diff --git a/paddle/fluid/operators/split_byref_op.cu.cc b/paddle/fluid/operators/split_byref_op.cu.cc
index 1faf4f55dd54a2dc28c19232643563a31850f38b..5ee6186f3541b7dcb845ce0c6d28081685925da0 100644
--- a/paddle/fluid/operators/split_byref_op.cu.cc
+++ b/paddle/fluid/operators/split_byref_op.cu.cc
@@ -15,4 +15,5 @@ limitations under the License. */
 #include "paddle/fluid/operators/split_byref_op.h"
 namespace ops = paddle::operators;
 REGISTER_OP_CUDA_KERNEL(
-    split, ops::SplitByrefOpKernel<paddle::platform::CUDADeviceContext, float>);
+    split_byref,
+    ops::SplitByrefOpKernel<paddle::platform::CUDADeviceContext, float>);