After READY -> NULL -> READY state change, the configured video
orientation didn't get applied on the new GstVaFilter instance.
Resettig prev_direction to default value in update_properties ensures
gst_va_filter_set_orientation() isn't inadvertently skipped.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8871>
Some frames are dequeued with error flag, which may cause AV unsync if decoder
does not drop them as soon as possible. So add "output-error-dequeued" and
"capture-error-dequeued" signal for v4l2 to drop such frames.
Fixes#3031
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5479>
Similar to NV12_10LE40, this is a 422 variant. This format is also named
NV20 (20bit per pixels) in other stack and is produced by rkvdec
decoder.
Co-authored-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5612>
Make sure to always update next_cluster_offset, if next cluster offset
isn't known set it to zero. If next_cluster_offfset isn't updated it will
be the same as current and if the cluster parsing fails the same cluster
will be parsed again leading to duplication of the data in the cluster.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8855>
When plugin_loader_load_and_sync returns false in plugin_loader_replay_pending,
the cur Glist l->pending_plugins will be added to the blacklist.
However, the l->pending_plugins might have already been loaded and freed in handle_rx_packet,
so causing a use-after-free issue.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8747>
The only thing that can be used in the chain function is the input segment. The
output segment might not be available at all yet or out of sync with the current
input segment.
Also because of that, the unadjusted timestamp has to be used for the
calculations as the adjustment is only part of the output segment.
This fixes the deadline calculation and the handling of force-keyunit events for
encoders using frame reordering (i.e. setting a minimum PTS).
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8842>
Calculating intersection-of-union (IoU) is a very common operation used by
tensor-decoder handling tensors from vision models. Having this in a library
will improve maintainability and ease of writing tensor-decoder.
- Post-fix _uint: We might eventually want to handle different datatype that we
woule post-fix with _type
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8805>
Due to race condition it was previously possible that
gst_element_add_pad was not completed for each RTSP stream before
signal 'no-more-pads' was emitted.
Race condition explained:
Lets say two RTSP streams are created: Video and Audio.
1. Callback new_manager_pad is called for the Video stream =>
stream->added=TRUE.
all_added=FALSE because both streams are not yet added.
Call gst_element_add_pad and emit signal 'pad-added' for Video stream.
2. Callback new_manager_pad is called for Audio stream =>
stream->added=TRUE.
all_added=TRUE because both streams are added.
Call gst_element_add_pad and emit signal 'pad-added' for Audio stream.
3. Lets say gst_element_add_pad for the audio stream completes before
the video stream. Since the audio stream already has all_added==TRUE
this will result in the signal 'no-more-pads' to be emitted before
gst_element_add_pad for the video stream is completed.
Solution is to move the logic that sets added=True and checks if all
streams are added to after gst_element_add_pad. This will make sure
signal 'no-more-pads' is not emitted until all code in
gst_element_add_pad is completed for all streams.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8832>
There was a race condition where the demuxer would seek back to beginning after
determining the duration and while that seek was in progress one pad would
attempt to push a new buffer downstream, leading to a critical warning in
gst_pad_push().
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8785>
driftsamples currently uses the requested skew directly, even if it
exceeds cexternal.
Use the approach that skew_slaving uses to fix this. As a side benefit,
this makes the custom_slaving and skew_slaving code easier to compare.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8605>
This reverts commit 8c017c79c5736c9e45e635df210e08550287646d.
1.22 was the correct pkg-config version. It's only the subproject
version that was wrong. Since we bumped libva.wrap to 2.22 version, h266
is now always available when using the subproject.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/8800>
