<Lyude>
I have spent all day bisecting a kernel regression on my laptop :(
<Lyude>
build #33
davidlt has joined #panfrost
<alyssa>
Oof
<alyssa>
So, trying to run any wayland egl apps (glmark, gears, etc) under sway results in a segfault in wl_proxy_marshal_constructor; we're fine under weston
<alyssa>
Wonder what's app
<Lyude>
alyssa: run the app with WAYLAND_DEBUG=1 and tell me what it says
raster has quit [Remote host closed the connection]
raster has joined #panfrost
afaerber has quit [Quit: Leaving]
afaerber has joined #panfrost
<tomeu>
robher: I see you have forked panfrost/linux and I have done that as well now
<tomeu>
robher: how would you like to share code at the end of our days? MRs? private branches?
<robher>
tomeu: probably easiest to just push branches in and we just check if there's any changes.
<tomeu>
sounds good
<tomeu>
looks like tomorrow will be my panfrost day :)
* Lyude
wishes she had a panfrost day :(
<alyssa>
daniels: Ahh... so Not My Bug (TM)? :P
<daniels>
alyssa: correct!
<alyssa>
Hooray! That happens about 0.1% of the time!
<alyssa>
:p
<alyssa>
robher: tomeu: BTW, for as many linux repos as we have cloned and forked and whatever, it's up to you if using any of them is actually a headstart
<robher>
alyssa: Are there others?
<alyssa>
The linux-panfrost changes never got to the point of, well, testability; it's certainly good reference code I think
<alyssa>
robher: I can't keep track :p hanetzer started writing a DRM driver from scratch (that's linux-panfrost, iirc). There's the fork of the mali_kbase driver. And there are... other forks of mali_kbase that are even more broken, do not touch
<alyssa>
:p
<robher>
alyssa: certainly with ARM's driver, there are no shortages of forks to use.
<alyssa>
Hehe
<alyssa>
robher: BTW, are you planning to write this module out-of-tree (at least for now)?
<alyssa>
At least on Chromebooks, reflashing the kernel is Not My Idea of Fun (TM), but with mali_kbase I can just recompile and insmod it over
<alyssa>
...although I guess I could do that too with an in-tree module, just setting the relevant CONFIG=m?
<davidlt>
Are there panfrost related talks at FOSDEM?
<robher>
alyssa: that's probably going to be a bit harder with a DRM driver because for one, ARM's driver probably has kernel version abstractions built in (#if LINUX_VERSION_CODE(...)) and is less dependent on kernel APIs.
<alyssa>
davidlt: Not to my knowledge
<alyssa>
robher: Oh.. well, in any event I can (and would probably prefer to) stick with Arm's driver until the DRM stuff is mature
<robher>
alyssa: It probably going to be easiest to maintain a CrOS specific backport while the development tracks mainline.
<raster>
davidlt: no idea, but i'd love to hear some... :)
<HdkR>
Graphics devroom is only one day this year as well
<raster>
lame
<raster>
:(O
<HdkR>
Not that big of a deal. I'm sure there will be more talks in the future :D
<raster>
people didnt submit enough talks to make it 2 days?
<HdkR>
I can only imagine
<HdkR>
Original call for talks only said one day as well. Could be the waning talks over the past couple years have made it be only one day
jernej has joined #panfrost
stikonas has joined #panfrost
pH5 has quit [Quit: bye]
paulk-leonov has quit [Ping timeout: 250 seconds]
paulk-leonov has joined #panfrost
Elpaulo has quit [Remote host closed the connection]
Elpaulo has joined #panfrost
pH5 has joined #panfrost
sy has quit [Ping timeout: 246 seconds]
sy has joined #panfrost
raster has quit [Remote host closed the connection]
BenG83 has quit [Quit: Leaving]
sy has quit [Quit: sy]
robertfoss_ has joined #panfrost
robertfoss_ has quit [Quit: WeeChat 2.3]
robertfoss_ has joined #panfrost
robertfoss_ has quit [Client Quit]
robertfoss_ has joined #panfrost
<ezequielg>
who wants hw decoding, rendered with panfrost ?
<HdkR>
hw decoding of what?
<robertfoss_>
vidya!
<ezequielg>
ｅｖｅｒｙｔｈｉｎｇ ?
<HdkR>
Oh snap, everything
<HdkR>
I don't know if I can open my mind that far
<ezequielg>
i'd like to decode the meaning of life. but we have to settle with video decoding.
<ezequielg>
mpeg2 at the moment.
