diff --git a/paddle/fluid/operators/sign_op.cc b/paddle/fluid/operators/sign_op.cc
index f3985dcc027f974e0213a73ea9a21e268d77615f..6837856a6da804e27af2cd6c83052c04f17140d8 100644
--- a/paddle/fluid/operators/sign_op.cc
+++ b/paddle/fluid/operators/sign_op.cc
@@ -67,4 +67,5 @@ namespace ops = paddle::operators;
 REGISTER_OPERATOR(sign, ops::SignOp, ops::SignOpMaker<float>,
                   ops::SignGradMaker);
 REGISTER_OP_CPU_KERNEL(
-    sign, ops::SignKernel<paddle::platform::CPUDeviceContext, float>);
+    sign, ops::SignKernel<paddle::platform::CPUDeviceContext, float>,
+    ops::SignKernel<paddle::platform::CPUDeviceContext, double>);
diff --git a/paddle/fluid/operators/sign_op.cu b/paddle/fluid/operators/sign_op.cu
index e0d7a87e6485a74dd1cdee1a05abc42eef460990..817e0fbbd511462f161633242d28e63062676eb9 100644
--- a/paddle/fluid/operators/sign_op.cu
+++ b/paddle/fluid/operators/sign_op.cu
@@ -13,7 +13,11 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/sign_op.h"
+#include "paddle/fluid/platform/float16.h"
 
 REGISTER_OP_CUDA_KERNEL(
     sign,
-    paddle::operators::SignKernel<paddle::platform::CUDADeviceContext, float>);
+    paddle::operators::SignKernel<paddle::platform::CUDADeviceContext, float>,
+    paddle::operators::SignKernel<paddle::platform::CUDADeviceContext, double>,
+    paddle::operators::SignKernel<paddle::platform::CUDADeviceContext,
+                                  paddle::platform::float16>);