Merge remote-tracking branch 'baidu/develop' into get_places_op

doc/api/v2/fluid/layers.rst

@@ -312,3 +312,9 @@ sequence_softmax
 ..  autofunction:: paddle.v2.fluid.layers.sequence_softmax
     :noindex:
 
+
+reduce_sum
+---------
+..  autofunction:: paddle.v2.fluid.layers.reduce_sum
+    :noindex:
+

doc/design/refactor/multi_cpu.md

+# Design Doc: Execute the Program with Multi CPU
+
+## Abstract
+
+This Design Doc propose an approach to make the user-defined Op graph
+running with multi-CPU, we will use an auto transpiler to convert the user-defined
+Op graph to a multi-CPU Op graph, and run `ParallelDo` Op to run the graph.
+
+## Transpiler
+
+<img src="src/multi-threads/single-thread@3x.png" width="300">
+
+After converted:
+
+<img src="src/multi-threads/multi-threads@3x.png" width="1000">
+
+## Implement
+
+- `Multi-CPU Transpiler` will convert the graph to a multi-CPU graph
+  which would be executed with multi-threads.
+- `BlockingCounter` will `Init/Decrement` an atomic counter, and Blocking `Wait`
+  for the atomic counter become `0`:
+  ```cpp
+  BlockingCounter bc(thread_count);
+  for (int i = 0; i < thread_count; ++i) {
+    thread_pool->Start([&bc] {bc.DecrementCount(); })
+  }
+  bc.Wait();
+  ```
+- `ParallelDo` Operator
+  - Initialize a thread pool which is a Singleton.
+  - Use a block id as the input, and create run the specify Block on independent scope
+    with multi-threads.
+  - Initialize a `BlockingCounter` instance and wait until all threads are done.
+- `Split` Operator will split the Input Tensor into a TensorArray.
+- `Merge` merge all the gradients which calculated in different threads
+  with `mean/sum/max/min...` method, and then run the Optimizer Op to optimize `W`.
+
+## TODO
+
+- Improve the optimizer stage with multi-threads, since we could
+  assign the parameters to the different threads and execute
+  optimizer with multi-threads.

doc/design/switch_kernel.md

+## Background
+Every operator has many kernels because there are multiple data types, places, data layout that Fluid supports. We use the `KernelType` to describe kernel types that operators can hold. 
+
+The `KernelType` is as follows.
+
+```
+struct KernelType {
+  Place place_;
+  DataType data_type_;
+  LayoutType layout_;
+};
+```
+
+The `place_` is a descriptor of the device and the computational library, e.g., `MKLDNNPlace`, `CUDAPlace`.
+
+The `data_type_` is the data type that this kernel performs on, e.g., `FP32`, `INT64`. Note that one kernel may have inputs with different data types. However, it will be a major `data_type`. For example, the `cross_entropy` takes `int64` as it label, and `double`/`float` as its input logit and output cost. The major `data_type` of `cross_entropy` is `float`/`double`.
+
+The `layout` is useful for some computational library. One example is that MKLDNN uses many kinds of layout, such as `nChw8c`. Each kind of layout will invoke the different kernel.
+
+## Problem
+
+We register a kernel for every operator and every kernel type ideally. However, it is impracticable for the following situations.
+
+1. Some operators, like CRF, are complicated and inefficient to be implemented on GPU. The CRF operator will only have a CPU kernel.
+2. Some operators will take too many memory. It is better to force them into CPU. However, the rest of operators in this neural network will be performed on GPU, i.e., model parallel problem.
+3. Some layout and place are particular. One example is that MKLDNN uses `nChw8` and there is no other library uses `nChw8c`.
+
+Problems under these situations are similar. We can formalise this problem as follow.
+
+We register kernels with types $KT = \{kt_1, kt_2, kt_3, ...\}$ for one operator. The inputs of this operator should be run on kernel type $kt_{?}$, which the $kt_{?} \notin KT$. How to cast the input of this operator from $kt_{?}$ to any of kernel type in $KT$.
+
+## Solution
+
+It is clearly that transforming inputs of an operator toadapt another kernel type is not related to the particular operator. So we should register these transformation methods as global methods.
+
+We can infer a kernel type from the inputs of an operators. We let this kernel type as `actual kernel type`, which means this kernel type is the actually kernel type that operator should be performed.
+
+We can get a kernel type by 1) The configuration of operator description. (Users may want to force use `MKL` for `conv` operator). 2) The place of the current executor. (Executor is running on GPU). This kernel type is what we expect the operator will be performed on. We let this kernel type as `expect kernel type`.
+
+We transform the input data from `actual` to `expect` if the expect kernel type is not as same as actual kernel type.
+
+The algorithm is described as follow
+
+```cpp
+using DataTransformationFN = std::function<void(const Tensor& in, Tensor* out)>;
+using KernelTypePair = std::pair<KernelType, KernelType>;
+
+map<KernelTypePair, DataTransformationFN> g_data_transformation_;
+
+void OpWithKernel::Run() {
+  vec<Tensor> inputs = ...
+  auto actual_kernel_type = GetActualKernelType(inputs);
+  
+  // The expected kernel type is related to actual kernel type.
+  // For the most operators, the expected kernel type is as same as
+  // actual kernel type.
+  //
+  // So we pass `actual_kernel_type` as a parameter of 
+  // GetExpectedKernelType
+  auto expect_kernel_type = GetExpectedKernelType(actual_kernel_type);
+  
+  auto trans = g_data_transformation_[{actual_kernel_type, expect_kernel_type}];
+  
+  kernel.run(trans(inputs));
+}
+```

doc/getstarted/build_and_install/docker_install_cn.rst

@@ -128,7 +128,7 @@ PaddlePaddle Book是为用户和开发者制作的一个交互式的Jupyter Note
 
 AVX是一种CPU指令集，可以加速PaddlePaddle的计算。最新的PaddlePaddle Docker镜像默认
 是开启AVX编译的，所以，如果您的电脑不支持AVX，需要单独
-`编译 <./build_from_source_cn.rst>`_ PaddlePaddle为no-avx版本。
+`编译 <./build_from_source_cn.html>`_ PaddlePaddle为no-avx版本。
 
 以下指令能检查Linux电脑是否支持AVX：
 

doc/getstarted/build_and_install/docker_install_en.rst

@@ -137,7 +137,7 @@ GPU driver installed before move on.
 AVX is a kind of CPU instruction can accelerate PaddlePaddle's calculations.
 The latest PaddlePaddle Docker image turns AVX on by default, so, if your
 computer doesn't support AVX, you'll probably need to
-`build <./build_from_source_en.rst>`_ with :code:`WITH_AVX=OFF`.
+`build <./build_from_source_en.html>`_ with :code:`WITH_AVX=OFF`.
 
 The following command will tell you whether your computer supports AVX.
 

doc/howto/dev/new_op_cn.md

@@ -53,7 +53,7 @@ Kernel实现       | CPU、CUDA共享Kernel实现在`.h`文件中，否则，CPU
 ```cpp
 class MulOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  MulOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+  MulOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "(Tensor), 2D tensor of size (M x K)");
     AddInput("Y", "(Tensor), 2D tensor of size (K x N)");
@@ -82,7 +82,7 @@ The equation is: Out = X * Y
 template <typename AttrType>
 class ScaleOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ScaleOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+  ScaleOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "The input tensor of scale operator.").NotInGradient();
     AddOutput("Out", "The output tensor of scale operator.").NotInGradient();

doc/howto/dev/new_op_en.md

@@ -50,7 +50,7 @@ First, define `ProtoMaker` to describe the Operator's input, output, and additio
 ```cpp
 class MulOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  MulOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+  MulOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "(Tensor), 2D tensor of size (M x K)");
     AddInput("Y", "(Tensor), 2D tensor of size (K x N)");
@@ -79,7 +79,7 @@ An additional example [`ScaleOp`](https://github.com/PaddlePaddle/Paddle/blob/de
 template <typename AttrType>
 class ScaleOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ScaleOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+  ScaleOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "The input tensor of scale operator.").NotInGradient();
     AddOutput("Out", "The output tensor of scale operator.").NotInGradient();

doc/howto/usage/cluster/k8s_distributed_cn.md

@@ -2,8 +2,6 @@
 
 前一篇文章介绍了如何在Kubernetes集群上启动一个单机PaddlePaddle训练作业 (Job)。在这篇文章里，我们介绍如何在Kubernetes集群上进行分布式PaddlePaddle训练作业。关于PaddlePaddle的分布式训练，文章 [Cluster Training](http://www.paddlepaddle.org/docs/develop/documentation/zh/howto/usage/cluster/cluster_train_cn.html)介绍了一种通过SSH远程分发任务，进行分布式训练的方法，与此不同的是，本文将介绍在Kubernetes容器管理平台上快速构建PaddlePaddle容器集群，进行分布式训练的方案。
 
-有关Kubernetes相关概念以及如何搭建和配置Kubernetes集群，可以参考[k8s_basis](./k8s_basis_cn.md)。
-
 ## 整体方案
 
 在训练之前，用户将配置与训练数据切分好放在分布式文件系统预先分配好的目录中(不同的分布式文件系统，需要使用其制定的方式挂载后并导入数据)，训练时，程序从此目录拷贝文件到容器内进行训练，将结果保存到此目录里。整体的结构图如下：

paddle/framework/attribute.cc

@@ -19,42 +19,42 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
-Attribute GetAttrValue(const OpDesc::Attr& attr_desc) {
+Attribute GetAttrValue(const proto::OpDesc::Attr& attr_desc) {
   switch (attr_desc.type()) {
-    case framework::AttrType::BOOLEAN: {
+    case proto::AttrType::BOOLEAN: {
       return attr_desc.b();
     }
-    case framework::AttrType::INT: {
+    case proto::AttrType::INT: {
       return attr_desc.i();
     }
-    case framework::AttrType::FLOAT: {
+    case proto::AttrType::FLOAT: {
       return attr_desc.f();
     }
-    case framework::AttrType::STRING: {
+    case proto::AttrType::STRING: {
       return attr_desc.s();
     }
-    case framework::AttrType::BOOLEANS: {
+    case proto::AttrType::BOOLEANS: {
       std::vector<bool> val(attr_desc.bools_size());
       for (int i = 0; i < attr_desc.bools_size(); ++i) {
         val[i] = attr_desc.bools(i);
       }
       return val;
     }
-    case framework::AttrType::INTS: {
+    case proto::AttrType::INTS: {
       std::vector<int> val(attr_desc.ints_size());
       for (int i = 0; i < attr_desc.ints_size(); ++i) {
         val[i] = attr_desc.ints(i);
       }
       return val;
     }
-    case framework::AttrType::FLOATS: {
+    case proto::AttrType::FLOATS: {
       std::vector<float> val(attr_desc.floats_size());
       for (int i = 0; i < attr_desc.floats_size(); ++i) {
         val[i] = attr_desc.floats(i);
       }
       return val;
     }
-    case framework::AttrType::STRINGS: {
+    case proto::AttrType::STRINGS: {
       std::vector<std::string> val(attr_desc.strings_size());
       for (int i = 0; i < attr_desc.strings_size(); ++i) {
         val[i] = attr_desc.strings(i);

paddle/framework/attribute.h

@@ -27,12 +27,12 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 template <typename T>
-inline AttrType AttrTypeID() {
+inline proto::AttrType AttrTypeID() {
   Attribute tmp = T();
-  return static_cast<AttrType>(tmp.which() - 1);
+  return static_cast<proto::AttrType>(tmp.which() - 1);
 }
 
-Attribute GetAttrValue(const OpDesc::Attr& attr_desc);
+Attribute GetAttrValue(const proto::OpDesc::Attr& attr_desc);
 
 class AttrReader {
  public:

paddle/framework/backward.cc

@@ -341,7 +341,7 @@ static void CreateGradVarInBlock(
       auto* param = block_desc->FindVarRecursive(pname);
       auto* grad = block_desc->FindVar(arg);
       if (param == nullptr) {
-        grad->SetDataType(DataType::FP32);
+        grad->SetDataType(proto::DataType::FP32);
       } else {
         grad->SetDataType(param->GetDataType());
       }

paddle/framework/backward_test.cc

@@ -166,7 +166,7 @@ class FillZeroOpMaker : public OpProtoAndCheckerMaker {
 
 class SumOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  SumOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+  SumOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "the input tensors of sum operator.").AsDuplicable();
     AddOutput("Out", "the output tensor of sum operator.");

paddle/framework/block_desc.cc

@@ -128,22 +128,22 @@ BlockDescBind *BlockDescBind::ParentBlock() const {
   return prog_->MutableBlock(static_cast<size_t>(this->desc_->parent_idx()));
 }
 
-BlockDesc *BlockDescBind::Proto() {
+proto::BlockDesc *BlockDescBind::Proto() {
   Flush();
   return desc_;
 }
 
-BlockDescBind::BlockDescBind(ProgramDescBind *prog, BlockDesc *desc)
+BlockDescBind::BlockDescBind(ProgramDescBind *prog, proto::BlockDesc *desc)
     : prog_(prog), desc_(desc), need_update_(false) {
-  for (const VarDesc &var_desc : desc_->vars()) {
+  for (const proto::VarDesc &var_desc : desc_->vars()) {
     vars_[var_desc.name()].reset(new VarDescBind(var_desc));
   }
-  for (const OpDesc &op_desc : desc_->ops()) {
+  for (const proto::OpDesc &op_desc : desc_->ops()) {
     ops_.emplace_back(new OpDescBind(op_desc, prog));
   }
 }
 
