00:20anholt: apologies for any cheza flakes in the last hour. fritz is setting up a pile of new ones and I had to restart the runner a couple times.
01:27jenatali: There we go, VK1.2 :)
01:30zmike: but can it pass cts with zink ?
01:30jenatali: Doubtful :P
01:30jenatali: At least, not yet. Maybe we'll get there?
01:31zmike: I believe in you 🙏
01:31jenatali: I'm not even 100% passing CTS yet on any D3D driver, though it's pretty close
01:31jenatali: I'm getting into some super nitty-gritty details of spec differences in sampling behavior to try to close that gap though
01:45airlied: Lynne: nope, but day off today!
01:45airlied: pretty sure if you fix the lengths properly it should all work
01:57Lynne: IIRC I did try overallocating, but it didn't work
04:49Lynne: airlied: oh, happy australia day, I knew there was something on the 26th!
10:28vsyrjala: narmstrong: your last drm-misc-next->drm-tip merge is broken
10:36narmstrong: vsyrjala: hmm weird, how to solve this ?
13:15jani: tzimmermann: http://paste.debian.net/1268527/
13:15jani: drm-tip
13:15tzimmermann: grr
13:16tzimmermann: jani, working on it
13:16tzimmermann: thanks
13:16jani: tzimmermann: thanks
13:16tzimmermann: comes from a failed merge
13:19javierm: tzimmermann, jani: argh, sorry :(
13:19tzimmermann: javierm, that's entirely my fault
13:19tzimmermann: i merged the fbdev patches and they conflicted with your fixes
13:20tzimmermann: while fixing the merge commit, i didn't realize that it says 'fbi' instead of 'info'
13:20javierm: tzimmermann: I see
14:05tzimmermann: jani, javierm, fixes
14:05tzimmermann: fixed
14:16glennk: p[]
14:20javierm: tzimmermann: thanks
14:29jani: tzimmermann: works
14:29tzimmermann: ok
16:23DemiMarie: How do the BSDs work around the lack of dma-buf?
16:25X512: FreeBSD have Linux kernel drivers compatibility layer that can be used with DRM kernel drivers.
16:26X512: Probably it have dma-buf too.
16:26emersion: yes
16:27X512: Expermental Haiku implementation use opaque opaque FDs (actually pipes) for referencing GPU buffers.
16:27X512: So fseek can't be used to get buffer size, special API should be used.
16:29emersion: that will Break Stuff™
16:30DemiMarie: I wonder if there should be a kernel-independent graphics driver project that has a well-defined interface to its host.
16:30DemiMarie: So that non-Linux is not a second-class citizen.
16:31jenatali: Are you including Windows and MacOS in that non-Linux category?
16:32X512: Recently opened Nvidia kernel drivers are Linux independent.
16:32X512: > that will Break Stuff™
16:33X512: It can be unbroken with patching.
16:34X512: https://github.com/NVIDIA/open-gpu-kernel-modules
16:34emersion: it can be patched, but it won't cause a compilation error like some other Linux-specific API usage
16:35X512: Direct dma-buf FD access is rare in client applications. It works fine if imported to Vulakn/OpenGL.
16:36X512: Regular applications usually don't care what GPU buffer FD means exactly.
16:37X512: Things like Wayland compositors are not regular applications.
16:38X512: Usual Wayland compositors can't be easily ported to Haiku because of using epoll/eventfd/timerfd.
16:38X512: Unusual Wayland compototor exists: https://github.com/X547/wayland-server.
16:39X512: compototor -> compositor
20:19danvet: mlankhorst, mripard ack for adding ogabbay to drm-misc?
20:21danvet: ogabbay, ping daniels or another fdo admin in #freedesktop once you have the ack to get added to the drm-misc group
20:22ogabbay: danvet, sure, thx
21:12jenatali: I don't suppose there's some database of which apps need which VK versions/extensions?
21:13jenatali: Trying out DOOM Eternal and it wants extensions that we could implement, but we'd have to emulate, and I'm trying to decide between always-supporting it with emulation, or trying to maintain a database of which apps should be able to use it...
21:56anholt: jenatali: don't think I've seen one. tried to do that for gl long long ago and it was real messy.
21:57jenatali: That's what I expected but still sad nonetheless :P
21:57jenatali: Guess I'm hooking up xmlconfig to Dozen
21:58airlied: jenatali: just expose the exts always, and turn off the emu paths if applications don't enable the exts
21:59jenatali: Right, this is specifically about 8bit loads/stores. I don't really want some app seeing that it's supported and using it, when it'd be almost-definitely faster not to
21:59jenatali: Unless the app refuses to run without it and then sure we can emulate it
22:43swick[m]: MrCooper zamundaaa emersion: EDID/DisplayID can contain the bit depth with and without temporal dithering but they generally just don't because why would any display maker give a fuck
22:44swick[m]: max bpc is also still a maximum and the driver is supposed to lower it to get the mode you specified. if you ever have to specify a max bpc something is broken. it really is a workaround for broken cables etc.
22:46swick[m]: and MrCooper is right, as soon as the pipeline is not identity you might actually want the higher bit depth
22:48swick[m]: we really need a min bpc property. drivers can then just free up bandwidth without any issues