00:46 airlied[d]: gfxstrand[d]: so looks like the underlying tiling for 16bpp/8bpp surfaces has changed somehow, and host image copy breaks
00:48 airlied[d]: https://paste.centos.org/view/raw/d0a088e7 has a dump with the different pattern in it, if you can make sense of it
03:43 airlied[d]: okay brought the big guns, 2 hrs on the 5080 with a real cpu
03:58 mangodev[d]: question
03:58 mangodev[d]: is there a way to get mesa to build with llvm 20 on arch? currently, my whole system is held up because mesa apparently hard depends on llvm 19, but i can't find any references to that
04:02 airlied[d]: should work fine
04:03 mangodev[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1382933251823370326/image.png?ex=684cf474&is=684ba2f4&hm=2ea10d3681aefbe53377c6bfdccf06df12ed8a3561d131d5ea7fc1207486763a&
04:03 mangodev[d]: this is the error that is given when i try to upgrade my system
04:03 airlied[d]: something wonky with arch 32-bit packages
04:03 mangodev[d]: it's also for the 64-bit one
04:03 airlied[d]: f42 ships llvm 20 + mesa 25.0.6
04:04 mangodev[d]: oh wait i see
04:04 mangodev[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1382933650496163841/image.png?ex=684cf4d3&is=684ba353&hm=fc3d9c25257080eaa0bee29543519e127828de17f9359a4d3bf0e7e20e3a9b58&
04:04 mangodev[d]: nvm
04:05 mangodev[d]: ig it's safe enough to just remove the version check?
04:09 mhenning[d]: mangodev[d]: you'll need to rebuild with new llvm
04:10 mhenning[d]: `pacman -Syu mesa lib32-mesa` to update and switch to the pre-built packages at the same time
04:10 mhenning[d]: then pull the mesa-git repo and re-build
04:41 mangodev[d]: hmmmm
04:41 mangodev[d]: bricked my drivers
04:46 mangodev[d]: okay hopefully it's fixing rn
04:50 mangodev[d]: fixed :D
05:27 airlied[d]: Pass: 1325445, Fail: 88, Crash: 15, Skip: 1470557, Timeout: 13, Flake: 1, Duration: 1:46:17, Remaining: 0
05:42 airlied[d]: https://paste.centos.org/view/raw/f12aa0f6
06:09 airlied: anarsoul: if you are feeling up for 6.15 + 4570355f8eaa476164cfb7ca959fdbf0cebbc9eb + 4570355f8eaa476164cfb7ca959fdbf0cebbc9eb
06:11 anarsoul[m]: Would it apply to 6.16-rc?
06:18 airlied: anarsoul: no, it has a different bug, I've sent them to stable team to pick up for 6.15
06:25 anarsoul[m]: OK, thanks, I'll try it tomorrow
06:27 airlied: I'll try and figure out 6.16 next, might be early next week :-P
06:52 phomes_[d]: VK_AMD_buffer_marker: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35507
07:03 skeggsb9778[d]: airlied[d]: wow, nice!
08:07 mohamexiety[d]: airlied[d]: We have a way to get the GOB layout so I can check in a few hours
08:07 mohamexiety[d]: airlied[d]: Oooh nice!!
08:10 mohamexiety[d]: The host image copy stuff passes too or did you skip?
08:19 LinuxLover471: Hello!
08:19 LinuxLover471: Is anyone here?
08:23 LinuxLover471: Hello!
08:23 LinuxLover471: Hi!
08:25 tiredchiku[d]: yes
08:30 phasta: tiredchiku[d]: now you chased him off <.<
08:31 tiredchiku[d]: I have that effect on people
08:33 LinuxLover471: Hi!
08:33 LinuxLover471: I use archlinux and wanted to ask whether to use the normal kernel or the linux-git kernel to perform tests on my system and help with development?
08:36 airlied[d]: Just Vulkan 1.3, I turned off host image copy
08:37 mohamexiety[d]: airlied[d]: Ok will take a look at that then, thanks!
08:41 mangodev[d]: LinuxLover471: i personally get by with regular kernel, but the git kernel may be needed for stuff like testing on blackwell
08:42 LinuxLover471: My device is old, a GT 740 with Kepler 2 architecture as reported by inxi
08:42 LinuxLover471: Also Is it important to use the normal kernel and avoid flavours like the Zen kernel?
08:54 LinuxLover471: Nvm, I have installed the normal linux kernel, will compile mesa from source and perform the piglit tests.
09:12 asdqueerfromeu[d]: mangodev[d]: NVK/zink don't need LLVM (but rusticl does)
09:13 LinuxLover471: Hey! Can I use the mesa-git from the AUR? I find it easier to use AUR packages.
10:01 LinuxLover471: Hey, can I use the mesa-git package in the AUR instead of compiling myself?
10:01 LinuxLover471: Because it will be so much more convenient.
10:07 LinuxLover471: There are constant power cuts here, so I need to ask again, can I use mesa-git from the AUR?
10:07 LinuxLover471: Extremely sorry for repeated questions.
10:10 asdqueerfromeu[d]: LinuxLover471: There was a recent LLVM version fixup (so it should probably work)
10:12 LinuxLover471: Okay thanks!
