diff --git a/CMakeLists.txt b/CMakeLists.txt
index e2b4173dfeae1e645993924a10fd1cdedb4ba81b..f506823db3351c8d50f5a4c3c9c093338062af7d 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -7,7 +7,7 @@ option(DEBUGING "enable debug mode" ON)
 option(USE_EXCEPTION "use std exception" OFF)
 option(LOG_PROFILE "log profile" OFF)
 # select the platform to build
-option(CPU "armv7 with neon" OFF)
+option(CPU "armv7 with neon" ON)
 option(GPU_MALI "mali gpu" OFF)
 option(GPU_CL "opencl gpu" ON)
 option(FPGA "fpga" OFF)
@@ -72,8 +72,8 @@ endif()
 
 if (GPU_CL)
     add_definitions(-DPADDLE_MOBILE_CL)
-    include_directories(third_party/opecl/OpenCL-Headers)
-    link_libraries(third_party/opecl/libOpenCL.so)
+    link_libraries(/Users/liuruilong/GitHub/paddle-mobile/third_party/opencl/libOpenCL.so)
+    include_directories(third_party/opencl/OpenCL-Headers)
 else()
     file(GLOB_RECURSE _tmp_list src/framework/cl/*.cpp src/operators/kernel/cl/*.cpp)
     foreach(f ${_tmp_list})
@@ -82,6 +82,7 @@ else()
 
     file(GLOB_RECURSE _tmp_list_h src/framework/cl/*.h)
     foreach(f ${_tmp_list_h})
+        message(STATUS ${f})
         list(REMOVE_ITEM PADDLE_MOBILE_H ${f})
     endforeach()
 endif()
diff --git a/src/framework/cl/cl_engine.cpp b/src/framework/cl/cl_engine.cpp
index b98b8ed8a3e0185a5d21c73a026f382ca94e5efb..045f50c059d8735893062d7e9c0b5c0af0289121 100644
--- a/src/framework/cl/cl_engine.cpp
+++ b/src/framework/cl/cl_engine.cpp
@@ -65,7 +65,7 @@ bool CLEngine::SetClDeviceId() {
       clGetDeviceIDs(platform_, CL_DEVICE_TYPE_GPU, 0, NULL, &numDevices);
 
   if (numDevices > 0) {
-    devices_ = reinterpret_cast<cl_platform_id *>(
+    devices_ = reinterpret_cast<cl_device_id *>(
         malloc(numDevices * sizeof(cl_device_id)));
     status = clGetDeviceIDs(platform_, CL_DEVICE_TYPE_GPU, numDevices, devices_,
                             NULL);
diff --git a/src/framework/cl/cl_engine.h b/src/framework/cl/cl_engine.h
index 7e18f612245b19ba7ee0f279dd21e9e095ffd94e..e77f6b47d6bed307349e73a0d0f852e6f93afa17 100644
--- a/src/framework/cl/cl_engine.h
+++ b/src/framework/cl/cl_engine.h
@@ -38,8 +38,8 @@ class CLEngine {
 
   std::unique_ptr<_cl_command_queue, CLCommQueueDeleter> CreateClCommandQueue() {
     cl_int status;
-    cl_command_queue = clCreateCommandQueue(context_.get(), devices_[0], 0, &status);
-    std::unique_ptr<_cl_command_queue, CLCommQueueDeleter> command_queue_ptr(cl_command_queue);
+    cl_command_queue queue = clCreateCommandQueue(context_.get(), devices_[0], 0, &status);
+    std::unique_ptr<_cl_command_queue, CLCommQueueDeleter> command_queue_ptr(queue);
     return std::move(command_queue_ptr);
   }
 
@@ -67,7 +67,7 @@ class CLEngine {
     return std::move(program_ptr);
   }
 
-  bool CLEngine::BuildProgram(cl_program program) {
+  bool BuildProgram(cl_program program) {
     cl_int status;
     status = clBuildProgram(program, 0, 0, "-cl-fast-relaxed-math", 0, 0);
     CL_CHECK_ERRORS(status);
diff --git a/src/framework/cl/cl_helper.h b/src/framework/cl/cl_helper.h
index 5326793e089b4e5080c1a6acff6ca76ed899afde..7562ab78695376d9766a72eefc2d5482283c49b7 100644
--- a/src/framework/cl/cl_helper.h
+++ b/src/framework/cl/cl_helper.h
@@ -25,6 +25,7 @@ namespace framework {
 
 class CLHelper {
  public:
+  CLHelper() = default;
 
   CLHelper(CLScope *scope): scope_(scope) {
   }
diff --git a/src/framework/cl/cl_scope.h b/src/framework/cl/cl_scope.h
index e7f961d05857a0f74936324b5cdcd8afad61dbea..b461bf65fb6e708951b09bc6f2fb8546b9e30160 100644
--- a/src/framework/cl/cl_scope.h
+++ b/src/framework/cl/cl_scope.h
@@ -57,7 +57,7 @@ class CLScope {
     auto program = CLEngine::Instance()->CreateProgramWith(context_.get(), file_name);
     programs_[file_name] = std::move(program);
 
-    status_ =  clBuildProgram(program, 0, 0, 0, 0, 0);
+    status_ =  clBuildProgram(program.get(), 0, 0, 0, 0, 0);
     CL_CHECK_ERRORS(status_);
     return program.get();
   }
diff --git a/src/framework/cl/cl_tool.h b/src/framework/cl/cl_tool.h
index 3244c796e54076cae40e45adb30ec9a9f774eac9..a2c1eb628b81588bf8bdf85da8b341410ce5a0c0 100644
--- a/src/framework/cl/cl_tool.h
+++ b/src/framework/cl/cl_tool.h
@@ -29,7 +29,6 @@ const char* opencl_error_to_str (cl_int error);
           opencl_error_to_str(ERR), __FILE__, __LINE__                                    \
          );                                                                               \
         }
-#endif
 
 }
 }
diff --git a/src/framework/op_registry.h b/src/framework/op_registry.h
index 8a7beae993be1a9f2a52fb48d4930754aba784e1..a5a2c2d23cea88f9697c5230d582cbb795622983 100644
--- a/src/framework/op_registry.h
+++ b/src/framework/op_registry.h
@@ -15,7 +15,9 @@ limitations under the License. */
 #pragma once
 
 #include <string>
+#include <memory>
 #include <tuple>
+
 #include "common/log.h"
 #include "common/type_define.h"
 #include "framework/op_info.h"
@@ -90,6 +92,7 @@ class OpRegistry {
       const std::string& type, const VariableNameMap& inputs,
       const VariableNameMap& outputs, const AttributeMap attrs,
       std::shared_ptr<paddle_mobile::framework::Scope> scope) {
+
     auto& info = OpInfoMap<Dtype>::Instance()->Get(type);
     auto op = info.Creator()(type, inputs, outputs, attrs, scope);
     return std::shared_ptr<OperatorBase<Dtype>>(op);
diff --git a/src/framework/operator.cpp b/src/framework/operator.cpp
index dd865fb27d4345f16ddca8005463986787d681be..c8d83c95998190e43ea8020518892453a2b7ab0e 100644
--- a/src/framework/operator.cpp
+++ b/src/framework/operator.cpp
@@ -56,7 +56,7 @@ template <typename Dtype>
 void OperatorBase<Dtype>::CheckAllInputOutputSet() const {}
 
 template <typename Dtype>
-void OperatorBase<Dtype>::Run() const {
+void OperatorBase<Dtype>::Run() {
   RunImpl();
 #ifdef PADDLE_MOBILE_DEBUG
   DLOG << "-------------" << type_ << "----------------------------";
@@ -87,6 +87,7 @@ void OperatorBase<Dtype>::Run() const {
 template class OperatorBase<CPU>;
 template class OperatorBase<FPGA>;
 template class OperatorBase<GPU_MALI>;
+template class OperatorBase<GPU_CL>;
 
 }  // namespace framework
 }  // namespace paddle_mobile
diff --git a/src/framework/operator.h b/src/framework/operator.h
index a83a95bbf8ccbff8ba57d11dc57bf6612930579c..d4ff39a4239ff28631c497467760e627a68042f3 100644
--- a/src/framework/operator.h
+++ b/src/framework/operator.h
@@ -31,9 +31,10 @@ limitations under the License. */
 #include "framework/scope.h"
 #include "framework/tensor.h"
 #include "framework/variable.h"
+#ifdef PADDLE_MOBILE_CL
 #include "framework/cl/cl_scope.h"
 #include "framework/cl/cl_helper.h"
-
+#endif
 namespace paddle_mobile {
 namespace framework {
 using std::string;
@@ -61,10 +62,10 @@ class OperatorBase {
                const VariableNameMap &outputs, const AttributeMap &attrs,
                std::shared_ptr<Scope> scope);
   virtual ~OperatorBase() {}
-  void Run() const;
+  void Run();
   std::vector<string> GetOutKeys() const;
   std::vector<string> GetInputKeys() const;
-  virtual void RunImpl() const = 0;
+  virtual void RunImpl() = 0;
 
   virtual void Init() = 0;
   /*
@@ -114,10 +115,9 @@ class OperatorWithKernel : public OperatorBase<Dtype> {
                      const VariableNameMap &outputs, const AttributeMap &attrs,
                      std::shared_ptr<Scope> scope)
       : OperatorBase<Dtype>(type, inputs, outputs, attrs, scope),
-        param_(inputs, outputs, attrs, *scope),
-        kernel_(scope->GetCLScpoe()) {}
+        param_(inputs, outputs, attrs, *scope) {}
 
-  virtual void RunImpl() const { this->kernel_.Compute(this->param_); }
+  virtual void RunImpl() { this->kernel_.Compute(this->param_); }
 
   virtual void InferShape() const = 0;
 
@@ -143,31 +143,33 @@ class OpKernelBase {
  public:
   OpKernelBase() = default;
 
-  OpKernelBase(CLScope *clscope): cl_helper_(clscope) {
-  }
+//  OpKernelBase(CLScope *clscope): cl_helper_(CLHelper(clscope)) {
+//  }
 
   /*
    * @b 所有kernel 需实现 Compute 方法
    * @p para 这个参数为 kernel 运算时所需要用到参数组成的一个结构体,
    *    所有结构体存在与: paddle-mobile/src/operators/op_param.h
    * */
-#ifdef PADDLE_MOBILE_MALI_GPU
+#ifdef PADDLE_McOBILE_MALI_GPU
   OpKernelBase() { acl_op_ = nullptr; }
   void *GetAclOp() const { return acl_op_; }
   void SetAclOp(void *op, void *ob) const {
     reinterpret_cast<OpKernelBase<Dtype, P> *>(ob)->acl_op_ = op;
   }
 #endif
-  virtual void Compute(const P &para) const = 0;
+  virtual void Compute(const P &para) = 0;
   virtual bool Init(P *para) { return true; }
   virtual ~OpKernelBase() = default;
 
+ protected:
+//  CLHelper cl_helper_;
+
  private:
 #ifdef PADDLE_MOBILE_MALI_GPU
   void *acl_op_;
 #endif
 
-  CLHelper cl_helper_;
 
 };
 
diff --git a/src/framework/scope.h b/src/framework/scope.h
index 7e8a9a38bf84135b79cc428e01c20712fa71e6d3..a984b9d5096da267938d09e4fc796b4217eabfec 100644
--- a/src/framework/scope.h
+++ b/src/framework/scope.h
@@ -16,9 +16,9 @@ limitations under the License. */
 
 #include <list>
 
-//#ifdef  PADDLE_MOBILE_OCL
+#ifdef  PADDLE_MOBILE_CL
 #include "framework/cl/cl_scope.h"
-//#endif
+#endif
 
 #include <unordered_map>
 #include "variable.h"
@@ -39,9 +39,9 @@ class Scope {
     }
     kids_.clear();
 
-//#ifdef  PADDLE_MOBILE_OCL
+#ifdef PADDLE_MOBILE_CL
     delete cl_scope_;
-//#endif
+#endif
 
   }
 
@@ -82,9 +82,11 @@ class Scope {
 
   Variable *FindVarLocally(const std::string &name) const;
 
+#ifdef PADDLE_MOBILE_CL
   CLScope *GetCLScpoe() {
     return cl_scope_;
   }
+#endif
 
  private:
   // Call Scope::NewScope for a sub-scope.
@@ -93,9 +95,11 @@ class Scope {
   mutable std::unordered_map<std::string, Variable *> vars_;
   mutable std::list<Scope *> kids_;
   Scope const *parent_{nullptr};
-//#ifdef  PADDLE_MOBILE_CL
+
+#ifdef PADDLE_MOBILE_CL
   CLScope *cl_scope_ = new CLScope();
-//#endif
+#endif
+
 };
 }  // namespace framework
 }  // namespace paddle_mobile
diff --git a/src/io/executor.cpp b/src/io/executor.cpp
index 926917d178569525f8bf878748dc0b8085ac1152..f94ae37e5124785760fda4d49a4ac760e6d86484 100644
--- a/src/io/executor.cpp
+++ b/src/io/executor.cpp
@@ -720,7 +720,8 @@ void Executor<Dtype, P>::Predict_To(int end) {
 #endif
 
 template class Executor<CPU, Precision::FP32>;
-template class Executor<GPU_MALI, Precision::FP32>;
 template class Executor<FPGA, Precision::FP32>;
+template class Executor<GPU_CL, Precision::FP32>;
+template class Executor<GPU_MALI, Precision::FP32>;
 
 }  // namespace paddle_mobile
diff --git a/src/io/loader.cpp b/src/io/loader.cpp
index 48a2b5cfdaa5f53cd9611dd0be1ce3df05988311..1faeb9d3ea2b29756a49c7ed676e431f9b0b6b4e 100644
--- a/src/io/loader.cpp
+++ b/src/io/loader.cpp
@@ -193,5 +193,6 @@ const framework::Program<Dtype, P> Loader<Dtype, P>::LoadCombinedMemory(
 template class Loader<CPU, Precision::FP32>;
 template class Loader<FPGA, Precision::FP32>;
 template class Loader<GPU_MALI, Precision::FP32>;
+template class Loader<GPU_CL, Precision::FP32>;
 
