17:28fdobridge: <gouz> draw-indirect instance now work!
17:29fdobridge: <gouz> draw-indirect instanced\ now work! (edited)
17:29fdobridge: <gouz> draw-indirect instanced now work! (edited)
17:37RSpliet: as good an idea as it may seem, this matrix bridge is at the same time starting to look like an attempt at embrace, extend, extinguish on IRC
17:37fdobridge: <jekstrand> Awesome!
17:38fdobridge: <gouz> 😋
17:38fdobridge: <gouz> now some more testing for draw indirects
17:49fdobridge: <jekstrand> Left you a comment
18:01fdobridge: <gouz> Thanks @jekstrand i will fix this
18:04fdobridge: <jekstrand> There's a couple other places where we do that. Maybe we should add an `nvk_root_descriptor_offset()` helper macro.
18:09karolherbst: RSpliet: see, already getting ignored :P
18:17RSpliet: karolherbst :-P Also not sure which bridge this is tbh. Comparing it with the matrix bridge in dri-devel, this looks different. Could be bridging something else instead... MS Teams? :>
18:18alyssa: jekstrand: See now I want MME Doom
18:18RSpliet: But it's ok. I don't have a lot of skin in the game, being on the sidelines of OSS driver dev. Just wanted to have said it once, I'd find it a shame if stuff slowly starts gravitating towards a non-open standard :-)
18:19karolherbst: RSpliet: if you'd know....
18:23alyssa: RSpliet: +1
18:23alyssa: (she says from both sides of the bridge)
18:23karolherbst: :D
18:25alyssa: It's pretty clear that IRC is fading, the question is what comes next... and yeah, I hope we have a FOSS answer for that.
18:26alyssa: There are answers -- Matrix, Rocket.chat, Mattermost -- but so far no singular good answer.
18:28i509VCB: A lot of irc servers aren't configured very well either. I have to use the matrix bridge right now to prevent the server from announcing my actual ip. And bouncers are just annoying.
18:30karolherbst: yeah.. all the FOSS alternatives I looked into it are just broken one way or another
18:30karolherbst: it's very sad
18:30karolherbst: either their permission system is utter trash
18:30karolherbst: they don't support gitlab SSO out of the box
18:30karolherbst: they UI are trash
18:30karolherbst: they don't allow what we need...
18:30karolherbst: it's frustrating
18:30karolherbst: there is a good solution covering almost everything and those are not FOSS
18:31i509VCB: The only solutions I see are the compromise principles or features
18:31alyssa: karolherbst: none of the nonfree ones do gitlab SSO out of the box either ...
18:31alyssa: so that's a moot point
18:31karolherbst: yeah..
18:32RSpliet: I don't have a particularly strong opinion. In general by the way
18:32karolherbst: that's like the only drawback and painful
18:32alyssa: RSpliet: I do :(
18:32alyssa: (she says from both sides of the bridge)
18:32RSpliet: I can see IRCs shortcomings, but I also don't care about them. It gets the messages across
18:32karolherbst: yeah.. it needs to be foss
18:32karolherbst: but atm the strongest candidate would be zulip
18:32alyssa: is that the one with the weird threading?
18:32karolherbst: yeah
18:32i509VCB: Foss is a hard requirement
18:32RSpliet: Bouncers annoying, meh, got a home server. Pretty sure one could be set up on FDO infra if need be
18:32karolherbst: it actually has a useful permission system
18:32i509VCB: We have to compromise the UI/UX somewhere
18:33karolherbst: and they host FOSS projects for free
18:33karolherbst: with all the features
18:33karolherbst: and unlimited
18:33karolherbst: just have to ask about custom gitlab SSO
18:34i509VCB: With zulip nouveau/asahi/etc can have their own streams
18:34i509VCB: A monolithic fdo zulip instance is possible
18:34karolherbst: yeah
18:34karolherbst: that's the strongest point for zulip
18:34karolherbst: and you can have different permissions per stream/channel/whatever
18:34karolherbst: so people can organize their own sub project
18:36alyssa: which matches our current IRC model nicely
18:36alyssa: although honestly if we were doing IRC for fd.o today I'd almost wish we had a dedicated irc.freedesktop.org server
18:38karolherbst: yep
18:38karolherbst: but hosting...