-BlockDescBind::BlockDescBind(const BlockDescBind &other, BlockDesc *desc,
+BlockDescBind::BlockDescBind(const BlockDescBind &other, proto::BlockDesc *desc,
                              ProgramDescBind *prog)
     : prog_(prog), desc_(desc) {
   need_update_ = true;

paddle/framework/block_desc.h

@@ -36,9 +36,9 @@ class ProgramDescBind;
 
 class BlockDescBind {
  public:
-  BlockDescBind(ProgramDescBind *prog, BlockDesc *desc);
+  BlockDescBind(ProgramDescBind *prog, proto::BlockDesc *desc);
 
-  BlockDescBind(const BlockDescBind &other, BlockDesc *desc,
+  BlockDescBind(const BlockDescBind &other, proto::BlockDesc *desc,
                 ProgramDescBind *prog);
 
   ~BlockDescBind() {
@@ -88,7 +88,7 @@ class BlockDescBind {
 
   void Flush();
 
-  BlockDesc *Proto();
+  proto::BlockDesc *Proto();
 
   ProgramDescBind *Program() { return this->prog_; }
 
@@ -98,7 +98,7 @@ class BlockDescBind {
 
  private:
   ProgramDescBind *prog_;   // not_own
-  BlockDesc *desc_;        // not_own
+  proto::BlockDesc *desc_;  // not_own
   bool need_update_;
 
   std::deque<std::unique_ptr<OpDescBind>> ops_;

paddle/framework/data_type.h

@@ -20,7 +20,8 @@
 namespace paddle {
 namespace framework {
 
-inline DataType ToDataType(std::type_index type) {
+inline proto::DataType ToDataType(std::type_index type) {
+  using namespace paddle::framework::proto;
   if (typeid(float).hash_code() == type.hash_code()) {
     return DataType::FP32;
   } else if (typeid(double).hash_code() == type.hash_code()) {
@@ -36,7 +37,8 @@ inline DataType ToDataType(std::type_index type) {
   }
 }
 
-inline std::type_index ToTypeIndex(DataType type) {
+inline std::type_index ToTypeIndex(proto::DataType type) {
+  using namespace paddle::framework::proto;
   switch (type) {
     case DataType::FP32:
       return typeid(float);
@@ -54,7 +56,8 @@ inline std::type_index ToTypeIndex(DataType type) {
 }
 
 template <typename Visitor>
-inline void VisitDataType(DataType type, Visitor visitor) {
+inline void VisitDataType(proto::DataType type, Visitor visitor) {
+  using namespace paddle::framework::proto;
   switch (type) {
     case DataType::FP32:
       visitor.template operator()<float>();

paddle/framework/details/op_registry.h

@@ -90,7 +90,7 @@ struct OpInfoFiller<T, kOperator> {
 template <typename T>
 struct OpInfoFiller<T, kOpProtoAndCheckerMaker> {
   void operator()(const char* op_type, OpInfo* info) const {
-    info->proto_ = new OpProto;
+    info->proto_ = new proto::OpProto;
     info->checker_ = new OpAttrChecker();
     auto maker = T(info->proto_, info->checker_);
     maker.Validate();

paddle/framework/executor.cc

@@ -41,20 +41,20 @@ Executor::Executor(const std::vector<platform::Place>& places) {
   device_contexts_.swap(borrowed_contexts);
 }
 
-static void CreateTensor(Variable* var, VarDesc::VarType var_type) {
-  if (var_type == VarDesc::LOD_TENSOR) {
+static void CreateTensor(Variable* var, proto::VarDesc::VarType var_type) {
+  if (var_type == proto::VarDesc::LOD_TENSOR) {
     var->GetMutable<LoDTensor>();
-  } else if (var_type == VarDesc::SELECTED_ROWS) {
+  } else if (var_type == proto::VarDesc::SELECTED_ROWS) {
     var->GetMutable<SelectedRows>();
-  } else if (var_type == VarDesc::FEED_MINIBATCH) {
+  } else if (var_type == proto::VarDesc::FEED_MINIBATCH) {
     var->GetMutable<FeedFetchList>();
-  } else if (var_type == VarDesc::FETCH_LIST) {
+  } else if (var_type == proto::VarDesc::FETCH_LIST) {
     var->GetMutable<FeedFetchList>();
-  } else if (var_type == VarDesc::STEP_SCOPES) {
+  } else if (var_type == proto::VarDesc::STEP_SCOPES) {
     var->GetMutable<std::vector<framework::Scope>>();
-  } else if (var_type == VarDesc::LOD_RANK_TABLE) {
+  } else if (var_type == proto::VarDesc::LOD_RANK_TABLE) {
     var->GetMutable<LoDRankTable>();
-  } else if (var_type == VarDesc::LOD_TENSOR_ARRAY) {
+  } else if (var_type == proto::VarDesc::LOD_TENSOR_ARRAY) {
     var->GetMutable<LoDTensorArray>();
   } else {
     PADDLE_THROW(

paddle/framework/framework.proto

@@ -14,7 +14,7 @@ limitations under the License. */
 
 syntax = "proto2";
 option optimize_for = LITE_RUNTIME;
-package paddle.framework;
+package paddle.framework.proto;
 
 enum AttrType {
   INT = 0;

paddle/framework/lod_rank_table.cc

@@ -46,4 +46,13 @@ void LoDRankTable::Reset(const LoD& lod, size_t level) {
 }
 
 }  // namespace framework
+
+std::ostream& operator<<(std::ostream& out,
+                         const framework::LoDRankTable& table) {
+  out << "NumOfSequence " << table.items().size() << "\n";
+  for (auto& each_item : table.items()) {
+    out << "\tSeq #" << each_item.index << ", Len=" << each_item.length << "\n";
+  }
+  return out;
+}
 }  // namespace paddle

paddle/framework/lod_rank_table.h

@@ -13,6 +13,7 @@
    limitations under the License. */
 
 #pragma once
+#include <iosfwd>
 #include "paddle/framework/lod_tensor.h"
 
 namespace paddle {
@@ -52,4 +53,8 @@ class LoDRankTable {
 };
 
 }  // namespace framework
+
+std::ostream& operator<<(std::ostream& out,
+                         const framework::LoDRankTable& table);
+
 }  // namespace paddle

paddle/framework/lod_tensor.cc

@@ -197,7 +197,7 @@ void SerializeToStream(std::ostream &os, const LoDTensor &tensor,
   {  // the 2nd field, tensor description
      // int32_t  size
      // void*    protobuf message
-    framework::TensorDesc desc;
+    proto::TensorDesc desc;
     desc.set_data_type(framework::ToDataType(tensor.type()));
     auto dims = framework::vectorize(tensor.dims());
     auto *pb_dims = desc.mutable_dims();
@@ -262,7 +262,7 @@ void DeserializeFromStream(std::istream &is, LoDTensor *tensor) {
   uint32_t version;
   is.read(reinterpret_cast<char *>(&version), sizeof(version));
   PADDLE_ENFORCE_EQ(version, 0U, "Only version 0 is supported");
-  framework::TensorDesc desc;
+  proto::TensorDesc desc;
   {  // int32_t size
      // proto buffer
     int32_t size;
@@ -281,16 +281,16 @@ void DeserializeFromStream(std::istream &is, LoDTensor *tensor) {
     void *buf;
     platform::Place cpu = platform::CPUPlace();
     switch (desc.data_type()) {
-      case framework::FP32:
+      case proto::FP32:
         buf = tensor->mutable_data<float>(cpu);
         break;
-      case framework::FP64:
+      case proto::FP64:
         buf = tensor->mutable_data<double>(cpu);
         break;
-      case framework::INT32:
+      case proto::INT32:
         buf = tensor->mutable_data<int>(cpu);
         break;
-      case framework::INT64:
+      case proto::INT64:
         buf = tensor->mutable_data<int64_t>(cpu);
         break;
       default:

paddle/framework/lod_tensor.h

@@ -184,6 +184,18 @@ LoDTensor LodExpand(const LoDTensor& source, const LoD& lod, size_t level,
   return tensor;
 }
 
+// Get the absolute offset of a lod[start_level][start_idx:end_idx] and
+// relative length of details for every levels(i.e., [start_level: ]).
+//
+// For example,
+//   lod = [[0, 3, 4, 8], [0, 9, 10, 11, 13, 17, 19, 22, 24]]
+//   start_level = 0
+//   start_idx = 1
+//   end_idx = 3
+//
+// Returns:
+//  LoD = [[1, 4], [2, 4, 2, 3, 2]]
+//  pair<size_t, size_t> = {11, 24}
 std::pair<LoD, std::pair<size_t, size_t>> GetSubLoDAndAbsoluteOffset(
     const LoD& lod, size_t start_idx, size_t end_idx, size_t start_level);
 

paddle/framework/op_desc.cc

@@ -58,11 +58,11 @@ class CompileTimeInferShapeContext : public InferShapeContext {
     PADDLE_ENFORCE_LT(j, Outputs(out).size());
     auto *in_var = block_.FindVarRecursive(Inputs(in)[i]);
     auto *out_var = block_.FindVarRecursive(Outputs(out)[j]);
-    if (in_var->GetType() != VarDesc::LOD_TENSOR) {
+    if (in_var->GetType() != proto::VarDesc::LOD_TENSOR) {
       VLOG(3) << "input " << in << " is not LodTensor";
       return;
     }
-    PADDLE_ENFORCE_EQ(in_var->GetType(), VarDesc::LOD_TENSOR,
+    PADDLE_ENFORCE_EQ(in_var->GetType(), proto::VarDesc::LOD_TENSOR,
                       "The %d-th output of Output(%s) must be LoDTensor.", j,
                       out);
     out_var->SetLoDLevel(in_var->GetLodLevel());
@@ -70,7 +70,7 @@ class CompileTimeInferShapeContext : public InferShapeContext {
   bool IsRuntime() const override;
 
  protected:
-  VarDesc::VarType GetVarType(const std::string &name) const override;
+  proto::VarDesc::VarType GetVarType(const std::string &name) const override;
 
   DDim GetDim(const std::string &name) const override;
 
@@ -90,12 +90,12 @@ OpDescBind::OpDescBind(const std::string &type, const VariableNameMap &inputs,
   need_update_ = true;
 }
 
-OpDescBind::OpDescBind(const OpDesc &desc, ProgramDescBind *prog)
+OpDescBind::OpDescBind(const proto::OpDesc &desc, ProgramDescBind *prog)
     : desc_(desc), need_update_(false) {
   // restore inputs_
   int input_size = desc_.inputs_size();
   for (int i = 0; i < input_size; ++i) {
-    const OpDesc::Var &var = desc_.inputs(i);
+    const proto::OpDesc::Var &var = desc_.inputs(i);
     std::vector<std::string> &args = inputs_[var.parameter()];
     int argu_size = var.arguments_size();
     args.reserve(argu_size);
@@ -106,7 +106,7 @@ OpDescBind::OpDescBind(const OpDesc &desc, ProgramDescBind *prog)
   // restore outputs_
   int output_size = desc_.outputs_size();
   for (int i = 0; i < output_size; ++i) {
-    const OpDesc::Var &var = desc_.outputs(i);
+    const proto::OpDesc::Var &var = desc_.outputs(i);
     std::vector<std::string> &args = outputs_[var.parameter()];
     int argu_size = var.arguments_size();
     args.reserve(argu_size);
@@ -115,9 +115,9 @@ OpDescBind::OpDescBind(const OpDesc &desc, ProgramDescBind *prog)
     }
   }
   // restore attrs_
-  for (const OpDesc::Attr &attr : desc_.attrs()) {
+  for (const proto::OpDesc::Attr &attr : desc_.attrs()) {
     std::string attr_name = attr.name();
-    if (attr.type() != AttrType::BLOCK) {
+    if (attr.type() != proto::AttrType::BLOCK) {
       attrs_[attr_name] = GetAttrValue(attr);
     } else {
       auto bid = attr.block_idx();
@@ -126,7 +126,7 @@ OpDescBind::OpDescBind(const OpDesc &desc, ProgramDescBind *prog)
   }
 }
 
-OpDesc *OpDescBind::Proto() {
+proto::OpDesc *OpDescBind::Proto() {
   Flush();
   return &desc_;
 }
@@ -175,10 +175,10 @@ void OpDescBind::SetOutput(const std::string &param_name,
   this->outputs_[param_name] = args;
 }
 
-AttrType OpDescBind::GetAttrType(const std::string &name) const {
+proto::AttrType OpDescBind::GetAttrType(const std::string &name) const {
   auto it = attrs_.find(name);
   PADDLE_ENFORCE(it != attrs_.end(), "Attribute %s is not found", name);
-  return static_cast<AttrType>(it->second.which() - 1);
+  return static_cast<proto::AttrType>(it->second.which() - 1);
 }
 
 std::vector<std::string> OpDescBind::AttrNames() const {
@@ -253,8 +253,8 @@ void OpDescBind::RenameInput(const std::string &old_name,
 }
 
 struct SetAttrDescVisitor : public boost::static_visitor<void> {
-  explicit SetAttrDescVisitor(OpDesc::Attr *attr) : attr_(attr) {}
-  mutable OpDesc::Attr *attr_;
+  explicit SetAttrDescVisitor(proto::OpDesc::Attr *attr) : attr_(attr) {}
+  mutable proto::OpDesc::Attr *attr_;
   void operator()(int v) const { attr_->set_i(v); }
   void operator()(float v) const { attr_->set_f(v); }
   void operator()(const std::string &v) const { attr_->set_s(v); }
@@ -272,7 +272,9 @@ struct SetAttrDescVisitor : public boost::static_visitor<void> {
   void operator()(const std::vector<bool> &v) const {
     VectorToRepeated(v, attr_->mutable_bools());
   }
-  void operator()(BlockDesc *desc) const { attr_->set_block_idx(desc->idx()); }
+  void operator()(proto::BlockDesc *desc) const {
+    attr_->set_block_idx(desc->idx());
+  }
   void operator()(boost::blank) const { PADDLE_THROW("Unexpected branch"); }
 };
 