 }  // namespace paddle_mobile
diff --git a/src/operators/bilinear_interp_op.h b/src/operators/bilinear_interp_op.h
index dbbf24eeac7a900d49f49242fddb8e568968dddc..e20c56cb3aab07a17f6c929358c221fb27b85898 100644
--- a/src/operators/bilinear_interp_op.h
+++ b/src/operators/bilinear_interp_op.h
@@ -40,10 +40,6 @@ class BilinearOp : public framework::OperatorWithKernel<
             DeviceType, BilinearInterpParam<DeviceType>,
             operators::BilinearInterpKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, BilinearInterpParam<DeviceType>,
-      operators::BilinearInterpKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 };
 
diff --git a/src/operators/box_coder_op.h b/src/operators/box_coder_op.h
index 5d475c98b6859a33b39e6b36419fa055cde7a1d3..89e0f4a67f7f35012732f159afe19a660a52b5fd 100644
--- a/src/operators/box_coder_op.h
+++ b/src/operators/box_coder_op.h
@@ -39,10 +39,6 @@ class BoxCoderOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, BoxCoderParam<DeviceType>,
                                       operators::BoxCoderKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, BoxCoderParam<DeviceType>,
-      operators::BoxCoderKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/concat_op.h b/src/operators/concat_op.h
index a169c17dc468dd06ed344a0c7a6ef3cb2c977a27..e6794b0e88c78b968ca59a71a358aa5b1f9a86f8 100644
--- a/src/operators/concat_op.h
+++ b/src/operators/concat_op.h
@@ -34,10 +34,6 @@ class ConcatOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, ConcatParam<DeviceType>,
                                       operators::ConcatKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, ConcatParam<DeviceType>,
-      operators::ConcatKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/conv_op.h b/src/operators/conv_op.h
index 267abfeb614dc8e19a2cf0cf43e7c5f232a62072..299c12cdeaed25c114be5c7d3dc4fe74044f9298 100644
--- a/src/operators/conv_op.h
+++ b/src/operators/conv_op.h
@@ -34,10 +34,6 @@ class ConvOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, ConvParam<DeviceType>,
                                       operators::ConvKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, ConvParam<DeviceType>,
-      operators::ConvKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  private:
diff --git a/src/operators/crf_op.h b/src/operators/crf_op.h
index 9c966c9077273282bbcb4f25674e8df401956967..d813c6f1d9d3b8e9e0ae091280a08eedefebb500 100644
--- a/src/operators/crf_op.h
+++ b/src/operators/crf_op.h
@@ -37,10 +37,6 @@ class CrfOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, CrfParam<DeviceType>,
                                       operators::CrfKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, CrfParam<DeviceType>,
-      operators::CrfKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 };
 
diff --git a/src/operators/depthwise_conv_op.h b/src/operators/depthwise_conv_op.h
index 40e87a9b1bf9d2b5102a56ff59821b9d122563c5..b2d20a6d1a9a3aa550fb8093b21f7cfd815dd0a4 100644
--- a/src/operators/depthwise_conv_op.h
+++ b/src/operators/depthwise_conv_op.h
@@ -36,10 +36,6 @@ class DepthwiseConvOp : public framework::OperatorWithKernel<
             DeviceType, ConvParam<DeviceType>,
             operators::DepthwiseConvKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, ConvParam<DeviceType>,
-      operators::DepthwiseConvKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  private:
diff --git a/src/operators/dropout_op.h b/src/operators/dropout_op.h
index 7523fd5fa8f21dfce20bce963be4b3bc323948e9..e1974f343f4ccbe63994afd2b6a4437e8aa095ff 100644
--- a/src/operators/dropout_op.h
+++ b/src/operators/dropout_op.h
@@ -38,10 +38,6 @@ class DropoutOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, DropoutParam<DeviceType>,
                                       operators::DropoutKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  // using framework::OperatorWithKernel<DeviceType, DropoutParam<DeviceType>,
-  //                                    operators::DropoutKernel<DeviceType,
-  //                                    T>>;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/elementwise_add_op.h b/src/operators/elementwise_add_op.h
index 14bcd5264d136007e2eb2ffe917697570b32e40b..dedac33210038e72f8dbc41e9aea5b60534f16cc 100644
--- a/src/operators/elementwise_add_op.h
+++ b/src/operators/elementwise_add_op.h
@@ -37,10 +37,6 @@ class ElementwiseAddOp : public framework::OperatorWithKernel<
             DeviceType, ElementwiseAddParam<DeviceType>,
             operators::ElementwiseAddKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, ElementwiseAddParam<DeviceType>,
-      operators::ElementwiseAddKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/feed_op.h b/src/operators/feed_op.h
index cccd4f52ebdc368e4f68eaf9dc3f25ee3693fdd2..2fae189d3c65f705d67e990660c9766b0bd0438c 100644
--- a/src/operators/feed_op.h
+++ b/src/operators/feed_op.h
@@ -74,7 +74,7 @@ class FeedOp : public framework::OperatorBase<DeviceType> {
 
 #else
   void Init() {}
-  void RunImpl() const {
+  void RunImpl() {
     param_.Out()->ShareDataWith(*param_.InputX());
     param_.Out()->set_lod(param_.InputX()->lod());
   }
diff --git a/src/operators/fetch_op.h b/src/operators/fetch_op.h
index 1efe0832b1fc4b2ce240ed838e2f4554c29dccd9..959beff632291e401026bec4917745f95af4cbf2 100644
--- a/src/operators/fetch_op.h
+++ b/src/operators/fetch_op.h
@@ -31,7 +31,7 @@ class FetchOp : public framework::OperatorBase<DeviceType> {
       : framework::OperatorBase<DeviceType>(type, inputs, outputs, attrs,
                                             scope),
         param_(inputs, outputs, attrs, *scope) {}
-  void RunImpl() const { param_.Out()->ShareDataWith(*param_.InputX()); }
+  void RunImpl() { param_.Out()->ShareDataWith(*param_.InputX()); }
 
   void Init() {}
 
diff --git a/src/operators/flatten_op.h b/src/operators/flatten_op.h
index 4c1f6ff8a0f2b3212750f3be4d1a6aa2bad790ee..98416413791696480c3945994666d0d2fa689b2a 100644
--- a/src/operators/flatten_op.h
+++ b/src/operators/flatten_op.h
@@ -53,10 +53,6 @@ class FlattenOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, FlattenParam<DeviceType>,
                                       operators::FlattenKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, FlattenParam<DeviceType>,
-      operators::FlattenKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 };
 
diff --git a/src/operators/fusion_conv_add.h b/src/operators/fusion_conv_add.h
index ba1ca997662ce67fdcd8f39d2a12e2f535c5b1a7..24bcfb52ca8d190807a3cf845bc997d11096d7fe 100644
--- a/src/operators/fusion_conv_add.h
+++ b/src/operators/fusion_conv_add.h
@@ -56,10 +56,6 @@ class FusionConvAddOp : public framework::OperatorWithKernel<
                                       FusionConvAddParam<DeviceType>,
                                       operators::ConvAddKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, FusionConvAddParam<DeviceType>,
-      operators::ConvAddKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/fusion_conv_add_add_prelu_op.h b/src/operators/fusion_conv_add_add_prelu_op.h
index d91b4d28d728efb4ecf817294f37e67ac19cfe72..fba981fce93a95c4cee4dac2ed9bd122f3cd8f71 100644
--- a/src/operators/fusion_conv_add_add_prelu_op.h
+++ b/src/operators/fusion_conv_add_add_prelu_op.h
@@ -67,10 +67,6 @@ class FusionConvAddAddPReluOp
             DeviceType, FusionConvAddAddPReluParam<DeviceType>,
             operators::ConvAddAddPReluKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, FusionConvAddAddPReluParam<DeviceType>,
-      operators::ConvAddAddPReluKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/fusion_conv_add_bn_relu_op.h b/src/operators/fusion_conv_add_bn_relu_op.h
index 4dee4416622e8dee8ca495026843c7506d084617..c6544debd35ac214c247f94cd05520ff49adfa9e 100644
--- a/src/operators/fusion_conv_add_bn_relu_op.h
+++ b/src/operators/fusion_conv_add_bn_relu_op.h
@@ -66,10 +66,6 @@ class FusionConvAddBNReluOp
             DeviceType, FusionConvAddBNReluParam<DeviceType>,
             operators::ConvAddBNReluKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, FusionConvAddBNReluParam<DeviceType>,
-      operators::ConvAddBNReluKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/fusion_conv_add_prelu_op.h b/src/operators/fusion_conv_add_prelu_op.h
index 4c968be68230fe6252e72655f47b2a347f720526..dda9019c8c876b32ab91c795bc53507ce4ad6b4d 100644
--- a/src/operators/fusion_conv_add_prelu_op.h
+++ b/src/operators/fusion_conv_add_prelu_op.h
@@ -63,9 +63,6 @@ class FusionConvAddPReluOp
             operators::ConvAddPReluKernel<DeviceType, T>>(type, inputs, outputs,
                                                           attrs, scope) {}
 
-  using framework::OperatorWithKernel<
-      DeviceType, FusionConvAddPReluParam<DeviceType>,
-      operators::ConvAddPReluKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/fusion_conv_add_relu_op.h b/src/operators/fusion_conv_add_relu_op.h
index 926f309403d37fa8ec1f15f7cb955c1c13842405..85d7d216379c5d36662d3b06b8c01b61dc5c6f07 100644
--- a/src/operators/fusion_conv_add_relu_op.h
+++ b/src/operators/fusion_conv_add_relu_op.h
@@ -57,9 +57,6 @@ class FusionConvAddReluOp : public framework::OperatorWithKernel<
             operators::ConvAddReluKernel<DeviceType, T>>(type, inputs, outputs,
                                                          attrs, scope) {}
 
-  using framework::OperatorWithKernel<
-      DeviceType, FusionConvAddReluParam<DeviceType>,
-      operators::ConvAddReluKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/fusion_conv_bn_add_relu_op.h b/src/operators/fusion_conv_bn_add_relu_op.h
index 62f3ccf37dfbff9720f39fb96b099f6d7eb5ddcc..ad91fb6c35055b8d2a6d795b664774b2176330fc 100644
--- a/src/operators/fusion_conv_bn_add_relu_op.h
+++ b/src/operators/fusion_conv_bn_add_relu_op.h
@@ -71,10 +71,6 @@ class FusionConvBNAddReluOp
             DeviceType, FusionConvBNAddReluParam<DeviceType>,
             operators::ConvBNAddReluKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, FusionConvBNAddReluParam<DeviceType>,
-      operators::ConvBNAddReluKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/fusion_conv_bn_relu_op.h b/src/operators/fusion_conv_bn_relu_op.h
index 2b5ff4ea9d3e77ad9449b3968667ecc4558c2147..139df4bb862f197af0429b3d91fe65dd32bf69fc 100644
--- a/src/operators/fusion_conv_bn_relu_op.h
+++ b/src/operators/fusion_conv_bn_relu_op.h
@@ -63,10 +63,6 @@ class FusionConvBNReluOp : public framework::OperatorWithKernel<
             DeviceType, FusionConvBNReluParam<DeviceType>,
             operators::ConvBNReluKernel<DeviceType, T>>(type, inputs, outputs,
                                                         attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, FusionConvBNReluParam<DeviceType>,
-      operators::ConvBNReluKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/fusion_dwconv_bn_relu_op.h b/src/operators/fusion_dwconv_bn_relu_op.h
index dd1f85688f576106a46cd3070ab2034ec8f55881..5782e6a0fe8233a3b097ce423ff8317330c5d28e 100644
--- a/src/operators/fusion_dwconv_bn_relu_op.h
+++ b/src/operators/fusion_dwconv_bn_relu_op.h
@@ -65,9 +65,6 @@ class FusionDWConvBNReluOp
             operators::DWConvBNReluKernel<DeviceType, T>>(type, inputs, outputs,
                                                           attrs, scope) {}
 
-  using framework::OperatorWithKernel<
-      DeviceType, FusionDWConvBNReluParam<DeviceType>,
-      operators::DWConvBNReluKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/fusion_fc_op.h b/src/operators/fusion_fc_op.h
index 076a95d745e8d44a417dd95fb75844a67b11e653..58c8f08a7a8fea62513242af4fa19594eb70f496 100644
--- a/src/operators/fusion_fc_op.h
+++ b/src/operators/fusion_fc_op.h
@@ -56,10 +56,6 @@ class FusionFcOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, FusionFcParam<DeviceType>,
                                       operators::FusionFcKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, FusionFcParam<DeviceType>,
-      operators::FusionFcKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/fusion_fc_relu_op.h b/src/operators/fusion_fc_relu_op.h
index fa7d4045fc10d6e240d93e129aa736be793f7bbf..dfedfb1b5c733f4984a4c102660c62b89cf59a7f 100644
--- a/src/operators/fusion_fc_relu_op.h
+++ b/src/operators/fusion_fc_relu_op.h
@@ -56,9 +56,6 @@ class FusionFcReluOp : public framework::OperatorWithKernel<
             operators::FusionFcReluKernel<DeviceType, T>>(type, inputs, outputs,
                                                           attrs, scope) {}
 
-  using framework::OperatorWithKernel<
-      DeviceType, FusionFcReluParam<DeviceType>,
-      operators::FusionFcReluKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/gru_op.h b/src/operators/gru_op.h
index d348b6c52431f93673f1b772f8c8a9462878cfd5..c26e32b8d7b90d03d8019bb7fa0c42bf1be40c96 100644
--- a/src/operators/gru_op.h
+++ b/src/operators/gru_op.h
@@ -37,10 +37,6 @@ class GruOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, GruParam<DeviceType>,
                                       operators::GruKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, GruParam<DeviceType>,
-      operators::GruKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 };
 
diff --git a/src/operators/im2sequence_op.h b/src/operators/im2sequence_op.h
index edb87d0012e5514cb5541f94a965965f3dc02825..6412b5d5a089989c8cd547960b91f1a025d551e7 100644
--- a/src/operators/im2sequence_op.h
+++ b/src/operators/im2sequence_op.h
@@ -39,9 +39,6 @@ class Im2SequenceOp : public framework::OperatorWithKernel<
             operators::Im2SequenceKernel<DeviceType, T>>(type, inputs, outputs,
                                                          attrs, scope) {}
 
-  // using framework::OperatorWithKernel<
-  //    DeviceType, Im2SequenceParam<DeviceType>,
-  //    operators::Im2SequenceKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  private:
diff --git a/src/operators/kernel/arm/batchnorm_kernel.cpp b/src/operators/kernel/arm/batchnorm_kernel.cpp
index c420727f425092240994ee834117225c72abeec2..fe7233088eaaf303af7c85f1702cf0381af33887 100644
--- a/src/operators/kernel/arm/batchnorm_kernel.cpp
+++ b/src/operators/kernel/arm/batchnorm_kernel.cpp
@@ -27,7 +27,7 @@ bool BatchNormKernel<CPU, float>::Init(BatchNormParam<CPU> *param) {
 
 template <>
 void BatchNormKernel<CPU, float>::Compute(
-    const BatchNormParam<CPU> &param) const {
+    const BatchNormParam<CPU> &param) {
   BatchnormCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/bilinear_interp_kernel.cpp b/src/operators/kernel/arm/bilinear_interp_kernel.cpp
index ea5ff627d7ea2e0fa5434f9f7fc9f5ec44ce60a7..3c00ee7c48fad08f2c7b2ddfee3ad4a074d0718c 100644
--- a/src/operators/kernel/arm/bilinear_interp_kernel.cpp
+++ b/src/operators/kernel/arm/bilinear_interp_kernel.cpp
@@ -28,7 +28,7 @@ bool BilinearInterpKernel<CPU, float>::Init(BilinearInterpParam<CPU> *param) {
 
 template <>
 void BilinearInterpKernel<CPU, float>::Compute(
-    const BilinearInterpParam<CPU> &param) const {
+    const BilinearInterpParam<CPU> &param) {
   BilinearInterpCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/box_coder_kernel.cpp b/src/operators/kernel/arm/box_coder_kernel.cpp
index b769d4fbbaa7570ee741476f960d9e5b60c61917..2071b0ee996b6a96509c8d0f8556ddc5ce12daac 100644
--- a/src/operators/kernel/arm/box_coder_kernel.cpp
+++ b/src/operators/kernel/arm/box_coder_kernel.cpp
@@ -27,7 +27,7 @@ bool BoxCoderKernel<CPU, float>::Init(BoxCoderParam<CPU> *param) {
 
 template <>
 void BoxCoderKernel<CPU, float>::Compute(
-    const BoxCoderParam<CPU> &param) const {
+    const BoxCoderParam<CPU> &param) {
   BoxCoderCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/concat_kernel.cpp b/src/operators/kernel/arm/concat_kernel.cpp
index 04c590e6b432fbf88cd136eac942485adf9a9003..8cdf6cb01afa85239bfd0d48bbce02790ba5250d 100644
--- a/src/operators/kernel/arm/concat_kernel.cpp
+++ b/src/operators/kernel/arm/concat_kernel.cpp
@@ -26,7 +26,7 @@ bool ConcatKernel<CPU, float>::Init(ConcatParam<CPU> *param) {
 }
 
 template <>
-void ConcatKernel<CPU, float>::Compute(const ConcatParam<CPU> &param) const {
+void ConcatKernel<CPU, float>::Compute(const ConcatParam<CPU> &param) {
   ConcatCompute<float>(param);
   param.Out()->set_lod(param.Inputs()[0]->lod());
 }
diff --git a/src/operators/kernel/arm/conv_add_add_prelu_kernel.cpp b/src/operators/kernel/arm/conv_add_add_prelu_kernel.cpp
index 74b88f5d4f5e24b1401803c8c48d99319f412d1b..2f6f5f3ac719b3fd32aac54ce36eb534f7d99dd7 100644
--- a/src/operators/kernel/arm/conv_add_add_prelu_kernel.cpp
+++ b/src/operators/kernel/arm/conv_add_add_prelu_kernel.cpp
@@ -28,7 +28,7 @@ bool ConvAddAddPReluKernel<CPU, float>::Init(
 
