00:12FireBurn: I have pixels :D
00:12FireBurn: They aren't the correct ones, but I have them all the same :D
06:27CounterPillow: all pixels are beautiful in their own, unique way
13:39FireBurn: Pixel perfect but slow as hell
17:01marex: daniels: I came across a comment from three years ago , that mesa libgbm does not support allocating NV12 BOs , because mesa does not support NV12 as render target , https://github.com/waydroid/android_external_mesa3d/issues/4#issuecomment-1422354484
17:01marex: daniels: I wanted to ask, is that still the case ?
17:02marex: daniels: I would like to use the panfrost driver with G310 GPU and render video frames into NV12 or I420 DMABUF buffers for follow up encoding with HW encoder
17:02marex: it seems to me the panfrost driver does support the NV12 render target, but I feel I am missing something somewhere
17:03marex: (and no, I am not using any downstream stuff)
17:32daniels: marex: there’s now YUV target support very close to landing for the next release for fdno and quite close for Mali
17:32daniels: just beware that it is really quite slow on Mali so may not necessarily be a win for you
17:47marex: daniels: mali I assume that is the new mali, not lima mali400, right ?
17:50marex: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40191 this I guess
17:52marex: and https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/42765 , I see
17:52marex: daniels: thank you
17:57daniels: yeah, that’s only the panfrost GPUs; I’m not sure if 4xx can do it, but I would assume so
17:57marex: the G310 I have is panfrost
18:42Lyude: Now that I'm looking at rust KMS stuff finally, there's something I'm trying to figure out - do we actually expect to support drivers where some CRTCs implement vblank support and others don't? I kind of assume the answer is no, in which case I wonder if we can make `drm_vblank_init()` not require the CRTC count?
18:46Lyude: The main thing making me ask this: with the current KMS bindings, I have to track whether or not any CRTCs with vblank support get created at runtime during driver init. And subsequently since I don't really see any value in the oddball possibility of "only some CRTCs have vblank support" and I more or less assume things would be broken like that, it also means I have to ensure
18:46Lyude: that after we allow a driver to create modesetting objects that we actually loop through all of the CRTCs and ensure that if any implement vblank, they all do.
18:46Lyude: and this seems like maybe something we should just be doing in the DRM core instead of in the rust bindings
18:48alyssa: daniels: ..why is it slow on mali?
18:54Lyude: you know, perhaps it is simply easier for me to just make the change and see if anyone complains on the ML
18:54alyssa: Lyude: legit
19:08glehmann: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/43141 I'm surprised that no static analysis tool flags func(union->member1) || func2(union->member2)
23:05DragoonAethis: glehmann: Coverity has a check like that, but it matches on a very specific pattern (write to member2 after read from member1 or something like that)
23:09karolherbst: anybody any complains on adding a fmul3 instruction that act as a fused fmul of three operands? or should we just vendor it? (re https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/42186/diffs?commit_id=c43627e149f2412ff99db1dcc95c4164e872fe17)