00:27 mhenning[d]: gfxstrand[d]: Do you remember why spilling RegFile::Bar takes two temporary gprs? https://gitlab.freedesktop.org/mesa/mesa/-/blob/d455d5d86cd940ea16208007c6260b4cab63738f/src/nouveau/compiler/nak/assign_regs.rs#L1429
00:28 mhenning[d]: I don't see why it would take more temp space than spilling a pred
00:33 mhenning[d]: Never mind, I see the difference in lower_par_copy now
00:33 gfxstrand[d]: One foot the address
00:33 gfxstrand[d]: Or that
00:33 gfxstrand[d]: Yeah, predicate swaps are easy.
06:38 ukleinek: gfxstrand[d]: so I need to subscribe first? Or might the mail still be in a moderator queue? Who would I have to ask for forwarding it to the list then?
10:42 karolherbst: normally most posts go through, but sometimes messages are held (if they are too big, or whatever other reason)
11:03 ukleinek: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1092189#30 and https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1092189#35 are the messages that didn't make it to the list
11:04 ukleinek: I would expect they pass the usual checks for mailing lists.
11:28 karolherbst: interesting
11:31 karolherbst: ukleinek: but anywya, if you are willing you could help out with figuring out what broke it. Most of the devs here are focusing on vulkan/newer hardware, and others are kind of either burned out too much or have other priorities atm to look at issues like that. Though I don't think this will require more than doing a git bisect of mesa at this
11:31 karolherbst: point in time to know what regressed this or causing it. Well.. assuming it did work in the past.
13:39 ukleinek: karolherbst: ok, I'll forward that info and ask the reporter if they can bisect
14:16 gfxstrand[d]: NVIDIA is having website trouble this morning. 😕
14:17 tiredchiku[d]: all the bots and scripts and people lining up
14:19 gfxstrand[d]: No, like https problems.
14:20 notthatclippy[d]: The website itself or one of the downloads repositories?
14:20 gfxstrand[d]: marketplace.nvidia.com
14:21 tiredchiku[d]: global customers and bots/scripts lining up :P
14:22 notthatclippy[d]: gfxstrand[d]: Huh. It works or doesn't work for me depending on which firefox container I use...
14:22 gfxstrand[d]: neither ff nor chrome are working here
14:23 notthatclippy[d]: I don't understand, there's nothing meaningfully different between the two containers here. Chrome works for me.
14:24 gfxstrand[d]: Well if it works for anyone then they're probably sold out and I got screwed by the internet. 😡
14:25 mohamexiety[d]: sadly nah it's pretty messed up
14:25 mohamexiety[d]: bestbuy et al are also having severe issues
14:26 marysaka[d]: right it's today 🙃
14:26 mohamexiety[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1334530201194991666/image.png?ex=679cdd9a&is=679b8c1a&hm=b285c123834c7d6cf7b7a100feb74e100056324df0e8f621ae6f37c89f6da986&
14:26 mohamexiety[d]: welp it works but
14:26 marysaka[d]: it was working for me for a little bit and now it's failing too
14:26 mohamexiety[d]: marysaka[d]: I did hear retail (in store) availability is better in Europe so could try that I guess
14:28 marysaka[d]: mohamexiety[d]: hmm already out of stock everywhere I look
14:28 mohamexiety[d]: sad 😦
14:29 marysaka[d]: yeah no all out of stock it seems :blobcatnotlikethis:
14:30 gfxstrand[d]: I have a local BestBuy. They open in 1.5 hours. I wonder how many people are lined up out front.
14:30 tiredchiku[d]: I heard people are selling spots in the lines at some MicroCenters for upwards of 2500$
14:31 mohamexiety[d]: yeah microcenter had people camping for _days_
14:31 mohamexiety[d]: insanity
14:34 gfxstrand[d]: It's been half an hour and the NVIDIA website is still busted. Do I give up no?
14:34 tiredchiku[d]: out of stock in germany
14:34 gfxstrand[d]: from nvidia?
