diff --git a/paddle/fluid/framework/details/nan_inf_utils_detail.h b/paddle/fluid/framework/details/nan_inf_utils_detail.h
index f4ee2c20b9d2f812095e3aeb2b2ca8656ababc06..ed2fa25a5ae6056aabc1d37e95789b371be16c35 100644
--- a/paddle/fluid/framework/details/nan_inf_utils_detail.h
+++ b/paddle/fluid/framework/details/nan_inf_utils_detail.h
@@ -61,7 +61,9 @@ HOSTDEVICE bool NeedPrint(MT max_value, MT min_value, int check_nan_inf_level) {
 template <typename T,
           typename MT,
           std::enable_if_t<!std::is_same<T, float>::value, bool> = true>
-HOSTDEVICE bool NeedPrint(MT max_value, MT min_value, int check_nan_inf_level) {
+HOSTDEVICE bool NeedPrint(MT max_value UNUSED,
+                          MT min_value UNUSED,
+                          int check_nan_inf_level) {
   if (check_nan_inf_level >= 3) {
     return true;
   }
diff --git a/paddle/fluid/framework/ir/fusion_group/code_generator_helper.h b/paddle/fluid/framework/ir/fusion_group/code_generator_helper.h
index 0980b17853cad858db62809a9ad73ec853822629..e068403181b1b7b950be6c7987e24918f7b0d41b 100644
--- a/paddle/fluid/framework/ir/fusion_group/code_generator_helper.h
+++ b/paddle/fluid/framework/ir/fusion_group/code_generator_helper.h
@@ -95,7 +95,7 @@ class TemplateVariable {
     strings_[identifier] = expression;
   }
 
-  void Remove(std::string identifier, std::string expression) {
+  void Remove(std::string identifier, std::string expression UNUSED) {
     for (auto it = strings_.begin(); it != strings_.end();) {
       if (it->first == identifier) {
         it = strings_.erase(it);
diff --git a/paddle/fluid/imperative/data_loader.cc b/paddle/fluid/imperative/data_loader.cc
index 65f6ef4e53ce6b372bc593ec3e01d40ee5e3f66b..f225b871b6860654ab23980dc64710d978d882be 100644
--- a/paddle/fluid/imperative/data_loader.cc
+++ b/paddle/fluid/imperative/data_loader.cc
@@ -73,7 +73,8 @@ void EraseLoadProcessPIDs(int64_t key) {
   } while (0)
 
 #define REGISTER_SIGNAL_HANDLER(SIGNAL, HANDLER_NAME, ERROR_MSG)           \
-  static void HANDLER_NAME(int sig, siginfo_t *info, void *ctx) {          \
+  static void HANDLER_NAME(                                                \
+      int sig UNUSED, siginfo_t *info UNUSED, void *ctx UNUSED) {          \
     auto _w =                                                              \
         write(STDERR_FILENO, ERROR_MSG, sizeof(ERROR_MSG) / sizeof(char)); \
     (void)_w;                                                              \
diff --git a/paddle/fluid/imperative/infer_shape_context.h b/paddle/fluid/imperative/infer_shape_context.h
index 6f1f54de8a95c348a5c3294352cdf4f2c23e909a..80c52ab4ac10f8032c1136d3a1504532c82e1a7d 100644
--- a/paddle/fluid/imperative/infer_shape_context.h
+++ b/paddle/fluid/imperative/infer_shape_context.h
@@ -237,14 +237,14 @@ class DygraphInferShapeContext : public framework::InferShapeContext {
     }
   }
 
-  void ShareAllLoD(const std::string& in,
-                   const std::string& out) const override {
+  void ShareAllLoD(const std::string& in UNUSED,
+                   const std::string& out UNUSED) const override {
     // do nothing
   }
-  void ShareLoD(const std::string& in,
-                const std::string& out,
-                size_t i = 0,
-                size_t j = 0) const override {
+  void ShareLoD(const std::string& in UNUSED,
+                const std::string& out UNUSED,
+                size_t i UNUSED = 0,
+                size_t j UNUSED = 0) const override {
     // do nothing
   }
 
@@ -415,14 +415,15 @@ class DygraphInferShapeContext : public framework::InferShapeContext {
     }
   }
 
-  int32_t GetLoDLevel(const std::string& in, size_t i = 0) const override {
+  int32_t GetLoDLevel(const std::string& in UNUSED,
+                      size_t i UNUSED = 0) const override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "GetLoDLevel function not support in dygraph mode"));
   }
 
-  void SetLoDLevel(const std::string& out,
-                   int32_t lod_level,
-                   size_t j = 0) const override {
+  void SetLoDLevel(const std::string& out UNUSED,
+                   int32_t lod_level UNUSED,
+                   size_t j UNUSED = 0) const override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "SetLoDLevel function not support in dygraph mode"));
   }
@@ -452,7 +453,8 @@ class DygraphInferShapeContext : public framework::InferShapeContext {
     }
   }
 
-  std::vector<DDim> GetRepeatedDims(const std::string& name) const override {
+  std::vector<DDim> GetRepeatedDims(
+      const std::string& name UNUSED) const override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "GetRepeatedDims not support in dygraph runtime"));
   }
@@ -486,8 +488,8 @@ class DygraphInferShapeContext : public framework::InferShapeContext {
     }
   }
 
