diff --git a/paddle/fluid/operators/mean_op.cu b/paddle/fluid/operators/mean_op.cu
index 5a0afb68d63f1b99658fc649c33c5831cfcf35b5..63b5b871aabb5424ae6c8636ed40c0e5aa547205 100644
--- a/paddle/fluid/operators/mean_op.cu
+++ b/paddle/fluid/operators/mean_op.cu
@@ -59,8 +59,7 @@ class MeanCUDAKernel : public framework::OpKernel<T> {
       return;
     }
 
-    using MT = typename details::MPTypeTrait<T>::Type;
-    using Div = kernel_primitives::DivideFunctor<T, MT>;
+    using Div = kernel_primitives::DivideFunctor<T, T>;
     std::vector<int> reduce_dims;
     reduce_dims.reserve(rank);
     for (decltype(rank) i = 0; i < rank; ++i) {
diff --git a/paddle/pten/kernels/gpu/math_kernel.cu b/paddle/pten/kernels/gpu/math_kernel.cu
index 80931db56c3debaab6720b5dd7ec338da8040642..d7a16ac49b1c9d191cfe4c1799bc3de4e72d8529 100644
--- a/paddle/pten/kernels/gpu/math_kernel.cu
+++ b/paddle/pten/kernels/gpu/math_kernel.cu
@@ -52,21 +52,6 @@ namespace pten {
         dev_ctx, inputs, &outputs, axis, funcs::name##Functor<T>()); \
   }
 
-/**
- * Util Functors
- */
-
-template <typename T>
-struct DivideFunctor {
-  HOSTDEVICE explicit inline DivideFunctor(int n)
-      : n_inv(static_cast<T>(1.0 / n)) {}
-
-  HOSTDEVICE inline T operator()(const T x) const { return x * n_inv; }
-
- private:
-  T n_inv;
-};
-
 /**
  * Kernels
  */