warpme_ has quit [Quit: Connection closed for inactivity]
_whitelogger has joined #panfrost
WoC has quit [Remote host closed the connection]
kaspter has quit [Quit: kaspter]
raster has joined #panfrost
leah has quit [Quit: WeeChat 2.8]
unoccupied has joined #panfrost
unoccupied is now known as leah
* macc24
found a bug
<macc24>
when 2 firefox windows are opened, one "normal" and other "private", then they leak contents from each other
<macc24>
on G72
<macc24>
kinkinkijkin: can you replicate this?
<alyssa>
macc24: Hey, can I conscript you into testing something on G72 for me? 🐈
<macc24>
maybe
<macc24>
gl3?
yann has joined #panfrost
<macc24>
does it make supertuxkart work better?
davidlt has quit [Ping timeout: 272 seconds]
<alyssa>
...no
<alyssa>
something that strongly smells like a hardware bug, reproduces reliably on G52, want to know if it's also an issue on G72
<macc24>
ok
<alyssa>
would require compiling dEQP so maybe more work than you want
<macc24>
is deqp big?
<alyssa>
unless I cooked up an API trace? I could probably do that
yann has quit [Ping timeout: 264 seconds]
<macc24>
ill test later today
<macc24>
can't now
<macc24>
alyssa: should i start compiling deqp?
<alyssa>
uhhh
davidlt has joined #panfrost
WoC has joined #panfrost
WoC has quit [Client Quit]
WoC has joined #panfrost
yann has joined #panfrost
yann has quit [Ping timeout: 272 seconds]
cphealy has quit [Ping timeout: 240 seconds]
cphealy has joined #panfrost
cphealy has quit [Read error: Connection reset by peer]
cphealy has joined #panfrost
warpme_ has joined #panfrost
<HdkR>
I don't believe content leaking in firefox is a bug. If the application tried reading back the framebuffer before writing to it then it would still get zero
<macc24>
HdkR: i didn't see anything like that on my desktop system
<macc24>
or on any of my laptops that aren't running panfrost
<macc24>
icecream95: can i has chromeos dmesg?
alpernebbi has quit [Remote host closed the connection]
<HdkR>
alyssa: How close is Bifrost to ES/GL 3 land atm?
<alyssa>
HdkR: ES3.0 is very close, big thing missing is MRT
<alyssa>
<1 month I expect to get that flipped on, but time tables are hard
<alyssa>
Let me see what big GL needs
<macc24>
:OOOOOOOOOOOOO
<alyssa>
HdkR: Big GL 3.1 can follow shortly after ES3.0 is flipped on. we're blocking on like one MR for midgard, just never got around to finishing that up
<alyssa>
why?
<HdkR>
Mostly just wondering if it is coming up soon that I should put newer panfrost on one of my boards :D
<alyssa>
soon™
<HdkR>
A month is a good amount of time to store some motivation towards figuring out kernel updating issues :P
<alyssa>
Hehe
<macc24>
hehe
<macc24>
in 5.11 mediatek people got some mt8183 stuff into mainline and that broke cadmium
<HdkR>
Time to bisect? :)
<macc24>
it's something in my patches
<macc24>
and dpi didn't get in so i have to piece it together from patches, and mainline code
<macc24>
mtk_dpi.c has stuff to handle mt8183, but mt8183.dtsi has no dpi
<macc24>
currently i'm just waiting for my -hopefully- male usb c breakout pcb
<HdkR>
bweh, DT seems completely non-viable long term. Hopefully ACPI sorts itself out going forward
<alyssa>
:<
<HdkR>
ooo, imported chocolate arriving today. Nothing can ruin this day
<macc24>
HdkR: pls no acpi
<HdkR>
The alternative is having vendors ship DT on their devices which isn't really viable thougih?
<macc24>
acpi is a mess imo
<HdkR>
Sure, which is where the "sorts itself out" request holds :P
<macc24>
especially when vendors don't use #include
<macc24>
>The alternative is having vendors ship DT on their devices which isn't really viable thougih?
<macc24>
why it won't be viable?
<HdkR>
Because Linux won't ever be large enough for mainstream vendors to ship DT versus ACPI and instead running Windows
<macc24>
google isn't mainstream enough?
