Updated CUDA and PyTorch versions

[ahariri13]

Hello, I am interested in this code but I have CUDA 10.1 version and PyTorch 1.5. Is there a way to modify the setup accordingly ?

Thank you in advance

[cihanongun]

I've made a pull request but in the meantime, just change the deprecated AT_CHECK with TORCH_CHECK in the following lines and follow the installation steps.

PyTorchEMD/cuda/emd_kernel.cu

Lines 16 to 17 in e5c0feb

	#define CHECK_CUDA(x) AT_CHECK(x.type().is_cuda(), #x " must be a CUDA tensor")
	#define CHECK_CONTIGUOUS(x) AT_CHECK(x.is_contiguous(), #x " must be contiguous")

[asurada404]

Thanks a lot! This work for me.

[QhelDIV]

This works for me!(Pytorch 1.7 CUDA 10.2) Thanks a lot.

[XinYu-Andy]

Hi, does anyone know how to compile with CUDA 11.3 PyTorch 1.11?

[bluestyle97]

@XinYu-Andy I build PyTorchEMD successfully with PyTorch 1.11 and CUDA 11.0. I modified cuda/emd_kernel.cu as follows:

1. Comment #include <THC/THC.h>
2. Replace all THCudaCheck with C10_CUDA_CHECK

I hope this can be helpful!

[XinYu-Andy]

2. C10_CUDA_CHECK

Thank you very much! It works for me!

[Alhambra-H]

i successfully build PyTorchEMD，but when i run test_emd_loss.py， there comes another problem：Segmentation fault (core dumped)。does anyone know how to solve this？

@XinYu-Andy I build PyTorchEMD successfully with PyTorch 1.11 and CUDA 11.0. I modified cuda/emd_kernel.cu as follows:

1. Comment #include <THC/THC.h>
2. Replace all THCudaCheck with C10_CUDA_CHECK

I hope this can be helpful!

i successfully build PyTorchEMD，but when i run test_emd_loss.py， there comes another problem：Segmentation fault (core dumped)。does anyone know how to solve this？

[zhouzheyuan]

Thanks to @cihanongun and @bluestyle97, I build successfully with torch 2.0.0 and CUDA 11.7 by modified cuda/emd_kernel.cu as:

1. Comment #include <THC/THC.h>
2. Repalce all AT_CHECK with TORCH_CHECK
3. Replace all THCudaCheck with C10_CUDA_CHECK
4. Replace all CHECK_EQ with TORCH_CHECK_EQ