Guest169 has quit [Remote host closed the connection]
abby has joined #asahi-dev
abby is now known as Guest170
Guest170 is now known as abby
jn has quit [Server closed connection]
jn has joined #asahi-dev
aleasto has quit [Server closed connection]
aleasto has joined #asahi-dev
hxliew has quit [Server closed connection]
hxliew has joined #asahi-dev
Esmil has quit [Server closed connection]
Esmil has joined #asahi-dev
nullroute has quit [Server closed connection]
nullroute has joined #asahi-dev
chadmed has quit [Quit: Konversation terminated!]
chadmed has joined #asahi-dev
sjg1 has quit [Server closed connection]
sjg1 has joined #asahi-dev
systwi_ has quit [Server closed connection]
systwi has joined #asahi-dev
os has quit [Server closed connection]
os has joined #asahi-dev
Skirmisher has quit [Server closed connection]
Skirmisher has joined #asahi-dev
mmarchini____ has quit [Server closed connection]
mmarchini____ has joined #asahi-dev
Manouchehri has quit [Server closed connection]
Manouchehri has joined #asahi-dev
nathanchance has quit [Server closed connection]
nathanchance has joined #asahi-dev
jole_ has quit [Server closed connection]
jole has joined #asahi-dev
mini has quit [Server closed connection]
mini has joined #asahi-dev
blasty_ has quit [Server closed connection]
blasty has joined #asahi-dev
nyx_o has quit [Server closed connection]
nyx_o has joined #asahi-dev
paddatrapper_ has quit [Server closed connection]
paddatrapper_ has joined #asahi-dev
ids1024 has quit [Server closed connection]
ids1024 has joined #asahi-dev
zjstraus has quit [Server closed connection]
zjstraus has joined #asahi-dev
gabuscus has quit [Server closed connection]
gabuscus has joined #asahi-dev
codingkoopa9 has quit [Server closed connection]
codingkoopa9 has joined #asahi-dev
greguu has quit [Server closed connection]
greguu has joined #asahi-dev
Poplar has quit [Server closed connection]
Poplar has joined #asahi-dev
grange_c has quit [Server closed connection]
grange_c has joined #asahi-dev
xcpy0 has quit [Server closed connection]
xcpy0 has joined #asahi-dev
as400 has quit [Server closed connection]
as400 has joined #asahi-dev
kode54 has quit [Server closed connection]
kode54 has joined #asahi-dev
DragoonAethis has quit [Server closed connection]
DragoonAethis has joined #asahi-dev
rqou_ has quit [Server closed connection]
rqou_ has joined #asahi-dev
tmlind has quit [Server closed connection]
tmlind has joined #asahi-dev
kloenk has quit [Server closed connection]
kloenk has joined #asahi-dev
tpw_rules has quit [Server closed connection]
tpw_rules has joined #asahi-dev
chadmed_ has joined #asahi-dev
chadmed_ has quit []
Dcow has quit [Remote host closed the connection]
Dcow has joined #asahi-dev
SSJ_GZ has joined #asahi-dev
blazra has quit [Server closed connection]
blazra has joined #asahi-dev
amarioguy has quit [Server closed connection]
amarioguy has joined #asahi-dev
chadmed_ has joined #asahi-dev
Mrmaxmeier has quit [Server closed connection]
Mrmaxmeier has joined #asahi-dev
opticron has quit [Server closed connection]
opticron has joined #asahi-dev
gff_ has quit [Server closed connection]
gff_ has joined #asahi-dev
Gaelan has quit [Server closed connection]
Gaelan has joined #asahi-dev
greguu has quit [Ping timeout: 480 seconds]
winter has quit [Server closed connection]
winter has joined #asahi-dev
Dcow_ has joined #asahi-dev
khilman has quit [Server closed connection]
khilman has joined #asahi-dev
Dcow has quit [Ping timeout: 480 seconds]
jonmasters_ has quit [Server closed connection]
jonmasters_ has joined #asahi-dev
liszt has quit [Server closed connection]
liszt has joined #asahi-dev
jbowen has quit [Server closed connection]
jbowen has joined #asahi-dev
qyousef has quit [Server closed connection]
qyousef has joined #asahi-dev
Puto has quit [Server closed connection]
Puto has joined #asahi-dev
msteffen has quit [Server closed connection]
msteffen has joined #asahi-dev
jluthra has quit [Remote host closed the connection]
jluthra has joined #asahi-dev
Avaflow has quit [Server closed connection]
Avaflow has joined #asahi-dev
JTL has quit [Server closed connection]
JTL has joined #asahi-dev
Dcow_ has quit [Remote host closed the connection]
princesszoey has quit [Server closed connection]
princesszoey has joined #asahi-dev
ricekot has joined #asahi-dev
ricekot has quit []
ricekot has joined #asahi-dev
ricekot has quit []
<sven>
jannau: alright, will send out the fix later today then
turo has quit [Server closed connection]
turo has joined #asahi-dev
Graypup_ has quit [Server closed connection]
<sven>
looks like it also works with my altmode code which is probably the worst possible situation since you get a few successive role switches without any i2c traffic inbetween
Graypup_ has joined #asahi-dev
MajorBiscuit has joined #asahi-dev
yrlf has joined #asahi-dev
chadmed_ has quit [Remote host closed the connection]
leitao has joined #asahi-dev
chadmed_ has joined #asahi-dev
MajorBiscuit has quit [Ping timeout: 480 seconds]
kloenk has quit [Remote host closed the connection]
kloenk has joined #asahi-dev
Dcow has joined #asahi-dev
Dcow has quit [Remote host closed the connection]
Dcow has joined #asahi-dev
leitao has quit [Ping timeout: 480 seconds]
Guest29 is now known as null
null is now known as Guest225
Guest225 has quit [Quit: oh no.]
