This element refactors functionality from gstonnxinference element,
namely separating out the ONNX inference from the subsequent analysis.
The new element runs an ONNX model on each video frame, and then
attaches a TensorMeta meta with the output tensor data. This tensor data
will then be consumed by downstream elements such as gstobjectdetector.
At the moment, a provisional TensorMeta is used just in the ONNX
plugin, but in future this will upgraded to a GStreamer API for other
plugins to consume.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4916>
- Don't try to make the parameters match `GHFunc`. Use a dedicated
callback for `g_hash_table_foreach`.
- Don't try to be clever with buffer memories. We're allocating a full
packet anyway, might as well memcpy and save on a lot of complexity.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5496>
After rendering a QML scene the qmlglsrc element copies the contents of
the scene to a GStreamer buffer. This happens on the Qt render thread.
Then it attaches a sync point to the destination buffer. This sync point
must be awaited by other threads which use the buffer later on. The
current implementation relies on the downstream elements to wait for the
sync point. However, there are situation where this does not work. The
GstBaseTransform e.g. copies the buffer metadata (which overwrites the
sync point without waiting for it) *before* waiting for the sync point.
This commit waits for the sync point inside the qmlglsrc element before
sending it downstream. The wait command is issued on the streaming
thread with the pipeline OpenGL context, i.e. it will synchronize with
the GStreamer OpenGL thread.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5506>
There are a bunch of plugins that you need for webrtc support, and
it's not obvious at all to users which those are.
With this commit, srtp, sctp and dtls options will be auto-enabled if
the webrtc option is enabled.
Requires meson 1.1
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5505>
In _gl_memory_upload_propose_allocation(), when output target is "external-oes",
then we should not provide GL allocator and pool in the allocation query.
This is because the "external-oes" kind memory can never be mapped directly
and the upstream element may misuse it.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5468>
The propose and decide allocation vfuncs are called directly from
basetransform and need to use the locked accessor function for
retrieving a reliable reference to the GstGLContext (if available)
Fixes spurious crashes on shutdown during pad reconfiguration
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5437>
The v4l2codecs H.265 decoder uses the
GstH265SliceHdr::entry_point_offset_minus1 array so make sure that it is not
freed before decoding the frame.
Before this patch, some H.265 input would segfault in
gst_v4l2_codec_h265_dec_fill_slice_params() when executing the line:
guint32 entry_point_offset = slice_hdr->entry_point_offset_minus1[i] + 1;
Make sure that the array is not freed before using it.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5499>
The number of planes is a meta we carry around in the GstVideoMeta with
DMA_DRM format. In cannot be decuded correctly from knowledge of the
base format. Notably, some compression modifier may introduce an extra
plane to store the compression parameters.
So use n_planes from GstVideoMeta and pass this explicitly when
importing to EGLImage.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5461>
The DMAbuf accept function was ensuring the in_dma_info values was valid if
the in_caps have change. But the check was bogus since the in_caps was being
modified without a pointer change. As a side effect, on the second accept
call, the drm_fourcc was reset to 0, which cause the uploader to fallback.
Fix this by ensuring we always have a valid dma_frm info directly in the
set_caps() function. Also remove the bogus caps changed check and remove any
modification to the info structure and always do that inner checks.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5461>
DRM Modifiers are not generically transferrable from a format like NV12 to
their indirect shading format (R8 / RG88). So the helper to this do needs
to be removed from our API.
To make things worse, we support indirect formats that aren't DRM format in
the first place. Notably NV12_16L32 (aka MM21) is not (yet) a DRM format. Yet,
each plane can be indirectly imported using R8/RG88 and a detiling shader.
This patch also removes this constraint restoring zero-copy playback on
Mediatek SoC.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5461>
With one regular image file path provided (without %05d),
the element was stuck in a dead loop counting the frames:
gst_image_sequence_src_count_frames
This allows to display any image file out of the element
for a given number of buffers.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5471>
Update connection-speed at runtime in playbin, uridecodebin and decodebin
also do the same thing in urisourcebin.
With contributions from Philippe Normand <philn@igalia.com> (build fixes and
rebase on mono-repo).
