11:42 <alyssa> ironically gl_BaryCoordEXT is probably implementable on Mali

11:42 <alyssa> all the way back to Midgard

11:43 <alyssa> dont remember if newer Mali drops support tho

11:44 <kusma> Huh, really? I thought I'd looked for ISA support, but not found it...

11:44 <alyssa> kusma: Midgard has it

11:44 <alyssa> if you read point coord in a TRIANGLE you get barys

11:44 <alyssa> not sure if that works on bifrost+ though

11:44 <kusma> Aha

11:45 <alyssa> i was also surprised about that, i think hans-kristian was the one who told me about that trick lol

11:56 <kusma> It actually makes a lot of sense, knowing abit about how this works in HW...

16:14 <leo60228> i'm trying to run mainline linux on an mt8183/mali-g72 device (lenovo ideapad duet/krane) and get a garbled output when using anything other than xorg: https://vriska.club/i/htkxo.png

16:15 <leo60228> i've tried kwin_wayland, weston, and mpv with the same results (`mpv --vo=drm` is fine but `mpv --vo=gpu --gpu-context=drm` to use opengl isn't)

16:15 <leo60228> this seems like it's probably just an issue with linux on this hardware *but* the bizarre thing is that it only happens on some distros

16:16 <leo60228> (and userspace is what causes the difference, using a working distro's kernel doesn't help)

16:16 <leo60228> i've run into this issue on nixos and based on discussion with others chimera linux has the same issue, but it's fine on debian, fedora, and postmarketos

16:18 <leo60228> there's some iommu warnings in dmesg but nothing else relevant-looking in any logs: https://gist.github.com/leo60228/83f16373c36176d40bb5e0d2e013184b

17:51 <leo60228> PAN_MESA_DEBUG=noafbc seems to work around it?

17:54 <leo60228> probably this issue https://gitlab.collabora.com/mediatek/aiot/linux/-/merge_requests/9

17:56 <kusma> That's working around a bug in the kernel display driver

17:58 <kusma> The display driver claims to support modifiers it doesn't support

18:10 <daniels> https://gitlab.collabora.com/mediatek/aiot/linux/-/commits/wip/adalessandro/mediatek-fix-afbc?ref_type=heads should fix that

18:13 <alyssa> daniels: https://gitlab.collabora.com/mediatek/aiot/linux/-/commit/23acc39c7c45d91f7ac7f51554aad2142616f954 feels like we should also be requiring YTR in the modifier bit no?

18:14 <alyssa> i distinctly remember fixing this bug on rk3399 years ago

18:18 <daniels> yeah, that just needs to look at the YTR bit in the modifier

20:28 <olivial> I'm fairly sure that gl_BaryCoordEXT is not implementable on bifrost/valhall

23:34 <mntirc> olivial: sniff... guess i'll need to come up with a way to emulate it in blender then, without their geometry shader workaround