cavis/libnd4j/include/ops/declarable/generic/shape/expand_dims.cpp

103 lines
3.7 KiB
C++

/*******************************************************************************
* Copyright (c) 2015-2018 Skymind, Inc.
*
* 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.
*
* 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
******************************************************************************/
//
// Created by raver119 on 02.11.2017.
//
#include <op_boilerplate.h>
#if NOT_EXCLUDED(OP_expand_dims)
#include <ops/declarable/CustomOperations.h>
namespace nd4j {
namespace ops {
CUSTOM_OP_IMPL(expand_dims, 1, 1, false, 0, -2) {
auto input = INPUT_VARIABLE(0);
auto output = OUTPUT_VARIABLE(0);
if (input->isScalar()) {
output->assign(input);
return Status::OK();
}
Nd4jLong axis = block.numI() > 0 ? INT_ARG(0) : INPUT_VARIABLE(1)->e<int>(0);
if (axis < 0)
axis += input->rankOf() + 1;
REQUIRE_TRUE(axis >= 0 && axis <= input->rankOf()+1, 0, "ExpandDims: axis should be in range of 0...%i in this case, but got %i instead", input->rankOf() + 1, axis);
std::vector<Nd4jLong> shape(input->rankOf());
for(int e = 0; e < input->rankOf(); e++)
shape[input->sizeAt(e)];
shape.insert(shape.begin() + axis, 1);
if (input->ews() == 1 && output->ews() == 1 && input->ordering() == output->ordering()) {
output->dataBuffer()->copyBufferFrom(*input->dataBuffer().get(), output->lengthOf() * DataTypeUtils::sizeOfElement(output->dataType()), 0, input->bufferOffset());
} else {
auto tmp = input->reshape(input->ordering(), shape);
output->assign(tmp);
}
return Status::OK();
}
DECLARE_TYPES(expand_dims) {
getOpDescriptor()
->setAllowedInputTypes(nd4j::DataType::ANY)
->setSameMode(true);
}
DECLARE_SHAPE_FN(expand_dims) {
auto inShape = inputShape->at(0);
// 0D scalar edge case
if (shape::rank(inShape) == 0) {
Nd4jLong x = 1;
auto newShape = ConstantShapeHelper::getInstance()->createShapeInfo(ArrayOptions::dataType(inShape), 'c', 1, &x);
return SHAPELIST(newShape);
}
// FIXME: temp workaround for TF
if (shape::isScalar(inShape)) {
auto newShape = ConstantShapeHelper::getInstance()->createShapeInfo(ArrayOptions::dataType(inShape), 'c', 2, shape::shapeOf(inShape));
return SHAPELIST(newShape);
}
auto x_rank = shape::rank(inShape);
char order = shape::order(inShape);
Nd4jLong axis = block.numI() > 0 ? INT_ARG(0) : INPUT_VARIABLE(1)->e<int>(0);
if (axis < 0)
axis += x_rank + 1;
std::vector<Nd4jLong> shape;
for(int e = 0; e < x_rank; e++)
shape.emplace_back(shape::shapeOf(inShape)[e]);
shape.insert(shape.begin() + axis, 1);
auto newShape = ConstantShapeHelper::getInstance()->createShapeInfo(ArrayOptions::dataType(inShape), order, shape);
return SHAPELIST(newShape);
}
}
}
#endif