alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://freenode.irclog.whitequark.org/panfrost - <daniels> avoiding X is a huge feature
uis has quit [Ping timeout: 240 seconds]
uis has joined #panfrost
<HdkR> alyssa: probably need to pad to some size after because of prefetching behaviour?
<alyssa> HdkR: Guess so 🤷
<HdkR> Nvidia has the same limitation :P
<alyssa> boring
<chewitt> felipealmeida Khadas VIM3L should be well within budget as is a Google AOSP reference device (meaning lots of people care about Linux support for it)
<chewitt> technically there is no upstream support for the BPi M5 board yet, only vendor kernel
<macc24> oh there isn't
<chewitt> although it appears to be a clone/rip-off of the Odroid C4
<chewitt> and I have a sample en-route to test the theory.. should arrive on 27/12
<chewitt> so likely there's a dts in my 5.11 branch before new year
<macc24> you know mt8183?
<chewitt> nope, not at all
<macc24> so there is a fast arm chip used in new chromebooks with bifrost gpu
<macc24> and it has crap mainline linux support
<chewitt> I hear good things about mediatek though, and compared to Amlogic, Allwinner, Rockchip, etc. they are clearly engaged in upstreaming their stuff, which looks good
<macc24> pls anyone help with it
<macc24> i can't get sound to work on it
<chewitt> is this a new chip or something around for a while?
<macc24> oldest device released this year
<alyssa> so... new chip
<HdkR> SoC itself was technically released late last year :P
<macc24> so new, that i have to use kernel fork
dstzd has quit [Quit: ZNC - https://znc.in]
<chewitt> looks like support is underway..
<macc24> still isn't merged
jernej has quit [Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net]
<macc24> it works in cadmium, but not in the mainline kernel
<chewitt> looks to be fairly active
<chewitt> "Rome wasn't built in a day"
<chewitt> upstreaming takes ... time
<macc24> i got almost everything working in my linux distro, aside from internal speakers and microphones
jernej has joined #panfrost
<chewitt> media capabilities are always last to arrive
jernej has quit [Client Quit]
<macc24> there are drivers in linux, they just refuse to work lol
<chewitt> stuff often gets written in isolation or with a particular use case in mind
<chewitt> have you pinged the mediatek mailing list with a bug report?
<macc24> i tried to subscribe to mediatek mailing list
<chewitt> no need to subscribe .. just send an email
<macc24> tbh i never thought about that
<chewitt> use ./scripts/get_maintainer.pl on the driver code and cc the author of the driver too
<macc24> ill write them an email after winter holidays this year
<chewitt> some lists/authors are great, others suck, but "nothing ventured, nothing gained"
<chewitt> always share a link to your kernel sources in a public repo so people can review what you included
<macc24> i use 100% upstream audio code
icecream95 has joined #panfrost
stikonas has quit [Remote host closed the connection]
vstehle has quit [Ping timeout: 264 seconds]
kaspter has joined #panfrost
<felipealmeida> chewitt: I bought a odroid c4, but that is likely going to take a month or more to arrive from USA. I'll be borrowing two Hikey 960 from Ricardo Salveti for now.
<felipealmeida> chewitt: he mentioned hikey 960 are also aosp references, so I'm likely to be good on upstream support if I'm not unlucky
<felipealmeida> thanks for your tip
<chewitt> I know nothing about the HiKey boards, but 96boards devices usually have good support
jernej has joined #panfrost
jernej has quit [Client Quit]
jernej has joined #panfrost
jernej has quit [Client Quit]
jernej has joined #panfrost
jernej has quit [Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net]
dstzd has joined #panfrost
jernej has joined #panfrost
<macc24> felipealmeida: if soc has mainline support it shouldn't be that hard to mainline a board which has open source kernel availabl
<macc24> available*
<alyssa> icecream95: finally got new IR to clear GLES2 ci
<macc24> \o/
<alyssa> haven't tried stk yet but I fixed a bunch of RA issues today so 🤷
<macc24> alyssa: do you know if bifrost will be good enough to not crash on me for no reason when mesa 21.0.0 releases?
<macc24> s/will/should/
<alyssa> probably not
<macc24> i'd like to have all hardware working in cadmium before stable mesa with good enough bifrost support releases
<macc24> oh good so i have lots of time
<alyssa> lol
<icecream95> alyssa: I've rebased my CL branch: https://gitlab.freedesktop.org/icecream95/mesa/-/commits/cl
<icecream95> I have noticed a lot of "compute" instructions (atomics, barriers, etc.) disappearing during DCE…
<HdkR> Sounds like you need to tag instructions with side-effects and not DCE them away
<alyssa> uhhhh
<alyssa> on which arch?
* macc24 guesses arm64
<alyssa> gpu
<icecream95> alyssa: Bifrost
<macc24> g72?
