00:38 fdobridge: <M​isyl with Max-Q Design> Vulkan API version is kinda useless imo
00:38 HdkR: 1.3 with all the optionals not implemented would be a very limited implementation
00:40 fdobridge: <z​mike.> IMG wanted it, they're shipping zink as their official GL driver, I don't have a strong opinion so here we are
00:41 fdobridge: <a​irlied> I've got a patch to reorder things
00:41 fdobridge: <a​irlied> just so we can paint a proper bikeshed
00:42 fdobridge: <a​irlied> unleash the painting https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29296
00:42 fdobridge: <r​edsheep> I might have found another zink session issue on nvk but I am not sure if I should file it, because it migth not be actionable... Building a kernel puts the cpu under so much load that the session basically locks up, and it does not happen on an nvc0 session
00:43 fdobridge: <z​mike.> https://cdn.discordapp.com/attachments/1034184951790305330/1242276451844362362/a7e1dee7-69af-41aa-a7f5-8246230f5519_1.png?ex=664d3f9a&is=664bee1a&hm=d93e983df93832e716743d0dba75102a21ef1d584294c2d8a7172ecd469d540d&
00:44 fdobridge: <r​edsheep> Does anybody have a zink session on gnome to test if it will lock up there when under really high cpu load?
00:45 fdobridge: <z​mike.> damn I jumped the gun
00:46 fdobridge: <z​mike.> now I need a rubber stamp that says SHEDCRAFTING IN PROGRESS
00:49 fdobridge: <a​irlied> oops pushed too much
00:49 fdobridge: <a​irlied> repushed
00:59 fdobridge: <c​langcat> Fun fact getting the ink off skin from these is hard.
01:58 fdobridge: <r​edsheep> I am testing the branch again and I don't see obvious bugs so far, so that is good. On the other hand, it occurred to me that I was not testing no_cbuf on mesa main before which is kinnd of important to getting useful data. Here's mesa main > cbuf0 branch:
01:58 fdobridge: <r​edsheep>
01:58 fdobridge: <r​edsheep> Without NVK_DEBUG=no_cbuf
01:58 fdobridge: <r​edsheep> Witness 58 > 58 fps
01:58 fdobridge: <r​edsheep> Talos 82 > 84 fps
01:58 fdobridge: <r​edsheep> With NVK_DEBUG=no_cbuf
01:58 fdobridge: <r​edsheep> Witness 104 > 107 fps
01:58 fdobridge: <r​edsheep> Talos 74 > 75 fps
01:59 fdobridge: <r​edsheep> It does seem to just raise fps a bit everywhere which is nice, but with the extra data the results are not as drastic in context
02:00 fdobridge: <r​edsheep> I don't know that this actually provides a concrete counterexample to this being the direction to go for good performance, but it did make me a little sad.
02:05 fdobridge: <r​edsheep> Is the performance bump on mesa main from no cbuf expected, and still indicates there's something there?
02:09 fdobridge: <g​fxstrand> A small bump from no_cbuf on a big card isn't surprising
02:39 genderfreedom: in the public space so retarded people were being delt by Finnish afro viking prostitute bums, who harassed our businesses and investments on tropical areas, and in so obnoxious ways disgustingly enough together with their stuffers that all the country is out of tourism as of now. retard is few to say about you, as idiots is terminology for best match. I do not intend to talk to you but
02:39 genderfreedom: stuff a barrel under your noses if you start committing your garbage at my direction, be that another verbal or whatever favourite sexual humiliating remark based of your heroics as aids carriers. you are entirely mindill folks and need to be thrown to imprisonment.
12:00 fdobridge: <m​agic_rb.> How could i debug nouveau not turning off the fans on a desktop gtx 1060? The proprietary driver does it. Im fine with manual control. The gpu is unused and in the future its gonna be used for encoding, i can script that
12:00 fdobridge: <m​agic_rb.> Though, encoding on nouveau probably sucks :) but like for now to shut it up
12:00 fdobridge: <m​agic_rb.> (Its quite hard to sleep next to 3 rog strix fans going at 50%)
12:18 fdobridge: <!​DodoNVK (she) 🇱🇹> The fans do turn off occasionally on my laptop
12:35 fdobridge: <r​inlovesyou> Could probably become good when we get vulkan video in properly
14:36 fdobridge: <g​fxstrand> It'll never be good on a 1060. We can't re-clock or fan control those cards
14:38 fdobridge: <g​fxstrand> If it's a 1060 with nouveau, then it's running in the BIOS "make sure it doesn't burn up mode" so minimum clocks and, apparently, half fan. The reason it's doing that is because no power management or fan control firmware is loaded. That's also why it's slow. Unfortunately, nouveau cannot and will likely never be able to do power management or fan control on those cards.
