Merge pull request #1027 from codeWorm2015/opencl

complie success

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()

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);

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);

src/framework/cl/cl_helper.h

@@ -25,6 +25,7 @@ namespace framework {
 
 class CLHelper {
  public:
+  CLHelper() = default;
 
   CLHelper(CLScope *scope): scope_(scope) {
   }

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();
   }

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
 
 }
 }

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);

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

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_;
 
 };
 

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

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

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

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;
 };
 

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:

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:

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:

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;
 };
 

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:

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:

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:

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());
   }

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() {}
 

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;
 };
 

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:

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:

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:

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:

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:

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:

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:

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:

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:

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:

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;
 };
 

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:

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);
 }
 

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);
 }
 

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);
 }
 

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());
 }

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>;

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>;

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);
 }
 

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>;

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>;

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>;

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>;

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);
 }
 

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);
 }
 

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);
 }
 

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);
 }
 

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();

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>;

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());
 }

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);
 }
 

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());
 }

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();

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>();

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());
 }

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);
 }
 

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());
 }

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);
 }
 

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

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();

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);
 }
 

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);
 }
 

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);
 }
 

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();

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();

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);
 }
 

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);
 }
 

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);
 }
 

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);
 }
 

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);
 }
 

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);
 };
 

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

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

src/operators/kernel/cl/conv_kernel.cpp

-/* 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

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);
 };
 

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);
 };
 

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);
 };
 

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);
 };
 

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);
 };
 

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);
 };
 

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);
 };
 

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);
 };
 

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);
 };
 

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);
 };
 

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);
 };
 

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);
 };

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

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

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

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);
 };
 

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

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

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

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

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>;

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());
 }
 

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());
 }
 

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());
 }
 

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());
 }
 

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());
 }
 

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

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

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());
 };
 

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

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

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();
 

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

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

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

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

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

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());

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());

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());

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());

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();

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();

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());

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();

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());

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());

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();

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());

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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;
 };
 

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:

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:

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:

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:

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:

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:

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:

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:

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:

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:

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;
 };
 

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

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:

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:

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

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;
 };
 

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"