01:50 karolherbst: I've added the pattern here and it seems to help a bit. Just mostly confused why there wasn't a rule already, but for inexact ones that should be fine? mhh https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/42186/diffs?commit_id=00e894f8adbf26521798dcfea74de779e6125af7
01:54 karolherbst: or uhm.. was that nir_opt_reassociate?
09:30 MrCooper: zf: FWIW, you can add the fence representing the client GPU work (in the form of a sync_file fd) to the implicit synchronization object using the DMA_BUF_IOCTL_IMPORT_SYNC_FILE on the dma-buf fd before submitting the PixmapPresent request, and extract the fence representing the server GPU work using the DMA_BUF_IOCTL_EXPORT_SYNC_FILE ioctl after receiving the PresentIdleNotify event, and handle those fences as required by the
09:30 MrCooper: Vulkan API
09:35 Tom^: MrCooper: is the fence in multiplane buffers the same on all fds, or can they be different ones?
09:37 Tom^: just a out of context, but related question :p
09:37 MrCooper: it should be possible to have a single fence which ensures correct synchronization for all of them
09:39 Tom^: so its no general rule regarding that then, depends i guess