01:29rinlovesyou[d]: Don't wanna be a bother with this because it's honestly a minor thing but how is dp audio looking?
01:29rinlovesyou[d]: Would be nice to get that working
04:36redsheep[d]: rinlovesyou[d]: I never did get the old 535 version of openrm working to confirm, but I feel like this is most likely something where the gsp from that old version has an issue
04:38redsheep[d]: If that hunch is right then the issue will be fixed by the same work that would be needed to support blackwell
08:33valentineburley[d]: esdrastarsis[d]: FYI, you don't need to rebase unless there are conflicts, Marge will do that automatically
08:34valentineburley[d]: And you can do `git commit --amend` if you don't want to do that fixup autosquash dance if it's just one commit 😄
08:37valentineburley[d]: And https://gitlab.freedesktop.org/mesa/mesa/-/issues/10981 is still open if anyone's looking for a quickie
08:37valentineburley[d]: There's a helper in the common runtime you can use and after that it's just a matter of going though NVK and plugging in a couple of `vk_object_base`s
10:39esdrastarsis[d]: valentineburley[d]: I thought about doing this, but I followed the documentation to learn this method with rebase, I had never done it before 🙂
13:00NyB: hello! I have an old Toshiba laptop with a gt525m GPU, and "recently" it has developed an issue where it will completely freeze, unless I blacklist nouveau or use nomodeset on boot
13:01NyB: does this ring any bells for anyone?
13:02NyB: it had been in a drawer for a few months, I turned it on, upgraded the OS (kernel went from 6.4->6.6 LTS) and now this started happening
13:04NyB: complete freeze within 10m from boot, CapsLock/SAK not responsive, nothing in the logs, nothing in netconsole. It just... stops. And that's regardless of X being loaded.
14:46juri_: as is usual with old laptops, i'd suspect thermals. take it apart, clean it, get it new thermal paste, etc.
14:49tiredchiku[d]: NyB does downgrading to 6.4 LTS make it work fine again?
15:29kuter7639[d]: Anyone working on improving instruction delays (to avoid WAR RAW hazards)? I am super nerd sniped by this problem.
15:30kuter7639[d]: The current implementation doesn't seem to take into account things like register bank conflicts
15:30kuter7639[d]: (and is overly pessimistic right?)
15:40NyB: tiredchiku[d]: no, both kernels have the same issue - it was the first thing I tried. Downgrading the firmware package made no difference either (does nouveau even need a firmware for this GPU?)
15:40NyB: I suppose it _could_ be some sort of hardware failure that is only triggered once a proper GPU driver is loaded
15:40tiredchiku[d]: or could be something in the userspace drivers
15:41NyB: are there any userspace drivers loaded if X is not active?
15:41tiredchiku[d]: that would depend. I imagine mesa gets loaded if you use a graphical display manager
15:42tiredchiku[d]: but it shouldn't be loaded if you run `startx` from the tty
15:42NyB: this failure happens even with just a VT, no DM loaded
15:42tiredchiku[d]: ..that's very strange
15:42NyB: all I have to do is wait up to 10m
15:42NyB: stress-testing the CPU showed no issues
15:43NyB: I'm debating with myself whether I should bite the bullet and open the thing up to clean it... it's be practically a full disassembly :/
15:44NyB: the weird thing is, there are no logs. Not in the framebuffer, not in the filesystem, not in netconsole. I haven't seen this kind of hard lock in ages.
15:44tiredchiku[d]: I'm stuck wondering why it doesn't freeze with nomodeset
15:45NyB: tiredchiku[d]: same here :)
15:45NyB: nomodeset _still_ loads nouveau, the module
15:45tiredchiku[d]: well, yeah
15:45NyB: so I'm not sure what the difference is
15:46tiredchiku[d]: I'm also stuck wondering what could be affecting it if it's not the kernel, because afaik there's not much userspace components involved in kernel modesetting
15:46NyB: I suppose I should try nomodeset once again, to make sure it really doesn't freeze
15:46tiredchiku[d]: though I could be wrong
15:46NyB: just in case I hit a fluke or two
15:46tiredchiku[d]: that would be worth checking, yes
15:47NyB: hmm... does X work with nomodeset?
