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
JulianGro has quit [Remote host closed the connection]
<daniels>
looks like that one flake does in fact go back to at least Feb, it's just drowned by irritating t720 noise
<daniels>
alyssa: fwiw I think there are two different things at play ehre maybe? MT8183 flakes on a lot of stuff, but RK3399 also flakes specifically on ASTC (much less frequently)
<alyssa>
arrrf
WoC` has joined #panfrost
WoC has quit [Ping timeout: 480 seconds]
Freedom12345 has joined #panfrost
Freedom12345 has quit []
WoC` has quit []
WoC has joined #panfrost
<greenjustin>
I'm still getting weird stuttering on my MT8183 with panfrost. I'm getting these in my dmesg, any of you folks happen to see something like this before? https://pastebin.com/FLrVp7aE
<greenjustin>
It doesn't happen with like webgl aquarium, but it's really bad when I'm scrolling, or typing, or trying to log in. I'm guessing it has something to do with the display controller?
<greenjustin>
it's fairly standard for our mediatek 5.10 builds, just minus the kbase drivers I think
<robmur01>
yup, doesn't look like much to do with panfrost, more like the display behaving weirdly either by itself or due to something userspace is doing
<greenjustin>
the behavior only starts when I use panfrost drivers though
<robmur01>
of course said weirdness may be tangential to any difference in how panfrost vs. kbase interact with the display driver, though
<robmur01>
(i.e. principally DRM vs. whatever kbase does)
<greenjustin>
the weird thing is it largely doesn't seem to happen with 3D applications, but panfrost is really the only thing I've changed
<robmur01>
hmm, might it doing something like aggressively sticking the panel into self-refresh and pausing scanout when it thinks it's idle, I wonder?
<greenjustin>
that's what I was thinking, because sometimes it doesn't present as a stutter so much as a black screen, or it will weirdly cycle through 2 or 3 frames when idle
<robmur01>
I wouldn't be at all surprised if kbase doing some non-DRM thing prevents it from ever getting down to that idle state. Can you compare with no GPU driver at all, maybe?
<greenjustin>
let me try that
<macc24>
greenjustin: what dts file are you using?
<macc24>
is the driver for 'mtk command queue' or whatever, enabled?
<robmur01>
oh man, I'd forgotten about MTK_CMDQ, that thing's bonkers
<macc24>
if it's enabled - disable it, if it's disabled - enable it
<macc24>
might kick dsi stuff into working fine
<macc24>
iirc there was something else too related to that, forgot about it now tho
<greenjustin>
looks like I can't actually disable that for some reason
<greenjustin>
I found it in the options, but it won't let me change it :\
rasterman has quit [Quit: Gettin' stinky!]
floof58 has quit [Quit: floof58]
floof58 has joined #panfrost
<macc24>
idk then
<macc24>
pure mainline kernel works well enough for me
<macc24>
pure-ish
guillaume_g has quit []
<alyssa>
with the old RA, but to make room for the new one :]
<alyssa>
err
<alyssa>
wrong buffer sorry
spawacz has quit [Quit: WeeChat 3.0]
spawacz has joined #panfrost
<greenjustin>
well, it looks like chrome may not start without GL, so I'm not sure I can really test this setup without a GPU driver
<macc24>
greenjustin: LIBGL_ALWAYS_SOFTWARE=1 or MESA_LOADER_DRIVER_OVERRIDE=llvmpipe
<macc24>
not sure about 2nd env var... i don't remember the spelling of it but it's around what i said
<robclark>
env vars are a bit more... interesting.. with chrome because of sandboxing.. but you can add things to /etc/chrome_dev.conf
<robclark>
that said, we probably don't (by default) enable llvmpipe in our builds.. since no need to
<robclark>
(and llvmpipe doesn't really work that well last time I tried.. probably due to lack of fence-fd's)
<greenjustin>
robclark: so I need more than just the env vars, I need to rebuild mesa to enable llvmpipe?
<macc24>
greenjustin: check out if softpipe renderer exists lol
Danct12 has quit [Quit: Quitting]
Danct12 has joined #panfrost
Danct12 has quit []
Danct12 has joined #panfrost
<robclark>
greenjustin: likely, we usually just enable a single driver, since no need to waste disk space.. and like I said, don't be surprised if swrast/llvmpipe tears badly
<greenjustin>
I'm having a hard time actually convincing chrome to use swrast...
<greenjustin>
Dumb plan: could I just compile panfrost as a module and then unload the module after chrome starts up?