Files
deeplearning4j--deeplearning4j/libnd4j/include/ops/declarable/generic/shape/transpose.cpp
T
2026-07-13 12:47:05 +08:00

134 lines
4.1 KiB
C++

/* ******************************************************************************
*
*
* 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
******************************************************************************/
//
// @author raver119@gmail.com
// @author Yurii Shyrma (iuriish@yahoo.com)
//
#include <system/op_boilerplate.h>
#if NOT_EXCLUDED(OP_transpose)
#include <helpers/ShapeUtils.h>
#include <ops/declarable/CustomOperations.h>
namespace sd {
namespace ops {
//////////////////////////////////////////////////////////////////////////
CUSTOM_OP_IMPL(transpose, 1, 1, false, 0, 0) {
auto x = INPUT_VARIABLE(0);
auto z = OUTPUT_VARIABLE(0);
// Special case: empty.reshape(<other empty shape>) -> return empty
if (x->isEmpty()) {
REQUIRE_TRUE(z->isEmpty(), 0, "TRANSPOSE OP: when input is empty, output must also be empty");
return Status::OK; // No op
}
NDArray* castedPermute = nullptr;
std::vector<LongType> permutationVector;
if (block.width() > 1) {
castedPermute = INPUT_VARIABLE(1)->cast(INT64);
permutationVector = castedPermute->asVectorT<LongType>();
} else {
permutationVector = *block.getIArguments();
}
if (permutationVector.size() == 0) {
NDArray *t = x->transpose();
z->assign(t);
// FIXED: transpose() returns a view - only delete if not a view
if (t != nullptr && !t->isView()) {
delete t;
}
if (castedPermute != nullptr) delete castedPermute;
return Status::OK;
}
bool isPermuteNecessary = false;
int rank = permutationVector.size();
//handles empty permute vector case as well as case where array rank and permute vector rank
//are different
for (LongType i = 0; i < rank; ++i) {
if (permutationVector[i] != i) {
isPermuteNecessary = true;
break;
}
}
if(!isPermuteNecessary) {
z->assign(x);
return Status::OK;
}
z->assign(x->permute(permutationVector, false, false));
if (castedPermute != nullptr) delete castedPermute;
return Status::OK;
}
DECLARE_TYPES(transpose) { getOpDescriptor()->setAllowedInputTypes(ANY)->setSameMode(true); }
DECLARE_SHAPE_FN(transpose) {
auto x = INPUT_VARIABLE(0);
const LongType rank = x->rankOf();
if(rank < 1)
return SHAPELIST(ConstantShapeHelper::getInstance().scalarShapeInfo(x->dataType()));
std::vector<LongType> permutationVector = block.width() > 1 ? INPUT_VARIABLE(1)->cast(INT64)->asVectorT<LongType>() : *block.getIArguments();
if (permutationVector.size() == 0) {
auto temp = ShapeUtils::evalTransposeShapeInfo(*x, nullptr, true);
auto ret = ConstantShapeHelper::getInstance().createFromExisting(temp);
RELEASE(temp, nullptr);
return SHAPELIST(ret);
}
bool isPermuteNecessary = false;
if(permutationVector.size() == static_cast<size_t>(rank))
for (LongType i = 0; i < rank; ++i) {
if (permutationVector[i] != i) {
isPermuteNecessary = true;
break;
}
}
if(!isPermuteNecessary) {
//note: do not deallocate this buffer. they are kept around.
auto permEvalShapeInfo = ConstantShapeHelper::getInstance().createFromExisting(inputShape->at(0));
return SHAPELIST(permEvalShapeInfo);
}
//note: do not deallocate this buffer. they are kept around.
auto permEvalShapeInfo = ShapeUtils::evalPermShapeInfo(permutationVector.data(), x->rankOf(), x, nullptr, true);
if(x->isEmpty()) {
ArrayOptions::setPropertyBit(permEvalShapeInfo, ARRAY_EMPTY);
}
auto ret = CONSTANT(permEvalShapeInfo);
return SHAPELIST(ret);
}
} // namespace ops
} // namespace sd
#endif