14:34 tiredchiku[d]: yes
14:35 HdkR: github also having issues. Seems like the bots are hammering the world :P
14:35 karolherbst[d]: OpenAI got owned, now they try 100 times harder
14:35 tiredchiku[d]: freedesktop gitlab is also slower than usual
14:35 tiredchiku[d]: :froge:
14:36 karolherbst[d]: yeah.. some of the admins took a look and it's a lot of bot traffic...
14:36 karolherbst[d]: I hope it gets outlawed, lol
14:39 gfxstrand[d]: Yeah, I should probably just give up at this point.
14:39 gfxstrand[d]: grrrr
14:40 mohamexiety[d]: hope you can find it locally at least 😦
14:40 gfxstrand[d]: I can try BestBuy. They open in 1.5 hours.
14:41 gfxstrand[d]: I should drive over there now and see if there's a line.
14:41 gfxstrand[d]: I don't even know if they have any in-store. The website isn't being helpful.
14:41 tiredchiku[d]: all the best
14:42 gfxstrand[d]: Yeah, looks like they don't have anything better than a 4060 in store
14:45 gfxstrand[d]: Yeah, I'm giving up
14:46 gfxstrand[d]: I guess I should have tried through BestBuy instead of hoping NVIDIA would work.
14:46 gfxstrand[d]: Not that BestBuy wouldn't have gotten DOSd too
14:46 tiredchiku[d]: we've got non-FE 5080s available here in India, if you're interested /j
14:46 tiredchiku[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1334535280404860978/jnTXvyj.png?ex=679ce255&is=679b90d5&hm=5d697a6c9d1c659aa1962aab625e8650db6cc77df940f4edb52b7f48fca61131&
14:46 tiredchiku[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1334535280711041085/Hzk8kDQ.png?ex=679ce255&is=679b90d5&hm=4a872135543b15a7ff075ab512a4fbb6beefef5db12874fd8e6a225b7457d96c&
14:46 gfxstrand[d]: non-FE?
14:46 tiredchiku[d]: vendor cards, yeah
14:51 gfxstrand[d]: Actually in stock even?
14:52 tiredchiku[d]: yup
14:54 tiredchiku[d]: the market here is small, and those prices are exorbitant for most of the population
14:54 tiredchiku[d]: considering our economy
14:55 gfxstrand[d]: annoyingly, all those are 16GB cards. 😢
14:55 gfxstrand[d]: Oh, they're 5080s
14:55 gfxstrand[d]: that makes the price make sense.
14:55 tiredchiku[d]: tiredchiku[d]: yup
14:55 tiredchiku[d]: 5090 should be available by this weekend
14:56 gfxstrand[d]: Watch for the 5090s and maybe I'll do that.
14:57 tiredchiku[d]: oh, um
14:57 tiredchiku[d]: sure :)
14:57 gfxstrand[d]: That's still almost going to be double MSRP. 😢
14:59 gfxstrand[d]: ugh
15:01 gfxstrand[d]: Not that we're ever going to see that MSRP again now that Trump tariff'd the shit out of TSMC.
15:01 tiredchiku[d]: yeah
15:01 tiredchiku[d]: gfxstrand[d]: + India markup
15:02 tiredchiku[d]: tiredchiku[d]: the cheapest one here has a list price of 1500 usd on asus' website
15:29 karolherbst[d]: have you seen that one? https://mastodon.social/@CalcProgrammer1/113916128544084215 random person doing random testing
15:59 zmike[d]: all I see there is 💪 ZINK :muscleright:
16:02 tiredchiku[d]: yeah zink with nvk is quite nice
16:03 Jasper[m]: I do kinda hope he tests the framerate outputting on the 3080 aswell. iirc the render offloading stuff limits pcie bandwidth a bit
16:12 tiredchiku[d]: oh no wonder the name was familiar
16:12 tiredchiku[d]: karolherbst[d]: he's the OpenRGB dev
16:12 karolherbst[d]: ahhh
16:36 mohamexiety[d]: does latest Fedora use zink on nvk automatically now?
16:37 gfxstrand[d]: I should look at the DOOM games. I have both of them.
