02:04mhenning[d]: I'd be curious if anyone can measure a perf difference with https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33105
02:16gfxstrand[d]: I really need to review and land that one. We should probably at least use it on Turing+
02:19gfxstrand[d]: It might help with pipelining
02:20mhenning[d]: Yeah, that was my hope, but I haven't measured a perf difference so far (although I've only tried like two apps so far, so who knows)
02:22gfxstrand[d]: Around the Turing timeframe, Nvidia was really pushing the "multi-draw everything!" line and I suspect that might have been part of it.
02:22gfxstrand[d]: But IDK how much to expect it to help. Might need microbenchmaking to see it.
09:26marysaka[d]: It would be nice to also integrate the new TURING_A+ variant using `SET_DRAW_CONTROL_A`/`DRAW_VERTEX_ARRAY_BEGIN_END_A` (had some MR around <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26288>)
13:11phomes_[d]: mhenning[d]: I can test with various games tonight. I just want to wrap up the thing with copy-prop of fabs. I think I have a fix
16:13phomes_[d]: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33107
16:15karolherbst[d]: mhhhhh
16:15karolherbst[d]: nak takes the abs restriction into account, right?
16:16karolherbst[d]: ohh wait.. was that `IADD3` only...
16:17karolherbst[d]: yeah.. don't mind me
16:19tiredchiku[d]: somehow the holidays sapped my motivation entirely and I'm struggling to get it back e-e
16:20tiredchiku[d]: but I did find a great resource on how to interface with nvidia's kernel modules
16:21tiredchiku[d]: https://github.com/elFarto/nvidia-vaapi-driver/blob/master/src/direct/nv-driver.c
16:21tiredchiku[d]: I'll get to it.. one of these days..
16:48avhe[d]: tiredchiku[d]: I have some PoC code working up to job submission (to the dma engine but it should be generic). I'm not super happy with its current state and there's no good sync primitive. I tried to use the NV01_EVENT_OS_EVENT stuff but couldn't get it working yet
16:48avhe[d]: https://gist.github.com/averne/54d46a1df1960e9961528b427629e82d
16:48avhe[d]: I'm planning to get back to it at some point in the coming weeks, I'm on something else atm. In particular I'd like to remove the UVM stuff since it adds a lot of code and notthatclippy[d] said it wasn't strictly required
16:49avhe[d]: Anyway I think it can help
16:49tiredchiku[d]: I'll take a look, thanks!
16:54avhe[d]: Also check out tinygrad
16:58tiredchiku[d]: will do
17:05asdqueerfromeu[d]: avhe[d]: Hopefully it works well on NVIDIA GPUs (considering it has prioritized AMD ones with that MES thing)
17:07rathernicebag: I feel bad for not sharing the method to the public, but i can not do this anymore. It's gonna turn too much upside down. But i say it has nothing to do with discrete fourier transform nor fast versions of this. It's simplest possible hack. I believe the ones who can appreciate others favours or can think their own, are safe to others too. And those people who did wank tunes to the world,
17:07rathernicebag: are none of this, they never worked, never showed behaviour nor can actually at all think. And this is very likely that they get murdered in the war. I have everything related to heaven , but it is going to make matters only worse if shared. So i maintain them in secure environment instead like i founded a company and am doing the right thing as it seems i am able to think faster and
17:07rathernicebag: better, i will also maintain the complex parts my own.
17:08rathernicebag: complex to you, i find it very simple.
17:10orowith2os[d]: why is it always Discord channel #Nouveau
17:11ermine1716[d]: usually it's Discord channel #dri-devel on irc
17:12kayliemoony[d]: huh.jpg
17:31rathernicebag: yeah dears, i am sure it's gonna escalate into full scale world wide conflict, and instead it has to be managed in a different way, as said, we need to coordinate this tech to be accessed by controlled parties. I hope you'd forgive me, it's all for better.
17:35rathernicebag: i asked from the biggest headed guys if this can be good idea to give such and such features to the public, and everyone said no dude, it's not, it would end up as catastrophe.
17:47alex____: So I’m here because Kepler still has horrible bugs and OpenGL issues under modern Linux
17:49alex____: For example Minecraft won’t even load the launcher using the Nouveau driver. The NVIDIA 390 driver doesn’t support the 6.x.x kernel series. These bugs have been present for a long time and were never fixed.
17:50alex____: These cards should be much better off now that it’s been 13 freaking years since they were released.
17:52alex____: Downgrading to Ubuntu 20.04.5 LTS just to use a proprietary driver is terrible.
17:58orowith2os[d]: alex____: I doubt anybody's working on legacy NVIDIA cards nowadays, considering Turing and newer are much easier to support
17:59orowith2os[d]: I know NVK has support for running on Kepler, but it's behind a broken flag
17:59alex____: orowith2os[d]: Turing and newer don’t support macOS
18:00orowith2os[d]: where does maOS come into play here?
18:00orowith2os[d]: *macOS
18:01asdqueerfromeu[d]: orowith2os[d]: I assume dualbooting
18:02orowith2os[d]: ah
18:02orowith2os[d]: probably on a desktop-y system
18:02orowith2os[d]: makes sense
18:04alex____: I like to dual boot my systems as far as running certain games and clearly macOS runs Minecraft terribly on said card… I’m running 1.21.4 just fine on 20.04 LTS with the 390 driver. This iMac was upgraded to a K1100M from one of these AMD cards that are prone to failure.
