Make Chamfer Distance available for both float and double.

Author: hzxie

README.md

@@ -62,3 +62,5 @@ See [this useful chart](http://arnon.dk/matching-sm-architectures-arch-and-genco
 * [```Pointnet2_Tensorflow```](https://github.com/charlesq34/pointnet2) by [Charles R. Qi](https://github.com/charlesq34)
 
 * [```Pointnet2_PyTorch```](https://github.com/erikwijmans/Pointnet2_PyTorch) by [Erik Wijmans](https://github.com/erikwijmans)
+
+* [```GRNet```](https://github.com/hzxie/GRNet) by [Haozhe Xie](https://github.com/hzxie)

cuda/src/chamfer_dist.cu

@@ -158,8 +158,8 @@ std::vector<torch::Tensor> chamfer_dist_kernel_wrapper(torch::Tensor xyz1, torch
     const int batch_size = xyz1.size(0);
     const int n = xyz1.size(1); // num_points point cloud A
     const int m = xyz2.size(1); // num_points point cloud B
-    torch::Tensor dist1 = torch::zeros({batch_size, n}, torch::CUDA(torch::kFloat));
-    torch::Tensor dist2 = torch::zeros({batch_size, m}, torch::CUDA(torch::kFloat));
+    torch::Tensor dist1 = torch::zeros({batch_size, n}, torch::CUDA(xyz1.scalar_type()));
+    torch::Tensor dist2 = torch::zeros({batch_size, m}, torch::CUDA(xyz1.scalar_type()));
     torch::Tensor idx1 = torch::zeros({batch_size, n}, torch::CUDA(torch::kInt));
     torch::Tensor idx2 = torch::zeros({batch_size, m}, torch::CUDA(torch::kInt));
 
@@ -219,8 +219,8 @@ std::vector<torch::Tensor> chamfer_dist_grad_kernel_wrapper(torch::Tensor xyz1,
     const int batch_size = xyz1.size(0);
     const int n = xyz1.size(1); // num_points point cloud A
     const int m = xyz2.size(1); // num_points point cloud B
-    torch::Tensor grad_xyz1 = torch::zeros_like(xyz1, torch::CUDA(torch::kFloat));
-    torch::Tensor grad_xyz2 = torch::zeros_like(xyz2, torch::CUDA(torch::kFloat));
+    torch::Tensor grad_xyz1 = torch::zeros_like(xyz1);
+    torch::Tensor grad_xyz2 = torch::zeros_like(xyz2);
 
     AT_DISPATCH_FLOATING_TYPES(
         xyz1.scalar_type(), "chamfer_dist_grad_cuda", ([&] {