Write custom NumPy ufunc in Rust?

[mhostetter]

Is it possible to write a custom NumPy ufunc in Rust and then apply it on arrays (with the requisite broadcasting) from either Rust or Python?

My application is writing finite field arithmetic ufuncs. I previously implemented this in Python and wrapped in Numba. I'm curious if I can replicate in Rust and achieve better performance.

Forgive the question if it's obvious. I didn't find an answer in documentation.

[adamreichold]

I think the only ways to current construct Rust-based ufuncs is to use the slow numpy.vectorize or use the FFI layer from numpy::npffi directly as we do not have any high-level wrapper for PyUFunc_FromFuncAndData and friends yet.

I do think we should have that and I would be glad if you could open a feature request for that. We can then try to figure our a useful API based on the information available in the tutorial.

[mhostetter]

Thanks for the response. Unfortunately, I'm not the right person for that PR. I'm a super beginner in Rust. Porting some Python+Numba code to Rust was going to be my first Rust project.

[adamreichold]

I meant just an issue for the feature to be added, not a pull request actually implementing it.

[kngwyu]

Thank you for your question. Closing in favor of #399. Let's continue the discussion in the issue thread.