[PATCH 1/2] pipeline: Add support for dumping capture script and metadata
Laurent Pinchart
laurent.pinchart at ideasonboard.com
Thu Oct 3 15:15:05 CEST 2024
On Thu, Oct 03, 2024 at 09:46:38AM +0900, Paul Elder wrote:
> On Thu, Oct 03, 2024 at 02:41:27AM +0300, Laurent Pinchart wrote:
> > On Wed, Oct 02, 2024 at 07:20:48PM +0900, Paul Elder wrote:
> > > On Wed, Oct 02, 2024 at 08:28:26AM +0200, Jacopo Mondi wrote:
> > > > On Tue, Oct 01, 2024 at 02:09:06AM GMT, Paul Elder wrote:
> > > > > Add support for dumping capture scripts and metadata. The capture
> > > > > scripts can then be fed into the cam application and a capture can thus
> > > > > be "replayed". Metadata can also be dumped.
> > > > >
> > > > > Camera configuration is also dumped to the capture script. The cam
> > > > > application currently does not support loading configuration from the
> > > > > capture script, but support for that will be added in a subsequent
> > > > > patch.
> > > > >
> > > > > These can be enabled by a new environment variable.
> > > >
> > > > Which needs to be documented :)
> > >
> > > I... opened the tab ready to write it then forgot to write it :)
> > >
> > > > > Signed-off-by: Paul Elder <paul.elder at ideasonboard.com>
> > > > > ---
> > > > > include/libcamera/internal/camera.h | 3 +
> > > > > include/libcamera/internal/pipeline_handler.h | 16 ++++
> > > > > src/libcamera/camera.cpp | 13 +++
> > > > > src/libcamera/pipeline_handler.cpp | 93 ++++++++++++++++++-
> > > > > 4 files changed, 124 insertions(+), 1 deletion(-)
> > > > >
> > > > > diff --git a/include/libcamera/internal/camera.h b/include/libcamera/internal/camera.h
> > > > > index 0add0428bb5d..a42f03d4c755 100644
> > > > > --- a/include/libcamera/internal/camera.h
> > > > > +++ b/include/libcamera/internal/camera.h
> > > > > @@ -19,6 +19,8 @@
> > > > >
> > > > > namespace libcamera {
> > > > >
> > > > > +enum class Orientation;
> >
> > You should include orientation.h, the compiler will need to to determine
> > the size of the enum, to determine the layout of the Camera class. It's
> > included indirectly at the moment, which explains why you don't get a
> > compilation error.
> >
> > > > > +
> > > > > class CameraControlValidator;
> > > > > class PipelineHandler;
> > > > > class Stream;
> > > > > @@ -65,6 +67,7 @@ private:
> > > > > std::string id_;
> > > > > std::set<Stream *> streams_;
> > > > > std::set<const Stream *> activeStreams_;
> > > > > + Orientation orientation_;
> > > > >
> > > > > bool disconnected_;
> > > > > std::atomic<State> state_;
> > > > > diff --git a/include/libcamera/internal/pipeline_handler.h b/include/libcamera/internal/pipeline_handler.h
> > > > > index 0d38080369c5..fb3914185a01 100644
> > > > > --- a/include/libcamera/internal/pipeline_handler.h
> > > > > +++ b/include/libcamera/internal/pipeline_handler.h
> > > > > @@ -9,6 +9,7 @@
> > > > >
> > > > > #include <memory>
> > > > > #include <queue>
> > > > > +#include <set>
> > > > > #include <string>
> > > > > #include <sys/types.h>
> > > > > #include <vector>
> > > > > @@ -20,6 +21,8 @@
> > > > >
> > > > > namespace libcamera {
> > > > >
> > > > > +enum class Orientation;
> > > > > +
> > > > > class Camera;
> > > > > class CameraConfiguration;
> > > > > class CameraManager;
> > > > > @@ -68,6 +71,9 @@ public:
> > > > >
> > > > > CameraManager *cameraManager() const { return manager_; }
> > > > >
> > > > > + void dumpConfiguration(const std::set<const Stream *> &streams,
> > > > > + const Orientation &orientation);
> >
> > Pass it by value, it's an enum.
