How can I implement the mapping operations on CPU cores?

[fkh12345]

Dear authors:
I want to execute torchsparse to analyze the performance of some operators on CPUs.
However, I notice that the mapping operations like "F.build_kernel_map" can only be executed on the GPUs.
This may be because the "build_mask_from_kmap" in the sparsemapping.cu only has the cuda implementation without CPU implementation.
May I ask what can I do if I want to execute the build_kernel_map operation on the CPU?

Thanks!

[zhijian-liu]

@kentang-mit, do we have a CPU implementation of build_kernel_map?