15:47NyB: I had assumed no, but did not try
15:47tiredchiku[d]: it does
15:48tiredchiku[d]: X11 pre-dates kernel modesetting by a good few decades 😅
15:48NyB: yes, I remember :p
15:53NyB: nope, X does not work with nomodeset: (EE) [drm] Failed to open DRM device for pci:0000:01:00.0: -19
15:54NyB: OK, it's on, with a trusty "watch" command err... watching its uptime
15:55NyB: one thing I noticed is that it seems to run cooler with nomodeset
15:55tiredchiku[d]: do you have an x config?
15:55tiredchiku[d]: I'm pretty sure x11 works with nomodeset too
15:55NyB: good point, I do. One that wants nouveau.
15:57tiredchiku[d]: what happens if there's _no_ x config
15:57tiredchiku[d]: back up your current one, and then try running it again
15:57tiredchiku[d]: afaik x11 can do auto-config at runtime in most scenarios
15:58NyB: just tried it, it starts. Resolution seems weird.
15:59Sid127: (I'm tiredchiku) the resolution being weird should be trivial to fix (xrandr)
16:00NyB: the other tiredchiku[d] got too tired? :p
16:00Sid127: correct :D (and I had to fire up my IRC client for another channel too)
16:01NyB: hmff...1024x768
16:01NyB: _probably_ the framebuffer resolution
16:02NyB: yup, vga=791 is 1024x768
16:03Sid127: I still can't think of what it'd be that's causing the system to lock up like that
16:03Sid127: maybe karolherbst has better ideas
16:05NyB: let's see if it will freeze again
16:06Sid127: godspeed
16:06NyB: what does modeset do anyway? I was never clear on that :)
16:07NyB: is it about not relying on the VBIOS for resolution changes, or is it deeper than that?
16:08NyB: (is it even that? My knowledge of the GPU susbsystem is... not there :) )
16:08Sid127: yeah, the kernel sets the display mode
16:08Sid127: kernel modesetting :P
16:09NyB: so... it should only be triggered on mode changes
16:09NyB: not randomly 10m later
16:10Sid127: or at boot time, yeah
16:10NyB: maybe I should see if this thing has an IOMMU and see about enabling it
16:10NyB: if the GPU goes crazy and starts scribbling in main memory, maybe that would catch it and tell me
16:11NyB: nope, no VT-d on this thing :/
16:13redsheep[d]: kuter7639[d]: Getting that stuff dialed in has value, but I don't think most of the compiler people are as active on weekends to discuss. Faith is the most likely to know the details and be able to make use of the correct latencies and such.
16:17kuter7639[d]: redsheep[d]: Yeah. You are right. Thanks
16:20NyB: hmm... still not frozen
16:20NyB: already 15min with Xorg loaded, at the original weird resolution...
16:45NyB: so, if the OpenGL rendered is listed as llvmpipe, that means I have no hardware acceleration right now, correct?
16:45NyB: *renderer
16:46NyB: hmm... "Dedicated video memory: 0 MB" -> yeah, sounds like software rendering to me...
16:48tiredchiku[d]: correct
16:48NyB: OK, so nomodeset means no hardware acceleration on this machine, AFAICT
16:49tiredchiku[d]: it _shouldn't_ be that way, hm
16:51mhenning[d]: kuter7639[d]: I believe the hardware stalls on bank conflicts for us, so we don't need to model them for correctness
16:51mhenning[d]: what we have so far is very much pessimistic - focused on just getting things correct, not as fast as possible
16:51mhenning[d]: karol has access to some documentation on how some of this works, but I'm not sure how detailed it is
16:53mhenning[d]: I'm not aware of anyone reverse engineering the latency tables right now
16:58kuter7639[d]: mhenning[d]: http://arxiv.org/abs/2208.11174 there is this paper where they used PTX to measure latencies. Not sure if their methodology is correct + they seem to have confused uniform and regular data paths
16:59mhenning[d]: Yeah, there's also this paper which has some stuff for volta: https://arxiv.org/abs/1804.06826
16:59mhenning[d]: but I don't think either of them are anywhere near detailed enough for our purposes eg. that one is missing a lot of instructions
16:59kuter7639[d]: Already read that one but might have skipped the part about latencies
17:00kuter7639[d]: I have some ideas I am thinking on to measure or deduce the latencies automatically.
17:02mhenning[d]: Yeah, I was thinking about doing something to automatically generate ptx and read the latencies from the output of ptxas, but I haven't gotten around to implementing it
17:05kuter7639[d]: mhenning[d]: I have two ideas.