@@ -297,7 +299,7 @@ void OpDescBind::Flush() {
       auto *attr_desc = desc_.add_attrs();
       attr_desc->set_name(attr.first);
       attr_desc->set_type(
-          static_cast<framework::AttrType>(attr.second.which() - 1));
+          static_cast<proto::AttrType>(attr.second.which() - 1));
       SetAttrDescVisitor visitor(attr_desc);
       boost::apply_visitor(visitor, attr.second);
     }
@@ -375,7 +377,7 @@ void OpDescBind::InferVarType(BlockDescBind *block) const {
     for (auto &out_pair : this->outputs_) {
       for (auto &out_var_name : out_pair.second) {
         block->FindRecursiveOrCreateVar(out_var_name)
-            ->SetType(VarDesc::LOD_TENSOR);
+            ->SetType(proto::VarDesc::LOD_TENSOR);
       }
     }
   }
@@ -484,7 +486,7 @@ void CompileTimeInferShapeContext::SetDim(const std::string &name,
 }
 bool CompileTimeInferShapeContext::IsRuntime() const { return false; }
 
-VarDesc::VarType CompileTimeInferShapeContext::GetVarType(
+proto::VarDesc::VarType CompileTimeInferShapeContext::GetVarType(
     const std::string &name) const {
   return block_.FindVarRecursive(name)->GetType();
 }

paddle/framework/op_desc.h

@@ -33,9 +33,9 @@ class OpDescBind {
   OpDescBind(const std::string &type, const VariableNameMap &inputs,
              const VariableNameMap &outputs, const AttributeMap &attrs);
 
-  OpDescBind(const OpDesc &desc, ProgramDescBind *prog);
+  OpDescBind(const proto::OpDesc &desc, ProgramDescBind *prog);
 
-  OpDesc *Proto();
+  proto::OpDesc *Proto();
 
   std::string Type() const { return desc_.type(); }
 
@@ -59,7 +59,7 @@ class OpDescBind {
     return attrs_.find(name) != attrs_.end();
   }
 
-  AttrType GetAttrType(const std::string &name) const;
+  proto::AttrType GetAttrType(const std::string &name) const;
 
   std::vector<std::string> AttrNames() const;
 
@@ -126,7 +126,7 @@ class OpDescBind {
     return ret_val;
   }
 
-  OpDesc desc_;
+  proto::OpDesc desc_;
   VariableNameMap inputs_;
   VariableNameMap outputs_;
   AttributeMap attrs_;

paddle/framework/op_info.h

@@ -34,7 +34,7 @@ class InferShapeBase {
 struct OpInfo {
   OpCreator creator_;
   GradOpMakerFN grad_op_maker_;
-  OpProto* proto_{nullptr};
+  proto::OpProto* proto_{nullptr};
   OpAttrChecker* checker_{nullptr};
   InferVarTypeFN infer_var_type_;
   InferShapeFN infer_shape_;
@@ -43,7 +43,7 @@ struct OpInfo {
     return proto_ != nullptr && checker_ != nullptr;
   }
 
-  const OpProto& Proto() const {
+  const proto::OpProto& Proto() const {
     PADDLE_ENFORCE_NOT_NULL(proto_, "Operator Proto has not been registered");
     PADDLE_ENFORCE(proto_->IsInitialized(),
                    "Operator Proto must be initialized in op info");

paddle/framework/op_proto_maker.h

@@ -22,6 +22,8 @@ namespace framework {
 // this class not only make proto but also init attribute checkers.
 class OpProtoAndCheckerMaker {
  public:
+  using OpProto = proto::OpProto;
+  using OpAttrChecker = framework::OpAttrChecker;
   OpProtoAndCheckerMaker(OpProto* proto, OpAttrChecker* op_checker)
       : proto_(proto), op_checker_(op_checker) {}
 
@@ -80,7 +82,7 @@ class OpProtoAndCheckerMaker {
 
 class NOPMaker : public OpProtoAndCheckerMaker {
  public:
-  NOPMaker(framework::OpProto* proto, framework::OpAttrChecker* op_checker)
+  NOPMaker(OpProto* proto, framework::OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {}
 };
 

paddle/framework/op_proto_maker_test.cc

@@ -18,7 +18,7 @@ limitations under the License. */
 
 class TestAttrProtoMaker : public paddle::framework::OpProtoAndCheckerMaker {
  public:
-  TestAttrProtoMaker(paddle::framework::OpProto* proto,
+  TestAttrProtoMaker(paddle::framework::proto::OpProto* proto,
                      paddle::framework::OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddAttr<float>("scale", "scale of test op");
@@ -27,7 +27,7 @@ class TestAttrProtoMaker : public paddle::framework::OpProtoAndCheckerMaker {
 };
 
 TEST(ProtoMaker, DuplicatedAttr) {
-  paddle::framework::OpProto op_proto;
+  paddle::framework::proto::OpProto op_proto;
   paddle::framework::OpAttrChecker op_checker;
   auto proto_maker = TestAttrProtoMaker(&op_proto, &op_checker);
   ASSERT_THROW(proto_maker.Validate(), paddle::platform::EnforceNotMet);
@@ -35,7 +35,7 @@ TEST(ProtoMaker, DuplicatedAttr) {
 
 class TestInOutProtoMaker : public paddle::framework::OpProtoAndCheckerMaker {
  public:
-  TestInOutProtoMaker(paddle::framework::OpProto* proto,
+  TestInOutProtoMaker(paddle::framework::proto::OpProto* proto,
                       paddle::framework::OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("input", "input of test op");
@@ -44,7 +44,7 @@ class TestInOutProtoMaker : public paddle::framework::OpProtoAndCheckerMaker {
 };
 
 TEST(ProtoMaker, DuplicatedInOut) {
-  paddle::framework::OpProto op_proto;
+  paddle::framework::proto::OpProto op_proto;
   paddle::framework::OpAttrChecker op_checker;
   auto proto_maker = TestInOutProtoMaker(&op_proto, &op_checker);
   ASSERT_THROW(proto_maker.Validate(), paddle::platform::EnforceNotMet);

paddle/framework/op_registry.cc

@@ -31,7 +31,8 @@ std::unique_ptr<OperatorBase> OpRegistry::CreateOp(
 }
 
 static VariableNameMap ConvertOpDescVarsToVarNameMap(
-    const google::protobuf::RepeatedPtrField<OpDesc::Var>& op_desc_vars) {
+    const google::protobuf::RepeatedPtrField<proto::OpDesc::Var>&
+        op_desc_vars) {
   VariableNameMap ret_val;
   for (auto& var : op_desc_vars) {
     auto& var_names = ret_val[var.parameter()];
@@ -43,7 +44,8 @@ static VariableNameMap ConvertOpDescVarsToVarNameMap(
   return ret_val;
 }
 
-std::unique_ptr<OperatorBase> OpRegistry::CreateOp(const OpDesc& op_desc) {
+std::unique_ptr<OperatorBase> OpRegistry::CreateOp(
+    const proto::OpDesc& op_desc) {
   VLOG(1) << "CreateOp directly from OpDesc is deprecated. It should only be"
              "used in unit tests. Use CreateOp(const OpDescBind& op_desc) "
              "instead.";

paddle/framework/op_registry.h

@@ -77,7 +77,7 @@ class OpRegistry {
                                                 const VariableNameMap& outputs,
                                                 AttributeMap attrs);
 
-  static std::unique_ptr<OperatorBase> CreateOp(const OpDesc& op_desc);
+  static std::unique_ptr<OperatorBase> CreateOp(const proto::OpDesc& op_desc);
 
   static std::unique_ptr<OperatorBase> CreateOp(const OpDescBind& op_desc);
 };

paddle/framework/op_registry_test.cc

@@ -51,7 +51,7 @@ class MyTestOpProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
 
 static void BuildVar(const std::string& param_name,
                      std::initializer_list<const char*> arguments,
-                     paddle::framework::OpDesc::Var* var) {
+                     paddle::framework::proto::OpDesc::Var* var) {
   var->set_parameter(param_name);
   for (auto& arg_name : arguments) {
     var->add_arguments(arg_name);
@@ -63,7 +63,7 @@ REGISTER_OP_WITHOUT_GRADIENT(my_test_op, paddle::framework::MyTestOp,
                              paddle::framework::MyTestOpProtoAndCheckerMaker);
 
 TEST(OpRegistry, CreateOp) {
-  paddle::framework::OpDesc op_desc;
+  paddle::framework::proto::OpDesc op_desc;
   op_desc.set_type("cos_sim");
   BuildVar("input", {"aa"}, op_desc.add_inputs());
   BuildVar("output", {"bb"}, op_desc.add_outputs());
@@ -71,7 +71,7 @@ TEST(OpRegistry, CreateOp) {
   float scale = 3.3;
   auto attr = op_desc.mutable_attrs()->Add();
   attr->set_name("scale");
-  attr->set_type(paddle::framework::AttrType::FLOAT);
+  attr->set_type(paddle::framework::proto::AttrType::FLOAT);
   attr->set_f(scale);
 
   auto op = paddle::framework::OpRegistry::CreateOp(op_desc);
@@ -83,14 +83,14 @@ TEST(OpRegistry, CreateOp) {
 }
 
 TEST(OpRegistry, IllegalAttr) {
-  paddle::framework::OpDesc op_desc;
+  paddle::framework::proto::OpDesc op_desc;
   op_desc.set_type("cos_sim");
   BuildVar("input", {"aa"}, op_desc.add_inputs());
   BuildVar("output", {"bb"}, op_desc.add_outputs());
 
   auto attr = op_desc.mutable_attrs()->Add();
   attr->set_name("scale");
-  attr->set_type(paddle::framework::AttrType::FLOAT);
+  attr->set_type(paddle::framework::proto::AttrType::FLOAT);
   attr->set_f(-2.0);
 
   bool caught = false;
@@ -108,7 +108,7 @@ TEST(OpRegistry, IllegalAttr) {
 }
 
 TEST(OpRegistry, DefaultValue) {
-  paddle::framework::OpDesc op_desc;
+  paddle::framework::proto::OpDesc op_desc;
   op_desc.set_type("cos_sim");
   BuildVar("input", {"aa"}, op_desc.add_inputs());
   BuildVar("output", {"bb"}, op_desc.add_outputs());
@@ -123,7 +123,7 @@ TEST(OpRegistry, DefaultValue) {
 }
 
 TEST(OpRegistry, CustomChecker) {
-  paddle::framework::OpDesc op_desc;
+  paddle::framework::proto::OpDesc op_desc;
   op_desc.set_type("my_test_op");
   BuildVar("input", {"ii"}, op_desc.add_inputs());
   BuildVar("output", {"oo"}, op_desc.add_outputs());
@@ -145,7 +145,7 @@ TEST(OpRegistry, CustomChecker) {
   // set 'test_attr' set to an illegal value
   auto attr = op_desc.mutable_attrs()->Add();
   attr->set_name("test_attr");
-  attr->set_type(paddle::framework::AttrType::INT);
+  attr->set_type(paddle::framework::proto::AttrType::INT);
   attr->set_i(3);
   caught = false;
   try {
@@ -164,7 +164,7 @@ TEST(OpRegistry, CustomChecker) {
   op_desc.mutable_attrs()->Clear();
   attr = op_desc.mutable_attrs()->Add();
   attr->set_name("test_attr");
-  attr->set_type(paddle::framework::AttrType::INT);
+  attr->set_type(paddle::framework::proto::AttrType::INT);
   attr->set_i(4);
   auto op = paddle::framework::OpRegistry::CreateOp(op_desc);
   paddle::platform::CPUDeviceContext dev_ctx;

paddle/framework/operator.cc

@@ -377,7 +377,7 @@ class RuntimeInferShapeContext : public InferShapeContext {
     }
   }
 
-  VarDesc::VarType GetVarType(const std::string& name) const override {
+  proto::VarDesc::VarType GetVarType(const std::string& name) const override {
     auto* var = scope_.FindVar(name);
     return ToVarType(var->Type());
   }
@@ -417,7 +417,7 @@ OpKernelType OperatorWithKernel::GetKernelType(
     const ExecutionContext& ctx) const {
   return OpKernelType(IndicateDataType(ctx), ctx.GetPlace());
 }
-DataType OperatorWithKernel::IndicateDataType(
+proto::DataType OperatorWithKernel::IndicateDataType(
     const ExecutionContext& ctx) const {
   auto& scope = ctx.scope();
   int data_type = -1;
@@ -443,7 +443,7 @@ DataType OperatorWithKernel::IndicateDataType(
     }
   }
   PADDLE_ENFORCE(data_type != -1, "DataType should be indicated by input");
-  return static_cast<DataType>(data_type);
+  return static_cast<proto::DataType>(data_type);
 }
 
 }  // namespace framework

paddle/framework/operator.h

@@ -358,12 +358,13 @@ struct OpKernelType {
   };
 
   platform::Place place_;
-  DataType data_type_;
+  proto::DataType data_type_;
 
-  OpKernelType(DataType data_type, platform::Place place)
+  OpKernelType(proto::DataType data_type, platform::Place place)
       : place_(place), data_type_(data_type) {}
 
