add double type for kernel

paddle/operators/conv_cudnn_op.cc

@@ -40,7 +40,8 @@ REGISTER_OP(conv_cudnn, ops::ConvOp, ops::CudnnConvOpMaker, conv_cudnn_grad,
             ops::ConvOpGrad);
 
 REGISTER_OP_CPU_KERNEL(conv_cudnn,
-                       ops::GemmConvKernel<paddle::platform::CPUPlace, float>);
+                       ops::GemmConvKernel<paddle::platform::CPUPlace, float>,
+                       ops::GemmConvKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(
-    conv_cudnn_grad,
-    ops::GemmConvGradKernel<paddle::platform::CPUPlace, float>);
+    conv_cudnn_grad, ops::GemmConvGradKernel<paddle::platform::CPUPlace, float>,
+    ops::GemmConvGradKernel<paddle::platform::CPUPlace, double>);

paddle/operators/conv_cudnn_op.cu.cc

@@ -259,6 +259,8 @@ class CudnnConvGradOpKernel : public framework::OpKernel<T> {
 }  // namespace operators
 }  // namespace paddle
 
-REGISTER_OP_GPU_KERNEL(conv_cudnn, paddle::operators::CudnnConvOpKernel<float>);
+REGISTER_OP_GPU_KERNEL(conv_cudnn, paddle::operators::CudnnConvOpKernel<float>,
+                       paddle::operators::CudnnConvOpKernel<double>);
 REGISTER_OP_GPU_KERNEL(conv_cudnn_grad,
-                       paddle::operators::CudnnConvGradOpKernel<float>);
+                       paddle::operators::CudnnConvGradOpKernel<float>,
+                       paddle::operators::CudnnConvGradOpKernel<double>);

paddle/operators/conv_transpose_cudnn_op.cc

@@ -61,10 +61,12 @@ REGISTER_OP(conv2d_transpose_cudnn, ops::ConvTransposeOp,
 
 REGISTER_OP_CPU_KERNEL(
     conv2d_transpose_cudnn,
-    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, float>);
+    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, float>,
+    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(
     conv2d_transpose_cudnn_grad,
-    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, float>);
+    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, float>,
+    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, double>);
 
 REGISTER_OP(conv3d_transpose_cudnn, ops::ConvTransposeOp,
             ops::CudnnConv3DTransposeOpMaker, conv3d_transpose_cudnn_grad,
@@ -72,7 +74,9 @@ REGISTER_OP(conv3d_transpose_cudnn, ops::ConvTransposeOp,
 
 REGISTER_OP_CPU_KERNEL(
     conv3d_transpose_cudnn,
-    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, float>);
+    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, float>,
+    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(
     conv3d_transpose_cudnn_grad,
-    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, float>);
+    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, float>,
+    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, double>);

paddle/operators/conv_transpose_cudnn_op.cu.cc

@@ -235,11 +235,15 @@ class CudnnConvTransposeGradOpKernel : public framework::OpKernel<T> {
 namespace ops = paddle::operators;
 
 REGISTER_OP_GPU_KERNEL(conv2d_transpose_cudnn,
-                       ops::CudnnConvTransposeOpKernel<float>);
+                       ops::CudnnConvTransposeOpKernel<float>,
+                       ops::CudnnConvTransposeOpKernel<double>);
 REGISTER_OP_GPU_KERNEL(conv2d_transpose_cudnn_grad,
-                       ops::CudnnConvTransposeGradOpKernel<float>);
+                       ops::CudnnConvTransposeGradOpKernel<float>,
+                       ops::CudnnConvTransposeGradOpKernel<double>);
 
 REGISTER_OP_GPU_KERNEL(conv3d_transpose_cudnn,
-                       ops::CudnnConvTransposeOpKernel<float>);
+                       ops::CudnnConvTransposeOpKernel<float>,
+                       ops::CudnnConvTransposeOpKernel<double>);
 REGISTER_OP_GPU_KERNEL(conv3d_transpose_cudnn_grad,
-                       ops::CudnnConvTransposeGradOpKernel<float>);
+                       ops::CudnnConvTransposeGradOpKernel<float>,
+                       ops::CudnnConvTransposeGradOpKernel<double>);

paddle/operators/pool_cudnn_op.cc

@@ -20,14 +20,18 @@ REGISTER_OP(pool2d_cudnn, ops::PoolOp, ops::Pool2dOpMaker, pool2d_cudnn_grad,
             ops::PoolOpGrad);
 