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4713>
If the v4l2videoenc receives an QUERY_ALLOCATION, it must not propose a
currently used pool, because it cannot be sure that the allocation query came
from exactly the same upstream element. The QUERY_ALLOCATION will not contain
the internal OUTPUT pool.
The upstream element (the basesrc) detects that the newly proposed pool differs
from the old pool. It deactivates the old pool and switches to the new pool.
If there was a format change, a new OUTPUT buffer pool will be allocated in
gst_v4l2_object_set_format_full() and the CAPTURE task will be stopped to switch
the format. If there hasn't been a format change,
gst_v4l2_object_set_format_full() will not be called. The old pool will be kept
and reused.
Without a format change, the processing task continues running.
This leads to the situation that the processing task is running, but the OUTPUT
buffer pool (the old pool) is deactivated. Therefore, the encoder is not able to
get buffers from the OUTPUT pool and encoding cannot continue.
This situation can be triggered by sending a RECONFIGURE event without a format
change.
Resolve this situation by ensuring that the OUTPUT buffer pool is always
activated when frames arrive at the encoder.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4235>
There is a CAPTURE pool in the same function. While the CAPTURE pool is called
cpool, using pool for the OUTPUT pool is confusing.
Using opool for the OUTPUT pool makes it more obvious, which pool is used.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4235>
We were already converting the pad last timestamp to running time but
not the segment position.
This segment position is used by gst_aggregator_simple_get_next_time()
to compute the waiting time when aggregating.
Those waiting times were wrong in my live pipeline using the system
clock, resulting in the aggregator to never wait at all.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5460>
While adding arbitrary tile support, a round up operation was badly
converter. This caused the Y component of the stride to be 0. This
eventually lead to a crash in glupoad preceded by the following
assertion.
gst_gl_buffer_allocation_params_new: assertion 'alloc_size > 0' failed
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5458>
While adding arbitrary tile support, a round up operation was badly
converter. This caused the Y component of the stride to be 0. This
eventually lead to a crash in glupoad preceded by the following
assertion.
gst_gl_buffer_allocation_params_new: assertion 'alloc_size > 0' failed
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5458>
Do not update timelevel on segment. Segment itself does not tell
anything about the amount of buffered time duration in the element
but buffer timestamp/duration is required to measure actual bufferred time.
Moreover, at the time when new segment is applied to sink/srcpad,
segment.position would point to random value.
Therefore calculating running time using the random value does not
make sense and it will result in wrong timelevel report.
This patch updates queue/queue2's timelevel measuring logic so that
it can be updated only on buffer/buffer-list/gap-event flow.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5430>
This was causing a memory leak in cases like `gltestsrc ! gltransformation scale-x=0.5 ! glimagesink`.
Parent meta was being added in assumption that those buffers are different, which was not the case here,
creating a reference loop and never freeing the buffer.
Co-authored-by: Matthew Waters <matthew@centricular.com>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5445>
The caps that were sent by the caps event can be retrieved from the sinkpad
using gst_pad_get_current_caps(). This is more reliable than using cur_caps as
we know exactly which caps upstream selected when the UVC host didn't select a
format, yet.
This further allows to simplify the check, if the uvcsink has to wait for the
caps event before switching to the internal v4l2sink.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4994>
The probe passes all events except the EVENT_CAPS. Installing and removing the
probe doesn't provide any additional value.
Install an event function and always handle EVENT_CAPS. Use the caps_changed
field, to decide, if the element has to do anything special on a EVENT_CAPS.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4994>
Move the sanity checks to the beginning of the function. Make the actual effect
of the function more obvious and reset the flags in the end.
This should make it easier to understand what this function is doing.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4994>
The probe that installs the buffer probe is already on the correct pad. There is
no need for a separate function to install the probe.
While at it, change the signature of the probe functions to GstPadProbeCallback
to avoid the cast when installing the probes.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4994>
The uvcsink calculates the caps for the format that the UVC host selected. The
gst_uvc_sink_parse_cur_caps() sets these caps as cur_caps as a side effect. This
behavior is surprising as cur_caps is later updated to reflect the actually used
caps.
