unique id for wayland objects
nancy.lan.zou at gmail.com
Mon May 21 15:06:12 UTC 2018
I debug the presentation time recently. I have checked each user call
drmAtomicCommit get the corresponding pageflip event.
But some presentation results (c2p) are a little weird. They are 1~5ms or
-1 ~ -5 ms. some results are normal(8 ~ 20 ms).
I used to explain the weird result is because pageflip return last frame's
timestamp, but I can't explain the normal results.
I suspect whether the kernel and user have the same time start point. It's
just a guess, I have no evidence.
Do you have some tips about the c2p results? Thank you.
2018-05-12 23:17 GMT+08:00 Pekka Paalanen <ppaalanen at gmail.com>:
> On Thu, 10 May 2018 12:53:49 +0800
> zou lan <nancy.lan.zou at gmail.com> wrote:
> > Hi pekka
> > I test presentation on my side. The presentation results is not accurate
> > because our pageflip event send last frame's timestamp every time. So
> > weston_output_finish_frame send last frame's present tine
> > to current frame's feedback. For this situation, I just delay one frame
> > send present event. Does this get the right c2p time? Thank you.
> Hi Nancy,
> I think your driver needs to get fixed, so that it provides the
> correct timestamp instead of an outdated one. If this is a kernel
> DRM driver, then the behaviour you describe is clearly a bug. The
> timestamp of the pageflip event signifies the time the new frame
> starts transmitting out of the connector. I believe there should be
> DRM driver API documentation stating that.
> This is also what weston_output_finish_frame() expects. Anything
> else will cause the frame scheduling to misbehave.
> It may be possible to work around the bug as you suggest, but that
> will make the compositor incorrect on any other driver, and it may
> cause other subtle breakage (the timestamp may be correct, but the
> time the event gets sent is now wrong and might even be delayed
> indefinitely if nothing causes a repaint).
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the wayland-devel