00:07 fdobridge_: <g​fxstrand> sweet!
00:07 fdobridge_: <g​fxstrand> Okay, docked works way better.
00:08 fdobridge_: <b​ylaws> Oh great!
00:09 fdobridge_: <b​ylaws> Are you looking into nvk on it or just for testing?
00:09 fdobridge_: <g​fxstrand> Gotta love the little glitches while docked. I'm going to guess that's X11 blitting to 4K and the GPU just not keeping up.
00:09 fdobridge_: <p​ac85> Nvk on switch? That's exciting
00:09 fdobridge_: <g​fxstrand> Planning to play with NVK, yeah.
00:09 fdobridge_: <g​fxstrand> Right now I'm just trying to get a feel for Linux+Switch at all.
00:09 fdobridge_: <p​ac85> Is it one of those with the vulnerable boot ROM?
00:10 fdobridge_: <g​fxstrand> Yeah. I bought this one about 3 months after launch so it's got all the vulnerabilities. 👿
00:10 fdobridge_: <g​fxstrand> Yeah. I bought this one about 3 months after launch so it's got all the vulnerabilities. 😈 (edited)
00:11 fdobridge_: <p​ac85> Nice lol, I lost that train
00:11 fdobridge_: <p​ac85> Nice lol, I lost that train (don't have a switch at all) (edited)
00:12 HdkR: Watch out for battery expansion. Both of my launch day Switches had to get their batteries replaced over the past weeks
00:13 fdobridge_: <g​fxstrand> Noted
00:13 fdobridge_: <b​ylaws> Neat, I remember trying mainline for a bit years back then gave up and ported l4t
00:14 fdobridge_: <b​ylaws> Will be cool to have that actually be usable
00:14 fdobridge_: <b​ylaws> Will be cool to have that actually be usable now (edited)
00:14 fdobridge_: <b​ylaws> Will be cool to have that actually be usable in the future with nvk (edited)
00:14 fdobridge_: <g​fxstrand> Yeah, in theory it sorta works but there's sync problems.
00:14 fdobridge_: <g​fxstrand> I want to poke about at it a bit myself.
00:19 fdobridge_: <g​fxstrand> I would play with a different Tegra board but all I have is a Switch
00:20 fdobridge_: <a​irlied> Kernel patches accepted :-p unless it's obviously broken, I just got rid of the two tegras I had
00:22 fdobridge_: <a​irlied> Don't think my switch is old enough or my kids would let me touch it 🙂
00:25 linkmauve: The last time I tried to upstream some Switch patches in mainline, the maintainers told me the Nvidia drivers were of very poor quality and I haven’t had the time to rewrite them yet.
00:26 linkmauve: And I’m back to playing with other parts of the stack currently.
00:28 linkmauve: Plus there were issues with copyright, so I kind of dropped that.
02:34 fdobridge_: <g​fxstrand> Well, the plan is to use stock nouveau, not whatever thing NV open-sourced.
05:58 fdobridge_: <S​id> @airlied we've been trying to reach you regarding ~~your car's extended warranty~~ the job timeouts on nouveau+GSP
06:03 Ellenor: I managed to get useful information just by cutting up some malformed CSV from the weather service here
06:05 fdobridge_: <a​irlied> Yeah I might have time next week to dig in, I need to figure out if my patch to fix the locking could cause it
06:06 fdobridge_: <S​id> the one that just went in/I tested a couple weeks ago? nope, not it
06:06 fdobridge_: <S​id> the job timeouts occur even on mainline, unpatched
06:07 fdobridge_: <a​irlied> Don't think it went in yet,but was going to push it, but if timeouts occur always then that is good to know,
06:09 fdobridge_: <S​id> timeouts are consistent, yeah. they happen without the whole patchset, with the whole patchset, with every commit on the patchset except the 10th patch, and also with *just* the 10th patch
06:09 fdobridge_: <S​id> on my setup I also saw a gsp mmu fault which @/RedSheep didn't see
06:14 fdobridge_: <a​irlied> Is there an issue filed with reproducer info?
