Jason Ekstrand <ja...@jlekstrand.net> writes: > Yeah, I suppose an application could ask for 10k frames in the future or > something ridiculous like that. For sync_file, people strongly want a > finite time guarantee for security/deadlock reasons. I don't know how > happy they would be with a finite time of 10 minutes...
Sure, we've put arbitrary finite limits on vblank counters in other places, so adding some kind of arbitrary limit like a couple of seconds would be entirely reasonable here. The Vulkan API doesn't need any of this complexity at all; the one I'm doing only lets you create a fence for the next vblank. > Ok, that's not expected... Part of the point of sync objects is that > they're re-usable. The client is free to not re-use them or to be very > careful about the recycling process. Heh. I thought the opposite -- lightweight objects that you'd use once and delete. I can certainly change the API to pass in an existing syncobj rather than creating a new one. That would be easier in some ways as it reduces the failure paths a bit. > Is the event the important part or the moderately accurate timestamp? I need the timestamp and sequence number, getting them in an event would mean not having to make another syscall. > We could probably modify drm_syncobj to have a "last signaled" > timestamp that's queryable through an IOCTL. That's exactly what I did, but it only works for these new syncobjs. The fields are global and could be filled in by other bits of the code. > Is the sequence number returned necessary? Will it ever be different from > the one requested? Yes, when the application queues it just slightly too late. The application doesn't actually need either of these values directly, but the system needs them to respond to requests to queue presentation at a specific time, so the vulkan driver wants 'recent' vblank time/sequence data to estimate when a vblank will occur. -- -keith
Description: PGP signature