diff --git a/cmake/flags.cmake b/cmake/flags.cmake
index 3dffacdacbdb18a2f3fbc670234022dddf360e06..d11094e90fcbf6af71bb35379bd2aef80b3e0e99 100644
--- a/cmake/flags.cmake
+++ b/cmake/flags.cmake
@@ -160,7 +160,7 @@ if(UNIX AND NOT APPLE)
   set(LINUX TRUE)
 endif(UNIX AND NOT APPLE)
 
-set(GPU_COMMON_FLAGS -std=c++11 ${GPU_COMMON_FLAGS})
+set(GPU_COMMON_FLAGS /std:c++14 ${GPU_COMMON_FLAGS})
 foreach(flag ${COMMON_FLAGS})
     safe_set_cflag(CMAKE_C_FLAGS ${flag})
     safe_set_cxxflag(CMAKE_CXX_FLAGS ${flag})
diff --git a/paddle/fluid/inference/api/helper.h b/paddle/fluid/inference/api/helper.h
index 4b64c2dc25213b8d6a3af3c35005b86e3b0c68b4..1c28428a8f6e7bb548c4218189c2d251941784b3 100644
--- a/paddle/fluid/inference/api/helper.h
+++ b/paddle/fluid/inference/api/helper.h
@@ -15,6 +15,7 @@
 #pragma once
 
 #include <algorithm>
+#include <iterator>
 #include <numeric>
 #include <sstream>
 #include <string>
@@ -25,8 +26,8 @@
 namespace paddle {
 namespace inference {
 
-
-static void split(const std::string &str, char sep, std::vector<std::string> *pieces) {
+static void split(const std::string &str, char sep,
+                  std::vector<std::string> *pieces) {
   pieces->clear();
   if (str.empty()) {
     return;
diff --git a/paddle/fluid/operators/cum_op.h b/paddle/fluid/operators/cum_op.h
index 999fdcff90784ed089cd620a4f0a908f196bcdda..51227b4907663ebbd1f1ba5ad7c9242e99dfc43d 100644
--- a/paddle/fluid/operators/cum_op.h
+++ b/paddle/fluid/operators/cum_op.h
@@ -85,14 +85,17 @@ class CumKernel : public framework::OpKernel<typename Functor::ELEMENT_TYPE> {
   template <typename Device, typename Dim, typename X, typename Out>
   void ComputeImp(Device d, const Dim& dims, X x, Out out, int axis,
                   bool reverse, bool exclusive) const {
+    Functor func();
     if (!reverse) {
-      out.reshape(dims).device(d) = Functor()(x.reshape(dims), axis, exclusive);
+      out.reshape(dims).device(d) =
+          func.apply(x.reshape(dims), axis, exclusive);
     } else {
       std::array<bool, Dim::count> rev;
       rev.fill(false);
       rev[axis] = reverse;
       out.reshape(dims).device(d) =
-          Functor()(x.reshape(dims).reverse(rev), axis, exclusive).reverse(rev);
+          func.apply(x.reshape(dims).reverse(rev), axis, exclusive)
+              .reverse(rev);
     }
   }
 };
@@ -101,8 +104,7 @@ template <typename T>
 struct CumsumFunctor {
   using ELEMENT_TYPE = T;
   template <typename X>
-  const typename X::TensorScanSumOp operator()(X x, int axis,
-                                               bool exclusive) const {
+  const typename X::TensorScanSumOp apply(X x, int axis, bool exclusive) const {
     return x.cumsum(axis, exclusive);
   }
 };