-  OpKernelType(DataType data_type, const platform::DeviceContext& dev_ctx)
+  OpKernelType(proto::DataType data_type,
+               const platform::DeviceContext& dev_ctx)
       : place_(dev_ctx.GetPlace()), data_type_(data_type) {}
 
   bool operator==(const OpKernelType& o) const {
@@ -409,7 +410,7 @@ class OperatorWithKernel : public OperatorBase {
  private:
   // indicate kernel DataType by input data. Defaultly all input data must be
   // same.
-  DataType IndicateDataType(const ExecutionContext& ctx) const;
+  proto::DataType IndicateDataType(const ExecutionContext& ctx) const;
 };
 
 std::ostream& operator<<(std::ostream& os, const OpKernelType& kernel_key);

paddle/framework/operator_test.cc

@@ -58,7 +58,7 @@ class OpeWithoutKernelTestProtoAndCheckerMaker : public OpProtoAndCheckerMaker {
 
 static void BuildVar(const std::string& param_name,
                      std::initializer_list<const char*> arguments,
-                     paddle::framework::OpDesc::Var* var) {
+                     paddle::framework::proto::OpDesc::Var* var) {
   var->set_parameter(param_name);
   for (auto& arg_name : arguments) {
     *var->mutable_arguments()->Add() = arg_name;
@@ -70,14 +70,14 @@ REGISTER_OP_WITHOUT_GRADIENT(
     paddle::framework::OpeWithoutKernelTestProtoAndCheckerMaker);
 
 TEST(OperatorBase, all) {
-  paddle::framework::OpDesc op_desc;
+  paddle::framework::proto::OpDesc op_desc;
   op_desc.set_type("test_operator");
   BuildVar("input", {"IN1"}, op_desc.add_inputs());
   BuildVar("output", {"OUT1"}, op_desc.add_outputs());
 
   auto attr = op_desc.mutable_attrs()->Add();
   attr->set_name("scale");
-  attr->set_type(paddle::framework::AttrType::FLOAT);
+  attr->set_type(paddle::framework::proto::AttrType::FLOAT);
   attr->set_f(3.14);
 
   paddle::platform::CPUDeviceContext device_context;
@@ -115,7 +115,7 @@ class OpWithKernelTest : public OperatorWithKernel {
  protected:
   void InferShape(framework::InferShapeContext* ctx) const override {}
   OpKernelType GetKernelType(const ExecutionContext& ctx) const override {
-    return OpKernelType(DataType::FP32, ctx.GetPlace());
+    return OpKernelType(proto::DataType::FP32, ctx.GetPlace());
   }
 };
 
@@ -195,14 +195,14 @@ REGISTER_OP_CPU_KERNEL(op_with_kernel,
 
 // test with single input
 TEST(OpKernel, all) {
-  paddle::framework::OpDesc op_desc;
+  paddle::framework::proto::OpDesc op_desc;
   op_desc.set_type("op_with_kernel");
   BuildVar("x", {"IN1"}, op_desc.add_inputs());
   BuildVar("y", {"OUT1"}, op_desc.add_outputs());
 
   auto attr = op_desc.mutable_attrs()->Add();
   attr->set_name("scale");
-  attr->set_type(paddle::framework::AttrType::FLOAT);
+  attr->set_type(paddle::framework::proto::AttrType::FLOAT);
   attr->set_f(3.14);
 
   paddle::platform::CPUDeviceContext cpu_device_context;
@@ -224,7 +224,7 @@ REGISTER_OP_CPU_KERNEL(op_multi_inputs_with_kernel,
 TEST(OpKernel, multi_inputs) {
   using namespace paddle::framework;
 
-  OpDesc op_desc;
+  proto::OpDesc op_desc;
   op_desc.set_type("op_multi_inputs_with_kernel");
   BuildVar("xs", {"x0", "x1", "x2"}, op_desc.add_inputs());
   BuildVar("k", {"k0"}, op_desc.add_inputs());
@@ -232,7 +232,7 @@ TEST(OpKernel, multi_inputs) {
 
   auto attr = op_desc.mutable_attrs()->Add();
   attr->set_name("scale");
-  attr->set_type(paddle::framework::AttrType::FLOAT);
+  attr->set_type(paddle::framework::proto::AttrType::FLOAT);
   attr->set_f(3.14);
 
   paddle::platform::CPUDeviceContext cpu_device_context;

paddle/framework/program_desc.cc

@@ -26,7 +26,7 @@ BlockDescBind *ProgramDescBind::AppendBlock(const BlockDescBind &parent) {
   return blocks_.back().get();
 }
 
-ProgramDesc *ProgramDescBind::Proto() {
+proto::ProgramDesc *ProgramDescBind::Proto() {
   for (auto &block : blocks_) {
     block->Flush();
   }
@@ -49,7 +49,7 @@ ProgramDescBind::ProgramDescBind(const ProgramDescBind &o) {
   }
 }
 
-ProgramDescBind::ProgramDescBind(const ProgramDesc &desc) {
+ProgramDescBind::ProgramDescBind(const proto::ProgramDesc &desc) {
   desc_ = desc;
   for (auto &block_desc : *desc_.mutable_blocks()) {
     blocks_.emplace_back(new BlockDescBind(this, &block_desc));

paddle/framework/program_desc.h

@@ -29,7 +29,7 @@ class ProgramDescBind {
  public:
   ProgramDescBind();
 
-  explicit ProgramDescBind(const ProgramDesc &desc);
+  explicit ProgramDescBind(const proto::ProgramDesc &desc);
 
   ProgramDescBind(const ProgramDescBind &o);
 
@@ -43,10 +43,10 @@ class ProgramDescBind {
 
   size_t Size() const { return blocks_.size(); }
 
-  ProgramDesc *Proto();
+  proto::ProgramDesc *Proto();
 
  private:
-  ProgramDesc desc_;
+  proto::ProgramDesc desc_;
 
   std::vector<std::unique_ptr<BlockDescBind>> blocks_;
 };

paddle/framework/program_desc_test.cc

@@ -22,15 +22,15 @@ TEST(ProgramDesc, copy_ctor) {
   ProgramDescBind program;
   auto* global_block = program.MutableBlock(0);
   auto* x = global_block->Var("X");
-  x->SetType(VarDesc_VarType_LOD_TENSOR);
+  x->SetType(proto::VarDesc_VarType_LOD_TENSOR);
   x->SetLoDLevel(0);
-  x->SetDataType(FP32);
+  x->SetDataType(proto::FP32);
   x->SetShape({1000, 784});
 
   auto* y = global_block->Var("Y");
-  y->SetType(VarDesc_VarType_LOD_TENSOR);
+  y->SetType(proto::VarDesc_VarType_LOD_TENSOR);
   y->SetLoDLevel(0);
-  y->SetDataType(FP32);
+  y->SetDataType(proto::FP32);
   y->SetShape({784, 100});
 
   auto* op = global_block->AppendOp();
@@ -39,7 +39,7 @@ TEST(ProgramDesc, copy_ctor) {
   op->SetInput("Y", {y->Name()});
 
   auto* out = global_block->Var("Out");
-  out->SetType(VarDesc_VarType_LOD_TENSOR);
+  out->SetType(proto::VarDesc_VarType_LOD_TENSOR);
   op->SetOutput("Y", {out->Name()});
 
   ProgramDescBind program_copy(program);
@@ -84,15 +84,15 @@ TEST(ProgramDescBind, serialize_and_deserialize) {
   ProgramDescBind program_origin;
   auto* global_block = program_origin.MutableBlock(0);
   auto* x = global_block->Var("X");
-  x->SetType(VarDesc_VarType_LOD_TENSOR);
+  x->SetType(proto::VarDesc_VarType_LOD_TENSOR);
   x->SetLoDLevel(0);
-  x->SetDataType(FP32);
+  x->SetDataType(proto::FP32);
   x->SetShape({1000, 784});
 
   auto* y = global_block->Var("Y");
-  y->SetType(VarDesc_VarType_LOD_TENSOR);
+  y->SetType(proto::VarDesc_VarType_LOD_TENSOR);
   y->SetLoDLevel(0);
-  y->SetDataType(FP32);
+  y->SetDataType(proto::FP32);
   y->SetShape({784, 100});
 
   auto* op = global_block->AppendOp();
@@ -101,7 +101,7 @@ TEST(ProgramDescBind, serialize_and_deserialize) {
   op->SetInput("Y", {y->Name()});
 
   auto* out = global_block->Var("Out");
-  out->SetType(VarDesc_VarType_LOD_TENSOR);
+  out->SetType(proto::VarDesc_VarType_LOD_TENSOR);
   op->SetOutput("Y", {out->Name()});
 
   std::string binary_str;

paddle/framework/prune.cc

@@ -29,7 +29,7 @@ const std::string kFetchOpType = "fetch";
 const std::string kDropOutOpType = "dropout";
 const std::string kBatchNormOpType = "batch_norm";
 
-bool HasDependentVar(const OpDesc& op_desc,
+bool HasDependentVar(const proto::OpDesc& op_desc,
                      const std::set<std::string>& dependent_vars) {
   for (auto& var : op_desc.outputs()) {
     for (auto& argu : var.arguments()) {
@@ -41,14 +41,15 @@ bool HasDependentVar(const OpDesc& op_desc,
   return false;
 }
 
-bool IsTarget(const OpDesc& op_desc) {
+bool IsTarget(const proto::OpDesc& op_desc) {
   if (op_desc.has_is_target()) {
     return op_desc.is_target();
   }
   return false;
 }
 
-void prune_impl(const ProgramDesc& input, ProgramDesc* output, int block_id) {
+void prune_impl(const proto::ProgramDesc& input, proto::ProgramDesc* output,
+                int block_id) {
   // TODO(tonyyang-svail):
   //    - will change to use multiple blocks for RNN op and Cond Op
 
@@ -104,12 +105,12 @@ void prune_impl(const ProgramDesc& input, ProgramDesc* output, int block_id) {
 }
 
 // TODO(fengjiayi): Prune() could be inplaced to avoid unnecessary copies
-void Prune(const ProgramDesc& input, ProgramDesc* output) {
+void Prune(const proto::ProgramDesc& input, proto::ProgramDesc* output) {
   prune_impl(input, output, 0);
 }
 
-void inference_optimize_impl(const ProgramDesc& input, ProgramDesc* output,
-                             int block_id) {
+void inference_optimize_impl(const proto::ProgramDesc& input,
+                             proto::ProgramDesc* output, int block_id) {
   *output = input;
   auto* op_field = output->mutable_blocks(block_id)->mutable_ops();
   for (auto& op_desc : *op_field) {
@@ -125,7 +126,8 @@ void inference_optimize_impl(const ProgramDesc& input, ProgramDesc* output,
   }
 }
 
-void InferenceOptimize(const ProgramDesc& input, ProgramDesc* output) {
+void InferenceOptimize(const proto::ProgramDesc& input,
+                       proto::ProgramDesc* output) {
   inference_optimize_impl(input, output, 0);
 }
 

paddle/framework/prune.h

@@ -20,9 +20,10 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
-void Prune(const ProgramDesc& input, ProgramDesc* output);
+void Prune(const proto::ProgramDesc& input, proto::ProgramDesc* output);
 
-void InferenceOptimize(const ProgramDesc& input, ProgramDesc* output);
+void InferenceOptimize(const proto::ProgramDesc& input,
+                       proto::ProgramDesc* output);
 
 }  // namespace framework
 }  // namespace paddle

paddle/framework/prune_test.cc

@@ -34,7 +34,7 @@ void AddOp(const std::string &type, const f::VariableNameMap &inputs,
   for (auto kv : outputs) {
     for (auto v : kv.second) {
       auto var = block->Var(v);
-      var->SetDataType(paddle::framework::DataType::FP32);
+      var->SetDataType(paddle::framework::proto::DataType::FP32);
     }
   }
 
@@ -57,14 +57,14 @@ TEST(Prune, one_operator) {
   AddOp("one_one", {{"input", {"a"}}}, {{"output", {"b"}}}, f::AttributeMap{},
         block);
 
-  f::ProgramDesc *pdesc = program.Proto();
-  f::ProgramDesc pruned;
+  f::proto::ProgramDesc *pdesc = program.Proto();
+  f::proto::ProgramDesc pruned;
 
-  Prune(*pdesc, &pruned);
+  f::Prune(*pdesc, &pruned);
   PADDLE_ENFORCE_EQ(pruned.blocks(0).ops_size(), 0);
 
   pdesc->mutable_blocks(0)->mutable_ops(0)->set_is_target(true);
-  Prune(*pdesc, &pruned);
+  f::Prune(*pdesc, &pruned);
   PADDLE_ENFORCE_EQ(pruned.blocks(0).ops_size(), 1);
 }
 
@@ -81,12 +81,12 @@ TEST(Prune, forward) {
   AddOp("one_one", {{"input", {"d"}}}, {{"output", {"e"}}}, f::AttributeMap{},
         block);
 
-  f::ProgramDesc *pdesc = program.Proto();
+  f::proto::ProgramDesc *pdesc = program.Proto();
 
   for (int i = 0; i < pdesc->blocks(0).ops_size(); ++i) {
-    f::ProgramDesc pruned;
+    f::proto::ProgramDesc pruned;
     pdesc->mutable_blocks(0)->mutable_ops(i)->set_is_target(true);
-    Prune(*pdesc, &pruned);
+    f::Prune(*pdesc, &pruned);
     PADDLE_ENFORCE_EQ(pruned.blocks(0).ops_size(), i + 1);
   }
 }
@@ -104,11 +104,11 @@ TEST(Prune, multi_input_op) {
   AddOp("three_one", {{"input", {"b0", "b1", "b2"}}}, {{"output", {"c"}}},
         f::AttributeMap{}, block);
 
