Three recent blog postings on DRI3000

[Maarten Lankhorst]

Op 25-07-13 03:44, Keith Packard schreef:
> Maarten Lankhorst <maarten.lankhorst at canonical.com> writes:
>
>> drm_calc_vbltimestamp_from_scanoutpos already uses CLOCK_MONOTONIC
>> for the timestamps afaict. If you round to the nearest frame I don't see how you could
>> possibly get glitches.
> Because the display clock and the time clock are not slaved to each
> other, so they will drift, possibly by quite a bit given the general
> goal of using the cheapest components that meet the timing specs
> everywhere. Plus, you'll presumably have NTP running, which will slew
> even the monotonic clock rate around.
>
> You'd have to constantly update your notion of the expected time for
> each frame or you'd be off from time to time.
It's going to be wrong in the case where userspace relies on being able to convert from 1 to the other
anyway in video applications, regardless of whether it happens in kernelspace or userspace, but this
will not be the common case.

I just want the option of being able to specify an absolute time to synchronize to for the users,
If opengl that wants to sync to a relative frame interval it can keep doing that.

~Maarten