You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
65 lines
2.3 KiB
65 lines
2.3 KiB
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://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. */
|
|
|
|
#include "paddle/fluid/operators/gather.cu.h"
|
|
#include "paddle/fluid/operators/gather_op.h"
|
|
#include "paddle/fluid/operators/scatter.cu.h"
|
|
#include "paddle/fluid/operators/scatter_op.h"
|
|
|
|
namespace paddle {
|
|
namespace operators {
|
|
|
|
template <typename T>
|
|
class ScatterOpCUDAKernel : public framework::OpKernel<T> {
|
|
public:
|
|
void Compute(const framework::ExecutionContext &ctx) const override {
|
|
PADDLE_ENFORCE(platform::is_gpu_place(ctx.GetPlace()),
|
|
"This kernel only runs on GPU device.");
|
|
auto *X = ctx.Input<Tensor>("X");
|
|
auto *Ids = ctx.Input<Tensor>("Ids");
|
|
auto *Updates = ctx.Input<Tensor>("Updates");
|
|
auto *Out = ctx.Output<Tensor>("Out");
|
|
|
|
Out->ShareDataWith(*X);
|
|
|
|
GPUScatterAssign<T>(ctx.device_context(), *Updates, *Ids, Out);
|
|
}
|
|
};
|
|
|
|
template <typename T>
|
|
class ScatterGradOpCUDAKernel : public framework::OpKernel<T> {
|
|
public:
|
|
void Compute(const framework::ExecutionContext &ctx) const override {
|
|
PADDLE_ENFORCE(platform::is_gpu_place(ctx.GetPlace()),
|
|
"This kernel only runs on GPU device.");
|
|
auto *dX = ctx.Output<Tensor>(framework::GradVarName("X"));
|
|
auto *dUpdates = ctx.Output<Tensor>(framework::GradVarName("Updates"));
|
|
auto *Ids = ctx.Input<Tensor>("Ids");
|
|
auto *dOut = ctx.Input<Tensor>(framework::GradVarName("Out"));
|
|
|
|
// In place gradient: dX = dO
|
|
dX->ShareDataWith(*dOut);
|
|
dUpdates->mutable_data<T>(ctx.GetPlace());
|
|
// Gradient by Gather: dUpdates = dO[Ids]
|
|
GPUGather<T>(ctx.device_context(), *dOut, *Ids, dUpdates);
|
|
}
|
|
};
|
|
|
|
} // namespace operators
|
|
} // namespace paddle
|
|
|
|
namespace ops = paddle::operators;
|
|
REGISTER_OP_CUDA_KERNEL(scatter, ops::ScatterOpCUDAKernel<float>);
|
|
REGISTER_OP_CUDA_KERNEL(scatter_grad, ops::ScatterGradOpCUDAKernel<float>);
|