06:00 mangodev[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1421375870232297483/image.png?ex=68d8ceef&is=68d77d6f&hm=9eb56ab470d3741c302bffecc26e3ce0d8513789d711f9cae72dc62aec2431c8&
06:00 mangodev[d]: welp
06:00 mangodev[d]: i found why mesa won't build for me
06:00 HdkR: what
06:00 HdkR: how even
06:01 mangodev[d]: idek
06:01 mangodev[d]: my system wayland is 1.23.1 🫠
06:01 mangodev[d]: and it seems to only happen in lib32?
06:02 HdkR: But it says the opposite, needs 1.23.1 but found 1.24.0?
06:02 mangodev[d]: yeah
06:02 mangodev[d]: i'm wondering where it's getting 1.24.0
06:02 HdkR: Unless maybe that version print is just backwards.
06:02 mangodev[d]: wait
06:03 mangodev[d]: my system wayland *is* 1.24?
06:03 mangodev[d]: i guess is should downgrade it
06:03 mangodev[d]: but idk what else would like that or not
06:03 mangodev[d]: idk if downgrading a major version would be a good idea for other wayland stuff on the system
08:38 mohamexiety[d]: mhenning[d]: since you have been looking into this, if I wanted to test out Blackwell's barrier/WFI semantics, what stuff should I look into? I was thinking of a test program that does some 3d+compute stuff on one queue and then inspecting it with envyhook on ada and blackwell and looking at the differences. or are there more complex patterns we should look out for?
14:36 gfxstrand[d]: I'd like to have a test that tests the actual hardware, too. The way I'd do that would be to do compute, 3D, compute with no barriers in between and have all of them bang on the same atomic counter and report out their results somehow. If there is no implicit WFI, we should see all three intermixed.
15:40 mhenning[d]: mohamexiety[d]: I'm actively working on this. I already wrote a program that dumps nvidia's sequences for vk_events, and we have data on that for blackwell here: https://gitlab.freedesktop.org/mhenning/re/-/tree/main/vk_event?ref_type=heads and I'm in the middle of writing a test that deliberately tries to create raw hazards and reports what value it reads
15:40 mhenning[d]: https://gitlab.freedesktop.org/mhenning/re/-/tree/main/vk_test_overlap_exec?ref_type=heads
17:38 orowith2os[d]: mangodev[d]: if it was already built, it'll be fine.
17:38 orowith2os[d]: that error seems wrong, though.