diff --git a/paddle/framework/tensor_util.h b/paddle/framework/tensor_util.h
index 5b474e4aef6d8ef714a59052abd55bc439605f6d..ebfb0e553877b776afe13d8a4e7a7ffa8710405e 100644
--- a/paddle/framework/tensor_util.h
+++ b/paddle/framework/tensor_util.h
@@ -82,6 +82,28 @@ inline void CopyFrom(const Tensor& src, const platform::Place& dst_place,
 #endif
 }
 
+/**
+ * @brief CopyFrom support CPU <-> CPU
+ */
+inline void CopyFrom(const Tensor& src, const platform::Place& dst_place,
+                     Tensor* dst) {
+  src.check_memory_size();
+  dst->Resize(src.dims());
+
+  auto src_place = src.place();
+  auto src_ptr = src.data<void>();
+
+  auto dst_ptr = dst->mutable_data(dst_place, src.type());
+
+  auto size = src.numel() * SizeOfType(src.type());
+
+  PADDLE_ENFORCE(platform::is_cpu_place(src_place) &&
+                 platform::is_cpu_place(dst_place));
+
+  memory::Copy(boost::get<platform::CPUPlace>(dst_place), dst_ptr,
+               boost::get<platform::CPUPlace>(src_place), src_ptr, size);
+}
+
 /**
  * @brief   Copy the content of an external vector to a tensor.
  *
@@ -115,6 +137,21 @@ inline void CopyFromVector(const std::vector<T>& src,
 #endif
 }
 
+/**
+ * @brief CopyFromVector CPU vector -> CPU Tensor
+ */
+template <typename T>
+inline void CopyFromVector(const std::vector<T>& src, Tensor* dst) {
+  platform::CPUPlace dst_place = platform::CPUPlace();
+  auto src_ptr = static_cast<const void*>(src.data());
+  platform::CPUPlace src_place;
+  dst->Resize({static_cast<int64_t>(src.size())});
+  auto dst_ptr = static_cast<void*>(dst->mutable_data<T>(dst_place));
+  auto size = src.size() * sizeof(T);
+
+  memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size);
+}
+
 /**
  * @brief   Copy the content of a tensor to a vector
  *
@@ -148,5 +185,23 @@ inline void CopyToVector(const Tensor& src, const platform::DeviceContext& ctx,
 #endif
 }
 
+/**
+ * @brief CopyToVector CPUTensor <-> CPU Vector
+ */
+template <typename T>
+inline void CopyToVector(const Tensor& src, std::vector<T>* dst) {
+  auto src_ptr = static_cast<const void*>(src.data<T>());
+  auto size = src.numel() * sizeof(T);
+
+  platform::CPUPlace dst_place;
+  dst->resize(src.numel());
+  auto dst_ptr = static_cast<void*>(dst->data());
+
+  PADDLE_ENFORCE(platform::is_cpu_place(src.place()));
+
+  memory::Copy(dst_place, dst_ptr, boost::get<platform::CPUPlace>(src.place()),
+               src_ptr, size);
+}
+
 }  // namespace framework
 }  // namespace paddle
diff --git a/paddle/framework/tensor_util_test.cc b/paddle/framework/tensor_util_test.cc
index 3afb98a4a52f8cff335833cfb589a87b2d8d2ef2..6fc243aaf6ea68d716e7bb6a73289197fde56247 100644
--- a/paddle/framework/tensor_util_test.cc
+++ b/paddle/framework/tensor_util_test.cc
@@ -17,6 +17,7 @@
 
 namespace paddle {
 namespace framework {
+
 TEST(CopyFrom, Tensor) {
   Tensor src_tensor;
   Tensor dst_tensor;
@@ -29,7 +30,7 @@ TEST(CopyFrom, Tensor) {
   memcpy(src_ptr, arr, 9 * sizeof(int));
 
   auto cpu_place = new platform::CPUPlace();
-  CopyFrom(src_tensor, *cpu_place, cpu_ctx, &dst_tensor);
+  CopyFrom(src_tensor, *cpu_place, &dst_tensor);
 
   const int* dst_ptr = dst_tensor.data<int>();
   ASSERT_NE(src_ptr, dst_ptr);
@@ -104,8 +105,7 @@ TEST(CopyFromVector, Tensor) {
     // Copy to CPU Tensor
     cpu_tensor.Resize(make_ddim({3, 3}));
     auto cpu_place = new paddle::platform::CPUPlace();
-    CPUDeviceContext cpu_ctx(*cpu_place);
-    CopyFromVector<int>(src_vec, cpu_ctx, &cpu_tensor);
+    CopyFromVector<int>(src_vec, &cpu_tensor);
 
     // Compare Tensors
     const int* cpu_ptr = cpu_tensor.data<int>();
@@ -117,7 +117,7 @@ TEST(CopyFromVector, Tensor) {
 
     src_vec.erase(src_vec.begin(), src_vec.begin() + 5);
     cpu_tensor.Resize(make_ddim({2, 2}));
-    CopyFromVector<int>(src_vec, cpu_ctx, &cpu_tensor);
+    CopyFromVector<int>(src_vec, &cpu_tensor);
     cpu_ptr = cpu_tensor.data<int>();
     src_ptr = src_vec.data();
     ASSERT_NE(src_ptr, cpu_ptr);
@@ -198,9 +198,8 @@ TEST(CopyToVector, Tensor) {
     }
 
     CPUPlace place;
-    CPUDeviceContext cpu_ctx(place);
     std::vector<int> dst;
-    CopyToVector<int>(src, cpu_ctx, &dst);
+    CopyToVector<int>(src, &dst);
 
     for (int i = 0; i < 3 * 3; ++i) {
       EXPECT_EQ(src_ptr[i], dst[i]);