-  f::ProgramDesc *pdesc = program.Proto();
+  f::proto::ProgramDesc *pdesc = program.Proto();
   pdesc->mutable_blocks(0)->mutable_ops(3)->set_is_target(true);
 
-  f::ProgramDesc pruned;
-  Prune(*pdesc, &pruned);
+  f::proto::ProgramDesc pruned;
+  f::Prune(*pdesc, &pruned);
   PADDLE_ENFORCE_EQ(pruned.blocks(0).ops_size(), 4);
 }
 
@@ -123,11 +123,11 @@ TEST(Prune, multi_output_op) {
   AddOp("one_one", {{"input", {"c"}}}, {{"output", {"c1"}}}, f::AttributeMap{},
         block);
 
-  f::ProgramDesc *pdesc = program.Proto();
+  f::proto::ProgramDesc *pdesc = program.Proto();
   pdesc->mutable_blocks(0)->mutable_ops(2)->set_is_target(true);
 
-  f::ProgramDesc pruned;
-  Prune(*pdesc, &pruned);
+  f::proto::ProgramDesc pruned;
+  f::Prune(*pdesc, &pruned);
   PADDLE_ENFORCE_EQ(pruned.blocks(0).ops_size(), 2);
 }
 
@@ -142,11 +142,11 @@ TEST(Prune, multi_target) {
   AddOp("one_one", {{"input", {"c"}}}, {{"output", {"c1"}}}, f::AttributeMap{},
         block);
 
-  f::ProgramDesc *pdesc = program.Proto();
+  f::proto::ProgramDesc *pdesc = program.Proto();
   pdesc->mutable_blocks(0)->mutable_ops(1)->set_is_target(true);
   pdesc->mutable_blocks(0)->mutable_ops(2)->set_is_target(true);
 
-  f::ProgramDesc pruned;
-  Prune(*pdesc, &pruned);
+  f::proto::ProgramDesc pruned;
+  f::Prune(*pdesc, &pruned);
   PADDLE_ENFORCE_EQ(pruned.blocks(0).ops_size(), 3);
 }

paddle/framework/shape_inference.cc

@@ -57,17 +57,17 @@ void InferShapeContext::SetDims(const std::vector<std::string> &names,
     SetDim(names[i], dims[i]);
   }
 }
-std::vector<VarDesc::VarType> InferShapeContext::GetInputsVarType(
+std::vector<proto::VarDesc::VarType> InferShapeContext::GetInputsVarType(
     const std::string &name) const {
   return GetVarTypes(Inputs(name));
 }
-std::vector<VarDesc::VarType> InferShapeContext::GetOutputsVarType(
+std::vector<proto::VarDesc::VarType> InferShapeContext::GetOutputsVarType(
     const std::string &name) const {
   return GetVarTypes(Outputs(name));
 }
-std::vector<VarDesc::VarType> InferShapeContext::GetVarTypes(
+std::vector<proto::VarDesc::VarType> InferShapeContext::GetVarTypes(
     const std::vector<std::string> &names) const {
-  std::vector<VarDesc::VarType> retv;
+  std::vector<proto::VarDesc::VarType> retv;
   retv.resize(names.size());
   std::transform(names.begin(), names.end(), retv.begin(),
                  std::bind(std::mem_fn(&InferShapeContext::GetVarType), this,

paddle/framework/shape_inference.h

@@ -27,8 +27,9 @@ class InferShapeContext {
   virtual bool HasInput(const std::string &name) const = 0;
   virtual bool HasOutput(const std::string &name) const = 0;
 
-  std::vector<VarDesc::VarType> GetInputsVarType(const std::string &name) const;
-  std::vector<VarDesc::VarType> GetOutputsVarType(
+  std::vector<proto::VarDesc::VarType> GetInputsVarType(
+      const std::string &name) const;
+  std::vector<proto::VarDesc::VarType> GetOutputsVarType(
       const std::string &name) const;
 
   virtual bool HasInputs(const std::string &name) const = 0;
@@ -65,10 +66,10 @@ class InferShapeContext {
   std::vector<framework::DDim> GetDims(
       const std::vector<std::string> &names) const;
 
-  std::vector<VarDesc::VarType> GetVarTypes(
+  std::vector<proto::VarDesc::VarType> GetVarTypes(
       const std::vector<std::string> &names) const;
 
-  virtual VarDesc::VarType GetVarType(const std::string &name) const = 0;
+  virtual proto::VarDesc::VarType GetVarType(const std::string &name) const = 0;
 };
 
 }  // namespace framework

paddle/framework/var_desc.cc

@@ -18,15 +18,17 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
-VarDesc::VarType VarDescBind::GetType() const { return desc_.type(); }
+proto::VarDesc::VarType VarDescBind::GetType() const { return desc_.type(); }
 
-void VarDescBind::SetType(VarDesc::VarType type) { desc_.set_type(type); }
+void VarDescBind::SetType(proto::VarDesc::VarType type) {
+  desc_.set_type(type);
+}
 
 void VarDescBind::SetShape(const std::vector<int64_t> &dims) {
   VectorToRepeated(dims, mutable_tensor_desc()->mutable_dims());
 }
 
-void VarDescBind::SetDataType(DataType data_type) {
+void VarDescBind::SetDataType(proto::DataType data_type) {
   mutable_tensor_desc()->set_data_type(data_type);
 }
 
@@ -34,14 +36,16 @@ std::vector<int64_t> VarDescBind::Shape() const {
   return RepeatedToVector(tensor_desc().dims());
 }
 
-DataType VarDescBind::GetDataType() const { return tensor_desc().data_type(); }
+proto::DataType VarDescBind::GetDataType() const {
+  return tensor_desc().data_type();
+}
 
 void VarDescBind::SetLoDLevel(int32_t lod_level) {
   switch (desc_.type()) {
-    case VarDesc::LOD_TENSOR:
+    case proto::VarDesc::LOD_TENSOR:
       desc_.mutable_lod_tensor()->set_lod_level(lod_level);
       break;
-    case VarDesc::LOD_TENSOR_ARRAY:
+    case proto::VarDesc::LOD_TENSOR_ARRAY:
       desc_.mutable_tensor_array()->set_lod_level(lod_level);
       break;
     default:
@@ -52,9 +56,9 @@ void VarDescBind::SetLoDLevel(int32_t lod_level) {
 
 int32_t VarDescBind::GetLodLevel() const {
   switch (desc_.type()) {
-    case VarDesc::LOD_TENSOR:
+    case proto::VarDesc::LOD_TENSOR:
       return desc_.lod_tensor().lod_level();
-    case VarDesc::LOD_TENSOR_ARRAY:
+    case proto::VarDesc::LOD_TENSOR_ARRAY:
       return desc_.tensor_array().lod_level();
     default:
       PADDLE_THROW("Tensor type=%d does not support LoDLevel",
@@ -62,29 +66,29 @@ int32_t VarDescBind::GetLodLevel() const {
   }
 }
 
-const TensorDesc &VarDescBind::tensor_desc() const {
+const proto::TensorDesc &VarDescBind::tensor_desc() const {
   PADDLE_ENFORCE(desc_.has_type(), "invoke TensorDesc must after set type");
   switch (desc_.type()) {
-    case VarDesc::SELECTED_ROWS:
+    case proto::VarDesc::SELECTED_ROWS:
       return desc_.selected_rows();
-    case VarDesc::LOD_TENSOR:
+    case proto::VarDesc::LOD_TENSOR:
       return desc_.lod_tensor().tensor();
-    case VarDesc::LOD_TENSOR_ARRAY:
+    case proto::VarDesc::LOD_TENSOR_ARRAY:
       return desc_.tensor_array().tensor();
     default:
       PADDLE_THROW("Unexpected branch.");
   }
 }
 
-TensorDesc *VarDescBind::mutable_tensor_desc() {
+proto::TensorDesc *VarDescBind::mutable_tensor_desc() {
   PADDLE_ENFORCE(desc_.has_type(),
                  "invoke MutableTensorDesc must after set type");
   switch (desc_.type()) {
-    case VarDesc::SELECTED_ROWS:
+    case proto::VarDesc::SELECTED_ROWS:
       return desc_.mutable_selected_rows();
-    case VarDesc::LOD_TENSOR:
+    case proto::VarDesc::LOD_TENSOR:
       return desc_.mutable_lod_tensor()->mutable_tensor();
-    case VarDesc::LOD_TENSOR_ARRAY:
+    case proto::VarDesc::LOD_TENSOR_ARRAY:
       return desc_.mutable_tensor_array()->mutable_tensor();
     default:
       PADDLE_THROW("Unexpected branch.");

paddle/framework/var_desc.h

@@ -57,40 +57,40 @@ class VarDescBind {
  public:
   explicit VarDescBind(const std::string &name) {
     desc_.set_name(name);
-    desc_.set_type(VarDesc::LOD_TENSOR);
+    desc_.set_type(proto::VarDesc::LOD_TENSOR);
   }
 
-  explicit VarDescBind(const VarDesc &desc) : desc_(desc) {}
+  explicit VarDescBind(const proto::VarDesc &desc) : desc_(desc) {}
 
-  VarDesc *Proto() { return &desc_; }
+  proto::VarDesc *Proto() { return &desc_; }
 
   std::string Name() const { return desc_.name(); }
 
   void SetShape(const std::vector<int64_t> &dims);
 
-  void SetDataType(DataType data_type);
+  void SetDataType(proto::DataType data_type);
 
   std::vector<int64_t> Shape() const;
 
-  DataType GetDataType() const;
+  proto::DataType GetDataType() const;
 
   void SetLoDLevel(int32_t lod_level);
 
   int32_t GetLodLevel() const;
 
-  VarDesc::VarType GetType() const;
+  proto::VarDesc::VarType GetType() const;
 
-  void SetType(VarDesc::VarType type);
+  void SetType(proto::VarDesc::VarType type);
 
   bool Persistable() const { return desc_.persistable(); }
 
   void SetPersistable(bool persistable) { desc_.set_persistable(persistable); }
 
  private:
-  const TensorDesc &tensor_desc() const;
-  TensorDesc *mutable_tensor_desc();
+  const proto::TensorDesc &tensor_desc() const;
+  proto::TensorDesc *mutable_tensor_desc();
 
-  VarDesc desc_;
+  proto::VarDesc desc_;
 };
 }  // namespace framework
 }  // namespace paddle

paddle/framework/var_type.h

@@ -20,15 +20,15 @@
 
 namespace paddle {
 namespace framework {
-inline VarDesc::VarType ToVarType(std::type_index type) {
+inline proto::VarDesc::VarType ToVarType(std::type_index type) {
   if (type.hash_code() == typeid(LoDTensor).hash_code()) {
-    return VarDesc_VarType_LOD_TENSOR;
+    return proto::VarDesc_VarType_LOD_TENSOR;
   } else if (type.hash_code() == typeid(LoDRankTable).hash_code()) {
-    return VarDesc_VarType_LOD_RANK_TABLE;
+    return proto::VarDesc_VarType_LOD_RANK_TABLE;
   } else if (type.hash_code() == typeid(LoDTensorArray).hash_code()) {
-    return VarDesc_VarType_LOD_TENSOR_ARRAY;
+    return proto::VarDesc_VarType_LOD_TENSOR_ARRAY;
   } else if (type.hash_code() == typeid(SelectedRows).hash_code()) {
-    return VarDesc_VarType_SELECTED_ROWS;
+    return proto::VarDesc_VarType_SELECTED_ROWS;
   } else {
     PADDLE_THROW("ToVarType:Unsupported type %s", type.name());
   }
@@ -37,16 +37,16 @@ inline VarDesc::VarType ToVarType(std::type_index type) {
 template <typename Visitor>
 inline void VisitVarType(const Variable& var, Visitor visitor) {
   switch (ToVarType(var.Type())) {
-    case VarDesc_VarType_LOD_TENSOR:
+    case proto::VarDesc_VarType_LOD_TENSOR:
       visitor(var.Get<framework::LoDTensor>());
       return;
-    case VarDesc_VarType_LOD_RANK_TABLE:
+    case proto::VarDesc_VarType_LOD_RANK_TABLE:
       visitor(var.Get<LoDRankTable>());
       return;
-    case VarDesc_VarType_LOD_TENSOR_ARRAY:
+    case proto::VarDesc_VarType_LOD_TENSOR_ARRAY:
       visitor(var.Get<LoDTensorArray>());
       return;
-    case VarDesc_VarType_SELECTED_ROWS:
+    case proto::VarDesc_VarType_SELECTED_ROWS:
       visitor(var.Get<SelectedRows>());
       return;
     default:

paddle/framework/var_type_inference_test.cc

@@ -36,14 +36,14 @@ class SumOpVarTypeInference : public VarTypeInference {
   void operator()(const OpDescBind &op_desc,
                   BlockDescBind *block) const override {
     auto &inputs = op_desc.Input("X");
-    auto default_var_type = VarDesc::SELECTED_ROWS;
+    auto default_var_type = proto::VarDesc::SELECTED_ROWS;
 
     bool any_input_is_lod_tensor = std::any_of(
         inputs.begin(), inputs.end(), [block](const std::string &name) {
-          return block->Var(name)->GetType() == VarDesc::LOD_TENSOR;
+          return block->Var(name)->GetType() == proto::VarDesc::LOD_TENSOR;
         });
     if (any_input_is_lod_tensor) {
-      default_var_type = VarDesc::LOD_TENSOR;
+      default_var_type = proto::VarDesc::LOD_TENSOR;
     }
 
     auto out_var_name = op_desc.Output("Out").front();
@@ -68,19 +68,19 @@ TEST(InferVarType, sum_op) {
   op->SetInput("X", {"test_a", "test_b", "test_c"});
   op->SetOutput("Out", {"test_out"});
 