 REGISTER_OP_CPU_KERNEL(pool2d_cudnn,
-                       ops::PoolKernel<paddle::platform::CPUPlace, float>);
+                       ops::PoolKernel<paddle::platform::CPUPlace, float>,
+                       ops::PoolKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(pool2d_cudnn_grad,
-                       ops::PoolGradKernel<paddle::platform::CPUPlace, float>)
+                       ops::PoolGradKernel<paddle::platform::CPUPlace, float>,
+                       ops::PoolGradKernel<paddle::platform::CPUPlace, double>)
 
 REGISTER_OP(pool3d_cudnn, ops::PoolOp, ops::Pool3dOpMaker, pool3d_cudnn_grad,
             ops::PoolOpGrad);
 
 REGISTER_OP_CPU_KERNEL(pool3d_cudnn,
-                       ops::PoolKernel<paddle::platform::CPUPlace, float>);
+                       ops::PoolKernel<paddle::platform::CPUPlace, float>,
+                       ops::PoolKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(pool3d_cudnn_grad,
-                       ops::PoolGradKernel<paddle::platform::CPUPlace, float>)
+                       ops::PoolGradKernel<paddle::platform::CPUPlace, float>,
+                       ops::PoolGradKernel<paddle::platform::CPUPlace, double>)

paddle/operators/pool_cudnn_op.cu.cc

@@ -162,8 +162,12 @@ class PoolCudnnGradOpKernel : public framework::OpKernel<T> {
 
 namespace ops = paddle::operators;
 
-REGISTER_OP_GPU_KERNEL(pool2d_cudnn, ops::PoolCudnnOpKernel<float>);
-REGISTER_OP_GPU_KERNEL(pool2d_cudnn_grad, ops::PoolCudnnGradOpKernel<float>);
-
-REGISTER_OP_GPU_KERNEL(pool3d_cudnn, ops::PoolCudnnOpKernel<float>);
-REGISTER_OP_GPU_KERNEL(pool3d_cudnn_grad, ops::PoolCudnnGradOpKernel<float>);
+REGISTER_OP_GPU_KERNEL(pool2d_cudnn, ops::PoolCudnnOpKernel<float>,
+                       ops::PoolCudnnOpKernel<double>);
+REGISTER_OP_GPU_KERNEL(pool2d_cudnn_grad, ops::PoolCudnnGradOpKernel<float>,
+                       ops::PoolCudnnGradOpKernel<double>);
+
+REGISTER_OP_GPU_KERNEL(pool3d_cudnn, ops::PoolCudnnOpKernel<float>,
+                       ops::PoolCudnnOpKernel<double>);
+REGISTER_OP_GPU_KERNEL(pool3d_cudnn_grad, ops::PoolCudnnGradOpKernel<float>,
+                       ops::PoolCudnnGradOpKernel<double>);

paddle/operators/pool_op.cc

@@ -217,14 +217,18 @@ REGISTER_OP(pool2d, ops::PoolOp, ops::Pool2dOpMaker, pool2d_grad,
             ops::PoolOpGrad);
 
 REGISTER_OP_CPU_KERNEL(pool2d,
-                       ops::PoolKernel<paddle::platform::CPUPlace, float>);
+                       ops::PoolKernel<paddle::platform::CPUPlace, float>,
+                       ops::PoolKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(pool2d_grad,
-                       ops::PoolGradKernel<paddle::platform::CPUPlace, float>)
+                       ops::PoolGradKernel<paddle::platform::CPUPlace, float>,
+                       ops::PoolGradKernel<paddle::platform::CPUPlace, double>)
 
 REGISTER_OP(pool3d, ops::PoolOp, ops::Pool3dOpMaker, pool3d_grad,
             ops::PoolOpGrad);
 
 REGISTER_OP_CPU_KERNEL(pool3d,
-                       ops::PoolKernel<paddle::platform::CPUPlace, float>);
+                       ops::PoolKernel<paddle::platform::CPUPlace, float>,
+                       ops::PoolKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(pool3d_grad,
-                       ops::PoolGradKernel<paddle::platform::CPUPlace, float>);
+                       ops::PoolGradKernel<paddle::platform::CPUPlace, float>,
+                       ops::PoolGradKernel<paddle::platform::CPUPlace, double>);

paddle/operators/pool_op.cu.cc

@@ -17,11 +17,15 @@ limitations under the License. */
 namespace ops = paddle::operators;
 
