Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add data transform fn #6953

Merged
merged 19 commits into from
Dec 26, 2017
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 4 additions & 1 deletion paddle/framework/CMakeLists.txt
Original file line number Diff line number Diff line change
Expand Up @@ -64,4 +64,7 @@ cc_test(threadpool_test SRCS threadpool_test.cc DEPS threadpool)
cc_library(init SRCS init.cc DEPS gflags device_context place stringpiece)
cc_test(init_test SRCS init_test.cc DEPS init)

cc_test(op_kernel_type_test SRCS op_kernel_type_test.cc DEPS place device_context)
cc_test(op_kernel_type_test SRCS op_kernel_type_test.cc DEPS place device_context framework_proto)

cc_library(data_transform SRCS data_transform.cc DEPS tensor framework_proto)
cc_test(data_transform_test SRCS data_transform_test.cc DEPS data_transform device_context)
26 changes: 26 additions & 0 deletions paddle/framework/data_transform.cc
Original file line number Diff line number Diff line change
@@ -0,0 +1,26 @@
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.

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/framework/data_transform.h"

namespace paddle {
namespace framework {

DataTransformFnMap& DataTransformFnMap::Instance() {
static DataTransformFnMap data_transform_map;
return data_transform_map;
}

} // namespace framework
} // namespace paddle
110 changes: 110 additions & 0 deletions paddle/framework/data_transform.h
Original file line number Diff line number Diff line change
@@ -0,0 +1,110 @@
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.

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 <functional>
#include <utility>
#include <vector>

#include "paddle/framework/op_kernel_type.h"
#include "paddle/framework/tensor.h"
#include "paddle/framework/variable.h"
#include "paddle/platform/device_context.h"
#include "paddle/platform/macros.h"

namespace paddle {
namespace framework {

using DataTransformFN =
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

DataTransform should not only use Tensor as its argument type.

How about SelectedRows? LoDTensor? vector<Scope>? It should take Variable as its argument.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ok

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

std::function<void(const std::vector<platform::DeviceContext*> ctx,
const Variable& in, Variable* out)>;
using KernelTypePair = std::pair<OpKernelType, OpKernelType>;

static void hash_combine(std::size_t& seed, const OpKernelType& t) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code style is:

template <class T>
inline void HashCombine(const T& v, std::size_t* seed) {
  std::hash<T> hasher;
  *seed ^= hasher(v) + 0x9e3779b9 + (*seed << 6) + (*seed >> 2);
}

We can fix it later.

OpKernelType::Hash kernel_type_hasher;
seed ^= kernel_type_hasher(t) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
}

struct KernelTypePairHash {
size_t operator()(const KernelTypePair& kernel_pair) const {
std::size_t seed = 0;
hash_combine(seed, kernel_pair.first);
hash_combine(seed, kernel_pair.second);

return seed;
}
};

using DataTransformMap =
std::unordered_map<KernelTypePair, DataTransformFN, KernelTypePairHash>;

class DataTransformFnMap {
public:
static DataTransformFnMap& Instance();

bool Has(const KernelTypePair& key_pair) const {
return map_.find(key_pair) != map_.end();
}

void Insert(const OpKernelType& left, const OpKernelType& right,
const DataTransformFN& data_tranform_fn) {
Insert(std::make_pair(left, right), data_tranform_fn);
}

void Insert(const KernelTypePair& kernel_type_pair,
const DataTransformFN& data_tranform_fn) {
PADDLE_ENFORCE(!Has(kernel_type_pair),
"KernelTypePair %s has been registered", "");
map_.insert({kernel_type_pair, data_tranform_fn});
}

const DataTransformFN& Get(const KernelTypePair& key_pair) const {
auto data_transformer = GetNullable(key_pair);
PADDLE_ENFORCE_NOT_NULL(data_transformer,
"DataTransformFN should not be NULL");
return *data_transformer;
}

const DataTransformFN* GetNullable(const KernelTypePair& key_pair) const {
auto it = map_.find(key_pair);
if (it == map_.end()) {
return nullptr;
} else {
return &(it->second);
}
}

const DataTransformMap& Map() const { return map_; }

private:
DataTransformFnMap() = default;
DataTransformMap map_;
DISABLE_COPY_AND_ASSIGN(DataTransformFnMap);
};

// generate unique name with __LINE__
// refs https://stackoverflow.com/questions/1597007
#define TOKENPASTE(x, y) x##y
#define TOKENPASTE2(x, y) TOKENPASTE(x, y)
#define REGISTER_DATA_TRANSFORM_FN(from, to, fn) \
static int TOKENPASTE2(fn_, __LINE__)() { \
::paddle::framework::DataTransformFnMap::Instance().Insert(from, to, fn); \
return 0; \
} \
static int TOKENPASTE2(var_, __LINE__) __attribute__((unused)) = \
TOKENPASTE2(fn_, __LINE__)()

} // namespace framework
} // namespace paddle
78 changes: 78 additions & 0 deletions paddle/framework/data_transform_test.cc
Original file line number Diff line number Diff line change
@@ -0,0 +1,78 @@
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.

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/framework/data_transform.h"
#include <gtest/gtest.h>

namespace paddle {
namespace framework {

using namespace platform;

int test_value = 0;

OpKernelType kernel_type_1(proto::DataType::FP32, CPUPlace(), DataLayout::kNCHW,
LibraryType::kCUDNN);
OpKernelType kernel_type_2(proto::DataType::FP32, CUDAPlace(0),
DataLayout::kNCHW, LibraryType::kCUDNN);
OpKernelType kernel_type_3(proto::DataType::FP16, CUDAPlace(0),
DataLayout::kNCHW, LibraryType::kCUDNN);

void type1_to_type2(std::vector<platform::DeviceContext*> ctx,
const Variable& in, Variable* out) {
test_value++;
}

void type2_to_type3(std::vector<platform::DeviceContext*> ctx,
const Variable& in, Variable* out) {
test_value--;
}

void type1_to_type3(std::vector<platform::DeviceContext*> ctx,
const Variable& in, Variable* out) {
test_value += 2;
}

} // namespace framework
} // namespace paddle

namespace frw = paddle::framework;

REGISTER_DATA_TRANSFORM_FN(frw::kernel_type_1, frw::kernel_type_2,
frw::type1_to_type2);
REGISTER_DATA_TRANSFORM_FN(frw::kernel_type_2, frw::kernel_type_3,
frw::type2_to_type3);
REGISTER_DATA_TRANSFORM_FN(frw::kernel_type_1, frw::kernel_type_3,
frw::type1_to_type3);

TEST(DataTransform, Register) {
using namespace paddle::framework;
using namespace paddle::platform;

auto& instance = DataTransformFnMap::Instance();
ASSERT_EQ(instance.Map().size(), 3UL);
std::vector<DeviceContext*> ctx;
paddle::framework::Variable in;
paddle::framework::Variable out;

instance.Get(std::make_pair(frw::kernel_type_1, frw::kernel_type_2))(ctx, in,
&out);
ASSERT_EQ(test_value, 1);
instance.Get(std::make_pair(frw::kernel_type_2, frw::kernel_type_3))(ctx, in,
&out);
ASSERT_EQ(test_value, 0);
instance.Get(std::make_pair(frw::kernel_type_1, frw::kernel_type_3))(ctx, in,
&out);
ASSERT_EQ(test_value, 2);
}