14:39 fdobridge: <!​DodoNVK (she) 🇱🇹> It can run GTA 5 at 4K on Windows (so it's still relatively powerful and not e-waste; it might be even faster than my GSP-supported GTX 1650 Ti)
14:40 fdobridge: <c​langcat> Not unless Nvidia actually help. But also Rin if you want good performance on 1060 on Linux only option is NVIDIA drivers(assuming their drivers still support the care).
14:40 fdobridge: <g​fxstrand> Oh, sure. I'm sure it's totally fine on windows. It'll also be fine with the blob.
14:41 fdobridge: <g​fxstrand> Nvidia won't help. So please don't get anyone's hopes up that that's even a possibility.
14:41 fdobridge: <p​ac85> Does it work fi the blob fw is extracted manually or does the fw have some interface that needs to match with the driver's?
14:41 fdobridge: <g​fxstrand> The later
14:41 fdobridge: <p​ac85> Does it work if the blob fw is extracted manually or does the fw have some interface that needs to match with the driver's? (edited)
14:41 fdobridge: <p​ac85> I see, that's unfortunate
14:42 fdobridge: <c​langcat> Not saying they will just pretty much is the only option. But also yea I doubt NVIDIA will do anything even if they can.
14:42 fdobridge: <m​agic_rb.> I know i can set the pci power state manually, to increase the clocks. Cant i lower it somehow to get it to shut up?
14:42 fdobridge: <g​fxstrand> Probably not
14:43 fdobridge: <p​ac85> ~~Just jam something in the fans to stop them ~~
14:43 fdobridge: <g​fxstrand> It's running in safe mode
14:43 fdobridge: <p​ac85> ~~Just jam something in the fans to stop them or cut this wirrs ~~ (edited)
14:43 fdobridge: <p​ac85> ~~Just jam something in the fans to stop them or cut this wires ~~ (edited)
14:43 fdobridge: <m​agic_rb.> Ill just unplug them :)
14:43 fdobridge: <m​agic_rb.> /s
14:44 fdobridge: <c​langcat> But yea as sad as it is proprietary drivers may be the only good option for performance for that card currently.
14:44 fdobridge: <m​agic_rb.> Okay so what im getting from this, is that i could increase the power state and the clocks, but i cant make it shut up. Ill check whats it in currently, maybe d3cold could do it?
14:44 fdobridge: <m​agic_rb.> Im rn interested in being able to sleep :) performance will come later
14:45 fdobridge: <p​ac85> ~~Just jam something in the fans to stop them or cut the wires ~~ (edited)
14:48 fdobridge: <c​langcat> Though I am curious @gfxstrand would there be no way to scrape firmware from proprietary drivers and hack it to work or would that fall into breaking Nvidia licenses?
14:51 fdobridge: <m​agic_rb.> Has been discussed before
14:51 fdobridge: <g​fxstrand> Not reliably. It would also have to be totally R/E'd. People have looked into this. It's not feasible.
14:51 fdobridge: <m​agic_rb.> Defintely falls into "you get sued"
14:51 fdobridge: <c​langcat> R/E'd?
14:52 fdobridge: <g​fxstrand> Reverse engineered
14:52 fdobridge: <c​langcat> Ahhh yea fair enough I was just curious
15:13 fdobridge: <r​inlovesyou> Ah, rip
15:18 fdobridge: <b​abblebones> With the 10 series basically being EOL for this, is there an EU based RE argument for compatibility purposes?
15:27 fdobridge: <g​fxstrand> It's not really about legality so much as the fact that a lot of people have tried and no one has managed.
15:27 fdobridge: <g​fxstrand> I mean, there are legalities but even if you account for those the technical problem isn't trivial.
16:44 fdobridge: <r​edsheep> The Nvidia people here have been really useful lately. From previous discussion here the effort on their end would be huge to create a firmware that does everything needed for nouveau, and it would have to be bespoke.