-  void SetRepeatedDims(const std::string& name,
-                       const std::vector<DDim>& dims) override {
+  void SetRepeatedDims(const std::string& name UNUSED,
+                       const std::vector<DDim>& dims UNUSED) override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "SetRepeatedDims not support in dygraph runtime"));
   }
diff --git a/paddle/fluid/imperative/infer_var_type_context.h b/paddle/fluid/imperative/infer_var_type_context.h
index 36558366e1d2f4f7e22aa142f206721c39beb5bf..fcf9c1d3ab4c9792f6683371e2b5ca746e04499b 100644
--- a/paddle/fluid/imperative/infer_var_type_context.h
+++ b/paddle/fluid/imperative/infer_var_type_context.h
@@ -145,8 +145,8 @@ class RuntimeInferVarTypeContext : public framework::InferVarTypeContext {
   }
 
   void SetOutputDataType(const std::string& name,
-                         framework::proto::VarType::Type type,
-                         int index = 0) override {
+                         framework::proto::VarType::Type type UNUSED,
+                         int index UNUSED = 0) override {
     VLOG(10) << "Set data type in infer var type of Eager mode is meaning less "
                 "for var: "
              << name;
@@ -155,77 +155,79 @@ class RuntimeInferVarTypeContext : public framework::InferVarTypeContext {
   bool IsDygraph() const override { return true; }
 
  protected:
-  bool HasVar(const std::string& name) const override {
+  bool HasVar(const std::string& name UNUSED) const override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "HasVar is not supported in runtime InferVarType"));
   }
 
   const std::vector<std::string>& InputVars(
-      const std::string& name) const override {
+      const std::string& name UNUSED) const override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "InputVars is not supported in runtime InferVarType"));
   }
 
   const std::vector<std::string>& OutputVars(
-      const std::string& name) const override {
+      const std::string& name UNUSED) const override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "OutputVars is not supported in runtime InferVarType"));
   }
 
   framework::proto::VarType::Type GetVarType(
-      const std::string& name) const override {
+      const std::string& name UNUSED) const override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "Do not manipulate var in runtime InferVarType"));
   }
 
-  void SetVarType(const std::string& name,
-                  framework::proto::VarType::Type type) override {
+  void SetVarType(const std::string& name UNUSED,
+                  framework::proto::VarType::Type type UNUSED) override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "Do not manipulate var in runtime InferVarType"));
   }
 
   framework::proto::VarType::Type GetVarDataType(
-      const std::string& name) const override {
+      const std::string& name UNUSED) const override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "Do not manipulate var in runtime InferVarType"));
   }
 
-  void SetVarDataType(const std::string& name,
-                      framework::proto::VarType::Type type) override {
+  void SetVarDataType(const std::string& name UNUSED,
+                      framework::proto::VarType::Type type UNUSED) override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "Do not manipulate var in runtime InferVarType"));
   }
 
   std::vector<framework::proto::VarType::Type> GetVarDataTypes(
-      const std::string& name) const override {
+      const std::string& name UNUSED) const override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "GetVarDataTypes is not supported in runtime InferVarType"));
   }
 
-  void SetVarDataTypes(const std::string& name,
+  void SetVarDataTypes(const std::string& name UNUSED,
                        const std::vector<framework::proto::VarType::Type>&
-                           multiple_data_type) override {
+                           multiple_data_type UNUSED) override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "SetVarDataTypes is not supported in runtime InferVarType"));
   }
 
-  std::vector<int64_t> GetVarShape(const std::string& name) const override {
+  std::vector<int64_t> GetVarShape(
+      const std::string& name UNUSED) const override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "Do not handle Shape in runtime InferVarType"));
   }
 
-  void SetVarShape(const std::string& name,
-                   const std::vector<int64_t>& dims) override {
+  void SetVarShape(const std::string& name UNUSED,
+                   const std::vector<int64_t>& dims UNUSED) override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "Do not handle Shape in runtime InferVarType"));
   }
 
-  int32_t GetVarLoDLevel(const std::string& name) const override {
+  int32_t GetVarLoDLevel(const std::string& name UNUSED) const override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "Do not handle LoDLevel in runtime InferVarType"));
   }
 
-  void SetVarLoDLevel(const std::string& name, int32_t lod_level) override {
+  void SetVarLoDLevel(const std::string& name UNUSED,
+                      int32_t lod_level UNUSED) override {
     PADDLE_THROW(platform::errors::PermissionDenied(
         "Do not handle LoDLevel in runtime InferVarType"));
   }
diff --git a/paddle/fluid/operators/collective/c_concat_op.h b/paddle/fluid/operators/collective/c_concat_op.h
index d9fe5c861a4da4fe7eb5ea235f25ec60ed4bed02..39bdc4c2740de8090125080ef5bf027d2e35ce6a 100644
--- a/paddle/fluid/operators/collective/c_concat_op.h
+++ b/paddle/fluid/operators/collective/c_concat_op.h
@@ -28,7 +28,7 @@ namespace operators {
 template <typename T, typename DeviceContext>
 class CConcatOpCPUKernel : public framework::OpKernel<T> {
  public:
-  void Compute(const framework::ExecutionContext& ctx) const override {
+  void Compute(const framework::ExecutionContext& ctx UNUSED) const override {
     PADDLE_THROW(platform::errors::Unavailable(
         "Do not support c_concat for cpu kernel now."));
   }
diff --git a/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op.h b/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op.h
index c8311bdab813cbcf57f248bf6663e60ce8625dd5..9b6a2c86897cbecc7cead39b10585e0a43a73cfb 100644
--- a/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op.h
+++ b/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op.h
@@ -32,7 +32,7 @@ namespace operators {
 template <typename T, typename DeviceContext>
 class CSoftmaxWithCrossEntropyOpCPUKernel : public framework::OpKernel<T> {
  public:
-  void Compute(const framework::ExecutionContext& ctx) const override {
+  void Compute(const framework::ExecutionContext& ctx UNUSED) const override {
     PADDLE_THROW(platform::errors::Unavailable(
         "Do not support c_embedding for cpu kernel now."));
   }
diff --git a/paddle/fluid/operators/controlflow/feed_op.cc b/paddle/fluid/operators/controlflow/feed_op.cc
index 698ca4e02e6290f3003dfe615f9f78d6633b086a..c2deeb4190986ce2310ec9a4fd529bc55e6c764b 100644
--- a/paddle/fluid/operators/controlflow/feed_op.cc
+++ b/paddle/fluid/operators/controlflow/feed_op.cc
@@ -97,7 +97,7 @@ void FeedSparseCooTensorKernel(const Context& dev_ctx,
 }
 
