diff --git a/paddle/fluid/framework/data_layout_transform.cc b/paddle/fluid/framework/data_layout_transform.cc
index cd00b7de7338982308acfa1f1e8c38e010c6a43b..c9e3a8ac1d1e5228725bff49ecc6d91e640dfe57 100644
--- a/paddle/fluid/framework/data_layout_transform.cc
+++ b/paddle/fluid/framework/data_layout_transform.cc
@@ -46,7 +46,7 @@ struct CastDataLayout {
   const std::vector<int> axis_;
 
   template <typename T>
-  void operator()() {
+  void apply() {
     auto place = ctx_->GetPlace();
 
     if (platform::is_cpu_place(place)) {
diff --git a/paddle/fluid/framework/data_type.h b/paddle/fluid/framework/data_type.h
index 84691a2059124960a3213802fec0863f8abe6df7..8ad2fb5f3ffd9641932bbbb024a31e81d31dc9bb 100644
--- a/paddle/fluid/framework/data_type.h
+++ b/paddle/fluid/framework/data_type.h
@@ -26,75 +26,40 @@ namespace framework {
 extern proto::VarType::Type ToDataType(std::type_index type);
 extern std::type_index ToTypeIndex(proto::VarType::Type type);
 
-#if !defined(_WIN32)
 template <typename Visitor>
 inline void VisitDataType(proto::VarType::Type type, Visitor visitor) {
   switch (type) {
     case proto::VarType::FP16:
-      visitor.template operator()<platform::float16>();
+      visitor.template apply<platform::float16>();
       break;
     case proto::VarType::FP32:
-      visitor.template operator()<float>();
+      visitor.template apply<float>();
       break;
     case proto::VarType::FP64:
-      visitor.template operator()<double>();
+      visitor.template apply<double>();
       break;
     case proto::VarType::INT32:
-      visitor.template operator()<int>();
+      visitor.template apply<int>();
       break;
     case proto::VarType::INT64:
-      visitor.template operator()<int64_t>();
+      visitor.template apply<int64_t>();
       break;
     case proto::VarType::BOOL:
-      visitor.template operator()<bool>();
+      visitor.template apply<bool>();
       break;
     case proto::VarType::UINT8:
-      visitor.template operator()<uint8_t>();
+      visitor.template apply<uint8_t>();
       break;
     case proto::VarType::INT16:
-      visitor.template operator()<int16_t>();
+      visitor.template apply<int16_t>();
       break;
     case proto::VarType::INT8:
-      visitor.template operator()<int8_t>();
+      visitor.template apply<int8_t>();
       break;
     default:
       PADDLE_THROW("Not supported %d", type);
   }
 }
-#else
-// the msvc compiler do not implement two-stage name lookup correctly.
-template <typename Visitor>
-inline void VisitDataType(proto::VarType::Type type, Visitor visitor) {
-  switch (type) {
-    case proto::VarType::FP16:
-      visitor.operator()<platform::float16>();
-      break;
-    case proto::VarType::FP32:
-      visitor.operator()<float>();
-      break;
-    case proto::VarType::FP64:
-      visitor.operator()<double>();
-      break;
-    case proto::VarType::INT32:
-      visitor.operator()<int>();
-      break;
-    case proto::VarType::INT64:
-      visitor.operator()<int64_t>();
-      break;
-    case proto::VarType::BOOL:
-      visitor.operator()<bool>();
-      break;
-    case proto::VarType::UINT8:
-      visitor.operator()<uint8_t>();
-      break;
-    case proto::VarType::INT16:
-      visitor.operator()<int16_t>();
-      break;
-    default:
-      PADDLE_THROW("Not supported %d", type);
-  }
-}
-#endif  // _WIN32
 
 extern std::string DataTypeToString(const proto::VarType::Type type);
 extern size_t SizeOfType(std::type_index type);
diff --git a/paddle/fluid/framework/data_type_transform.cc b/paddle/fluid/framework/data_type_transform.cc
index 5a57ec20585c26dbcd4251464718fc819148a7a5..d79f8cacb5f4727defc77380371e57bcea65f068 100644
--- a/paddle/fluid/framework/data_type_transform.cc
+++ b/paddle/fluid/framework/data_type_transform.cc
@@ -37,7 +37,7 @@ struct CastDataType {
   const platform::DeviceContext* ctx_;
 
   template <typename OutType>
-  void operator()() {
+  void apply() {
     auto* in_begin = in_.data<InType>();
     auto* in_end = in_begin + in_.numel();
     auto* out_begin = out_->mutable_data<OutType>(in_.place());
diff --git a/paddle/fluid/framework/details/reduce_and_gather.h b/paddle/fluid/framework/details/reduce_and_gather.h
index e28264eb32756f77ef5baed3dff77ba9f0943160..bd6153c0c736f6e32378eebcbf6c4d7e402c9b42 100644
--- a/paddle/fluid/framework/details/reduce_and_gather.h
+++ b/paddle/fluid/framework/details/reduce_and_gather.h
@@ -31,7 +31,7 @@ struct ReduceLoDTensor {
       : src_tensors_(src), dst_tensor_(*dst) {}
 