-  prog.MutableBlock(0)->Var("test_a")->SetType(VarDesc::SELECTED_ROWS);
-  prog.MutableBlock(0)->Var("test_b")->SetType(VarDesc::SELECTED_ROWS);
-  prog.MutableBlock(0)->Var("test_c")->SetType(VarDesc::SELECTED_ROWS);
+  prog.MutableBlock(0)->Var("test_a")->SetType(proto::VarDesc::SELECTED_ROWS);
+  prog.MutableBlock(0)->Var("test_b")->SetType(proto::VarDesc::SELECTED_ROWS);
+  prog.MutableBlock(0)->Var("test_c")->SetType(proto::VarDesc::SELECTED_ROWS);
   prog.MutableBlock(0)->Var("test_out");
 
   op->InferVarType(prog.MutableBlock(0));
 
-  ASSERT_EQ(VarDesc::SELECTED_ROWS,
+  ASSERT_EQ(proto::VarDesc::SELECTED_ROWS,
             prog.MutableBlock(0)->Var("test_out")->GetType());
 
-  prog.MutableBlock(0)->Var("test_b")->SetType(VarDesc::LOD_TENSOR);
+  prog.MutableBlock(0)->Var("test_b")->SetType(proto::VarDesc::LOD_TENSOR);
   op->InferVarType(prog.MutableBlock(0));
-  ASSERT_EQ(VarDesc::LOD_TENSOR,
+  ASSERT_EQ(proto::VarDesc::LOD_TENSOR,
             prog.MutableBlock(0)->Var("test_out")->GetType());
 }
 
@@ -91,14 +91,14 @@ TEST(InferVarType, sum_op_without_infer_var_type) {
   op->SetInput("X", {"test2_a", "test2_b", "test2_c"});
   op->SetOutput("Out", {"test2_out"});
 
-  prog.MutableBlock(0)->Var("test2_a")->SetType(VarDesc::SELECTED_ROWS);
-  prog.MutableBlock(0)->Var("test2_b")->SetType(VarDesc::SELECTED_ROWS);
-  prog.MutableBlock(0)->Var("test2_c")->SetType(VarDesc::SELECTED_ROWS);
+  prog.MutableBlock(0)->Var("test2_a")->SetType(proto::VarDesc::SELECTED_ROWS);
+  prog.MutableBlock(0)->Var("test2_b")->SetType(proto::VarDesc::SELECTED_ROWS);
+  prog.MutableBlock(0)->Var("test2_c")->SetType(proto::VarDesc::SELECTED_ROWS);
   prog.MutableBlock(0)->Var("test2_out");
 
   op->InferVarType(prog.MutableBlock(0));
 
-  ASSERT_EQ(VarDesc_VarType_LOD_TENSOR,
+  ASSERT_EQ(proto::VarDesc_VarType_LOD_TENSOR,
             prog.MutableBlock(0)->Var("test2_out")->GetType());
 }
 

paddle/operators/accuracy_op.cc

@@ -63,8 +63,7 @@ class AccuracyOp : public framework::OperatorWithKernel {
 
 class AccuracyOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  AccuracyOpMaker(framework::OpProto *proto,
-                  framework::OpAttrChecker *op_checker)
+  AccuracyOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     // TODO(typhoonzero): support both inference value and indices.
     AddInput("Out", "The network output of topk (inferences)");

paddle/operators/activation_op.cc

@@ -38,9 +38,8 @@ class ActivationOpGrad : public framework::OperatorWithKernel {
 
 class SigmoidOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  SigmoidOpMaker(framework::OpProto *proto,
-                 framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  SigmoidOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Sigmoid operator");
     AddOutput("Y", "Output of Sigmoid operator");
     AddComment(R"DOC(
@@ -54,9 +53,8 @@ $$y = \frac{1}{1 + e^{-x}}$$
 
 class LogSigmoidOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  LogSigmoidOpMaker(framework::OpProto *proto,
-                    framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  LogSigmoidOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of LogSigmoid operator");
     AddOutput("Y", "Output of LogSigmoid operator");
     AddComment(R"DOC(
@@ -70,8 +68,8 @@ $$y = \log \frac{1}{1 + e^{-x}}$$
 
 class ExpOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ExpOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  ExpOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Exp operator");
     AddOutput("Y", "Output of Exp operator");
     AddComment(R"DOC(
@@ -85,8 +83,8 @@ $y = e^x$
 
 class ReluOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ReluOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  ReluOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Relu operator");
     AddOutput("Y", "Output of Relu operator");
     AddComment(R"DOC(
@@ -100,9 +98,8 @@ $y = \max(x, 0)$
 
 class LeakyReluOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  LeakyReluOpMaker(framework::OpProto *proto,
-                   framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  LeakyReluOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of LeakyRelu operator");
     AddOutput("Y", "Output of LeakyRelu operator");
     AddAttr<float>("alpha", "The small negative slope").SetDefault(0.02f);
@@ -117,9 +114,8 @@ $y = \max(x, \alpha * x)$
 
 class SoftShrinkOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  SoftShrinkOpMaker(framework::OpProto *proto,
-                    framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  SoftShrinkOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Softshrink operator");
     AddOutput("Y", "Output of Softshrink operator");
     AddAttr<float>("lambda", "non-negative offset").SetDefault(0.5f);
@@ -140,8 +136,8 @@ $$
 
 class TanhOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  TanhOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  TanhOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Tanh operator");
     AddOutput("Y", "Output of Tanh operator");
     AddComment(R"DOC(
@@ -155,9 +151,8 @@ $$y = \frac{e^{x} - e^{-x}}{e^{x} + e^{-x}}$$
 
 class TanhShrinkOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  TanhShrinkOpMaker(framework::OpProto *proto,
-                    framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  TanhShrinkOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of TanhShrink operator");
     AddOutput("Y", "Output of TanhShrink operator");
     AddComment(R"DOC(
@@ -171,9 +166,8 @@ $$y = x - \frac{e^{x} - e^{-x}}{e^{x} + e^{-x}}$$
 
 class HardShrinkOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  HardShrinkOpMaker(framework::OpProto *proto,
-                    framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  HardShrinkOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of HardShrink operator");
     AddOutput("Y", "Output of HardShrink operator");
     AddAttr<float>("threshold", "The value of threshold for HardShrink")
@@ -195,8 +189,8 @@ $$
 
 class SqrtOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  SqrtOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  SqrtOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Sqrt operator");
     AddOutput("Y", "Output of Sqrt operator");
     AddComment(R"DOC(
@@ -210,8 +204,8 @@ $y = \sqrt{x}$
 
 class AbsOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  AbsOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  AbsOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Abs operator");
     AddOutput("Y", "Output of Abs operator");
     AddComment(R"DOC(
@@ -225,8 +219,8 @@ $y = |x|$
 
 class CeilOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  CeilOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  CeilOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Ceil operator");
     AddOutput("Y", "Output of Ceil operator");
     AddComment(R"DOC(
@@ -240,8 +234,8 @@ $y = ceil(x)$
 
 class FloorOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  FloorOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  FloorOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Floor operator");
     AddOutput("Y", "Output of Floor operator");
     AddComment(R"DOC(
@@ -255,8 +249,8 @@ $y = floor(x)$
 
 class RoundOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  RoundOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  RoundOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Round operator");
     AddOutput("Y", "Output of Round operator");
     AddComment(R"DOC(
@@ -270,9 +264,8 @@ $y = [x]$
 
 class ReciprocalOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ReciprocalOpMaker(framework::OpProto *proto,
-                    framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  ReciprocalOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Reciprocal operator");
     AddOutput("Y", "Output of Reciprocal operator");
     AddComment(R"DOC(
@@ -286,8 +279,8 @@ $$y = \frac{1}{x}$$
 
 class LogOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  LogOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  LogOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Log operator");
     AddOutput("Y", "Output of Log operator");
     AddComment(R"DOC(
@@ -303,8 +296,8 @@ Natural logarithm of x.
 
 class SquareOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  SquareOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  SquareOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Square operator");
     AddOutput("Y", "Output of Square operator");
     AddComment(R"DOC(
@@ -318,9 +311,8 @@ $y = x^2$
 
 class SoftplusOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  SoftplusOpMaker(framework::OpProto *proto,
-                  framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  SoftplusOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Softplus operator");
     AddOutput("Y", "Output of Softplus operator");
     AddComment(R"DOC(
@@ -334,9 +326,8 @@ $y = \ln(1 + e^{x})$
 
 class SoftsignOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  SoftsignOpMaker(framework::OpProto *proto,
-                  framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  SoftsignOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Softsign operator");
     AddOutput("Y", "Output of Softsign operator");
     AddComment(R"DOC(
@@ -350,8 +341,8 @@ $$y = \frac{x}{1 + |x|}$$
 
 class BReluOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  BReluOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  BReluOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of BRelu operator");
     AddOutput("Y", "Output of BRelu operator");
     AddAttr<float>("t_min", "The min marginal value of BRelu")
@@ -369,9 +360,8 @@ $y = \max(\min(x, t_{min}), t_{max})$
 
 class SoftReluOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  SoftReluOpMaker(framework::OpProto *proto,
-                  framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  SoftReluOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of SoftRelu operator");
     AddOutput("Y", "Output of SoftRelu operator");
     AddAttr<float>("threshold", "The threshold value of SoftRelu")
@@ -387,8 +377,8 @@ $y = \ln(1 + \exp(\max(\min(x, threshold), threshold))$
 
 class ELUOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ELUOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  ELUOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of ELU operator");
     AddOutput("Y", "Output of ELU operator");
     AddAttr<float>("alpha", "The alpha value of ELU").SetDefault(1.0f);
@@ -406,8 +396,8 @@ $y = \max(0, x) + \min(0, \alpha * (e^x - 1))$
 
 class Relu6OpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  Relu6OpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  Relu6OpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Relu6 operator");
     AddOutput("Y", "Output of Relu6 operator");
     AddAttr<float>("threshold", "The threshold value of Relu6")
@@ -423,8 +413,8 @@ $y = \min(\max(0, x), 6)$
 
 class PowOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  PowOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  PowOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Pow operator");
     AddOutput("Y", "Output of Pow operator");
     AddAttr<float>("factor", "The exponential factor of Pow").SetDefault(1.0f);
@@ -439,8 +429,8 @@ $y = x^{factor}$
 
 class STanhOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  STanhOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  STanhOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of STanh operator");
     AddOutput("Y", "Output of STanh operator");
     AddAttr<float>("scale_a", "The scale parameter of a for the input")
@@ -458,9 +448,8 @@ $$y = b * \frac{e^{a * x} - e^{-a * x}}{e^{a * x} + e^{-a * x}}$$
 
 class ThresholdedReluOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ThresholdedReluOpMaker(framework::OpProto *proto,
-                         framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  ThresholdedReluOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of ThresholdedRelu operator");
     AddOutput("Y", "Output of ThresholdedRelu operator");
     AddAttr<float>("threshold", "The threshold location of activation")
@@ -481,9 +470,8 @@ $$
 
 class HardSigmoidOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  HardSigmoidOpMaker(framework::OpProto *proto,
-                     framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  HardSigmoidOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of HardSigmoid operator");
     AddOutput("Y", "Output of HardSigmoid operator");
     AddAttr<float>("slope", "Slope for linear approximation of sigmoid")
@@ -508,8 +496,8 @@ It is recommended to use the defaults for this activation.
 
 class SwishOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  SwishOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  SwishOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input of Swish operator");
     AddOutput("Y", "Output of Swish operator");
     AddAttr<float>("beta", "Constant beta of swish operator").SetDefault(1.0f);

paddle/operators/adadelta_op.cc

@@ -59,8 +59,7 @@ class AdadeltaOp : public framework::OperatorWithKernel {
 
 class AdadeltaOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  AdadeltaOpMaker(framework::OpProto *proto,
-                  framework::OpAttrChecker *op_checker)
+  AdadeltaOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Param", "(Tensor) Input parameter");
     AddInput("Grad", "(Tensor) Input gradient");

paddle/operators/adagrad_op.cc

@@ -59,8 +59,7 @@ class AdagradOp : public framework::OperatorWithKernel {
 
 class AdagradOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  AdagradOpMaker(framework::OpProto* proto,
-                 framework::OpAttrChecker* op_checker)
+  AdagradOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Param", "(Tensor) Input parameter");
     AddInput("Grad", "(Tensor) Input gradient");

paddle/operators/adam_op.cc

@@ -73,7 +73,7 @@ class AdamOp : public framework::OperatorWithKernel {
 
 class AdamOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  AdamOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+  AdamOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Param", "(Tensor) Input parameter");
     AddInput("Grad", "(Tensor) Input gradient");

paddle/operators/adamax_op.cc

@@ -67,7 +67,7 @@ class AdamaxOp : public framework::OperatorWithKernel {
 
 class AdamaxOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  AdamaxOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+  AdamaxOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Param", "(Tensor) Input parameter");
     AddInput("Grad", "(Tensor) Input gradient");

paddle/operators/array_to_lod_tensor_op.cc

@@ -114,8 +114,7 @@ class ArrayToLoDTensorOp : public framework::OperatorBase {
 
 class ArrayToLoDTensorOpProtoMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ArrayToLoDTensorOpProtoMaker(framework::OpProto *proto,
-                               framework::OpAttrChecker *op_checker)
+  ArrayToLoDTensorOpProtoMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X",
              "(std::vector<LodTensor>) A vector of tensors that is going to "

paddle/operators/assign_op.cc

@@ -86,8 +86,7 @@ class AssignOp : public framework::OperatorBase {
 
 class AssignOpProtoMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  AssignOpProtoMaker(framework::OpProto *proto,
-                     framework::OpAttrChecker *op_checker)
+  AssignOpProtoMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X",
              "(LoDTensor, SelectedRows or LoDTensorArray) The input variable "
@@ -109,8 +108,8 @@ class AssignInferShape : public framework::InferShapeBase {
   void operator()(framework::InferShapeContext *context) const override {
     if (context->HasInput("X")) {
       auto type = context->GetInputsVarType("X")[0];
-      if (type == framework::VarDesc_VarType_SELECTED_ROWS ||
-          type == framework::VarDesc_VarType_LOD_TENSOR) {
+      if (type == framework::proto::VarDesc_VarType_SELECTED_ROWS ||
+          type == framework::proto::VarDesc_VarType_LOD_TENSOR) {
         context->SetOutputDim("Out", context->GetInputDim("X"));
       }
     }

paddle/operators/auc_op.cc

@@ -49,7 +49,7 @@ class AucOp : public framework::OperatorWithKernel {
 
 class AucOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  AucOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+  AucOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Out",
              "A floating point 2D tensor, values are in the range [0, 1]."