> >
> > > > > +
> > > >
> > > > To define the expected configuration format, I would make 2/2 precede 1/2
> > >
> > > Indeed 2/2 going first doesn't actually break anything. Sure.
> > >
> > > > > protected:
> > > > > void registerCamera(std::shared_ptr<Camera> camera);
> > > > > void hotplugMediaDevice(MediaDevice *media);
> > > > > @@ -81,6 +87,11 @@ protected:
> > > > > CameraManager *manager_;
> > > > >
> > > > > private:
> > > > > + enum DumpMode {
> >
> > Make it an enum class, you already prefix the enumerators with the enum
> > name when you use them.
> >
> > > > > + Controls,
> > > > > + Metadata,
> > > > > + };
> > > > > +
> > > > > void unlockMediaDevices();
> > > > >
> > > > > void mediaDeviceDisconnected(MediaDevice *media);
> > > > > @@ -89,6 +100,8 @@ private:
> > > > > void doQueueRequest(Request *request);
> > > > > void doQueueRequests();
> > > > >
> > > > > + void dumpRequest(Request *request, DumpMode mode);
> > > > > +
> > > > > std::vector<std::shared_ptr<MediaDevice>> mediaDevices_;
> > > > > std::vector<std::weak_ptr<Camera>> cameras_;
> > > > >
> > > > > @@ -97,6 +110,9 @@ private:
> > > > > const char *name_;
> > > > > unsigned int useCount_;
> > > > >
> > > > > + std::ostream *dumpCaptureScript_;
> > > > > + std::ostream *dumpMetadata_;
> > > > > +
> > > > > friend class PipelineHandlerFactoryBase;
> > > > > };
> > > > >
> > > > > diff --git a/src/libcamera/camera.cpp b/src/libcamera/camera.cpp
> > > > > index a86f552a47bc..1282f99d839b 100644
> > > > > --- a/src/libcamera/camera.cpp
> > > > > +++ b/src/libcamera/camera.cpp
> > > > > @@ -1215,6 +1215,9 @@ int Camera::configure(CameraConfiguration *config)
> > > > > d->activeStreams_.insert(stream);
> > > > > }
> > > > >
> > > > > + /* TODO Save sensor configuration for dumping it to capture script */
> > > > > + d->orientation_ = config->orientation;
> > > > > +
> > > > > d->setState(Private::CameraConfigured);
> > > > >
> > > > > return 0;
> > > > > @@ -1356,6 +1359,16 @@ int Camera::start(const ControlList *controls)
> > > > >
> > > > > ASSERT(d->requestSequence_ == 0);
> > > > >
> > > > > + /*
> > > > > + * Invoke method in blocking mode to avoid the risk of writing after
> > > > > + * streaming has started.
> > > > > + * This needs to be here as PipelineHandler::start is a virtual function
> > > > > + * so it is impractical to add the dumping there.
> > > > > + * TODO Pass the sensor configuration, once it is supported
> > > > > + */
> > > > > + d->pipe_->invokeMethod(&PipelineHandler::dumpConfiguration,
> > > > > + ConnectionTypeBlocking, d->activeStreams_, d->orientation_);
> > > >
> > > > Can't you pass in the CameraConfiguration which containts the
> > > > StreamConfigurations and the the Orientation already ?
> > >
> > > CameraConfiguration can't be copied because it has virtual functions.
> > > And CameraConfiguration is only available in Camera::configure(). So, no.
> > >
> > > >
> > > > > +
> > > > > ret = d->pipe_->invokeMethod(&PipelineHandler::start,
> > > > > ConnectionTypeBlocking, this, controls);
> > > > > if (ret)
> > > > > diff --git a/src/libcamera/pipeline_handler.cpp b/src/libcamera/pipeline_handler.cpp
> > > > > index e5940469127e..7002b4323bdd 100644
> > > > > --- a/src/libcamera/pipeline_handler.cpp
> > > > > +++ b/src/libcamera/pipeline_handler.cpp
> > > > > @@ -8,6 +8,7 @@
> > > > > #include "libcamera/internal/pipeline_handler.h"
> > > > >
> > > > > #include <chrono>
> > > > > +#include <fstream>
> > > > > #include <sys/stat.h>
> > > > > #include <sys/sysmacros.h>
> > > > >
> > > > > @@ -68,14 +69,36 @@ LOG_DEFINE_CATEGORY(Pipeline)
> > > > > * through the PipelineHandlerFactoryBase::create() function.
