tensor-tang
7 years ago
commit
35e820dc2b
@ -0,0 +1,51 @@
|
||||
// Copyright (c) 2018 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/framework/details/fuse_vars_op_handle.h"
|
||||
|
||||
namespace paddle {
|
||||
namespace framework {
|
||||
namespace details {
|
||||
|
||||
void FuseVarsOpHandle::RunImpl() {
|
||||
WaitInputVarGenerated(place_);
|
||||
|
||||
auto in_var_handles = DynamicCast<VarHandle>(this->Inputs());
|
||||
auto out_var_handles = DynamicCast<VarHandle>(this->Outputs());
|
||||
PADDLE_ENFORCE_EQ(in_var_handles.size(), 0);
|
||||
PADDLE_ENFORCE_EQ(out_var_handles.size() - 1, inputs_numel_.size(), "");
|
||||
|
||||
auto scope = local_scope_->FindVar(kLocalExecScopeName)->Get<Scope *>();
|
||||
|
||||
auto out_var_handle = out_var_handles[0];
|
||||
auto out_var = scope->Var(out_var_handle->name_);
|
||||
|
||||
auto out_tensor = out_var->GetMutable<LoDTensor>();
|
||||
out_tensor->Resize({total_numel_}).mutable_data(this->place_, type_);
|
||||
|
||||
int64_t s = 0;
|
||||
for (size_t i = 1; i < out_var_handles.size(); ++i) {
|
||||
auto out_name = out_var_handles[i]->name_;
|
||||
auto out_t = scope->Var(out_name)->GetMutable<LoDTensor>();
|
||||
auto numel = this->inputs_numel_.at(out_name);
|
||||
out_t->ShareDataWith(out_tensor->Slice(s, s + numel));
|
||||
s += numel;
|
||||
}
|
||||
this->RunAndRecordEvent([this] {});
|
||||
}
|
||||
|
||||
std::string FuseVarsOpHandle::Name() const { return "fuse vars"; }
|
||||
} // namespace details
|
||||
} // namespace framework
|
||||
} // namespace paddle
|
||||
@ -0,0 +1,63 @@
|
||||
// Copyright (c) 2018 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 <map>
|
||||
#include <string>
|
||||
#include <vector>
|
||||
|
||||
#include "paddle/fluid/framework/details/container_cast.h"
|
||||
#include "paddle/fluid/framework/details/op_handle_base.h"
|
||||
#include "paddle/fluid/framework/lod_tensor.h"
|
||||
#include "paddle/fluid/framework/scope.h"
|
||||
#include "paddle/fluid/platform/device_context.h"
|
||||
|
||||
namespace paddle {
|
||||
namespace framework {
|
||||
namespace details {
|
||||
|
||||
struct FuseVarsOpHandle : public OpHandleBase {
|
||||
public:
|
||||
FuseVarsOpHandle(Scope *local_scope, const platform::Place &place,
|
||||
const std::unordered_map<std::string, int64_t> &inputs_numel,
|
||||
const std::type_index &var_type)
|
||||
: local_scope_(local_scope),
|
||||
place_(place),
|
||||
inputs_numel_(inputs_numel),
|
||||
type_(var_type) {
|
||||
total_numel_ = 0;
|
||||
for (auto in_numel : inputs_numel) {
|
||||
PADDLE_ENFORCE_GT(in_numel.second, 0);
|
||||
total_numel_ += in_numel.second;
|
||||
}
|
||||
}
|
||||
|
||||
std::string Name() const override;
|
||||
|
||||
bool IsMultiDeviceTransfer() override { return false; };
|
||||
|
||||
protected:
|
||||
void RunImpl() override;
|
||||
|
||||
private:
|
||||
Scope *local_scope_;
|
||||
const platform::Place place_;
|
||||
const std::unordered_map<std::string, int64_t> inputs_numel_;
|
||||
const std::type_index type_;
|
||||
int64_t total_numel_;
|
||||
};
|
||||
} // namespace details
|
||||
} // namespace framework
|
||||
} // namespace paddle
|
||||
@ -1,106 +1,47 @@
|
||||
/* Copyright (c) 2018 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
|
||||
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
|
||||
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. */
|
||||
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 <gtest/gtest.h>
|
||||
#include "paddle/fluid/framework/lod_tensor.h"
|
||||
#include "paddle/fluid/framework/op_registry.h"
|
||||
#include "paddle/fluid/framework/program_desc.h"
|
||||
#include "paddle/fluid/inference/tensorrt/convert/io_converter.h"
