04:07 zzyiwei: I feel so tired of reviewing https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36732 (Android surfaceless for gfxstream). Whoever would love to review or approve, please go ahead.
08:54 glehmann: cmarcelo: gfxstrand: can we replace SPV_KHR_workgroup_memory_explicit_layout with untyped pointers in vtn? Then we could remove the concept of aliased shared variables from nir.
08:56 glehmann: I think the features are equivalent, but untyped pointers are more powerful and fit nir's variable/deref model better.
09:00 mlankhorst: sima: airlied: https://patchwork.freedesktop.org/series/149550/#rev8 should I create a topic branch for the patches in that series that are not xe so that I can merge those to drm-misc too?
09:00 mlankhorst: it's too big to merge to drm-misc, and some of the early patches affect more drivers
09:24 sima: mlankhorst, if it's not super urgent (and shouldn't be since we're in -rc1) I'd just land the prep work in -misc and then land the xe stuff once you can backmerge drm-next
09:24 sima: shouldn't be more than a week of delay
09:25 sima: or do the topic branch if you feel super bored, but usually we haven't bothered with those within drm outside of the merge window freeze
10:02 mlankhorst: Yeah that's doable too, thanks. :)
11:59 mlankhorst: sima, airlied: https://patchwork.freedesktop.org/series/153134/ fwiw
14:28 cmarcelo: glehmann: you mean implement that extension in terms of untyped pointers?
14:30 cmarcelo: i.e. replace the aliased section with a single shared variable, and all the accesses to them with untyped ptr access to that single one? or you are thinking something else?
14:31 glehmann: yes, exactly that
14:37 cmarcelo: glehmann: I think it should work, one thing to be mindful is the lowering code for shared mem in lower_io. in one place we do use a driver based callback to decide alignment of the grouped aliased memory. see if we have enough plumbing to do it from vtn.
15:22 karolherbst: I have posix_memalign stomp over memory allocated by os_malloc_aligned_fd :')
15:26 karolherbst: mhh, maybe my debugger lied to me