diff --git a/paddle/memory/CMakeLists.txt b/paddle/memory/CMakeLists.txt
index 8035d93bfec75b20a54c5af0521ab724cafba8ca..9cc4233e43267472d405c3e4e617f0782e1430ea 100644
--- a/paddle/memory/CMakeLists.txt
+++ b/paddle/memory/CMakeLists.txt
@@ -1,7 +1,7 @@
 add_subdirectory(detail)
 
 cc_library(memory SRCS memory.cc)
-cc_library(memcpy SRCS memcpy.cc DEPS device_context)
+cc_library(memcpy SRCS memcpy.cc)
 
 cc_library(paddle_memory
     DEPS
diff --git a/paddle/memory/memcpy.cc b/paddle/memory/memcpy.cc
index 098931c887479ce6f1afc8b90e4003758d88c018..aaab1142ca18d3319469a4d685fde9d30929113f 100644
--- a/paddle/memory/memcpy.cc
+++ b/paddle/memory/memcpy.cc
@@ -35,7 +35,7 @@ void Copy<platform::CPUPlace, platform::GPUPlace>(platform::CPUPlace dst_place,
                                                   platform::GPUPlace src_place,
                                                   const void* src, size_t num,
                                                   cudaStream_t stream) {
-  platform::GPUPlaceGuard g(src_place.device);
+  platform::SetDeviceId(src_place.device);
   platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyDeviceToHost, stream);
 }
 
@@ -45,7 +45,7 @@ void Copy<platform::GPUPlace, platform::CPUPlace>(platform::GPUPlace dst_place,
                                                   platform::CPUPlace src_place,
                                                   const void* src, size_t num,
                                                   cudaStream_t stream) {
-  platform::GPUPlaceGuard g(dst_place.device);
+  platform::SetDeviceId(dst_place.device);
   platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyHostToDevice, stream);
 }
 
@@ -56,7 +56,7 @@ void Copy<platform::GPUPlace, platform::GPUPlace>(platform::GPUPlace dst_place,
                                                   const void* src, size_t num,
                                                   cudaStream_t stream) {
   if (dst_place == src_place) {
-    platform::GPUPlaceGuard g(src_place.device);
+    platform::SetDeviceId(src_place.device);
     platform::GpuMemcpyAsync(dst, src, num, cudaMemcpyDeviceToDevice, stream);
   } else {
     platform::GpuMemcpyPeer(dst, dst_place.device, src, src_place.device, num,