16:37 mohamexiety[d]: soon they shall be triplets even
16:38 mohamexiety[d]: think Dark Ages has mandatory RT though
16:39 zmike[d]: fwiw DG2 was pulling like 150fps in doom2016 last I checked
16:39 zmike[d]: though that was a year or two ago
16:40 zmike[d]: v embarrassing to be losing there
16:40 zmike[d]: (on zink)
17:10 gfxstrand[d]: I suspect we're losing a bunch of perf to our naive descriptor handling. I really need to figure out a better plan there. We have too many indirections right now.
18:02 asdqueerfromeu[d]: gfxstrand[d]: Do you mean descriptor pools or descriptors in general?
18:15 gfxstrand[d]: in general
18:45 gfxstrand[d]: Ugh... We're kinda screwed for D3D12. 😢
18:46 gfxstrand[d]: There's two whole indirects I can't get rid of
18:46 karolherbst[d]: painful
18:46 gfxstrand[d]: Yeah.
18:46 gfxstrand[d]: EDB is evil
18:46 karolherbst[d]: what's the indirection?
18:46 gfxstrand[d]: Well, EDB itself would be kind of fine
18:47 gfxstrand[d]: The problem is that VKD3D makes huge EDB buffers
18:47 karolherbst[d]: I think some of it is fixable with Turing+ if you are up for some trickery
18:48 karolherbst[d]: but that kinda depends on what the issue there is
18:48 gfxstrand[d]: karolherbst[d]: It's a texture op so I need the handle. I have to load that from a descriptor buffer. Because it's EDB with a huge buffer, my max offset is 8MB range, way outside CBuf size. So I have to load the descriptor set address from a cbuf, add the offset, load the descriptor, and then texture.
18:49 karolherbst[d]: why is it 8MB? spec?
18:49 gfxstrand[d]: Normally, if you have normal descriptors or even EDB with a reasonably sized buffer, we can optimize all that to a cbuf and then to a bound texture op.
18:49 gfxstrand[d]: karolherbst[d]: It's D3D12 max * descriptor size or similar
18:49 karolherbst[d]: I see...
18:50 karolherbst[d]: so one indirection for the buffer itself, another for the tex handle.. mhh
18:50 karolherbst[d]: sounds evil
18:50 gfxstrand[d]: Yeah
18:50 gfxstrand[d]: All my tricks to get rid of indirections depend on me being able to upgrade stuff to cbufs
18:51 gfxstrand[d]: It's all ldc or ld.constant but stifll
18:51 karolherbst[d]: yeah.. that's not gonna work with 8MiB buffers...
18:51 gfxstrand[d]: Nope
18:51 karolherbst[d]: have you checked if nvidia does some trickery there?
18:52 gfxstrand[d]: I think this is why proton sucks on NVIDIA
18:52 gfxstrand[d]: that and structure buffers
18:52 karolherbst[d]: sounds likely
18:52 gfxstrand[d]: On D3D12, the D3D heap is just the hardware texture table
18:52 karolherbst[d]: I was wondering why my a lot faster nvidia GPU didn't perform that much better than my slower AMD GPU for some time 🙃
18:53 karolherbst[d]: could shader variant it.... but that's kinda painful
18:53 karolherbst[d]: though might not even work
18:54 gfxstrand[d]: I don't think that works
18:54 gfxstrand[d]: I can maybe do something nuts with some sort of preamble shaders
18:54 karolherbst[d]: mhhh
18:55 karolherbst[d]: not sure if that ends up more expensive
18:55 karolherbst[d]: but probably
18:55 karolherbst[d]: it's not like you win _that_ much
18:55 karolherbst[d]: the tex ops just pull data from a cbuf anyway
18:56 gfxstrand[d]: I did preamble shaders for ANV back in the day. It helped a lot, except when the increased push size blew out the register file. We don't have that problem with NV>
18:56 gfxstrand[d]: The big problem with NV is stalls due to repeated cbuf binding
18:56 karolherbst[d]: sooooo...
18:56 karolherbst[d]: could you put the stuff the tex ops pulled, into the desc put into the edb?
