09:05 <tolszak> Hello there, how to report a bug on Odroid N2+ using mesa 21.0.2? There seems something wrong with shader handling I think. Simple scaling done by Qt (using shader internally) behaves strangely, it looks like it confuses coordinated on some scale range. E.g. for 5104x3920 image it "distorts" an image from scale 5 to 12

09:37 <urja> .... that's a big image. (But nah i dont have anything insightful to add)

09:51 <tolszak> urja: On smaller images it does the same just scale needs to be bigger

09:51 <tolszak> e.g. for 1920x1080 the issues happens from 16 to ~30

09:54 <HdkR> Make sure to also test with latest mesa versions. The issue could have already been fixed

12:21 <tolszak> HdkR: It still happens with 21.2.4

12:24 <macc24> tolszak: can you try with main branch of mesa?

12:25 <tolszak> macc24: I can easily test it with yocto hardknott so perhaps I can try to build 21.3.0-rc2. Is it enough?

12:28 <macc24> tolszak: try to aim for as new as it can possibly be

12:32 <tolszak> macc24: The newest tarball is 21.3.0-rc2 on https://archive.mesa3d.org/. YOcto hardknott recipe is pretty easy to modify and use such bit. Incorporating crosscompilation on host and deployment to device is probably a work for a few hours so will first give a try tarball.

12:32 <macc24> tolszak: https://gitlab.freedesktop.org/mesa/mesa/-/archive/main/mesa-main.tar.gz

12:34 <tolszak> macc24: Perhaps I can use this one too. thanks

12:46 <macc24> tolszak: if you can capture an apitrace and if apitrace can replicate the bug i can try replicating it on my G72 machine, if i can find my usb-c - usb-a adapter

12:51 <tolszak> macc24: Will try to do it, thanks. Although I've never used apitrace. But I'm sure I figure it out. Thank you for guides.

15:13 <tolszak> macc24: Have the apitrace *.trace file

15:17 <tolszak> macc24: I forgot to mention that I don't use X11 just plain writing to egl display with Qt's eglfs platform plugin

15:18 <macc24> tolszak: can you see the bug with LIBGL_ALWAYS_SOFTWARE=1 in your environment?

15:19 <tolszak> Yes

15:20 <tolszak> macc24: But I use GLES

15:20 <tolszak> I don't know if the same var is responsible for that

15:20 <macc24> tolszak: so if it still replicates with software rendering then it's not panfrost bug

15:21 <tolszak> Is the LIBGL_ALWAYS_SOFTWARE variable used in non debug builds?

15:21 <macc24> it's then *checks backlog* meson_drm bug

15:21 <macc24> tolszak: yes, it forces software rendering

15:22 <tolszak> Great to know that.

15:22 <tolszak> macc24: Thanks for narrowing down the issue!

15:23 <macc24> tolszak: sure no prob

15:25 <alyssa> macc24: ❤️

15:26 <alyssa> I love problems that solve themselves 😋

15:27 <macc24> alyssa: gotta love being "jack of all trades" lol

16:22 <alyssa> pipewire of all trades now

16:47 <macc24> imagine using pipewire, L // this post was made by the alsa gang

18:11 <alyssa> cphealy: if you have an Android DDK lying around, https://rosenzweig.io/wat.c might do something

18:11 <alyssa> per https://android.googlesource.com/device/linaro/hikey/+/16100f6%5E%21/

18:11 <alyssa> +# GPU support for AFBC 1.0

18:11 <alyssa> +# GPU support for AFBC 1.1 block split

18:11 <alyssa> +# GPU support for AFBC 1.1 wide block

18:12 <alyssa> +# GPU support for AFBC 1.2 tiled headers

18:12 <alyssa> I suppose those defines are the useful part of that commit, not the dlopen stuff

18:13 <alyssa> Midgard can't do wide blocks or split blocks, so I guess that means it's AFBC 1.0

18:13 <alyssa> that makes sense, it's the first gen with AFBC

18:14 <alyssa> G72 can do wide blocks and split blocks but not tiled headers, so that means it's AFBC 1.1

18:14 <alyssa> while Arm says here https://community.arm.com/arm-community-blogs/b/graphics-gaming-and-vr-blog/posts/the-mali-g51-gpu-brings-premium-performance-to-mainstream-mobile that G51 is AFBC 1.2

18:14 <alyssa> and G76 can do tiled headers, it's AFBC 1.2

18:15 <alyssa> So I guess in practice, the magic table is

18:15 <alyssa> "v4 = no afbc, v5 = AFBC 1.0, v6 = AFBC 1.1, v7 = AFBC 1.2"

18:15 <alyssa> Idk if there's an AFBC 1.3 but that would be Valhall I guess

18:16 <alyssa> ah, yep

18:16 <alyssa> https://community.arm.com/arm-community-blogs/b/graphics-gaming-and-vr-blog/posts/introducing-arm-mali-g77-with-new-valhall-architecture

18:16 <alyssa> "The new features in Valhall through AFBC 1.3. include support for 2-plane YUV, improved support for front-buffer rendering, support for separate depth/ stencil encoding (a better match for Vulkan) and opacity/ transparency hints"

18:16 <alyssa> "v4 = no afbc, v5 = AFBC 1.0, v6 = AFBC 1.1, v7 = AFBC 1.2, v9 = AFBC 1.3"

18:17 <alyssa> So there you have it