 template <>
 void ConvAddAddPReluKernel<CPU, float>::Compute(
-    const FusionConvAddAddPReluParam<CPU> &param) const {
+    const FusionConvAddAddPReluParam<CPU> &param) {
   ConvAddAddPReluCompute<float>(param);
 }
 template class ConvAddAddPReluKernel<CPU, float>;
diff --git a/src/operators/kernel/arm/conv_add_bn_relu_kernel.cpp b/src/operators/kernel/arm/conv_add_bn_relu_kernel.cpp
index ca53ebea8e4577fdc52fad066691d4351eaf12f9..eb55920621db34d191a9536f287ec50747e1ce3c 100644
--- a/src/operators/kernel/arm/conv_add_bn_relu_kernel.cpp
+++ b/src/operators/kernel/arm/conv_add_bn_relu_kernel.cpp
@@ -55,7 +55,7 @@ bool ConvAddBNReluKernel<CPU, float>::Init(
 
 template <>
 void ConvAddBNReluKernel<CPU, float>::Compute(
-    const FusionConvAddBNReluParam<CPU> &param) const {
+    const FusionConvAddBNReluParam<CPU> &param) {
   ConvAddBNReluCompute<float>(param);
 }
 template class ConvAddBNReluKernel<CPU, float>;
diff --git a/src/operators/kernel/arm/conv_add_kernel.cpp b/src/operators/kernel/arm/conv_add_kernel.cpp
index 1af1c3db1159cd4fed007ebf153ba15b804eee75..8672678882ef15be8538312e0526c57d5956f10f 100644
--- a/src/operators/kernel/arm/conv_add_kernel.cpp
+++ b/src/operators/kernel/arm/conv_add_kernel.cpp
@@ -26,7 +26,7 @@ bool ConvAddKernel<CPU, float>::Init(FusionConvAddParam<CPU> *param) {
 
 template <>
 void ConvAddKernel<CPU, float>::Compute(
-    const FusionConvAddParam<CPU> &param) const {
+    const FusionConvAddParam<CPU> &param) {
   ConvAddCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/conv_add_prelu_kernel.cpp b/src/operators/kernel/arm/conv_add_prelu_kernel.cpp
index 5930cfdcfc0f983c9f07754113dc37d5122d19f0..f04a9a7d746f2d970196945707bd05409c5fa340 100644
--- a/src/operators/kernel/arm/conv_add_prelu_kernel.cpp
+++ b/src/operators/kernel/arm/conv_add_prelu_kernel.cpp
@@ -27,7 +27,7 @@ bool ConvAddPReluKernel<CPU, float>::Init(FusionConvAddPReluParam<CPU> *param) {
 
 template <>
 void ConvAddPReluKernel<CPU, float>::Compute(
-    const FusionConvAddPReluParam<CPU> &param) const {
+    const FusionConvAddPReluParam<CPU> &param) {
   ConvAddPReluCompute<float>(param);
 }
 template class ConvAddPReluKernel<CPU, float>;
diff --git a/src/operators/kernel/arm/conv_add_relu_kernel.cpp b/src/operators/kernel/arm/conv_add_relu_kernel.cpp
index f50e1e3900bb5fce35a29100d6c2cb6004b4af74..211d6d8487bfd4afc71d74e5ecbff149ad34e466 100644
--- a/src/operators/kernel/arm/conv_add_relu_kernel.cpp
+++ b/src/operators/kernel/arm/conv_add_relu_kernel.cpp
@@ -27,7 +27,7 @@ bool ConvAddReluKernel<CPU, float>::Init(FusionConvAddReluParam<CPU> *param) {
 
 template <>
 void ConvAddReluKernel<CPU, float>::Compute(
-    const FusionConvAddReluParam<CPU> &param) const {
+    const FusionConvAddReluParam<CPU> &param) {
   ConvAddReluCompute<float>(param);
 }
 template class ConvAddReluKernel<CPU, float>;
diff --git a/src/operators/kernel/arm/conv_bn_add_relu_kernel.cpp b/src/operators/kernel/arm/conv_bn_add_relu_kernel.cpp
index 785b13dde2ec1196792d17b253bb0d904da799f5..a0f21dd6126ed81cf5e96f99bd0f8ed5211f96a4 100644
--- a/src/operators/kernel/arm/conv_bn_add_relu_kernel.cpp
+++ b/src/operators/kernel/arm/conv_bn_add_relu_kernel.cpp
@@ -55,7 +55,7 @@ bool ConvBNAddReluKernel<CPU, float>::Init(
 
 template <>
 void ConvBNAddReluKernel<CPU, float>::Compute(
-    const FusionConvBNAddReluParam<CPU> &param) const {
+    const FusionConvBNAddReluParam<CPU> &param) {
   ConvBNAddReluCompute<float>(param);
 }
 template class ConvBNAddReluKernel<CPU, float>;
diff --git a/src/operators/kernel/arm/conv_bn_relu_kernel.cpp b/src/operators/kernel/arm/conv_bn_relu_kernel.cpp
index 6b9ea0428fa496980a234c7c895ef9cbf1245b51..d8acb8d2083b732da026a9bff19c2d7732568597 100644
--- a/src/operators/kernel/arm/conv_bn_relu_kernel.cpp
+++ b/src/operators/kernel/arm/conv_bn_relu_kernel.cpp
@@ -57,7 +57,7 @@ bool ConvBNReluKernel<CPU, float>::Init(FusionConvBNReluParam<CPU> *param) {
 
 template <>
 void ConvBNReluKernel<CPU, float>::Compute(
-    const FusionConvBNReluParam<CPU> &param) const {
+    const FusionConvBNReluParam<CPU> &param) {
   ConvBNReluCompute<float>(param);
 }
 template class ConvBNReluKernel<CPU, float>;
diff --git a/src/operators/kernel/arm/conv_kernel.cpp b/src/operators/kernel/arm/conv_kernel.cpp
index 4e9d3a34f231485685bc8f7b087382cb99a3b036..0d67bdc656f2ba9ad674c18c9cefbd7d9cd711df 100644
--- a/src/operators/kernel/arm/conv_kernel.cpp
+++ b/src/operators/kernel/arm/conv_kernel.cpp
@@ -26,7 +26,7 @@ bool ConvKernel<CPU, float>::Init(ConvParam<CPU> *param) {
 }
 
 template <>
-void ConvKernel<CPU, float>::Compute(const ConvParam<CPU> &param) const {
+void ConvKernel<CPU, float>::Compute(const ConvParam<CPU> &param) {
   ConvCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/conv_transpose_kernel.cpp b/src/operators/kernel/arm/conv_transpose_kernel.cpp
index d695e6144b40d945857d547f7c208f1192481e8f..852502965d6d043329b824a12868242e83f2f83d 100644
--- a/src/operators/kernel/arm/conv_transpose_kernel.cpp
+++ b/src/operators/kernel/arm/conv_transpose_kernel.cpp
@@ -27,7 +27,7 @@ bool ConvTransposeKernel<CPU, float>::Init(ConvTransposeParam<CPU> *param) {
 
 template <>
 void ConvTransposeKernel<CPU, float>::Compute(
-    const ConvTransposeParam<CPU> &param) const {
+    const ConvTransposeParam<CPU> &param) {
   ConvTransposeCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/crf_kernel.cpp b/src/operators/kernel/arm/crf_kernel.cpp
index 89769c50a6fc05b28192ebf584ba3cb12f19ac2c..d30c28b3576e2a8a8a108ae6c86edc2f4310b83f 100644
--- a/src/operators/kernel/arm/crf_kernel.cpp
+++ b/src/operators/kernel/arm/crf_kernel.cpp
@@ -27,7 +27,7 @@ bool CrfKernel<CPU, float>::Init(CrfParam<CPU> *param) {
 }
 
 template <>
-void CrfKernel<CPU, float>::Compute(const CrfParam<CPU> &param) const {
+void CrfKernel<CPU, float>::Compute(const CrfParam<CPU> &param) {
   CrfCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/depthwise_conv_kernel.cpp b/src/operators/kernel/arm/depthwise_conv_kernel.cpp
index fd5e068afb6f7f2a069a7d8fccc459d4c2a6828d..d3ada81b774d23d4b91270d2ae11da86dbe0a209 100644
--- a/src/operators/kernel/arm/depthwise_conv_kernel.cpp
+++ b/src/operators/kernel/arm/depthwise_conv_kernel.cpp
@@ -27,7 +27,7 @@ bool DepthwiseConvKernel<CPU, float>::Init(ConvParam<CPU> *param) {
 
 template <>
 void DepthwiseConvKernel<CPU, float>::Compute(
-    const ConvParam<CPU> &param) const {
+    const ConvParam<CPU> &param) {
   DepthwiseConvCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/dropout_kernel.cpp b/src/operators/kernel/arm/dropout_kernel.cpp
index 4578ac6607d87c316853f6201f02f8204bc41de1..ea59519edc1ca336e5271af2cabb41e47298fc29 100644
--- a/src/operators/kernel/arm/dropout_kernel.cpp
+++ b/src/operators/kernel/arm/dropout_kernel.cpp
@@ -35,7 +35,7 @@ struct DropoutFunctor {
 };
 
 template <>
-void DropoutKernel<CPU, float>::Compute(const DropoutParam<CPU> &param) const {
+void DropoutKernel<CPU, float>::Compute(const DropoutParam<CPU> &param) {
   const auto *input_x = param.InputX();
   auto *input_x_ptr = input_x->data<float>();
   auto *out = param.Out();
diff --git a/src/operators/kernel/arm/dwconv_bn_relu_kernel.cpp b/src/operators/kernel/arm/dwconv_bn_relu_kernel.cpp
index b85701bb936b2ccc0323e4d534424abb726a69be..f92d9a273467bf15d9d7fad43237af5385d3d54e 100644
--- a/src/operators/kernel/arm/dwconv_bn_relu_kernel.cpp
+++ b/src/operators/kernel/arm/dwconv_bn_relu_kernel.cpp
@@ -54,7 +54,7 @@ bool DWConvBNReluKernel<CPU, float>::Init(FusionDWConvBNReluParam<CPU> *param) {
 
 template <>
 void DWConvBNReluKernel<CPU, float>::Compute(
-    const FusionDWConvBNReluParam<CPU> &param) const {
+    const FusionDWConvBNReluParam<CPU> &param) {
   DWConvBNReluCompute<float>(param);
 }
 template class DWConvBNReluKernel<CPU, float>;
diff --git a/src/operators/kernel/arm/elementwise_add_kernel.cpp b/src/operators/kernel/arm/elementwise_add_kernel.cpp
index 9c6f4a3316385b803a8fdb833490f1fe9e7f41ac..043d27e72f16ab4b38f31d6cff60bd2f4e89a649 100644
--- a/src/operators/kernel/arm/elementwise_add_kernel.cpp
+++ b/src/operators/kernel/arm/elementwise_add_kernel.cpp
@@ -27,7 +27,7 @@ bool ElementwiseAddKernel<CPU, float>::Init(ElementwiseAddParam<CPU> *param) {
 
 template <>
 void ElementwiseAddKernel<CPU, float>::Compute(
-    const ElementwiseAddParam<CPU> &param) const {
+    const ElementwiseAddParam<CPU> &param) {
   ElementwiseAddCompute<float>(param);
   param.Out()->set_lod(param.InputX()->lod());
 }
diff --git a/src/operators/kernel/arm/flatten_kernel.cpp b/src/operators/kernel/arm/flatten_kernel.cpp
index 6866b740aa945852050e7fca4991489f48435150..4b470658f1f9a037a18d9deea78f5cea85af773b 100644
--- a/src/operators/kernel/arm/flatten_kernel.cpp
+++ b/src/operators/kernel/arm/flatten_kernel.cpp
@@ -27,7 +27,7 @@ bool FlattenKernel<CPU, float>::Init(FlattenParam<CPU> *param) {
 }
 
 template <>
-void FlattenKernel<CPU, float>::Compute(const FlattenParam<CPU> &param) const {
+void FlattenKernel<CPU, float>::Compute(const FlattenParam<CPU> &param) {
   FlattenCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/fusion_fc_kernel.cpp b/src/operators/kernel/arm/fusion_fc_kernel.cpp
index d9d112e7a762705efe041c74eea9ddb7d5162918..0f13a1e84b8e80bc09e9bbf2c6554108567f00f0 100644
--- a/src/operators/kernel/arm/fusion_fc_kernel.cpp
+++ b/src/operators/kernel/arm/fusion_fc_kernel.cpp
@@ -27,7 +27,7 @@ bool FusionFcKernel<CPU, float>::Init(FusionFcParam<CPU> *param) {
 
 template <>
 void FusionFcKernel<CPU, float>::Compute(
-    const FusionFcParam<CPU> &param) const {
+    const FusionFcParam<CPU> &param) {
   FusionFcCompute<float>(param);
   param.Out()->set_lod(param.InputX()->lod());
 }
diff --git a/src/operators/kernel/arm/gru_kernel.cpp b/src/operators/kernel/arm/gru_kernel.cpp
index 168471185e07a9c1814c708238996a82c1ee0891..a4e89ff42a3d70c0a9a3d1bd7316e18d015a0926 100644
--- a/src/operators/kernel/arm/gru_kernel.cpp
+++ b/src/operators/kernel/arm/gru_kernel.cpp
@@ -26,7 +26,7 @@ bool GruKernel<CPU, float>::Init(GruParam<CPU> *param) {
 }
 
 template <>
-void GruKernel<CPU, float>::Compute(const GruParam<CPU> &param) const {
+void GruKernel<CPU, float>::Compute(const GruParam<CPU> &param) {
   GruCompute<float>(param);
   param.OutHidden()->set_lod(param.InputInput()->lod());
   //  DLOG << "________________" << param.OutHidden()->dims();
diff --git a/src/operators/kernel/arm/im2sequence_kernel.cpp b/src/operators/kernel/arm/im2sequence_kernel.cpp
index 8295fd94a31db2ad1c10d32a8c639b067e422f45..3d078f13312a05bf1f0c8b5c26e52775667fd685 100644
--- a/src/operators/kernel/arm/im2sequence_kernel.cpp
+++ b/src/operators/kernel/arm/im2sequence_kernel.cpp
@@ -33,7 +33,7 @@ inline int Im2SeqOutputSize(int input_size, int filter_size, int padding_0,
 
 template <>
 void Im2SequenceKernel<CPU, float>::Compute(
-    const Im2SequenceParam<CPU> &param) const {
+    const Im2SequenceParam<CPU> &param) {
   const Tensor *in_x = param.Input();
   Tensor *out = param.Output();
   out->mutable_data<float>();
diff --git a/src/operators/kernel/arm/lookup_kernel.cpp b/src/operators/kernel/arm/lookup_kernel.cpp
index 584c497c701bd0598e0a151774fe60b7c7fee718..0e6df6ab6bf19f67b0c5f5a873d4a47215167e45 100644
--- a/src/operators/kernel/arm/lookup_kernel.cpp
+++ b/src/operators/kernel/arm/lookup_kernel.cpp
@@ -25,7 +25,7 @@ bool LookupKernel<CPU, float>::Init(LookupParam<CPU> *param) {
 }
 
 template <>
-void LookupKernel<CPU, float>::Compute(const LookupParam<CPU> &param) const {
+void LookupKernel<CPU, float>::Compute(const LookupParam<CPU> &param) {
   LookupCompute<float>(param);
   param.Out()->set_lod(param.InputIds()->lod());
 }
diff --git a/src/operators/kernel/arm/lrn_kernel.cpp b/src/operators/kernel/arm/lrn_kernel.cpp
index 3ec1bdd9a0e2ebbce555eef944fe56750505430f..bf049053fc5b9157f24c50233742eea3c0ca2de1 100644
--- a/src/operators/kernel/arm/lrn_kernel.cpp
+++ b/src/operators/kernel/arm/lrn_kernel.cpp
@@ -26,7 +26,7 @@ bool LrnKernel<CPU, float>::Init(LrnParam<CPU> *param) {
 }
 