18:38karolherbst: apparently zulip is also more FOSS than the others, so we won't have that kind of discussion either
18:39karolherbst: I'll probably contact their sales to upgrade the test freedesktop server I've setup or something... maybe I need an official email for that (lol)
18:39karolherbst: I have IRC enough to actually spend time on this
18:39karolherbst: *hate
18:39i509VCB: https://zulip.readthedocs.io/en/stable/production/authentication-methods.html
18:40i509VCB: I'm not sure if the gitlab sso there is for gitlab.com or will work on any self hosted gitlab
18:44karolherbst: that's for gitlab.com
18:44karolherbst: but it also has SAML and OpenID connect
19:23mntirc: i'm the designer of the open hardware mnt reform laptop and looking for some advice; i'm thinking to make a version based around nvidia xavier nx (or later orin) and mainline linux, mesa. as nouveau devs would you say this is a good idea or nah? i'm also considering rk3588(s) which is (somewhat?) supported by panfrost if i'm not mistaken
19:24mntirc: currently the soc is imx8mq which has a vivante gpu (gc7000). it's a pretty stable platform but limited by the a53 and gles 2.
19:24karolherbst: mntirc: nvidia stopped shipping firmware for tegra devices, and xavier is not supported
19:24mntirc: karolherbst: that sounds like it's a dead end?
19:25karolherbst: yeah, pretty much. Unless you can convince Nvidia to care (I tried in my position as a Red Hat developer, and that didn't work)
19:25mntirc: ugh, that's disappointing
19:25mntirc: did they ship the fw before as a separate download?
19:25karolherbst: if you want to use panfrost, _maybe_ talking to daniels on #panfrost is a good plan forward
19:25karolherbst: no, it's all in linux-firmware
19:26mntirc: i wonder why they put in the effort to support the platform on mainline linux at all then
19:26karolherbst: but "gp10b" is the last they shipped for tegras
19:26karolherbst: which is the generation before xavier afaik
19:26mntirc: yeah it should be gv11b
19:26karolherbst: mntirc: there was this deal with google and the pixel.... something
19:27karolherbst: they tried using nouveau, nvidia didn't care enough to make that a success and...
19:28HdkR: You /really/ wouldn't want Xavier anyway :P
19:28karolherbst: orin only supports one display :P
19:28mntirc: HdkR: why is that?
19:28HdkR: Closed source firmware blobs for the CPU
19:28HdkR: Wacky CPU is wacky
19:29HdkR: Also both Xavier and Orin are like 50-60w TDP parts, they're HUUge
19:29mntirc: fwiw i have a NX sample and i was surprised about the amount of chaos around booting it
19:29karolherbst: mntirc: there is UEFI firmware btw
19:29mntirc: ah, i have the jetson xavier nx, it can run on 10w
19:29mntirc: yes, i installed the UEFI one yesterday
19:29karolherbst: ahh, cool
19:29mntirc: it was able to boot a debian installer iso
19:29karolherbst: I have it on my jetson nano :3
19:29HdkR: Yea, it scales down a bit at least
19:30mntirc: so, there's a framebuffer and all... and i thought the next step could be to convince nouveau to work with the gpu
19:30karolherbst: yeah.. no dice
19:30HdkR: * Sad Trombone Music *
19:30mntirc: i found some irc logs from hier from tagr who seemed to have made some initial work
19:30mntirc: s/hier/here
19:30karolherbst: yeah, so in theory it should all work
19:30karolherbst: just missing the firmware
19:31karolherbst: I asked nvidia, and they said they don't care
19:31karolherbst: soo....
19:31mntirc: ok, and i couldn't extract that from the binary blob driver stuff?
19:31mntirc: (i trust your judgement, just curious about the technical details)
19:32karolherbst: but I wasn't much caring either, just said if it's not much work for them to release that firmware they should figure out if they could, but sadly apparently it's a lot of work, so without anybody actually wanting that it was kind of in this limbo of not caring and not wanting to put in the effort
19:33mntirc: aha, so it's more like a legal hurdle, because extracting it would be some copyright problem, or... is it just that noone was eager enough to go down that path?
19:33karolherbst: though I'd probably would have supported it, but nouveau doesn't have enough developers to actually support it well anyway :'(
19:33karolherbst: mhh.. soo the firmware we got from nvidia is different than what they use
19:33mntirc: ah?
