17:18 chikuwad[d]: mhenning[d]: leaving *swap\* intrinsics alone caused CTS failures, and I ended up modifying a minimal set
17:19 karolherbst[d]: oh yeah.. any decisions about https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/41042 ? Or anything I should try out or test to make reviewing easier?
17:31 mhenning[d]: chikuwad[d]: Do you know why you see cts failures? I don't really understand why that would make a difference.
17:31 mhenning[d]: karolherbst[d]: I'm working on reviewing it today
17:31 karolherbst[d]: okay, cool, thanks
18:17 chikuwad[d]: mhenning[d]: I'll grab a log soon
22:42 karolherbst[d]: I probably want to add shared memory usage to the shader stats, and also adjust `max_warps_per_sm` to account for that...
22:43 karolherbst[d]: I know some compute workloads that are caped by shared memory usage, but I have no idea if that's also the case with certain games...
22:43 karolherbst[d]: and aliasing shared memory variables could help there (I know it will for some compute stuff)
22:45 karolherbst[d]: might also help with async compute generally..
22:50 karolherbst[d]: and the prepass scheduler will also benefit from this information
22:54 chikuwad[d]: mhenning[d]: https://pastebin.com/5QFb0qeg
22:55 karolherbst[d]: phomes_[d]: if you want to try something out: https://gitlab.freedesktop.org/karolherbst/mesa/-/commit/a681a8f6414e6ecb9453cb971ccc6dd9c5794d53
22:55 karolherbst[d]: I'm curious if that would help any games
22:57 mhenning[d]: chikuwad[d]: that looks like we'd just need to remove the `instr->num_components = 1` for the comp_swap case
22:59 mhenning[d]: It's probably specifically the `atomic->num_components = glsl_get_vector_elements(deref_type);` line in ntv
23:01 jStefan: karolherbst[d], I filed my bug in launchpad, does that get enough eyeballs ?
23:12 chikuwad[d]: mhenning[d]: that still causes a failure in the same case https://pastebin.com/muuxrd8q
23:12 chikuwad[d]: I'd poke more at it right now but it's 01:12 and I have a lecture at 09:00 😅
23:14 chikuwad[d]: also do I also leave nir_intrinsics.py:846 alone?
23:14 chikuwad[d]: `image("atomic_swap", src_comp=[4, 1, 1, 1], dest_comp=1, extra_indices=[ATOMIC_OP])`
23:15 chikuwad[d]: image atomic swap
23:32 mhenning[d]: might need more adjustments then
23:32 mhenning[d]: or I suppose going back to the initial "adjust everything" version is fine
23:33 mhenning[d]: and no rush. the MR can wait until after lecture 😛
23:33 mhenning[d]: image atomic swap should be the same as the rest of the atomic swaps
23:33 chikuwad[d]: oh yeah absolutely, it's waited a few months already, what's a day more
23:33 chikuwad[d]: mhenning[d]: ok so I had the right idea in mind
23:34 chikuwad[d]: mhenning[d]: I'm open to doing either, I suppose I'll leave it to your preference here 😅
23:35 karolherbst: jStefan: probably not, but you could link it here or file a bug against https://gitlab.freedesktop.org/drm/nouveau/-/work_items?sort=created_date&state=opened&first_page_size=100
23:41 jStefan: https://bugs.launchpad.net/ubuntu/+source/xserver-xorg-video-nouveau/+bug/2148792
23:42 jStefan: i'm waiting on ubuntu 26.04 to release to try again with it, this is not a production system, i can experiment.