Just return the configured caps to avoid side effects. This makes the function
easier to understand. Update the function name to reflect the new behavior.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4994>
The only job of the event peer probe is to catch the upcoming caps event
and be able to react with the sink change. All other events that are
passing the pad shall be passed and ignored.
Since the probe is a blocking probe, there is no use in returning
with GST_PAD_PROBE_OK on other events. Otherwise the event would just
be blocked.
Since we are handling the probe removal of the probe already in the
event switch, we can remove the second explicit probe removal.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4994>
There is a race condition where transfer has not been submitted yet while the
request is cancelled which leads to the transfer state going back to
`DOWNLOAD_REQUEST_STATE_OPEN` and the user of the request to get signalled about
its completion (and the task actually happening after it was cancelled) leading
to assertions and misbehaviours.
To ensure that this race can't happen, we start differentiating between the
UNSENT and CANCELLED states as in the normal case, when entering `submit_request`
the state is UNSENT and at that point we need to know that it is not because
the request has been cancelled.
In practice this case lead to an assertion in
`gst_adaptive_demux2_stream_begin_download_uri` because in a previous call to
`gst_adaptive_demux2_stream_stop_default` we cancelled the previous request and
setup a new one while it had not been submitted yet and then got a `on_download_complete`
callback called from that previous cancelled request and then we tried to do
`download_request_set_uri` on a request that was still `in_use`, leading to
something like:
```
#0: 0x0000000186655ec8 g_assert (request->in_use == FALSE)assert.c:0
#1: 0x00000001127236b8 libgstadaptivedemux2.dylib`download_request_set_uri(request=0x000060000017cc00, uri="https://XXX/chunk-stream1-00002.webm", range_start=0, range_end=-1) at downloadrequest.c:361
#2: 0x000000011271cee8 libgstadaptivedemux2.dylib`gst_adaptive_demux2_stream_begin_download_uri(stream=0x00000001330f1800, uri="https://XXX/chunk-stream1-00002.webm", start=0, end=-1) at gstadaptivedemux-stream.c:1447
#3: 0x0000000112719898 libgstadaptivedemux2.dylib`gst_adaptive_demux2_stream_load_a_fragment [inlined] gst_adaptive_demux2_stream_download_fragment(stream=0x00000001330f1800) at gstadaptivedemux-stream.c:0
#4: 0x00000001127197f8 libgstadaptivedemux2.dylib`gst_adaptive_demux2_stream_load_a_fragment(stream=0x00000001330f1800) at gstadaptivedemux-stream.c:1969
#5: 0x000000011271c2a4 libgstadaptivedemux2.dylib`gst_adaptive_demux2_stream_next_download(stream=0x00000001330f1800) at gstadaptivedemux-stream.c:2112
```
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5435>
Back in the mists of time[1], we switched `giostream*` elements to not close the
stream on stop() so that applications that needed a handle to the stream after
the element stopped had it.
Unfortunately, we also have cases[2] where waiting for the element to be
finalized is too late for the stream to be closed.
In order to not change the behaviour of the element, we add a property to allow
users to select the desired behaviour.
[1]: https://bugzilla.gnome.org/show_bug.cgi?id=587896
[2]: gst-plugins-rs#423
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5372>
Formatters might call "loaded" from the `gessrc` streaming thread
meaning that the `->formatters` field need to be protected.
Several other APIs are called from gesbasedemux, in some radom
thread, so we should ensure that this is all MT. safe, and the API
makes it simple.
Co-authored-by: Philippe Normand <philn@igalia.com>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5431>
By using the gst_caps_set_simple() to set the format on all structures, the
compositor may create invalid combinations as the caps may contain passthrough
caps. Avoid this issue by intersecting the resul with its original.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5386>
Adds list of formats that should be used by element in needs to passthrough
video. It contains the full list of video format plus DMA_DRM format
and will be extended in the future as needed. This patches includes 3 new
symbols:
- GST_VIDEO_FORMATS_ANY_STR
- GST_VIDEO_FORMATS_ANY
- gst_video_formats_any()
The last one can be used by bindings or for code that prefers having
GstVideoFormat values instead of strings.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5386>
This commit ports functionality from the `rtpsrc` to make the `ristsrc`
work with dynamic payload types.
