CPU version

[kamo-naoyuki]

Thank you for creating great library.
Would you support the CPU implementation in this library? It's not necessary, but I think more useful for comparing the performance, for example.

[DavidDiazGuerra]

Thanks for your feedback. I don't have plans to include a CPU implementation since I would need to program it from scratch; as far as I know, there is no way to compile CUDA code to run on CPU or anything like that. It is true that that it could be interesting for comparing the performance but making a vanilla implementation wouldn't be fair and writing an optimized CPU version would need more time than I have for this project.

For comparing the performance I took as reference this Python library: https://github.com/Marvin182/rir-generator and this Matlab library: http://www.eric-lehmann.com/fast_ISM_code/index.html

[kamo-naoyuki]

I see.

I think it's enough useful even if it's not optimized implementation.
I imaged that the users would like to switch the mode between CUDA and CPU version, like pytorch.
The API of gpuRIR is different from the other CPU RIR-generator module,
so we need to write only a little more codes for it.

However, it's O.K If you don't have the plan. I'm lazy.
Thanks.

[DavidDiazGuerra]

It's true that it would be nice having the option to switch between CUDA and CPU just like PyTorch does... maybe I do it in the future if I have time.