|
||||
#include "paddle/fluid/inference/tensorrt/convert/op_converter.h"
|
||||
#include "paddle/fluid/platform/device_context.h"
|
||||
#include "paddle/fluid/platform/place.h"
|
||||
|
||||
USE_OP(relu);
|
||||
#include "paddle/fluid/inference/tensorrt/convert/ut_helper.h"
|
||||
|
||||
namespace paddle {
|
||||
namespace inference {
|
||||
namespace tensorrt {
|
||||
|
||||
void Compare(const std::string op_type, float input, float expect) {
|
||||
TEST(ReluOpConverter, main) {
|
||||
framework::Scope scope;
|
||||
platform::CUDAPlace place;
|
||||
platform::CUDADeviceContext ctx(place);
|
||||
|
||||
// init fluid op and variable
|
||||
auto x_var = scope.Var("X");
|
||||
auto x_tensor = x_var->GetMutable<framework::LoDTensor>();
|
||||
x_tensor->Resize({1, 1});
|
||||
x_tensor->mutable_data<float>(place);
|
||||
std::vector<float> init;
|
||||
init.push_back(input);
|
||||
framework::TensorFromVector(init, ctx, x_tensor);
|
||||
|
||||
auto out_var = scope.Var("Out");
|
||||
auto out_tensor = out_var->GetMutable<framework::LoDTensor>();
|
||||
out_tensor->Resize({1, 1});
|
||||
out_tensor->mutable_data<float>(place);
|
||||
|
||||
framework::OpDesc op_desc;
|
||||
op_desc.SetType(op_type);
|
||||
op_desc.SetInput("X", {"X"});
|
||||
op_desc.SetOutput("Out", {"Out"});
|
||||
|
||||
auto op = framework::OpRegistry::CreateOp(*op_desc.Proto());
|
||||
|
||||
// run fluid op
|
||||
op->Run(scope, place);
|
||||
// get fluid output
|
||||
std::vector<float> out1;
|
||||
framework::TensorToVector(*out_tensor, ctx, &out1);
|
||||
|
||||
// init tensorrt op
|
||||
cudaStream_t stream;
|
||||
ASSERT_EQ(0, cudaStreamCreate(&stream));
|
||||
TensorRTEngine* engine = new TensorRTEngine(1, 1 << 10, &stream);
|
||||
engine->InitNetwork();
|
||||
engine->DeclareInput("X", nvinfer1::DataType::kFLOAT,
|
||||
nvinfer1::DimsCHW{1, 1, 1});
|
||||
// convert op
|
||||
OpConverter op_converter;
|
||||
op_converter.ConvertOp(*op_desc.Proto(), engine);
|
||||
|
||||
engine->DeclareOutput("Out");
|
||||
engine->FreezeNetwork();
|
||||
|
||||
// convert LoDTensor to ITensor
|
||||
size_t size = x_tensor->memory_size();
|
||||
EngineIOConverter::ConvertInput(op_type, *x_tensor,
|
||||
engine->buffer("X").buffer, size, &stream);
|
||||
// run tensorrt Outp
|
||||
engine->Execute(1);
|
||||
// convert ITensor to LoDTensor
|
||||
EngineIOConverter::ConvertOutput(op_type, engine->buffer("Out").buffer,
|
||||
out_tensor, size, &stream);
|
||||
// get tensorrt output
|
||||
std::vector<float> out2;
|
||||
framework::TensorToVector(*out_tensor, ctx, &out2);
|
||||
|
||||
// compare
|
||||
ASSERT_EQ(out1[0], out2[0]);
|
||||
ASSERT_EQ(out1[0], expect);
|
||||
|
||||
delete engine;
|
||||
cudaStreamDestroy(stream);
|
||||
}
|
||||
|
||||
TEST(OpConverter, ConvertRelu) {
|
||||
Compare("relu", 1, 1); // relu(1) = 1
|
||||
Compare("relu", -5, 0); // relu(-5) = 0
|
||||
std::unordered_set<std::string> parameters;
|
||||
TRTConvertValidation validator(10, parameters, scope, 1000);
|
||||
validator.DeclInputVar("relu-X", nvinfer1::Dims2(10, 6));
|
||||
validator.DeclOutputVar("relu-Out", nvinfer1::Dims2(10, 6));
|
||||
|
||||
// Prepare Op description
|
||||
framework::OpDesc desc;
|
||||
desc.SetType("relu");
|
||||
desc.SetInput("X", {"relu-X"});
|
||||
desc.SetOutput("Out", {"relu-Out"});
|
||||
|
||||
LOG(INFO) << "set OP";
|
||||
validator.SetOp(*desc.Proto());
|
||||
LOG(INFO) << "execute";
|
||||
|
||||
validator.Execute(10);
|
||||
}
|
||||
|
||||
} // namespace tensorrt
|
||||
} // namespace inference
|
||||
} // namespace paddle
|
||||
|
||||
USE_OP(activation);
|
||||
USE_OP(relu);
|
||||
|
||||
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in new issue