07:17 <junari> kusma: so what I can do next?

07:23 <kusma> Not sure what you mean... I don't think the G31 supports minmax filtering, so I guess the emulator would need code to emulate that...

07:25 <kusma> Atomics are supported by the GPU, so it should probably be possible to wire that up properly for Vulkan, but that needs running the CTS to find out what's missing etc...

07:34 <junari> I understand. But in fact, with these changes I get a working application

07:36 <kusma> Yes, and that's cool. But we can't just take those changes as-is, as they will probably break other applications.

07:39 <junari> Yes, but not many real apps are working now, I think

07:41 <junari> I will double check that dolphin-emu works only with enabled atomics

07:58 <kusma> We'd be violating the spec. We don't do that.

07:59 <kusma> There's a fair amount of applications that does work.

08:01 <kusma> (well, we try to reduce the cases we violate the spec, not increase them)

08:01 <HdkR> It's probably the fragment atomics, since dolphin-emu uses them to emulate boundingbox queries

08:02 <HdkR> I don't think it uses minmax filtering

08:03 <kusma> Fragment atomics is implementable

08:29 <junari> yep, looks like thar only atomics are matter

08:33 <kusma> Well, then run the CTS with that change, and see if anything breaks?

08:35 <kusma> It might "just work"... I think it's wired up for GL, and I think it might have "just worked" on v10 (ignoring some iffy details)

08:54 <junari> Are you talking about this thing? https://github.com/KhronosGroup/VK-GL-CTS

09:18 <kusma> Yeah... Although, it might be easier to just submit a CI run on the Mesa infrastructure, because that will test against known failures etc.

09:22 <kusma> I just kicked off a CI run here on a minimal patch: https://gitlab.freedesktop.org/kusma/mesa/-/pipelines/1345950

09:23 <kusma> Most likely, there'll be a few new fails here, probably the ones included here: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32879/diffs?commit_id=d0a95541fefdd334ae73b2bb4b30826bf5d3885e

09:24 <HdkR> The game to test on the dolphin-emu side to see if it is working as expected would be Paper Mario: The Thousand Year door. A cutscene right at the start of the game where it draws a couple hundred sprites on screen would look incorrect.

09:24 <kusma> Could of course be more, and that they need investigating.

09:25 <kusma> I don't do emulators, so I'll leave emulator testing to others :)

09:25 <HdkR> Figure if the user wanted to test :)

09:26 <kusma> Yep, good plan :)

09:43 <junari> I'm doing this for fun, I only encountered emulators a few months ago. I don't think I'd want to play this.

09:44 <junari> but a portable linux computer is interesting

09:44 <junari> https://drive.google.com/file/d/1nEcfqgR7grn-2045i8bDfEucpkizudsJ/view?usp=drive_link

10:00 <kusma> Nah, lots of failures in that CI run, seems there's stuff that needs investigating etc to enable this. I won't look futher at this right now, as I have other things to deal with, sorry.

10:22 <junari> ok, no problem. I'll open an issue later when I have free time

10:22 <kusma> Sounds great!

12:17 <pac85> Wow dolphin running on panvk? That's cool

12:30 <linkmauve> Reminds me it still fails to render on Wayland, I should fix that someday…

12:49 <mntirc> linkmauve: ah yeah, i always have to force QT to xcb before using dolphin

12:53 <linkmauve> Yeah, that’s super annoying, especially since we had that working about ten years ago, but then someone just removed it…

12:53 <linkmauve> I wonder if Qt’s Wayland integration is less crappy than back then though.

12:54 <linkmauve> The Vulkan widget plain didn’t work IIRC, and the OpenGL widget had major limitations, so people reimplemented them using plain xcb/win32/macos and EGL.

12:56 <mntirc> ah meh

17:01 <soreau> linkmauve: yea, that was disappointing. looks like it's been 11 years since I got the -nogui working (on wl_shell protocol before xdg-shell IIRC) https://www.youtube.com/watch?v=69k2ceiO5PY https://www.youtube.com/watch?v=3ZdXu1VM_VQ