null_ has joined #asahi-dev
null_ is now known as null
chadmed_ has quit [Remote host closed the connection]
<kettenis>
marcan: time for another update to the cpufreq patch series?
<marcan>
kettenis: Already on it
<marcan>
doing a test build now :)
<marcan>
jannau: do you have M2/misc DCP updates? I should probably spin some new kernels
<_jannau_>
macos powers dcp off completely on the imac. I don't remember seeing that on mac mini or mbp 14"
<sven>
doesn‘t it even keep dcpext alive when nothing’s connected?
<_jannau_>
I haven't traced dcpext. this is just dcp with turn the display off after 1 minute but don't sleep
<marcan>
_jannau_: fun. thanks!
<marcan>
cpufreq v4 sent, wonder if it'll make it this week
<marcan>
should be pretty much ready, if I get the acks and Vishar merges the driver per se I can push the DT stuff and send arnd one last pull for it
<marcan>
I really should start putting stuff in a TODO list, I probably should've sent this last week
<marcan>
_jannau_: do I need to merge both for m2?
<marcan>
I mean are the independent branches?
<marcan>
*they
<_jannau_>
they are independent, the m2 branch has the panel dimensions in the dt which are only used by a commit in the fixes branch but they shouldn't cause problems
<eiln>
it's really just MAC hardware, so whatever it can be hooked to
<eiln>
also some exynos stuff
<ChaosPrincess>
i thought until somewhat recently, the approach for this kind of stuff was to just dump it in char/misc, an do all the fancy stuff in userspace.
<ChaosPrincess>
but that made kernel folks unhappy
<sven>
as i said, i'd talk to the drm folks and/or alyssa
<eiln>
alright ty!
<sven>
alyssa should be here or in #asahi-gpu sometimes
<povik>
running opencl on ane when
<sven>
I wonder if it's versatile enough for that :D
chengsun_ has quit [Read error: Connection reset by peer]
chengsun has joined #asahi-dev
eiln has quit [Remote host closed the connection]
eiln has joined #asahi-dev
<povik>
jannau: alyssa: i wrote up what is per my understanding the boilerplate to add audio to DCP. take a look if curious
<sven>
povik: I think the proper solution to that drvdata thing is to just create a new device from code. nvme does that as well for its nvme_crrl
<sven>
*nvme_ctrl
chengsun has quit [Ping timeout: 480 seconds]
chengsun has joined #asahi-dev
<povik>
yeah, makes sense
<sven>
in other thunderbolt and „sven is dumb“ news: part of the acio MMIO wasn’t „enabled“ because I literally got the address wrong in the device tree *sigh*
<sven>
pciec still SErrors though, but maybe I need something connected before that comes up
<povik>
i guess you double-checked the address there? :D
<sven>
yup :D
chengsun has quit [Ping timeout: 480 seconds]
leitao has joined #asahi-dev
gff_ has quit [Ping timeout: 480 seconds]
leitao has quit []
<ChaosPrincess>
Wasnt there some people who tried to put asahi on checkrainable ipads?
alyssa has joined #asahi-dev
<alyssa>
eiln (for when they're back)--
<alyssa>
NPUs in mainline are in a bit of a weird spot right now
<alyssa>
Stuff isn't too standardized yet, as you're seeing
<alyssa>
I think I'd like to understand the capabilities of the ANE more before I could give any advice
<alyssa>
Some NPUs (e.g. Vivante) are general purpose compute accelerators
<alyssa>
the Vivante NPU is literally just a Vivante GPU with all the graphics support stripped out
<alyssa>
hence is supported by the existing etnaviv GPU stack, supporting (only) OpenCL in userspace via mesa
<alyssa>
IMHO that design doesn't make a lot of sense on SoCs that have competent CPUs and GPUs, either of which can run OpenCL excellently, why add NPU into the mix?
