diff --git a/paddle/fluid/framework/fleet/nccl_wrapper.cc b/paddle/fluid/framework/fleet/nccl_wrapper.cc
index 9a39da6dcae1e788d696c92f863e36f41e59ba4b..0df6aca8b12eefb32aec99efc50c1a678e99e656 100644
--- a/paddle/fluid/framework/fleet/nccl_wrapper.cc
+++ b/paddle/fluid/framework/fleet/nccl_wrapper.cc
@@ -24,9 +24,9 @@ std::shared_ptr<NCCLWrapper> NCCLWrapper::s_instance_ = NULL;
 bool NCCLWrapper::is_initialized_ = false;
 
 void NCCLWrapper::InitNCCL() {
-  platform::dynload::ncclCommInitRank(
+  PADDLE_ENFORCE(platform::dynload::ncclCommInitRank(
       &(nccl_info_.comm_), nccl_info_.global_ranks_, nccl_info_.nccl_id_,
-      nccl_info_.my_global_rank_);
+      nccl_info_.my_global_rank_));
   return;
 }
 
@@ -55,9 +55,9 @@ void NCCLWrapper::SyncVar(const int root_rank, const Scope& scope,
     auto var = scope.FindVar(name);
     LoDTensor* tensor = var->GetMutable<LoDTensor>();
     int32_t total_size = tensor->numel();
-    platform::dynload::ncclBcast(reinterpret_cast<void*>(tensor->data<float>()),
-                                 total_size, ncclFloat, root_rank,
-                                 nccl_info_.comm_, nccl_info_.stream_);
+    PADDLE_ENFORCE(platform::dynload::ncclBcast(
+        reinterpret_cast<void*>(tensor->data<float>()), total_size, ncclFloat,
+        root_rank, nccl_info_.comm_, nccl_info_.stream_));
     cudaStreamSynchronize(nccl_info_.stream_);
   }
 }
diff --git a/paddle/fluid/framework/fleet/nccl_wrapper.h b/paddle/fluid/framework/fleet/nccl_wrapper.h
index 5d73cba892d4f0358141c2b882a461f45d313f2c..eb4e5e19a3a2f7ad79b260261c8c1aa5c5103fa8 100644
--- a/paddle/fluid/framework/fleet/nccl_wrapper.h
+++ b/paddle/fluid/framework/fleet/nccl_wrapper.h
@@ -24,6 +24,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/program_desc.h"
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/framework/variable_helper.h"
+#include "paddle/fluid/platform/dynload/nccl.h"
 #include "paddle/fluid/platform/macros.h"  // for DISABLE_COPY_AND_ASSIGN
 
 namespace paddle {
diff --git a/paddle/fluid/pybind/fleet_wrapper_py.cc b/paddle/fluid/pybind/fleet_wrapper_py.cc
index 77f15db8d68da131c892b1a65946c1994b90fd04..2f6a7d2480aedd5bd37d0dbd5ccf64447e4a21ff 100644
--- a/paddle/fluid/pybind/fleet_wrapper_py.cc
+++ b/paddle/fluid/pybind/fleet_wrapper_py.cc
@@ -36,7 +36,6 @@ limitations under the License. */
 #include "paddle/fluid/pybind/fleet_wrapper_py.h"
 
 namespace py = pybind11;
-namespace pd = paddle::framework;
 
 namespace paddle {
 namespace pybind {
diff --git a/paddle/fluid/pybind/nccl_wrapper_py.cc b/paddle/fluid/pybind/nccl_wrapper_py.cc
index aec4f4463540df294694aa7a6b1d702867815eac..bbba03f6660fe9ddb14764709ea81a9a82b1b386 100644
--- a/paddle/fluid/pybind/nccl_wrapper_py.cc
+++ b/paddle/fluid/pybind/nccl_wrapper_py.cc
@@ -29,6 +29,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/async_executor.h"
 #include "paddle/fluid/framework/data_feed.h"
 #include "paddle/fluid/framework/data_feed.pb.h"
+#include "paddle/fluid/framework/fleet/nccl_wrapper.h"
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/inference/io.h"
 #include "paddle/fluid/platform/place.h"