02:18 fdobridge: <j​ekstrand> NVK doesn't really do pipeline caching
05:20 anholt: skeggsb: so, I made the mistake of letting my desktop display content from nvk while also running cts tests, and it seems that reset isn't canceling fences as it ought to? https://paste.centos.org/view/66036171
05:21 anholt: jekstrand: presumably not really pipeline caching should at least not corrupt a pipeline cache, though :)
10:22 karolherbst: anholt: are you on upstream or one of skeggsbs branches?
10:22 karolherbst: you should use this tree https://gitlab.freedesktop.org/skeggsb/nouveau/-/tree/00.06-gr-ampere
15:59 fdobridge: <j​ekstrand> anholt: Yeah, there may be a memory corruption in there, I guess. I've never observed one but bugs do happen and we've got a bunch of flakes yet. Do you have a specific test that does it or is it just a general phenomena you've observed?
16:28 anholt: jekstrand: was really consistent doing just deqp-runner -t points iirc.
16:43 fdobridge: <j​ekstrand> Ok, cool. That may be enough to go on.