18:56 karolherbst[d]: could just do a bindless thing in the tex instead
18:57 gfxstrand[d]: I'm not sure what you mean
18:58 karolherbst[d]: like in full bindless mode the tex doesn't have to fetch stuff from a cbuf as you simply give it the header + sampler ptr directly
18:58 gfxstrand[d]: Yeah. That's what we're doing now.
18:58 gfxstrand[d]: the header+sampler goes in the EDB
18:59 karolherbst[d]: ahh, I see
18:59 gfxstrand[d]: But we have to load it from the EDB and we can't promote the EDB to a cbuf anything
18:59 karolherbst[d]: yeah then not quite sure what else you could improve here
18:59 karolherbst[d]: yeah.. fair
18:59 gfxstrand[d]: We could hallucinate a bindless cbuf handle, I suppose.
18:59 karolherbst[d]: mhhhhhhh
19:00 karolherbst[d]: could mess with the cache a lot
19:00 karolherbst[d]: but like if only small ranges of the EDB are used, then yeah, that could certainly work
19:00 karolherbst[d]: I'm sure it could trash perf if it's all random
19:01 karolherbst[d]: you can predict the access ranges for some shaders, right? it's not all super indirect, is it?
19:02 gfxstrand[d]: It's super indirect
19:03 karolherbst[d]: what a pity
19:04 karolherbst[d]: is it at least uniform?
19:04 karolherbst[d]: oh right.. speaking of uniform...
19:04 karolherbst[d]: uhm...
19:04 karolherbst[d]: how much do you do non uniform accesses into constant buffers?
19:04 karolherbst[d]: or rather.. how likely is it with d3d12 and such
19:04 karolherbst[d]: because uhm... don't do that
19:05 gfxstrand[d]: Well, yeah, I know it sucks
19:05 karolherbst[d]: yeah, but a global load can be faster in such cases
19:06 karolherbst[d]: though I'm not sure how bad it is in the average case
19:06 karolherbst[d]: but there are some stalls on the subgroup level when it's all indirect
19:08 mhenning[d]: maybe vkd3d could be modified to use less than the full 8mb if the application doesn't use that much?
19:10 karolherbst[d]: I wonder if some of the rebinding can also be prevented by having the index indirectly and doing smart things (tm) in the driver by only binding new cbs when needed or so
19:11 karolherbst[d]: with the model GL had, applications could e.g. always use the same buffer in slot 4 or whatever
19:11 karolherbst[d]: so you didn't need to rebind for applications caring about perf
19:11 karolherbst[d]: so I wonder if having the index more dynamic could help with perf
19:12 karolherbst[d]: might burn some uniform registers, but those are for free anyway
19:12 karolherbst[d]: uhh wait..
19:13 karolherbst[d]: the problem with that is, that alus require constant index...
19:13 karolherbst[d]: and offset
19:13 karolherbst[d]: uhh..
19:13 karolherbst[d]: annoying
19:21 gfxstrand[d]: mhenning[d]: I think most apps ask for 1M descriptors
19:22 gfxstrand[d]: I think a lot of our perf problems are stalling. I'm still feeling no heat coming off my 4060
19:22 gfxstrand[d]: It should be yelling at me
19:27 tiredchiku[d]: yeah, I tried running alan wake 2 on my 3070 and cranked everything up, and while perf was at 3fps, I didn't hear the fans ramp up
19:28 gfxstrand[d]: I wish Veilguard didn't take 15 minutes to compile shaders. 😭
19:30 mhenning[d]: yeah, who wrote that compiler anyway? terribly slow 😛
19:31 gfxstrand[d]: IDK, from what my little brother (a prop user) tells me, it's about on par with the blob.
