Those debug infos have proved to be very helpful when debugging
timestamp issues. They are often linked to gst-omx picking the wrong
frame when trying to map from OMX.
One big restriction of the OMX buffer pool has always been
that the GstMemory objects were flagged with NO_SHARE.
This was because the buffer pool needed to be sure that when
a buffer returned to the pool, it would be safe to release the
OMX buffer back to OpenMAX.
With this change, this is no longer a restriction. What this
commit introduces is a new allocator that allows us to track
the GstMemory objects independently. Now, when a buffer returns
to the pool, it is not necessary for the memory to be released
as well. We simply track the memory's ref count in the allocator
and we return the OMX buffer back when the memory's ref count
drops to 0.
The reason for doing this is to allow implementing zero-copy
transfers in situations where we may need to copy or map a
certain region of the buffer. For instance, omxh264enc ! h264parse
should be possible to be zero-copy by using an OMX buffer pool
between them.
gstomxvideoenc.c:2874:7: error: "USE_OMX_TARGET_ZYNQ_USCALE_PLUS" is not defined, evaluates to 0 [-Werror=undef]
#elif USE_OMX_TARGET_ZYNQ_USCALE_PLUS
Works on meson because it doesn't use -Wundef
We used to track the 'allocating' status on the pool. It is used while
allocating so output buffers aren't passed right away to OMX and input
ones are not re-added to the pending queue.
This was causing a bug when exporting buffers to v4l2src. On start
v4l2src acquires a buffer, read its stride and release it right away.
As no buffer was received by the encoder element at this point, 'allocating'
was still on TRUE and so the the buffer wasn't put back to the pending
queue and, as result, no longer available to the pool.
Fix this by checking the active status of the pool instead of manually
tracking it down. The pool is considered as active at the very end of
the activation process so we're good when buffers are released during
the activation.
The method we call in the context of pushing a buffer are all thread
safe. Holding a lock would prevent input buffers from being queued while
pushing.
https://bugzilla.gnome.org/show_bug.cgi?id=715192
Propose pool upstream so input buffers can be allocated by the port and
exported as dmabuf.
The actual OMX buffers are allocated when the pool is activated, so we
don't end up doing useless allocations if the pool isn't used.
https://bugzilla.gnome.org/show_bug.cgi?id=796918
This is no longer needed since we implemented close() vfuncs as the
encoder/decoder base class already take care of calling close() (which
is calling shutdown()) in its own change_state implementation.
We also move the shut down of the component from PAUSED_TO_READY to READY_TO_NULL.
By doing so upstream will have already deactivated the pool from the
encoder and so won't be preventing the OMX state change as the buffers
will all be released.
https://bugzilla.gnome.org/show_bug.cgi?id=796918
As stated in the spec ("6.1.3 Seek Event Sequence") we should pause
before flushing.
We were pausing the decoder but not the encoder so I just aligned the
two code paths.
https://bugzilla.gnome.org/show_bug.cgi?id=797038
According to the OMX spec (3.1.3.7.1) nFilledLen is meant to include any
padding. We use to include the horizontal one (stride) but not the
vertical one if nSliceHeight is bigger than the actual height.
The calculated nFilledLen was wrong as it didn't include the padding
between planes.
https://bugzilla.gnome.org/show_bug.cgi?id=796749
Increase the number of output buffers by the number of buffers requested
downstream.
Prevent buffers starvation if downstream is going to use dynamic buffer
mode on its input.
https://bugzilla.gnome.org/show_bug.cgi?id=795746
The OMX specs states that the nBufferCountActual of a port has to default
to its nBufferCountMin. If we don't change nBufferCountActual we purely rely
on this default. But in some cases, OMX may change nBufferCountMin before we
allocate buffers. Like for example when configuring the input ports with the
actual format, it may decrease the number of minimal buffers required.
This method checks this and update nBufferCountActual if needed so we'll use
less buffers than the worst case in such scenarios.
SetParameter() needs to be called when the port is either disabled or
the component in the Loaded state.
Don't do this for the decoder output as
gst_omx_video_dec_allocate_output_buffers() already check
nBufferCountMin when computing the number of output buffers.
On some platform, like rpi, the default nBufferCountActual is much
higher than nBufferCountMin so only enable this using a specific gst-omx
hack.
https://bugzilla.gnome.org/show_bug.cgi?id=791211
Setting the input format and the associated encoder/decoder settings
may also affect the nBufferCountMin of the input port.
Refresh the input port so we'll use up to date values in propose/decide
allocation.
https://bugzilla.gnome.org/show_bug.cgi?id=796445
0xffffffff is the magic number in gst-omx meaning 'the default value
defined in OMX'. This works fine with OMX parameters which are only set
once when starting the component but not with configs which can be
changed while PLAYING.
Save the actual OMX default bitrate so we can restore it later if user
sets back 0xffffffff on the property.
Added GST_OMX_PROP_OMX_DEFAULT so we stop hardcoding magic numbers
everywhere.
https://bugzilla.gnome.org/show_bug.cgi?id=794998
We weren't using the usual pattern when re-setting the bitrate:
- get parameters from OMX
- update only the fields different from 0xffffffff (OMX defaults)
- set parameters
Also added a comment explaining why we re-set this param.
https://bugzilla.gnome.org/show_bug.cgi?id=794998
- Report the error from OMX if any (OMX_EventError)
- If not report the failing to the application (GST_ELEMENT_ERROR)
- return GST_FLOW_ERROR rather than FALSE
- don't leak @frame
https://bugzilla.gnome.org/show_bug.cgi?id=795352
We already have the exact same message at the beginning of
gst_omx_video_enc_handle_frame(). Having it twice is confusing when
reading/grepping logs.
I kept the earlier one to keep the symetry with
gst_omx_video_dec_handle_frame().
https://bugzilla.gnome.org/show_bug.cgi?id=794897
Check input buffers for ROI meta and pass them to the encoder by using
zynqultrascaleplus's custom OMX extension. Also add a new
"default-roi-quality" in order to tell the encoder what quality level
should be applied to ROI by default.
https://bugzilla.gnome.org/show_bug.cgi?id=793696
The 'target-bitrate' property can be changed while PLAYING
(GST_PARAM_MUTABLE_PLAYING). Make it thread-safe to prevent concurrent
accesses between the application and streaming thread.
https://bugzilla.gnome.org/show_bug.cgi?id=793458
The Zynq UltraScale+ encoder implements a custom OMX extension to
directly import dmabuf saving the need of mapping input buffers.
This can be use with either 'v4l2src io-mode=dmabuf' or an OMX video
decoder upstream.
https://bugzilla.gnome.org/show_bug.cgi?id=792361
Make use of the new GstVideoEncoder QoS API to drop late input frames. This may
help a live pipeline to catch up if it's being late and all frames end up
being dropped at the sink.
https://bugzilla.gnome.org/show_bug.cgi?id=792783
If something goes wrong while trying to manually copy the input buffer,
the 'break' was moving us out of the 'for' loop but not out of the switch block.
So we ended up calling gst_video_frame_unmap() a second time (raising
assertions) and returning TRUE rather than FALSE.
Reproduced with a WIP zynqultrascaleplus OMX branch reporting wrong
buffer sizes and so triggering this bug.
https://bugzilla.gnome.org/show_bug.cgi?id=792167
If the OMX component supports dynamic buffer mode and the input buffers
are properly aligned avoid copying each input frame between OMX and
GStreamer.
Tested on zynqultrascaleplus and rpi (without dynamic buffers).
https://bugzilla.gnome.org/show_bug.cgi?id=787093
