Re: How to fetch the implicit sync fence for a GPU buffer?
On Friday, June 2nd, 2023 at 12:29, Michel Dänzer wrote: > > I’m wondering whether there’s an API -- maybe something analogous to > > drmPrimeHandleToFD() – that allows userspace to fetch a waitable fence fd > > for a dmabuf? > > A dma-buf fd itself can serve this purpose; it polls as readable when the GPU > has finished drawing to the buffer. If you _really_ need a sync_file FD, you can extract it via DMA_BUF_IOCTL_EXPORT_SYNC_FILE. This is a somewhat recent kernel uAPI.
Re: How to fetch the implicit sync fence for a GPU buffer?
On 6/1/23 23:15, Hoosier, Matt wrote: > Hi, > > > > In the past there has been writing about Wayland’s model using implicit > synchronization of GPU renderbuffers and KMS commits [1] [2]. > > > > It would sometimes be nice to do that waiting explicitly in a compositor > before enqueueing a KMS pageflip that references a buffer who may go on > rendering for some time still. This stalls the pipeline. > > > > I’m wondering whether there’s an API -- maybe something analogous to > drmPrimeHandleToFD() – that allows userspace to fetch a waitable fence fd for > a dmabuf? A dma-buf fd itself can serve this purpose; it polls as readable when the GPU has finished drawing to the buffer. See https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1880 for how I used this to implement what you describe above in mutter. Note that this involves some Wayland state management challenges for correct operation in all cases though. -- Earthling Michel Dänzer| https://redhat.com Libre software enthusiast | Mesa and Xwayland developer