Does not work out of the box in Windows 10

[3dluvr]

Windows 10
VS 2017
Python 3.10

Aside from the missing instruction to install the CUDA Toolkit, which someone put a PR for here, using the default requirements fails to compile the CUDA kernel in Windows 10.
It returns the error: too few arguments for template template parameter "Tuple" detected, that many seem to be struggling with e.g. facebookresearch/pytorch3d#1024, for possibly the same reasons.

The combination that worked for me was CUDA-Toolkit 11.3.1 and torch==1.12.1+cu113 but even then the nvcc will fail to compile the kernel, complaining about mismatched parameters on atomicAdd in quant_cuda_kernel.cu.

I found that quick fix involves adding #include <THC/THCAtomics.cuh> to quant_cuda_kernel.cu, as per https://github.com/mit-han-lab/torchsparse/pull/75/files. Not sure what impact that has on performance, and perhaps atomicAdd would be there if correct platform flags are used for nvcc, but I wanted to see results fast and didn't bother.

In any case, this is a neat project, and looking forward to more updates.

[kuleshov]

Thanks for sharing this. Indeed, I haven't tested this on Windows. If you think you figured it out, and want to add Windows instructions as a separate doc linked from the README, I will be happy merge this as a PR.

[3dluvr]

I've dropped native Windows because I read it's slower than native Linux.

Right now I'm setting up WSL Ubuntu which shows promise, although it's another ordeal to get setup correctly with the right versions of everything.

[kuleshov]

setting up a virtualenv with pip -r requirement.txt is causing issues?

[3dluvr]

It was a combination of requirements because as-is, it was trying to install Torch 2.0. Also the bitsandbytes needed attention as well.

Overall, it's not worth running in Windows because it's slower than WSL, and anyone can setup WSL if they want to keep using Windows (like myself) for other things.

[halfpast-learn]

On windows 10 everything builds and runs fine except for quant cuda library. Gonna try to build gptq some other ways and see if it can be fixed