 99 for c ordering and (char) 'f' is 102
//note it has to be negative for the long array case only
//to flag the difference between an ordering being specified
//and a dimension.
if(iArguments.isEmpty())
addIArgument(C_ORDER);
addIArgument(shape);
this.reshapeWithViewPossible = org.nd4j.linalg.api.shape.Shape.ableToReshapeWithView(i_v.getArr(), iArguments.get(0) == F_ORDER, Longs.toArray(iArguments.subList(1,iArguments.size())));
}
public Reshape(SameDiff sameDiff, SDVariable i_v, long[] shape,char c) {
super(null, sameDiff, new SDVariable[]{i_v});
Preconditions.checkState(c == 'c' || c == 'f', "Invalid order: must be 'c' or 'f', got %s", c);
this.shape = shape;
//c ordering: see (char) 99 for c ordering and (char) 'f' is 102
//note it has to be negative for the long array case only
//to flag the difference between an ordering being specified
//and a dimension.
addIArgument(-c);
addIArgument(shape);
}
public Reshape(SameDiff sameDiff, SDVariable i_v, SDVariable shape) {
super(null, sameDiff, new SDVariable[]{i_v, shape});
if(iArguments.isEmpty())
addIArgument(C_ORDER);
}
public Reshape(INDArray in, long... shape) {
super(new INDArray[]{in}, null);
this.shape = shape;
//c ordering: see (char) 99 for c ordering and (char) 'f' is 102
//note it has to be negative for the long array case only
//to flag the difference between an ordering being specified
//and a dimension.
if(iArguments.isEmpty())
addIArgument(C_ORDER);
addIArgument(shape);
this.reshapeWithViewPossible = org.nd4j.linalg.api.shape.Shape.ableToReshapeWithView(in, iArguments.get(0) == F_ORDER, Longs.toArray(iArguments.subList(1,iArguments.size())));
}
public Reshape(INDArray in, char order,long... shape) {
super(new INDArray[]{in}, null);
Preconditions.checkState(order == 'c' || order == 'f', "Invalid order: must be 'c' or 'f', got %s", order);
this.shape = shape;
//c ordering: see (char) 99 for c ordering and (char) 'f' is 102
//note it has to be negative for the long array case only
//to flag the difference between an ordering being specified
//and a dimension.
addIArgument(-order);
addIArgument(shape);
this.reshapeWithViewPossible = org.nd4j.linalg.api.shape.Shape.ableToReshapeWithView(in, iArguments.get(0) == F_ORDER, Longs.toArray(iArguments.subList(1,iArguments.size())));
}
public Reshape(@NonNull INDArray in, @NonNull INDArray shape, INDArray out) {
super(null, new INDArray[]{in, shape}, wrapOrNull(out), null, (List<Long>)null);
if(iArguments.isEmpty())
addIArgument(C_ORDER);
this.reshapeWithViewPossible = org.nd4j.linalg.api.shape.Shape.ableToReshapeWithView(in, iArguments.get(0) == F_ORDER,shape.toLongVector());
}
public Reshape(INDArray in, INDArray shape) {
this(in, shape, null);
}
@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
if (!nodeDef.containsAttr("TShape") && nodeDef.getInputCount() == 1) {
this.shape = new long[]{};
return;
} else if(nodeDef.getInputCount() == 1){
val shape = nodeDef.getAttrOrThrow("Tshape");
if (!shape.hasShape()) {
val shapeRet = new long[2];
shapeRet[0] = 1;
shapeRet[1] = shape.getValueCase().getNumber();
this.shape = shapeRet;
} else {
val shapeVals = shape.getShape().getDimList();
if (shapeVals.size() > 1) {
this.shape = new long[shapeVals.size()];
for (int i = 0; i < shapeVals.size(); i++) {
this.shape[i] = (int) shapeVals.get(i).getSize();
}
} else {
this.shape = new long[2];
this.shape[0] = 1;
this.shape[1] = (int) shapeVals.get(0).getSize();
}
}
//all TF is c
if (this.shape != null) {
addIArgument(this.shape);
}
}
}
@Override
public void initFromOnnx(Onnx.NodeProto node, SameDiff initWith, Map<String, Onnx.AttributeProto> attributesForNode, Onnx.GraphProto graph) {
}
@Override
public Map<String, Map<String, PropertyMapping>> mappingsForFunction() {
Map<String, Map<String, PropertyMapping>> ret = new HashMap<>();
Map<String, PropertyMapping> map = new HashMap<>();
val shapeMapping = PropertyMapping.builder()
.onnxAttrName("shape")
.tfInputPosition(-1)
.propertyNames(new String[]{"shape"})
.build();
map.put("shape", shapeMapping);
ret.put(tensorflowName(), map);
ret.put(onnxName(), map);
return ret;
}
@Override
public String opName() {
return "reshape";
}
@Override
public String onnxName() {
return "Reshape";
}
@Override
public String tensorflowName() {
return "Reshape";
}
@Override
public void configureFromArguments() {
if(iArguments.size() > 1) {
//ordering comes first followed by the actual shape
this.shape = new long[iArguments.size() - 1];
for(int i = 0; i < shape.length; i++) {
this.shape[i] = iArguments.get(i + 1);
}
this.reshapeWithViewPossible = org.nd4j.linalg.api.shape.Shape.ableToReshapeWithView(getInputArgument(0), iArguments.get(0) == F_ORDER, Longs.toArray(iArguments.subList(1,iArguments.size())));
}
}
@Override
public void setPropertiesForFunction(Map<String, Object> properties) {
}
@Override
public List<SDVariable> doDiff(List<SDVariable> i_v) {
SDVariable origShape = sameDiff.shape(arg());
SDVariable ret = sameDiff.reshape(i_v.get(0), origShape);
return Collections.singletonList(ret);
}
@Override
public boolean initializeOutputs(OpContext ctx) {
if(!reshapeWithViewPossible)
return super.initializeOutputs(ctx);
else {
char newOrder = (char) -iArguments.get(0);
if(inputArguments.size() > 1)
shape = inputArguments.get(1).toLongVector();
//wrap an existing buffer to ensure that the original buffer doesn't get deallocated
INDArray arr = Nd4j.create(Nd4j.createBuffer(inputArguments().get(0).data(),0,inputArguments().get(0).data().length()),shape,Nd4j.getStrides(shape,newOrder),0,newOrder);
addOutputArgument(arr);
return false;
}
}
@Override
public List<DataType> calculateOutputDataTypes(List<DataType> dataTypes) {
//Output type is always same as input type
return Collections.singletonList(dataTypes.get(0));
}
}