diff --git a/paddle/operators/clip_op.cc b/paddle/operators/clip_op.cc
index fde05759a9040efcb7470fda0d6ecc59837cd090..86d79866a8e7c4cda036ce7e0f5527fd0086b482 100644
--- a/paddle/operators/clip_op.cc
+++ b/paddle/operators/clip_op.cc
@@ -80,5 +80,7 @@ class ClipOpGrad : public framework::OperatorWithKernel {
 namespace ops = paddle::operators;
 REGISTER_OP(clip, ops::ClipOp, ops::ClipOpMaker<float>, clip_grad,
             ops::ClipOpGrad);
-REGISTER_OP_CPU_KERNEL(clip, ops::ClipKernel<float>);
-REGISTER_OP_CPU_KERNEL(clip_grad, ops::ClipGradKernel<float>);
+REGISTER_OP_CPU_KERNEL(clip,
+                       ops::ClipKernel<paddle::platform::CPUPlace, float>);
+REGISTER_OP_CPU_KERNEL(clip_grad,
+                       ops::ClipGradKernel<paddle::platform::CPUPlace, float>);
diff --git a/paddle/operators/clip_op.cu b/paddle/operators/clip_op.cu
index 3a61841a56b30729bbf3db7f8e0cd41e009581eb..ca9701298fdae3fabe234925edaf9e4d775cc66e 100644
--- a/paddle/operators/clip_op.cu
+++ b/paddle/operators/clip_op.cu
@@ -15,5 +15,7 @@
 #include "paddle/operators/clip_op.h"
 
 namespace ops = paddle::operators;
-REGISTER_OP_GPU_KERNEL(clip, ops::ClipKernel<float>);
-REGISTER_OP_GPU_KERNEL(clip_grad, ops::ClipGradKernel<float>);
+REGISTER_OP_GPU_KERNEL(clip,
+                       ops::ClipKernel<paddle::platform::GPUPlace, float>);
+REGISTER_OP_GPU_KERNEL(clip_grad,
+                       ops::ClipGradKernel<paddle::platform::GPUPlace, float>);
diff --git a/paddle/operators/clip_op.h b/paddle/operators/clip_op.h
index 5d05959129eaf1e4fd44e2761e265242750650b2..5ca32da41ad85379d9358fa86ddd163d9906dde4 100644
--- a/paddle/operators/clip_op.h
+++ b/paddle/operators/clip_op.h
@@ -58,7 +58,7 @@ class ClipGradFunctor {
   T max_;
 };
 
-template <typename T>
+template <typename Place, typename T>
 class ClipKernel : public framework::OpKernel {
  public:
   void Compute(const framework::ExecutionContext& context) const override {
@@ -69,12 +69,13 @@ class ClipKernel : public framework::OpKernel {
     T* out_data = out->mutable_data<T>(context.GetPlace());
     const T* x_data = x->data<T>();
     int numel = x->numel();
-    Transform(context.device_context(), x_data, x_data + numel, out_data,
-              ClipFunctor<T>(min, max));
+    Transform<Place> trans;
+    trans(context.device_context(), x_data, x_data + numel, out_data,
+          ClipFunctor<T>(min, max));
   }
 };
 
-template <typename T>
+template <typename Place, typename T>
 class ClipGradKernel : public framework::OpKernel {
  public:
   void Compute(const framework::ExecutionContext& context) const override {
@@ -88,8 +89,9 @@ class ClipGradKernel : public framework::OpKernel {
       auto d_x_data = d_x->mutable_data<T>(context.GetPlace());
       const T* d_out_data = d_out->data<T>();
       const T* x_data = x->data<T>();
-      Transform(context.device_context(), d_out_data, d_out_data + numel,
-                x_data, d_x_data, ClipGradFunctor<T>(min, max));
+      Transform<Place> trans;
+      trans(context.device_context(), d_out_data, d_out_data + numel, x_data,
+            d_x_data, ClipGradFunctor<T>(min, max));
     }
   }
 };