<alyssa>
On the other extreme you have NPUs that are fixed function neural network hardware
<alyssa>
which is more sensible if you're going to have an ANE at all
<alyssa>
It does not make sense to support that in Mesa
<alyssa>
but we don't currently have a home for userspace drivers for hardware of that class (Mesa for GPUs, libcamera for camera processing, what for NPUs?)
<alyssa>
I think the current state-of-the-art is adding a tensorflow backend for your hardware and yeeting
<alyssa>
(Because there's not a standard API, ie a hypothetical Khronos OpenML, that this hardware would implement... In effect the tensorflow API would be that API in this case...)
<alyssa>
Usually that's downstream char/misc devices in the kernel with custom vendor ioctls and a downstream fork of tflite or whatever ...
<alyssa>
I think Habana Labs (only) are doing things "properly" with upstream/mainline approach now
<alyssa>
and ruminations of an accel subsystem as mentioned above
<alyssa>
so it's worth looking at what they're doing
<alyssa>
as the closest prior art for "ANE done right"
<alyssa>
("ANE done wrong" would be the downstream vendor approach to get some cool pics for Twitter/Mastodon and never end up getting shipped...)
<alyssa>
this isn't my 'area', I'm not the expert here
<alyssa>
but every so often the ML folk pass by my virtual cubicle at work :-p
<sven>
she said after giving amazing advise on what to do :D
<sven>
*advice
<alyssa>
sven: NPUs are just GPUs that suck at rasterization....
<sven>
huh, i guess
<alyssa>
I do want to understand the complexity of the ANE instruction set
<alyssa>
Is this "we have a couple of registers for network weights and the rest is fixed"?
<alyssa>
Is this "it can run C code"?
<alyssa>
Something in the middle?
<sven>
eiln wrote a bit in -gpu and -re
<alyssa>
Habanalabs is using LLVM for their commpiler ... while I won't stop you from doing so, as a maintainer of a competing project I won't be able to help if you go that route ;-)
<sven>
it sounded like a linear algebra accelerator for float16
<sven>
so a bunch of simple math operations that you can chain together
<marcan>
alyssa: eiln said it's basically a big MAC unit
<marcan>
it doesn't have load/store
<kettenis>
that's what most "machine learning" oriented instruction set extension seem to be
<marcan>
just a giant pile of registers or something
<alyssa>
right... so quite a bit simpler than Habana I suspect
<alyssa>
not sure there'd be any value in using LLVM or NIR, they may just get in the way..
<alyssa>
so yeah... nothing standardized at that point... do what's best in the kernel side (probably DRM driver), do your own thing in userspace (libane with your own cute little compiler?), plumb libane into tensorflow and pytorch or whatever? IDK
<alyssa>
you're not the only one having these problems
<alyssa>
lots of people are being paid big bucks to not solve these (-:
<marcan>
alyssa: also I didn't know you had random ML knowledge to drop :p
<alyssa>
marcan: go to sleep
<sven>
:D
<alyssa>
marcan: and I'm a GPU developer with a minor in math, of couse I have random ML knowledge to drop :p
<alyssa>
(...strictly i need to graduate first, but I have enough credits for a minor...)
<marcan>
lol
<marcan>
fine, fine
<sven>
all you need for ML is linear algebra 101 anyway :P
<sven>
and maybe some high school calculus
<alyssa>
thank you Dr. Peter
<marcan>
it's not hard like CRCs!
<sven>
lol
<marcan>
also I implemented waifu2x in opencl at some point so I'm supposed to know about this?
<marcan>
(not very well mind you)
<kettenis>
I did a course on NN 30 years ago, does that count?
<povik>
pfff, everybody knows ML was invited just a couple of years ago
<alyssa>
povik: i see your attempt at nerdsniping dcp audio
<marcan>
sven: dunno if I ever told you about that one? waifu2x was some mess based on one of those giant ML frameworks in lua or something
<marcan>
so here's an idea: sven can be povik's minion, povik can be alyssa's minion, and alyssa can be sven's minion!
