00:04RSpliet: karolherbst: the hangs and random behaviours did occur more frequently since upgrading to a 4K monitor, but they'd occur even in the highest pstate on that card of mine
00:05RSpliet: There's some precursors as well. I've seen a weird line at the bottom of the cursor plane from time to time randomly appear.
00:07RSpliet: And, when playing (accelerated) video, sometimes the GPU decides that all white is now a colour with 2 out of 3 channels saturated (so either yellow, magenta or baby-blue...).
00:07RSpliet: Which only a reboot fixes, if it survives that long :-p
01:10karolherbst: heh, guess there is a real bug then
01:35karolherbst: RSpliet: I suspect you upgraded to mesa 22.3 or something already?
11:34ice9: does nouveau supports vulkan?
12:02karolherbst: ice9: "support" might be a strong word, but we are working on it and some are already able to play some simpler games
12:39ice9: karolherbst, I have hybrid laptop, how can I start a game from the command line with nvidia card and enable vulkan?
12:44karolherbst: at the moment you can't unless you build it yourself from our vulkan fork
12:45karolherbst: also, unless you also run some very very early WIP GSP patches on the kernel side, performance will also be terrible anyway
12:53fdobridge: <Esdras Tarsis> How can I use these GSP patches? Is it just compiling the skeggsb kernel fork?
13:13RSpliet: karolherbst: nah, I upgraded to AMD. Sorry, not sorry :')
13:14RSpliet: But yes, I'm on Fedora 37, and it ships with Mesa 22.3
14:17fdobridge: <gfxstrand> @marysaka Where are we at on the Fermi MME stuff? No pressure. I've just not seen the MR move and wanted to be sure you didn't just forget to push or something.
15:20fdobridge: <marysaka> Will try to do that tonight hopefully, but don't we need a rebase of nvk/main first?
15:37fdobridge: <gfxstrand> I can rebase if you'd like or we can merge MME and then I can rebase and drop the old version of the isaspec patch.
15:37fdobridge: <gfxstrand> I can rebase if you'd like or we can merge MME and then I can rebase and drop the old version of the isaspec patch.
15:43fdobridge: <marysaka> Would prefer a rebase of nvk/main tbh
15:46fdobridge: <gfxstrand> Ok. I'll do that once I finish making breakfast
15:59fdobridge: <gfxstrand> Done!
16:59fdobridge: <Esdras Tarsis> The Talos Principle on NVK
16:59fdobridge: <Esdras Tarsis> https://cdn.discordapp.com/attachments/1034184951790305330/1062415341726220399/10-01-23-135318-SCREEN.png
17:17fdobridge: <karolherbst🐧🦀> huh, that's with dxvk even?
17:17fdobridge: <karolherbst🐧🦀> how does the native vulkan version run
17:18HdkR: Isn't Talos Vulkan only?
17:19karolherbst: nope, it does have a GL renderer
17:19karolherbst: and my PCIe "reclocking" patches are able to even improve its perf on nouveau :D
17:19karolherbst: I think I got 25%+ on laptops and 5%+ on desktop GPUs
17:19fdobridge: <Esdras Tarsis> this is the native vulkan version, dxvk needs more extensions to work
17:19karolherbst: one of the few games where it mattered on desktpu GPUs
17:20fdobridge: <karolherbst🐧🦀> ahh.. I was confused about the "wine" stuff in there
17:20HdkR: Nice
17:20fdobridge: <karolherbst🐧🦀> the fps though 🥲
17:20fdobridge: <karolherbst🐧🦀> does the GL version run faster or slower?
17:21HdkR: PCIe BW is starting to matter more with these low-end APIs because people just map things directly and become super inefficient :P
17:21karolherbst: that was on GL
17:21HdkR: bwah? But Mangohud said Vulkan
17:22karolherbst: on laptops it matters, because of prime
17:22karolherbst: just talos triggered some odd code paths in our driver I guess
17:22karolherbst: the gl driver isn't very optimized and stuff
17:22HdkR: Oh yea, prime matter more definitely
17:22karolherbst: and it does a lot of stupid things
17:22karolherbst: pcie link speed increases generally gave me a 5-10% perf increase on laptops
17:22karolherbst: talos was just an exception
17:23karolherbst: but I think this is due to something dumb we do
17:23HdkR: Or they just abuse buffer_storage :D
17:23karolherbst: heh
17:23karolherbst: I'd rather bet on our driver being stupid
17:24HdkR: non-zero chance definitely
17:24fdobridge: <Esdras Tarsis> Does the game have a native Linux port? I'm using the Windows version because it's from Epic Games
17:25karolherbst: more like 250% chance, because digging into it, will bring up 1.5 other bugs :P
17:25HdkR: Yea, it has a native Linux build
17:25fdobridge: <Esdras Tarsis> this is the wine vulkan version, dxvk needs more extensions to work (edited)
17:25fdobridge: <Esdras Tarsis> this is the windows vulkan version, dxvk needs more extensions to work (edited)
17:25fdobridge: <karolherbst🐧🦀> ohh.. the windows version supports vulkan.. duh. Right. But yeah, there is a native one doing GL and Vk
17:27HdkR: The Linux version has some bugs around window resizing, so it's not perfect
17:27HdkR: Tends to crash under i3wm because of it
17:27karolherbst: it wouldn't be the first game where running the windows version through wine is actually better
17:28fdobridge: <Esdras Tarsis> I'm using linux drm-next with skeggsb 00.06 patches, btw
17:28fdobridge: <karolherbst🐧🦀> right.. though I don't think you really need that and upstream should be good enough. The GSP stuff will be more interesting once we have a version which mostly works
17:31fdobridge: <Esdras Tarsis> What is the status of reclocking via GSP at the moment?
