04:14andres: Do you prefer bugs to the list or gitlab? Turned on lockdep on my laptop to debug something unrelated and got hit by reports in nouveau
04:24airlied: andres: yeah we have a few, but file a gitlab issue, I should actually try and fix some of them
06:49andres: airlied: done, as https://gitlab.freedesktop.org/drm/nouveau/-/issues/268
06:51airlied: andres: if you want to avoid that one you can probably do nouveau.runpm=0
06:52airlied: TimurTabi: I got a cts run to complete today, setting the ALL_PDB flag on the tlb flush seems to fix it, but not sure that is correct, but a good hint as to where the problem might be
10:01Sid127: I'm looking to start contributing to NVK, I was behind the no-op MR, and wanted to know if there's anything simple I can pick up to do right away since I'm still quite new to driver development
10:43fdobridge: <gfxstrand> Hi Sid127!
10:43fdobridge: <gfxstrand> Let me look
11:05Sid127: I thought I could take up VK_EXT_color_write_enable, but if there's anything more important that's languishing I'll gladly divert my efforts there
12:30fdobridge: <gfxstrand> Oh, that's probably a good first extension
13:46fdobridge: <conan_kudo> Where do nouveau patches get staged?
13:46fdobridge: <karolherbst🐧🦀> good question. We kinda want to allow gitlab MRs, but I also kinda want to have gitlab MRs on drm-misc 😄
13:46fdobridge: <conan_kudo> I'm trying to figure out if any of the nouveau GSP patches have landed yet
13:46fdobridge: <karolherbst🐧🦀> you should always check with drm-misc on that
13:46fdobridge: <conan_kudo> and this is... hard given the distinct lack of feedback on nouveau ML
13:47fdobridge: <karolherbst🐧🦀> yeah... I want to do everything in gitlab anyway, maybe we should just do it, it's just uhhh
13:48fdobridge: <conan_kudo> can we please 🙂
13:48fdobridge: <karolherbst🐧🦀> ... annoying, because you kinda have a next and fixes targetr
13:48fdobridge: <karolherbst🐧🦀> ... annoying, because you kinda have a next and fixes target (edited)
13:48fdobridge: <conan_kudo> these patch sets are already hard to track
15:11fdobridge: <tiredchiku> I have arrived 🐸
15:50benjaminl: gfxstrand: is NAK in a state where working on maxwell support would be helpful?
15:55fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> benjaminl: Every shader stage except RT stuff pass CTS (on Turing at least)
16:45fdobridge: <gfxstrand> Yeah, probably. I'm hoping to get NAK merged in the next couple weeks so I'll be focusing on that. @marysaka already has a Maxwell branch started.
16:46fdobridge: <gfxstrand> That was to benjaminl (not on discord)
16:54benjaminl: technically I have a discord account, but discord likes to OOM my laptop :(
16:54fdobridge: <karolherbst🐧🦀> pain
16:54fdobridge: <karolherbst🐧🦀> use it from within your webbrowser instead
16:54benjaminl: this is in firefox
16:54fdobridge: <karolherbst🐧🦀> pain
16:55benjaminl: it's slightly better these days now that I run firefox inside a memory-limited cgroup
16:55benjaminl: so the OOM-killer usually only kills the discord/element/whatever tab instead of freezing the whole machine
16:55fdobridge: <karolherbst🐧🦀> but yeah.. firefox is very very greedy in terms of RAM usage 🙃
17:00benjaminl: marysaka: https://gitlab.freedesktop.org/marysaka/mesa/-/tree/nak/sm50-rebased this the most recent sm50 branch?
17:01benjaminl: I'd been down to work on the remaining instructions and such, but don't want to step on your toes if you're already working on sm50 stuff?
17:07fdobridge: <gfxstrand> I think @dwlsalmeida is planning to spend a bit of time on sm50 stuff as well
17:07fdobridge: <gfxstrand> If that's going to be a collaborative branch, should we move it to nouveau/mesa?
17:34fdobridge: <marysaka> might be worth moving to nouveau/mesa yeah
17:36fdobridge: <marysaka> the encoding for ALU op really need some rewriting to make it less generic
17:37fdobridge: <karolherbst🐧🦀> yeah.. encoding pre volta is silly in many places 🙃
17:39fdobridge: <marysaka> we also need a variant of nak legaliser for it as quite a lot of instructions don't have the same modifier as Volta+
17:40fdobridge: <marysaka> (not to mention having only 24 bits for immediate...)
17:41fdobridge: <karolherbst🐧🦀> or are cursed, like the 32 bit immediate ffma
17:42fdobridge: <karolherbst🐧🦀> who doesn't like post RA optimizations
17:43fdobridge: <marysaka> well you also loose modifiers for sources if I remember correctly
17:43fdobridge: <karolherbst🐧🦀> you do? why tho
17:44fdobridge: <karolherbst🐧🦀> you don't have a rounding mode
17:44fdobridge: <karolherbst🐧🦀> but whatever
17:44fdobridge: <karolherbst🐧🦀> long immd ffma is just pure 🔥
17:45fdobridge: <karolherbst🐧🦀> usually most would be like "whatever", but that one actually matters for performance 🙃
17:45fdobridge: <karolherbst🐧🦀> though constant pull also works