 template <>
-void LrnKernel<CPU, float>::Compute(const LrnParam<CPU> &param) const {
+void LrnKernel<CPU, float>::Compute(const LrnParam<CPU> &param) {
   LrnCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/mul_kernel.cpp b/src/operators/kernel/arm/mul_kernel.cpp
index aa3ee7077eb7db440c8493eae5b95f03a42196a4..94200c5c241b20179c056a8b2569cddfc3601d22 100644
--- a/src/operators/kernel/arm/mul_kernel.cpp
+++ b/src/operators/kernel/arm/mul_kernel.cpp
@@ -26,7 +26,7 @@ bool MulKernel<CPU, float>::Init(MulParam<CPU> *param) {
 }
 
 template <>
-void MulKernel<CPU, float>::Compute(const MulParam<CPU> &param) const {
+void MulKernel<CPU, float>::Compute(const MulParam<CPU> &param) {
   MulCompute<float>(param);
   param.Out()->set_lod(param.InputX()->lod());
 }
diff --git a/src/operators/kernel/arm/multiclass_nms_kernel.cpp b/src/operators/kernel/arm/multiclass_nms_kernel.cpp
index 938f81cf485eb64f408c0fb274eeec673349e306..61638da0051c7b27b695752c445f0fd6b20114b5 100644
--- a/src/operators/kernel/arm/multiclass_nms_kernel.cpp
+++ b/src/operators/kernel/arm/multiclass_nms_kernel.cpp
@@ -27,7 +27,7 @@ bool MultiClassNMSKernel<CPU, float>::Init(MultiClassNMSParam<CPU> *param) {
 
 template <>
 void MultiClassNMSKernel<CPU, float>::Compute(
-    const MultiClassNMSParam<CPU> &param) const {
+    const MultiClassNMSParam<CPU> &param) {
   MultiClassNMSCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/pool_kernel.cpp b/src/operators/kernel/arm/pool_kernel.cpp
index 60d6f1401876b957649d08889218b88cf1fe5eef..58d6359efa48b0db215269a631e7e4cb57c429d9 100644
--- a/src/operators/kernel/arm/pool_kernel.cpp
+++ b/src/operators/kernel/arm/pool_kernel.cpp
@@ -25,7 +25,7 @@ bool PoolKernel<CPU, float>::Init(PoolParam<CPU> *param) {
 }
 
 template <>
-void PoolKernel<CPU, float>::Compute(const PoolParam<CPU> &param) const {
+void PoolKernel<CPU, float>::Compute(const PoolParam<CPU> &param) {
   PoolCompute<float>(param);
 }
 }  // namespace operators
diff --git a/src/operators/kernel/arm/prelu_kernel.cpp b/src/operators/kernel/arm/prelu_kernel.cpp
index e1ec927fb13d1f4a2e600d46f65f2806448059d9..591bd644165f1a271a879073b27429d1780cbfb5 100644
--- a/src/operators/kernel/arm/prelu_kernel.cpp
+++ b/src/operators/kernel/arm/prelu_kernel.cpp
@@ -35,7 +35,7 @@ struct PReluFunctor {
  * @b 特化到具体平台的实现, param 从 op 层传入
  * */
 template <>
-void PReluKernel<CPU, float>::Compute(const PReluParam<CPU> &param) const {
+void PReluKernel<CPU, float>::Compute(const PReluParam<CPU> &param) {
   auto *x = param.InputX();
   auto *alpha = param.InputAlpha();
   auto *out = param.Out();
diff --git a/src/operators/kernel/arm/prior_box_kernel.cpp b/src/operators/kernel/arm/prior_box_kernel.cpp
index 71011fa2112f36d573b5bdc55f1b5bf92318c448..43b0d4f16ac64f6a4fb8d08e8d9715d2f66ef683 100644
--- a/src/operators/kernel/arm/prior_box_kernel.cpp
+++ b/src/operators/kernel/arm/prior_box_kernel.cpp
@@ -27,7 +27,7 @@ bool PriorBoxKernel<CPU, float>::Init(PriorBoxParam<CPU> *param) {
 
 template <>
 void PriorBoxKernel<CPU, float>::Compute(
-    const PriorBoxParam<CPU> &param) const {
+    const PriorBoxParam<CPU> &param) {
   PriorBoxCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/relu_kernel.cpp b/src/operators/kernel/arm/relu_kernel.cpp
index 6e04e6013aa8dd5c50dcc22a720b470b08ecd648..8ee103484eb753913e5554b64d6dac523066322a 100644
--- a/src/operators/kernel/arm/relu_kernel.cpp
+++ b/src/operators/kernel/arm/relu_kernel.cpp
@@ -26,7 +26,7 @@ bool ReluKernel<CPU, float>::Init(ReluParam<CPU> *param) {
 }
 
 template <>
-void ReluKernel<CPU, float>::Compute(const ReluParam<CPU> &param) const {
+void ReluKernel<CPU, float>::Compute(const ReluParam<CPU> &param) {
   ReluCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/reshape_kernel.cpp b/src/operators/kernel/arm/reshape_kernel.cpp
index 235288ae13e2c557e6f7310727f5d8e6e83cedf6..800808f9c23cd07d17f8207b9b51e96d3feb34f3 100644
--- a/src/operators/kernel/arm/reshape_kernel.cpp
+++ b/src/operators/kernel/arm/reshape_kernel.cpp
@@ -26,7 +26,7 @@ bool ReshapeKernel<CPU, float>::Init(ReshapeParam<CPU> *param) {
 }
 
 template <>
-void ReshapeKernel<CPU, float>::Compute(const ReshapeParam<CPU> &param) const {
+void ReshapeKernel<CPU, float>::Compute(const ReshapeParam<CPU> &param) {
   ReshapeCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/resize_kernel.cpp b/src/operators/kernel/arm/resize_kernel.cpp
index 5c0c186554a31454447b1df47a1b7573fd948fb9..b53b7545e33c929fe0b55bccd68e7b955db0d676 100644
--- a/src/operators/kernel/arm/resize_kernel.cpp
+++ b/src/operators/kernel/arm/resize_kernel.cpp
@@ -108,7 +108,7 @@ void ResizeTensor(const Tensor* src, Tensor* dst) {
 }
 
 template <>
-void ResizeKernel<CPU, float>::Compute(const ResizeParam<CPU>& param) const {
+void ResizeKernel<CPU, float>::Compute(const ResizeParam<CPU>& param) {
   const auto* input_x = param.InputX();
   const auto& input_x_dims = input_x->dims();
   auto* out = param.Out();
diff --git a/src/operators/kernel/arm/scale_kernel.cpp b/src/operators/kernel/arm/scale_kernel.cpp
index 299132ea00f40838249022c45d994e7d88547eaa..bded56275f80741c552d4978bb238d6f0d6339db 100644
--- a/src/operators/kernel/arm/scale_kernel.cpp
+++ b/src/operators/kernel/arm/scale_kernel.cpp
@@ -23,7 +23,7 @@ namespace operators {
  * @b 特化到具体平台的实现, param 从 op 层传入
  * */
 template <>
-void ScaleKernel<CPU, float>::Compute(const ScaleParam<CPU> &param) const {
+void ScaleKernel<CPU, float>::Compute(const ScaleParam<CPU> &param) {
   const auto *input_x = param.InputX();
   auto *input_x_ptr = input_x->data<float>();
   auto *out = param.Out();
diff --git a/src/operators/kernel/arm/shape_kernel.cpp b/src/operators/kernel/arm/shape_kernel.cpp
index 69fd4021fe3110a7cea02a67443939048c1dddab..9702a7cf2dbfeb66b3959859f5dfada510a8c9f0 100644
--- a/src/operators/kernel/arm/shape_kernel.cpp
+++ b/src/operators/kernel/arm/shape_kernel.cpp
@@ -27,7 +27,7 @@ bool ShapeKernel<CPU, float>::Init(ShapeParam<CPU> *param) {
 }
 
 template <>
-void ShapeKernel<CPU, float>::Compute(const ShapeParam<CPU> &param) const {
+void ShapeKernel<CPU, float>::Compute(const ShapeParam<CPU> &param) {
   ShapeCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/sigmoid_kernel.cpp b/src/operators/kernel/arm/sigmoid_kernel.cpp
index 7912fd8762b693cd40c632d6b152406ed4b0c568..3d6e14ffea80169172431229e34309cde331d588 100644
--- a/src/operators/kernel/arm/sigmoid_kernel.cpp
+++ b/src/operators/kernel/arm/sigmoid_kernel.cpp
@@ -32,7 +32,7 @@ bool SigmoidKernel<CPU, float>::Init(SigmoidParam<CPU> *param) {
 }
 
 template <>
-void SigmoidKernel<CPU, float>::Compute(const SigmoidParam<CPU> &param) const {
+void SigmoidKernel<CPU, float>::Compute(const SigmoidParam<CPU> &param) {
   SigmoidCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/softmax_kernel.cpp b/src/operators/kernel/arm/softmax_kernel.cpp
index f86a10601aa3a67300736f2f4c751c05bf41a781..d5a1009fd79d57d8815d313ed61bbc5d7bf32bbe 100644
--- a/src/operators/kernel/arm/softmax_kernel.cpp
+++ b/src/operators/kernel/arm/softmax_kernel.cpp
@@ -26,7 +26,7 @@ bool SoftmaxKernel<CPU, float>::Init(SoftmaxParam<CPU> *param) {
 }
 
 template <>
-void SoftmaxKernel<CPU, float>::Compute(const SoftmaxParam<CPU> &param) const {
+void SoftmaxKernel<CPU, float>::Compute(const SoftmaxParam<CPU> &param) {
   SoftmaxCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/split_kernel.cpp b/src/operators/kernel/arm/split_kernel.cpp
index 292b5bda99a524615df4a8552e5617fd4470d8a0..1099edc14fdf19e272ef3194e948706a25fee35d 100644
--- a/src/operators/kernel/arm/split_kernel.cpp
+++ b/src/operators/kernel/arm/split_kernel.cpp
@@ -27,7 +27,7 @@ bool SplitKernel<CPU, float>::Init(SplitParam<CPU> *param) {
 }
 
 template <>
-void SplitKernel<CPU, float>::Compute(const SplitParam<CPU> &param) const {
+void SplitKernel<CPU, float>::Compute(const SplitParam<CPU> &param) {
   SplitCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/arm/transpose_kernel.cpp b/src/operators/kernel/arm/transpose_kernel.cpp
index bb7a881bdc1d2706a25a77833ca38695ede2fec7..43355abf0a6e10f3aca857cc94ac72d1827b47f2 100644
--- a/src/operators/kernel/arm/transpose_kernel.cpp
+++ b/src/operators/kernel/arm/transpose_kernel.cpp
@@ -26,7 +26,7 @@ bool TransposeKernel<CPU, float>::Init(TransposeParam<CPU> *param) {
 
 template <>
 void TransposeKernel<CPU, float>::Compute(
-    const TransposeParam<CPU> &param) const {
+    const TransposeParam<CPU> &param) {
   TransposeCompute<float>(param);
 }
 
