Krzysztof Drewniak

The expected behavior would be to prepend to LD_LIBRARY_PATH if it's set, not overwrite it

This does seem to be an issue with the desktop drawing over things like panels. `xkill`ing the desktop temporarily makes those things visible, but then `gnome-flashback` restarts and draws over...

Update: the fix, per https://github.com/regolith-linux/regolith-desktop/issues/276#issuecomment-625851388 is ```` gsettings set org.gnome.gnome-flashback root-background true gsettings set org.gnome.gnome-flashback desktop false ```

Since my changes aren't merged into SBCL's master yet, the API can still be adjusted. [Here](https://www.dropbox.com/s/n5qn4v8p13zg4dg/sbcl-manual-unicode-algorithms-for-github.pdf) is an SBCL manual that contains documentation for my API in section 7.8. Please...

I'll look through the `cl-unicode` api and try to add any functionality that might be needed to make integration easier.

As far as I know, my `unicode-algorithms` branch now puts all the data that `cl-unicode` uses in the SBCL database.

Re the perf degradation 1. We've got a known issue with how input fusions are codegen'd 2. Because matrix elements are inherently reused during a gemm - and even more...

How I'd like to reproduce the case where tuning generates a kernel that then has incorrect results: 1. Look at the failure log for a failing kernel config 2. Copy...

I'm afraid I left out some context. We call MIOpenDriver to test our tuning support and to make sure we can be called from MIOpen - the CI eventually runs...

I can collect more logs if folks think that'd be helpful