`torch.compile` support?

[MRiabov]

Hello Torchsparse devs,
I understand the torch.compile of torchsparse is impossible due to custom cuda kernels?

I understand we can wrap torch custom op over torchsparse, however that would do no grad.

Is there a workaround?

Thank you.