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

119 lines
3.3 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
//
#include <system/op_boilerplate.h>
#if NOT_EXCLUDED(OP_subtract)
#include <ops/declarable/CustomOperations.h>
#include <ops/declarable/generic/helpers/BroadcastHelper.h>
namespace sd {
namespace ops {
BROADCASTABLE_OP_IMPL(subtract, 0, 0) {
auto x = INPUT_VARIABLE(0);
auto y = INPUT_VARIABLE(1);
auto z = OUTPUT_VARIABLE(0);
BROADCAST_CHECK_EMPTY(x, y, z);
auto tZ = BroadcastHelper::broadcastApply(BroadcastOpsTuple::Subtract(), x, y, z);
if (tZ == nullptr)
return Status::KERNEL_FAILURE;
else if (tZ != z) {
OVERWRITE_RESULT(tZ);
}
return Status::OK;
}
DECLARE_SYN(Sub, subtract);
DECLARE_SYN(sub, subtract);
DECLARE_TYPES(subtract) {
getOpDescriptor()
->setAllowedInputTypes(0, ANY)
->setAllowedInputTypes(1, ANY)
->setAllowedOutputTypes(0, INHERIT);
}
CUSTOM_OP_IMPL(subtract_bp, 3, 2, false, 0, 0) {
auto x = INPUT_VARIABLE(0);
auto y = INPUT_VARIABLE(1);
auto epsNext = INPUT_VARIABLE(2);
auto gradX = OUTPUT_VARIABLE(0);
auto gradY = OUTPUT_VARIABLE(1);
if (x->isSameShape(y)) {
// PWT case case
epsNext->applyTransform(transform::Neg, gradY);
gradX->assign(epsNext);
} else if (y->isScalar()) {
// scalar case
auto reduce = epsNext->reduceNumber(reduce::Sum);
auto tmp = -(*reduce);
gradY->assign(&tmp);
gradX->assign(epsNext);
delete reduce;
} else {
// broadcastable
auto axisX = ShapeUtils::evalBroadcastBackwardAxis(x->shapeInfo(), epsNext->shapeInfo());
auto axisY = ShapeUtils::evalBroadcastBackwardAxis(y->shapeInfo(), epsNext->shapeInfo());
if (axisX.size() > 0) {
auto sum = epsNext->reduceAlongDimension(reduce::Sum, &axisX);
gradX->assign(sum);
delete sum;
} else
gradX->assign(epsNext);
if (axisY.size() > 0) {
auto sum = epsNext->reduceAlongDimension(reduce::Sum, &axisY);
sum->applyTransform(transform::Neg, gradY);
delete sum;
} else {
epsNext->applyTransform(transform::Neg, gradY);
}
}
return Status::OK;
}
DECLARE_TYPES(subtract_bp) {
getOpDescriptor()->setAllowedInputTypes(ANY)->setAllowedOutputTypes({ALL_FLOATS});
}
DECLARE_SHAPE_FN(subtract_bp) {
auto x = inputShape->at(0);
auto y = inputShape->at(1);
auto e = inputShape->at(2);
// eps always has shape of x
// grad always has shape of y
auto shapeList = SHAPELIST(CONSTANT(x), CONSTANT(y));
return shapeList;
}
} // namespace ops
} // namespace sd
#endif