19:31 gfxstrand[d]: It's just that there are a LOT of shaders and it compiles them all up-front
19:32 mhenning[d]: That's fair
19:32 mhenning[d]: Admittedly most users probably don't recompile shaders all that often
19:33 orowith2os[d]: My system has a weird thing where it'll download or redo the shaders whenever the drives get plugged/unplugged during and between steam runs
19:33 orowith2os[d]: But it's a niche case involving mimicking the steam deck's hot plugging of storage drives on a normal laptop
19:34 orowith2os[d]: And I haven't looked into it all too much
19:34 orowith2os[d]: Steam not being able to access the system bus might somewhat contribute to that, I don't know how it does things on the deck
19:35 tiredchiku[d]: steam shader pre-cache is a bit different to what's being talked about 😅
19:36 gfxstrand[d]: It looks like `NVK_DEBUG=no_cbuf` improves my DAI trace in RenderDoc. I'm launching the full game to see if there's an FPS difference.
19:36 orowith2os[d]: tiredchiku[d]: Oopsies
19:40 gfxstrand[d]: gfxstrand[d]: Maybe a little but not noticable.
19:42 mhenning[d]: Anyway, if anyone wants to make their shader compiles take even longer, my prepass scheduler is finally ready https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33306
19:43 gfxstrand[d]: Let's try that just for grins
19:45 avhe[d]: https://ffmpeg.org/pipermail/ffmpeg-devel/2025-January/339082.html
19:45 avhe[d]: dwlsalmeida[d]
19:45 avhe[d]: 10-bit h264 and 422 subsampling for h264 and hevc on blackwell
19:45 gfxstrand[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1334610500566384700/image.png?ex=679d2863&is=679bd6e3&hm=4f299ca691fc11b7448f89df163fd857f765891abb5fbaa03cac54c8d153c749&
19:45 gfxstrand[d]: mhenning[d]: Maybe not...
19:46 avhe[d]: pretty weird that they're adding 10-bit h264 in 2025 but ¯\_(ツ)_/¯
19:46 mhenning[d]: gfxstrand[d]: Want to send me that shader? I don't hit that on either shaderdb or cts
19:46 gfxstrand[d]: I'm not sure which shader it is. :blobcatnotlikethis:
19:47 mhenning[d]: You can also try commenting that assert out - it's a check that the heuristic is calculating things right but it won't affect correctness
19:48 mhenning[d]: gfxstrand[d]: A fossil with everything would work fine
19:48 gfxstrand[d]: Yeah. RenderDoc doesn't have a tool to dump all pipelines to a foz
19:48 gfxstrand[d]: I suppose I could run RenderDoc on top of fossilize. 🙃
19:49 mhenning[d]: or just the game on top of fossilize?
19:49 gfxstrand[d]: That's a LOT of shaders
19:49 phomes_[d]: gfxstrand[d]: doom 2016 crashes with `renderAPI: ../src/vulkan/runtime/vk_buffer.h:69: vk_buffer_range: Assertion 'range + offset <= buffer->size' failed.`
19:49 mhenning[d]: Ah, fair
19:49 gfxstrand[d]: At least with a RenderDoc trace, you only get the ones needed for that scene
19:50 gfxstrand[d]: phomes_[d]: Right. Have you tried with the PCI ID hack?
19:50 phomes_[d]: not yet. I did zero_memory
19:50 gfxstrand[d]: Or maybe we should just make that helper clamp
19:50 tiredchiku[d]: :thonk:
19:50 tiredchiku[d]: did doom 2016 break again
19:50 mohamexiety[d]: doomed
19:55 mhenning[d]: gfxstrand[d]: Oh, actually there's the free "dragon age the veilguard character creator". I'll test with that later and see if I can reproduce - no need to send me the shader
19:57 phomes_[d]: gfxstrand[d]: still the same crash with that
19:58 mhenning[d]: If it used to work, a bisect could be helpful
19:58 phomes_[d]: I am heading to FOSDEM soon. I can take a closer look and bisect monday unless somebody else wants to do it
20:22 gfxstrand[d]: mhenning[d]: That (with a rebase) maybe helps DA:TV by 3-5%. It's hard to tell since the FPS doesn't move.
20:23 asdqueerfromeu[d]: gfxstrand[d]: You can always use the frametime ⏲️
20:32 gfxstrand[d]: We're like 5x behind NVIDIA on geekbench
20:32 gfxstrand[d]: ugh
20:39 gfxstrand[d]: Oh, wait, no. We're just 1/2 NVIDIA
20:39 gfxstrand[d]: That's not nearly as bad.
