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
<macc24>
uh oh
<macc24>
alyssa
<macc24>
i think you should see this
<alyssa>
please dont be a bug report please dont be a bug reportplease dont be a bug report
<macc24>
icecream95: 1. open sway with panfrost from main branch 2. try to run glxinfo
<icecream95>
macc24: I can't run glxinfo, trying to open konsole fails with "Failed to initialize EGL display 3001"
<macc24>
and we got another victim
atler is now known as Guest617
atler has joined #panfrost
* macc24
runs away to the safe land of freedreno
Guest617 has quit [Ping timeout: 480 seconds]
<icecream95>
macc24: glxinfo is working fine for me. Maybe you need a newer Xwayland?
<macc24>
icecream95: xwayland 1.20.11 is not enough?
stano has joined #panfrost
<icecream95>
macc24: That must be ancient, it's got to version 21.1 now
* icecream95
is using Xwayland from xserver master branch
<macc24>
hm
<macc24>
il try with xwayland 21.1.1 on ubuntu impish when i wake up
* macc24
goes to sleep
stano_ has quit [Ping timeout: 480 seconds]
camus1 has joined #panfrost
camus has quit [Ping timeout: 480 seconds]
camus1 has quit [Remote host closed the connection]
<icecream95>
It seems that varying interpolation on Bifrost is too imprecise for Firefox, which uses linear interpolation to copy large textures (e.g. 3840 high) even when they are not scaled
camus has joined #panfrost
<icecream95>
Midgard is a bit better, but is not perfect. On Bifrost the linear interpolation means that pixels can be out by about 20 / 256, which is easily noticeable
<icecream95>
(This happens even with nofp16)
davidlt has joined #panfrost
<HdkR>
Uh oh
<HdkR>
Does Mali do the fun ArtX thing where MSAA sample pattern actually affects pixel centers when MSAA isn't enabled? :D
<robclark>
hmm, seems like glxinfo would be an easy thing to get going in CI
davidlt has quit [Ping timeout: 480 seconds]
camus1 has joined #panfrost
camus has quit [Ping timeout: 480 seconds]
vstehle has joined #panfrost
davidlt has joined #panfrost
davidlt has quit [Ping timeout: 480 seconds]
warpme_ has joined #panfrost
rasterman has joined #panfrost
* icecream95
sees D3D9 backend for STK, digs up old patches for Gallium Nine
<macc24>
icecream95: what if dxvk o.o
nlhowell has joined #panfrost
<macc24>
why is openjdk in ubuntu's linux build deps
camus has joined #panfrost
camus1 has quit [Ping timeout: 480 seconds]
davidlt has joined #panfrost
warpme_ has quit [Quit: Connection closed for inactivity]
<macc24>
*notices that oibaf ppa has arm64 builds with panfrost*
<HdkR>
But does its x86 builds have panfrost?
<macc24>
hmm
<macc24>
no
<HdkR>
Fail
<macc24>
there are no panfrost machines that anyone would want to run fex on *shrugs*
* HdkR
sideeyes RK3588
* macc24
points at mt8192
<HdkR>
That one as well
<HdkR>
RK3588 doesn't have announced clock speeds, could be faster :P
<macc24>
2.2ghz big cores on rk3588 would mean chernobyl in your hands
<HdkR>
That's what the MT8192 is running them at, what would the difference be?
<macc24>
uh
<robmur01>
who said anything about FEX?
* robmur01
is still on the lookout for an Atom x3-C3440
<HdkR>
Forbidden product
<macc24>
robmur01: i can let you know if i come across one :p
<macc24>
isn't it the chip that some xda folks ran windows 7 on?
<HdkR>
Did it ever actually ship in a product?