paddle/operators/batch_norm_op.cc

@@ -85,8 +85,7 @@ class BatchNormOp : public framework::OperatorWithKernel {
 
 class BatchNormOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  BatchNormOpMaker(framework::OpProto *proto,
-                   framework::OpAttrChecker *op_checker)
+  BatchNormOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddAttr<bool>("is_test", "").SetDefault(false);
     AddAttr<float>("momentum", "").SetDefault(0.9);

paddle/operators/beam_search_decode_op.cc

@@ -83,9 +83,8 @@ class BeamSearchDecodeOp : public framework::OperatorBase {
 
 class BeamSearchDecodeOpProtoMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  BeamSearchDecodeOpProtoMaker(framework::OpProto* proto,
-                               framework::OpAttrChecker* op_checker)
-      : OpProtoAndCheckerMaker(proto, op_checker) {
+  BeamSearchDecodeOpProtoMaker(OpProto* proto, OpAttrChecker* op_checker)
+      : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Ids",
              "(LodTensorArray)"
              "score of the candidate words in each step");
@@ -123,10 +122,10 @@ class BeamSearchDecodeInferVarType : public framework::VarTypeInference {
   void operator()(const framework::OpDescBind& op_desc,
                   framework::BlockDescBind* block) const override {
     for (auto& o : op_desc.Output("SentenceIds")) {
-      block->Var(o)->SetType(framework::VarDesc::LOD_TENSOR);
+      block->Var(o)->SetType(framework::proto::VarDesc::LOD_TENSOR);
     }
     for (auto& o : op_desc.Output("SentenceScores")) {
-      block->Var(o)->SetType(framework::VarDesc::LOD_TENSOR);
+      block->Var(o)->SetType(framework::proto::VarDesc::LOD_TENSOR);
     }
   }
 };

paddle/operators/beam_search_op.cc

@@ -153,8 +153,7 @@ bool BeamSearch::NextItemSet(std::vector<BeamSearch::Item> *items) {
 class BeamSearchProtoAndCheckerMaker
     : public framework::OpProtoAndCheckerMaker {
  public:
-  BeamSearchProtoAndCheckerMaker(framework::OpProto *proto,
-                                 framework::OpAttrChecker *op_checker)
+  BeamSearchProtoAndCheckerMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     // inputs and outputs stored in proto
     AddInput("pre_ids", "ids in previous step");

paddle/operators/bilinear_tensor_product_op.cc

@@ -65,8 +65,7 @@ class BilinearTensorProductOp : public framework::OperatorWithKernel {
 
 class BilinearTensorProductOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  BilinearTensorProductOpMaker(framework::OpProto* proto,
-                               framework::OpAttrChecker* op_checker)
+  BilinearTensorProductOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "The first input of bilinear_tensor_product operator.");
     AddInput("Y", "The second input of bilinear_tensor_product operator.");

paddle/operators/cast_op.cc

@@ -20,8 +20,7 @@ namespace operators {
 
 class CastOpProtoMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  CastOpProtoMaker(framework::OpProto *proto,
-                   framework::OpAttrChecker *op_checker)
+  CastOpProtoMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "The input tensor of cast op");
     AddOutput("Out", "The output tensor of cast op");

paddle/operators/cast_op.h

@@ -55,7 +55,7 @@ class CastOpKernel : public framework::OpKernel<InT> {
     auto* in = context.Input<framework::Tensor>("X");
     auto* out = context.Output<framework::Tensor>("Out");
     framework::VisitDataType(
-        static_cast<framework::DataType>(context.Attr<int>("out_dtype")),
+        static_cast<framework::proto::DataType>(context.Attr<int>("out_dtype")),
         CastOpFunctor<DeviceContext, InT>(
             in, out, context.template device_context<DeviceContext>()));
   }

paddle/operators/chunk_eval_op.cc

@@ -57,15 +57,14 @@ class ChunkEvalOp : public framework::OperatorWithKernel {
  protected:
   framework::OpKernelType GetKernelType(
       const framework::ExecutionContext &ctx) const override {
-    return framework::OpKernelType(framework::DataType::FP32,
+    return framework::OpKernelType(framework::proto::DataType::FP32,
                                    ctx.device_context());
   }
 };
 
 class ChunkEvalOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ChunkEvalOpMaker(framework::OpProto *proto,
-                   framework::OpAttrChecker *op_checker)
+  ChunkEvalOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Inference",
              "(Tensor, default: Tensor<int64_t>). "

paddle/operators/clip_by_norm_op.cc

@@ -37,8 +37,7 @@ class ClipByNormOp : public framework::OperatorWithKernel {
 
 class ClipByNormOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ClipByNormOpMaker(framework::OpProto* proto,
-                    framework::OpAttrChecker* op_checker)
+  ClipByNormOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X",
              "(Tensor) The input of clip_by_norm op."

paddle/operators/clip_op.cc

@@ -38,7 +38,7 @@ class ClipOp : public framework::OperatorWithKernel {
 template <typename AttrType>
 class ClipOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ClipOpMaker(framework::OpProto* proto, framework::OpAttrChecker* op_checker)
+  ClipOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X",
              "(Tensor)The input of clip op."

paddle/operators/compare_op.cc

@@ -20,8 +20,7 @@ namespace operators {
 template <typename OpComment>
 class CompareOpProtoMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  CompareOpProtoMaker(framework::OpProto *proto,
-                      framework::OpAttrChecker *op_checker)
+  CompareOpProtoMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     OpComment comment;
     AddInput("X",

paddle/operators/concat_op.cc

@@ -58,7 +58,7 @@ class ConcatOp : public framework::OperatorWithKernel {
 
 class ConcatOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ConcatOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+  ConcatOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "Input tensors of concat operator.").AsDuplicable();
     AddOutput("Out", "Output tensor of concat operator.");

paddle/operators/cond_op.cc

@@ -205,8 +205,7 @@ void CondOp::Run(const Scope& scope,
 
 class CondOpProtoAndCheckerMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  CondOpProtoAndCheckerMaker(framework::OpProto* proto,
-                             framework::OpAttrChecker* op_checker)
+  CondOpProtoAndCheckerMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Cond", "The condition, which is a bool vector");
     AddInput("Xs", "Inputs of Subnets").AsDuplicable();

paddle/operators/conditional_block_op.cc

@@ -74,8 +74,7 @@ class ConditionalBlockOp : public ConditionalOp {
 
 class ConditionalBlockOpProtoMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ConditionalBlockOpProtoMaker(framework::OpProto *proto,
-                               framework::OpAttrChecker *op_checker)
+  ConditionalBlockOpProtoMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X",
              "The conditional variable of this operator. If X is empty, the "

paddle/operators/conv_cudnn_op.cc

@@ -19,8 +19,7 @@ namespace operators {
 
 class CudnnConv2DOpMaker : public Conv2DOpMaker {
  public:
-  CudnnConv2DOpMaker(framework::OpProto* proto,
-                     framework::OpAttrChecker* op_checker)
+  CudnnConv2DOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : Conv2DOpMaker(proto, op_checker) {
     AddAttr<int>("workspace_size_MB",
                  "workspace size for cudnn, in MB, "
@@ -34,8 +33,7 @@ class CudnnConv2DOpMaker : public Conv2DOpMaker {
 
 class CudnnConv3DOpMaker : public Conv3DOpMaker {
  public:
-  CudnnConv3DOpMaker(framework::OpProto* proto,
-                     framework::OpAttrChecker* op_checker)
+  CudnnConv3DOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : Conv3DOpMaker(proto, op_checker) {
     AddAttr<int>("workspace_size_MB",
                  "workspace size for cudnn, in MB, "

paddle/operators/conv_op.cc

@@ -66,8 +66,7 @@ void ConvOp::InferShape(framework::InferShapeContext* ctx) const {
   ctx->SetOutputDim("Output", framework::make_ddim(output_shape));
 }
 
-Conv2DOpMaker::Conv2DOpMaker(framework::OpProto* proto,
-                             framework::OpAttrChecker* op_checker)
+Conv2DOpMaker::Conv2DOpMaker(OpProto* proto, OpAttrChecker* op_checker)
     : OpProtoAndCheckerMaker(proto, op_checker) {
   AddInput(
       "Input",
@@ -138,8 +137,7 @@ $$
 )DOC");
 }
 
-Conv3DOpMaker::Conv3DOpMaker(framework::OpProto* proto,
-                             framework::OpAttrChecker* op_checker)
+Conv3DOpMaker::Conv3DOpMaker(OpProto* proto, OpAttrChecker* op_checker)
     : OpProtoAndCheckerMaker(proto, op_checker) {
   AddInput(
       "Input",

paddle/operators/conv_op.h

@@ -50,14 +50,12 @@ inline bool IsExpand(std::vector<int64_t>& filter_dim,
 // operator implementations can reuse the code.
 class Conv2DOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  Conv2DOpMaker(framework::OpProto* proto,
-                framework::OpAttrChecker* op_checker);
+  Conv2DOpMaker(OpProto* proto, OpAttrChecker* op_checker);
 };
 
 class Conv3DOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  Conv3DOpMaker(framework::OpProto* proto,
-                framework::OpAttrChecker* op_checker);
+  Conv3DOpMaker(OpProto* proto, OpAttrChecker* op_checker);
 };
 
 class ConvOp : public framework::OperatorWithKernel {

paddle/operators/conv_shift_op.cc

@@ -75,8 +75,7 @@ class ConvShiftGradOp : public framework::OperatorWithKernel {
 
 class ConvShiftOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ConvShiftOpMaker(framework::OpProto *proto,
-                   framework::OpAttrChecker *op_checker)
+  ConvShiftOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X",
              "(Tensor, default Tensor<float>), a 2-D tensor with shape B x M, "

paddle/operators/conv_transpose_cudnn_op.cc

@@ -19,8 +19,7 @@ namespace operators {
 
 class CudnnConv2DTransposeOpMaker : public Conv2DTransposeOpMaker {
  public:
-  CudnnConv2DTransposeOpMaker(framework::OpProto* proto,
-                              framework::OpAttrChecker* op_checker)
+  CudnnConv2DTransposeOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : Conv2DTransposeOpMaker(proto, op_checker) {
     AddAttr<std::vector<int>>("dilations", "dilations of convolution operator.")
         .SetDefault({1, 1});
@@ -36,8 +35,7 @@ class CudnnConv2DTransposeOpMaker : public Conv2DTransposeOpMaker {
 
 class CudnnConv3DTransposeOpMaker : public Conv3DTransposeOpMaker {
  public:
-  CudnnConv3DTransposeOpMaker(framework::OpProto* proto,
-                              framework::OpAttrChecker* op_checker)
+  CudnnConv3DTransposeOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : Conv3DTransposeOpMaker(proto, op_checker) {
     AddAttr<std::vector<int>>("dilations", "dilations of convolution operator.")
         .SetDefault({1, 1, 1});

paddle/operators/conv_transpose_op.cc

@@ -53,8 +53,8 @@ void ConvTransposeOp::InferShape(framework::InferShapeContext* ctx) const {
   ctx->SetOutputDim("Output", framework::make_ddim(output_shape));
 }
 
-Conv2DTransposeOpMaker::Conv2DTransposeOpMaker(
-    framework::OpProto* proto, framework::OpAttrChecker* op_checker)
+Conv2DTransposeOpMaker::Conv2DTransposeOpMaker(OpProto* proto,
+                                               OpAttrChecker* op_checker)
     : OpProtoAndCheckerMaker(proto, op_checker) {
   AddInput(
       "Input",
@@ -112,8 +112,8 @@ Example:
 )DOC");
 }
 
-Conv3DTransposeOpMaker::Conv3DTransposeOpMaker(
-    framework::OpProto* proto, framework::OpAttrChecker* op_checker)
+Conv3DTransposeOpMaker::Conv3DTransposeOpMaker(OpProto* proto,
+                                               OpAttrChecker* op_checker)
     : OpProtoAndCheckerMaker(proto, op_checker) {
   AddInput("Input",
            "(Tensor) The input tensor of convolution transpose operator."