> > > > > */
> > > > > PipelineHandler::PipelineHandler(CameraManager *manager)
> > > > > - : manager_(manager), useCount_(0)
> > > > > + : manager_(manager), useCount_(0),
> > > > > + dumpCaptureScript_(nullptr), dumpMetadata_(nullptr)
> > > > > {
> > > > > + /* TODO Print notification that we're dumping capture script */
> >
> > \todo
> >
> > or better, address it.
> >
> > > > > + const char *file = utils::secure_getenv("LIBCAMERA_DUMP_CAPTURE_SCRIPT");
> > > > > + if (!file)
> > > > > + return;
> > > > > +
> > > > > + dumpCaptureScript_ = new std::ofstream(file);
> > > >
> > > > I was really expecting to be operating on File instances
> > >
> > > Oh ig we could do that too. I copied a few ideas from the logger.
> > >
> > > > > +
> > > > > + /*
> > > > > + * Metadata needs to go into a separate file because otherwise it'll
> > > > > + * flood the capture script
> > > > > + */
> > > > > + dumpMetadata_ = new std::ofstream(std::string(file) + ".metadata");
> > > > > + std::string str = "frames:\n";
> > > > > + dumpMetadata_->write(str.c_str(), str.size());
> > > >
> > > > and use the libyaml emitter for this instead of raw writes
> > > >
> > > > Have you considered it and decided it was better to use raw writes ?
> > >
> > > No I hadn't. I'll keep it raw for now though until you make the
> > > YamlEmitter :)
> > >
> > > > > + dumpMetadata_->flush();
> > > > > }
> > > > >
> > > > > PipelineHandler::~PipelineHandler()
> > > > > {
> > > > > for (std::shared_ptr<MediaDevice> media : mediaDevices_)
> > > > > media->release();
> > > > > +
> > > > > + if (dumpCaptureScript_)
> > > > > + delete dumpCaptureScript_;
> > > > > +
> > > > > + if (dumpMetadata_)
> > > > > + delete dumpMetadata_;
> > > > > }
> > > > >
> > > > > /**
> > > > > @@ -464,6 +487,8 @@ void PipelineHandler::doQueueRequest(Request *request)
> > > > >
> > > > > request->_d()->sequence_ = data->requestSequence_++;
> > > > >
> > > > > + dumpRequest(request, DumpMode::Controls);
> > > > > +
> > > > > if (request->_d()->cancelled_) {
> > > > > completeRequest(request);
> > > > > return;
> > > > > @@ -555,6 +580,8 @@ void PipelineHandler::completeRequest(Request *request)
> > > > >
> > > > > request->_d()->complete();
> > > > >
> > > > > + dumpRequest(request, DumpMode::Metadata);
> > > > > +
> > > > > Camera::Private *data = camera->_d();
> > > > >
> > > > > while (!data->queuedRequests_.empty()) {
> > > > > @@ -758,6 +785,70 @@ void PipelineHandler::disconnect()
> > > > > * \return The CameraManager for this pipeline handler
> > > > > */
> > > > >
> > > > > +void PipelineHandler::dumpConfiguration(const std::set<const Stream *> &streams,
> > > > > + const Orientation &orientation)
> > > > > +{
> > > > > + if (!dumpCaptureScript_)
> > > > > + return;
> > > > > +
> > > > > + std::stringstream ss;
> > > > > + ss << "configuration:" << std::endl;
> > > > > + ss << " orientation: " << orientation << std::endl;
> > > > > +
> > > > > + /* TODO Dump Sensor configuration */
> >
> > \todo
> >
> > Same below.