16:44 fdobridge: <r​edsheep> It's easy to forget, we're talking about a huge effort to support 7-8 year old hardware
16:45 fdobridge: <S​id> 555 beta is out: https://www.nvidia.com/Download/driverResults.aspx/224751/en-us/
16:46 fdobridge: <k​arolherbst🐧🦀> it's a hard No from nvidia providing PMU firmware for older hardware, so we'd need to dump the firmware and reverse engineer the interfaces
16:50 fdobridge: <r​edsheep> 555 defaulting gsp is huge, I wonder if it worth making nouveau usable with 555 to get updated there
16:51 fdobridge: <S​id> explicit sync seems to be broken with GSP enabled on the non-openrm modules right now
16:53 fdobridge: <S​id> the most interesting logline to me is `Added support for using EGL instead of GLX as the OpenGL ICD for NvFBC.`
16:54 fdobridge: <S​id> OBS NvFBC plugin resurrection soon?
16:55 fdobridge: <r​edsheep> Did it ever go away? Haven't touched it in a while
16:56 fdobridge: <S​id> obs v28 dropped GLX support, so yes
16:56 fdobridge: <r​edsheep> Some of that changelog really reads like firmware fixes to me
16:58 fdobridge: <p​ac85> Is the fw encrypted too?
16:59 fdobridge: <m​arysaka> Yes the final blob is, not sure about the actual "low security" part of it tho
16:59 fdobridge: <p​ac85> Uhm, I wonder how one would about to start doing the RE
17:00 fdobridge: <p​ac85> Uhm, I wonder how one would about it to start doing the RE (edited)
17:01 fdobridge: <m​arysaka> Probably with a good old mmiotrace and lot of patience, but the interface is likely unstable and the blobs couldn't be shipped as part of linux-firmware anyway
17:03 fdobridge: <p​ac85> Yeah I know I'm just curious how one would start REing something like that
17:04 fdobridge: <p​ac85> So do you mean doing blackbox re when you say mmio trace?
17:05 fdobridge: <m​arysaka> Yeah blackbox
17:05 fdobridge: <m​arysaka> that also assume loading up the firmware and testing the interface, making assumption and theories and so on
17:10 fdobridge: <p​ac85> Yeah I get it
17:11 fdobridge: <p​ac85> I was more thinking along yhr lines of how you'd get to decrypt it but I guess that could potentially take longer than doing blackbox
17:11 fdobridge: <p​ac85> I was more thinking along the lines of how you'd get to decrypt it but I guess that could potentially take longer than doing blackbox (edited)
17:11 fdobridge: <p​ac85> I was more thinking along the lines of how you'd get to decrypt it but I guess that could potentially take longer than doing blackbox RE (edited)
17:17 fdobridge: <m​tijanic> What do you mean? Can you file a bug report somewhere? That's the default config that got the most QA coverage for this release.
17:19 fdobridge: <S​id> https://cdn.discordapp.com/attachments/1034184951790305330/1242527083339448350/image.png?ex=664e2905&is=664cd785&hm=35724f9ad64f36ea6e5218d39603bb7f635053d89296f095e2d56f64d95b4e49&
17:20 fdobridge: <S​id> https://forums.developer.nvidia.com/t/major-kde-plasma-desktop-frameskip-lag-issues-on-driver-555/293606
17:24 fdobridge: <m​tijanic> Oh that [MAJOR] one. I've been collecting willpower to deal with that report for a few days now, wasn't sure if it was made in good faith or not. Thanks, I'll look into it.
17:25 fdobridge: <S​id> it's 3 hours old, I think you're referring to a different report 😅
17:25 fdobridge: <S​id> oh you mean the github counterpart
17:27 fdobridge: <m​tijanic> Yeah.
17:27 fdobridge: <m​tijanic> It's got some inconsistencies in the attached log
17:28 fdobridge: <m​tijanic> We unfortunately have to deal with fake reports that only exist to waste our time and troll, so sometimes regular ones also raise red flags.
17:30 fdobridge: <S​id> valid
17:36 fdobridge: <m​agic_rb.> Completely unrelated, but there are nvidia people here, so uh, im gonna ask. But on the proprietary driver on nixos we have a fun bug where changing the mode of any output twice (turn monitor on, reposition) is enough to segfault X11. I did trace it back in the day and it seems to fault in glibc while loading the `xf86FreePixmap` symbol. But after that i got stuck. Even though im on nouveau it might still be good to fix it. Any ideas on how
17:38 fdobridge: <m​agic_rb.> I also compared the obvious: glibc versions and some of the configure flags between arch and nixos. The versioned matched, nixos had the same flags except for some extra hardenings
17:38 fdobridge: <m​agic_rb.> And i think it only happens with reverse prime
17:47 fdobridge: <m​tijanic> I feel that's partially directed at me, so replying just so it doesn't seem like I'm ignoring you.. but I have no clue. That's out of my zone of expertise, sorry.