18:05alex____: The AMD HD 6770m to be precise
18:06alex____: orowith2os[d]: exactly 😊
18:06alex____: iMac 12,2
18:07orowith2os[d]: I would like to say just buy a new AMD card and use that, if I can trust Google here, macOS supports recent RDNA cards
18:08alex____: Problem is flashing said cards and getting into these systems is a PITA
18:08alex____: Sourcing MXM-A cards are a real PITA too
18:09alex____: This is a 27 inch
18:11orowith2os[d]: ouchie
18:11orowith2os[d]: unless you're willing to work with *and* find people who want to work on nouveau for pre-Turing cards, you might be out of luck
18:20alex____: Yeah the mid2011 iMac requires complete disassembly to get at the GPU
18:22alex____: I’m at least thankful that the system won’t fail on me now. Those AMD cards they came with were hot garbage reliability wise.
18:33alex____: 1:11 PM <orowith2os[d]> unless you're willing to work with *and* find people who want to work on nouveau for pre-Turing cards, you might be out of luck
18:37ermine1716[d]: I've just gave my kepler away
18:39orowith2os[d]: I think I have a Fermi somewhere
18:58karolherbst[d]: of all the gens nouveau supports, fermi is the most cursed one
19:40alex____: Fermi I’ve never dealt with… I wonder what the 7300GT’s gen is? That’s the card that originally came in my 1,1 Mac Pro
19:41alex____: I have a GTX 950 in said Mac Pro now
19:41karolherbst[d]: 7300 is curie, I think the first generation with shaders
19:42alex____: Naturally I would run Linux on my 1,1 if I could see anything lol PC Cards in a Mac never output anything on here until the OS kicks in
19:44alex____: OpenCore would help if it were feasible for EFI32 Macs and an EnableGOP solution for that
20:09awilfox[d]: orowith2os[d]: I'm working exclusively on older cards and I was told that these contributions would be welcome. Has something changed in the last months?
20:09orowith2os[d]: nope
20:09orowith2os[d]: I'm not saying contributions aren't welcome, but that they're not the primary focus for nouveau as of late
20:10orowith2os[d]: you can do your thing, I'm sure people out there appreciate your work ^^
20:11orowith2os[d]: er, nouveau *as a whole*
20:41mhenning[d]: awilfox[d]: Yep, contributions are welcome for older cards!
20:42mhenning[d]: We still maintain the old gl driver enough that we try to avoid regressions, but there are known bugs in it and relatively little activity in that codebase
20:56gfxstrand[d]: awilfox[d]: Contributions are definitely welcome. Maxwell is pretty much done for NVK. There's a few issues and the kernel isn't stable enough for full CTS runs so we haven't enabled it for users yet.
20:56gfxstrand[d]: Kepler should work on the API side but it needs a compiler.
21:05awilfox[d]: I just hope that Vulkan doesn't become the next "minimum bar" for starting a basic Linux desktop. I have hundreds of users on cards older than Kepler and LXQt and XFCE etc all work perfectly fine and are even performant. Honestly, I sometimes wish fbdev was still a thing, because for many people there's no need to use GL or anything, but drawing a picture on the screen is crucial…
21:10ermine1716[d]: well, accel is not the only fbdev issue
21:12awilfox[d]: Oh, fbdev is horrible, I agree, that's why I said 'sometimes' lol
21:47orowith2os[d]: awilfox[d]: that would depend on the software you want to use
21:48orowith2os[d]: most likely
21:48orowith2os[d]: I know GNOME is going to, for example, just use OpenGL in Mutter (no point in using Vulkan in their eyes), and GTK still has its ngl backend
21:49orowith2os[d]: there's an org I'm working with that's going to, most likely, use Vulkan/Metal (and WGPU on web platforms) for their UI toolkit because OpenGL is just hellish on mobile platforms for them
22:10ermine1716[d]: you're not going to have vulkan backend? Kwin seems to want one
22:14orowith2os[d]: ermine1716[d]: I filed an issue in gnome/mutter and they didn't want it, but also weren't against it
22:15orowith2os[d]: clarifying: I'm not an official GNOME dev, I'm just a third party that really likes it and tries to help out if I can
22:15orowith2os[d]: https://gitlab.gnome.org/GNOME/mutter/-/issues/2606
22:15ermine1716[d]: Oh
22:16orowith2os[d]: basically, the biggest bottleneck in mutter are CPU-bound (allocs in clutter), and Vulkan wouldn't provide any benefit with those in place
22:16orowith2os[d]: and direct scanout means you aren't compositing while gaming or watching a video anyways
22:19ermine1716[d]: In Kwin: https://invent.kde.org/plasma/kwin/-/issues/169
22:20orowith2os[d]: interesting
22:20orowith2os[d]: is kwin really that limited with opengl?
22:23ermine1716[d]: Tbh i haven't observed any opengl-related limitations in kwin, so it's better to ask involved people
22:23ermine1716[d]: I'm just excited with Vulkan 😄
22:23orowith2os[d]: oh, for sure
22:24orowith2os[d]: I referenced the kwin issue on the mutter issue, we'll see if anybody changes their mind on it
22:30orowith2os[d]: I do wonder if relying on zink would resolve some of those opengl issues though?
22:30orowith2os[d]: without needing to rewrite everything to use vulkan, just use some gl<->vk interop
22:31ermine1716[d]: idk, last time i've tried to launch plasma with zink it failed to start
22:31orowith2os[d]: it did on my 6th gen intel laptop
22:31orowith2os[d]: worked on my new amd one though
22:32orowith2os[d]: it did -> it didn't start
22:32orowith2os[d]: Firefox and Cinny gave me shit though
22:33ermine1716[d]: btw, maybe we should move to another channel, since it's not nvidia-related strictly speaking
22:33orowith2os[d]: true
22:33orowith2os[d]: I'll be in Discord channel #tech-talk
22:34ermine1716[d]: ok