 template <typename Context>
-void FeedStringsKernel(const Context& dev_ctx,
+void FeedStringsKernel(const Context& dev_ctx UNUSED,
                        const phi::ExtendedTensor& x,
                        int col,
                        phi::ExtendedTensor* out) {
diff --git a/paddle/fluid/operators/elementwise/elementwise_op.h b/paddle/fluid/operators/elementwise/elementwise_op.h
index 4ef896ff01b87e414c3b687165a433099ab73e0c..2f8db0e737a21a8c2bddeb4c1d809b1a6d4d3b79 100644
--- a/paddle/fluid/operators/elementwise/elementwise_op.h
+++ b/paddle/fluid/operators/elementwise/elementwise_op.h
@@ -159,7 +159,7 @@ class ElementwiseOp : public framework::OperatorWithKernel {
   }
 
   phi::KernelKey GetKernelTypeForVar(
-      const std::string &var_name,
+      const std::string &var_name UNUSED,
       const phi::DenseTensor &tensor,
       const phi::KernelKey &expected_kernel_type) const override {
     if (framework::IsComplexType(expected_kernel_type.dtype())) {
@@ -305,7 +305,7 @@ class ElementwiseOpGrad : public framework::OperatorWithKernel {
   }
 
   phi::KernelKey GetKernelTypeForVar(
-      const std::string &var_name,
+      const std::string &var_name UNUSED,
       const phi::DenseTensor &tensor,
       const phi::KernelKey &expected_kernel_type) const override {
     if (framework::IsComplexType(expected_kernel_type.dtype())) {
@@ -346,7 +346,7 @@ class ElementwiseOpDoubleGrad : public framework::OperatorWithKernel {
   }
 
   phi::KernelKey GetKernelTypeForVar(
-      const std::string &var_name,
+      const std::string &var_name UNUSED,
       const phi::DenseTensor &tensor,
       const phi::KernelKey &expected_kernel_type) const override {
     if (framework::IsComplexType(expected_kernel_type.dtype())) {
@@ -394,7 +394,7 @@ class ElementwiseOpDoubleGradWithoutDXDY
   }
 
   phi::KernelKey GetKernelTypeForVar(
-      const std::string &var_name,
+      const std::string &var_name UNUSED,
       const phi::DenseTensor &tensor,
       const phi::KernelKey &expected_kernel_type) const override {
     if (framework::IsComplexType(expected_kernel_type.dtype())) {
@@ -442,7 +442,7 @@ class ElementwiseOpTripleGrad : public framework::OperatorWithKernel {
   }
 
   phi::KernelKey GetKernelTypeForVar(
-      const std::string &var_name,
+      const std::string &var_name UNUSED,
       const phi::DenseTensor &tensor,
       const phi::KernelKey &expected_kernel_type) const override {
     if (framework::IsComplexType(expected_kernel_type.dtype())) {
diff --git a/paddle/fluid/operators/fused/mkldnn/multi_gru_mkldnn_op.cc b/paddle/fluid/operators/fused/mkldnn/multi_gru_mkldnn_op.cc
index dba3087f1a01bf73de9307cae30594384ae2c755..ff6613d8ae94f8beab6d5afef8ece7ff9b96bc55 100644
--- a/paddle/fluid/operators/fused/mkldnn/multi_gru_mkldnn_op.cc
+++ b/paddle/fluid/operators/fused/mkldnn/multi_gru_mkldnn_op.cc
@@ -598,7 +598,7 @@ class MultiGRUHandler {
   }
 
   template <typename Tout>
-  void reorderOutput(std::shared_ptr<dnnl::memory> mem, int layer) {
+  void reorderOutput(std::shared_ptr<dnnl::memory> mem, int layer UNUSED) {
     auto* data = mem->get_data_handle();
     auto* hidden_data =
         phi::funcs::to_void_cast(hidden_->mutable_data<Tout>(place_));
diff --git a/paddle/fluid/platform/device_context.cc b/paddle/fluid/platform/device_context.cc
index e14ba8b1710cf3cf0a8a5164de28ee355a031284..456abd55ef68fe339df94be97fade11dc81b9353 100644
--- a/paddle/fluid/platform/device_context.cc
+++ b/paddle/fluid/platform/device_context.cc
@@ -57,7 +57,8 @@ DeviceType Place2DeviceType(const platform::Place& place) {
 template <typename DevCtx>
 typename std::enable_if<!std::is_same<DevCtx, phi::GPUContext>::value,
                         DevCtx*>::type
-ConstructDevCtx(const phi::Place& p, /*unused*/ int stream_priority = 0) {
+ConstructDevCtx(const phi::Place& p,
+                /*unused*/ int stream_priority UNUSED = 0) {
   return new DevCtx(p);
 }
 
