18:36 <jannau> omg it works!

18:37 <jannau> 8GB imac, booted via hypervisor

19:51 <joske> \o/

19:54 <joske> jannau: can it boot over u-boot too?

20:24 <jannau> joske: I haven't tried, it's a little bit annoying to go from remote booting to u-boot

20:25 <joske> couldn't you chainload a m1n1 with u-boot?

20:25 <joske> you may not have any grub installed I guess

20:27 <joske> just asking theoretically, if you don't want to for whatever reason, no worries

20:40 <karolherbst> lina: how does your willingness of reviewing userspace code looks like? (or anybody who wants to get into mesa development). Would be cool to get some Rust related reviews on the stuff I write, but at the same time I also don't want to distract you from keeping up with your incredible kernel work :D So, my hope is others here who want to get into mesa and/or linux development might be interested.. dunno. I think I am just thinking out

20:40 <karolherbst> loud here

20:52 <jannau> joske: works through u-boot with this http://paste.debian.net/1255646/ patch, haven't tried without the "no-map" and u-boot

20:53 <jannau> this only works if the system has 8GB ram, for 16GB systems the reg values need to adjusted

20:56 <joske> jannau: thx a lot!

20:57 <joske> any idea what these adjustments might be?

20:58 <jannau> yes

20:59 <jannau> I know how to retrieve the necessary values from the ADT so that m1n1 can fill the values

20:59 <joske> for that I would have to boot macOS?

21:00 <joske> or can you do that from a non-16GB machine?

22:07 <jannau> joske: https://github.com/jannau/linux/tree/omg-it-works-16gb and https://github.com/jannau/m1n1/tree/omg-it-works-16gb still works on the 8GB system and removed one mem layout dependency

22:07 <jannau> I can test tomorrow on a 16gb mac mini

22:09 <jannau> lina: ^^^ also updated dcp to {asahi-,}dcp-september which has a few fixes

22:16 <jannau> removed t600x dcp dt changes since I didn't want deal with my t600x dt changes for linux 6.1. I suppose you will rebase eventually

22:20 <jannau> hmpf, display doesn't wakes up after weston disabled it. I've seen the same with xrandr --off

22:58 <alyssa> jannau: Oh my, did I troll you into becoming a graphics person? :-p

23:00 <alyssa> anyway, thinking out loud here...

23:00 <karolherbst> :D

23:00 <alyssa> My rough unprioritized AGX driver todo list

23:01 <alyssa> - Finish off the nir_opt_preamble code

23:01 <alyssa> That's probably the single most important shader compiler optimization for AGX

23:01 <alyssa> I wrote up an implementation but haven't tested it yet, that will wait until after XDC

23:02 <karolherbst> alyssa: please teach me master, how can I also learn the skill of trolling people into working on stuff (here: rust stuff in mesa)

23:02 <alyssa> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18813

23:02 <alyssa> - land random AGX MRs I have flying around

23:03 <alyssa> - build lina's kernel and replicate the gnome demo. Probably daily drive the AGX driver from here on.

23:03 <alyssa> - Upstream as much of lina's mesa fork as I can.

23:04 <alyssa> I don't think I can upstream any of the linux UAPI bits

23:04 <alyssa> but there's a pile of boring WSI patches that lina and I wrote, that can probably go in now

23:04 <alyssa> - plumb through compute shader support

23:05 <alyssa> I already reverse-engineered the most important bits for the mesa driver, just need to actually do a demo implementation

23:05 <alyssa> that will unblock lina working on the kernel side of this.

23:05 <karolherbst> ✨ compute shaders ✨

23:06 <alyssa> it will not be enough to get started with OpenCL, because we still need reverse-engineer atomics and image writes

23:06 <alyssa> (neither are hard, just more things to do)

23:06 <karolherbst> I have good news for you

23:06 <karolherbst> that's entirely optional

23:06 <karolherbst> both

23:06 <alyssa> uhhh.. okay

23:06 <karolherbst> 1.1 needs 32 bit atomics

23:06 <alyssa> I need them both for GLES3.1 anyway

23:06 <karolherbst> but 1.0? nope

23:07 <karolherbst> though I assume they are there, because all drivers exposing compute shaders also do atomics

23:07 <karolherbst> but in theory I could add code detecting it and just not advertize the atomic extensions and cap to CL 1.0

23:07 <alyssa> - fix bugs affecting glamor, firefox, chrome, supertuxkart, mpv, whatever I come across for day-to-day stuff

23:08 <alyssa> (those are all blockers to releasing)

23:09 <alyssa> - try to get rid of some "unknowns" in G13, ahead of forking off the XML for G14

23:09 <alyssa> I don't plan to do the initial g14 port myself but I do need to make sure the driver is in good shape for supporting multiple gens

23:10 <alyssa> (right now it's not)

23:10 <alyssa> might also be nice to do some recon on g15, though that's mostly blocked on getting my hands on an m3 mac I think

23:11 <alyssa> (I'm assuming that the newly announced A16 is using G15. But maybe that's also G14 and G15 won't happen until A17..)