   template <typename T>
-  void operator()() const {
+  void apply() const {
     PADDLE_ENFORCE(!src_tensors_.empty());
     auto &t0 = *src_tensors_[0];
     PADDLE_ENFORCE_NE(t0.numel(), 0);
diff --git a/paddle/fluid/framework/selected_rows.cc b/paddle/fluid/framework/selected_rows.cc
index a4319ffabb04f39437b76d97845e021ef9de66d3..8c290bb095d554a973e66a3a19606a06759fd668 100644
--- a/paddle/fluid/framework/selected_rows.cc
+++ b/paddle/fluid/framework/selected_rows.cc
@@ -49,7 +49,7 @@ struct TensorCopyVisitor {
         size_(size) {}
 
   template <typename T>
-  void operator()() const {
+  void apply() const {
     // TODO(Yancey1989): support other place
     platform::CPUPlace cpu;
     memory::Copy(cpu, dst_->mutable_data<T>(cpu) + dst_offset_, cpu,
diff --git a/paddle/fluid/framework/tensor_util.cc b/paddle/fluid/framework/tensor_util.cc
index ab693004cfb038fd92afd9c60e0fcb4e16b9f8a9..05c4a17a01c6fabe48f3fe18544c13153feb0673 100644
--- a/paddle/fluid/framework/tensor_util.cc
+++ b/paddle/fluid/framework/tensor_util.cc
@@ -149,7 +149,7 @@ struct AnyDTypeVisitor {
       : predicate_(predicate), tensor_(tensor), ctx_(ctx), out_(out) {}
 
   template <typename T>
-  void operator()() const {
+  void apply() const {
     auto t = EigenVector<T>::Flatten(tensor_);
     auto o = EigenScalar<bool>::From(*out_);
     // return any of predicate_(t) is true.
@@ -302,7 +302,7 @@ struct DeserializedDataFunctor {
       : buf_(buf), tensor_(tensor), place_(place) {}
 
   template <typename T>
-  void operator()() {
+  void apply() {
     *buf_ = tensor_->mutable_data<T>(place_);
   }
 
diff --git a/paddle/fluid/operators/beam_search_decode_op.cc b/paddle/fluid/operators/beam_search_decode_op.cc
index 10d678111f5325e495b24286e6ecf651230393fe..b6cb935814e25b31d4104f9ce24fe952680cb491 100644
--- a/paddle/fluid/operators/beam_search_decode_op.cc
+++ b/paddle/fluid/operators/beam_search_decode_op.cc
@@ -74,7 +74,7 @@ struct BeamSearchDecodeFunctor {
   }
 
   template <typename T>
-  void operator()() const;
+  void apply() const;
 
   bool tensor_on_gpu_;
   size_t beam_size_;
@@ -88,7 +88,7 @@ struct BeamSearchDecodeFunctor {
 };
 
 template <typename T>
-void BeamSearchDecodeFunctor::operator()() const {
+void BeamSearchDecodeFunctor::apply() const {
   BeamSearchDecoder<T> beam_search_decoder(beam_size_, end_id_);
   // Check if the tensor is on GPU. If so, use the CPU copy instead
   if (tensor_on_gpu_) {
@@ -101,7 +101,7 @@ void BeamSearchDecodeFunctor::operator()() const {
 }
 
 template <>
-void BeamSearchDecodeFunctor::operator()<bool>() const {
+void BeamSearchDecodeFunctor::apply<bool>() const {
   PADDLE_THROW("beam search decode op does not support bool!");
 }
 
diff --git a/paddle/fluid/operators/cast_op.h b/paddle/fluid/operators/cast_op.h
index 6220e57f5941d89cbf0aea268b85ad55af6132cc..8fa0416049f8fa128d7ab61f8350b41960f07263 100644
--- a/paddle/fluid/operators/cast_op.h
+++ b/paddle/fluid/operators/cast_op.h
@@ -37,7 +37,7 @@ struct CastOpFunctor {
       : in_(in), out_(out), ctx_(ctx) {}
 
   template <typename OutT>
-  void operator()() const {
+  void apply() const {
     auto* in_begin = in_->data<InT>();
     auto numel = in_->numel();
     auto* in_end = in_begin + numel;
diff --git a/paddle/fluid/operators/detection/generate_proposals_op.cc b/paddle/fluid/operators/detection/generate_proposals_op.cc
index d29b0153389574de8992b93ac6795e91556af870..fcdcafae7273afa6887ee531dfc37ef833b92d68 100644
--- a/paddle/fluid/operators/detection/generate_proposals_op.cc
+++ b/paddle/fluid/operators/detection/generate_proposals_op.cc
@@ -33,7 +33,7 @@ struct AppendProposalsFunctor {
       : out_(out), offset_(offset), to_add_(to_add) {}
 