17:05kuter7639[d]: A) Try analyzing pre-compiled binaries where we look at data dependent instructions and count cycles.
17:05kuter7639[d]: B) Automatically generate kernels where we increase the delay/stall count until we get the correct result. The correct result here would also be automatically found using maximum delay.
17:05kuter7639[d]: 4
17:06DodoGTA: 4
17:07mhenning[d]: kuter7639[d]: yeah, those approaches would also probably work
17:12redsheep[d]: DodoGTA: 4?
17:13DodoGTA: redsheep[d]: kuter7639[d] said 4 for some reason
17:38NyB: so, every post I have found so far, although not specifically for nouveau, is that nomodeset basically disables hardware acceleration
17:39NyB: which is kinda what I'm seeing :)
18:53NyB: well, this is not confusing at all...
18:54NyB: blacklisting nouveau for boot, then loading it later... no issues
18:54NyB: X11, hwaccel, all fine.
18:54NyB: something happens _during_ boot that messes things up. W.T.H.?
18:55tiredchiku[d]: :thonk:
18:57NyB: OK, let's confirm this...
19:01NyB: step 1 - unblacklist nouveau, reboot with no X -> dies after 2m
19:02NyB: let's powercycle and try again...
19:02NyB: (I kinda miss a reset switch right now)
19:04NyB: I think this time it didn't even make it through boot. How the heck is it getting worse?
19:25NyB: hmm, OK, so blacklist nouveau, reboot, modprobe nouveau, start X11... 5 minutes in, all good. let's see :)
19:26NyB: let's kick off a CPU stress test, for fun
19:26NyB: hmmm... is there something to stress-test my GPU
19:42NyB: still fine...
19:43NyB: CPU is at 100C, GPU at 86C (same heatsink assembly)
19:43NyB: I'll tentatively say that it's not a hardware problem
19:43NyB: (yay?)
20:12selaaaa[d]: i should add i've been having very similar issues to NyB on my plasma wayland session on a modern PC build (running a 970, though), weirdly not in recent times
20:12selaaaa[d]: but i haven't had the time to actually debug or post anything on it just because i'm not home for enough time to diagnose it properly, but my system has locked up, completely unresponsive, can't switch tty nor anything but sysrq+b works
20:14NyB: in my cases SAK was not responding either
20:14selaaaa[d]: haven't been able to dump logs because i'm on alpine and the default kernel isn't built with kgdb support, so i can't load into a recovery kernel :p
20:14selaaaa[d]: ah, i see, interesting
20:15NyB: hmm... it's clocking 46.6 FPS on openarena
20:15clee: I got NVK working last night and I've been going through my Steam library and testing different games (so far everything runs, at least!) is there anywhere that would be useful for me to log the results of each game, like protondb maybe?
20:16tiredchiku[d]: https://docs.google.com/spreadsheets/d/1RuHD3Z_nBKCp618HHC5I9hOu0lqCoFYwQ4FM69M-Ajg/edit#gid=469568508
20:16tiredchiku[d]: you'll have to ask gfxstrand[d] for edit access to the spreadsheep
20:16tiredchiku[d]: spreadsheet*
20:17gfxstrand[d]: Yeah, just click the request access button and ping me when you have.
20:18tiredchiku[d]: hi faith :gura_wave:
20:18NyB: nvk sounds promising, although I suspect all of my nvidia GPUs are too old for it
20:18NyB: my latest one is a gtx970
20:18clee: gfxstrand[d]: clicked!
20:18NyB: switched to AMD for my last build
20:19clee: NyB: yeah, that's a few generations too old. 1660/20x0 or higher
20:19selaaaa[d]: nvk is still messy on 970. i tried it, it wasn't too great :p
20:19selaaaa[d]: most things don't work, but i expected as much, i was just curious really
20:19clee: (at least according to the public docs)
20:19gfxstrand[d]: clee: Granted!
20:19clee: thanks gfxstrand[d]!
20:20clee: currently running Tunic and it's a little stuttery but it looks gorgeous. no visual glitches.
20:20gfxstrand[d]: 🎉
20:21NyB: btw, does NVK have any support for compute?
20:21gfxstrand[d]: selaaaa[d]: Yeah, Maxwell is still a bit of a mess. We've got line of sight to getting it working for the most part but no one's hacked on it for a while.
20:21gfxstrand[d]: NyB: Yup. Compute shaders are a core part of Vulkan.
20:21clee: how do I tell which version of DirectX a game is using? google?