19:33karolherbst: so even if there wouldn't be any legal issues in terms of distribution, we'd have to figure out how to use that firmware
19:34karolherbst: and it's this signed firmware mess with some super secure bootloader
19:34mntirc: so the fw that might be in a blob somewhere is not like older tegra fw?
19:34karolherbst: and if it fails to boot the only thing we know is: it didn't boot
19:34karolherbst: yeah.. it's quite close to their discrete GPU model
19:34karolherbst: but it's a pita to rely on that
19:34karolherbst: and we don't have the people to make it run well in the short term
19:35HdkR: mntirc: I'm curious how you feel about Snapdragon as a SoC choice :)
19:35mntirc: HdkR: bad because afaik there is no documentation for those
19:35karolherbst: anyway.. you probably want to talk to daniels if you have specific SoCs in mind
19:36karolherbst: daniels will be able to help you.. maybe.. depending on the hardware
19:36mntirc: karolherbst: ok, thanks a ton for your perspective
19:36mntirc: karolherbst: i kinda know daniels, so that sounds goot
19:36mntirc: good
19:36karolherbst: cool
19:36mntirc: (what's with the typos today)
19:37HdkR: ah
19:37mntirc: HdkR: i was surprised one can just download the reference manual for xavier
19:37mntirc: 8000+ pages
19:38mntirc: so, that was a good thing at least. similar to the nxp imx family we use
19:38HdkR: Whoa, didn't know that was available
19:38fdobridge: <🌺 ¿butterflies? 🌸> For Xavier the firmware is shipped in a separate download
19:39fdobridge: <🌺 ¿butterflies? 🌸> But is available
19:39karolherbst: ahh cool
19:39fdobridge: <🌺 ¿butterflies? 🌸> The NVIDIA L4T kernel also has zero binary only modules
19:39fdobridge: <🌺 ¿butterflies? 🌸> (nvgpu being GPLv2)
19:39karolherbst: maybe one could make it work with that
19:39karolherbst: but I don't have the hardware anyway
19:40karolherbst: and I have way too many things to work on already :'(
19:41mntirc: karolherbst: i can send you the hardware if you want to play with it
19:41fdobridge: <🌺 ¿butterflies? 🌸> Oh yeah… I wonder for me if the right route is making nouveau be able to use the nvgpu KMD
19:41mntirc: it's no use for me without nouveau anyway
19:41karolherbst: I can also buy it, it's not the problem. The proble is time and random other things :(
19:42mntirc: ok i see ^^
19:44fdobridge: <🌺 ¿butterflies? 🌸> I have Xavier around but
19:44fdobridge: <🌺 ¿butterflies? 🌸> Not remotely enough time
19:45fdobridge: <🌺 ¿butterflies? 🌸> Also remember one thing: Jetson Nano can be booted around entirely blob less. This doesn’t apply to Xavier
19:45mntirc: oh, i didn't know that
19:45fdobridge: <🌺 ¿butterflies? 🌸> In fact Xavier has a mandatory multimeg CPU microcode to load 😛
19:45karolherbst: uhhhhhh
19:45karolherbst: pain
19:45mntirc: the nano gpu is also not supported by nouveau, right?
19:45karolherbst: yeah I don't have the time to fight this annoying security thing
19:46karolherbst: it actually is
19:46mntirc: it is?
19:46karolherbst: yeah
19:46karolherbst: it's just slow and weak hardware
19:46fdobridge: <🌺 ¿butterflies? 🌸> It’s a Transmeta-style machine with a translator from arm to native ISA
19:46mntirc: ah yeah
19:46karolherbst: and quite dated
19:46fdobridge: <🌺 ¿butterflies? 🌸> So you need to bundle and have that JIT around
19:46mntirc: do you get GL 4.x on it?
19:46mntirc: or vulkan?
