Add GPU support to direct linear solver in M3

[nonhermitian]

I am done with the code for direct solutions using GPUs via CuPy. Now need to integrate. Not sure how to do the unit testing in CI though.

[mtreinish]

I don't think there is going to be a way to run this in CI that exercises CuPy code unless you write things an agnostic way (like https://docs.cupy.dev/en/stable/user_guide/basic.html#how-to-write-cpu-gpu-agnostic-code) and it ends up just calling numpy under the covers. None of the publicly available CI services offer GPUs.

I checked what CuPy itself uses and it looks like they have a custom solution for gpu testing and just gha for build testing and other things like lint and docs.

[nonhermitian]

Yeah the agnostic numpy / cupy is also not really an option because things require custom kernels. The matrix free version requires a couple more kernels which is why that is not going to be in part 1

[nonhermitian]

Closing as no bandwidth to work on.