<povik>
(or jannau of course, that's a viable target too)
<alyssa>
minions are great for doing the stuff you were supposed to do yourself months ago
<alyssa>
povik: pretty sure sven is already jannau's minion
<marcan>
alyssa: wait come to think of it isn't jannau your minion?
<sven>
lol
<alyssa>
19:22 <@alyssa> minions are great for doing the stuff you were supposed to do yourself months ago
<povik>
at this point i'm confused and need a graph before we start discussing minions further
<alyssa>
marcan: hmm by that defintiion yes
<alyssa>
fun fact: all the RK3399 chromebooks are named after Despicable Me characters
<marcan>
we're already in non-DAG territory, this isn't going to end well
<alyssa>
this laptop is "Kevin", the baseboard is "Gru"
<marcan>
ha
Guest237 has quit [Quit: Guest237]
<jn>
chromebook codenames are fun
<alyssa>
truly
<alyssa>
idk what "asurada" references
nicolas17 has joined #asahi-dev
* mps
owns Kevin-Gru :)
<alyssa>
mps: still my favourite laptop
<mps>
alyssa: yes, easy to carry around
<mps>
less than 1Kg and small dimensions
<alyssa>
so good
<alyssa>
and it runs one of my graphicsdrivers! ;-p
<alyssa>
yknow before you could just get a macbook for that :-p
<mps>
alyssa: yes, I use your GPU driver on it. never told thank you and now I have chance: thank you for good work
<alyssa>
aww
<alyssa>
you're very welcome :-)
<mps>
(though we argued one on #panfrost but I hope this is forgotten ;) )
<mps>
s/one/once/
<alyssa>
if it didn't result in a ban, we good :-p
<mps>
ah, so no ban is good. there were no ban
<alyssa>
:-D
<alyssa>
I was talking to dougall about ^ yesterday, how in some ways panfrost has been wildly successful, and as a result is wildly 'boring'
<alyssa>
Its success means that it's become infrastructure, you know, how often do you think about the people maintaining your USB drivers? or NVMe?
<alyssa>
when they break and you go to the bug tracker/mailing list/yell into the twitter void, sure
<alyssa>
other than that? \shrug/
<mps>
I was early user when it had bugs so it was not borring :)
<alyssa>
ahahaha yes well :)
<alyssa>
dougall: observed that success for Asahi means becoming boring in precisely the same way
<mps>
well, asahi becoming somewhat boring for me now
<alyssa>
in 5 years or so, some 13yo kid is going to install Fedora on his hand-me-down M1 MacBook Air and everything will just work and she'll never know we existed :-)
<mps>
now is full year I'm using m1 as daily driver, started with usb wifi and usb sound
<alyssa>
:-D
<alyssa>
1yr + 3 months for me
<alyssa>
lol, I started using M1 at work before any graphics were alive, just using a TTY to compile stuff and run vim
<mps>
if I don't need firefox I could still be on tty
<mps>
oh, btw, gpu driver on m1pro works very good with wayland/sway
<alyssa>
Excellent ^^
<Tramtrist>
good news mps
<mps>
Tramtrist: xorg works also but not so fine as wayland
<Tramtrist>
i onlky use swaywm now ;)
<alyssa>
i like gnome
<alyssa>
on the mac anyway
<alyssa>
sway on the chromebooks
<Tramtrist>
im holding out on getting the gpu enhancement running until its in edge at the earliest
<alyssa>
Tramtrist: thank you
<Tramtrist>
Im assuming youll let folks know when we can help in testing
zzywysm_ has joined #asahi-dev
zzywysm has quit [Ping timeout: 480 seconds]
<alyssa>
Yep!
chengsun has joined #asahi-dev
<alyssa>
lina and I have a mostly-unwritten todo list of release blockers
<kettenis>
alyssa: are the UAPI ioctls taking shape?
<kettenis>
there are a few subtle differences between Linux and *BSD in ioctl handling and it would be good if we can avoid those
<kettenis>
the most important one is that an ioctl that fails (returns an error) can not copy out its argument struct to userland
Dcow has quit [Remote host closed the connection]
chengsun_ has joined #asahi-dev
chengsun has quit [Ping timeout: 480 seconds]
chengsun_ has quit [Quit: Quit]
<alyssa>
lina: ^^
chengsun has joined #asahi-dev
Dcow has joined #asahi-dev
chengsun has quit [Ping timeout: 480 seconds]
Dcow has quit [Ping timeout: 480 seconds]
SSJ_GZ has quit [Ping timeout: 480 seconds]
kettenis has quit [Remote host closed the connection]