It adds two properties:
- `caps`
- `encoding-name`
These can be used to make the `ristsrc` receive other payload types than
the MPEG TS one.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5422>
While the suspend modes NONE and PAUSED provided a low startup latency
for connecting clients they did not ensure that streams started on
fresh data.
With this property we can maintain the low startup latency of those
suspend modes while also ensuring that a stream starts on a key unit.
Furthermore, by modifying the value of a new property,
ensure-keyunit-on-start-timeout, it is possible to accept a keyunit of
a certain age but discard it if too much time has passed and instead
force a new keyunit.
Fixes#2443
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4334>
The counter was using a signed 8 bit integer, which was overflowing
after 127 entries. That was then passed as an unsigned 32 bit integer to
libflac, which caused it to be converted to a huge unsigned number.
That then caused an invalid memory access inside libflac.
As a bonus, signed integer overflow is undefined behaviour.
Instead, use an unsigned 8 bit integer. Once this overflows the existing
code already catches it and stops adding the cue. While FLAC__metadata_object_cuesheet_insert_track()
takes an unsigned 32 bit integer for the track number, FLAC__StreamMetadata_CueSheet_Track is
limiting it to an unsigned 8 bit integer.
Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/2921
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5420>
Since DXVA does not support some profiles such as HEVC RExt,
vendor specific decoding API is still required.
When decoder is negotiated with d3d11 caps, decoder will convert
semi-planar frame to planar since semi-planar format (e.g.,
DXGI_FORMAT_NV12) is not supported by CUDA/D3D11 interop.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5409>
Found that osxaudiosink could not be added standalone in gst-full build
using
-Dgst-full-elements=osxaudio:osxaudiosink because element registration
was
done at the plugin level. Now src/sink elements and deviceprovider have
their
individual registration.
Copied/adapted from the alsa plugin.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5419>
Use gst_codec_utils_caps_get_mime_codec() in pbutils for codec
strings. That function gives more elaborate RFC 6381 compatible
strings than the helper functions in gstmdphelper.c, such as
"avc1.F4000D".
Remove the helper functions, as they were only used from dashsink.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5404>
When the property "start-time-selection" is set to "first", it
calculates the start time of the output from the buffer pts
(converting it to running time of the segment), but if the
rate is negative, the real start is not the pts, but the
pts + duration, because it plays from the end of the buffer
to it's start.
As a result of this bug, in the negative rate, when the
start-time-selection=first, the first frame is dropped
by the videoaggregator (reproduced on d3d11compositor).
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5276>
The interaudiosrc might take buffers of different sizes from the audio adapter,
so keeping metas consistency would be an issue. So the sink now strips the audio
metas away and the src adds them back (for non-interleaved layouts only) when
taking buffers from the adapter.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5324>
Move the GstStructure field into public struct for direct access, that's
easier than having to call a function to get it. It is not an API/ABI
breakage to extend the public structure of a GstMeta because they are
always allocated by inside GStreamer. The structure is exposed already
by gst_custom_meta_get_structure() which does not return a copy/ref, so
it is locked into holding a GstStructure forever anyway.
Also add gst_meta_register_custom_simple() because most of the time only
a name is required, tags and transform functions are more niche
use-case.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5385>
If there are multiple Wayland event listeners in different threads we
get the formats and modifiers pushed concurrently which leads to
segfault from GArray methods. This patch protects the array.
The problem occurs e.g. when using vaapipostproc together with Qt
qmlglsink, QtWayland will get the events as well as VAAPI.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5280>
Moves outputting frames to a task on the source pad, bringing vtdec in line with vtenc.
This brings possible performance improvements thanks to decoupling queueing new frames from outputting processed ones.
The queue length is limited to `2*DBP` to prevent decoding too far ahead compared to what we're pushing downstream.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5163>
Right now we split the RTP header from the current buffer into a new
buffer and aggregate those buffers for later processing if the
depayloader creates an output buffer.
