01:02 dcbaker: mareko: pretty sure that was the slogan of my grade school baseball team
01:18 mareko: lol
01:19 mareko: I'm surprised Raja didn't bring up those words
09:14 marcan: well that was a nice rabbit hole for today: https://gitlab.freedesktop.org/mesa/drm/-/issues/85
09:25 ccr: heh
15:16 Lynne: offtopic, do I need a vulkan memory barrier between compute->compute dispatches if I don't care about the contents of a buffer (which I'm only using for scratch)?
15:17 Lynne: or do I need a memory barrier to indicate I don't care about the contents of a buffer? the spec doesn't really say anything
15:57 ishitatsuyuki: Lynne, you need an execution barrier to ensure writes from previous dispatch doesn't bleed into the later dispatch
15:59 ishitatsuyuki: memory barrier is not necessary since you don't need writes to be visible to later read
16:01 Lynne: so an event or a semaphore, huh?
16:01 ishitatsuyuki: just a pipeline barrier
16:04 Lynne: what if I also had another buffer that I read+wrote in the 2 dispatches? pipeline + memory barrier?
16:06 Lynne: err, nevermind, pipeline barriers is what I'm already using
16:07 Lynne: (I've abstracted them enough to forget what they're called)
16:07 ishitatsuyuki: after reading stuff i'm actually unsure whether it's safe to omit the memory barrier now
16:08 ishitatsuyuki: maybe you could add a COMPUTE->COMPUTE memory barrier for everything you wrote to
16:10 HdkR: I was today days old when I learned xcb_connect* isn't allowed to return nullptr. Very quirky.
17:22 Lynne: ishitatsuyuki: I had the impression that using stage = NONE and access = NONE indicated you don't care about the contents, but I'll just put barriers in as if the contents did matter, thanks
17:23 Lynne: afaik barriers are ignored on most implementations anyway, like with image barriers
19:32 DavidHeidelberg[m]: eric_engestrom: how much do we need that cc/cxx wrapper now? We can enable ccache everywhere, only think I see useful there is passing ` -Werror` to the linker for LTO
19:33 DavidHeidelberg[m]:asks because is lazy; tried to deal with clang to clang-15 change (by dropping -15 from the CC and CXX), but meson is still not happy
22:51 DavidHeidelberg[m]: What do you think about more HORIZONTAL CI pipelines? https://gitlab.freedesktop.org/okias/mesa/-/pipelines/839046
22:54 DavidHeidelberg[m]: what I like about it, is that horizontal scrollbar fits at the bottom of the view, so you see all the jobs without having to scroll up & down
22:56 psykose: doesn't it take longer because of less parallelism and more stages
23:02 DavidHeidelberg[m]: Hmm, I think we have dependencies on individual jobs; but I'll recheck.
23:03 psykose: (i'm only guessing, since i'm used to the gitlab 'all in stage depends on entire previous stage' kinda thing)
23:03 psykose: but if it's not restricted by that then there's no change of course and all works fine