> >
> > > > > +
> > > > > + ss << " streams:" << std::endl;
> > > > > + for (const auto &stream : streams) {
> > > > > + const StreamConfiguration &streamConfig = stream->configuration();
> > > > > + ss << " - pixelFormat: " << streamConfig.pixelFormat << std::endl;
> > > > > + ss << " size: " << streamConfig.size << std::endl;
> > > > > + ss << " stride: " << streamConfig.stride << std::endl;
> > > > > + ss << " frameSize: " << streamConfig.frameSize << std::endl;
> > > > > + ss << " bufferCount: " << streamConfig.bufferCount << std::endl;
> > > > > + if (streamConfig.colorSpace)
> > > > > + ss << " colorSpace: " << streamConfig.colorSpace->toString() << std::endl;
> > > > > + }
> > > > > +
> > > > > + dumpCaptureScript_->write(ss.str().c_str(), ss.str().size());
> >
> > I wonder if you could write
> >
> > *dumpCaptureScript_ << ss.rdbuf();
> >
> > But you could use dumpCaptureScript_ directly with the << operator
> > above, no need for an intermediate stringstream.
> >
>
> ack
>
> > > > > +
> > > > > + std::string str = "frames:\n";
> > > > > + dumpCaptureScript_->write(str.c_str(), str.size());
> >
> > *dumpCaptureScript_ << "frames:\n";
> >
> > > > > + dumpCaptureScript_->flush();
> > > > > +}
> > > > > +
> > > > > +void PipelineHandler::dumpRequest(Request *request, DumpMode mode)
> > > > > +{
> > > > > + ControlList &controls =
> >
> > const
> >
> > > > > + mode == DumpMode::Controls ? request->controls()
> > > > > + : request->metadata();
> > > > > + std::ostream *output =
> > > > > + mode == DumpMode::Controls ? dumpCaptureScript_
> > > > > + : dumpMetadata_;
> > > > > +
> > > > > + if (!output || controls.empty())
> > > > > + return;
> > > > > +
> > > > > + std::stringstream ss;
> >
> > Use output directly, don't create an intermediate stringstream.
> >
> > > > > + /* TODO Figure out PFC */
> > > > > + ss << " - " << request->sequence() << ":" << std::endl;
> > > > > +
> > > > > + const ControlIdMap *idMap = controls.idMap();
> > > > > + for (const auto &pair : controls) {
> > > > > + const ControlId *ctrlId = idMap->at(pair.first);
> > > > > + /* TODO Prettify enums (probably by upgrading ControlValue::toString()) */
> >
> > Could you elaborate on this for my information ?
>
> This will print enums just as numbers, so you'd get "AeMeteringMode: 0"
> instead of "AeMeteringMode: MeteringCentreWeighted".
The latter would be more readable indeed. At the cost of more CPU cycles
to output the value, as well as to parse the capture script. We can't
have our cake and eat it :(
> > > > > + ss << " " << ctrlId->name() << ": " << pair.second.toString() << std::endl;
> > > > > + }
> > > > > +
> > > > > + /*
> > > > > + * TODO Investigate the overhead of flushing this frequently
> > > > > + * Controls aren't going to be queued too frequently so it should be
> >
> > I'm not too sure about that, I think we'll see some types of
> > applications (or frameworks) setting controls in every request. We need
> > to be ready for it.
>
> Good point. Although I'm not sure how else to determine how often to
> flush. Would it be good enough to add another environment variable to
> set every how many frames to dump? And default to a value > 1? Or count
> how often it's getting queued and adjust the dump rate dynamically (with
> an environment variable to override to "dump every frame" for debugging
> purposes)?
Let's not make it overly complicated, you can start with a constant
flush interval expressed as a number of requests. Store it in a static
constexpr, and set it to 1 to start with.
> In either case ig it's time to investigate the actual overhead :/
Having numbers will certainly help discussions. No need to get the
numbers to post v2 though.
> > > > > + * fine to dump controls every frame. Metadata on the other hand needs
> > > > > + * to be investigated.
> > > > > + */
> > > > > + output->write(ss.str().c_str(), ss.str().size());
> > > > > + output->flush();
> > > > > +}
> > > > > +
> > > > > /**
> > > > > * \class PipelineHandlerFactoryBase
> > > > > * \brief Base class for pipeline handler factories
--
Regards,
Laurent Pinchart
More information about the libcamera-devel
mailing list