diff --git a/paddle/phi/kernels/cpu/margin_cross_entropy_kernel.cc b/paddle/phi/kernels/cpu/margin_cross_entropy_kernel.cc
index 06d74471dd992244b6ee0b9d3b6572073c3cdc32..b967d7232254c586dbfc7c0079d297220ca29b35 100644
--- a/paddle/phi/kernels/cpu/margin_cross_entropy_kernel.cc
+++ b/paddle/phi/kernels/cpu/margin_cross_entropy_kernel.cc
@@ -21,19 +21,19 @@
 namespace phi {
 
 template <typename T, typename Context>
-void MarginCrossEntropyKernel(const Context& dev_ctx,
-                              const DenseTensor& logits,
-                              const DenseTensor& labels,
-                              bool return_softmax,
-                              int ring_id,
-                              int rank,
-                              int nranks,
-                              float margin1,
-                              float margin2,
-                              float margin3,
-                              float scale,
-                              DenseTensor* softmax,
-                              DenseTensor* loss) {
+void MarginCrossEntropyKernel(const Context& dev_ctx UNUSED,
+                              const DenseTensor& logits UNUSED,
+                              const DenseTensor& labels UNUSED,
+                              bool return_softmax UNUSED,
+                              int ring_id UNUSED,
+                              int rank UNUSED,
+                              int nranks UNUSED,
+                              float margin1 UNUSED,
+                              float margin2 UNUSED,
+                              float margin3 UNUSED,
+                              float scale UNUSED,
+                              DenseTensor* softmax UNUSED,
+                              DenseTensor* loss UNUSED) {
   PADDLE_THROW(
       errors::Unavailable("Do not support margin_cross_entropy for cpu kernel "
                           "now."));
diff --git a/paddle/phi/kernels/cpu/rnn_functor.h b/paddle/phi/kernels/cpu/rnn_functor.h
index 49b0c7655c6df21f5af4c45f4523bd42237ce25e..511021c00860a6ac3d73625bf5d1fd45594c6617 100644
--- a/paddle/phi/kernels/cpu/rnn_functor.h
+++ b/paddle/phi/kernels/cpu/rnn_functor.h
@@ -79,7 +79,7 @@ void CreateMaskMatrix(const CPUContext& dev_ctx,
 template <typename TensorType>
 void ResetParameterVector(const std::vector<TensorType>& raw_params_vec,
                           int num_layers,
-                          int gate_num,
+                          int gate_num UNUSED,
                           bool is_bidirec,
                           std::vector<std::vector<DenseTensor>>* params_vec) {
   // the parameter raw seuquence is [FWhi, FWhh, BWhi, BWhh] * num_layers
diff --git a/paddle/phi/kernels/cpu/sgd_kernel.cc b/paddle/phi/kernels/cpu/sgd_kernel.cc
index e63e2f2637bdcb69ddb32dc696e7538111ffd026..30a3a13a50d695ad87e1fefd6a913ecf47de5ea3 100644
--- a/paddle/phi/kernels/cpu/sgd_kernel.cc
+++ b/paddle/phi/kernels/cpu/sgd_kernel.cc
@@ -117,10 +117,10 @@ void SGDDenseKernel(const Context& dev_ctx,
                     const DenseTensor& param,
                     const DenseTensor& learning_rate,
                     const DenseTensor& grad,
-                    const paddle::optional<DenseTensor>& master_param,
-                    bool multi_precision,
+                    const paddle::optional<DenseTensor>& master_param UNUSED,
+                    bool multi_precision UNUSED,
                     DenseTensor* param_out,
-                    DenseTensor* master_param_out) {
+                    DenseTensor* master_param_out UNUSED) {
   dev_ctx.template Alloc<T>(param_out);
   sgd_dense_param_dense_grad_impl<T>(param, learning_rate, grad, param_out);
 }
@@ -131,24 +131,24 @@ void SGDDenseParamSparseGradKernel(
     const DenseTensor& param,
     const DenseTensor& learning_rate,
     const SelectedRows& grad,
-    const paddle::optional<DenseTensor>& master_param,
-    bool multi_precision,
+    const paddle::optional<DenseTensor>& master_param UNUSED,
+    bool multi_precision UNUSED,
     DenseTensor* param_out,
-    DenseTensor* master_param_out) {
+    DenseTensor* master_param_out UNUSED) {
   dev_ctx.template Alloc<T>(param_out);
   sgd_dense_param_sparse_grad_impl<T>(param, learning_rate, grad, param_out);
 }
 
 template <typename T, typename Context>
 void SGDSparseParamSparseGradKernel(
-    const Context& dev_ctx,
+    const Context& dev_ctx UNUSED,
     const SelectedRows& param,
     const DenseTensor& learning_rate,
     const SelectedRows& grad,
-    const paddle::optional<SelectedRows>& master_param,
-    bool multi_precision,
+    const paddle::optional<SelectedRows>& master_param UNUSED,
+    bool multi_precision UNUSED,
     SelectedRows* param_out,
-    SelectedRows* master_param_out) {
+    SelectedRows* master_param_out UNUSED) {
   // for distributed training, a sparse var may be empty,
   // just skip updating.
   if (grad.rows().size() == 0) {
diff --git a/paddle/phi/kernels/cpu/uniform_kernel.cc b/paddle/phi/kernels/cpu/uniform_kernel.cc
index 17ee5459188249cca2416d5737740715f3706dc1..d850dc5074e033a0ed645b4c54c080d7ef8e3144 100644
--- a/paddle/phi/kernels/cpu/uniform_kernel.cc
+++ b/paddle/phi/kernels/cpu/uniform_kernel.cc
@@ -22,7 +22,7 @@ namespace phi {
 template <typename T, typename Context>
 void UniformKernel(const Context &dev_ctx,
                    const IntArray &shape,
-                   DataType dtype,
+                   DataType dtype UNUSED,
                    const Scalar &min,
                    const Scalar &max,
                    int seed,
diff --git a/paddle/phi/kernels/funcs/activation_functor.h b/paddle/phi/kernels/funcs/activation_functor.h
index df63c22c9460023a0055aed61fad10df9d99a291..b73ca44f7cacee303c791f8cc4c95562cf69e4a5 100644
--- a/paddle/phi/kernels/funcs/activation_functor.h
+++ b/paddle/phi/kernels/funcs/activation_functor.h
@@ -104,7 +104,7 @@ struct SinGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout * x.unaryExpr(Cosine<T>());
   }
 