diff --git a/src/operators/kernel/batchnorm_kernel.h b/src/operators/kernel/batchnorm_kernel.h
index beac7399583d074956fa4564fdd9312b2d7985f0..8f9f61c46b8edf57713c64e207bc4c0c4b70e90c 100644
--- a/src/operators/kernel/batchnorm_kernel.h
+++ b/src/operators/kernel/batchnorm_kernel.h
@@ -28,7 +28,7 @@ template <typename DeviceType, typename T>
 class BatchNormKernel
     : public framework::OpKernelBase<DeviceType, BatchNormParam<DeviceType>> {
  public:
-  void Compute(const BatchNormParam<DeviceType> &param) const;
+  void Compute(const BatchNormParam<DeviceType> &param);
   bool Init(BatchNormParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/bilinear_interp_kernel.h b/src/operators/kernel/bilinear_interp_kernel.h
index ac3dfcb16190315f72dc60da54c4f944874e4458..9a68fe65a562a8567dab2e5977506e083f7889a2 100644
--- a/src/operators/kernel/bilinear_interp_kernel.h
+++ b/src/operators/kernel/bilinear_interp_kernel.h
@@ -29,7 +29,7 @@ class BilinearInterpKernel
     : public framework::OpKernelBase<DeviceType,
                                      BilinearInterpParam<DeviceType>> {
  public:
-  void Compute(const BilinearInterpParam<DeviceType>& param) const;
+  void Compute(const BilinearInterpParam<DeviceType>& param);
   bool Init(BilinearInterpParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/box_coder_kernel.h b/src/operators/kernel/box_coder_kernel.h
index 58144a87349ed3a6504e0074903594be3aa6fe8f..eadb21b3d5ecb95ef82cfef2ac8c3245e925ec7c 100644
--- a/src/operators/kernel/box_coder_kernel.h
+++ b/src/operators/kernel/box_coder_kernel.h
@@ -29,7 +29,7 @@ template <typename DeviceType, typename T>
 class BoxCoderKernel
     : public framework::OpKernelBase<DeviceType, BoxCoderParam<DeviceType>> {
  public:
-  void Compute(const BoxCoderParam<DeviceType>& param) const;
+  void Compute(const BoxCoderParam<DeviceType>& param);
   bool Init(BoxCoderParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/cl/conv_kernel.cpp b/src/operators/kernel/cl/conv_kernel.cpp
index 6e5a4326261f1ad17d0c803ee3a4cfb4156eccf0..bd0dffe8af17b7e21fa83c1b78a45230ae355a58 100644
--- a/src/operators/kernel/cl/conv_kernel.cpp
+++ b/src/operators/kernel/cl/conv_kernel.cpp
@@ -1,39 +1,41 @@
-/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License. */
-
-#ifdef CONV_OP
-
-#include "operators/kernel/conv_kernel.h"
-#include "operators/kernel/central-arm-func/conv_arm_func.h"
-
-namespace paddle_mobile {
-namespace operators {
-
-template <>
-bool ConvKernel<GPU_CL, float>::Init(ConvParam<GPU_CL> *param) {
-
-  return true;
-}
-
-template <>
-void ConvKernel<GPU_CL, float>::Compute(const ConvParam<GPU_CL> &param) const {
-
-}
-
-template class ConvKernel<GPU_CL, float>;
-
-}  // namespace operators
-}  // namespace paddle_mobile
-
-#endif
+///* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+//
+//Licensed under the Apache License, Version 2.0 (the "License");
+//you may not use this file except in compliance with the License.
+//You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+//Unless required by applicable law or agreed to in writing, software
+//distributed under the License is distributed on an "AS IS" BASIS,
+//WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//See the License for the specific language governing permissions and
+//limitations under the License. */
+//
+//#ifdef CONV_OP
+//
+//#include "operators/kernel/conv_kernel.h"
+//#include "operators/kernel/central-arm-func/conv_arm_func.h"
+//
+//namespace paddle_mobile {
+//namespace operators {
+//
+//template <>
+//bool ConvKernel<GPU_CL, float>::Init(ConvParam<GPU_CL> *param) {
+//  this->cl_helper_.AddKernel("conv_3x3", "conv_kernel.cl");
+//  return true;
+//}
+//
+//template <>
+//void ConvKernel<GPU_CL, float>::Compute(const ConvParam<GPU_CL> &param) {
+//  auto kernel = this->cl_helper_.KernelAt(0);
+//
+//
+//}
+//
+//template class ConvKernel<GPU_CL, float>;
+//
+//}  // namespace operators
+//}  // namespace paddle_mobile
+//
+//#endif
diff --git a/src/operators/kernel/concat_kernel.h b/src/operators/kernel/concat_kernel.h
index 61100bf5f0e9de43bfb6295a0719f1be0954d128..ac9ebca4d5ab30307303b8720677e67470634b44 100644
--- a/src/operators/kernel/concat_kernel.h
+++ b/src/operators/kernel/concat_kernel.h
@@ -27,7 +27,7 @@ template <typename DeviceType, typename T>
 class ConcatKernel
     : public framework::OpKernelBase<DeviceType, ConcatParam<DeviceType>> {
  public:
-  void Compute(const ConcatParam<DeviceType> &param) const;
+  void Compute(const ConcatParam<DeviceType> &param);
   bool Init(ConcatParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/conv_add_add_prelu_kernel.h b/src/operators/kernel/conv_add_add_prelu_kernel.h
index 5715cd46d5a6c7e80ab5ff77ba83c7973e1db811..fadaf7564ceeb7a52215dc335135016be02bc1ab 100644
--- a/src/operators/kernel/conv_add_add_prelu_kernel.h
+++ b/src/operators/kernel/conv_add_add_prelu_kernel.h
@@ -35,7 +35,7 @@ template <typename DeviceType, typename T>
 class ConvAddAddPReluKernel
     : public OpKernelBase<DeviceType, FusionConvAddAddPReluParam<DeviceType>> {
  public:
-  void Compute(const FusionConvAddAddPReluParam<DeviceType> &param) const;
+  void Compute(const FusionConvAddAddPReluParam<DeviceType> &param);
   bool Init(FusionConvAddAddPReluParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/conv_add_bn_kernel.h b/src/operators/kernel/conv_add_bn_kernel.h
index ee73215c4688c3e604de69cda55b05e63844c0b8..7a921ecc7d0f4498cae80fbb9cea1b13e4c94101 100644
--- a/src/operators/kernel/conv_add_bn_kernel.h
+++ b/src/operators/kernel/conv_add_bn_kernel.h
@@ -35,7 +35,7 @@ template <typename DeviceType, typename T>
 class ConvAddBNKernel
     : public OpKernelBase<DeviceType, FusionConvAddBNParam<DeviceType>> {
  public:
-  void Compute(const FusionConvAddBNParam<DeviceType> &param) const;
+  void Compute(const FusionConvAddBNParam<DeviceType> &param);
   bool Init(FusionConvAddBNParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/conv_add_bn_relu_kernel.h b/src/operators/kernel/conv_add_bn_relu_kernel.h
index 9faaaedcf8d6f825f818ebf5121dc7685185d5d8..3f088528fc901987873038c7e1dd779dcc2019e7 100644
--- a/src/operators/kernel/conv_add_bn_relu_kernel.h
+++ b/src/operators/kernel/conv_add_bn_relu_kernel.h
@@ -35,7 +35,7 @@ template <typename DeviceType, typename T>
 class ConvAddBNReluKernel
     : public OpKernelBase<DeviceType, FusionConvAddBNReluParam<DeviceType>> {
  public:
-  void Compute(const FusionConvAddBNReluParam<DeviceType> &param) const;
+  void Compute(const FusionConvAddBNReluParam<DeviceType> &param);
   bool Init(FusionConvAddBNReluParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/conv_add_kernel.h b/src/operators/kernel/conv_add_kernel.h
index 360cbb6775168885e9c1a25db1f9ffb9e552324b..4e9ff0853f1d502ebb4dc4ef3641d0a879f32b60 100644
--- a/src/operators/kernel/conv_add_kernel.h
+++ b/src/operators/kernel/conv_add_kernel.h
@@ -40,7 +40,7 @@ template <typename DeviceType, typename T>
 class ConvAddKernel
     : public OpKernelBase<DeviceType, FusionConvAddParam<DeviceType>> {
  public:
-  void Compute(const FusionConvAddParam<DeviceType> &param) const;
+  void Compute(const FusionConvAddParam<DeviceType> &param);
   bool Init(FusionConvAddParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/conv_add_prelu_kernel.h b/src/operators/kernel/conv_add_prelu_kernel.h
index a109f84cf09b4d0e2469a1885b902c0f70acc6c8..631982789b09c57d0d21186d0a30df7368d2955f 100644
--- a/src/operators/kernel/conv_add_prelu_kernel.h
+++ b/src/operators/kernel/conv_add_prelu_kernel.h
@@ -35,7 +35,7 @@ template <typename DeviceType, typename T>
 class ConvAddPReluKernel
     : public OpKernelBase<DeviceType, FusionConvAddPReluParam<DeviceType>> {
  public:
-  void Compute(const FusionConvAddPReluParam<DeviceType> &param) const;
+  void Compute(const FusionConvAddPReluParam<DeviceType> &param);
   bool Init(FusionConvAddPReluParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/conv_add_relu_kernel.h b/src/operators/kernel/conv_add_relu_kernel.h
index f33b1dc312e1d94be0c23cff55e9e6789a556bc7..e001926b361da96ec3ff76e120bc3d1ad13714fa 100644
--- a/src/operators/kernel/conv_add_relu_kernel.h
+++ b/src/operators/kernel/conv_add_relu_kernel.h
@@ -35,7 +35,7 @@ template <typename DeviceType, typename T>
 class ConvAddReluKernel
     : public OpKernelBase<DeviceType, FusionConvAddReluParam<DeviceType>> {
  public:
-  void Compute(const FusionConvAddReluParam<DeviceType> &param) const;
+  void Compute(const FusionConvAddReluParam<DeviceType> &param);
   bool Init(FusionConvAddReluParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/conv_bn_add_relu_kernel.h b/src/operators/kernel/conv_bn_add_relu_kernel.h
index 820e5f8bcbf58676e8374e575044b10fe4676efa..dcd8fecf07fbb4ea75b382f5315e24e64e26e939 100644
--- a/src/operators/kernel/conv_bn_add_relu_kernel.h
+++ b/src/operators/kernel/conv_bn_add_relu_kernel.h
@@ -35,7 +35,7 @@ template <typename DeviceType, typename T>
 class ConvBNAddReluKernel
     : public OpKernelBase<DeviceType, FusionConvBNAddReluParam<DeviceType>> {
  public:
-  void Compute(const FusionConvBNAddReluParam<DeviceType> &param) const;
+  void Compute(const FusionConvBNAddReluParam<DeviceType> &param);
   bool Init(FusionConvBNAddReluParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/conv_bn_kernel.h b/src/operators/kernel/conv_bn_kernel.h
index f740ca836481c1331ea2e889865b3078d48644a6..e669f3bdd85dbd89e3a48d417dcd0cd6b9706062 100644
--- a/src/operators/kernel/conv_bn_kernel.h
+++ b/src/operators/kernel/conv_bn_kernel.h
@@ -35,7 +35,7 @@ template <typename DeviceType, typename T>
 class ConvBNKernel
     : public OpKernelBase<DeviceType, FusionConvBNParam<DeviceType>> {
  public:
-  void Compute(const FusionConvBNParam<DeviceType> &param) const;
+  void Compute(const FusionConvBNParam<DeviceType> &param);
   bool Init(FusionConvBNParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/conv_bn_relu_kernel.h b/src/operators/kernel/conv_bn_relu_kernel.h
index 225976aa5db31096ef691ecefa8b63d4ae3dc277..91b3413116ae22a8e212cf149c4e0c2a8924664a 100644
--- a/src/operators/kernel/conv_bn_relu_kernel.h
+++ b/src/operators/kernel/conv_bn_relu_kernel.h
@@ -35,7 +35,7 @@ template <typename DeviceType, typename T>
 class ConvBNReluKernel
     : public OpKernelBase<DeviceType, FusionConvBNReluParam<DeviceType>> {
  public:
-  void Compute(const FusionConvBNReluParam<DeviceType> &param) const;
+  void Compute(const FusionConvBNReluParam<DeviceType> &param);
   bool Init(FusionConvBNReluParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/conv_kernel.h b/src/operators/kernel/conv_kernel.h
index 93474adaa97743d1850b53df114ae08f144aebca..cac498c36bd5debef0ff996cdf017355a2371a18 100644
--- a/src/operators/kernel/conv_kernel.h
+++ b/src/operators/kernel/conv_kernel.h
@@ -31,7 +31,7 @@ using framework::OpKernelBase;
 template <typename DeviceType, typename T>
 class ConvKernel : public OpKernelBase<DeviceType, ConvParam<DeviceType>> {
  public:
-  void Compute(const ConvParam<DeviceType> &param) const;
+  void Compute(const ConvParam<DeviceType> &param);
   bool Init(ConvParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/conv_transpose_kernel.h b/src/operators/kernel/conv_transpose_kernel.h
index 9cbd7c8c3bafde8b4f4939e86ceabdd94dbd3bc8..8e321e2e2efdefa0d409d0918aec22d2b699e386 100644
--- a/src/operators/kernel/conv_transpose_kernel.h
+++ b/src/operators/kernel/conv_transpose_kernel.h
@@ -28,7 +28,7 @@ template <typename DeviceType, typename T>
 class ConvTransposeKernel
     : public OpKernelBase<DeviceType, ConvTransposeParam<DeviceType>> {
  public:
-  void Compute(const ConvTransposeParam<DeviceType> &param) const;
+  void Compute(const ConvTransposeParam<DeviceType> &param);
 
   bool Init(ConvTransposeParam<DeviceType> *param);
 };
diff --git a/src/operators/kernel/crf_kernel.h b/src/operators/kernel/crf_kernel.h
index 71c07cf0384d482522de3a6652c6d24a22af656a..1436aafc0603d4c7ba9ecae911f10bd8f297852a 100644
--- a/src/operators/kernel/crf_kernel.h
+++ b/src/operators/kernel/crf_kernel.h
@@ -28,7 +28,7 @@ template <typename DeviceType, typename T>
 class CrfKernel
     : public framework::OpKernelBase<DeviceType, CrfParam<DeviceType>> {
  public:
-  void Compute(const CrfParam<DeviceType>& param) const;
+  void Compute(const CrfParam<DeviceType>& param);
   bool Init(CrfParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/depthwise_conv_kernel.h b/src/operators/kernel/depthwise_conv_kernel.h
index 605b81cd6ed4ccd54b1803cf7a603b8f4576982d..3ee5bf86e97baa3970239e32b7fd5fc341e09f92 100644
--- a/src/operators/kernel/depthwise_conv_kernel.h
+++ b/src/operators/kernel/depthwise_conv_kernel.h
@@ -31,7 +31,7 @@ template <typename DeviceType, typename T>
 class DepthwiseConvKernel
     : public OpKernelBase<DeviceType, ConvParam<DeviceType>> {
  public:
-  void Compute(const ConvParam<DeviceType> &param) const;
+  void Compute(const ConvParam<DeviceType> &param);
   bool Init(ConvParam<DeviceType> *param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/dropout_kernel.h b/src/operators/kernel/dropout_kernel.h
index b7535095d4fef11ee628aea96a074abcc3562f7f..2f59d01b6723eea274b1ed059ae08863a4937961 100644
--- a/src/operators/kernel/dropout_kernel.h
+++ b/src/operators/kernel/dropout_kernel.h
@@ -26,7 +26,7 @@ template <typename DeviceType, typename T>
 class DropoutKernel
     : public framework::OpKernelBase<DeviceType, DropoutParam<DeviceType>> {
  public:
-  void Compute(const DropoutParam<DeviceType>& param) const;
+  void Compute(const DropoutParam<DeviceType>& param);
   bool Init(DropoutParam<DeviceType>* para);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/dwconv_bn_relu_kernel.h b/src/operators/kernel/dwconv_bn_relu_kernel.h
index 594c594cb00f8f4ddd8a511f3c992c4efbfcdfc6..f2e4c0afbd0aaafff5339816764f9e30592f122c 100644
--- a/src/operators/kernel/dwconv_bn_relu_kernel.h
+++ b/src/operators/kernel/dwconv_bn_relu_kernel.h
@@ -35,7 +35,7 @@ template <typename DeviceType, typename T>
 class DWConvBNReluKernel
     : public OpKernelBase<DeviceType, FusionDWConvBNReluParam<DeviceType>> {
  public:
-  void Compute(const FusionDWConvBNReluParam<DeviceType> &param) const;
+  void Compute(const FusionDWConvBNReluParam<DeviceType> &param);
   bool Init(FusionDWConvBNReluParam<DeviceType> *param);
 };
 
diff --git a/src/operators/kernel/elementwise_add_kernel.h b/src/operators/kernel/elementwise_add_kernel.h
index 67182af2e20e23c40effab6b87eefde1e0ab629d..8fa07e519ec0b78baffabd08fb7e524f8259c9eb 100644
--- a/src/operators/kernel/elementwise_add_kernel.h
+++ b/src/operators/kernel/elementwise_add_kernel.h
@@ -30,7 +30,7 @@ class ElementwiseAddKernel
     : public framework::OpKernelBase<DeviceType,
                                      ElementwiseAddParam<DeviceType>> {
  public:
-  void Compute(const ElementwiseAddParam<DeviceType> &param) const;
+  void Compute(const ElementwiseAddParam<DeviceType> &param);
   bool Init(ElementwiseAddParam<DeviceType> *param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/elementwise_add_relu_kernel.h b/src/operators/kernel/elementwise_add_relu_kernel.h
index 5eda5a0c56c228ad54c888b6faa82ce9417f2dc1..d18c4e27fa3345b1818d0e6149fc8fb83195f644 100644
--- a/src/operators/kernel/elementwise_add_relu_kernel.h
+++ b/src/operators/kernel/elementwise_add_relu_kernel.h
@@ -29,7 +29,7 @@ class ElementwiseAddReluKernel
     : public framework::OpKernelBase<DeviceType,
                                      ElementwiseAddReluParam<DeviceType>> {
  public:
-  void Compute(const ElementwiseAddReluParam<DeviceType> &param) const;
+  void Compute(const ElementwiseAddReluParam<DeviceType> &param);
   bool Init(ElementwiseAddReluParam<DeviceType> *param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/fc_relu_kernel.h b/src/operators/kernel/fc_relu_kernel.h
index 6e9446da37df4ba83db85d416aa87f216816c4a5..6735a50bee86e25d9f8d091b6218a472f3838aec 100644
--- a/src/operators/kernel/fc_relu_kernel.h
+++ b/src/operators/kernel/fc_relu_kernel.h
@@ -28,7 +28,7 @@ class FusionFcReluKernel
     : public framework::OpKernelBase<DeviceType,
                                      FusionFcReluParam<DeviceType>> {
  public:
-  void Compute(const FusionFcReluParam<DeviceType>& param) const;
+  void Compute(const FusionFcReluParam<DeviceType>& param);
   bool Init(FusionFcReluParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/flatten_kernel.h b/src/operators/kernel/flatten_kernel.h
index 80d66ccf87c21532c8b4590d992f5bccbe4f00dc..4846725bcb6522389d29e137980b9d53e63f9f32 100644
--- a/src/operators/kernel/flatten_kernel.h
+++ b/src/operators/kernel/flatten_kernel.h
@@ -28,7 +28,7 @@ template <typename DeviceType, typename T>
 class FlattenKernel
     : public framework::OpKernelBase<DeviceType, FlattenParam<DeviceType>> {
  public:
-  void Compute(const FlattenParam<DeviceType>& param) const;
+  void Compute(const FlattenParam<DeviceType>& param);
   bool Init(FlattenParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/fpga/concat_kernel.cpp b/src/operators/kernel/fpga/concat_kernel.cpp
index 86da2833ed6e1443707054896127e87c0ca297b9..5e01dc55907e5cc2fe9dade00f6ca7b727bdcd9d 100644
--- a/src/operators/kernel/fpga/concat_kernel.cpp
+++ b/src/operators/kernel/fpga/concat_kernel.cpp
@@ -57,7 +57,7 @@ bool ConcatKernel<FPGA, float>::Init(ConcatParam<FPGA> *param) {
 }
 