20:41 gfxstrand[d]: But running geekbench my fans barely ran. So maybe heat isn't a great indication?
20:42 tiredchiku[d]: oh yeah, default GSP fan curves are weird
20:43 tiredchiku[d]: I've seen my GPU be at 73C and still only around 55% fan
20:43 tiredchiku[d]: on prop
20:43 tiredchiku[d]: (openrm)
20:43 tiredchiku[d]: we really need hwmon e-e
20:49 gfxstrand[d]: https://browser.geekbench.com/v6/compute/compare/3574826?baseline=3574930
20:50 gfxstrand[d]: The easiest way to run the blob was to just boot Windows. 🙃
20:52 tiredchiku[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1334627233851244716/xhOFJHh.png?ex=679d37f8&is=679be678&hm=fdd4a219dd3730dd2d06be2317f810a94d3d65043189dcbee42321c96455e024&
20:52 tiredchiku[d]: tiredchiku[d]:
20:54 tiredchiku[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1334627805723627550/E0xXQ3C.png?ex=679d3881&is=679be701&hm=d7c22abc34690ec1ed9f63850ce6e7941c1f79f238d7af35eeb57bbef2b77f71&
20:54 tiredchiku[d]: yeah..
20:54 anholt: captured some traces (bevy demos, powervr sdk demos, hollow knight, bigger games failed at gfxr capture), didn't find any perf difference from the greedy instruction scheduler.
20:56 gfxstrand[d]: Oh, hey anholt!
20:56 anholt: hi! I'm looking at computers again!
20:56 gfxstrand[d]: Cool beans.
20:57 gfxstrand[d]: If you dropped in and hacked on NAK, I would not complain. 🙂
20:57 gfxstrand[d]: (Or NVK in general)
20:58 karolherbst[d]: hoi anholt!
20:58 anholt: We'll see how much time I get -- the nonprofit I've been busy with this month *should* settle down soon.
20:58 zmike[d]: anholt!
20:58 zmike[d]: we miss you
20:58 anholt: aw, thanks!
20:59 anholt: it took a long time to miss this whole thing, but I do now. burnout sucked.
20:59 zmike[d]: I believe it
21:00 gfxstrand[d]: Yeah, burnout is pretty horrible.
21:01 anholt: current interest is in building perf comparison tools I don't hate, and maaaybe perfetto but that doesn't seem like a great way to remember the joy of hacking on software.
21:01 gfxstrand[d]: I'm glad to hear from you again. 💜 I was a little worried you nooped out for good.
21:01 djdeath3483[d]: anholt: Ahahah 🙂
21:01 gfxstrand[d]: I'm a big fan of perf tools we don't hate.
21:01 zmike[d]: perfetto :fullheadache:
21:01 airlied:would love to see a really good gpu profiler :-P
21:01 djdeath3483[d]: perfetto, nice to use, not nice to look at
21:01 gfxstrand[d]: Because looking at GPU timestamp times in RenderDoc is NOT optimal. 🙃
21:02 anholt: airlied: clearly the solution is to build a new gpu trace and profiling tool. nothing has ever gone wrong when someone had that idea.
21:02 zmike[d]: rb
21:04 gfxstrand[d]: What I really want are GPU metrics. I wonder if it's even possible to get nsight to work with NVK.
21:04 gfxstrand[d]: I need to chat with some NV folks about that.
21:05 gfxstrand[d]: The fact that the RADV team can use AMD's profiling tools is pretty killer
21:09 mohamexiety[d]: yeah that would be really good
21:11 airlied:has no idea what a gpu aware version of perf would be like :-P
21:16 gfxstrand[d]: Uh... Wat?!?
