17:34 <mntirc> one can crash the rk3588 completely with shadertoy (chromium 128, wayland mode, sway, newest mesa from git)

17:35 <mntirc> with this shader: https://www.shadertoy.com/view/4t3SzN

17:36 <mntirc> the a311d with mali g52 also greatly suffers but doesn't crash

17:37 <mntirc> cc bbrezillon

17:38 <mntirc> the whole system freezes here when accessing that page

17:42 <mntirc> perhaps interesting: on g52 this yields > [ 1488.666493] panfrost ffe40000.gpu: gpu sched timeout, js=0, config=0x7301, status=0x8, head=0x17bbe900, tail=0x17bbe900, sched_job=00000000385ada72

18:15 <robclark> probably just runs too slow and job timeout kicks in.. ie. probably WAI

19:00 <mntirc> robclark: yeah, but is that expected to crash v10?

19:05 <daniels> crash, no

22:40 <alyssa> mntirc: 9fps on m1 so yeah i can't imagine that going well on a mali

22:42 <alyssa> 200 atan's a pixel is going to hurt

22:48 <alyssa> Mesa implements atan with an 11th degree polynomial, which... hurts

22:57 <mntirc> :3

23:03 <alyssa> mntirc: oops i wrote a patch speeding up atan for all mesa drivers

23:03 <mntirc> alyssa: nice

23:03 <alyssa> i blame u uwu :3

23:03 <mntirc> :333

23:30 <alyssa> ok, patch series takes an atan on agx from 24 -> 13 instructions

23:30 <alyssa> so about half. I expect mali to be similar

23:41 <alyssa> and now to find out what it does for that shadertoy

23:42 <alyssa> 9.6fps -> 10fps

23:43 <alyssa> so... 4% reduction in frametime. Not as dramatic as I hoped but not terrible

23:49 <alyssa> mntirc: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30934 this does not fix your issue :P

23:55 <alyssa> mali also helped a lot, though there's more on the table there