17:32fdobridge: <karolherbst🐧🦀> there is code, and it somewhat runs on Ben's machine, but it has bugs and is incomplete. Like some of the displaying work isn't finished yet and other random bits
17:37fdobridge: <Esdras Tarsis> I see
17:37karolherbst: Ben has the branches, but I didn't try them myself, and you need to get a specific gsp.bin file from the correct driver and stuff yourself I think? Dunno. If you are willing to waste a few hours that might be something to check out as well :D
17:52fdobridge: <Esdras Tarsis> the current correct driver is 525.60.11 iirc
17:54fdobridge: <karolherbst🐧🦀> knock yourself out then 🙃
17:56fdobridge: <Esdras Tarsis> I can try, which path should I put the firmware on? And will reclocking work automatically?
18:40fdobridge: <marysaka> @gfxstrand I rebased just now, haven't tested it yet again but will do a bit later and mark as ready
18:41fdobridge: <marysaka> (also need to add a test for immediate load with the edge cases)
18:41fdobridge: <gfxstrand> Cool
21:38airlied: anyone any idea on where nvk is wrt zink?
22:04fdobridge: <marysaka> Should be ready now 👍
22:05airlied: oh nice
22:06fdobridge: <marysaka> airlied: oh hmm I was replying to gfxstrand on discord
22:06airlied: oh oops :-P
22:08fdobridge: <Esdras Tarsis> airlied: I think there is an issue about zink in the repository
22:09fdobridge: <Esdras Tarsis> https://gitlab.freedesktop.org/nouveau/mesa/-/issues/24
22:18fdobridge: <gfxstrand> Cool. Let me take a look.
22:24airlied: jekstrand: I've pushed a cleaned up rebase to https://gitlab.freedesktop.org/nouveau/mesa/-/merge_requests/150 for the new uapi support enabling all the sparse stuff
22:24airlied: we should probably work out how we want to start integrating the new uapi
22:25airlied: and how long a cross over period to sustain
22:25airlied: dakr: ^ just fyi, branch has been rebased/cleaned up
22:27jekstrand: airlied: Ypu
22:27jekstrand: *Yup
22:27jekstrand: airlied: I think things should already be fairly split out.
22:27jekstrand: airlied: As far as cross-over goes, not long IMO.
22:28jekstrand: We should start validating on the new API ASAP
22:28jekstrand: I think NVK is in good enough shape that I'll consider it a decent test.
22:30airlied: jekstrand: should I make a meson option? or are we happy with a #define for now?
22:31jekstrand: Yeah, I think so
22:31jekstrand: Keeps us from accidentally testing with the wrong API
22:38fdobridge: <gfxstrand> Merged! Thanks a million! The one change I made was to squash your RA fix into my patch to add the `mme_reg_alloc` struct.
22:38fdobridge: <gfxstrand> Now time to land Maxwell support. 😄
22:45fdobridge: <marysaka> :AkkoYay:
22:45airlied: okay the new uapi MR has a meson option now
22:49fdobridge: <gfxstrand> 😄
22:50fdobridge: <gfxstrand> Kicking off a full Maxwell run to get a baseline and then we're gonna merge it. 😄
22:51fdobridge: <gfxstrand> Still have a bug with clears. IDK what yet.
22:51fdobridge: <gfxstrand> And maybe something with indexed draws
22:55fdobridge: <gfxstrand> https://gitlab.freedesktop.org/nouveau/mesa/-/merge_requests/177
22:57fdobridge: <Esdras Tarsis> Why does this MR have a geometry shaders commit?
23:02fdobridge: <mhenning> oh, cool to see pre-turing stuff moving along! Is https://gitlab.freedesktop.org/nouveau/mesa/-/merge_requests/118 still unmerged? I haven't been paying super close attention, but I don't see it in nvk/main
23:06fdobridge: <gfxstrand> Good question...
23:41fdobridge: <karolherbst🐧🦀> uhh... heh
23:42fdobridge: <karolherbst🐧🦀> I think we wanted to check if it actually works 🙂
23:42fdobridge: <karolherbst🐧🦀> but I'm also wondering why so much things pass without it
23:42fdobridge: <karolherbst🐧🦀> did we merge something similar or something?
23:42fdobridge: <gfxstrand> Nope
23:42fdobridge: <gfxstrand> I think things are defaulting to zero and working by accident. 🙃
23:42fdobridge: <gfxstrand> I just rebased on it
23:43fdobridge: <karolherbst🐧🦀> yeah well...
23:43fdobridge: <karolherbst🐧🦀> kick of another CTS run with those and see how much it fixes?
23:44fdobridge: <gfxstrand> Yup
23:45fdobridge: <gfxstrand> Kickec
23:48fdobridge: <karolherbst🐧🦀> do you still have the turing card which doesn't work with nouveau?
23:53fdobridge: <gfxstrand> Yes
23:54fdobridge: <gfxstrand> It's my blob card ATM