20:21selaaaa[d]: i'd love to contribute but i just have no knowledge in.. anything near hardware
20:21gfxstrand[d]: And we support a pretty competent feature set there.
20:21tiredchiku[d]: google is the best source, yes
20:21selaaaa[d]: i have electronics knowledge! and user-space programming knowledge!
20:21selaaaa[d]: just nothing imbetween lol
20:22redsheep[d]: clee: Enabling the dxvk HUD should tell you I believe
20:23clee: oh awesome, thanks
20:23tiredchiku[d]: it doesn't
20:24esdrastarsis[d]: tiredchiku[d]: Suggestion: date convention (dd/mm or mm/dd)
20:24redsheep[d]: tiredchiku[d]: Doesn't it list a feature level?
20:24NyB: esdrastarsis[d]: yyyy-mm-dd
20:24clee: esdrastarsis[d]: today is 7/7 so it doesn't matter
20:24selaaaa[d]: esdrastarsis[d]: ISO8601 :p
20:24tiredchiku[d]: oh wait nvm
20:24tiredchiku[d]: full dxvk hud does show what's being used
20:24redsheep[d]: Well anyway mangohud tells you that it's dxvk or vkd3d which at least gives you broad strokes
20:24tiredchiku[d]: am blind
20:25redsheep[d]: Ok so I'm not crazy
20:26redsheep[d]: esdrastarsis[d]: We probably need a legend/guide somewhere, all dates are intended to be day/month/year
20:26redsheep[d]: Painful as it is for us Americans to read
20:27tiredchiku[d]: fixed
20:27tiredchiku[d]: :D
20:28esdrastarsis[d]: some dates need fixes like GTA IV
20:28redsheep[d]: Having a better definition of the statuses would probably be good but I spent a whole thinking about it and decided it's just too subjective to draw good lines
20:29redsheep[d]: Basically it's just whether it was a good experience, a poor experience, or a complete non starter
20:29tiredchiku[d]: esdrastarsis[d]: done
20:31redsheep[d]: I do think there are a few entries where they're green where I'd have marked 🫤 if I'd done it
20:32redsheep[d]: I don't think we should be marking things green if performance is below 60 fps or a reasonable percentage or prop driver performance
20:32redsheep[d]: If the game is just hard to run, it's hard to run
21:01clee: redsheep[d]: have you tried Hollow Knight yet? :D
21:01clee: (I get 180-220fps at 3440x1440)
21:02tiredchiku[d]: clee what CPU do you have, out of sheer curiosity
21:02redsheep[d]: I don't have that game, that's pretty good performance though
21:02tiredchiku[d]: also I can try it out
21:02redsheep[d]: Lots of lighter games like that, especially if they're usually CPU bound, will run great
21:03redsheep[d]: I've noticed some scenarios where CPU bound seems slow on NVK but broadly speaking it's the GPU bound titles with the furthest way to go
21:03clee: tiredchiku[d]: 5950X, but my RAM is currently running at 2133MHz because it doesn't POST at its own XMP speeds
21:03tiredchiku[d]: oof :( that's unfun
21:04tiredchiku[d]: I should really finish adding all my games to that online library management thing
21:04tiredchiku[d]: so that I have one place to see all the games I own across platforms
21:04tiredchiku[d]: would be helpful for testing purposes 😅
21:05redsheep[d]: CPU bound perf on NVK usually seems to fall in the 60-120% of prop range, and GPU bound is usually more like 30-70%
21:05tiredchiku[d]: like, last week I learned I owned Talos 1
21:06clee: Horizon Zero Dawn has some real interesting rendering glitches
21:06clee: feels very trippy
21:06redsheep[d]: Yeah that one made me nauseated after a few minutes
21:09redsheep[d]: Still, it's impressive that it works at all being one of the dx12 titles on the list
21:50clee: Star Wars Jedi Fallen Order is totally playable as long as you don't need to enter the settings menu a second time
21:52NyB: so, I diffed the logs of a nouveau-blacklisted boot, and a nouveau-active boot. There might be ordering differences (did not go that deep), but I noticed nothing new that might explain my problem.
22:03NyB: I think I'll quit for now... maybe a future system upgrade will magically fix it :(
23:23xiphmont: rehi
23:33clee: running Hollow Knight under SteamPlay/DXVK, still over 100fps but not quite as much as native. https://youtu.be/qgd-O6Sff8s