17:49 fdobridge: <m​agic_rb.> Oh no worries i only realized youre an nvidia person after i posted the question :)
17:50 fdobridge: <m​agic_rb.> But yeah, makes two of us, ive absolutely no idea whats happening there
18:19 fdobridge: <S​id> for what its worth, things appear to be fine on kwin 6.0.4, nvidia-open 555
18:22 fdobridge: <b​abblebones> No lag?
18:22 fdobridge: <b​abblebones> How's gaming?
18:22 fdobridge: <c​langcat> Wait do you work at Nvidia?
18:23 fdobridge: <c​langcat> Oh wild.
18:23 fdobridge: <S​id> I couldn't tell you, the issue never affected me on my laptop. Just assimilating info from other friends/people I know
18:24 fdobridge: <r​edsheep> Hence me saying the Nvidia people have been helpful. Like 4-5 of the somewhat active people here are from nvidia
18:26 fdobridge: <m​agic_rb.> I humbly thank the nvidia people for helping the nouveau people so that i can run a opensource driver 🙏🙏
18:27 fdobridge: <m​agic_rb.> And i also humbly thank the nouveau people for the exact same reason
18:27 fdobridge: <S​id> going through the member list I see 4 members that I immediately recognize as NV employees, yeah
18:37 fdobridge: <c​langcat> Never said Nvidia haven't ever been helpful. Just that I doubt they would help with making firmware for an old card. :P
18:39 fdobridge: <p​avlo_kozlenko> And now he SPIR-V used for nouveau?
18:39 fdobridge: <p​avlo_kozlenko> https://cdn.discordapp.com/attachments/1034184951790305330/1242547388854436024/8cee81d388d80b0b.png?ex=664e3bee&is=664cea6e&hm=96f04391b84d568acd6726f4a68d95a8795ebacb1704ae951274db87f32d955f&
18:39 fdobridge: <p​avlo_kozlenko> Now he SPIR-V used for nouveau? (edited)
18:39 fdobridge: <p​avlo_kozlenko> https://cdn.discordapp.com/attachments/1034184951790305330/1242547388854436024/8cee81d388d80b0b.png?ex=664e3bee&is=664cea6e&hm=96f04391b84d568acd6726f4a68d95a8795ebacb1704ae951274db87f32d955f&
18:40 fdobridge: <p​avlo_kozlenko> Please explain what SPIR-V is used for nuvo
18:41 fdobridge: <p​avlo_kozlenko> Please explain what SPIR-V is used for nuvo. I am very far from these things (edited)
18:48 fdobridge: <!​DodoNVK (she) 🇱🇹> nouveau only consumes TGSI or NIR depending on the compiler
18:52 fdobridge: <p​avlo_kozlenko> Please explain where SPIR-V is used for nuvo. I am very far from these things (edited)
18:53 fdobridge: <a​irlied> It's been clearly stated and repeated ad nauseum that they won't. Can we pin a topic or something so it stops wasting time? If someone with goes off into a corner and makes it happen yay, but they need to do it all in some sort of supreme hyperfocus event and not ask any questions or talk about it until finished
18:54 fdobridge: <a​irlied> All Vulkan and GL drivers consume SPIRV just we have mostly generic code to translate it to NIR
19:10 fdobridge: <!​DodoNVK (she) 🇱🇹> So NAK would look different without the work of vtn/ntv, right?
19:11 fdobridge: <k​arolherbst🐧🦀> it goes a bit further than this. I'm sure if every driver would have written their own spirv to whatever thing, we wouldn't have advanced so quickly on everything
19:24 fdobridge: <a​irlied> Ah I remember the old SPIRV to codegen ideas :-p
19:27 fdobridge: <m​agic_rb.> note to self: putting a 1060 in d3hot still wont turn off the damn fans >:(
19:29 fdobridge: <m​agic_rb.> ill just suspend the machine when not in use, resolves the fan spin problem 👍 (ill get around to installing proprietary drivers one of these days)
20:28 fdobridge: <S​id> further reports suggest GSP is problematic only on some setups, not all
22:56 fdobridge: <g​fxstrand> Yeah, that idea was never going to pan out. 😂
22:59 fdobridge: <g​fxstrand> That's a really old slide...
22:59 fdobridge: <g​fxstrand>
22:59 fdobridge: <g​fxstrand> You shouldn't think of SPIR-V as an IR or as something directly used by driver at all. It's basically a binary form of GLSL. We consume it but as a high-level language. spirv_to_nir is more like a parser/AST than a compiler.