@@ -40,7 +40,6 @@ namespace libcamera {
LOG_DEFINE_CATEGORY(RkISP1)
class PipelineHandlerRkISP1;
-class RkISP1ActionQueueBuffers;
class RkISP1CameraData;
enum RkISP1ActionType {
@@ -203,7 +202,6 @@ private:
PipelineHandler::cameraData(camera));
}
- friend RkISP1ActionQueueBuffers;
friend RkISP1CameraData;
friend RkISP1Frames;
@@ -214,6 +212,7 @@ private:
void bufferReady(FrameBuffer *buffer);
void paramReady(FrameBuffer *buffer);
void statReady(FrameBuffer *buffer);
+ void frameStart(uint32_t sequence);
int allocateBuffers(Camera *camera);
int freeBuffers(Camera *camera);
@@ -348,53 +347,6 @@ RkISP1FrameInfo *RkISP1Frames::find(Request *request)
return nullptr;
}
-class RkISP1ActionQueueBuffers : public FrameAction
-{
-public:
- RkISP1ActionQueueBuffers(unsigned int frame, RkISP1CameraData *data,
- PipelineHandlerRkISP1 *pipe)
- : FrameAction(frame, QueueBuffers), data_(data), pipe_(pipe)
- {
- }
-
-protected:
- void run() override
- {
- RkISP1FrameInfo *info = data_->frameInfo_.find(frame());
- if (!info)
- LOG(RkISP1, Fatal) << "Frame not known";
-
- /*
- * \todo: If parameters are not filled a better method to handle
- * the situation than queuing a buffer with unknown content
- * should be used.
- *
- * It seems excessive to keep an internal zeroed scratch
- * parameters buffer around as this should not happen unless the
- * devices is under too much load. Perhaps failing the request
- * and returning it to the application with an error code is
- * better than queue it to hardware?
- */
- if (!info->paramFilled)
- LOG(RkISP1, Error)
- << "Parameters not ready on time for frame "
- << frame();
-
- pipe_->param_->queueBuffer(info->paramBuffer);
- pipe_->stat_->queueBuffer(info->statBuffer);
-
- if (info->mainPathBuffer)
- pipe_->mainPath_.queueBuffer(info->mainPathBuffer);
-
- if (info->selfPathBuffer)
- pipe_->selfPath_.queueBuffer(info->selfPathBuffer);
- }
-
-private:
- RkISP1CameraData *data_;
- PipelineHandlerRkISP1 *pipe_;
-};
-
int RkISP1CameraData::loadIPA()
{
ipa_ = IPAManager::createIPA(pipe_, 1, 1);
@@ -948,9 +900,14 @@ int PipelineHandlerRkISP1::queueRequestDevice(Camera *camera, Request *request)
op.controls = { request->controls() };
data->ipa_->processEvent(op);
- data->timeline_.scheduleAction(std::make_unique<RkISP1ActionQueueBuffers>(data->frame_,
- data,
- this));
+ param_->queueBuffer(info->paramBuffer);
+ stat_->queueBuffer(info->statBuffer);
+
+ if (info->mainPathBuffer)
+ mainPath_.queueBuffer(info->mainPathBuffer);
+
+ if (info->selfPathBuffer)
+ selfPath_.queueBuffer(info->selfPathBuffer);
data->frame_++;
@@ -1088,6 +1045,7 @@ bool PipelineHandlerRkISP1::match(DeviceEnumerator *enumerator)
mainPath_.bufferReady().connect(this, &PipelineHandlerRkISP1::bufferReady);
selfPath_.bufferReady().connect(this, &PipelineHandlerRkISP1::bufferReady);
stat_->bufferReady.connect(this, &PipelineHandlerRkISP1::statReady);
+ isp_->frameStart.connect(this, &PipelineHandlerRkISP1::frameStart);
param_->bufferReady.connect(this, &PipelineHandlerRkISP1::paramReady);
/*
@@ -1167,8 +1125,6 @@ void PipelineHandlerRkISP1::statReady(FrameBuffer *buffer)
if (!info)
return;
- data->timeline_.bufferReady(buffer);
-
if (data->frame_ <= buffer->metadata().sequence)
data->frame_ = buffer->metadata().sequence + 1;
@@ -1178,6 +1134,20 @@ void PipelineHandlerRkISP1::statReady(FrameBuffer *buffer)
data->ipa_->processEvent(op);
}
+void PipelineHandlerRkISP1::frameStart(uint32_t sequence)
+{
+ if (!activeCamera_)
+ return;
+
+ RkISP1CameraData *data = cameraData(activeCamera_);
+
+ RkISP1FrameInfo *info = data->frameInfo_.find(sequence);
+ if (!info || !info->paramFilled)
+ LOG(RkISP1, Info)
+ << "Parameters not ready on time for frame "
+ << sequence;
+}
+
REGISTER_PIPELINE_HANDLER(PipelineHandlerRkISP1)
} /* namespace libcamera */
In the Timeline approach the idea was to delay queuing buffers to the device until the IPA had a chance to prepare the parameters buffer. A check was still added to warn if the IPA queued buffers before the parameters buffer was filled in. This check happened much sooner then needed as the parameter buffers does not have to be ready when the buffer is queued but just before its consumed. As the pipeline now has true knowledge of each SOF we can move the check there and remove the delaying of queuing of buffers. This change really speeds up the IPA reactions as the delays used in the Timeline where approximated while with this change they are driven by events reported by the device. Signed-off-by: Niklas Söderlund <niklas.soderlund@ragnatech.se> --- src/libcamera/pipeline/rkisp1/rkisp1.cpp | 78 ++++++++---------------- 1 file changed, 24 insertions(+), 54 deletions(-)