 REGISTER_OP_GPU_KERNEL(pool2d,
-                       ops::PoolKernel<paddle::platform::GPUPlace, float>);
+                       ops::PoolKernel<paddle::platform::GPUPlace, float>,
+                       ops::PoolKernel<paddle::platform::GPUPlace, double>);
 REGISTER_OP_GPU_KERNEL(pool2d_grad,
-                       ops::PoolGradKernel<paddle::platform::GPUPlace, float>);
+                       ops::PoolGradKernel<paddle::platform::GPUPlace, float>,
+                       ops::PoolGradKernel<paddle::platform::GPUPlace, double>);
 
 REGISTER_OP_GPU_KERNEL(pool3d,
-                       ops::PoolKernel<paddle::platform::GPUPlace, float>);
+                       ops::PoolKernel<paddle::platform::GPUPlace, float>,
+                       ops::PoolKernel<paddle::platform::GPUPlace, double>);
 REGISTER_OP_GPU_KERNEL(pool3d_grad,
-                       ops::PoolGradKernel<paddle::platform::GPUPlace, float>);
+                       ops::PoolGradKernel<paddle::platform::GPUPlace, float>,
+                       ops::PoolGradKernel<paddle::platform::GPUPlace, double>);

paddle/operators/pool_with_index_op.cc

@@ -250,10 +250,12 @@ REGISTER_OP(max_pool2d_with_index, ops::MaxPoolWithIndexOp,
 
 REGISTER_OP_CPU_KERNEL(
     max_pool2d_with_index,
-    ops::MaxPoolWithIndexKernel<paddle::platform::CPUPlace, float>);
+    ops::MaxPoolWithIndexKernel<paddle::platform::CPUPlace, float>,
+    ops::MaxPoolWithIndexKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(
     max_pool2d_with_index_grad,
-    ops::MaxPoolWithIndexGradKernel<paddle::platform::CPUPlace, float>)
+    ops::MaxPoolWithIndexGradKernel<paddle::platform::CPUPlace, float>,
+    ops::MaxPoolWithIndexGradKernel<paddle::platform::CPUPlace, double>)
 
 REGISTER_OP(max_pool3d_with_index, ops::MaxPoolWithIndexOp,
             ops::MaxPool3dWithIndexOpMaker, max_pool3d_with_index_grad,
@@ -261,7 +263,9 @@ REGISTER_OP(max_pool3d_with_index, ops::MaxPoolWithIndexOp,
 
 REGISTER_OP_CPU_KERNEL(
     max_pool3d_with_index,
-    ops::MaxPoolWithIndexKernel<paddle::platform::CPUPlace, float>);
+    ops::MaxPoolWithIndexKernel<paddle::platform::CPUPlace, float>,
+    ops::MaxPoolWithIndexKernel<paddle::platform::CPUPlace, double>);
 REGISTER_OP_CPU_KERNEL(
     max_pool3d_with_index_grad,
-    ops::MaxPoolWithIndexGradKernel<paddle::platform::CPUPlace, float>)
+    ops::MaxPoolWithIndexGradKernel<paddle::platform::CPUPlace, float>,
+    ops::MaxPoolWithIndexGradKernel<paddle::platform::CPUPlace, double>)

paddle/operators/pool_with_index_op.cu.cc

@@ -18,14 +18,18 @@ namespace ops = paddle::operators;
 
 REGISTER_OP_GPU_KERNEL(
     max_pool2d_with_index,
-    ops::MaxPoolWithIndexKernel<paddle::platform::GPUPlace, float>);
+    ops::MaxPoolWithIndexKernel<paddle::platform::GPUPlace, float>,
+    ops::MaxPoolWithIndexKernel<paddle::platform::GPUPlace, double>);
 REGISTER_OP_GPU_KERNEL(
     max_pool2d_with_index_grad,
-    ops::MaxPoolWithIndexGradKernel<paddle::platform::GPUPlace, float>)
+    ops::MaxPoolWithIndexGradKernel<paddle::platform::GPUPlace, float>,
+    ops::MaxPoolWithIndexGradKernel<paddle::platform::GPUPlace, double>)
 
 REGISTER_OP_GPU_KERNEL(
     max_pool3d_with_index,
-    ops::MaxPoolWithIndexKernel<paddle::platform::GPUPlace, float>);
+    ops::MaxPoolWithIndexKernel<paddle::platform::GPUPlace, float>,
+    ops::MaxPoolWithIndexKernel<paddle::platform::GPUPlace, double>);
 REGISTER_OP_GPU_KERNEL(
     max_pool3d_with_index_grad,
-    ops::MaxPoolWithIndexGradKernel<paddle::platform::GPUPlace, float>)
+    ops::MaxPoolWithIndexGradKernel<paddle::platform::GPUPlace, float>,
+    ops::MaxPoolWithIndexGradKernel<paddle::platform::GPUPlace, double>)