difference between cuda kernel code and using torch.ops #179
Description
I would like to ask what is the difference between cuda kernel code and using torch.ops. Is it okay if I change to torch.ops implementation? How much will it affect speed and memory usage?