19:46fdobridge: <🌺 ¿butterflies? 🌸> Or you won’t exactly get very far
19:46karolherbst: atm just GL
19:46fdobridge: <🌺 ¿butterflies? 🌸> GL4.5 works
19:46karolherbst: but vulkan is in the making
19:47karolherbst: the GL driver isn't great though
19:47mntirc: fdobridge: seriously @ transmeta? lol
19:47fdobridge: <🌺 ¿butterflies? 🌸> Note that it would be by far the strongest GPU of any mnt reform
19:47mntirc: but the orin (?) will have cortex-a78 again
19:47karolherbst: and that clocking mess is a mess
19:47fdobridge: <🌺 ¿butterflies? 🌸> Yes. Xavier is the swan song of Transmeta-style designs
19:48fdobridge: <🌺 ¿butterflies? 🌸> The Estes core that will succeed Carmel is no longer using a dynamic code optimisation approach
19:48mntirc: i see
19:49fdobridge: <🌺 ¿butterflies? 🌸> Orin also needs blobs to boot, MB1, the first stage loaded from SPI is signed and encrypted
19:50fdobridge: <🌺 ¿butterflies? 🌸> But for Xavier onwards you have something cool: UEFI out of the box with edk2
19:51mntirc: yep, i was surprised that it works
19:51mntirc: and the edk2 fork is open source also
19:51fdobridge: <🌺 ¿butterflies? 🌸> The last BSP release for Jetson Nano came with U-Boot with the UEFI compat module on
19:52fdobridge: <🌺 ¿butterflies? 🌸> So that you are supposed to be able to use regular Fedora install media now without tricks
19:52mntirc: i was able to boot stock debian and fedora install isos
19:52mntirc: so mainline kernel def. works
19:53fdobridge: <🌺 ¿butterflies? 🌸> For Xavier and Orin ACPI is a work in progress too
19:53fdobridge: <🌺 ¿butterflies? 🌸> You won’t be stuck with FDT in the long run
19:53mntirc: well, it's all no use if there are no gpu drivers to go with mainline
19:54fdobridge: <🌺 ¿butterflies? 🌸> The plan so far is to split the nvgpu kernel module and make it available as DKMS
19:54fdobridge: <🌺 ¿butterflies? 🌸> As far as I understood
19:58fdobridge: <🌺 ¿butterflies? 🌸> If you want something with relatively complete mainline support - Jetson Nano/TX1 is where you might want to stop
19:59mhenning: hmm. I wonder if it's possible to use the out of tree l4t driver with something like https://gitlab.freedesktop.org/nouveau/mesa/-/merge_requests/65
19:59fdobridge: <🌺 ¿butterflies? 🌸> Tegra X1 is one of the few platforms with OSS DRAM training code available (was reverse engineeeed)
19:59fdobridge: <🌺 ¿butterflies? 🌸> Yes. The L4T driver set is fully OSS. It’d just take quite a bit of work…
20:00mntirc: butterflies: oh, hadn't heard that! so the tx1 is supported by nouveau as well?
20:01fdobridge: <🌺 ¿butterflies? 🌸> Yes TX1 is supported by nouveau. Such a config did ship in production on the Pixel C
20:03mntirc: i see. that is quite interesting, thanks! i also didn't realize it was a 8 core, 2 ghz cpu
20:03fdobridge: <🌺 ¿butterflies? 🌸> Quad
20:03fdobridge: <🌺 ¿butterflies? 🌸> Forget the A53 core complex on the Jetson Nano
20:03fdobridge: <🌺 ¿butterflies? 🌸> A53 core complex never shipped, it’s there on silicon tho
20:03mntirc: ok?
20:04fdobridge: <🌺 ¿butterflies? 🌸> Basically, the 4 little cores are always off.
20:04fdobridge: <🌺 ¿butterflies? 🌸> Under any os
20:04mntirc: weird
20:04fdobridge: <🌺 ¿butterflies? 🌸> Rumour mill says that it’s because of a cache coherency erratum
20:05fdobridge: <🌺 ¿butterflies? 🌸> Nothing prevents manually enabling them tho - but NVIDIA never did it for a reason
20:09fdobridge: <🌺 ¿butterflies? 🌸> Jetson Nano has the Switch SoC with half the GPU fused off - but running at significantly higher clocks
20:10fdobridge: <🌺 ¿butterflies? 🌸> It also has a relatively good memory subsystem - faster than RPi4 in MT
20:10fdobridge: <🌺 ¿butterflies? 🌸> (On the cpu side, GPU is not even a contest)
20:13mntirc: we often have problems on the cpu side because of javascript heavy web apps... those don't work well on "little" cores unfortunately. and on the GL side, it would be great to be able to run blender and some more modern games via box64. so that's attractive about tegra.