<macc24>
ngl i kinda want to see x86 android phone running linux
<robmur01>
HdkR: that's what I'm not sure about - I think at least the lower-end ones with Mali-4x0 did end up in some Android tablets
<HdkR>
I don't recall any either
<macc24>
hmmmmmmmmmmm
<robmur01>
but the weirdest integration of the weirdest Midgard would be a pretty fun trophy :)
<macc24>
huh, atom x3 c3405 has a mali t720 too
<HdkR>
We just need a CXL Mali dGPU ;)
<macc24>
rk3288 on a stick
<HdkR>
I'm sure with a max spec Mali it would compete well with the Intel DG1
<macc24>
or maybe even geforce gt 710
<HdkR>
1030 is the new lowend hotness from Nvidia
<macc24>
no that's too fast
<HdkR>
Max spec Mali should beat it :P
<macc24>
a 100ghz mali 400 would beat everything
<macc24>
reject modernity
<macc24>
return to utgard
<HdkR>
I'd rather go to Hel :)
<macc24>
sure, plane tickets are $100 to gdańsk and then you can go to hel by train ;)
<robmur01>
more like RK3399 on a stick - in principle I think you could run some minimal software on it to marshal data/IRQs between the PCIe in endpoint mode and the GPU...
camus1 has joined #panfrost
<alyssa>
F/scroll goto -10
camus has quit [Ping timeout: 480 seconds]
wolfshappen has joined #panfrost
wolfshappen has quit []
wolfshappen has joined #panfrost
camus1 has quit [Remote host closed the connection]
camus1 has joined #panfrost
karolherbst has quit [Remote host closed the connection]
karolherbst has joined #panfrost
urja has quit [Read error: Connection reset by peer]
urja has joined #panfrost
cphealy_ has joined #panfrost
stormyknight3000 has joined #panfrost
cphealy has quit [Ping timeout: 482 seconds]
stormyknight3000 has quit []
stormyknight3000 has joined #panfrost
<stormyknight3000>
there we go
<stormyknight3000>
Is this a channel where I can get support with an issue I have been having with my screen resolution when using panfrost with an rk3399?
<macc24>
stormyknight3000: screen resolution has nothing to do with panfrost
<stormyknight3000>
so it's a kernel problem then... where would I go to discuss rockchipdrm then?
<macc24>
#linux-rockchip on libera
<stormyknight3000>
sounds good. thank you :)
<alyssa>
macc24: hey do you have dEQP built on your Duet?
<macc24>
alyssa: no
<macc24>
my duet has been nuked yesterday, and last week and 2 weeks ago...
<alyssa>
Woof.
<macc24>
such is the life of distro dev
<stormyknight3000>
on another note, how would I go about helping with funding panfrost? I would contribute code, but I am such an amateur that I don't think my code would be worth anything at the moment (meaning I have no idea what I am doing)
<alyssa>
stormyknight3000: panfrost development is sponsored by Corporations™ nowadays, we're not setup for individual donations
<stormyknight3000>
ah ok
<alyssa>
there are lots of deserving free software projects out there (including that rk3399 support relies on), consider donating to some of them? :)
<macc24>
"including that rk3399 support relies on" <- what is it
pendingchaos has quit [Quit: No Ping reply in 180 seconds.]
<alyssa>
....exercise for the reader? O:)
<macc24>
i won't figure it out myself
<stormyknight3000>
this is why I wish we had at least touched on low-level and systems programming in my c programming class
<stormyknight3000>
that might have given me some useful information I could use to teach myself such things :/
<stormyknight3000>
I went to two years of college and ended up where I began in terms of technical knowledge
pendingchaos has joined #panfrost
<stormyknight3000>
there was a lot of stuff about data structures and OOP, but nothing actually useful for systems programming and writing kernel drivers... I guess I have to go back to teaching myself... that was more effective anyways XD
cphealy_ has quit [Ping timeout: 480 seconds]
camus1 has quit [Remote host closed the connection]
<alyssa>
Example test case: dEQP-GLES3.functional.fragment_ops.depth_stencil.stencil_depth_funcs.stencil_never_depth_equal
jernej has joined #panfrost
<icecream95>
alyssa: Okay, will investigate. I still have about 70% battery left
<icecream95>
Guess which module caused this: ---[ end Kernel panic - not syncing: Fatal exception in interrupt ]---
<macc24>
icecream95: something mediatek?
