diff --git a/paddle/operators/elementwise_add_op.cu b/paddle/operators/elementwise_add_op.cu
index 85d063a76b5592c716a5bdf23a0993976abc6ae4..7591428ac7c2f74f25f0f7d818eafcf59c8e4a4f 100644
--- a/paddle/operators/elementwise_add_op.cu
+++ b/paddle/operators/elementwise_add_op.cu
@@ -19,7 +19,13 @@ namespace ops = paddle::operators;
 
 REGISTER_OP_GPU_KERNEL(
     elementwise_add,
-    ops::ElementwiseAddKernel<paddle::platform::GPUPlace, float>);
+    ops::ElementwiseAddKernel<paddle::platform::GPUPlace, float>,
+    ops::ElementwiseAddKernel<paddle::platform::GPUPlace, double>,
+    ops::ElementwiseAddKernel<paddle::platform::GPUPlace, int>,
+    ops::ElementwiseAddKernel<paddle::platform::GPUPlace, int64_t>);
 REGISTER_OP_GPU_KERNEL(
     elementwise_add_grad,
-    ops::ElementwiseAddGradKernel<paddle::platform::GPUPlace, float>);
+    ops::ElementwiseAddGradKernel<paddle::platform::GPUPlace, float>,
+    ops::ElementwiseAddGradKernel<paddle::platform::GPUPlace, double>,
+    ops::ElementwiseAddGradKernel<paddle::platform::GPUPlace, int>,
+    ops::ElementwiseAddGradKernel<paddle::platform::GPUPlace, int64_t>);
diff --git a/paddle/operators/elementwise_div_op.cu b/paddle/operators/elementwise_div_op.cu
index b96aa31748c77f0d07f9bb7fb19235239983abd5..de4d0c33442a1fcfe0dd4c16df7ceeec737fbc6d 100644
--- a/paddle/operators/elementwise_div_op.cu
+++ b/paddle/operators/elementwise_div_op.cu
@@ -19,7 +19,13 @@ namespace ops = paddle::operators;
 
 REGISTER_OP_GPU_KERNEL(
     elementwise_div,
-    ops::ElementwiseDivKernel<paddle::platform::GPUPlace, float>);
+    ops::ElementwiseDivKernel<paddle::platform::GPUPlace, float>,
+    ops::ElementwiseDivKernel<paddle::platform::GPUPlace, double>,
+    ops::ElementwiseDivKernel<paddle::platform::GPUPlace, int>,
+    ops::ElementwiseDivKernel<paddle::platform::GPUPlace, int64_t>);
 REGISTER_OP_GPU_KERNEL(
     elementwise_div_grad,
-    ops::ElementwiseDivGradKernel<paddle::platform::GPUPlace, float>);
+    ops::ElementwiseDivGradKernel<paddle::platform::GPUPlace, float>,
+    ops::ElementwiseDivGradKernel<paddle::platform::GPUPlace, double>,
+    ops::ElementwiseDivGradKernel<paddle::platform::GPUPlace, int>,
+    ops::ElementwiseDivGradKernel<paddle::platform::GPUPlace, int64_t>);
diff --git a/paddle/operators/elementwise_mul_op.cu b/paddle/operators/elementwise_mul_op.cu
index 056f081d3e6ac349978ff00689700c035bed8e39..b0dfdee1ccef56c6cda06ae6759017294fa5115c 100644
--- a/paddle/operators/elementwise_mul_op.cu
+++ b/paddle/operators/elementwise_mul_op.cu
@@ -20,8 +20,12 @@ namespace ops = paddle::operators;
 REGISTER_OP_GPU_KERNEL(
     elementwise_mul,
     ops::ElementwiseMulKernel<paddle::platform::GPUPlace, float>,
-    ops::ElementwiseMulKernel<paddle::platform::GPUPlace, double>);
+    ops::ElementwiseMulKernel<paddle::platform::GPUPlace, double>,
+    ops::ElementwiseMulKernel<paddle::platform::GPUPlace, int>,
+    ops::ElementwiseMulKernel<paddle::platform::GPUPlace, int64_t>);
 REGISTER_OP_GPU_KERNEL(
     elementwise_mul_grad,
     ops::ElementwiseMulGradKernel<paddle::platform::GPUPlace, float>,
-    ops::ElementwiseMulGradKernel<paddle::platform::GPUPlace, double>);
+    ops::ElementwiseMulGradKernel<paddle::platform::GPUPlace, double>,
+    ops::ElementwiseMulGradKernel<paddle::platform::GPUPlace, int>,
+    ops::ElementwiseMulGradKernel<paddle::platform::GPUPlace, int64_t>);
diff --git a/paddle/operators/elementwise_sub_op.cu b/paddle/operators/elementwise_sub_op.cu
index 0efb92fce9975ed9fa029a3ce919589d09efb0d7..ec23bec35feae26f5463c575b1ab6f58d417e100 100644
--- a/paddle/operators/elementwise_sub_op.cu
+++ b/paddle/operators/elementwise_sub_op.cu
@@ -19,7 +19,13 @@ namespace ops = paddle::operators;
 
 REGISTER_OP_GPU_KERNEL(
     elementwise_sub,
-    ops::ElementwiseSubKernel<paddle::platform::GPUPlace, float>);
+    ops::ElementwiseSubKernel<paddle::platform::GPUPlace, float>,
+    ops::ElementwiseSubKernel<paddle::platform::GPUPlace, double>,
+    ops::ElementwiseSubKernel<paddle::platform::GPUPlace, int>,
+    ops::ElementwiseSubKernel<paddle::platform::GPUPlace, int64_t>);
 REGISTER_OP_GPU_KERNEL(
     elementwise_sub_grad,
-    ops::ElementwiseSubGradKernel<paddle::platform::GPUPlace, float>);
+    ops::ElementwiseSubGradKernel<paddle::platform::GPUPlace, float>,
+    ops::ElementwiseSubGradKernel<paddle::platform::GPUPlace, double>,
+    ops::ElementwiseSubGradKernel<paddle::platform::GPUPlace, int>,
+    ops::ElementwiseSubGradKernel<paddle::platform::GPUPlace, int64_t>);