<enunes>
it seems that liveness with live_in and live_out ranges is not great with loops and control flow
<anarsoul>
it's not since we don't have holes
<enunes>
I rewrote liveness analysis to generate sets of live variables at each program point, so we don't run into that problem
<anarsoul>
enunes: can it compile all the glamor shaders now?
<enunes>
also I think there was an issue with variables being written and read in the same instruction
<enunes>
so their live range was pretty much forever
<enunes>
yes, glamor passes
<anarsoul>
hope we still have some time to merge it before 20.0 branches out :)
<anarsoul>
otherwise we may want to backport these fixes into 20.0
<anarsoul>
with glamor shaders working fine 20.0 should be more or less usable
<enunes>
I still want to optimize this a bit and maybe use sets instead of lists, there is one large shader that takes a few seconds to compile now with my inefficient implementation
<enunes>
hopefully the algorithm is now ok and I just need better data structures
<MoeIcenowy>
anarsoul: I used 19.3 with backported patches from 20.0 on my Pinebook now