[libcamera-devel] [PATCH v10 1/3] libcamera: controls: Add frame duration control

[Laurent Pinchart]

Hi Naush,

On Mon, Dec 21, 2020 at 11:11:59AM +0000, Naushir Patuck wrote:
> On Sun, 20 Dec 2020 at 05:38, Laurent Pinchart wrote:
> > On Fri, Dec 18, 2020 at 10:06:25AM +0000, Naushir Patuck wrote:
> > > Add an int64_t array control (controls::FrameDurations) to specify the
> > > minimum and maximum (in that order) frame duration to be used by the
> > > camera sensor.
> > >
> > > Signed-off-by: Naushir Patuck <naush at raspberrypi.com>
> > > Reviewed-by: David Plowman <david.plowman at raspberrypi.com>
> > > Tested-by: David Plowman <david.plowman at raspberrypi.com>
> > > Reviewed-by: Jacopo Mondi <jacopo at jmondi.org>
> > > ---
> > >  src/libcamera/control_ids.yaml | 41 ++++++++++++++++++++++++++++++++++
> > >  1 file changed, 41 insertions(+)
> > >
> > > diff --git a/src/libcamera/control_ids.yaml b/src/libcamera/control_ids.yaml
> > > index 6d6f0fee..a58bff18 100644
> > > --- a/src/libcamera/control_ids.yaml
> > > +++ b/src/libcamera/control_ids.yaml
> > > @@ -306,6 +306,47 @@ controls:
> > >          maximum valid value is given by the properties::ScalerCropMaximum
> > >          property, and the two can be used to implement digital zoom.
> > >
> > > +  - FrameDurations:
> > > +      type: int64_t
> > > +      description: |
> > > +          The minimum and maximum (in that order) frame duration,
> > > +          expressed in micro-seconds.
> > > +
> > > +          When provided by applications, the control specifies the sensor frame
> > > +          duration interval the pipeline has to use. This could also limit the
> >
> > s/could also limit/limits/ ?
> 
> Ack
> 
> > > +          largest exposure time the sensor can use. For example, if a maximum
> > > +          frame duration of 33ms is requested (corresponding to 30 frames per
> > > +          second), the sensor will not be able to raise the exposure time above
> > > +          33ms. A fixed frame duration is achieved by setting the minimum and
> > > +          maximum values to be the same.
> > > +
> > > +          The maximum frame duration provides the absolute limit to the shutter
> > > +          speed computed by the AE algorithm and it overrides any exposure mode
> >
> > I think we should talk about exposure time instead of shutter speed, but
> > we'll need to go through the whole documentation and code base to ensure
> > consistency, so this can be done later.
> >
> > > +          setting specified with controls::AeExposureMode. Similarly, when a
> > > +          manual exposure time is set through controls::ExposureTime, it also
> > > +          gets clipped to the limits set by this control.
> > > +
> > > +          \sa AeExposureMode
> > > +          \sa ExposureTime
> > > +
> > > +          \todo Refer to the frame duration limits property to describe how
> > > +          application-provided values gets clipped and reset.
> > > +
> > > +          When reported by pipelines, the control expresses the minimum and
> >
> > Is this about Request::metadata(), or Camera::properties() ? I assume
> > the former as it's defined in control_ids.yaml.
> 
> This is through Request::metadata().  Should I perhaps reword to make this
> more explicit?
> 
> "When reported by pipelines through metadata, the control expresses...." ?

You could just say "When reported in metadata, ...".

> > > +          maximum frame durations used after being clipped to what the current
> > > +          sensor mode supports, and what is achievable based on the exposure
> > > +          mode setting specified with controls::AeExposureMode or manual
> > > +          exposure time set through controls::ExposureTime. The sensor frame
> >
> > I have a hard time parsing this, and a hard time seeing how it should be
> > used by applications :-S Do you need this for your use cases ?
> 
> Yes, this was also a point of confusion between Jacopo and me :-)
> 
> What I was trying to convey is that the values returned through the
> metadata is the actual limits that will be used (after being constrained by
> the exposure mode and/or shutter speed), and may not be the same as what
> the application requested.  I thought this might be useful for the
> application to know the actual limits used.  If you think this may not be
> needed, I could remove this control from the metadata?

Now I understand. Sorry for being slow :-)

One thing we should report in metadata for sure (but not necessarily as
part of this series) is the actual frame duration. I've so far thought
it could be done with this control, with the minimum and maximum values
set to the actual frame duration. That would however preclude from using
the control to report clamped values. We could introduce a new
FrameDuration control used in metadata only for this purpose, but it
could be a bit confusing when setting a particular frame duration to set
it through FrameDurations and read it back through FrameDuration.

I see pros and cons in both cases, even if I don't see at the moment use
cases for reporting the constrained limits. If you think it would be
useful, I'm fine keeping this usage, if possible with a slightly
clarified documentation. Maybe this paragraph could be reworked to first
explain that the current configuration of the camera sensor and of the
algorithms restrict possible frame durations, followed by the sentence
stating that, when used in metadata, the control reports the limits used
by the camera after those constraints are taken into account ?

> > > +          duration is one of the parameter that defines the capture frame rate
> > > +          but it does not alone provide enough information to fully calculate it
> > > +          as it does not account for pipeline processing delays.
> >
> > Delays in the pipeline don't affect the frame duration, do they ? We
> > could drop frames when processing streams in software (for instance for
> > JPEG compression), but that's out of scope as libcamera doesn't support
> > that.
> 
> If you have, e.g. a slow software denoise running in the IPA whose
> processing time limits your frame throughput, it could be limiting the
> overall pipeline framerate (increasing frame duration).  Do you think
> perhaps this should not be mentioned here to avoid complicating things for
> now?

While you're right, I'd drop it for now. We'll need to document how
certain processing blocks may not have any bandwidth to process all
frames at the sensor frame rate, but before we do so, I think partial
documentation would be confusing. The todo comment should be enough for
now.

> > > +
> > > +          \todo Define how to calculate the capture frame rate by
> > > +          defining controls to report additional delays introduced by
> > > +          the capture pipeline or post-processing stages (ie JPEG
> > > +          conversion, frame scaling).
> > > +      size: [2]
> > > +
> > >    # ----------------------------------------------------------------------------
> > >    # Draft controls section
> > >

-- 
Regards,

Laurent Pinchart