ChanServ changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://oftc.irclog.whitequark.org/panfrost - <macc24> i have been here before it was popular
<alyssa>
i'm still shook from that cache coherency thing so I guess debugging spilling isn't so bad
<alyssa>
first of all, it's an absolutely massive shader on bifrost..
<alyssa>
fails to find a register for node #15732 ... oy
<alyssa>
uh wait no
<alyssa>
failing Node 22560
<alyssa>
but I don't even see that node?
<alyssa>
unless wait ugh why do you have to be difficult alyssa
<alyssa>
index is 11280
<alyssa>
seemingly a normal ALU in a sea of ALU
<alyssa>
mostly fp16
<alyssa>
Nothing jumps out as wrong here.
<alyssa>
but, obviously it shouldn't be spilling and it is..
<alyssa>
oh.. right
<alyssa>
midgard makes better use of fp16
<alyssa>
fp16 on this shader keeps midgard from spilling, otherwise midgard does spill
<alyssa>
but even bifrost spilling nofp16 should still make progress?
<alyssa>
(23:202 on midgard with nofp16)
<alyssa>
It does seem to be making progress, as a matter of fact.
<alyssa>
uber slow, but progress
<alyssa>
oh wait it gets stuck at 1152.
<alyssa>
....Curious.
<alyssa>
with an improve heuristic it /is/ making progress but
<alyssa>
why is the allocator struggling this hard?
<alyssa>
that's all I got for now :\
<alyssa>
maybe I should revive that bifrost RA rewrite sometime
Bennett has quit [Read error: Connection reset by peer]