11:02 <itoral> apinheiro[m]1: remember that pass I added to optimize constant alu operations in the backend for optimized ldunifa chains? Turns out that can help more cases if we support more operations there

11:02 <itoral> A quick first version:

11:02 <itoral> total instructions in shared programs: 13733627 -> 13730592 (-0.02%)

11:02 <itoral> instructions in affected programs: 212353 -> 209318 (-1.43%)

11:02 <itoral> HURT: 317

11:02 <itoral> helped: 1721

11:03 <itoral> not a huge improvement, but hey..

11:03 <apinheiro[m]1> itoral: in fact I vaguley remember that one of your patches has a "FIXME: support more operations" comment

11:03 <apinheiro[m]1> is the same?

11:03 <itoral> yes

11:03 <itoral> at that point it was not clear to me if we would benefit because we don't usually emit arith in the backend

11:03 <itoral> so typically NIR would handle that

11:04 <apinheiro[m]1> one nice extra stat on shader-db would be how much it affects for HURT shaders

11:04 <apinheiro[m]1> although I guess that shader-db has already a lot of stats

11:05 <itoral> yeah, although you can see that if you want, it lists the hurt shaders with the stats for quick comparison