01:14TranquilIty[m]: <notthatclippy[d]> "At the very least you'd need..." <- What if we'd patch the kernel to not call the teardown sequence and just unload the module ?
01:14TranquilIty[m]: We are considering whether this'd be easier or harder than extracting the PMU firmware from the prop driver blob and reverse engineering (unless someone had already done so?) the interface for that...
01:19TranquilIty[m]: Our particular arch of interest is NV130/Pascal, unsure if there's already some workaround for those to get comparable performance (same order of magnitude) to the prop driver. This is mostly because nouveau is actually working much better than the prop driver already w.r.t. modesetting :D
01:22TranquilIty[m]: Ah we missed the rest of the convo that followed actually, so it should "just workTM" to load prop, unload, then load nouveau ?
01:23TranquilIty[m]: We really need to get a unit to test this with
13:57tiredchiku[d]: vulkan 1.4
13:57tiredchiku[d]: https://github.com/KhronosGroup/Vulkan-Docs/commit/c7a3955e47d223c6a37fb29e2061c973eec98d0a
15:52gfxstrand[d]: Yes, that is a thing. 🙂
17:17gfxstrand[d]: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32433
17:44mhenning[d]: gfxstrand[d]: Are the nvidia entries under the Software Freedom Conservancy on https://www.khronos.org/conformance/adopters/conformant-products all nvk entries? I notice they don't list a driver the way the RADV entries do.
17:46asdqueerfromeu[d]: mhenning[d]: Why would NVIDIA need to go through the SFC?
17:46gfxstrand[d]: mhenning[d]: Yes, they're NVK entries.
17:47gfxstrand[d]: All the metadata about drivers, CPUs, and the like is optional and free-form. There is no standard.
17:48mhenning[d]: Sure, that's what I assumed - it just requires a little reading between the lines right now
17:51gfxstrand[d]: Yeah, anything from SFC or SPI is Mesa
18:07gfxstrand[d]: gfxstrand[d]: Merged!
18:07tiredchiku[d]: nice
18:22karolherbst[d]: impressive speed run
18:23mohamexiety[d]: \o/
18:24karolherbst[d]: so if I got that right, nvidia was the quickest and then nvk?
18:27tiredchiku[d]: correct
18:27karolherbst[d]: nouveau being unstable prevented Faith to submit before nvidia probably, lol
18:48tiredchiku[d]: honeycrisp comes third
18:48tiredchiku[d]: https://rosenzweig.io/blog/vulkan-14-sur-asahi-linux.html
18:53Lyude: f_, karolherbst[d] what kind of system is this? FWIW: I'm not totally sure but it looks like an issue with the LUT arguments, but the LUT arguments don't really make sense to me according to nvidia's headers
18:59karolherbst[d]: good question
19:49f_: Lyude: HP EliteBook 8560w
20:07gfxstrand[d]: karolherbst[d]: Nah, NVIDIA cheated on their first round of submissions. That's the only way they got in before me.
20:08gfxstrand[d]: Which, TBF, I kinda wish I'd done. 😅
20:08karolherbst[d]: 🙃
20:09karolherbst[d]: what did they do?
20:11gfxstrand[d]: They ran on 1.4.0.0-rc4 and then just re-ran `git status` after pulling the 1.4.0.0 tag and everything was good because it was the same as 1.4.0.0-rc4.
20:12gfxstrand[d]: The CTS records the git SHA of the VK-GL-CTS package in every qpa file and then the `git status` you put in the package is checked against that to make sure the run matches the tag. So it's all fine.
20:13gfxstrand[d]: That's the only way Piers was able to upload 1.4.0.0 results less than an hour after the tag was pushed.
20:13gfxstrand[d]: That or they have a REALLY efficient farm with package generation automated.
20:14karolherbst[d]: or both
20:14gfxstrand[d]: Yeah
20:14karolherbst[d]: I mean, nvidia has an impressive CI system
20:15gfxstrand[d]: I did all my runs on the one desktop and swapped out cards each time.
20:15karolherbst[d]: so it's not even unplausible they could run the CTS through it quickly
20:15karolherbst[d]: oh yeah.. Nvidia can just run stuff in parallel 🙃
20:15gfxstrand[d]: I also ended up making a flash run to BestBuy for an RTX 4060 because my laptop wasn't stable enough. 🤡
20:15karolherbst[d]: heh
20:15karolherbst[d]: I'm sure somebody at nvidia might have scripted the entire CTS run thing and just pushed it through their CI system
20:17gfxstrand[d]: It's possible but I doubt it. Even sharding the maximum amount and assuming you're not fighting the GSP lock, it'd still be hard to get packages put together that fast.
20:17karolherbst[d]: mhh, apparently they are also 1.4 conformant on Maxwell 1st gen
20:17karolherbst[d]: their CI system is really impressive
20:18karolherbst[d]: but yeah.. could also cheat, hard to tell anyway
20:20gfxstrand[d]: And it doesn't matter.
20:33gfxstrand[d]: https://www.collabora.com/news-and-blog/news-and-events/nvk-now-supports-vulkan-14.html