<HdkR>
Also doesn't DT have compatibility breaking changes. Making someone shipping a device with long term kernel support untenable unless their DT is upstreamed and managed?
<macc24>
i have no idea
<HdkR>
For the handful of devices that Google ships, sure. They have enough money to throw people at the maintaining problem for a few years :P
<macc24>
i'asdfsdaf
<macc24>
i'd prefer to have stuff on discoverable buses, eliminating need to hardcode most devices in dt
<HdkR>
Sure
<HdkR>
I believe everyone would want that
<HdkR>
But it's never going to happen for everything
<HdkR>
and you have the problem of semi-customizable designs, so having just a SoC described in DT is just a template and won't work for everything
<macc24>
hmm
<HdkR>
Even the same model device having minor changes is causing problems in DT land. Meanwhile the manufacturer should be shipping ACPI definitions on device describing the hardware differences
<HdkR>
Apparently there are like five different minor revisions of Lenovo c630, changing panels, keyboards, and touchpads :P
icecream95 has joined #panfrost
warpme_ has quit [Quit: Connection closed for inactivity]
<alyssa>
I see no logical reason that test should fail, but it consistently does on G52. Trying to figure out if this is an architectural issue or just a G52 bug.
<alyssa>
(This connects to your "end a shader with a nop" patch as well)
<macc24>
icecream95: do you still have chromeos on your duet?
<icecream95>
macc24: yes
<macc24>
icecream95: pls dmesg log
<macc24>
i lost my one
<kinkinkijkin>
ok, that patch i applied last night
<kinkinkijkin>
now getting a fair few notable graphical glitches, one of the most important being that transparency no longer exists
<kinkinkijkin>
another is that quads with video on them are the wrong colour
<macc24>
kinkinkijkin: ?
<macc24>
which programs?
<kinkinkijkin>
to fix a SIGFPE
<kinkinkijkin>
panfrost
<kinkinkijkin>
everything
<macc24>
that's weird
<macc24>
unless it's not on duet
<kinkinkijkin>
it's on duet
<macc24>
for me git commit 4dbcf24f68 works mostly fine
<kinkinkijkin>
i applied a patch given to me by icecream95 written by alyssa that's likely not fully ready yet
<icecream95>
alyssa: The dEQP test fails, but passes if I disable DCE so the instruction count is no longer 128
<icecream95>
kinkinkijkin: It's merged in master now
<kinkinkijkin>
oh
<kinkinkijkin>
k ill update to master in a sec and see if that changes anything
<alyssa>
icecream95: lovely, so that means *all* of Bifrost is broken. thanks for checking that
<alyssa>
raster: ^^
davidlt has quit [Ping timeout: 272 seconds]
<macc24>
icecream95: is it 2021 in your timezone?
<alyssa>
so it is. happy new year!
<icecream95>
I've already typed 2020-01-01 a few times today
* alyssa
working on fp16 stuff since she ran out of energy to fight the scheduler
<alyssa>
icecream95: but you'll be happy to know the scheduler is coming along nicely
<alyssa>
Hoping to have something functional, if not highly optimized, finished next week...but it looks like that hw bug might disagree.
<alyssa>
[At which point implementing atomics would actually be doable
<alyssa>
]
<HdkR>
mmm, mt8192 dev board. gimme
<alyssa>
🎵 Gimme Gimme - Thoroughly Modern Millie 🎵
<macc24>
HdkR: you know who will mainline it happily ;)
<macc24>
icecream95: thanks
<HdkR>
Once I get my hands on a Linux devboard with drivers I'm going to drop everything and just spend a month on it
<macc24>
HdkR: that's literally what i did with duet
<macc24>
i spent like, all of my free time on it
<HdkR>
:D
<macc24>
it was like 8 hours per day
<kinkinkijkin>
i am willing to daily a linux devboard with cortex-a75 or better, and either a sterk gpu or working pcie, for at least a month
<kinkinkijkin>
not "work on it daily"
<kinkinkijkin>
"use it as my primary daily work device"
<macc24>
if not for a fact that duet doesn't support more than one external monitor i'd replace my xeon desktop with it
<kinkinkijkin>
i replaced my ryzen desktop with it when my ryzen desktop got destroyed and then now also stolen
<macc24>
RIP
tomboy64 has quit [Remote host closed the connection]
<mifritscher>
Happy New Year!
tomboy64 has joined #panfrost
<alyssa>
"use it as my primary daily work device"
<alyssa>
you called?