This is cumbersome as it happens even if none of the incoming RTP
buffers carries RTP header extensions at all just because header
aggregation has been enabled in the depayloader class.
This commit will start aggregation only in case that there really are
RTP header extensions available on an incoming RTP buffer. The check
is trivial and cheap. Once activated we keep aggregation active for
all buffers. The active state is reset on state change READY_TO_PAUSE.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5278>
scanlines->m1 = same line of the previous field
scanlines->t0 = line above of the current field
scanlines->b0 = line below of the current field
scanlines->mp = same line of the next field
Deinterlacing a field weaved frame:
When deinterlacing the top field, the next bottom field is available
(part of the same frame). but when deinterlacing the bottom field,
the next top field (part of the next frame) is not available and
scanlines->mp equals NULL.
In this case it's better to use greedy algorithm using the prevous field
(twice) rather then linear interpolation of the current field.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5331>
If we end up with GST_CLOCK_TIME_NONE as running time for an RTP packet
then this can't be used for bitrate estimation, and also not for
constructing the next RTCP SR. Both would end up with completely wrong
values, and an RTCP SR with wrong values can easily break
synchronization in receivers.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5329>
This was easy to trigger when testing with e.g. vtenc ! vtdec ! glimagesink and closing the sink via window button,
causing GST_FLOW_ERROR to be received by the output loop, stopping it with the queue still full. This made the
enqueue_buffer() callback to lock waiting for space in our queue, while handle_frame() was waiting for the internal
VideoToolbox queue to free up, so that VTCompressionSessionEncodeFrame could finish. As the output loop was not
running, both functions waited forever.
Fixed by 1) immediately emptying our queue when GST_FLOW_ERROR is received (like we already did with _FLUSHING)
and 2) unconditionally setting the flushing flag in finish_encoding() when it sees the output loop stopped because
of GST_FLOW_ERROR, so that enqueue_buffer() will immediately discard any new frames coming out of VideoToolbox.
Both of those make sure we never run into the both-queues-full scenario.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5303>
As a short-term solution before full d3d12 rendering feature,
copy decoded d3d12 texture to shared d3d11 texture in order to use
existing various d3d11 implementations such as conversion, resizing,
and videosink.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5356>
There is no need to use DRM dumb pool if buffer to
render is already a DMABuf, just import it and render it.
This fixes a DMAbuf memory leakage when waylandsink downstream
element exports DMABuf while waylandsink is configured to be
DMABuf exporter (drm-device=/drv/dri/card0):
gst-launch-1.0 v4l2src io-mode=4 ! gtkwaylandsink drm-device=/dev/dri/card0
leakage identfied with command:
watch "cat /sys/kernel/debug/dma_buf/bufinfo | grep attached "
Fixes#2729
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5350>
There is no need to use DRM dumb pool if buffer to
render is already a DMABuf, just import it and render it.
This fixes a DMAbuf memory leakage when waylandsink downstream
element exports DMABuf while waylandsink is configured to be
DMABuf exporter (drm-device=/drv/dri/card0):
gst-launch-1.0 v4l2src io-mode=4 ! waylandsink drm-device=/dev/dri/card0
leakage identfied with command:
watch "cat /sys/kernel/debug/dma_buf/bufinfo | grep attached "
Fixes#2729
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5350>
Don't update info's size with the VA image reported data size for single plane
images, since drivers might allocate bigger space than the strictly required to
store the image, but when we dump the buffer as is (using filesink, for example)
the produced stream is corrupted. For multi-plane images video meta is required
to read/write them.
We updated info's size because gstreamer-vaapi did it too, but the reason to
update it there was for uploading and rendering surfaces (commit c698a015).
Furthermore, this patch adds an error message if the allocated data size for the
image by the driver is lesser than the expected because it would be a buggy
driver.
Fixes: #2959
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5308>
Even if decoder is negotiated with CUDA memory feature, if downstream
proposed no buffer pool, assume that the pool size is unknown.
And disable zero-copy if there's no more free output surface.
Or, in case of reverse playback, always copy frames.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5338>