diff --git a/paddle/fluid/operators/reduce_ops/logsumexp_op.cu b/paddle/fluid/operators/reduce_ops/logsumexp_op.cu
index c25e5d01b2758a96192d6fbf8f4e881770cbbbf0..c9ad1075c0c3c1c6f405144dbfde2e81b85124aa 100644
--- a/paddle/fluid/operators/reduce_ops/logsumexp_op.cu
+++ b/paddle/fluid/operators/reduce_ops/logsumexp_op.cu
@@ -12,7 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/fluid/operators/reduce_ops/cub_reduce.h"
 #include "paddle/fluid/operators/reduce_ops/logsumexp_op.h"
 
 REGISTER_OP_CUDA_KERNEL(logsumexp,
@@ -20,8 +19,3 @@ REGISTER_OP_CUDA_KERNEL(logsumexp,
                                           float, ops::LogsumexpFunctor>,
                         ops::ReduceKernel<paddle::platform::CUDADeviceContext,
                                           double, ops::LogsumexpFunctor>);
-REGISTER_OP_CUDA_KERNEL(
-    logsumexp_grad, ops::ReduceGradKernel<paddle::platform::CUDADeviceContext,
-                                          float, ops::LogsumexpGradFunctor>,
-    ops::ReduceGradKernel<paddle::platform::CUDADeviceContext, double,
-                          ops::LogsumexpGradFunctor>);
diff --git a/paddle/fluid/operators/reduce_ops/logsumexp_op.part.cu b/paddle/fluid/operators/reduce_ops/logsumexp_op.part.cu
new file mode 100644
index 0000000000000000000000000000000000000000..d6ad4863092a50233b806c944db0b8c161ed9dd0
--- /dev/null
+++ b/paddle/fluid/operators/reduce_ops/logsumexp_op.part.cu
@@ -0,0 +1,22 @@
+// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// .part used to speed up nvcc compile
+#include "paddle/fluid/operators/reduce_ops/logsumexp_op.h"
+
+REGISTER_OP_CUDA_KERNEL(
+    logsumexp_grad, ops::ReduceGradKernel<paddle::platform::CUDADeviceContext,
+                                          float, ops::LogsumexpGradFunctor>,
+    ops::ReduceGradKernel<paddle::platform::CUDADeviceContext, double,
+                          ops::LogsumexpGradFunctor>);