15:05 snowycoder[d]: I'm building a simple tool to isolate dEQP co-failures (when a test fail depends on other tests running before it).
15:05 snowycoder[d]: If anyone is intrested I can share it.
17:33 gfxstrand[d]: Oh, that's neat. If you can make it robust, that'd be great
17:33 gfxstrand[d]: It's always a massive pain when you have to debug one of those
17:53 snowycoder[d]: gfxstrand[d]: Oh I have a test that failes if (and only if) it is preceded by 10 other specific tests.
17:53 snowycoder[d]: Remove any of them and boom, everything works 🤷‍♂️
18:55 kar1m0[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1397653331043946637/image.png?ex=68828199&is=68813019&hm=601f9b85339ad3c3b1f9b0ba1cab0ce30f3464593da7a57dc24b426ca9e1034f&
18:55 kar1m0[d]: thank you nvidia for this very informative description
18:59 notthatclippy[d]: kar1m0[d]: I feel you (it's really hard to get good user-facing changelogs with our flow) but.. it does say _Minor_. Like.. here's the diff with the previous version for the kernel part of it: <https://github.com/NVIDIA/open-gpu-kernel-modules/compare/575.64.03...575.64.05>
19:01 kar1m0[d]: notthatclippy[d]: well they did add something
19:01 kar1m0[d]: better than nothing
19:01 notthatclippy[d]: Pretty much always it's support for a new SKU and some kernel compatibility, and then you might get a bugfix or two like in the link above
20:07 mohamexiety[d]: The windows side usually has decent changelogs but it’s mostly specific game bugs and such which I guess is harder to do for Linux
20:23 gfxstrand[d]: Can someone with Arch run this for me so I can get an Arch account?
20:23 gfxstrand[d]: `LC_ALL=C pacman -V | tail -n3 | base32 | head -1`
20:27 avhe[d]: EAQCAIBAEAQCAIBAEAQCAIBAEAQCAIBAEAQCAVDINFZSA4DSN5TXEYLNEBWWC6JAMJSSAZTSMVSW
20:37 magic_rb[d]: gfxstrand[d]: Is that a captcha?
20:39 cubanismo[d]: If so, I prefer that over clicking on motorcyles and traffic lights.
20:41 zmike[d]: What about buses
20:49 magic_rb[d]: My favorite is "select all the squares that contain X", always has me guessing what percentage of the square must be covered by X for it count. I always get it wrong
20:53 ermine1716[d]: Or legs?
21:04 asdqueerfromeu[d]: Or trains?
21:06 mhenning[d]: or when they do this https://xkcd.com/1897/
21:06 asdqueerfromeu[d]: gfxstrand[d]: I assume this isn't related to my AUR packages
21:07 mhenning[d]: asdqueerfromeu[d]: nah, it's a libx11 packaging issue
21:07 mhenning[d]: https://gitlab.freedesktop.org/mesa/mesa/-/issues/13557#note_3023122 if you want to follow along
21:08 snowycoder[d]: gfxstrand[d]: `EAQCAIBAEAQCAIBAEAQCAIBAEAQCAIBAEAQCAVDINFZSA4DSN5TXEYLNEBWWC6JAMJSSAZTSMVSW`
21:10 snowycoder[d]: But it's just encoding the last few strings of the pacman GPL?
21:10 snowycoder[d]: ooooh, I'm stupid
21:21 gfxstrand[d]: mohamexiety[d]: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36336 if you want to do a bit of reviewing
21:22 gfxstrand[d]: asdqueerfromeu[d]: Not unless you're the package of `libx11`
21:29 gfxstrand[d]: Looks like it's fixed now. IDK how long it'll take to make its way through the build farm but the PKGBUILD has been fixed upstream.
21:31 mhenning[d]: I'm not sure arch has a build farm
21:31 mohamexiety[d]: gfxstrand[d]: ooh lets see
21:33 gfxstrand[d]: mhenning[d]: Git commits have to turn into packages somehow, don't they?
21:36 cubanismo[d]: gfxstrand[d]: Note the drm_fourcc.h patch there isn't applied or review yet. It's just from my local tree so far.
21:36 cubanismo[d]: But it's pretty straightforward.
21:37 mhenning[d]: gfxstrand[d]: Yeah, I think core devs build on their machine and upload the binary
21:37 gfxstrand[d]: cubanismo[d]: Yeah, but I needed something so I could type the NVK patch. I'll pull the real version once it's reviewed and upstream. That MR will stay Draft until then.
21:52 cubanismo[d]: I personally find it funny you already had the set_field2 method because the hardware itself has wonky fields.
21:59 gfxstrand[d]: There's a few in the ISA. Blackwell added like 3 more. So I broke down and added the helper. I don't have a corresponding get, though it wouldn't be hard to add.
21:59 gfxstrand[d]: And the image descriptors have several as well.
22:00 gfxstrand[d]: Intel also had a few though I don't think I could fire off an example without looking at code.
22:03 gfxstrand[d]: If I ever have to add `set_field3()` I might get annoyed. 😝
23:02 mohamexiety[d]: gfxstrand[d]: Yup, looks all good and simple. The new table is a lot nicer as an approach too
23:07 mohamexiety[d]: Running into a weird thing with the variable page stuff— so I was testing CTS to see how things are going now but with deqp-runner it just ran into a kernel rip in 9 minutes or so. Tried to figure out the tests causing issues by running a normal sequential 1T CTS buuut I hit a mmu fault after a few hours that killed deqp-vk
23:07 mohamexiety[d]: This would have been useful except the test passes solo. How do I debug this/follow along? :thonk:
23:59 gfxstrand[d]: I honestly don't know what the best practices there are. But a single thread repro is probably easier than threaded