<alyssa> oh, well, the DCE pass predates any of those instructions being there so.. ;)
<alyssa> midgard had LDST_SIDE_FX tagging
<alyssa> should be easy enough to add corresponding flags in the XML
<alyssa> will also need reordering flags, this is broken in midgard but I digress
<alyssa> "TODO: Why is batch->framebuffer.gpu set in the first place?" That... is a good question
<alyssa> The boring answer is "because the blob does that and we were RE'ing so monkey see monkey do"
<alyssa> The real answer (why does the blob set a framebuffer for compute job?) is unknown to me.
<macc24> b l o b s do weird stuff
<alyssa> IIRC the OpenGL blob does for compute shaders, but the OpenCL blob does not.
<alyssa> (There is no architectural reason for this that I know of.)
<alyssa> The shortest MBS parser in existence! :p
<alyssa> anyway, super neat stuff :) and you're now the first one other than me to use the new bifrost builder infrastructure, hope it doesn't suck too much
<alyssa> How's my drivering? ★★★☆☆
<HdkR> 3/5?! Who is the scoundrel that rated low?
<alyssa> ✋
<HdkR> Need to pay a bot network to raid this yelp listing. Boost the rating quick
camus has joined #panfrost
kaspter has quit [Ping timeout: 272 seconds]
camus is now known as kaspter
davidlt has joined #panfrost
camus has joined #panfrost
kaspter has quit [Ping timeout: 246 seconds]
camus is now known as kaspter
davidlt has quit [Ping timeout: 268 seconds]
vstehle has joined #panfrost
davidlt has joined #panfrost
awordnot has quit [Quit: Ping timeout (120 seconds)]
awordnot has joined #panfrost
phh has quit [Quit: No Ping reply in 180 seconds.]
phh has joined #panfrost
icecream95 has quit [Ping timeout: 240 seconds]
rando25892 has quit [Ping timeout: 256 seconds]
stikonas has joined #panfrost
raster has joined #panfrost
<robmur01> felipealmeida: beware that Hikey960 has G71, which doesn't work well with Panfrost since it needs various operations emulating in software which we don't handle (yet?)
<HdkR> Hikey970 to the rescue?
<robmur01> also I'm not sure there's an upstream display driver (although I think USB and UFS support may have landed since the last time I tried booting the one I have)
<macc24> robmur01: G72 is working mostly fine in panfrost from my experience
<HdkR> But that's why they were talking about G71 :)
<robmur01> macc24: so is T860, and to a lesser extent T620. I don't really have time today to evaluate *all* the GPUs that Hikey960 doesn't have, but maybe we can pick this up again in the new year :P
kaspter has quit [Ping timeout: 256 seconds]
camus has joined #panfrost
camus is now known as kaspter
<br_> where to set PAN_MESA_DEBUG=trace ?
<br_> 'PAN_MESA_DEBUG=trace startx' does not enable debugging
<HdkR> yep, in the environment variable just like that
<HdkR> Probably need PANDECODE_DUMP_FILE set to something as well
<HdkR> Otherwise it dumps to file
<HdkR> `pandecode.dump`
<HdkR> stderr there works if you just want output in stderr
<macc24> what does PAN_MESA_DEBUG=trace do?
<br_> hmm I see that dev->debug still does not include PAN_DBG_TRACE flag
<br_> macc24: decodes a jc. see PAN_DBG_TRACE in pan_job.c
archetech has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
davidlt has quit [Ping timeout: 240 seconds]
<br_> alyssa: I found that I had problems with drmSyncobjWait(), in particular converting args->timeout_nsecs to jiffies :)
<br_> I see that drmSyncobjWait() does actually waits now, but a new problem arise: the first job submitted never completed. So the job interrupt never arrives :(
alpernebbi has joined #panfrost
<alyssa> br_: PAN_MESA_DEBUG=trace as env suffices, but startx clobbers env vars
* tlwoerner has an odroid-c4 and odroid-n2+ arriving today (based on yesterday's conversation)
<alyssa> weston will be easier to bring up first
<tlwoerner> the hikey 960 doesn't seem to be available anywhere
<tlwoerner> nor the 970
* alyssa looking into fixing up the G71 issues
<alyssa> Should be relatively painless once the builder lands
<HdkR> Oh wow, the 970 is out of stock
<HdkR> Looks like someone is just going to have to make a Kirin 9000 board to replace it
<tlwoerner> HdkR: when was it released?
<HdkR> I ordered my 970 May 2018
<macc24> :o kirin 9000
<tlwoerner> i find seeed is really bad for stock
<tlwoerner> e.g. you can buy a rock pi x, but there aren't any heatsinks in stock (??)