 template <>
-void ConcatKernel<FPGA, float>::Compute(const ConcatParam<FPGA> &param) const {
+void ConcatKernel<FPGA, float>::Compute(const ConcatParam<FPGA> &param) {
   ComputeFPGAConcat(param.FpgaArgs());
 }
 template class ConcatKernel<FPGA, float>;
diff --git a/src/operators/kernel/fpga/conv_add_bn_kernel.cpp b/src/operators/kernel/fpga/conv_add_bn_kernel.cpp
index 671df76967b4537d111695cdbe091b9c7de2c5a2..c88812fa832a779ae0e5addf7d603bd376780a71 100644
--- a/src/operators/kernel/fpga/conv_add_bn_kernel.cpp
+++ b/src/operators/kernel/fpga/conv_add_bn_kernel.cpp
@@ -77,7 +77,7 @@ bool ConvAddBNKernel<FPGA, float>::Init(FusionConvAddBNParam<FPGA> *param) {
 
 template <>
 void ConvAddBNKernel<FPGA, float>::Compute(
-    const FusionConvAddBNParam<FPGA> &param) const {
+    const FusionConvAddBNParam<FPGA> &param) {
   fpga::ComputeFpgaConv(param.FpgaArgs());
 }
 
diff --git a/src/operators/kernel/fpga/conv_add_bn_relu_kernel.cpp b/src/operators/kernel/fpga/conv_add_bn_relu_kernel.cpp
index d435692db6b40568afc599733c2adb6b05b00ffa..21392600a2c04e5db31d4618a6631218b204a34f 100644
--- a/src/operators/kernel/fpga/conv_add_bn_relu_kernel.cpp
+++ b/src/operators/kernel/fpga/conv_add_bn_relu_kernel.cpp
@@ -74,7 +74,7 @@ bool ConvAddBNReluKernel<FPGA, float>::Init(
 
 template <>
 void ConvAddBNReluKernel<FPGA, float>::Compute(
-    const FusionConvAddBNReluParam<FPGA> &param) const {
+    const FusionConvAddBNReluParam<FPGA> &param) {
   fpga::ComputeFpgaConv(param.FpgaArgs());
 }
 
diff --git a/src/operators/kernel/fpga/conv_add_relu_kernel.cpp b/src/operators/kernel/fpga/conv_add_relu_kernel.cpp
index 32d90b36e4c14a60219a3779da03100651aa2f13..e099fe659c8556eb0ccbb7ed1b50715e3c2e09aa 100644
--- a/src/operators/kernel/fpga/conv_add_relu_kernel.cpp
+++ b/src/operators/kernel/fpga/conv_add_relu_kernel.cpp
@@ -56,7 +56,7 @@ bool ConvAddReluKernel<FPGA, float>::Init(FusionConvAddReluParam<FPGA> *param) {
 
 template <>
 void ConvAddReluKernel<FPGA, float>::Compute(
-    const FusionConvAddReluParam<FPGA> &param) const {
+    const FusionConvAddReluParam<FPGA> &param) {
   fpga::ComputeFpgaConv(param.FpgaArgs());
 }
 
diff --git a/src/operators/kernel/fpga/conv_bn_kernel.cpp b/src/operators/kernel/fpga/conv_bn_kernel.cpp
index 4263c9c40491366813d3c9a5bf7dbc8ae976d39e..4d5486fd6c45c0bcbdf047743f849ac3edfeb6b5 100644
--- a/src/operators/kernel/fpga/conv_bn_kernel.cpp
+++ b/src/operators/kernel/fpga/conv_bn_kernel.cpp
@@ -68,7 +68,7 @@ bool ConvBNKernel<FPGA, float>::Init(FusionConvBNParam<FPGA> *param) {
 
 template <>
 void ConvBNKernel<FPGA, float>::Compute(
-    const FusionConvBNParam<FPGA> &param) const {
+    const FusionConvBNParam<FPGA> &param) {
   fpga::ComputeFpgaConv(param.FpgaArgs());
 }
 
diff --git a/src/operators/kernel/fpga/conv_bn_relu_kernel.cpp b/src/operators/kernel/fpga/conv_bn_relu_kernel.cpp
index 3d6e0faa5fe3d4ef3514bbe1679298b11d96727c..d72f7f57c3599bff8d4a07ecb64bc94548991240 100644
--- a/src/operators/kernel/fpga/conv_bn_relu_kernel.cpp
+++ b/src/operators/kernel/fpga/conv_bn_relu_kernel.cpp
@@ -68,7 +68,7 @@ bool ConvBNReluKernel<FPGA, float>::Init(FusionConvBNReluParam<FPGA> *param) {
 
 template <>
 void ConvBNReluKernel<FPGA, float>::Compute(
-    const FusionConvBNReluParam<FPGA> &param) const {
+    const FusionConvBNReluParam<FPGA> &param) {
   fpga::ComputeFpgaConv(param.FpgaArgs());
 }
 
diff --git a/src/operators/kernel/fpga/dropout_kernel.cpp b/src/operators/kernel/fpga/dropout_kernel.cpp
index b0981c4254060996a16f4ae5beabb7c22edd6d34..24dbbaa2f412529b3ebb88749f03c03f391fece4 100644
--- a/src/operators/kernel/fpga/dropout_kernel.cpp
+++ b/src/operators/kernel/fpga/dropout_kernel.cpp
@@ -27,7 +27,7 @@ bool DropoutKernel<FPGA, float>::Init(DropoutParam<FPGA> *param) {
 
 template <>
 void DropoutKernel<FPGA, float>::Compute(
-    const DropoutParam<FPGA> &param) const {}
+    const DropoutParam<FPGA> &param) {}
 
 }  // namespace operators
 }  // namespace paddle_mobile
diff --git a/src/operators/kernel/fpga/elementwise_add_relu_kernel.cpp b/src/operators/kernel/fpga/elementwise_add_relu_kernel.cpp
index f0d8533641941fe43a6d06b49266ac06646a7b4d..b8c01ebf3fa4c2e0413a4e52c3e9330096d2456a 100644
--- a/src/operators/kernel/fpga/elementwise_add_relu_kernel.cpp
+++ b/src/operators/kernel/fpga/elementwise_add_relu_kernel.cpp
@@ -56,7 +56,7 @@ bool ElementwiseAddReluKernel<FPGA, float>::Init(
 
 template <>
 void ElementwiseAddReluKernel<FPGA, float>::Compute(
-    const ElementwiseAddReluParam<FPGA> &param) const {
+    const ElementwiseAddReluParam<FPGA> &param) {
   fpga::ComputeFpgaEWAdd(param.FpgaArgs());
 }
 }  // namespace operators
diff --git a/src/operators/kernel/fpga/fc_relu_kernel.cpp b/src/operators/kernel/fpga/fc_relu_kernel.cpp
index 38b39f982ce41c7d5a88b82f21e446b05c859a2c..53a9389cc4551d68a67ef50b2a0b37867e027544 100644
--- a/src/operators/kernel/fpga/fc_relu_kernel.cpp
+++ b/src/operators/kernel/fpga/fc_relu_kernel.cpp
@@ -59,7 +59,7 @@ bool FusionFcReluKernel<FPGA, float>::Init(FusionFcReluParam<FPGA> *param) {
 }
 template <>
 void FusionFcReluKernel<FPGA, float>::Compute(
-    const FusionFcReluParam<FPGA> &param) const {
+    const FusionFcReluParam<FPGA> &param) {
   fpga::ComputeFpgaConv(param.FpgaArgs());
 };
 
diff --git a/src/operators/kernel/fpga/fusion_fc_kernel.cpp b/src/operators/kernel/fpga/fusion_fc_kernel.cpp
index 6dee8ea6a7e1b26bec4ffd3ed324db4a4ac3be2d..b105c51c892d4ae32cc56a4bf0ed03d3c4f5ab51 100644
--- a/src/operators/kernel/fpga/fusion_fc_kernel.cpp
+++ b/src/operators/kernel/fpga/fusion_fc_kernel.cpp
@@ -61,7 +61,7 @@ bool FusionFcKernel<FPGA, float>::Init(FusionFcParam<FPGA> *param) {
 
 template <>
 void FusionFcKernel<FPGA, float>::Compute(
-    const FusionFcParam<FPGA> &param) const {
+    const FusionFcParam<FPGA> &param) {
   fpga::ComputeFpgaConv(param.FpgaArgs());
 }
 }  // namespace operators
diff --git a/src/operators/kernel/fpga/pool_kernel.cpp b/src/operators/kernel/fpga/pool_kernel.cpp
index 4dad2f789baeb6e381c66ed861b8a8360fa2996e..5e08fcfab6baa49c99b7a58a54d6e09f3d2a1433 100644
--- a/src/operators/kernel/fpga/pool_kernel.cpp
+++ b/src/operators/kernel/fpga/pool_kernel.cpp
@@ -49,7 +49,7 @@ bool PoolKernel<FPGA, float>::Init(PoolParam<FPGA> *param) {
 }
 
 template <>
-void PoolKernel<FPGA, float>::Compute(const PoolParam<FPGA> &param) const {
+void PoolKernel<FPGA, float>::Compute(const PoolParam<FPGA> &param) {
   fpga::ComputeFpgaPool(param.FpgaArgs());
 }
 }  // namespace operators
diff --git a/src/operators/kernel/fpga/softmax_kernel.cpp b/src/operators/kernel/fpga/softmax_kernel.cpp
index 7cfd0c7d76c1a8e73955dbec1971d86ceebde259..f4b5f535cf222a912bb8adeae1aed722b3a0b639 100644
--- a/src/operators/kernel/fpga/softmax_kernel.cpp
+++ b/src/operators/kernel/fpga/softmax_kernel.cpp
@@ -48,7 +48,7 @@ bool SoftmaxKernel<FPGA, float>::Init(SoftmaxParam<FPGA> *param) {
 
 template <>
 void SoftmaxKernel<FPGA, float>::Compute(
-    const SoftmaxParam<FPGA> &param) const {
+    const SoftmaxParam<FPGA> &param) {
   Tensor *in_x = param.FloatInput();
   Tensor *out = param.Out();
 
diff --git a/src/operators/kernel/fusion_fc_kernel.h b/src/operators/kernel/fusion_fc_kernel.h
index 39cfd898a203e742168a775ec892e562bd19f5db..43ca8a84ac0f809dd3a817bccd9b321dd9de74ba 100644
--- a/src/operators/kernel/fusion_fc_kernel.h
+++ b/src/operators/kernel/fusion_fc_kernel.h
@@ -27,7 +27,7 @@ template <typename DeviceType, typename T>
 class FusionFcKernel
     : public framework::OpKernelBase<DeviceType, FusionFcParam<DeviceType>> {
  public:
-  void Compute(const FusionFcParam<DeviceType>& param) const;
+  void Compute(const FusionFcParam<DeviceType>& param);
   bool Init(FusionFcParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/gru_kernel.h b/src/operators/kernel/gru_kernel.h
index 6b02663bd0e2982bdb2480c54632d2a8da9f67fc..b03b2e3ecb514fdf962bde9c06620fa6e64934df 100644
--- a/src/operators/kernel/gru_kernel.h
+++ b/src/operators/kernel/gru_kernel.h
@@ -28,7 +28,7 @@ template <typename DeviceType, typename T>
 class GruKernel
     : public framework::OpKernelBase<DeviceType, GruParam<DeviceType>> {
  public:
-  void Compute(const GruParam<DeviceType>& param) const;
+  void Compute(const GruParam<DeviceType>& param);
   bool Init(GruParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/im2sequence_kernel.h b/src/operators/kernel/im2sequence_kernel.h
index df93ea5abacda1a5291caa53dc5dae7ea2b5d710..b15eb68996a990f6bc770db6940be83a0eea0cbf 100644
--- a/src/operators/kernel/im2sequence_kernel.h
+++ b/src/operators/kernel/im2sequence_kernel.h
@@ -29,7 +29,7 @@ template <typename DeviceType, typename T>
 class Im2SequenceKernel
     : public framework::OpKernelBase<DeviceType, Im2SequenceParam<DeviceType>> {
  public:
-  void Compute(const Im2SequenceParam<DeviceType>& param) const;
+  void Compute(const Im2SequenceParam<DeviceType>& param);
   bool Init(Im2SequenceParam<DeviceType>* para);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/lookup_kernel.h b/src/operators/kernel/lookup_kernel.h
index 73f6cfcced078382b40526eae1f6560d7d168b97..8c29349e737b0fba95688e1ebb8fe893a29b2a4f 100644
--- a/src/operators/kernel/lookup_kernel.h
+++ b/src/operators/kernel/lookup_kernel.h
@@ -28,7 +28,7 @@ template <typename DeviceType, typename T>
 class LookupKernel
     : public framework::OpKernelBase<DeviceType, LookupParam<DeviceType>> {
  public:
-  void Compute(const LookupParam<DeviceType>& param) const;
+  void Compute(const LookupParam<DeviceType>& param);
   bool Init(LookupParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/lrn_kernel.h b/src/operators/kernel/lrn_kernel.h
index 164178f1dcc0ee2523fc9c5fdc4736c14a3e55ce..2cd35f1f92c8b619ca44f1d139b74935b54ca370 100644
--- a/src/operators/kernel/lrn_kernel.h
+++ b/src/operators/kernel/lrn_kernel.h
@@ -173,7 +173,7 @@ template <typename DeviceType, typename T>
 class LrnKernel
     : public framework::OpKernelBase<DeviceType, LrnParam<DeviceType>> {
  public:
-  void Compute(const LrnParam<DeviceType> &param) const;
+  void Compute(const LrnParam<DeviceType> &param);
   bool Init(LrnParam<DeviceType> *param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/mali/batchnorm_kernel.cpp b/src/operators/kernel/mali/batchnorm_kernel.cpp
index 50f6ef5f566347c089869c30b8f7534a4f8b6779..5d50ca9a7250f66f20b6bfaf0d93db18014d791c 100755
--- a/src/operators/kernel/mali/batchnorm_kernel.cpp
+++ b/src/operators/kernel/mali/batchnorm_kernel.cpp
@@ -145,7 +145,7 @@ bool BatchNormKernel<GPU_MALI, float>::Init(BatchNormParam<GPU_MALI>* param) {
 
 template <>
 void BatchNormKernel<GPU_MALI, float>::Compute(
-    const BatchNormParam<GPU_MALI>& param) const {
+    const BatchNormParam<GPU_MALI>& param) {
   std::cout << "init acl" << std::endl;
   AclBatchNormOp<GPU_MALI, float>* acl_op =
       reinterpret_cast<AclBatchNormOp<GPU_MALI, float>*>(this->GetAclOp());
diff --git a/src/operators/kernel/mali/concat_kernel.cpp b/src/operators/kernel/mali/concat_kernel.cpp
index 267c0101a8f66de3d508dbe5795c87ee5027a288..2fb05ab10eccf4e0dca9c74bbcc83067b438e981 100644
--- a/src/operators/kernel/mali/concat_kernel.cpp
+++ b/src/operators/kernel/mali/concat_kernel.cpp
@@ -118,7 +118,7 @@ bool ConcatKernel<GPU_MALI, float>::Init(ConcatParam<GPU_MALI>* param) {
 
 template <>
 void ConcatKernel<GPU_MALI, float>::Compute(
-    const ConcatParam<GPU_MALI>& param) const {
+    const ConcatParam<GPU_MALI>& param) {
   std::cout << "init acl" << std::endl;
   AclConcatOp<GPU_MALI, float>* acl_op =
       reinterpret_cast<AclConcatOp<GPU_MALI, float>*>(this->GetAclOp());
diff --git a/src/operators/kernel/mali/conv_add_kernel.cpp b/src/operators/kernel/mali/conv_add_kernel.cpp
index 74cace00dd2dead7a5d9ddfc76e2d48c67cccf89..427bcd596f71bf434ea155d04f192c5bdedfded5 100644
--- a/src/operators/kernel/mali/conv_add_kernel.cpp
+++ b/src/operators/kernel/mali/conv_add_kernel.cpp
@@ -212,7 +212,7 @@ bool ConvAddKernel<GPU_MALI, float>::Init(FusionConvAddParam<GPU_MALI>* param) {
 