   template <typename T>
-  void operator()() const {
+  void apply() const {
     auto *out_data = out_->data<T>();
     auto *to_add_data = to_add_->data<T>();
     memcpy(out_data + offset_, to_add_data, to_add_->numel() * sizeof(T));
diff --git a/paddle/fluid/operators/fill_op.cc b/paddle/fluid/operators/fill_op.cc
index 925dc19061e2196a40411f415eb6e5ad59ab52ff..adc7cb1f9e48ba5fabeb91c5e3ecec016db34a45 100644
--- a/paddle/fluid/operators/fill_op.cc
+++ b/paddle/fluid/operators/fill_op.cc
@@ -25,7 +25,7 @@ struct FillOpVisitor {
       : tensor_(tensor), value_(value) {}
 
   template <typename T>
-  void operator()() const {
+  void apply() const {
     platform::CPUPlace cpu;
     auto *data = tensor_->mutable_data<T>(cpu);
     std::transform(value_.data(), value_.data() + tensor_->numel(), data,
diff --git a/paddle/fluid/operators/math/math_function.cc b/paddle/fluid/operators/math/math_function.cc
index 9a6e646b28fdec78734eb4e7b98c8acf688b2645..5923792902a81521256de300f77955f1ea3d16c6 100644
--- a/paddle/fluid/operators/math/math_function.cc
+++ b/paddle/fluid/operators/math/math_function.cc
@@ -55,7 +55,7 @@ struct TensorSetConstantCPU {
   TensorSetConstantCPU(framework::Tensor* tensor, float value)
       : tensor_(tensor), value_(value) {}
   template <typename T>
-  void operator()() const {
+  void apply() const {
     auto cpu = platform::CPUPlace();
     auto* begin = tensor_->mutable_data<T>(cpu);
     std::fill(begin, begin + tensor_->numel(), static_cast<T>(value_));
diff --git a/paddle/fluid/operators/math/math_function.cu b/paddle/fluid/operators/math/math_function.cu
index 12d1baa8fb544a8b9684e43204c61ba410d1b295..79b7538ad05b0ff348b8264d50b63211b5254e80 100644
--- a/paddle/fluid/operators/math/math_function.cu
+++ b/paddle/fluid/operators/math/math_function.cu
@@ -52,7 +52,7 @@ struct TensorSetConstantGPU {
       : context_(context), tensor_(tensor), value_(value) {}
 
   template <typename T>
-  void operator()() const {
+  void apply() const {
     SetConstant<platform::CUDADeviceContext, T> functor;
     functor(reinterpret_cast<const platform::CUDADeviceContext&>(context_),
             tensor_, static_cast<T>(value_));
diff --git a/paddle/fluid/operators/one_hot_op.cu b/paddle/fluid/operators/one_hot_op.cu
index 625065692c1f32c89d9e566d00051e237ac9a3af..59d8b9b8a8d554eb16826712ff634eed5df2d648 100644
--- a/paddle/fluid/operators/one_hot_op.cu
+++ b/paddle/fluid/operators/one_hot_op.cu
@@ -41,7 +41,7 @@ struct OneHotOpCUDAFunctor {
       : in_(in), out_(out), depth_(depth), ctx_(ctx) {}
 
   template <typename OutT>
-  void operator()() const {
+  void apply() const {
     auto* p_in_data = in_->data<InT>();
     auto numel = in_->numel();
     auto* p_out_data = out_->mutable_data<OutT>(ctx_.GetPlace());
diff --git a/paddle/fluid/operators/one_hot_op.h b/paddle/fluid/operators/one_hot_op.h
index 7e77f25089c4bd0297b0eb5a0ed7555cc0af5a9f..1ebd2676496940ff8f90caaaded5c8227bd7ae78 100644
--- a/paddle/fluid/operators/one_hot_op.h
+++ b/paddle/fluid/operators/one_hot_op.h
@@ -31,7 +31,7 @@ struct OneHotOpFunctor {
       : in_(in), out_(out), depth_(depth), ctx_(ctx) {}
 
   template <typename OutT>
-  void operator()() const {
+  void apply() const {
     auto* p_in_data = in_->data<InT>();
     auto numel = in_->numel();
     auto* p_out_data = out_->mutable_data<OutT>(ctx_.GetPlace());
diff --git a/paddle/fluid/operators/sequence_mask_op.h b/paddle/fluid/operators/sequence_mask_op.h
index 0dd554adfe57e469c2fac17f27adae2db7003a6a..18acb735cecabd1e01f7821c880fd8ed5e52971f 100644
--- a/paddle/fluid/operators/sequence_mask_op.h
+++ b/paddle/fluid/operators/sequence_mask_op.h
@@ -99,7 +99,7 @@ struct SequenceMaskFunctor {
       : ctx_(ctx), x_(x), y_(y), limits_(limits), maxlen_(maxlen) {}
 
   template <typename Ty>
-  void operator()() const {
+  void apply() const {
     auto *y_data = y_->mutable_data<Ty>(ctx_.GetPlace());
     platform::ForRange<DeviceContext> for_range(ctx_, limits_);
     for_range(SequenceMaskForRangeFunctor<Tx, Ty>(x_, y_data, maxlen_));