21:16 gfxstrand[d]: [15:15:29.220] Loading module '/usr/lib64/libweston-14/gl-renderer.so'
21:16 gfxstrand[d]: MESA: error: ZINK: vkEnumeratePhysicalDevices failed (VK_ERROR_INITIALIZATION_FAILED)
21:16 gfxstrand[d]: MESA: error: ZINK: failed to choose pdev
21:16 gfxstrand[d]: libEGL warning: egl: failed to create dri2 screen
21:16 gfxstrand[d]: libEGL warning: egl: failed to create dri2 screen
21:16 gfxstrand[d]: [15:15:29.239] failed to initialize display
21:16 gfxstrand[d]: [15:15:29.239] EGL error state: EGL_NOT_INITIALIZED (0x3001)
21:16 gfxstrand[d]: How does `vkEnumeratePhysicalDevices` fail?
21:17 orowith2os[d]: computer goblins
21:17 orowith2os[d]: make sure no env vars are fucking with it
21:19 gfxstrand[d]: Yeah, this was literally just working
21:19 gfxstrand[d]: Now not even vulkaninfo works
21:22 gfxstrand[d]: Oh, my GPU fell off the bus.
21:25 gfxstrand[d]: Ugh... Weston+NVK+Zink doesn't work via the X11 back-end. It just fails but doesn't crash
21:27 gfxstrand[d]: No, that's the Wayland backend
21:27 gfxstrand[d]: X11 backend seems fine. But also it doesn't crash. 😢
21:27 tiredchiku[d]: wdym x11 backend
21:27 orowith2os[d]: probably running it nested
21:27 gfxstrand[d]: I mean Weston is running in an X11 window
21:27 tiredchiku[d]: oh, looked it up
21:27 tiredchiku[d]: neat
21:31 gfxstrand[d]: I guess I'm going to have to use an actual VT if I want to hunt this down. 😭
21:37 mohamexiety[d]: VT?
21:38 mhenning[d]: virtual terminal
21:38 mohamexiety[d]: ahhh damn
21:38 mhenning[d]: It's what you get when you hit control-alt-f5
21:46 gfxstrand[d]: Of course Weston doesn't want to respect my environment variables and keeps trying to load radeonsi...
21:47 gfxstrand[d]: I hate debugging compositors...
21:53 karolherbst[d]: same
21:59 airlied: I run a screen session on my vt, start the compostior in gdb and reconnect the screen sesion from ssh, I still use Xorg too much because I don't have to do that :-P
22:02 gfxstrand[d]: Oh, I hadn't thought of that
22:02 zmike[d]: that's literally what I've been saying to do lmao
22:02 zmike[d]: except from your phone
22:03 zmike[d]: have to do it from your phone to get best results.
22:04 gfxstrand[d]: Oh, it's the screen session bit that's interesting. You didn't suggest that, at least not as far as I remember.
22:04 zmike[d]: I did
22:04 gfxstrand[d]: Then I missed it
22:04 zmike[d]: it's the only way to debug this intersection
22:05 zmike[d]: and weston should respect MESA_LOADER_DRIVER_OVERRIDE ?
22:05 zmike[d]: or at least it always has for me
22:05 gfxstrand[d]: You can always put a sleep(5) somewhere and gdb --pid
22:05 zmike[d]: no that's steam debugging
22:05 zmike[d]: smh can't mix the debug streams like that
22:05 zmike[d]: no wonder your compositor is trying to rebel
22:07 zmike[d]: I assume it'll work fine for you on drm though
22:26 gfxstrand[d]: Okay, disabling the integrated AMD GPU. I think that's messing everything up
22:35 gfxstrand[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1334653291925340250/image.png?ex=679d503d&is=679bfebd&hm=9e4e58054027fb1ede802a5ffed39619ce06aeb1532924a8d406e7879bc7a13e&
22:36 gfxstrand[d]: That's on an RTX 2060
22:38 gfxstrand[d]: Wait... How are you even getting resizeable_bar on a 2070?!?
22:38 gfxstrand[d]: That shouldn't be possible.
22:42 gfxstrand[d]: Do I need to run xeyes or something else?
22:43 zmike[d]: idk
22:43 zmike[d]: I was doing about that
22:44 gfxstrand[d]: gears triggered it?