 template <>
 void ConvAddKernel<GPU_MALI, float>::Compute(
-    const FusionConvAddParam<GPU_MALI>& param) const {
+    const FusionConvAddParam<GPU_MALI>& param) {
   std::cout << "init acl" << std::endl;
   AclConvAddOp<GPU_MALI, float>* acl_op =
       reinterpret_cast<AclConvAddOp<GPU_MALI, float>*>(this->GetAclOp());
diff --git a/src/operators/kernel/mali/conv_kernel.cpp b/src/operators/kernel/mali/conv_kernel.cpp
index 7852e64990e5a2cd6f3d7e803e71c23c55aa7a27..7de2a755cbacc3c511de0fcadba9a8797ec76fbf 100644
--- a/src/operators/kernel/mali/conv_kernel.cpp
+++ b/src/operators/kernel/mali/conv_kernel.cpp
@@ -212,7 +212,7 @@ bool ConvKernel<GPU_MALI, float>::Init(ConvParam<GPU_MALI>* param) {
 
 template <>
 void ConvKernel<GPU_MALI, float>::Compute(
-    const ConvParam<GPU_MALI>& param) const {
+    const ConvParam<GPU_MALI>& param) {
   std::cout << "init acl" << std::endl;
   AclConvOp<GPU_MALI, float>* acl_op =
       reinterpret_cast<AclConvOp<GPU_MALI, float>*>(this->GetAclOp());
diff --git a/src/operators/kernel/mali/elementwise_add_kernel.cpp b/src/operators/kernel/mali/elementwise_add_kernel.cpp
index 5596476e1bb33ecc2b3122bf237090b099307156..3711a946b508c9ad71f59dd85f2e01c99bccc9e5 100644
--- a/src/operators/kernel/mali/elementwise_add_kernel.cpp
+++ b/src/operators/kernel/mali/elementwise_add_kernel.cpp
@@ -34,7 +34,7 @@ bool ElementwiseAddKernel<GPU_MALI, float>::Init(
 
 template <>
 void ElementwiseAddKernel<GPU_MALI, float>::Compute(
-    const ElementwiseAddParam<GPU_MALI> &param) const {
+    const ElementwiseAddParam<GPU_MALI> &param) {
   const Tensor *input_x = param.InputX();
   const Tensor *input_y = param.InputY();
   Tensor *Out = param.Out();
diff --git a/src/operators/kernel/mali/fushion_fc_kernel.cpp b/src/operators/kernel/mali/fushion_fc_kernel.cpp
index c3197f38c6c6ee1a4f4f684c824a9a9e43d69d4f..5e59215834ce00e902deb19e54e149b3b4cfb8ac 100755
--- a/src/operators/kernel/mali/fushion_fc_kernel.cpp
+++ b/src/operators/kernel/mali/fushion_fc_kernel.cpp
@@ -26,7 +26,7 @@ bool FusionFcKernel<GPU_MALI, float>::Init(FusionFcParam<GPU_MALI> *param) {
 
 template <>
 void FusionFcKernel<GPU_MALI, float>::Compute(
-    const FusionFcParam<GPU_MALI> &param) const {
+    const FusionFcParam<GPU_MALI> &param) {
   const Tensor *input_x = param.InputX();
   const Tensor *input_y = param.InputY();
   const Tensor *input_z = param.InputZ();
diff --git a/src/operators/kernel/mali/lrn_kernel.cpp b/src/operators/kernel/mali/lrn_kernel.cpp
index fc088f735c538bedc4d5c79593aa31c48acc4fc6..5a84ce1aad15468ec58020b3d5151e2e8791b2ac 100644
--- a/src/operators/kernel/mali/lrn_kernel.cpp
+++ b/src/operators/kernel/mali/lrn_kernel.cpp
@@ -128,7 +128,7 @@ bool LrnKernel<GPU_MALI, float>::Init(LrnParam<GPU_MALI>* param) {
 
 template <>
 void LrnKernel<GPU_MALI, float>::Compute(
-    const LrnParam<GPU_MALI>& param) const {
+    const LrnParam<GPU_MALI>& param) {
   std::cout << "init acl" << std::endl;
   AclLrnOp<GPU_MALI, float>* acl_op =
       reinterpret_cast<AclLrnOp<GPU_MALI, float>*>(this->GetAclOp());
diff --git a/src/operators/kernel/mali/mul_kernel.cpp b/src/operators/kernel/mali/mul_kernel.cpp
index a9e54dad2b51c595be4f68df3916a4803047617e..b3658d34545009b687e5d7738af0a04f59f926fd 100644
--- a/src/operators/kernel/mali/mul_kernel.cpp
+++ b/src/operators/kernel/mali/mul_kernel.cpp
@@ -28,7 +28,7 @@ bool MulKernel<GPU_MALI, float>::Init(MulParam<GPU_MALI> *param) {
 
 template <>
 void MulKernel<GPU_MALI, float>::Compute(
-    const MulParam<GPU_MALI> &param) const {
+    const MulParam<GPU_MALI> &param) {
   const Tensor *input_x = param.InputX();
   const Tensor *input_y = param.InputY();
   Tensor *out = param.Out();
diff --git a/src/operators/kernel/mali/pool_kernel.cpp b/src/operators/kernel/mali/pool_kernel.cpp
index 33b3bd7017739144a519bfb1be247b4751883779..0c94b595a85f8dcaa1ad9e3da274dfb07238c53c 100644
--- a/src/operators/kernel/mali/pool_kernel.cpp
+++ b/src/operators/kernel/mali/pool_kernel.cpp
@@ -196,7 +196,7 @@ bool PoolKernel<GPU_MALI, float>::Init(PoolParam<GPU_MALI>* param) {
 
 template <>
 void PoolKernel<GPU_MALI, float>::Compute(
-    const PoolParam<GPU_MALI>& param) const {
+    const PoolParam<GPU_MALI>& param) {
   std::cout << "init acl" << std::endl;
   AclPoolOp<GPU_MALI, float>* acl_op =
       reinterpret_cast<AclPoolOp<GPU_MALI, float>*>(this->GetAclOp());
diff --git a/src/operators/kernel/mali/relu_kernel.cpp b/src/operators/kernel/mali/relu_kernel.cpp
index 10b270800dee1a0ad8176da1f788100d29b60173..23334c4ffda0001817badf174b76d603e16aaacf 100644
--- a/src/operators/kernel/mali/relu_kernel.cpp
+++ b/src/operators/kernel/mali/relu_kernel.cpp
@@ -116,7 +116,7 @@ bool ReluKernel<GPU_MALI, float>::Init(ReluParam<GPU_MALI>* param) {
 
 template <>
 void ReluKernel<GPU_MALI, float>::Compute(
-    const ReluParam<GPU_MALI>& param) const {
+    const ReluParam<GPU_MALI>& param) {
   std::cout << "init acl" << std::endl;
   AclReluOp<GPU_MALI, float>* acl_op =
       reinterpret_cast<AclReluOp<GPU_MALI, float>*>(this->GetAclOp());
diff --git a/src/operators/kernel/mali/reshape_kernel.cpp b/src/operators/kernel/mali/reshape_kernel.cpp
index 69c077e252162017cb477a000b5f17f5a968fc10..f98906c0a982c10896e75101eaa2732d75d6cdf4 100644
--- a/src/operators/kernel/mali/reshape_kernel.cpp
+++ b/src/operators/kernel/mali/reshape_kernel.cpp
@@ -28,7 +28,7 @@ bool ReshapeKernel<GPU_MALI, float>::Init(ReshapeParam<GPU_MALI> *param) {
 
 template <>
 void ReshapeKernel<GPU_MALI, float>::Compute(
-    const ReshapeParam<GPU_MALI> &param) const {
+    const ReshapeParam<GPU_MALI> &param) {
   const auto *input_x = param.InputX();
   const auto &input_x_dims = input_x->dims();
   auto *out = param.Out();
diff --git a/src/operators/kernel/mali/softmax_kernel.cpp b/src/operators/kernel/mali/softmax_kernel.cpp
index d4f25c96cc47d7baa394645d4e0c84e0e3f7ad29..d6ce1ecb61c2790c68883231eb6b90dcde43a956 100644
--- a/src/operators/kernel/mali/softmax_kernel.cpp
+++ b/src/operators/kernel/mali/softmax_kernel.cpp
@@ -113,7 +113,7 @@ bool SoftmaxKernel<GPU_MALI, float>::Init(SoftmaxParam<GPU_MALI>* param) {
 
 template <>
 void SoftmaxKernel<GPU_MALI, float>::Compute(
-    const SoftmaxParam<GPU_MALI>& param) const {
+    const SoftmaxParam<GPU_MALI>& param) {
   std::cout << "init acl" << std::endl;
   AclSoftmaxOp<GPU_MALI, float>* acl_op =
       reinterpret_cast<AclSoftmaxOp<GPU_MALI, float>*>(this->GetAclOp());
diff --git a/src/operators/kernel/mul_kernel.h b/src/operators/kernel/mul_kernel.h
index e441de4d4495b736aec248c0ef85191b32bfcbf9..8deb4a2cb74786257ddfc12c805c4a7d56589bbf 100644
--- a/src/operators/kernel/mul_kernel.h
+++ b/src/operators/kernel/mul_kernel.h
@@ -29,7 +29,7 @@ template <typename DeviceType, typename T>
 class MulKernel
     : public framework::OpKernelBase<DeviceType, MulParam<DeviceType>> {
  public:
-  void Compute(const MulParam<DeviceType> &param) const;
+  void Compute(const MulParam<DeviceType> &param);
   bool Init(MulParam<DeviceType> *param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/multiclass_nms_kernel.h b/src/operators/kernel/multiclass_nms_kernel.h
index b1b20ddd81b395ea94ae62b1abf2fe861d9257db..6a4ac0c22941aa364f05e38c7abaf29948cd324b 100644
--- a/src/operators/kernel/multiclass_nms_kernel.h
+++ b/src/operators/kernel/multiclass_nms_kernel.h
@@ -28,7 +28,7 @@ class MultiClassNMSKernel
     : public framework::OpKernelBase<DeviceType,
                                      MultiClassNMSParam<DeviceType>> {
  public:
-  void Compute(const MultiClassNMSParam<DeviceType>& param) const;
+  void Compute(const MultiClassNMSParam<DeviceType>& param);
   bool Init(MultiClassNMSParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/pool_kernel.h b/src/operators/kernel/pool_kernel.h
index 2be254444cc410fb95a94125cccb224ca9505545..ff80e0e44536d924026dbbe80a09677c069a8f6b 100644
--- a/src/operators/kernel/pool_kernel.h
+++ b/src/operators/kernel/pool_kernel.h
@@ -26,7 +26,7 @@ using framework::OpKernelBase;
 template <typename DeviceType, typename T>
 class PoolKernel : public OpKernelBase<DeviceType, PoolParam<DeviceType>> {
  public:
-  void Compute(const PoolParam<DeviceType> &param) const override;
+  void Compute(const PoolParam<DeviceType> &param);
   bool Init(PoolParam<DeviceType> *param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/prelu_kernel.h b/src/operators/kernel/prelu_kernel.h
index f6c7c3ac7f139cf7eafe8843ef48e53c90292082..c043149243f21f2abceeed37c5d0e81a61e5059f 100644
--- a/src/operators/kernel/prelu_kernel.h
+++ b/src/operators/kernel/prelu_kernel.h
@@ -24,7 +24,7 @@ template <typename DeviceType, typename T>
 class PReluKernel
     : public framework::OpKernelBase<DeviceType, PReluParam<DeviceType>> {
  public:
-  void Compute(const PReluParam<DeviceType>& param) const;
+  void Compute(const PReluParam<DeviceType>& param);
 };
 }  // namespace operators
 }  // namespace paddle_mobile
diff --git a/src/operators/kernel/prior_box_kernel.h b/src/operators/kernel/prior_box_kernel.h
index 5640375483d42d52965986dab6795254bbf4b908..921d5901a8f24abab61f7aa94663385d91e597a7 100644
--- a/src/operators/kernel/prior_box_kernel.h
+++ b/src/operators/kernel/prior_box_kernel.h
@@ -54,7 +54,7 @@ template <typename DeviceType, typename T>
 class PriorBoxKernel
     : public framework::OpKernelBase<DeviceType, PriorBoxParam<DeviceType>> {
  public:
-  void Compute(const PriorBoxParam<DeviceType>& param) const;
+  void Compute(const PriorBoxParam<DeviceType>& param);
   bool Init(PriorBoxParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/relu_kernel.h b/src/operators/kernel/relu_kernel.h
index b0c32791d626f14b0840ce1c8f3f12f02b403d97..48f47c2de6df8d3aa9461fba915fd1a6406d4b9f 100644
--- a/src/operators/kernel/relu_kernel.h
+++ b/src/operators/kernel/relu_kernel.h
@@ -27,7 +27,7 @@ template <typename DeviceType, typename T>
 class ReluKernel
     : public framework::OpKernelBase<DeviceType, ReluParam<DeviceType>> {
  public:
-  void Compute(const ReluParam<DeviceType>& param) const;
+  void Compute(const ReluParam<DeviceType>& param);
   bool Init(ReluParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/reshape_kernel.h b/src/operators/kernel/reshape_kernel.h
index 73eb63f797f34ec4eb2baec8c4ab79fafb06f0e2..a5405654874320cdfe3432d16d3a8c6358d2d8e1 100644
--- a/src/operators/kernel/reshape_kernel.h
+++ b/src/operators/kernel/reshape_kernel.h
@@ -71,7 +71,7 @@ template <typename DeviceType, typename T>
 class ReshapeKernel
     : public framework::OpKernelBase<DeviceType, ReshapeParam<DeviceType>> {
  public:
-  void Compute(const ReshapeParam<DeviceType>& param) const;
+  void Compute(const ReshapeParam<DeviceType>& param);
   bool Init(ReshapeParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/resize_kernel.h b/src/operators/kernel/resize_kernel.h
index 4c06429858b9575ffc061c000e4a9343fa7eee26..249488c6ffbf88cadebefd9384b3ff3e97b14063 100644
--- a/src/operators/kernel/resize_kernel.h
+++ b/src/operators/kernel/resize_kernel.h
@@ -73,7 +73,7 @@ template <typename DeviceType, typename T>
 class ResizeKernel
     : public framework::OpKernelBase<DeviceType, ResizeParam<DeviceType>> {
  public:
-  void Compute(const ResizeParam<DeviceType> &param) const;
+  void Compute(const ResizeParam<DeviceType> &param);
 };
 }  // namespace operators
 }  // namespace paddle_mobile
diff --git a/src/operators/kernel/scale_kernel.h b/src/operators/kernel/scale_kernel.h
index 2da92d8d3c8b0d7867e7e6e628a04a853dd69464..a17e57652224992b2ee7127e6081804bf3253fb1 100644
--- a/src/operators/kernel/scale_kernel.h
+++ b/src/operators/kernel/scale_kernel.h
@@ -24,7 +24,7 @@ template <typename DeviceType, typename T>
 class ScaleKernel
     : public framework::OpKernelBase<DeviceType, ScaleParam<DeviceType>> {
  public:
-  void Compute(const ScaleParam<DeviceType>& param) const;
+  void Compute(const ScaleParam<DeviceType>& param);
 };
 }  // namespace operators
 }  // namespace paddle_mobile
diff --git a/src/operators/kernel/shape_kernel.h b/src/operators/kernel/shape_kernel.h
index 7caf3e427a4f3b469265248708a3090c52d1ca91..9d3c6e1701523acc43410fb0e3402b5679d4f19a 100644
--- a/src/operators/kernel/shape_kernel.h
+++ b/src/operators/kernel/shape_kernel.h
@@ -28,7 +28,7 @@ template <typename DeviceType, typename T>
 class ShapeKernel
     : public framework::OpKernelBase<DeviceType, ShapeParam<DeviceType>> {
  public:
-  void Compute(const ShapeParam<DeviceType>& param) const;
+  void Compute(const ShapeParam<DeviceType>& param);
   bool Init(ShapeParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/sigmoid_kernel.h b/src/operators/kernel/sigmoid_kernel.h
index e8cfe6cad9ce2f25b9f38e1784ded9ea0741ff9a..3d9aa3089ca2416dc56c7b1b55a68bcc48901025 100644
--- a/src/operators/kernel/sigmoid_kernel.h
+++ b/src/operators/kernel/sigmoid_kernel.h
@@ -26,7 +26,7 @@ template <typename DeviceType, typename T>
 class SigmoidKernel
     : public OpKernelBase<DeviceType, SigmoidParam<DeviceType>> {
  public:
-  void Compute(const SigmoidParam<DeviceType>& param) const override;
+  void Compute(const SigmoidParam<DeviceType>& param);
   bool Init(SigmoidParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/slice_kernel.h b/src/operators/kernel/slice_kernel.h
index 17f7fe4a9ebf5b78fc92c41abd4756a7bc6bff45..6ae6528622b37c2f2694d70da3e74540e3404c99 100644
--- a/src/operators/kernel/slice_kernel.h
+++ b/src/operators/kernel/slice_kernel.h
@@ -24,7 +24,7 @@ template <typename DeviceType, typename T>
 class SliceKernel
     : public framework::OpKernelBase<DeviceType, SliceParam<DeviceType>> {
  public:
-  void Compute(const SliceParam<DeviceType>& param) const {}
+  void Compute(const SliceParam<DeviceType>& param) {}
 };
 }  // namespace operators
 }  // namespace paddle_mobile
diff --git a/src/operators/kernel/softmax_kernel.h b/src/operators/kernel/softmax_kernel.h
index 67bd9167e8c717355fc326d3025cde410ce66010..d7d7435fd5145e702de848872f93087188fd31fc 100644
--- a/src/operators/kernel/softmax_kernel.h
+++ b/src/operators/kernel/softmax_kernel.h
@@ -27,7 +27,7 @@ template <typename DeviceType, typename T>
 class SoftmaxKernel
     : public OpKernelBase<DeviceType, SoftmaxParam<DeviceType>> {
  public:
-  void Compute(const SoftmaxParam<DeviceType> &param) const override;
+  void Compute(const SoftmaxParam<DeviceType> &param);
   bool Init(SoftmaxParam<DeviceType> *param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/split_kernel.h b/src/operators/kernel/split_kernel.h
index 03a418de59606e42684c67ca3053fa8e39b07940..3a2c03dce718e650ebf9127044f0db44d9d5c9a5 100644
--- a/src/operators/kernel/split_kernel.h
+++ b/src/operators/kernel/split_kernel.h
@@ -28,7 +28,7 @@ template <typename DeviceType, typename T>
 class SplitKernel
     : public framework::OpKernelBase<DeviceType, SplitParam<DeviceType>> {
  public:
-  void Compute(const SplitParam<DeviceType>& param) const;
+  void Compute(const SplitParam<DeviceType>& param);
   bool Init(SplitParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/kernel/transpose_kernel.h b/src/operators/kernel/transpose_kernel.h
index 56c41fd221e080a4db3b34fbd4ab208c9986c2a8..63ee6eb172ff691ff51dd3f74613cd3e412210bf 100644
--- a/src/operators/kernel/transpose_kernel.h
+++ b/src/operators/kernel/transpose_kernel.h
@@ -28,7 +28,7 @@ template <typename DeviceType, typename T>
 class TransposeKernel
     : public framework::OpKernelBase<DeviceType, TransposeParam<DeviceType>> {
  public:
-  void Compute(const TransposeParam<DeviceType>& param) const;
+  void Compute(const TransposeParam<DeviceType>& param);
   bool Init(TransposeParam<DeviceType>* param);
 };
 }  // namespace operators
diff --git a/src/operators/lookup_op.h b/src/operators/lookup_op.h
index 9c9d03c8d10e9b01ad958c12d31a49908075eb27..b5e14d3cb15985057ec2889b89661fe953ce0ff8 100644
--- a/src/operators/lookup_op.h
+++ b/src/operators/lookup_op.h
@@ -37,10 +37,6 @@ class LookupOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, LookupParam<DeviceType>,
                                       operators::LookupKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, LookupParam<DeviceType>,
-      operators::LookupKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 };
 