06:17 fdobridge_: <r​edsheep> I don't believe either of us have filed anything, and that is correct, I have not seen the mmu fault at all. I only ever saw device lost once as well so something different about my setup makes it far far more rare, and I have no way to know that's even related to the kernel.
06:19 fdobridge_: <r​edsheep> I am not even sure where to file one, would that be here? https://gitlab.freedesktop.org/drm/nouveau/-/issues
06:21 fdobridge_: <S​id> no issue filed but I do have everything for it ready, including an apitrace that consistently timesout on nouveau+gsp but not on nouveau-gsp or on the proprietary drivers
06:21 fdobridge_: <S​id> I'll file it in a while, post lunch
06:32 fdobridge_: <a​irlied> Yeah file it there
06:38 fdobridge_: <S​id> actually...
06:38 fdobridge_: <S​id> this is relevant. the thread itself is a mess but it is the relevant issue
06:38 fdobridge_: <S​id> https://gitlab.freedesktop.org/mesa/mesa/-/issues/10225
06:39 fdobridge_: <S​id> I'll attach my files there, maybe move it to drm/nouveau? 😅
06:53 fdobridge_: <r​edsheep> Oh yeah I forgot there was an issue I had been replying to there, I never connected that they were describing almost the exact results you have been.
07:47 fdobridge_: <a​zkali> I put a lot of effort into resuming your work on mainline
07:47 fdobridge_: <a​zkali> https://gitlab.azka.li/l4t-community/kernel/mainline/linux
07:47 fdobridge_: <a​zkali> It's pretty far in completion, basically only audio stack is completely missing, other than that some small bugs exists
07:48 fdobridge_: <a​zkali> I put a lot of effort into resuming your work on mainline
07:48 fdobridge_: <a​zkali> https://gitlab.azka.li/l4t-community/kernel/mainline/linux
07:48 fdobridge_: <a​zkali> It's pretty far in completion, basically only audio stack is completely missing, other than that some small bugs exists
07:48 fdobridge_: <a​zkali>
07:48 fdobridge_: <a​zkali> And I am currently adding Mariko support too (edited)
08:28 fdobridge_: <a​irlied> Is that bug got the same timeout? Do I need working dxvk or do we have a reproducer without it?
08:48 HdkR: Yay Mariko support
12:44 shulai: Hi~Does anyone know what is GPFIFO channel? Is it different from the compute channel? Which channel does CUDA kernel commands push to?
12:49 fdobridge_: <S​id> appears to be the same timeout, yes. you don't *need* dxvk but too few games run vulkan natively an the couple I have installed have it happen at random
13:46 shulai:
16:00 fdobridge_: <z​mike.> is anyone crazy enough to look into this for me or do I gotta play gpu shuffleup on my first day back https://gitlab.freedesktop.org/mesa/mesa/-/issues/10340
16:02 fdobridge_: <g​fxstrand> My only instance of the proprietary driver is a Haswell box I keep tucked into the corner that isn't even plugged into a screen.
16:16 fdobridge_: <z​mike.> :grimace:
16:32 fdobridge_: <g​fxstrand> Let's Goooooo! 🏎️
16:32 fdobridge_: <g​fxstrand> https://cdn.discordapp.com/attachments/1034184951790305330/1191781087958007869/image.png?ex=65a6afa1&is=65943aa1&hm=8f600aac8c237475a2e8efe22262aee7d63cf2389965d4055ebbb65517357638&
16:53 fdobridge_: <t​riang3l> But dynamic graphics state setting is not an operation on shaders 🙃
17:01 fdobridge_: <p​ac85> Uuuhm is this doing patching of shaders?
17:01 fdobridge_: <p​ac85> Uuuhm is this doing patching of shaders? (edited)
17:05 fdobridge_: <g​fxstrand> Yup. For ESO you get NULL. But I'm also planning to build a common pipeline impl on top of this and that'll provide state.