@@ -277,7 +277,7 @@ struct ReciprocalGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x UNUSED, Out out, dOut dout, dX dx) const {
     dx.device(d) = dout * static_cast<T>(-1) * out * out;
   }
 
@@ -310,7 +310,7 @@ struct CosGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = -dout * x.unaryExpr(Sine<T>());
   }
 
@@ -505,7 +505,7 @@ struct MishGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     auto sp = (x > static_cast<T>(threshold))
                   .select(x, (static_cast<T>(1) + x.exp()).log());
     auto gsp = static_cast<T>(1) - (-sp).exp();
@@ -544,7 +544,7 @@ struct STanhGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     auto a = static_cast<T>(scale_a);
     auto b = static_cast<T>(scale_b);
     auto temp = (a * x).tanh() * (a * x).tanh();
@@ -574,7 +574,7 @@ struct TanGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout / x.unaryExpr(Cosine<T>()).square();
   }
 
@@ -620,7 +620,7 @@ struct SqrtGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x UNUSED, Out out, dOut dout, dX dx) const {
     dx.device(d) = static_cast<T>(0.5) * dout / out;
   }
 
@@ -645,7 +645,7 @@ struct RsqrtGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x UNUSED, Out out, dOut dout, dX dx) const {
     dx.device(d) = static_cast<T>(-0.5) * dout * out * out * out;
   }
 
@@ -697,7 +697,7 @@ struct SoftplusGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     auto x_beta = static_cast<T>(beta) * x;
     dx.device(d) =
         (x_beta > static_cast<T>(threshold))
@@ -816,7 +816,7 @@ struct SinhGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout * x.unaryExpr(Cosh<T>());
   }
 
@@ -831,7 +831,7 @@ struct CoshGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout * x.unaryExpr(Sinh<T>());
   }
 
@@ -867,7 +867,7 @@ struct AcosGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) =
         -dout * static_cast<T>(1) / (static_cast<T>(1) - x.square()).sqrt();
   }
@@ -904,7 +904,7 @@ struct AsinGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) =
         dout * static_cast<T>(1) / (static_cast<T>(1) - x.square()).sqrt();
   }
@@ -941,7 +941,7 @@ struct AtanGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout * static_cast<T>(1) / (static_cast<T>(1) + x.square());
   }
 
@@ -989,7 +989,7 @@ struct AcoshGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) =
         dout * static_cast<T>(1) / (x * x - static_cast<T>(1)).sqrt();
   }
@@ -1026,7 +1026,7 @@ struct AsinhGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) =
         dout * static_cast<T>(1) / (x.square() + static_cast<T>(1)).sqrt();
   }
@@ -1063,7 +1063,7 @@ struct AtanhGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout * static_cast<T>(1) / (static_cast<T>(1) - x.square());
   }
 
@@ -1160,12 +1160,12 @@ template <typename T>
 struct ReluGradGradFunctor : public BaseActivationFunctor<T> {
   template <typename Device>
   void operator()(const Device& dev,
-                  const DenseTensor* X,
+                  const DenseTensor* X UNUSED,
                   const DenseTensor* Out,
                   const DenseTensor* ddX,
                   DenseTensor* ddOut,
-                  DenseTensor* dOut,
-                  DenseTensor* dX) const {
+                  DenseTensor* dOut UNUSED,
+                  DenseTensor* dX UNUSED) const {
     auto* d = dev.eigen_device();
     auto ddx = EigenVector<T>::Flatten(
         GET_DATA_SAFELY(ddX, "Input", "DDX", "ReluGradGrad"));
@@ -1375,7 +1375,7 @@ struct HardTanhGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) =
         dout * ((x > static_cast<T>(t_min)) * (x < static_cast<T>(t_max)))
                    .template cast<T>();
@@ -1431,11 +1431,11 @@ struct LeakyReluGradGradFunctor : public BaseActivationFunctor<T> {
   template <typename Device>
   void operator()(const Device& dev,
                   const DenseTensor* X,
-                  const DenseTensor* Out,
+                  const DenseTensor* Out UNUSED,
                   const DenseTensor* ddX,
                   DenseTensor* ddOut,
-                  DenseTensor* dOut,
-                  DenseTensor* dX) const {
+                  DenseTensor* dOut UNUSED,
+                  DenseTensor* dX UNUSED) const {
     if (ddOut) {
       auto* d = dev.eigen_device();
       auto ddx = EigenVector<T>::Flatten(
@@ -1479,7 +1479,7 @@ struct ThresholdedReluGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     auto th = static_cast<T>(threshold);
     dx.device(d) = dout * (x > th).template cast<T>();
   }
@@ -1511,7 +1511,7 @@ struct Relu6GradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x UNUSED, Out out, dOut dout, dX dx) const {
     float threshold = 6;
     dx.device(d) =
         dout * ((out > static_cast<T>(0)) * (out < static_cast<T>(threshold)))
@@ -1540,7 +1540,7 @@ struct TanhShrinkGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout * (x.tanh() * x.tanh());
   }
 
