17:36 anholt: karolherbst: there, I put up the NIR backend MR for you.
17:36 karolherbst: nice
17:37 karolherbst: the stats are massive
17:37 karolherbst: but I think reduction in local might be just a side effect
17:39 karolherbst: anholt: it seems like we do something wrong for GS shaders on nv50
17:40 karolherbst: huh edgeflag ...
17:44 anholt: I think I'm making some sense of the primid fail
17:45 karolherbst: I can take a look at all those nvc0+ regressions, as they should be fixed asap (because volta+)
17:49 karolherbst: anholt: I am a bit unhappy about 15597 as we are doing the wrong thing with tgsi as well
17:49 karolherbst: but I guess we can do the same wrong thing with nir
18:00 anholt: nv50 alphatest fixed.
18:01 karolherbst: anholt: more nir lowering is the correct answer anyway :P
18:15 anholt: nv50 primid sorted. now to see if this nvc0 fix worked.
18:17 karolherbst: anholt: what was wrong with primid?
18:17 anholt: it's an sv now, gotta look in the sv list for it.
18:18 karolherbst: ahh
18:18 karolherbst: makes sense
18:20 karolherbst: last time I checked using nir by default, the GPR usage increased by 40% for pixmark_piano, but speed improved by 10%
18:24 anholt: cuts instructions by 10%, so yeah.
18:24 karolherbst: well more gprs means less threads
18:25 karolherbst: but all those loop based opts are a huge win
18:25 karolherbst: so I wouldn't be surprised that now it's even more perf
18:25 anholt: sure, but if you're basically not doing memory access then your threadcount doesn't matter much.
18:26 karolherbst: I meant active threads
18:27 karolherbst: I would be surprised if running more threads at once doesn't make a difference
18:27 anholt: I'm assuming your threadcount is the "number of shaders logically active on the shader core where you thread switch between them on stalls"
18:27 anholt: is it not that?
18:27 karolherbst: nope
18:27 karolherbst: I meant like real threads
18:28 karolherbst: it's... a bit complicated, but we have a "logical threads being there" and a "threads actually running at the same time" thing
18:28 karolherbst: and used GPRs have an impact on the latter
18:30 agneli: guys maybe one of you woudl have some time and will to look into my issue, please? here is the copy of that I previously pasted in the channel https://paste.debian.net/plain/1237981
18:31 karolherbst: anholt: but there is a little bit more to it.. so if we fall below a certian threshold (I think 32 regs) it doesn't matter for real
21:16 karolherbst: anholt: the diff in fails looks very good now :)