16:28karolherbst[d]: mhenning[d]: you'll enjoy hearing that me digging into nvdisasm validation made me found two bugs already.. one existing one.. they moved the constant offset of OpAtom in Blackwell...
16:29karolherbst[d]: okay..
16:29karolherbst[d]: actually it's worse
16:29karolherbst[d]: they haven't moved it...
16:29karolherbst[d]: the constant offset is 23 bits in size 🙃
16:29karolherbst[d]: not 24
16:29karolherbst[d]: but only for ATOMG, not ATOMS
16:33gfxstrand[d]: Ugh... Working with Mark to try and set up an NVK demo and Chromium+NVK+Zink is really not working like it should. 😭
16:35karolherbst[d]: 😭
16:36karolherbst[d]: I'm innocent
16:36gfxstrand[d]: Some of the issues may come down to prime being sketchy? It is running on an Intel iGPU
16:37gfxstrand[d]: But NVK+Iris works great! NVK+Zink/NVK isn't stable.
16:37gfxstrand[d]: Disabling compression helped some (I was seeing corruptions before that) but even with that disabled, it's not where it needs to be.
16:38gfxstrand[d]: No idea if it's regressed or if I just haven't banged on Zink long enough.
17:16mhenning[d]: mango has been complaining about chromium issues for a bit. I've never been able to reproduce those.
17:17mhenning[d]: It's possible it's always been a bit broken
17:18mhenning[d]: marysaka[d]: Kernel zcull stuff landed if you want to do another round of review on the userspace side: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33861
17:19marysaka[d]: oh nice, will take a look at it tomorrow 👍
18:34mangodev[d]: mhenning[d]: weirdly enough, it's hit or miss
18:35mangodev[d]: electron has actually been behaving quite well since the compression patch for me
18:35mangodev[d]: and chrome itself seems to work fine more or less
18:36mangodev[d]: but firefox still breaks down with backdrop filter blur
18:36mangodev[d]: that hasn't changed
18:36mangodev[d]: easy to replicate, hard to find out why
18:38mangodev[d]: i'm hoping it's just that my system configuration is screwed and that it can be fixed with a reinstall, since atomic present is also broken on my system (even with the flag enabled)
18:55chikuwad[d]: could try Fallen Order in a couple hours
18:56gfxstrand[d]: We have one report for it but it's from like 2 years ago
18:56chikuwad[d]: yeah, I'll give it a spin in ~2h
20:54airlied[d]: hehe tried vulkan rendering in chrome, lots of white videos, no idea if it's an nvk problem
21:01airlied[d]: so at some point nvk+zink+chrome used to generate a CT WIDTH violation, any ideas what fixed it, since it doesn't with main
21:12mhenning[d]: airlied[d]: Yeah, that was a zink bug https://gitlab.freedesktop.org/mesa/mesa/-/issues/14647
21:20mangodev[d]: airlied[d]: i was wondering why my discord stopped crashing one day
21:20airlied[d]: firefox still flickers in theatre mode youtube, not a clue on how to figure it out
21:20mangodev[d]: and YouTube embeds work properly
21:21mangodev[d]: airlied[d]: something i have even less of an idea about is YouTube channel pages
21:21mangodev[d]: tons of graphical issues on Firefox, but I have zero clue what it *could* be
21:21mangodev[d]: there's no blur on channel pages as far as i'm aware
21:21mangodev[d]: yet there's egregious flickering and even screen tearing?
21:22airlied[d]: playing a video in youtube theatre mode I get a black square randomly instead of video content appearing once every bunch of frames
21:22airlied[d]: but it's like one 64x64 tile or something
21:22mangodev[d]: yeah, i've noticed that too
21:23mangodev[d]: ambient lighting mode doesn't seem to make the issue any better or worse
21:23mangodev[d]: which is weird
21:23mangodev[d]: because it's one big blur effect
21:23mangodev[d]: it's strange
21:23mangodev[d]: i really do wonder what is making webrender break down though
21:24mangodev[d]: the background blur issue makes sites that use it just break apart, not even just the blurred parts
21:24mangodev[d]: seemingly random parts of the page just die at random
22:16x512[m]: marysaka[d]: Any updates for NVK NVRM backend?
22:27Mary: x512[m]: I could push my WIP branch tomorrow if you want but it still need support for sync_fd and sparse
22:32x512[m]: Yes, it would be helpful. Are there big problem with implementing sparse memory support? I made some notes before which NVRM APIs should be used to implement it.
22:34Mary: I just didn't start looking at sparse yet because sync fd is required if you want zink to work or any decent WSI.
22:37x512[m]: I have my own WSI loadable add-on and I do not use dma-buf/syncobj, I am on Haiku.
22:38x512[m]: So sparse memory is expected to be synchronized with in-memory semaphores and NVRM events.