@@ -1577,7 +1577,7 @@ struct HardShrinkGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     auto temp1 = x < static_cast<T>(threshold * -1.f);
     auto temp2 = x > static_cast<T>(threshold);
     dx.device(d) = dout * (temp1 || temp2).template cast<T>();
@@ -1615,7 +1615,7 @@ struct SoftShrinkGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     auto lambdaT = static_cast<T>(lambda);
     auto temp1 = (x > lambdaT).template cast<T>();
     auto temp2 = (x < -lambdaT).template cast<T>();
@@ -1673,7 +1673,7 @@ struct ELUGradNegativeAlphaFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     // case 2: alpha < 0
     // dx = dout, if x > 0
     // dx = dout * (out + alpha), if x <=0
@@ -1742,7 +1742,7 @@ struct SiluGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     auto temp1 = static_cast<T>(1) + (-x).exp();  // 1+e^(-x)
     auto temp2 = x * (-x).exp();                  // x*e^(-x)
     dx.device(d) = dout * ((static_cast<T>(1) / temp1) *
@@ -1770,7 +1770,7 @@ struct SoftsignGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) =
         dout * (static_cast<T>(1) / (static_cast<T>(1) + x.abs()).square());
   }
@@ -1947,7 +1947,7 @@ struct LogSigmoidGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     auto temp = (-x).cwiseMax(static_cast<T>(0));  // temp = max(-x, 0)
     dx.device(d) =
         dout * ((-x - temp).exp() / ((-temp).exp() + (-x - temp).exp()));
@@ -1984,7 +1984,7 @@ struct HardSigmoidGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x UNUSED, Out out, dOut dout, dX dx) const {
     dx.device(d) = dout *
                    ((out > static_cast<T>(0)) * (out < static_cast<T>(1)))
                        .template cast<T>() *
@@ -2012,7 +2012,7 @@ struct LogGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout * (static_cast<T>(1) / x);
   }
 
@@ -2036,7 +2036,7 @@ struct Log2GradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout * static_cast<T>(1) / (x * static_cast<T>(log(2)));
   }
 
@@ -2060,7 +2060,7 @@ struct Log10GradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout * static_cast<T>(1) / (x * static_cast<T>(log(10)));
   }
 
@@ -2083,7 +2083,7 @@ struct Log1pGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout * (static_cast<T>(1) / (x + static_cast<T>(1)));
   }
 
@@ -2157,7 +2157,7 @@ struct HardSwishGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     auto tmp = ((x + static_cast<T>(offset)) < static_cast<T>(threshold))
                    .template cast<T>();
     dx.device(d) =
@@ -2193,7 +2193,7 @@ struct SwishGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out fake_out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out fake_out UNUSED, dOut dout, dX dx) const {
     float beta = 1.0;
     auto temp1 = static_cast<T>(1) /
                  (static_cast<T>(1) + (static_cast<T>(-beta) * x).exp());
@@ -2229,7 +2229,7 @@ struct PowGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     dx.device(d) = dout * static_cast<T>(factor) *
                    x.pow(static_cast<T>(factor) - static_cast<T>(1));
   }
@@ -2279,7 +2279,8 @@ struct ZeroGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(
+      Device d, X x UNUSED, Out out, dOut dout UNUSED, dX dx) const {
     dx.device(d) = static_cast<T>(0) * out;
   }
 
