Matteo Seclì

> Given that sorting is relatively expensive, and IIUC does not practically matter if only for convenience/reproducibility, I'd only do this in the tests. Great, I'll do that. > What...

> It's unfortunate that we (still) have these (again). The Base LAPACK interface isn't really meant to be extended; IME it's better to implement interfaces at a higher level (e.g....

Hello, are there any news about this?

> Why the added `CUSOLVER.version() >= v"11.7.1"` conditionals? `Xgeev` was introduced in CUDA 12.6.2, which should correspond to CUSOLVER 11.7.1: https://docs.nvidia.com/cuda/archive/12.6.2/cuda-toolkit-release-notes/index.html

Hi @maleadt, sorry, didn't have much time to work on this. I've removed the extra checks, leaving only the one that explicitly tests `Xgeev` (as we need CUSOLVER 11.7.1 for...

It seems CI is properly running again and tests are passing on older CUDA versions as well.

Note: I still have to add tests, evaluate performance, etc.

I've been running some benchmarks (on CPU) in the meantime. I'm just using the example for the vacuum Rabi splitting, with the following script that writes out the runtime (minus...

I should have implemented GPU support for the `Lanczos` solver, but testing it as-is (and as a matter of fact, _any_ solver) doesn't work, and you'd have to change this...

Btw maybe it's worth keeping `kwargs` to let the user specify the `steadystate` solver, in this case? I'm not sure about the other solvers, but this is basically effortlessly GPU-compatible...