13:23 karolherbst[d]: mhenning[d]: hit any regressions caused by your recent RA change? though not sure if it's related to mma instructions in my case, but on Ampere I have a scheduling related regression from your vector RA optimization
13:45 karolherbst[d]: it's a write after read hazard 🥲
13:47 karolherbst[d]: yeah.. maybe I should split the files already, because .. uhhh
13:47 karolherbst[d]: there are soo many subtle differences
14:57 LinuxLover471: Hello! I just installed nouveau and compiled mesa-git and lib32-mesa-git from AUR and installed waffle by using chatgpt as I didn't find any instructions. After this, I tried running piglet as non-root and it gave an error. Also when I tried with root it skipped all the 10000ish tests.
14:59 LinuxLover471: https://pastebin.com/jP84sWJs
14:59 LinuxLover471: This is where you can read the error, when I run it as root then it skips everything, is that normal?
15:06 LinuxLover471_: Sorry my network wonked.
15:08 LinuxLover471_: No, my host is still here
15:13 LinuxLover471_: I think this is because I compiled mesa from the AUR? I will try to compile it myself then.
16:01 anarsoul: airlied: it looks like you mentioned the same commit hash twice (4570355f8eaa476164cfb7ca959fdbf0cebbc9eb), I assume you meant some other commit?
16:09 mhenning[d]: karolherbst[d]: which RA change? https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35143 or https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35033 ?
16:10 mhenning[d]: either way, both pass cts and neither has known regressions
16:11 mhenning[d]: it's possible that one of the changes uncover an existing issue elsewhere so I'm not sure if it's the fault of the change or not
16:23 karolherbst[d]: mhenning[d]: yeah, no worries, it's just more buggy latencies for ampere
16:30 mohamexiety[d]: I have a feeling NV gave you the GA100 latencies and not the GA102 latencies 🐸
16:30 mohamexiety[d]: i.e. sm80 and not sm86/sm89
16:32 mhenning[d]: sm80 binaries are supposed to run unmodified on sm 86 and sm89
16:32 mohamexiety[d]: yeah was about to say
16:33 mohamexiety[d]: if they're supposed to be binary compatible it should be the same so :blobcatnotlikethis:
16:33 mhenning[d]: so the latencies are supposed to be sufficient unless it's a feature introduced in a later gen
16:57 karolherbst[d]: mohamexiety[d]: correct
16:57 karolherbst[d]: mhenning[d]: well....
16:57 karolherbst[d]: apparently they don't
16:58 karolherbst[d]: Dave also found a wrong latency with FP16
16:58 karolherbst[d]: even left a comment
16:58 mohamexiety[d]: the docs say they should so that's weird :thonk:
16:58 mohamexiety[d]: and if they run unmodified then the latencies should be similar cuz latency is encoded in the instruction
16:58 karolherbst[d]: anyway.. the required latencies for HMMA are higher on my GA102, soo..
16:59 karolherbst[d]: for reasons unkown it seems to match the hopper ones
16:59 mohamexiety[d]: either way :blobcatnotlikethis:
16:59 mhenning[d]: I'm aware of the wrong latencies, but I don't see any reason to suspect that's binary incompatibility
16:59 karolherbst[d]: well my GA102 needs higher latencies
16:59 karolherbst[d]: so that's incompatible
16:59 mhenning[d]: higher latencies than the spreadsheet though
17:00 karolherbst[d]: ohhh uhm... you know that nvidia sometimes inserts "bra" instructions?
17:00 karolherbst[d]: ehm
17:00 karolherbst[d]: noop I mean
17:00 karolherbst[d]: I wonder if they do that to stay binary compatible so they can patch latencies at runtime
17:00 mhenning[d]: all we know is that ga102 is incompatible with the spreadsheet
17:00 karolherbst[d]: nah
17:00 karolherbst[d]: I already got confirmation
17:00 mhenning[d]: of what?
17:01 karolherbst[d]: that GA102 is different and needs higher than GA100
17:01 mhenning[d]: ah, okay.
17:01 karolherbst[d]: yeah.. I just hope I get the spreadsheets soon
17:17 pavlo_kozlenko[d]: Will I be able to use my Kepler gt 630 video card to record video to an OBS via va-api?
17:24 karolherbst[d]: probably not
19:54 gfxstrand[d]: I hate it when CTS tests pass with `NVK_DEBUG=zero_vram`...
19:55 mhenning[d]: try trash_memory ?
19:55 gfxstrand[d]: I mean yes, that makes them fail
19:55 gfxstrand[d]: But why...
20:06 gfxstrand[d]: My bug.
20:07 gfxstrand[d]: I was doing `nvk_sampler_fill_header()` straight into GPU memory. It assumes the memory it's writing to is zero-initialized.
20:07 gfxstrand[d]: I think that means I need to rework it
20:09 mhenning[d]: yeah, that'll do it
20:14 gfxstrand[d]: I'm gonna do the same thing I did with NIL
22:34 phomes_[d]: I added a tracker issue for supporting the Crash diagnostics layer: https://gitlab.freedesktop.org/mesa/mesa/-/issues/13354
23:16 airlied: anarsoul: a738fa9105ac2897701ba4067c33e85faa27d1e2 was the other one