00:18 phomes_[d]: gfxstrand[d]: thanks. There were two things that NV sets different than NAK in the SPH. Setting it exactly the same did way as NV did not help
00:23 rinlovesyou[d]: redsheep[d]: I use native
00:39 gfxstrand[d]: karolherbst[d]: Yeah, it's really annoying. I screwed up Windows once when I was working on the Intel drivers.
01:23 airlied[d]: Flatpak won't have nvk in it or might have old nvk
08:51 Sid127: airlied[d]: what's the status on that SG_DEBUG issue?
08:52 airlied[d]: The fix I sent works but isn't really a proper fix, think we need to rewrite a few sg using interfaces
08:53 Sid127: could I get a link to that fix?
08:54 asdqueerfromeu[d]: airlied[d]: (it won't have it until 24.08)
08:54 Sid127: additionally, someone I know came across a null pointer dereference on 6.9 (void linux)
08:55 tiredchiku[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1246748887574904956/dmesglog?ex=665d84e2&is=665c3362&hm=878a200068d6d1a2e2309f29b6ced43f3dc0e85961f4066809fbbf40604bc2bb&
08:55 Sid127: the only difference from void's vanilla kernel is that she had the SG_DEBUG patch applied, the one that got reverted
08:55 Sid127: because without it her machine was unusable
08:57 airlied[d]: https://patchwork.freedesktop.org/patch/593723/
08:58 Sid127: another question for you: do we use different code paths for different devices, or do we try to find a code path that works for all of them (in the kernel, I mean)
08:59 magic_rb[d]: redsheep[d]: I might try zink then, since im on a laptop and plasma ends up being rendered using intel anyway. Ill set it up maybe today and let you know 👌
09:19 karolherbst[d]: looks like the bridge is working pretty well so far (it already tracks 17 users)
09:20 redsheep[d]: magic_rb[d]: I bet on a laptop with NOUVEAU_USE_ZINK=1 it will work great, if you also run a kernel that can do NVK modifiers like gfxstrand/Linux on the NVK branch in gitlab
09:21 magic_rb[d]: redsheep[d]: Yep i already am doing that, due to gamescope
09:21 magic_rb[d]: Cool! Ill play around with wayland, first time ever for me. Ive never ran any machine without nvidia proprietary so wayland has been difficult to play with
10:34 airlied[d]: Sid127: If a path works for multiple devices we try and use it
11:03 Sid127: gotcha, thanks
13:19 pac85[d]: I have a question regarding compression on nv hw. Is nv hw capable of applying compression to buffers or do compressed image have other requirements besides the right flags in the pte?
13:49 f_: ooo new bridge?
13:50 f_: >and also if oftc gets upset at some point for having too many IPv6 connections out of the same prefix
13:51 f_: karolherbst: https://oftc.net/LimitExceptions/
15:34 magic_rb[d]: For me it complains about morr than 5 connection from one IP
15:34 magic_rb[d]: Idk if they have something similar for ipv6
17:31 airlied[d]: pac85[d]: Only images
17:48 pac85[d]: I see thx!
19:34 airlied[d]: pac85[d]: by asking the question you probably don't understand how compression is used in this context
21:46 gfxstrand[d]: marysaka[d]: https://mastodon.gamedev.place/@danil/112549152403747930
21:52 pac85[d]: airlied[d]: Mmm, the reason I asked is that amd seems to be moving dcc bits to pte and apparently they can apply it to buffers too, though I'd still like to hear what my misconception is regarding nv hw
22:04 airlied[d]: no just that compression on buffers doesn't really make any sense, since it's usually not color data etc
22:05 airlied[d]: the compression mechanism usually relies on tiling and block shapes
22:33 gfxstrand[d]: Depends on the buffer, I'd guess. It could be that they think they can compress BVHs or something.
22:35 redsheep[d]: Wouldn't those be pretty hard to compress?
23:43 airlied[d]: gfxstrand[d]: but it really needs the block shapes etc, so you'd have to have some additional metadata about the buffer to tell the hw
23:55 gfxstrand[d]: Nah. Most compression schemes work in terms of compressing several cache lines to one. Tiling just gives you nice locality. It has nothing to do with the actual compression algorithm.
23:56 gfxstrand[d]: Sometimes the format is used to key which compression to use, though. Makes hash of texture views but that's the cost you pay.