Tested locally to not wait for CI. Though it's very odd this could slip by before... The basictests.jl should fail if the triangular solve is off but they didn't in the previous merges so there is still something I don't understand here.

@ChrisRackauckas Could this line prevent MKL from loading in the test? I think Github computers have AMD EPYC.

Oh true. Hmm We might need to force alternative hardware on here so that MKL is hit. I used to always be testing that locally, but now I do more dev on my Mac so that is under tested 😅. I'll setup some self-hosted hardware to make sure this stays stable.

I hit #679 on Github CI for my package, and manually enabled MKL to produce this documentation page that includes MKLLUFactorization and prints hardware info showing it was run on AMD EPYC.

I'm not 100% sure what happens when MKL is force-enabled on AMD hardware, though. Maybe it falls back to something else? Regardless, the crash on my Github CI was indeed MKL-related and fixed with this PR. 🤷

This package also directly uses MKLLUFactorization(), so I'm not sure why that case wouldn't've done it 😅