<icecream95>
macc24: This is speedy
<macc24>
hm
<alyssa>
dwc2
<macc24>
brcmfmac
<icecream95>
alyssa: yep
<macc24>
D:
<alyssa>
icecream95: i really hate that driver
<alyssa>
so many USB issues just disappeared when I upgraded from rk3288 to rk3399.
<macc24>
should i even care about veyron in 2021?
<macc24>
it's not broken last time i checked
<alyssa>
"it's not broken last time i checked"
<macc24>
literally today
<alyssa>
A sentence you never want to hear in code review
<icecream95>
alyssa: Even the whole of dEQP-GLES3.functional.fragment ops.depthstencil doesn't fail any tests for me...
* icecream95
searches for the link to the initramfs used for the CI job
<alyssa>
Curious
<macc24>
initramfs is waste of space
<alyssa>
icecream95: I wonder if there's a kernel bug in whatever version CI is using?
<robclark>
macc24: I think the CI runners use initramfs for the entire (rather small) filesys.. or at least the ones that aren't using tftpboot+nfsroot.. since it avoids reflashing things 100x a day
<urja>
yeah i was about to comment that but i wasnt sure of that so ...
<urja>
but yeah i was like... not when it's the entire system :P
* icecream95
does not consider 274M "rather small"
<alyssa>
macc24: also ttbomk initramfs is needed for full disk encryption in 99% of cases
<alyssa>
(the 1% being suitably configured grub/coreboot/whatever)
<macc24>
alyssa: well you can do effectively same thing by just booting from a small partition that would have same contents as initramfs
<anholt>
robclark: nope, we don't initramfs in ci at all these days.
<robclark>
oh, I know we aren't for the freedreno runners, didn't realize everyone else moved off of that
<alyssa>
macc24: Depending on threat model that could increase risk of evil maid attacks, but we're quite off topic now :)
<macc24>
alyssa: >off-topic i can connect it to panfrost xD
<icecream95>
It seems Panfrost CI doesn't use an initramfs any more either, but just uses NFS root pointing to the extracted rootfs tarball
<icecream95>
The rootfs tarball has increased in size by 75% since the start of the year when I last downloaded it
<icecream95>
(That comes from including a copy of Mesa including .git and build/, and piglit)
<icecream95>
Where did my Mesa checkout go?
<macc24>
/dev/null
<robmur01>
alyssa: on the topic of off-topic, I just spent my whole evening reviewing the DART driver; I hope you're happy now :P
<icecream95>
("You moved it into the rootfs directory, then deleted it")
<macc24>
robmur01: DART?
<robmur01>
macc24: M1 IOMMU
<macc24>
huh
<robmur01>
for some reason people want I/O device to be usable on their fruitputers
<robmur01>
*devices
<macc24>
for some reason people want their fruitputers to be usable
<icecream95>
(Failing/flaky tests are dEQP-GLES3.functional.fragment out.array.fixed.rgb5, dEQP-GLES3.functional.fragment_out.random.33)
<icecream95>
The fragment_out flakes are weird.. One run I got 0 fails, the next was 46/940
* robmur01
wonders if this is the kevin saga all over again WRT non-functional DVFS
<macc24>
robmur01: uhh... what?
<macc24>
i think it worked for me
<icecream95>
Is deciding whether to use a blend shader or not nondeterministic now?
<icecream95>
Answer: yes
<icecream95>
Conditional jump or move depends on uninitialised value(s)
<macc24>
ew
<anarsoul>
:)
<icecream95>
at 0x6300054: blend_factor_constant_mask (pan_blend.c:77)
<robmur01>
macc24: as in when we were failing to scale voltages, some RK3399s managed to work OK by default but kevins (with a different regulator setup) were particularly flaky
<macc24>
robmur01: lol
<robmur01>
since we can't change OPPs at all on jacuzzi, I wonder if the default regulator voltages end up being marginal for whatever the default clock rate is
<macc24>
>since we can't change OPPs at all on jacuzzi
<macc24>
does it relate to kukui having multiple regulators for gpu and stuff?
<robmur01>
from the CI log it looks like ~512MHz, which is certainly above the middle of the table
<robmur01>
right, the OPP core won't touch >1 regulator since it can't know whether ordering matters