<ezequielg>
h264 if the afternoon is long enough
<HdkR>
Nice
<ezequielg>
i needed to force the vop big to be used. should check if there's a way of asking it politely.
afaerber has quit [Quit: Leaving]
raster has joined #panfrost
<cyrozap>
<alyssa> At least on Chromebooks, reflashing the kernel is Not My Idea of Fun (TM)
<cyrozap>
kexec, yo
<cyrozap>
Load and boot Linux from Linux
<Lyude>
if you're lucky
<Lyude>
:)
<cyrozap>
What do you mean?
<anarsoul>
ezequielg: what platform?
<anarsoul>
rk3399?
<Lyude>
cyrozap: I have many a machine where kexec doesn't work, although it'll probably work fine on arm
<Lyude>
some hardware is just bad at resets
<anarsoul>
Lyude: rather some drivers are bad at shutting down hardware properly
<cyrozap>
Lyude: Oh, interesting... I've only ever used kexec on my Talos, so I just assumed it worked everywhere.
<Lyude>
anarsoul: I dunno, I've yet to see someone reset r600 series GPUs correctly, even when going off hw docs
<Lyude>
cyrozap: you have a talos?
<HdkR>
Lyude: Isn't that reset problem a firmware issue?
<Lyude>
HdkR: kind of?
<ezequielg>
anarsoul: yep
<anarsoul>
ezequielg: nice!
<cyrozap>
Lyude: I do :)
<Lyude>
HdkR: I tried getting radeon to reset those GPUs properly at one point and the UVD/VCE always gets stuck
<Lyude>
granted though, I have suspecions about that since i haven't tried again at that for a while
<cyrozap>
Unfortunately, I can't get my RX Vega 64 to work with Debian Testing, but at least llvmpipe and pocl are pretty fast on this machine...
<Lyude>
cyrozap: seriously? D:, or are these ppc specific related amdgpu issues
<HdkR>
Lyude: Like, last I knew you could shut it down completely sanely from the host side and it wouldn't matter regardless since the firmware just won't take bits down correctly :P
<HdkR>
But I don't do AMD things, could be mistaken
<Lyude>
yeah, I need to give another shot at trying to get that reset working
<Lyude>
last time I tried it I knew a lot less about drivers
<Lyude>
/hw/etc
<HdkR>
:D
<Lyude>
cyrozap: huh, was hoping i could help but I'm not sure I can D:
adjtm has joined #panfrost
adjtm_ has quit [Ping timeout: 240 seconds]
<cyrozap>
Lyude: That said, my larger concern is that I can't use it for any GPU compute stuff because ROCm only supports x86 and ARM :/
<Lyude>
ugh
<Lyude>
gpu compute sucks atm
<HdkR>
Lyude: But just think, OpenCL-Next will save us
<Lyude>
wait
<Lyude>
they're still going with OpenCL?
<HdkR>
Yep
<cyrozap>
I wish Clover wasn't a steaming pile of garbage (no offense to anyone who works on that, but seriously I've gotten segfaults every time I've tried to use it) and I wish all the CL runtimes that use LLVM would statically link it so that OpenCL ICDs would work properly (though really it's LLVM's fault for having global state in the first place).
<HdkR>
Supposedly OpenCL-next is going to be a significant API break
<Lyude>
cyrozap: I think most of the people working on clover will agree with you :p
<Lyude>
it has been suffering from lack of people power for a long time
<HdkR>
I think there is also an initiative to make Vulkan compute be able to take on things that OpenCL is used for as well?
<Lyude>
that woul be very nice :)
<Lyude>
I'd like to see a world where gpu compute is a bit more then which manufacturer can monopolize things first
* HdkR
throws cuda under a bus quick
<Lyude>
HdkR: hehe, to be fair though rocm isn't much better
<Lyude>
in more ways then one
<HdkR>
Let's just use HIP
<HdkR>
Oh, that's technically part of rocm
<cyrozap>
Let's all just switch to GPU assembly and ioctl(). What could go wrong?
afaerber has joined #panfrost
<alyssa>
robher: I didn't mean a CrOS kernel (my dev machine is running 4.20 mainline with no patches minus, you know, mali_kbase). Just... you know what, nvm, I was being silly :p
<robher>
alyssa: That makes it a lot easier as a CrOS kernel would also have lots of backports and is its own thing.
<alyssa>
Indeed
pH5 has quit [Ping timeout: 240 seconds]
<alyssa>
robher: I noticed anholt pushed some but not at all of the kmsro patches -- does that mean I'm safe to merge your MR, or do the other patches matter too?
<robher>
alyssa: I pushed the rest too. You kind of need the etnaviv one only because it restructures things to add other drivers like panfrost.