<macc24> that kirin chip is fast
<HdkR> It's a pretty cool chip
<robmur01> thermally? :P
<alyssa> ⛷️
<HdkR> Nah, it wears sunglasses at night
<alyssa> 😎
davidlt has joined #panfrost
camus has joined #panfrost
kaspter has quit [Ping timeout: 240 seconds]
camus has quit [Read error: Connection reset by peer]
kaspter has joined #panfrost
kaspter has quit [Quit: kaspter]
tgall_foo has quit [Quit: My iMac has gone to sleep. ZZZzzz…]
archetech has quit [Quit: Konversation terminated!]
popolon has joined #panfrost
kaspter has joined #panfrost
nlhowell has joined #panfrost
tgall_foo has joined #panfrost
tomboy64 has quit [Remote host closed the connection]
tomboy64 has joined #panfrost
kaspter has quit [Ping timeout: 264 seconds]
kaspter has joined #panfrost
urjaman has quit [Read error: Connection reset by peer]
urjaman has joined #panfrost
kaspter has quit [Quit: kaspter]
tlwoerner has quit [Ping timeout: 256 seconds]
alpernebbi has quit [Quit: alpernebbi]
davidlt has quit [Ping timeout: 240 seconds]
<solidhal> alyssa daniels and anyone else interested in gru-kevin/samsung chromebook v1 cursor lag/display lag
<solidhal> found the culprit
<solidhal> Writeup for those that are curious here: https://github.com/SolidHal/PrawnOS/issues/225#issuecomment-750451173
<solidhal> TL;DR We spent too long confirming that Panel Self Refresh was disabled
<solidhal> Working on getting the change upstreamed
<alyssa> solidhal: oh boy
<daniels> solidhal: nicei! :)
<solidhal> thanks for suggesting perf btw, I had completely forgotten about how awesome it can be for nailing issues like this
<daniels> really glad you managed to figure out what it was
<alyssa> solidhal: worth uprevving my kernel over? :<
<solidhal> alyssa depends how much the display lag bothers you :D
<solidhal> dri-devel@lists.freedesktop.org is blocking me for some reason... great
<alyssa> solidhal: subscribe?
<solidhal> seems they need to manually confirm my subscription
<alyssa> ah.. apparently there's a holiday this week in the US?
<solidhal> there is indeed some sort of holiday hahaha
<solidhal> ah, there we go. Time to try again...
<solidhal> intsta-fail great
tlwoerner has joined #panfrost
<solidhal> guess I'll mail the list owner...
<solidhal> oh, mail to the list owner is *also* blocked... cool
raster has quit [Quit: Gettin' stinky!]
<macc24> time to NUKE chromeos out of my laptop
<alyssa> macc24: careful that partner :p
<macc24> alyssa: the lag issue was literally the only reason why i kept chromeos on my kevin
<alyssa> gotcha
<alyssa> i never noticed it until solidhal pointed it out lol
<macc24> lol
<solidhal> Sorry again for that alyssa, ignorance is bliss ;)
<macc24> so when i borrow kevin's firmware from chromeos, add solidhal's patches and make installing to emmc work on it, cadmium would have 2 (mostly) functioning devices
<macc24> mt8183-mt6358-ts3a227-max98357a not working is a privacy feature in case some malicious code gets on device to record anything that user says
raster has joined #panfrost
mixfix41_ has quit [Ping timeout: 256 seconds]
icecream95 has joined #panfrost
<icecream95> I still have ChromeOS on my duet. For research purposes only, of course ;)
<alyssa> macc24: what is mt8183-mt6358-ts3a227-max98357a?
<macc24> icecream95: unacceptable
<alyssa> icecream95: hey, ChromeOS is decent for its use case
<macc24> ;D
<alyssa> but apparently writing gfx drivers is not its use case
<macc24> alyssa: sound driver for duet, i probably have a typo somewhere in this longboi name
<alyssa> macc24: jokes on you i don't say anything anyway! :p
<macc24> oh btw, merry <insert here local winter holiday name> everyone!
<alyssa> happy festivus
<icecream95> Winter? That's six months away!
<macc24> icecream95: i meant in the "default" hemisphere
<alyssa> macc24: Hemisphere supremacy is against the project CoC.
<macc24> D:
<solidhal> < 4th quarter of year holiday >
<icecream95> solidhal: The start of the year is in ~June, so that would still be three months away for me
<solidhal> ahhhhhhhhhh
<solidhal> June 1st is 1/1/XXXX ?
<alyssa> solstice :)
<macc24> <around this month holiday>
<alyssa> icecream95: happy summer solstice regardless
<solidhal> no more happy nothings for anyone, cancel all holiday things :D
<alyssa> icecream95: sad summer solstice regardless
<solidhal> 👍
rcf has quit [Quit: WeeChat 2.9]
* icecream95 doesn't know what all this "cancelling" buisness is all about
<HdkR> Winter is cancelled, pack it up
<macc24> 22
<macc24> oops
<solidhal> but for icecream95 summer is canceled, *now* is canceled
<icecream95> Is summer cancelled? What for? Is there a war going on or something?
<macc24> icecream95: you'd better hope that summer is canceled
uis has quit [Ping timeout: 246 seconds]
<macc24> (that was reffering to fact that parts of australia tend to catch on fire :( in summer)
* icecream95 wonders where this "Australia" is. Probably it's just a government conspiracy theory
uis has joined #panfrost
rcf has joined #panfrost
<HdkR> But if AU doesn't exist, that means NZ can't exist, and I adore the NZ accent D:
raster has quit [Quit: Gettin' stinky!]