16:45 <pac85> Inside zink_gfx_program_update_optimal there ara branches that look like if (ctx->gfx_dirty) { ... } else if (ctx->dirty_gfx_stages) { ... } where the second branch is more or less just a simplified version of the one above that only handles variants. Correct? If so does it really make sense to keep the two branches after my changes?

16:46 <zmike> I'd have to check, but currently the point is to avoid unnecessary locking and hash table lookups

16:50 <pac85> I see what you mean, I'll see if I can implement this optimization in my branch

16:51 <zmike> there's a drawoverhead case for program switching, and a lot of this code has been optimized for that

16:53 <pac85> I guess I'll run that when we get closer merging to make sure there are no significant performance regressions