17:14 fdobridge_: <!​DodoNVK (she) 🇱🇹> Barbie background 🤔
17:40 fdobridge_: <g​fxstrand> Yup! 💖
17:42 fdobridge_: <S​id> best movie I watched in 2023
18:35 fdobridge_: <g​fxstrand> Such a good movie. I haven't laughed and cried that hard doing a movie in, like, ever.
18:36 fdobridge_: <r​hed0x> I thought it was mediocre at best
18:36 fdobridge_: <r​hed0x> really didn't work for me
18:38 fdobridge_: <r​hed0x> especially the second half. The whole feminist topic is a good fit for a Barbie movie, I just wish they had any subtlety at all in how they get that message across
18:38 fdobridge_: <r​hed0x> I loved Oppenheimer though
18:42 fdobridge_: <g​fxstrand> And that's exactly why they didn't go subtle. The patriarchy isn't subtle. It's a big noisy hammer that beats people around. Meanwhile, a huge part of feminine oppression is the expectation that every time we speak up, every time we oppose a man, we have to be very subtle and hidden about it so as to not hurt male egos. So, yeah, it was a role reversal. A lack of subtlety was the point.
18:43 fdobridge_: <r​hed0x> fair enough, just didn't make for an enjoyable movie for me
18:43 fdobridge_: <r​hed0x> but I wasn't the target audience anyway and lots of people loved it, so whatever
18:44 fdobridge_: <g​fxstrand> That's okay. Not everyone likes every movie.
18:46 fdobridge_: <r​hed0x> (Just for the record, I'm not one of those "this movie hates men" idiots)
18:48 fdobridge_: <t​riang3l> Unnormalized coordinate samplers when it's toggled in sample instructions :blobcatnotlikethis: :blobcatnotlikethis: :blobcatnotlikethis: (not on Nvidia probably, but Nvidia has 3D texture sampling with explicit gradients or something like that which is also fun)
18:51 fdobridge_: <p​rop_energy_ball> Agree, I think the lack of subtlety kinda defeats the purpose. Things are much more impactful if you let people think for themselves and come to their own conclusion as to the message that was trying to be put across.
21:08 fdobridge_: <t​riang3l> Will that include a magic function that runs all NIR loweings needed to get a shader you'd typically work with in a Vulkan driver? 🥵 😜
21:08 fdobridge_: <t​riang3l> Will that include a magic function that runs all NIR lowerings needed to get a shader you'd typically work with in a Vulkan driver? 🥵 😜 (edited)
21:09 fdobridge_: <t​riang3l> I'm at this stage when I'm looking for what specifies the `base` for vertex input loads because they are all the same 0 now…
21:10 fdobridge_: <t​riang3l> and where that's invoked in `compiler`, `mesa` or some `amd/common` or `amd/vulkan` :blobcatnotlikethis:
21:10 fdobridge_: <g​fxstrand> Not quite but I'm planning to try and standardize things somewhat.
21:11 fdobridge_: <t​riang3l> and I also have lots of comments saying where I steal from like `/* st_link_glsl_to_nir, but without linkage. */`, `/* preprocess_shader. */`, `/* st_link_glsl_to_nir. */`
21:37 fdobridge_: <t​riang3l> oooh it's `st_nir_assign_vs_in_locations` 👀
21:37 airlied: for vulkan a lot pf
21:37 airlied: of drivers have a loop assigning them
21:43 airlied: not near pc, and code has probably moved since i cant find it
21:49 fdobridge_: <t​riang3l> Yeah, I see `brw_nir_lower_vs_inputs`, similar code in RADV, Lavapipe
21:51 fdobridge_: <t​riang3l> 64-bit inputs are scary in Vulkan btw, but they are something that you really need to care about just once probably so I'll spend a few minutes/hours adding them some time later
21:53 fdobridge_: <t​riang3l> if they're even needed really
22:37 fdobridge_: <g​fxstrand> They're arguably less scary in Vulkan than GL