paddle/operators/conv_transpose_op.h

@@ -30,14 +30,12 @@ using DDim = framework::DDim;
 // operator implementations can reuse the code.
 class Conv2DTransposeOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  Conv2DTransposeOpMaker(framework::OpProto* proto,
-                         framework::OpAttrChecker* op_checker);
+  Conv2DTransposeOpMaker(OpProto* proto, OpAttrChecker* op_checker);
 };
 
 class Conv3DTransposeOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  Conv3DTransposeOpMaker(framework::OpProto* proto,
-                         framework::OpAttrChecker* op_checker);
+  Conv3DTransposeOpMaker(OpProto* proto, OpAttrChecker* op_checker);
 };
 
 class ConvTransposeOp : public framework::OperatorWithKernel {

paddle/operators/cos_sim_op.cc

@@ -62,7 +62,7 @@ class CosSimOp : public framework::OperatorWithKernel {
 
 class CosSimOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  CosSimOpMaker(framework::OpProto* proto, framework::OpAttrChecker* op_checker)
+  CosSimOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "The 1st input of cos_sim op.");
     AddInput("Y", "The 2nd input of cos_sim op.");

paddle/operators/crf_decoding_op.cc

@@ -18,8 +18,7 @@ namespace paddle {
 namespace operators {
 class CRFDecodingOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  CRFDecodingOpMaker(framework::OpProto* proto,
-                     framework::OpAttrChecker* op_checker)
+  CRFDecodingOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Emission",
              "(LoDTensor, default: LoDTensor<float>). A LoDTensor with shape "

paddle/operators/crop_op.cc

@@ -52,7 +52,7 @@ class CropOp : public framework::OperatorWithKernel {
 
 class CropOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  CropOpMaker(framework::OpProto* proto, framework::OpAttrChecker* op_checker)
+  CropOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X",
              "The input of pad op. "

paddle/operators/cross_entropy_op.cc

@@ -111,8 +111,7 @@ class CrossEntropyGradientOp : public framework::OperatorWithKernel {
 
 class CrossEntropyOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  CrossEntropyOpMaker(framework::OpProto* proto,
-                      framework::OpAttrChecker* op_checker)
+  CrossEntropyOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X",
              "(Tensor, default Tensor<float>), a 2-D tensor with shape N x D, "

paddle/operators/decayed_adagrad_op.cc

@@ -55,8 +55,7 @@ class DecayedAdagradOp : public framework::OperatorWithKernel {
 
 class DecayedAdagradOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  DecayedAdagradOpMaker(framework::OpProto *proto,
-                        framework::OpAttrChecker *op_checker)
+  DecayedAdagradOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Param", "(Tensor) Input parameter");
     AddInput("Grad", "(Tensor) Input gradient");

paddle/operators/dropout_op.cc

@@ -40,8 +40,7 @@ class DropoutOp : public framework::OperatorWithKernel {
 template <typename AttrType>
 class DropoutOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  DropoutOpMaker(framework::OpProto* proto,
-                 framework::OpAttrChecker* op_checker)
+  DropoutOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "The input of dropout op.");
     AddOutput("Out", "The output of dropout op.");

paddle/operators/elementwise_add_op.cc

@@ -19,8 +19,7 @@ namespace paddle {
 namespace operators {
 class ElementwiseAddOpMaker : public ElementwiseOpMaker {
  public:
-  ElementwiseAddOpMaker(framework::OpProto* proto,
-                        framework::OpAttrChecker* op_checker)
+  ElementwiseAddOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : ElementwiseOpMaker(proto, op_checker) {
     SetComment("Add", "$Out = X + Y$");
     AddComment(comment_);

paddle/operators/elementwise_div_op.cc

@@ -19,8 +19,7 @@ namespace paddle {
 namespace operators {
 class ElementwiseDivOpMaker : public ElementwiseOpMaker {
  public:
-  ElementwiseDivOpMaker(framework::OpProto* proto,
-                        framework::OpAttrChecker* op_checker)
+  ElementwiseDivOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : ElementwiseOpMaker(proto, op_checker) {
     SetComment("Div", "$Out = X / Y$");
     AddComment(comment_);

paddle/operators/elementwise_mul_op.cc

@@ -20,8 +20,7 @@ namespace operators {
 
 class ElementwiseMulOpMaker : public ElementwiseOpMaker {
  public:
-  ElementwiseMulOpMaker(framework::OpProto* proto,
-                        framework::OpAttrChecker* op_checker)
+  ElementwiseMulOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : ElementwiseOpMaker(proto, op_checker) {
     SetComment("Mul", "$Out = X \\odot\\ Y$");
     AddComment(comment_);

paddle/operators/elementwise_op.h

@@ -43,8 +43,7 @@ class ElementwiseOp : public framework::OperatorWithKernel {
 
 class ElementwiseOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ElementwiseOpMaker(framework::OpProto* proto,
-                     framework::OpAttrChecker* op_checker)
+  ElementwiseOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "(Tensor) The first input tensor of elementwise op");
     AddInput("Y", "(Tensor) The second input tensor of elementwise op");

paddle/operators/elementwise_sub_op.cc

@@ -19,8 +19,7 @@ namespace paddle {
 namespace operators {
 class ElementwiseSubOpMaker : public ElementwiseOpMaker {
  public:
-  ElementwiseSubOpMaker(framework::OpProto* proto,
-                        framework::OpAttrChecker* op_checker)
+  ElementwiseSubOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : ElementwiseOpMaker(proto, op_checker) {
     SetComment("Sub", "$Out = X - Y$");
     AddComment(comment_);

paddle/operators/expand_op.cc

@@ -55,7 +55,7 @@ class ExpandOp : public framework::OperatorWithKernel {
 
 class ExpandOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  ExpandOpMaker(framework::OpProto* proto, framework::OpAttrChecker* op_checker)
+  ExpandOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X",
              "(Tensor, default Tensor<float>) A tensor with rank in [1, 6]."

paddle/operators/feed_op.cc

@@ -54,8 +54,7 @@ class FeedOp : public framework::OperatorBase {
 
 class FeedOpInfoMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  FeedOpInfoMaker(framework::OpProto *proto,
-                  framework::OpAttrChecker *op_checker)
+  FeedOpInfoMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "The input of feed op");
     AddOutput("Out", "The output of feed op");

paddle/operators/fetch_op.cc

@@ -61,8 +61,7 @@ class FetchOp : public framework::OperatorBase {
 
 class FetchOpInfoMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  FetchOpInfoMaker(framework::OpProto *proto,
-                   framework::OpAttrChecker *op_checker)
+  FetchOpInfoMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "The input of fetch op");
     AddOutput("Out", "The output of fetch op");

paddle/operators/fill_constant_batch_size_like_op.cc

@@ -52,7 +52,7 @@ class FillConstantBatchSizeLikeOp : public framework::OperatorWithKernel {
   framework::OpKernelType GetKernelType(
       const framework::ExecutionContext &ctx) const override {
     return framework::OpKernelType(
-        static_cast<framework::DataType>(ctx.Attr<int>("dtype")),
+        static_cast<framework::proto::DataType>(ctx.Attr<int>("dtype")),
         ctx.device_context());
   }
 };
@@ -60,13 +60,12 @@ class FillConstantBatchSizeLikeOp : public framework::OperatorWithKernel {
 class FillConstantBatchSizeLikeOpMaker
     : public framework::OpProtoAndCheckerMaker {
  public:
-  FillConstantBatchSizeLikeOpMaker(framework::OpProto *proto,
-                                   framework::OpAttrChecker *op_checker)
+  FillConstantBatchSizeLikeOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddAttr<int>("dtype",
                  "(int, default 5 (FP32)) "
                  "Output data type")
-        .SetDefault(framework::DataType::FP32);
+        .SetDefault(framework::proto::DataType::FP32);
     AddInput("Input",
              "(Tensor) Tensor "
              "whose dim_idx th dimension is used to specify the batch_size");

paddle/operators/fill_constant_op.cc

@@ -34,7 +34,8 @@ class FillConstantOp : public framework::OperatorBase {
   using framework::OperatorBase::OperatorBase;
   void Run(const framework::Scope &scope,
            const platform::DeviceContext &dev_ctx) const override {
-    auto data_type = static_cast<framework::DataType>(Attr<int>("dtype"));
+    auto data_type =
+        static_cast<framework::proto::DataType>(Attr<int>("dtype"));
     auto value = Attr<float>("value");
     auto force_cpu = Attr<bool>("force_cpu");
     auto &out =
@@ -52,13 +53,12 @@ class FillConstantOp : public framework::OperatorBase {
 
 class FillConstantOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  FillConstantOpMaker(framework::OpProto *proto,
-                      framework::OpAttrChecker *op_checker)
+  FillConstantOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddAttr<int>("dtype",
                  "(int, default 5 (FP32)) "
                  "Output data type")
-        .SetDefault(framework::DataType::FP32);
+        .SetDefault(framework::proto::DataType::FP32);
     AddAttr<std::vector<int>>("shape", "(vector<int>) The shape of the output");
     AddAttr<float>("value", "(float, default 0) The value to be filled")
         .SetDefault(0.0f);

paddle/operators/fill_op.cc

@@ -48,7 +48,7 @@ class FillOp : public framework::OperatorBase {
                                 "Cannot find variable %s", Output("Out"))
                         .GetMutable<framework::LoDTensor>());
     out.Resize(framework::make_ddim(Attr<std::vector<int>>("shape")));
-    auto dtype = static_cast<framework::DataType>(Attr<int>("dtype"));
+    auto dtype = static_cast<framework::proto::DataType>(Attr<int>("dtype"));
     platform::CPUPlace cpu;
     auto force_cpu = Attr<bool>("force_cpu");
     out.mutable_data(force_cpu ? cpu : dev_ctx.GetPlace(),
@@ -76,7 +76,7 @@ class FillOp : public framework::OperatorBase {
 
 class FillOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  FillOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+  FillOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddComment(R"DOC(Fill operator
 
@@ -88,7 +88,7 @@ Fill an tensor with `value` and `shape`. The type of the tensor is specify by
         "value", "The float values of tensor, which are flatten in row major");
     AddAttr<std::vector<int>>("shape", "The shape of output tensor");
     AddAttr<int>("dtype", "The data type of output tensor, Default is float")
-        .SetDefault(framework::DataType::FP32);
+        .SetDefault(framework::proto::DataType::FP32);
     AddAttr<bool>("force_cpu",
                   "Whether the output tensor must be at CPU memory or not. "
                   "Default is false.")

paddle/operators/fill_zeros_like_op.cc

@@ -33,8 +33,7 @@ class FillZerosLikeOp : public framework::OperatorWithKernel {
 
 class FillZerosLikeOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  FillZerosLikeOpMaker(framework::OpProto *proto,
-                       framework::OpAttrChecker *op_checker)
+  FillZerosLikeOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "The input of fill-zeros-like op.");
     AddOutput("Y", "The variable will be filled up with zeros.");

paddle/operators/ftrl_op.cc

@@ -57,7 +57,7 @@ class FTRLOp : public framework::OperatorWithKernel {
 
 class FTRLOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  FTRLOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
+  FTRLOpMaker(OpProto *proto, OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Param",
              "(Tensor, default Tensor<float>) "

paddle/operators/gather_op.cc

@@ -67,7 +67,7 @@ class GatherGradOp : public framework::OperatorWithKernel {
 
 class GatherOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  GatherOpMaker(framework::OpProto* proto, framework::OpAttrChecker* op_checker)
+  GatherOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "The source input of gather op");
     AddInput("Index", "The index input of gather op");

paddle/operators/gaussian_random_op.cc

@@ -60,15 +60,14 @@ class GaussianRandomOp : public framework::OperatorWithKernel {
   framework::OpKernelType GetKernelType(
       const framework::ExecutionContext& ctx) const override {
     return framework::OpKernelType(
-        static_cast<framework::DataType>(ctx.Attr<int>("dtype")),
+        static_cast<framework::proto::DataType>(ctx.Attr<int>("dtype")),
         ctx.device_context());
   }
 };
 
 class GaussianRandomOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  GaussianRandomOpMaker(framework::OpProto* proto,
-                        framework::OpAttrChecker* op_checker)
+  GaussianRandomOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : framework::OpProtoAndCheckerMaker(proto, op_checker) {
     AddOutput("Out", "Output matrix of gaussian random op");
 
@@ -91,7 +90,7 @@ class GaussianRandomOpMaker : public framework::OpProtoAndCheckerMaker {
     AddAttr<int>("dtype",
                  "(int, default 5(FP32)) "
                  "Output data type.")
-        .SetDefault(framework::DataType::FP32);
+        .SetDefault(framework::proto::DataType::FP32);
 
     AddComment(R"DOC(
 GaussianRandom Operator.

paddle/operators/gru_op.cc

@@ -67,7 +67,7 @@ class GRUOp : public framework::OperatorWithKernel {
 
 class GRUOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  GRUOpMaker(framework::OpProto* proto, framework::OpAttrChecker* op_checker)
+  GRUOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Input",
              "(LoDTensor) The first input is a LodTensor, which supports "

paddle/operators/gru_unit_op.cc

@@ -71,8 +71,7 @@ class GRUUnitOp : public framework::OperatorWithKernel {
 
 class GRUUnitOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
-  GRUUnitOpMaker(framework::OpProto* proto,
-                 framework::OpAttrChecker* op_checker)
+  GRUUnitOpMaker(OpProto* proto, OpAttrChecker* op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("Input",
              "(Tensor) Matrix with shape [batch_size, frame_size * 3] for the "