<macc24>
kinkinkijkin: does your right shift sometimes feel s
<macc24>
spongy*
<macc24>
and icecream95
<icecream95>
macc24: The right shift key does feel a bit different to the left shift
<kinkinkijkin>
macc24 not spongy but different yeah
<macc24>
does it get better after blowing air through it?
<kinkinkijkin>
it doesn't exactly feel bad... just like different
<macc24>
hmm
<HdkR>
keyboard feel is surpringingly difficult to get right in laptops
<macc24>
HdkR: that's why i'm making mechanical keyboard for duet when i get something to build casing of;D
<HdkR>
Key haha
<HdkR>
s/Key //
<macc24>
similar thing has been done with pinephone
<HdkR>
keykap stabilization not being right destroys my ability to use a keyboard. I hit the corner of keys far too often
<macc24>
which switches were you typing on?
<macc24>
outemus and other cherry copies are pretty bad in stabilization
<HdkR>
From my laptop it is Dell's XPS13 line. Desktop is using Apple's Scissor Switches in their Magic Keyboard 2. Which I went to from a Cherry MX-Red keyboard which had shite keycaps
<macc24>
,_,
<HdkR>
:P
<macc24>
my current keyboard has kailh red and it's so much better than any membrame keyboard i typed on
<HdkR>
Oh definitely
<macc24>
"membrame" includes scissor switches
<kinkinkijkin>
kailh box are pretty good for stabiliztion, as far as mx compatibles go
<kinkinkijkin>
i had a keyboard with navies
<kinkinkijkin>
real nice keyboard
<macc24>
my t400 was the best laptop that i typed on
<kinkinkijkin>
i miss it =[
<HdkR>
I mean, I'm not going to claim my keyboards have the absolute best. It's just what I'm current running. I dropped my Razer Blade 13" and one of the major reasons was because of it's absolute trash tier keyboard :P
<HdkR>
Key consistency, key stabilization, layout. All bad
<macc24>
HdkR: is it possible to have more cramped keyboard layout than in duet? xD
<macc24>
some keys there are for ants!
<kinkinkijkin>
apple makes the best scissor-switch keyboards
<kinkinkijkin>
you're right to enjoy your apple magic keyboard 2
<HdkR>
It's non-abusive when switching between my laptop and desktop which is what I was wanting after that other keyboard being bad
<kinkinkijkin>
my box navies board had a custom capset
<macc24>
i use a stock spc gear gk540
<kinkinkijkin>
i was using two tada 68s with different switches, caps, and plates
<macc24>
before that i had literally cheapest mechanical keyboard on the market xDD
<kinkinkijkin>
one was gateron yellows on the ansi plate with really, really good stabs (amazing stabs, super good stabs), and a generic styled pbt capset that were alright i guess
<kinkinkijkin>
the other was kailh box navies, iso plate, less good stabs (just noisier, still stabilized just as well), and the custom capset
<kinkinkijkin>
they were oem profile dark grey pbt caps with white front printing, and a dark cyan enter, lshift, iso extra key, and one other key
<HdkR>
Oh right, non-mechanical is nice for not having noise during voice calls :D
<kinkinkijkin>
unless you spend a lot of money and time :p
<macc24>
HdkR: my current keyboard is quite quiet
<macc24>
i no longer wake up neighbours on my late night coding sessions
<kinkinkijkin>
the gat yellows board was insanely quiet unless i hammered it, at which point it had a very nice sound
<kinkinkijkin>
not shrill, actually pretty deep and subdued
<macc24>
if i had a nickel for every time i heard "shut up your keyboard" at night i would have 2 nickels
<kinkinkijkin>
and it felt like pure wonderment
<macc24>
which isnt much but it happened 2 times
<kinkinkijkin>
tbh preferred the feel over my navies board but the navies board was easier to service and something broke on the yellows board that wouldve taken a few hours and some tools i didnt have to fix
<kinkinkijkin>
cause i put it together wrong
<icecream95>
alyssa: Can you kick all these spammers?