diff --git a/src/operators/lrn_op.h b/src/operators/lrn_op.h
index 6c609c7654cca022f473dba0aad1f4214a4e43e3..b5e7a3de6faeea47d03a4a0cbdb8fc19923d0514 100644
--- a/src/operators/lrn_op.h
+++ b/src/operators/lrn_op.h
@@ -35,10 +35,6 @@ class LrnOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, LrnParam<DeviceType>,
                                       operators::LrnKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, LrnParam<DeviceType>,
-      operators::LrnKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/mul_op.h b/src/operators/mul_op.h
index 127048efbacf2da87de9371cd8e54875f8554d61..c0a28054f4a5d1a2e6852de7fdb5fb7b01a1af9f 100644
--- a/src/operators/mul_op.h
+++ b/src/operators/mul_op.h
@@ -35,10 +35,6 @@ class MulOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, MulParam<DeviceType>,
                                       operators::MulKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, MulParam<DeviceType>,
-      operators::MulKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/multiclass_nms_op.h b/src/operators/multiclass_nms_op.h
index b40ef5ee009f6c16c685479ffcf58186958bb4cc..6d3058a9605439e66b4852292750c5e512248853 100644
--- a/src/operators/multiclass_nms_op.h
+++ b/src/operators/multiclass_nms_op.h
@@ -40,10 +40,6 @@ class MultiClassNMSOp : public framework::OperatorWithKernel<
             DeviceType, MultiClassNMSParam<DeviceType>,
             operators::MultiClassNMSKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, MultiClassNMSParam<DeviceType>,
-      operators::MultiClassNMSKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/pool_op.h b/src/operators/pool_op.h
index 4f76fb8f800dea43432b48562cca563505a1af76..0b43b607a2788955e82a3e98084defc81d2ec58d 100644
--- a/src/operators/pool_op.h
+++ b/src/operators/pool_op.h
@@ -38,9 +38,6 @@ class PoolOp : public OperatorWithKernel<DeviceType, PoolParam<DeviceType>,
       : OperatorWithKernel<DeviceType, PoolParam<DeviceType>,
                            operators::PoolKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-  using OperatorWithKernel<
-      DeviceType, PoolParam<DeviceType>,
-      operators::PoolKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  private:
diff --git a/src/operators/prelu_op.h b/src/operators/prelu_op.h
index 7b6b778fa6e8f0951faffda6803b25b6b23ea17c..83ecf73906c137d0911ce5f40da949bac6e566fd 100644
--- a/src/operators/prelu_op.h
+++ b/src/operators/prelu_op.h
@@ -38,10 +38,6 @@ class PReluOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, PReluParam<DeviceType>,
                                       operators::PReluKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, PReluParam<DeviceType>,
-      operators::PReluKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/prior_box_op.h b/src/operators/prior_box_op.h
index 00fc8e039c9958e4b43653d6360c0f54c78648a1..7a19266f0435e0df30ebf19c2c4cd89f44cb3d45 100644
--- a/src/operators/prior_box_op.h
+++ b/src/operators/prior_box_op.h
@@ -40,9 +40,6 @@ class PriorBoxOp : public framework::OperatorWithKernel<
                                       operators::PriorBoxKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
 
-  using framework::OperatorWithKernel<
-      DeviceType, PriorBoxParam<DeviceType>,
-      operators::PriorBoxKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/relu_op.h b/src/operators/relu_op.h
index 0364dd7f8ec4b3861200380597e18ede0819e8b6..403de6f3613628878e2c7f1c7e2aaa82ecdd7d9e 100644
--- a/src/operators/relu_op.h
+++ b/src/operators/relu_op.h
@@ -41,10 +41,6 @@ class ReluOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, ReluParam<DeviceType>,
                                       operators::ReluKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, ReluParam<DeviceType>,
-      operators::ReluKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/reshape_op.h b/src/operators/reshape_op.h
index 9284e94f346ed0f225d6dabe16077b1fb2034c64..7df6890018ca34d95999124fb473cf7a5d214c2c 100644
--- a/src/operators/reshape_op.h
+++ b/src/operators/reshape_op.h
@@ -39,10 +39,6 @@ class ReshapeOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, ReshapeParam<DeviceType>,
                                       operators::ReshapeKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, ReshapeParam<DeviceType>,
-      operators::ReshapeKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/resize_op.h b/src/operators/resize_op.h
index c0b38bb1cf4048af4b07d05f28a88a5ac8056ea3..954b3a82f8d2b5ccba242045c3d5e0f28553d484 100644
--- a/src/operators/resize_op.h
+++ b/src/operators/resize_op.h
@@ -38,10 +38,6 @@ class ResizeOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, ResizeParam<DeviceType>,
                                       operators::ResizeKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, ResizeParam<DeviceType>,
-      operators::ResizeKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/scale_op.h b/src/operators/scale_op.h
index 4c5f5e620f25bef88533e80cdd78b243fef9bc70..56265259fe3a10feda67cc5c5732b2ba44e0730e 100644
--- a/src/operators/scale_op.h
+++ b/src/operators/scale_op.h
@@ -38,10 +38,6 @@ class ScaleOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, ScaleParam<DeviceType>,
                                       operators::ScaleKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, ScaleParam<DeviceType>,
-      operators::ScaleKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/shape_op.h b/src/operators/shape_op.h
index 2f88c807d3c331f83cf87e6c77a65fa5d90a9f4e..a46f3d47ffc6daa26fd40c09e1ed357772910f9e 100644
--- a/src/operators/shape_op.h
+++ b/src/operators/shape_op.h
@@ -38,10 +38,6 @@ class ShapeOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, ShapeParam<DeviceType>,
                                       operators::ShapeKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, ShapeParam<DeviceType>,
-      operators::ShapeKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 };
 
diff --git a/src/operators/sigmoid_op.h b/src/operators/sigmoid_op.h
index 533ea587958e8766b1469c73b909cfa2fcb60696..1f64f15b9046367216f986a923eceb311bc64d5b 100644
--- a/src/operators/sigmoid_op.h
+++ b/src/operators/sigmoid_op.h
@@ -36,11 +36,6 @@ class SigmoidOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, SigmoidParam<DeviceType>,
                                       operators::SigmoidKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, SigmoidParam<DeviceType>,
-      operators::SigmoidKernel<DeviceType, T>>::OperatorWithKernel;
-
   void InferShape() const override;
 };
 }  // namespace operators
diff --git a/src/operators/slice_op.h b/src/operators/slice_op.h
index 6bcb6fa0b9e88cefb3c88dfc096e1073ad261c1b..c45061696577dbe6948fb9cab7edebbaf8e15f2f 100644
--- a/src/operators/slice_op.h
+++ b/src/operators/slice_op.h
@@ -38,10 +38,6 @@ class SliceOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, SliceParam<DeviceType>,
                                       operators::SliceKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, SliceParam<DeviceType>,
-      operators::SliceKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 
  protected:
diff --git a/src/operators/softmax_op.h b/src/operators/softmax_op.h
index 579a2ed605cb3f3c8c4a3d0c2f1ccc7bd9595fc2..d532332992a7df332e267355d148aefa2175998b 100644
--- a/src/operators/softmax_op.h
+++ b/src/operators/softmax_op.h
@@ -36,11 +36,6 @@ class SoftmaxOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, SoftmaxParam<DeviceType>,
                                       operators::SoftmaxKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, SoftmaxParam<DeviceType>,
-      operators::SoftmaxKernel<DeviceType, T>>::OperatorWithKernel;
-
   void InferShape() const override;
 
  private:
diff --git a/src/operators/split_op.h b/src/operators/split_op.h
index f7d60b37441e77c5d47ac6040404535a841bcf8e..ecb18918c55b0bcc954b11149ca15937b516c2f1 100644
--- a/src/operators/split_op.h
+++ b/src/operators/split_op.h
@@ -38,10 +38,6 @@ class SplitOp : public framework::OperatorWithKernel<
       : framework::OperatorWithKernel<DeviceType, SplitParam<DeviceType>,
                                       operators::SplitKernel<DeviceType, T>>(
             type, inputs, outputs, attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, SplitParam<DeviceType>,
-      operators::SplitKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 };
 }  // namespace operators
diff --git a/src/operators/transpose_op.h b/src/operators/transpose_op.h
index b96ce4e17ca4b0d0e321cefb3175b973cd7df307..058d2d59eee1a33f27027daeada1f1eaad7b5042 100644
--- a/src/operators/transpose_op.h
+++ b/src/operators/transpose_op.h
@@ -40,10 +40,6 @@ class TransposeOp : public framework::OperatorWithKernel<
             DeviceType, TransposeParam<DeviceType>,
             operators::TransposeKernel<DeviceType, T>>(type, inputs, outputs,
                                                        attrs, scope) {}
-
-  using framework::OperatorWithKernel<
-      DeviceType, TransposeParam<DeviceType>,
-      operators::TransposeKernel<DeviceType, T>>::OperatorWithKernel;
   void InferShape() const override;
 };
 
diff --git a/test/operators/test_sigmoid_op.cpp b/test/operators/test_sigmoid_op.cpp
index c8fac6b9eee5c5777ddb0147bc81d361d4dd09f5..37d05a44b5b66f2428eedd8e8719cd127747ee08 100644
--- a/test/operators/test_sigmoid_op.cpp
+++ b/test/operators/test_sigmoid_op.cpp
@@ -13,6 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "../../src/operators/kernel/sigmoid_kernel.h"
+#include "../../src/operators/kernel/central-arm-func/sigmoid_arm_func.h"
 #include "../test_helper.h"
 #include "io/executor.h"
 
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl.h b/third_party/opencl/OpenCL-Headers/CL/cl.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl.h
rename to third_party/opencl/OpenCL-Headers/CL/cl.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_d3d10.h b/third_party/opencl/OpenCL-Headers/CL/cl_d3d10.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_d3d10.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_d3d10.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_d3d11.h b/third_party/opencl/OpenCL-Headers/CL/cl_d3d11.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_d3d11.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_d3d11.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_dx9_media_sharing.h b/third_party/opencl/OpenCL-Headers/CL/cl_dx9_media_sharing.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_dx9_media_sharing.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_dx9_media_sharing.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_dx9_media_sharing_intel.h b/third_party/opencl/OpenCL-Headers/CL/cl_dx9_media_sharing_intel.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_dx9_media_sharing_intel.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_dx9_media_sharing_intel.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_egl.h b/third_party/opencl/OpenCL-Headers/CL/cl_egl.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_egl.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_egl.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_ext.h b/third_party/opencl/OpenCL-Headers/CL/cl_ext.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_ext.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_ext.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_ext_intel.h b/third_party/opencl/OpenCL-Headers/CL/cl_ext_intel.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_ext_intel.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_ext_intel.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_gl.h b/third_party/opencl/OpenCL-Headers/CL/cl_gl.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_gl.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_gl.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_gl_ext.h b/third_party/opencl/OpenCL-Headers/CL/cl_gl_ext.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_gl_ext.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_gl_ext.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_platform.h b/third_party/opencl/OpenCL-Headers/CL/cl_platform.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_platform.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_platform.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_va_api_media_sharing_intel.h b/third_party/opencl/OpenCL-Headers/CL/cl_va_api_media_sharing_intel.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_va_api_media_sharing_intel.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_va_api_media_sharing_intel.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/cl_version.h b/third_party/opencl/OpenCL-Headers/CL/cl_version.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/cl_version.h
rename to third_party/opencl/OpenCL-Headers/CL/cl_version.h
diff --git a/third_party/opecl/OpenCL-Headers/CL/opencl.h b/third_party/opencl/OpenCL-Headers/CL/opencl.h
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/CL/opencl.h
rename to third_party/opencl/OpenCL-Headers/CL/opencl.h
diff --git a/third_party/opecl/OpenCL-Headers/LICENSE b/third_party/opencl/OpenCL-Headers/LICENSE
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/LICENSE
rename to third_party/opencl/OpenCL-Headers/LICENSE
diff --git a/third_party/opecl/OpenCL-Headers/README.md b/third_party/opencl/OpenCL-Headers/README.md
similarity index 100%
rename from third_party/opecl/OpenCL-Headers/README.md
rename to third_party/opencl/OpenCL-Headers/README.md