nyilas has quit [Remote host closed the connection]
nyilas has joined #asahi-gpu
nyilas has quit [Remote host closed the connection]
nyilas has joined #asahi-gpu
Major_Biscuit has joined #asahi-gpu
MajorBiscuit has quit [Ping timeout: 480 seconds]
bisko has joined #asahi-gpu
<lina>
alyssa: MR with the stuff (I think) is already upstreamable is !21620, and asahilina:agx/next should be your rebased next ^^
<lina>
There was both some fence breakage in Mesa and a kernel bug (or rather a kernel I-didn't-know-about-a-firmware-corner-case). The latter was causing the hangs, so please uprev the kernel too ^^
<lina>
256FPS in terrain!
<lina>
alyssa: On that branch + latest firefox, quakejs renders without faults but all geometry is jittery, as if vertices are being computed in fp16.
<lina>
This is not a regression, it's always been like that apparently...
V has quit [Remote host closed the connection]
V has joined #asahi-gpu
hightower2 has quit [Ping timeout: 480 seconds]
c10l9 has quit []
c10l has joined #asahi-gpu
<lina>
Shell Shockers has the same problem ^^
amarioguy has quit [Remote host closed the connection]
<lina>
Other than that I haven't been able to repro any pink glitches, or any other weirdness, with everything I've thrown at it so far ^^
aleasto- has quit [Read error: Connection reset by peer]
sneak has joined #asahi-gpu
kenzie7 has quit [Read error: Connection reset by peer]
Emantor has quit [Ping timeout: 480 seconds]
TheLink has quit [Ping timeout: 480 seconds]
TheLink2 is now known as TheLink
hightower2 has joined #asahi-gpu
Stary has quit [Read error: Connection reset by peer]
aleasto has quit [Write error: connection closed]
aleasto- has joined #asahi-gpu
Stary has joined #asahi-gpu
Yamakaja_ has joined #asahi-gpu
jix_2 has quit [Remote host closed the connection]
jix_2 has joined #asahi-gpu
Yamakaja has quit [Read error: Connection reset by peer]
jannau has quit [Remote host closed the connection]
jannau has joined #asahi-gpu
kenzie795 has joined #asahi-gpu
kenzie79 has quit [Ping timeout: 480 seconds]
MajorBiscuit has joined #asahi-gpu
Major_Biscuit has quit [Ping timeout: 480 seconds]
possiblemeatball has joined #asahi-gpu
Major_Biscuit has joined #asahi-gpu
kesslerd has quit [Remote host closed the connection]
kesslerd has joined #asahi-gpu
MajorBiscuit has quit [Ping timeout: 480 seconds]
chipxxx has quit [Ping timeout: 480 seconds]
mairacanal9 is now known as mairacanal
Major_Biscuit has quit [Ping timeout: 480 seconds]
karolherbst is now known as karolherbst_
karolherbst_ is now known as karolherbst
Major_Biscuit has joined #asahi-gpu
kesslerd_ has joined #asahi-gpu
louisca has joined #asahi-gpu
louisca has quit [Remote host closed the connection]
bluetail9 has joined #asahi-gpu
bluetail has quit [Ping timeout: 480 seconds]
kesslerd_ has quit [Remote host closed the connection]
nela has joined #asahi-gpu
<alyssa>
lina: Woohoo!
<alyssa>
lina: Try ASAHI_MESA_DEBUG=nofp16
kesslerd has quit [Remote host closed the connection]
<lina>
alyssa: Huh, that does not fix it...
stsmwg has joined #asahi-gpu
<alyssa>
er wait
<alyssa>
lina: ASAHI_MESA_DEBUG=no16
stsmwg has quit []
<alyssa>
it's no16 for asahi, nofp16 for panfrost, becuase i like confusing myself apparently
<lina>
That does ^^
<alyssa>
woof
<alyssa>
Likely an application bug tbh
<lina>
Yeah, one of those "all apps get this wrong" things I bet...
<alyssa>
Yep
<alyssa>
If you only test on desktop GPUs you never test your precision assumptions
<alyssa>
If you test on mobile GPUs but only in phones, you meet your precision assumptions for screen resolution (and hence stuff like position and texture coordinates) accidentally
<lina>
This is <(modern) phone res and it's pretty bad, like decals sliding into walls bad and vertices jumping multiple pixels...
<alyssa>
maybe didn't test on phones
<alyssa>
Also, anytime you set an ASAHI_MESA_DEBUG related to shaders you need to also set MESA_SHADER_CACHE_DISABLE=1
<alyssa>
that keeps biting me >_>
<lina>
Oh, it didn't break here...
<alyssa>
oh nvm we did get that right
<lina>
So I guess this is going to have to be a quirks thing? And probably enable it for all browsers because WebGL...
<alyssa>
IDK
<alyssa>
It is an "I don't care" thing
<lina>
Is there a good way to prove it's an app bug so we can confidently say we don't care? ^^
<alyssa>
not easily
<alyssa>
but if you can reproduce the same brokenness on other mobile GPUs it's a pretty solid indication
<lina>
Would it happen on panfrost too?
<alyssa>
Plausibly?
<lina>
I can't test that easily but I think you can ^^
<alyssa>
I only have hardware that I write the drivers for,
<alyssa>
oh, shit
* alyssa
is running super late for a million things will test later
alyssa has quit [Quit: leaving]
Major_Biscuit has quit []
<lina>
I can't reproduce it on my phone at least (Firefox on Android, Pixel 4a 5G)
<lina>
Anyway, I should sleep ^^
kesslerd has joined #asahi-gpu
possiblemeatball has quit [Quit: Quit]
<povik>
the issue not reproducing on macos is not surprising?
alyssa has joined #asahi-gpu
<alyssa>
quakejs is too slow on rk3399 for me to bother to find out tbh
<alyssa>
seems ok on firefox because firefox is going through big gl and hence doesn't get fp16
<alyssa>
ok yes I can reproduce on asahi now that's not great
<alyssa>
idk and don't have the spoons to care about a webgl demo
<alyssa>
i'm running the webgl cts now, not 100% pass but results looks good
<alyssa>
so it probably is an app bug
<alyssa>
and again
<alyssa>
i do not care about quakejs
<alyssa>
it is a tech demo
<alyssa>
it is probably a buggy one at that
<alyssa>
it's not worth either of us wasting our time debugging it