Inverse Matrix on GPU

Hello Guys, there are some feature that I’m not aware of that perform matrix inversion using numba in GPU? my code runs everything on GPU already, but at some point I need to invert a matrix and I would like to know if this is possible with numba or if I’ll have to use other package.

Thanks in advance

Perhaps using CuPy instead? e.g. cupy.linalg.inv — CuPy 9.4.0 documentation

1 Like

yes, I was wondering if this would be possible just with numba, but thats okay. Do you know if this is something that may come to numba in the future? some linear algebra functions

I’d like it if there were support for linear algebra functions that worked at a thread / warp / block granularity, but I’m presently not aware of a library that does these things that can easily be integrated with Numba.

For larger data (e.g. where you’d like to use the whole GPU to invert a large matrix) it’s unlikely that support would be added in Numba, since it would duplicate the functionality of other libraries. It’s worth noting that most Python libraries that use CUDA can transparently operate on each others’ arrays without copying data because they implement the CUDA Array Interface: CUDA Array Interface (Version 3) — Numba 0.55.0.dev0+238.g6c01bbcb3.dirty-py3.7-linux-x86_64.egg documentation

1 Like