22:44 zmike[d]: no, trying to run weston triggered it
22:45 gfxstrand[d]: 😢
22:45 gfxstrand[d]: Here's my Weston command: `mesa-run -w nvk --zink gdb --args weston -Bdrm --continue-without-input --xwayland -i0`
22:46 gfxstrand[d]: That's running latest NVK and latest Zink
22:46 gfxstrand[d]: origin/main from earlier today
22:46 gfxstrand[d]: On a 12GB RTX 2060
22:46 gfxstrand[d]: Is your NVIDIA GPU the only GPU in the machine?
22:46 zmike[d]: yep
22:47 zmike[d]: well there's intel igpu also
22:47 gfxstrand[d]: Hrm...
22:47 zmike[d]: but I've never had issues with that before
22:47 gfxstrand[d]: What if you disable the Intel in bios?
22:47 gfxstrand[d]: Just trying to limit the parameter space here.
22:47 zmike[d]: I guess I can test that tomorrow
22:47 gfxstrand[d]: That above is with my AMD iGPU disabled
22:47 gfxstrand[d]: I couldn't get it to run with my AMD iGPU enabled, regardless of whether or not I used Zink.
22:48 zmike[d]: yeah but that was your entire gpu not being detected
22:50 gfxstrand[d]: I still suspect this has something to do with dma-buf. Cross-GPU sharing would trigger that
22:52 gfxstrand[d]: And now it's working even with the AMD iGPU re-enabled. :facepalm:
22:55 zmike[d]: I guess it's not a big deal to disable it anymore now that I won't have janesma screaming at me every time I break gen9
22:55 gfxstrand[d]: 🤭
22:56 zmike[d]: in any case this week was a bit of a personal disappointment
22:57 zmike[d]: was really looking forward to 🤠 time but there were no actual bugs I could repro
22:57 zmike[d]: except this one
22:57 zmike[d]: and I can't even launch steam because I can't do a 32bit cross-compile
23:14 orowith2os[d]: zmike[d]: Have you tried adapting the arch manifest for mesa-git (with NVK) to it, and running steam inside of arch?
23:14 orowith2os[d]: Not sure what you're doing now to make things (not) work
23:15 zmike[d]: what I'm doing is trying to build 32bit mesa like usual, but now that nvk requires so many rust subprojects it can no longer be built
23:16 zmike[d]: steam is a mixed 32/64 app, so I can no longer run it
23:16 gfxstrand[d]: zmike[d]: And I would love to fix this one for you but I can't repro. 🤡
23:16 zmike[d]: the struggle is immense
23:16 zmike[d]: gfxstrand[d]: we are a matched set
23:16 orowith2os[d]: zmike[d]: I feel like there's a 32-bit setup somewhere on the aur you could steal
23:16 gfxstrand[d]: zmike[d]: Is yours in a laptop?
23:16 zmike[d]: I'm also not using arch
23:17 zmike[d]: no it's a desktop
23:17 orowith2os[d]: Distrobox ✨
23:17 gfxstrand[d]: Okay, so we can discount power management making your maps go away
23:19 orowith2os[d]: orowith2os[d]: Should work for your needs - I used it to mess with mesa-git a while back, since I couldn't do so natively. Steam doesn't have any issues inside of it that I've encountered, compared to the flatpak.
23:19 zmike[d]: 🤔
23:19 orowith2os[d]: I can link my setup command if you want to give it a try
23:19 zmike[d]: maybe next week, I'm already out of hours this week so probably I won't be doing much tomorrow
23:20 orowith2os[d]: Alrighty, just hmu if you want to give it a shot
23:20 zmike[d]: will do, thanks
23:23 mhenning[d]: zmike[d]: 32-bit cross also works for me (on arch, following what the arch official 32-bit package does)
23:23 zmike[d]: bizarre
23:29 gfxstrand[d]: Okay, both DOOMs are downloading. I will soon have a desktop of DOOM.
23:29 gfxstrand[d]: https://tenor.com/view/tonight-doom-gif-19873326
23:34 orowith2os[d]: Both? I thought there were at least five :ferrisClueless:
23:34 orowith2os[d]: If you aren't testing og doom what are you even doing smh my head
23:42 mhenning[d]: the software rendered dooms dont use NVK™ Certified Organic Triangles