feature/DC asgd (#12722)
* wip * add ref_by_trainer_id op * ready to test * fix ref inputs * refine rpc_op_handle * fix merge bugfix_recordio_link
Wu Yi
7 years ago
committed by
GitHub
parent
c3cbf0b8ef
commit
306236c2c0
@ -0,0 +1,79 @@
|
|||||||
|
/* 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/ref_by_trainer_id_op.h"
|
||||||
|
#include <string>
|
||||||
|
|
||||||
|
namespace paddle {
|
||||||
|
namespace operators {
|
||||||
|
|
||||||
|
class RefByTrainerIdOp : public framework::OperatorWithKernel {
|
||||||
|
public:
|
||||||
|
RefByTrainerIdOp(const std::string &type,
|
||||||
|
const framework::VariableNameMap &inputs,
|
||||||
|
const framework::VariableNameMap &outputs,
|
||||||
|
const framework::AttributeMap &attrs)
|
||||||
|
: OperatorWithKernel(type, inputs, outputs, attrs) {}
|
||||||
|
|
||||||
|
void InferShape(framework::InferShapeContext *ctx) const override {
|
||||||
|
PADDLE_ENFORCE(ctx->HasInputs("X"),
|
||||||
|
"Input(X) of RefByTrainerIdOp should not be null.");
|
||||||
|
PADDLE_ENFORCE(ctx->HasInput("TrainerId"),
|
||||||
|
"Input(TrainerId) of RefByTrainerIdOp should not be null.");
|
||||||
|
PADDLE_ENFORCE(ctx->HasOutput("Out"),
|
||||||
|
"Output(Out) of RefByTrainerIdOp should not be null.");
|
||||||
|
PADDLE_ENFORCE_EQ(ctx->GetInputDim("TrainerId").size(), 1,
|
||||||
|
"TrainerId should be a scalar.");
|
||||||
|
// Out's shape is determined at runtime.
|
||||||
|
}
|
||||||
|
|
||||||
|
protected:
|
||||||
|
framework::OpKernelType GetExpectedKernelType(
|
||||||
|
const framework::ExecutionContext &ctx) const override {
|
||||||
|
return framework::OpKernelType(
|
||||||
|
framework::ToDataType(
|
||||||
|
ctx.MultiInput<framework::Tensor>("X")[0]->type()),
|
||||||
|
ctx.GetPlace());
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
class RefByTrainerIdOpMaker : public framework::OpProtoAndCheckerMaker {
|
||||||
|
public:
|
||||||
|
void Make() override {
|
||||||
|
AddInput("X", "(Tensor) Input tensor list.").AsDuplicable();
|
||||||
|
AddInput("TrainerId", "(Tensor) Scalar int, the trainer id runtime value.");
|
||||||
|
AddOutput("Out", "(Tensor) Return one tensor reference of X[trainer_id]");
|
||||||
|
AddComment(R"DOC(
|
||||||
|
**RefByTrainerId operator**
|
||||||
|
|
||||||
|
Return a reference of a tensor, using trainer_id as the index to find from the input.
|
||||||
|
|
||||||
|
$$Out = X[TrainerId]$$
|
||||||
|
)DOC");
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
} // namespace operators
|
||||||
|
} // namespace paddle
|
||||||
|
|
||||||
|
namespace ops = paddle::operators;
|
||||||
|
|
||||||
|
REGISTER_OP_WITHOUT_GRADIENT(ref_by_trainer_id, ops::RefByTrainerIdOp,
|
||||||
|
ops::RefByTrainerIdOpMaker);
|
||||||
|
REGISTER_OP_CPU_KERNEL(
|
||||||
|
ref_by_trainer_id,
|
||||||
|
ops::RefByTrainerIdKernel<paddle::platform::CPUDeviceContext, float>,
|
||||||
|
ops::RefByTrainerIdKernel<paddle::platform::CPUDeviceContext, double>,
|
||||||
|
ops::RefByTrainerIdKernel<paddle::platform::CPUDeviceContext, int>,
|
||||||
|
ops::RefByTrainerIdKernel<paddle::platform::CPUDeviceContext, int64_t>);
|
||||||
@ -0,0 +1,26 @@
|
|||||||
|
/* 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/ref_by_trainer_id_op.h"
|
||||||
|
|
||||||
|
REGISTER_OP_CUDA_KERNEL(
|
||||||
|
ref_by_trainer_id,
|
||||||
|
paddle::operators::RefByTrainerIdKernel<paddle::platform::CUDADeviceContext,
|
||||||
|
float>,
|
||||||
|
paddle::operators::RefByTrainerIdKernel<paddle::platform::CUDADeviceContext,
|
||||||
|
double>,
|
||||||
|
paddle::operators::RefByTrainerIdKernel<paddle::platform::CUDADeviceContext,
|
||||||
|
int>,
|
||||||
|
paddle::operators::RefByTrainerIdKernel<paddle::platform::CUDADeviceContext,
|
||||||
|
int64_t>);
|
||||||
@ -0,0 +1,49 @@
|
|||||||
|
/* 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. */
|
||||||
|
|
||||||
|
#pragma once
|
||||||
|
|
||||||
|
#include <stdio.h>
|
||||||
|
#include "paddle/fluid/framework/op_registry.h"
|
||||||
|
|
||||||
|
namespace paddle {
|
||||||
|
namespace operators {
|
||||||
|
template <typename DeviceContext, typename T>
|
||||||
|
class RefByTrainerIdKernel : public framework::OpKernel<T> {
|
||||||
|
public:
|
||||||
|
virtual void Compute(const framework::ExecutionContext& context) const {
|
||||||
|
auto* out = context.Output<framework::Tensor>("Out");
|
||||||
|
auto in_list = context.MultiInput<framework::Tensor>("X");
|
||||||
|
auto* trainer_id_t = context.Input<framework::Tensor>("TrainerId");
|
||||||
|
int64_t trainer_id;
|
||||||
|
auto* trainer_id_data = trainer_id_t->data<int64_t>();
|
||||||
|
if (platform::is_gpu_place(context.GetPlace())) {
|
||||||
|
#ifdef PADDLE_WITH_CUDA
|
||||||
|
auto stream = context.cuda_device_context().stream();
|
||||||
|
memory::Copy<>(platform::CPUPlace(), &trainer_id,
|
||||||
|
boost::get<platform::CUDAPlace>(context.GetPlace()),
|
||||||
|
trainer_id_data, sizeof(int64_t), stream);
|
||||||
|
#endif
|
||||||
|
} else {
|
||||||
|
trainer_id = *trainer_id_data;
|
||||||
|
}
|
||||||
|
printf("after get trainer_id %lu\n", trainer_id);
|
||||||
|
PADDLE_ENFORCE_LT(trainer_id, in_list.size());
|
||||||
|
out->mutable_data<T>(context.GetPlace());
|
||||||
|
out->ShareDataWith(*(in_list[trainer_id]));
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
} // namespace operators
|
||||||
|
} // namespace paddle
|
||||||
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in new issue