20:15fdobridge: <🌺 ¿butterflies? 🌸> If you are happy with blobs you might want to go Orin Nano
20:15fdobridge: <🌺 ¿butterflies? 🌸> But know that’s certainly not a blob less system 😛
20:15mntirc: i'm ok with boot blobs but not with blob drivers running on top of the OS
20:16fdobridge: <🌺 ¿butterflies? 🌸> Yeah for all Tegras no binary Linux kernel modules around
20:16fdobridge: <🌺 ¿butterflies? 🌸> So you don’t have that concern at least
20:16mntirc: ok, so this i probably didn't get. the l4t GPU driver for nx is _not_ closed source?
20:16mntirc: xavier nx i mean
20:17fdobridge: <🌺 ¿butterflies? 🌸> Kernel mode part is OSS. User mode is the good old proprietary nvidia driver stack
20:17fdobridge: <🌺 ¿butterflies? 🌸> But that can be potentially replaced with Mesa
20:17mntirc: ah, so that is also quite unacceptable :D
20:17mntirc: yeah, but it sounds like that's not gonna happen
20:17fdobridge: <🌺 ¿butterflies? 🌸> I have a Xavier around so maybe I should take a shot…
20:19mntirc: well, that would be amazing.
20:19fdobridge: <🌺 ¿butterflies? 🌸> (I’m a CUDA user so I rely on the nvidia KMD anyway… but indeed getting Mesa to work with that KMD would be nice)
20:20mntirc: i have the xavier nx so i can test things, and i can also contribute code, but only in 2.5 weeks after a short vacation
20:21fdobridge: <🌺 ¿butterflies? 🌸> Hah - meanwhile currently a bit (too) busy since I’m moving to Berlin
20:21mntirc: oh, i am based in berlin
20:21fdobridge: <🌺 ¿butterflies? 🌸> nice. finding good housing there is a bit of a mess tho
20:22mntirc: it is very hard currently yes :/
20:22mntirc: did you find something?
20:22fdobridge: <🌺 ¿butterflies? 🌸> Still stuck. A lot of the pricing is nuts
20:22mntirc: yes, i live here since 2004 but i had to move ~2 years ago and now pay 1050 eur for like 37sqm
20:23mntirc: anyway, i guess this is going kind of off topic for the channel... if you would like to stay in touch and discuss this and other stuff, my email is lukas@mntre.com
20:25fdobridge: <🌺 ¿butterflies? 🌸> true… so far did you just use an unmodified Jetson BSP on your boards?
20:25mntirc: yes, it came with an old one with cboot (?) and i reflashed it to the latest one that turned it into UEFI
20:26fdobridge: <🌺 ¿butterflies? 🌸> CBoot is NVIDIA’s fork of LK
20:26fdobridge: <🌺 ¿butterflies? 🌸> comes from the Android world
20:26fdobridge: <🌺 ¿butterflies? 🌸> Jetson got transitioned to UEFI in 2022. Will make a lot of things nicer in the future, including stable bindings for nouveau
20:27mntirc: ok, so i will not abandon the project yet but put it in research mode for a while longer
20:44karolherbst: yeah.. uefi is great
20:46hell__: great pain in the rear
20:47fdobridge: <🌺 ¿butterflies? 🌸> UEFI is so much nicer for the ecosystem than the alternatives - allows to share install media between distros
20:47karolherbst: no, it's literally great, stop arguing about it
20:47karolherbst: yeah
20:47karolherbst: UEFI is super cool
20:47karolherbst: I have my ventoy USB driver with tons of random isos and I can just boot and install stuff anywhere
20:49hell__: ← coreboot dev
20:49karolherbst: as long as coreboot implements UEFI all is well
20:49hell__: you can use edk2 as payload, then you can boot things using UEFI
20:50karolherbst: finally sanity
20:50hell__:had to dig into edk2 code today to figure out why some feature for $CUSTOMER was not working
20:51karolherbst: I also kind of like being able to just dig into ACPI firmware and figure out what OEMs are terrible at
20:51fdobridge: <🌺 ¿butterflies? 🌸> I don’t care if you use EDK2 or u-boot UEFI or whatever as long as what is provided to the OS (bootloader) is UEFI
20:51karolherbst: same
23:54mhenning: mntirc: I'd be willing to poke at orin or xavier a bit if you want to give me ssh access
23:56mntirc: mhenning: cool, i can set it up on monday. should i reach you here?
23:57mhenning: mntirc: You can reach me here if I'm on the channel, or email at mhenning at darkrefraction dot com if I'm not
23:58mntirc: great, will do! thanks!
23:58mhenning: sure. no promises though :)
23:59mntirc: sure!