diff --git a/paddle/operators/conv_op.cc b/paddle/operators/conv_op.cc
index 687d741cb2..7a36a9b21a 100644
--- a/paddle/operators/conv_op.cc
+++ b/paddle/operators/conv_op.cc
@@ -225,11 +225,15 @@ REGISTER_OP(conv3d, ops::ConvOp, ops::Conv3DOpMaker, conv3d_grad,
             ops::ConvOpGrad);
 
 REGISTER_OP_CPU_KERNEL(conv2d,
-                       ops::GemmConvKernel<paddle::platform::CPUPlace, float>);
+                       ops::GemmConvKernel<paddle::platform::CPUPlace, float>,
+                       ops::GemmConvKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(
-    conv2d_grad, ops::GemmConvGradKernel<paddle::platform::CPUPlace, float>);
+    conv2d_grad, ops::GemmConvGradKernel<paddle::platform::CPUPlace, float>,
+    ops::GemmConvGradKernel<paddle::platform::CPUPlace, double>);
 
 REGISTER_OP_CPU_KERNEL(conv3d,
-                       ops::GemmConvKernel<paddle::platform::CPUPlace, float>);
+                       ops::GemmConvKernel<paddle::platform::CPUPlace, float>,
+                       ops::GemmConvKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(
-    conv3d_grad, ops::GemmConvGradKernel<paddle::platform::CPUPlace, float>);
+    conv3d_grad, ops::GemmConvGradKernel<paddle::platform::CPUPlace, float>,
+    ops::GemmConvGradKernel<paddle::platform::CPUPlace, double>);
diff --git a/paddle/operators/conv_op.cu.cc b/paddle/operators/conv_op.cu.cc
index 8e6f9da455..546451234a 100644
--- a/paddle/operators/conv_op.cu.cc
+++ b/paddle/operators/conv_op.cu.cc
@@ -17,11 +17,15 @@
 namespace ops = paddle::operators;
 
 REGISTER_OP_GPU_KERNEL(conv2d,
-                       ops::GemmConvKernel<paddle::platform::GPUPlace, float>);
+                       ops::GemmConvKernel<paddle::platform::GPUPlace, float>,
+                       ops::GemmConvKernel<paddle::platform::GPUPlace, double>);
 REGISTER_OP_GPU_KERNEL(
-    conv2d_grad, ops::GemmConvGradKernel<paddle::platform::GPUPlace, float>);
+    conv2d_grad, ops::GemmConvGradKernel<paddle::platform::GPUPlace, float>,
+    ops::GemmConvGradKernel<paddle::platform::GPUPlace, double>);
 
 REGISTER_OP_GPU_KERNEL(conv3d,
-                       ops::GemmConvKernel<paddle::platform::GPUPlace, float>);
+                       ops::GemmConvKernel<paddle::platform::GPUPlace, float>,
+                       ops::GemmConvKernel<paddle::platform::GPUPlace, double>);
 REGISTER_OP_GPU_KERNEL(
-    conv3d_grad, ops::GemmConvGradKernel<paddle::platform::GPUPlace, float>);
+    conv3d_grad, ops::GemmConvGradKernel<paddle::platform::GPUPlace, float>,
+    ops::GemmConvGradKernel<paddle::platform::GPUPlace, double>);
diff --git a/paddle/operators/conv_transpose_op.cc b/paddle/operators/conv_transpose_op.cc
index 310e3f5c93..3e55ef036a 100644
--- a/paddle/operators/conv_transpose_op.cc
+++ b/paddle/operators/conv_transpose_op.cc
@@ -185,17 +185,21 @@ REGISTER_OP(conv2d_transpose, ops::ConvTransposeOp, ops::Conv2DTransposeOpMaker,
 
 REGISTER_OP_CPU_KERNEL(
     conv2d_transpose,
-    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, float>);
+    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, float>,
+    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(
     conv2d_transpose_grad,
-    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, float>);
+    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, float>,
+    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, double>);
 
 REGISTER_OP(conv3d_transpose, ops::ConvTransposeOp, ops::Conv3DTransposeOpMaker,
             conv3d_transpose_grad, ops::ConvTransposeOpGrad);
 
 REGISTER_OP_CPU_KERNEL(
     conv3d_transpose,
-    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, float>);
+    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, float>,
+    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(
     conv3d_transpose_grad,
-    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, float>);
+    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, float>,
+    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, double>);
diff --git a/paddle/operators/conv_transpose_op.cu.cc b/paddle/operators/conv_transpose_op.cu.cc
index 401cddb379..4165eb0c7b 100644
--- a/paddle/operators/conv_transpose_op.cu.cc
+++ b/paddle/operators/conv_transpose_op.cu.cc
@@ -18,14 +18,18 @@ namespace ops = paddle::operators;
 
 REGISTER_OP_GPU_KERNEL(
     conv2d_transpose,
-    ops::GemmConvTransposeKernel<paddle::platform::GPUPlace, float>);
+    ops::GemmConvTransposeKernel<paddle::platform::GPUPlace, float>,
+    ops::GemmConvTransposeKernel<paddle::platform::GPUPlace, double>);
 REGISTER_OP_GPU_KERNEL(
     conv2d_transpose_grad,
-    ops::GemmConvTransposeGradKernel<paddle::platform::GPUPlace, float>);
+    ops::GemmConvTransposeGradKernel<paddle::platform::GPUPlace, float>,
+    ops::GemmConvTransposeGradKernel<paddle::platform::GPUPlace, double>);
 
 REGISTER_OP_GPU_KERNEL(
     conv3d_transpose,
-    ops::GemmConvTransposeKernel<paddle::platform::GPUPlace, float>);
+    ops::GemmConvTransposeKernel<paddle::platform::GPUPlace, float>,
+    ops::GemmConvTransposeKernel<paddle::platform::GPUPlace, double>);
 REGISTER_OP_GPU_KERNEL(
     conv3d_transpose_grad,
-    ops::GemmConvTransposeGradKernel<paddle::platform::GPUPlace, float>);
+    ops::GemmConvTransposeGradKernel<paddle::platform::GPUPlace, float>,
+    ops::GemmConvTransposeGradKernel<paddle::platform::GPUPlace, double>);