07:26 dollythelamb: it reminds me the Cambodia where similar Quasimodo aids fuckers were left alone by everyone who left their tyranny where’s that you continue your absurd here until you receive the death penalty, I am not worried about this red hat loses in court after these sanctions are lifted to your illborn people. I am fixing your bootrom exploits every now and then, the more you do that the bigger
07:26 dollythelamb: your relatives left behind will be tortured.
13:30 karolherbst[d]: anyway.. I bought X4, so maybe next week or so I'll be able to see if I can improve perf there a bit
13:42 asdqueerfromeu[d]: Lyude: I'm surprised how much they're resisting though ✊
14:27 phomes_[d]: wrt X4 and compression. When I launch the game and load into the scene that I have benchmarked it calls nvk_CreateImage 13 times. Of these 2 can use compression
14:34 phomes_[d]: enabling compression for all 13 did not change performance or cause any issues
14:36 mohamexiety[d]: yeah there are a few cases where we can enable comp but dont do it as it wouldnt be beneficial
14:42 phomes_[d]: yes. I was just experimenting a bit because it was strange that the game did not improve one bit when compression support landed
14:50 mohamexiety[d]: yeah some of those games are weird
14:51 mohamexiety[d]: i'd guess something else is bottlenecking us before compression becomes relevant but i dont know
14:52 karolherbst[d]: there are tons of desc sets updates
15:03 karolherbst[d]: but also many rather big copies, so I'm wondering if those two things matter here
16:22 gfxstrand[d]: New Fedora version. New Rust build errors.
16:22 gfxstrand[d]: karolherbst[d]: Do you happen to already have an MR?
16:22 karolherbst[d]: gfxstrand[d]: I use rustup
16:22 gfxstrand[d]: These look like they're coming from bindgen
16:23 karolherbst[d]: what's the error?
16:23 karolherbst[d]: bindgen had a bug with llvm-22
16:23 gfxstrand[d]: Lots of unknown fields and stuff
16:24 karolherbst[d]: ahh yeah, sounds like that bug
16:24 karolherbst[d]: https://github.com/rust-lang/rust-bindgen/issues/3264
16:24 karolherbst[d]: update bindgen
16:24 karolherbst[d]: 0.72 is affected, fixed in 0.72.1
16:25 gfxstrand[d]: Got it. Thanks!
18:44 _lyude[d]: Does anyone remember why all of the interfaces in drivers/gpu/drm/nouveau/include/nvif/if****.d are numbered like this? Like, is this something that we're copying from nvidia or are they just numbers we came up with on our own?
18:46 _lyude[d]: writing up a patch hooking up the interfaces atm for the WRITE_AWAKEN stuff that got mentioned here the other day
18:50 mhenning[d]: I think we came up with them on our own, although I really don't understand why the code is structured the way it it.
18:50 _lyude[d]: alright - I guess I can just make a new if file then for adding something for getting awaken events from display channels
18:51 _lyude[d]: though I'd love to know if there was a reason this never got hooked up in the past, it seems like a big enough omission to be intentional
19:17 airlied[d]: NVIF is all nouveau, it was Ben's attempt at making an abstraction layer like NVIDIA ctrls that would be usable from userspace drivers like NVIDIA ctrls.
19:18 airlied[d]: In the end he realised it probably wasnt what we needed, and there was a patch series to rip it out, but he left before finishing it
19:28 _lyude[d]: wonder if maybe we should try to finish it at some point
19:29 airlied[d]: With nova I just didn't feel like investing that much effort into nouveau
19:29 _lyude[d]: oh gitlab.freedesktop.org/mesa/mesa/-/work_items/15248 marysaka[d] i assume you still need someone to look at this at some point?
19:29 airlied[d]: It has a high chance for regressions
19:29 airlied[d]: Fermi is already broken enough :-p
19:30 _lyude[d]: that is true, fwiw: if you're curious about the thing I'm working on: https://github.com/Dirbaio/linux/commit/0f1d715f1ef8cbe8c2b96b6658cb422daaad7176 someone sent a claude diff for hooking up AWAKEN stuff, so i've been using it as reference for just writing a proper implementation of this. it seems on some platforms we're getting really long busy-wait times for some display dma channels
19:30 dirbaio: o/
19:30 _lyude[d]: since apparently the busy wait is long enough to actually start eating cpu
19:31 x512[m]: > and there was a patch series to rip it out, but he left before finishing it
19:32 marysaka[d]: _lyude[d]: ah yes! it seems related to atomic modeset not being enabled and vulkaninfo now querying certain infos that it wasn't before
19:32 x512[m]: Is it even possible considering Linux kernel policy to not break UAPI?
19:32 _lyude[d]: marysaka[d]: ah lol, gotcha
19:32 marysaka[d]: the vulkan common side have been enforcing atomic modeset for 3 years btw
19:32 _lyude[d]: re uapi: no, it's not UAPI - it's internal kernel module APIs
19:32 marysaka[d]: we never saw that it seems
19:33 _lyude[d]: yeah - I mean it makes sense. we should have turned it on with nouveau ages ago, and it's one of the reasons I've been testing it on my desktop quite a lot
19:33 marysaka[d]: ... I forgot to update the issue isn't it, I should probably type that down in a more proper way
19:33 _lyude[d]: honestly once i figure out wtf is going on with the screen flahsing issue I'll probably feel comfortable turning it on
19:34 _lyude[d]: *flashing
19:35 marysaka[d]: that would be great, display aren't my things tbh 😅
19:36 x512[m]: But Mesa driver is calling NVIF interfaces with ioctl: https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/nouveau/winsys/nouveau_device.c?ref_type=heads#L355
19:36 _lyude[d]: the thing is we don't actually have to change the kernel ABI to remove nvif/nvkm
19:36 _lyude[d]: we can still keep all the same ioctls
19:37 _lyude[d]: my assumption is we just named them NVIF ioctls because the original intent was to have the ability to have a separate kernel driver with nouveau that represented the host GPU
19:38 marysaka[d]: The NVIF userspace side of things was dropped a while ago, what is actually exposed to userspace is an hardcoded subset those days I think
19:39 _lyude[d]: oh right - there was probably a more involved NVIF interface that ben used for ben things at some point, wasn't there
19:39 marysaka[d]: yeah it even had TSG and ZBC support but those were never wired anywhere :maxpoeSweat:
19:40 airlied[d]: The patch series stubbed in all the NVIF uapi bits as hard coded
19:40 airlied[d]: Since actual users of the uapi were only a few calls
19:46 airlied[d]: https://gitlab.freedesktop.org/bskeggs/nouveau/-/tree/03.02-remove-ioctl?ref_type=heads
22:59 _lyude[d]: btw dirbaio - I -think- I'm pretty close to having a patch written up to turn on the awaken notifiers for all of the DMA channels (we might as well). Just figuring out how to get the right interrupts enabled on volta+ now, but hopefully should have a patch soon
23:12 dirbaio: <3
23:12 dirbaio: Awesome :D Let me know if you want me to test anything.