@@ -2384,7 +2385,7 @@ struct CELUGradFunctor : public BaseActivationFunctor<T> {
             typename Out,
             typename dOut,
             typename dX>
-  void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
+  void operator()(Device d, X x, Out out UNUSED, dOut dout, dX dx) const {
     auto temp_a_pos = static_cast<T>(alpha > 0);
     auto temp_a_neg = static_cast<T>(alpha <= 0);
     auto temp_x_pos = (x > static_cast<T>(0)).template cast<T>();
diff --git a/paddle/phi/kernels/funcs/detection/poly_util.cc b/paddle/phi/kernels/funcs/detection/poly_util.cc
deleted file mode 100644
index fd8037a82eabcdb1756122f38cb52f030280e083..0000000000000000000000000000000000000000
--- a/paddle/phi/kernels/funcs/detection/poly_util.cc
+++ /dev/null
@@ -1,138 +0,0 @@
-/* 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. */
-#ifndef POLY_UTIL_CC_
-#define POLY_UTIL_CC_
-
-#include "paddle/phi/kernels/funcs/detection/poly_util.h"
-
-namespace phi {
-namespace funcs {
-
-using phi::funcs::gpc_free_polygon;
-using phi::funcs::gpc_polygon_clip;
-
-template <class T>
-void Array2PointVec(const T* box,
-                    const size_t box_size,
-                    std::vector<Point_<T>>* vec) {
-  size_t pts_num = box_size / 2;
-  (*vec).resize(pts_num);
-  for (size_t i = 0; i < pts_num; i++) {
-    (*vec).at(i).x = box[2 * i];
-    (*vec).at(i).y = box[2 * i + 1];
-  }
-}
-
-template <class T>
-void Array2Poly(const T* box,
-                const size_t box_size,
-                phi::funcs::gpc_polygon* poly) {
-  size_t pts_num = box_size / 2;
-  (*poly).num_contours = 1;
-  (*poly).hole = reinterpret_cast<int*>(malloc(sizeof(int)));
-  (*poly).hole[0] = 0;
-  (*poly).contour =
-      (phi::funcs::gpc_vertex_list*)malloc(sizeof(phi::funcs::gpc_vertex_list));
-  (*poly).contour->num_vertices = pts_num;
-  (*poly).contour->vertex =
-      (phi::funcs::gpc_vertex*)malloc(sizeof(phi::funcs::gpc_vertex) * pts_num);
-  for (size_t i = 0; i < pts_num; ++i) {
-    (*poly).contour->vertex[i].x = box[2 * i];
-    (*poly).contour->vertex[i].y = box[2 * i + 1];
-  }
-}
-
-template <class T>
-void PointVec2Poly(const std::vector<Point_<T>>& vec,
-                   phi::funcs::gpc_polygon* poly) {
-  int pts_num = vec.size();
-  (*poly).num_contours = 1;
-  (*poly).hole = reinterpret_cast<int*>(malloc(sizeof(int)));
-  (*poly).hole[0] = 0;
-  (*poly).contour =
-      (phi::funcs::gpc_vertex_list*)malloc(sizeof(phi::funcs::gpc_vertex_list));
-  (*poly).contour->num_vertices = pts_num;
-  (*poly).contour->vertex =
-      (phi::funcs::gpc_vertex*)malloc(sizeof(phi::funcs::gpc_vertex) * pts_num);
-  for (size_t i = 0; i < pts_num; ++i) {
-    (*poly).contour->vertex[i].x = vec[i].x;
-    (*poly).contour->vertex[i].y = vec[i].y;
-  }
-}
-
-template <class T>
-void Poly2PointVec(const phi::funcs::gpc_vertex_list& contour,
-                   std::vector<Point_<T>>* vec) {
-  int pts_num = contour.num_vertices;
-  (*vec).resize(pts_num);
-  for (int i = 0; i < pts_num; i++) {
-    (*vec).at(i).x = contour.vertex[i].x;
-    (*vec).at(i).y = contour.vertex[i].y;
-  }
-}
-
-template <class T>
-T GetContourArea(const std::vector<Point_<T>>& vec) {
-  size_t pts_num = vec.size();
-  if (pts_num < 3) return T(0.);
-  T area = T(0.);
-  for (size_t i = 0; i < pts_num; ++i) {
-    area += vec[i].x * vec[(i + 1) % pts_num].y -
-            vec[i].y * vec[(i + 1) % pts_num].x;
-  }
-  return std::fabs(area / 2.0);
-}
-
-template <class T>
-T PolyArea(const T* box, const size_t box_size, const bool normalized) {
-  // If coordinate values are is invalid
-  // if area size <= 0,  return 0.
-  std::vector<Point_<T>> vec;
-  Array2PointVec<T>(box, box_size, &vec);
-  return GetContourArea<T>(vec);
-}
-
-template <class T>
-T PolyOverlapArea(const T* box1,
-                  const T* box2,
-                  const size_t box_size,
-                  const bool normalized) {
-  phi::funcs::gpc_polygon poly1;
-  phi::funcs::gpc_polygon poly2;
-  Array2Poly<T>(box1, box_size, &poly1);
-  Array2Poly<T>(box2, box_size, &poly2);
-  phi::funcs::gpc_polygon respoly;
-  phi::funcs::gpc_op op = phi::funcs::GPC_INT;
-  phi::funcs::gpc_polygon_clip(op, &poly2, &poly1, &respoly);
-
-  T inter_area = T(0.);
-  int contour_num = respoly.num_contours;
-  for (int i = 0; i < contour_num; ++i) {
-    std::vector<Point_<T>> resvec;
-    Poly2PointVec<T>(respoly.contour[i], &resvec);
-    // inter_area += std::fabs(cv::contourArea(resvec)) + 0.5f *
-    // (cv::arcLength(resvec, true));
-    inter_area += GetContourArea<T>(resvec);
-  }
-
-  phi::funcs::gpc_free_polygon(&poly1);
-  phi::funcs::gpc_free_polygon(&poly2);
-  phi::funcs::gpc_free_polygon(&respoly);
-  return inter_area;
-}
-
-}  // namespace funcs
-}  // namespace phi
-
-#endif
diff --git a/paddle/phi/kernels/funcs/detection/poly_util.h b/paddle/phi/kernels/funcs/detection/poly_util.h
index 6d527d2d95f9eb65cbea4c385403716e10d24908..608f373f3d6a3dbcfa49d3acedc03e6753463a53 100644
--- a/paddle/phi/kernels/funcs/detection/poly_util.h
+++ b/paddle/phi/kernels/funcs/detection/poly_util.h
@@ -13,6 +13,9 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 #pragma once
 
+#ifndef POLY_UTIL_H_
+#define POLY_UTIL_H_
+
 #include <vector>
 
 #include "paddle/phi/kernels/funcs/gpc.h"
@@ -20,6 +23,9 @@ limitations under the License. */
 namespace phi {
 namespace funcs {
 
