Fix a couple of places where notify_one() was called with the lock
held. Also restartAsync doesn't need the lock for its entire duration.
This change exactly matches commit db552b0b92 ("libcamera: ipa:
raspberrypi: ALSC: Improve locking in a few places") where we do the
same for ALSC (the asynchronous thread arrangement there is identical).
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Add an empty IPA skeleton for the IPU3 pipeline. The skeleton IPA
handles the flow of parameter and statistic buffers but does not read or
write anything in the buffers. It also allows the IPA to set sensor
controls but does not implement any logic to set optimal values and
instead sets the V4L2 exposure and gain controls to max and keeps them
at that setting.
This IPA is meant as a base to allow the pipeline handler to be wired up
to an IPA. The image algorithms can then later be added to the IPA
independently from also having to add plumbing to the pipeline handler.
Signed-off-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
In order to provide an optimal split between shutter speed and gain, the
AGC must know the maximum allowable shutter speed, as limited by the
maximum frame duration (either application provided or the default).
Add a new API function, SetMaxShutter, to the AgcAlgorithm class. The
IPA provides the maximum shutter speed for AGC calculations. This
applies to both the manual and auto AGC modes.
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
The existing framerate/vblank calculations did not account for the
sensor mode limits. This commit extracts vblank limits from the sensor
v4l2 controls, and stores it in the camera modes structure.
Exposure and vblank value calculations now use values clamped to the
sensor mode limits. The libcamera::controls::FrameDurations metadata
return values now reflect the minimum and maximum after clamping.
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Users are free to avoid loading certain control algorithms from the
json tuning file if they wish, and this could mean that some libcamera
controls will therefore not be available. Currently we don't have a
good means of indicating which these are, therefore failing completely
when an application tries to use one is unhelpful - it is better just
to issue a warning.
Note that once we can indicate this properly, applications should
check for supported controls as this change may be reverted.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Add support for setting V4L2_CID_VBLANK appropriately when setting
V4L2_CID_EXPOSURE. This will allow adaptive framerates during
viewfinder use cases (e.g. when the exposure time goes above 33ms, we
can reduce the framerate to lower than 30fps).
The minimum and maximum frame durations are provided via libcamera
controls, and will prioritise exposure time limits over any AGC request.
V4L2_CID_VBLANK is controlled through the staggered writer, just like
the exposure and gain controls.
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: David Plowman <david.plowman@raspberrypi.com>
Tested-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Previously the CamHelper was returning the number of frames to drop
(on account of AGC/AWB converging). This wasn't really appropriate,
it's better for the algorithms to do it as they know how many frames
they might need.
The CamHelper::HideFramesStartup method should now just be returning
the number of frames to hide because they're bad/invalid in some way,
not worrying about the AGC/AWB. For many sensors, the correct value
for this is zero. But the ov5647 needs updating as it must return 2.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
When the AWB is started from "cold" with fixed colour gains, we try to
estimate the colour temperature this corresponds to (if a calibrated
CT curve was supplied). When fixed colour gains are set after the AWB
has been running, we leave the CT estimate alone, as the one we have
is probably sensible.
This estimated colour is passed out in the metadata for other
algorithms - notably ALSC - to use.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Add a method to the piecewise linear function (Pwl) class to compute
the inverse of a given Pwl. If the input function is non-monotonic we
can only produce a best effort "pseudo" inverse, and we signal this to
the caller.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
We add a GetConvergenceFrames method to the AwbAlgorithm class which
can be called when the AWB is started from scratch. It suggests how
many frames should be dropped before displaying any (while the AWB
converges).
The Raspberry Pi specific implementation makes this customisable from
the tuning file.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
We add a GetConvergenceFrames method to the AgcAlgorithm class which
can be called when the AGC is started from scratch. It suggests how
many frames should be dropped before displaying any (while the AGC
converges).
The Raspberry Pi specific implementation makes this customisable from
the tuning file.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
The number of frames to drop (not display) is passed back now from the
start method, not configure. This means applications have a chance to
set fixed exposure/gain before starting the camera and this can affect
the frame drop count that is returned.
Note how we need to be able to tell the very first time we start the
camera from subsequent restarts, hence addition of the "firstStart_"
flag.
Both the IPA implementation file and the pipeline handler need
matching modifications.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
This change allows controls passed into PipelineHandler::start to be
forwarded onto IPAInterface::start(). We also add a return channel if the
pipeline handler must action any of these controls, e.g. setting the
analogue gain or shutter speed in the sensor device.
The IPA interface wrapper isn't addressed as it will soon be replaced by
a new mechanism to handle IPC.
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: David Plowman <david.plowman@raspberrypi.com>
Tested-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
If the IPA fails during configuration, return an error flag to the
pipeline handler and fail the use case gracefully.
At present, the IPA configuration can fail for the following reasons:
- The sensor is not recognised, and fails to open a CamHelper object.
- The pipeline handler did not pass in controls for the ISP and sensor.
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
The exposure times in the exposure modes were causing AGC oscillations
because the algorithm was demanding long unachievable exposure times
but, without working sensor metadata, thought it was getting them when
actually it was not. We fix it by making the exposure profile request
only achievable exposure times, as we do for the ov5647 tuning.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
AGC, when paused, sets the last exposure/gain it wrote to be its
"fixed" values and will therefore continue to return them. When
resumed, we clear them so that both will float again.
This approach is better because AGC can be paused and we can
subsequently change (for example) the exposure and the gain won't
float again.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Previously we required that the sensor absolutely reaches the target
exposure, but this can fail if frame rates or analogue gains are
limited. Instead insist only that we get several frames with the same
exposure time, analogue gain and that the algorithm's target exposure
hasn't changed either.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Acked-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
When parts of an image saturate then the image brightness no longer
increases linearly with increased exposure/gain. Having calculated a
linear gain value it's better then to try it, allowing for saturating
regions, and if necessary increase the gain some more. We repeat this
several times.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Acked-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
When an application has specified fixed exposure time and/or gain they
must be programmed into the sensor immediately, even before the sensor
has been started. For this to happen they must be written into the
image metadata when the SwitchMode method is invoked.
We also make the default exposure/gain, when nothing has been set,
customisable in the tuning file.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Acked-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Previously the calculation computed Y for each region before returning
the weighted average, which "baked in" the over-importance of small
statistics regions. The revised calculation will treat all pixels
equally when the region weights are the same, making it easier to
use. With the previous scheme, proper "average" metering was difficult
to implement.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Naushir Patuck <naush@raspberrypi.com>
Acked-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
During configure() we update the ScalerCropMaximum to the correct
value for this camera mode and work out the minimum crop size allowed
by the ISP.
Whenever a new ScalerCrop request is received we check it's valid and
apply it to the ISP V4L2 device. When the IPA returns its metadata to
us we add the ScalerCrop information, rescaled to sensor native
pixels.
Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
The function gauss_seidel2_SOR() makes use of a function scoped iterator
'i', for several loops, and has a precedence of re-using the function
scoped iterator declaration in the majority of cases, except the first
where it is declared in the loop scope before the function scope, and
later which aliases a new declaration.
Re-use the existing iterator variable for consistency, and to prevent
variable aliasing.
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
