00:03 robclark: DemiMarie: yes, I believe there is uabi to go btwn syncobj and fence.. drm_syncobj is just a fancy wrapper for dma_fence.. at the driver level when it comes to pageflip/submit/etc it is all just dma_fence
01:20 DemiMarie: robclark: Would OpenGL (even with obscure extensions) be simpler and safer than Vulkan?
01:21 DemiMarie: Also this is in OCaml, possibly with some Rust thrown in.
01:23 robclark: for just doing a blit, I don't think there is any safety difference.. for directly using c gl vs vk, gl will be less lines of code to type... idk what the situation is with ocaml or rust bindings or helper libs
01:24 robclark: gl probably gives you wider hw support (no idea if you care about older things where vk is not possible)
03:23 DemiMarie: Is OpenGL easier to use correctly for this non-expert?
03:24 DemiMarie: My understanding is that OpenGL was originally meant to be used by applications directly, whereas Vulkan was meant to be used by other libraries or by experts. I am very much in the former position.
07:15 K900: Hey folks, NixOS Mesa maintainer here, who do I talk to about libgbm related things? I have a refactoring idea that's been floated in NixOS spaces for a while but I think it really belongs upstream
07:25 zamundaaa[m]: Demi: getting a fence fd from a syncobj requires first waiting for the timeline point to materialize a fence
07:25 zamundaaa[m]: You can either copy what most Wayland compositors do - keep all Wayland state for each surface delayed until the syncobj is materialized and signaled
07:26 zamundaaa[m]: Or do a blocking wait for it to materialize and then extract the fence fd
07:26 zamundaaa[m]: ... or use Vulkan and avoid the issue
07:45 K900: (sorry for the join spam, trying to figure out why the Matrix bridge is not working)
08:02 emersion: zamundaaa[m]: vulkan waits and can't import syncobjs
08:02 emersion: waits as in blocks
08:12 zamundaaa[m]: emersion ah, so it can only export them?
08:32 zamundaaa[m]: Emma [it/its]: yes, but I'm on Matrix
08:32 zamundaaa[m]: You need to authenticate yourself with NickServ to get your messages to IRC
08:47 daniels: ah and K900 is gone again
08:48 K900: I am still fighting the bridge
09:01 K900: Anyway I can read the IRC side from Matrix, just not write, I'll ask the Matrix people about this
09:06 K900: Oh god
09:06 K900: I fought the bridge and won
09:10 K900: Anyway I'll just bump this because I made a bunch of noise:
09:10 K900: Hey folks, NixOS Mesa maintainer here, who do I talk to about libgbm related things? I have a refactoring idea that's been floated in NixOS spaces for a while but I think it really belongs upstream
09:31 karolherbst: if you already coded it out you can always submit the patches and people would discuss it there
09:32 K900: I wasn't sure about the etiquette around that
09:34 K900: I have a draft of a patch (and some notes) on https://github.com/NixOS/nixpkgs/pull/338109
09:34 K900: I can just clean it up and submit to Mesa gitlab
09:37 karolherbst: K900: I think my concern here is, if we make such a API public we'll also need to add support for versioning, not sure if that's already part of it or not
09:38 K900: It is
09:38 karolherbst: okay
09:38 K900: The GBM backend API is already public
09:38 K900: Because Nvidia
09:38 karolherbst: right...
09:38 K900: (and ABI even!)
09:49 emersion: zamundaaa[m]: it can only import/export sync_files
09:49 emersion: zamundaaa[m]: there is the opaque FD ext which happens to import/export syncobjs but that's completely implementation defined and can change any time
09:54 K900: Anyway, I'll submit it as a proper patch then
10:03 emersion: K900: in principle it sounds fine to me but I haven't looked in detail
10:18 K900: All right, submitted https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31074
10:19 K900: Also took the liberty to ping @_oftc_zmike:matrix.org who has done a lot of other cleanups in that area
11:08 bluetail: Hello! Reporting invisible textures in dolphin-emu in different games! (regardless of version) starting from 24.1.4. Installing 24.1.3 of mesa fixed the issue. Before: https://0x0.st/Xwxj.png after: https://0x0.st/XwxQ.png
11:10 bluetail: This is ONLY on OpenGL backend
11:10 bluetail: not on Vulkan
11:12 K900: Does it also happen on 24.2? I think we had a user report something like that on NixOS and 24.2 fixed it for them
11:12 bluetail: yea its also happening on the latest version
11:12 bluetail: mesa-git is unaffected though
11:12 bluetail: but we talking about stable
11:13 bluetail: amdgpu, archlinux latest, hardened kernel
11:21 bluetail: see logs from earlier when I tested vulkan - so its evident it was also happening on latest: https://0x0.st/Xwxc.txt
12:52 HdkR: bluetail: Almost looks culling related rather than invisible textures
12:53 bluetail: HdkR I see. thanks for looking into it
12:53 HdkR: Well, looked at the pictures anyway
12:55 HdkR: Definitely been some changes around radeonsi's culling over the past couple versions. Should be bisectable
12:59 bluetail: HdkR well I found that it started to be an issue after 24.1.3. Is that good enough ?
12:59 bluetail: And that it is no issue with git version
13:00 K900: If you know it happens on 24.1.4 and doesn't on 24.1.3, you could bisect it pretty quickly
13:00 K900: The change is very small
13:00 bluetail: wait, I thought knowing the version is the same as bisect
13:00 bluetail: What is bisect?
13:01 HdkR: bisecting is finding out which change exactly broke it
13:01 bluetail: But I opened the version under tags and see no git diff
13:01 bluetail: why is that
13:01 bluetail: does mesa on gitlab not have diffs?
13:01 K900: It does
13:01 K900: gitlab.freedesktop.org/mesa/mesa/-/compare/mesa-24.1.3...mesa-24.1.4
13:01 K900: * https://gitlab.freedesktop.org/mesa/mesa/-/compare/mesa-24.1.3...mesa-24.1.4
13:02 HdkR: Assuming it is a change between those two versions
13:02 HdkR: But if it is also fixed in HEAD, then just need to wait for a new release which includes whatever the fix is
13:03 bluetail: it is fixed in git version, idk wdym with HEAD
13:04 bluetail: I just don't wanna expand more effort if it doesn't make sense
13:04 K900: "HEAD" means "latest git version"
13:04 bluetail: oh, ok I just wait