+using phi::funcs::gpc_free_polygon;
+using phi::funcs::gpc_polygon_clip;
+
 template <class T>
 class Point_ {
  public:
@@ -43,33 +49,114 @@ class Point_ {
 template <class T>
 void Array2PointVec(const T* box,
                     const size_t box_size,
-                    std::vector<Point_<T>>* vec);
+                    std::vector<Point_<T>>* vec) {
+  size_t pts_num = box_size / 2;
+  (*vec).resize(pts_num);
+  for (size_t i = 0; i < pts_num; i++) {
+    (*vec).at(i).x = box[2 * i];
+    (*vec).at(i).y = box[2 * i + 1];
+  }
+}
 
 template <class T>
 void Array2Poly(const T* box,
                 const size_t box_size,
-                phi::funcs::gpc_polygon* poly);
+                phi::funcs::gpc_polygon* poly) {
+  size_t pts_num = box_size / 2;
+  (*poly).num_contours = 1;
+  (*poly).hole = reinterpret_cast<int*>(malloc(sizeof(int)));
+  (*poly).hole[0] = 0;
+  (*poly).contour =
+      (phi::funcs::gpc_vertex_list*)malloc(sizeof(phi::funcs::gpc_vertex_list));
+  (*poly).contour->num_vertices = pts_num;
+  (*poly).contour->vertex =
+      (phi::funcs::gpc_vertex*)malloc(sizeof(phi::funcs::gpc_vertex) * pts_num);
+  for (size_t i = 0; i < pts_num; ++i) {
+    (*poly).contour->vertex[i].x = box[2 * i];
+    (*poly).contour->vertex[i].y = box[2 * i + 1];
+  }
+}
 
 template <class T>
 void PointVec2Poly(const std::vector<Point_<T>>& vec,
-                   phi::funcs::gpc_polygon* poly);
+                   phi::funcs::gpc_polygon* poly) {
+  int pts_num = vec.size();
+  (*poly).num_contours = 1;
+  (*poly).hole = reinterpret_cast<int*>(malloc(sizeof(int)));
+  (*poly).hole[0] = 0;
+  (*poly).contour =
+      (phi::funcs::gpc_vertex_list*)malloc(sizeof(phi::funcs::gpc_vertex_list));
+  (*poly).contour->num_vertices = pts_num;
+  (*poly).contour->vertex =
+      (phi::funcs::gpc_vertex*)malloc(sizeof(phi::funcs::gpc_vertex) * pts_num);
+  for (size_t i = 0; i < pts_num; ++i) {
+    (*poly).contour->vertex[i].x = vec[i].x;
+    (*poly).contour->vertex[i].y = vec[i].y;
+  }
+}
 
 template <class T>
 void Poly2PointVec(const phi::funcs::gpc_vertex_list& contour,
-                   std::vector<Point_<T>>* vec);
+                   std::vector<Point_<T>>* vec) {
+  int pts_num = contour.num_vertices;
+  (*vec).resize(pts_num);
+  for (int i = 0; i < pts_num; i++) {
+    (*vec).at(i).x = contour.vertex[i].x;
+    (*vec).at(i).y = contour.vertex[i].y;
+  }
+}
 
 template <class T>
-T GetContourArea(const std::vector<Point_<T>>& vec);
+T GetContourArea(const std::vector<Point_<T>>& vec) {
+  size_t pts_num = vec.size();
+  if (pts_num < 3) return T(0.);
+  T area = T(0.);
+  for (size_t i = 0; i < pts_num; ++i) {
+    area += vec[i].x * vec[(i + 1) % pts_num].y -
+            vec[i].y * vec[(i + 1) % pts_num].x;
+  }
+  return std::fabs(area / 2.0);
+}
 
 template <class T>
-T PolyArea(const T* box, const size_t box_size, const bool normalized);
+T PolyArea(const T* box, const size_t box_size, const bool normalized UNUSED) {
+  // If coordinate values are is invalid
+  // if area size <= 0,  return 0.
+  std::vector<Point_<T>> vec;
+  Array2PointVec<T>(box, box_size, &vec);
+  return GetContourArea<T>(vec);
+}
 
 template <class T>
 T PolyOverlapArea(const T* box1,
                   const T* box2,
                   const size_t box_size,
-                  const bool normalized);
+                  const bool normalized UNUSED) {
+  phi::funcs::gpc_polygon poly1;
+  phi::funcs::gpc_polygon poly2;
+  Array2Poly<T>(box1, box_size, &poly1);
+  Array2Poly<T>(box2, box_size, &poly2);
+  phi::funcs::gpc_polygon respoly;
+  phi::funcs::gpc_op op = phi::funcs::GPC_INT;
+  phi::funcs::gpc_polygon_clip(op, &poly2, &poly1, &respoly);
+
+  T inter_area = T(0.);
+  int contour_num = respoly.num_contours;
+  for (int i = 0; i < contour_num; ++i) {
+    std::vector<Point_<T>> resvec;
+    Poly2PointVec<T>(respoly.contour[i], &resvec);
+    // inter_area += std::fabs(cv::contourArea(resvec)) + 0.5f *
+    // (cv::arcLength(resvec, true));
+    inter_area += GetContourArea<T>(resvec);
+  }
+
+  phi::funcs::gpc_free_polygon(&poly1);
+  phi::funcs::gpc_free_polygon(&poly2);
+  phi::funcs::gpc_free_polygon(&respoly);
+  return inter_area;
+}
+
 }  // namespace funcs
 }  // namespace phi
 
-#include "paddle/phi/kernels/funcs/detection/poly_util.cc"
+#endif
diff --git a/paddle/phi/kernels/legacy/cpu/uniform_kernel.cc b/paddle/phi/kernels/legacy/cpu/uniform_kernel.cc
index ecea86874a7a283c07443f1c80adb11800209c16..3aa697b2409ee9da9de8a84278a390eccc17f537 100644
--- a/paddle/phi/kernels/legacy/cpu/uniform_kernel.cc
+++ b/paddle/phi/kernels/legacy/cpu/uniform_kernel.cc
@@ -20,7 +20,7 @@ namespace phi {
 template <typename T, typename Context>
 void UniformRawKernel(const Context &dev_ctx,
                       const IntArray &shape,
-                      DataType dtype,
+                      DataType dtype UNUSED,
                       const Scalar &min,
                       const Scalar &max,
                       int seed,