01:36 cwabbott: mareko: it seems there's a corner case nir_opt_varyings doesn't handle and drivers are hacking around it with mesh shaders
01:37 cwabbott: some builtin varyings like PrimitiveID, ViewportIndex etc. are never declared as PerPrimitive in the FS (glslang will throw it away if you try) but nir_opt_varyings is asserting that they are matching
01:39 cwabbott: I tried to hack something in and avoided the assert but it fails, whereas the fast-linked version of the test that doesn't call nir_opt_varyings passes
01:42 cwabbott: I think turnip is the only shader to store NIR instead of SPIR-V with LTO, and it's kinda screwing us over here, since we can't do the same workaround other drivers do
01:43 cwabbott: by the time we know that mesh & FS are being linked together we've lowered IO, and overriding the inputs to per-primitive is way more annoying with lowered IO
01:45 cwabbott: seems like nir_opt_varyings should just handle this...
02:24 mareko: cwabbott: I see, yes, I agree
02:29 mareko: there could be a little intrinsic pass before the first init_linkage that would fix up the FS input loads
13:46 jani: mripard: thanks for the MAINTAINERS update series. it made me wonder about the project member management at gitlab. we have lots of folks with unnecessary and/or stale permissions.
13:47 mripard: yeah I wondered about that as well
13:47 jani: we have 360+ people who can push to drm/tip for example
13:47 mripard: I'm not entirely sure it's a bit deal. We've been pretty laxed with the committers requirement so far anyway, so anybody who is a committer would still be elligible in like 5 years
13:47 mripard: even if they weren't active
13:49 jani: it's so easy to add folks back again, feels like there should be no qualms about "demoting" to Reporter
13:49 jani: *shrug*
13:49 mripard: but yeah, maybe we can start to prune people from that list if they haven't committed something in the last, say, 3 years or some other abritary period
13:50 mripard: yeah, that's true
13:50 jani: the other alternative is to add an expiry date, and then periodically push that forward if they're active
13:51 jani: also, should Mesa membership automatically give push access to drm-tip?
13:51 jani: anyway, putting this out there, let's see what people thing
13:52 jani: *think
16:19 zmike: jesus somebody went wild with llm reporting
16:40 jani: news at 11
16:41 zmike: 😬🔫
19:03 dcbaker: Would anyone with owner status of crucible be awesome and go into settings -> general and turn on CI? Current master has a build break after the Vulkan headers were updated but the video headers weren't. I wrote a CI patch but it doesn't do anything, as it seems that CI is off for the repo
19:07 Kayden: done
19:09 dcbaker: Thanks!
19:09 Kayden: I didn't turn on the docker registry stuff
19:12 dcbaker: Okay. I have no idea what all we need, just that the project needs to have CI enabled to actually run CI. Maybe one of our fearless fdo types will come along and fix everything, lol
19:31 jclsn: Btw I have some flickering on the RX 9070 during the bootscreen and login manager when I have two monitors on DP and one TV over 20m HDMI cable connected. I just realized this only happens when the TV is connected. This did not occur with my RDNA4 card. Should I file a report about this?
19:31 jclsn: Okay that btw was a bit pointless ^^
19:34 bl4ckb0ne: is it relevant to check the return value of `eglDestroyContext`? I have a weird issue im debugging where the EGLDisplay has already been destroyed by _eglFiniDisplay
19:58 bl4ckb0ne: i finally found the issue, `_eglAtExit` is called before my unique_ptr holding my display and context gets destroyed
21:25 bcheng: jclsn: open an issue on drm/amd, sounds like a display driver issue