Qualcomm GPU works fine with current implementation now.
Noticeable difference between when it was disabled and current
d3d11 implementation is that we now support GstD3D11Memory
pool, so there will be no more frequent re-binding decoder surface anymore.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2377>
Adding the compatile profiles when we decide the final profile used for decoding.
The final profile candidates include:
1. The profile directly specified by SPS, which is the exact one.
2. The compatile profiles decided by the upstream element such as the h265parse.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2322>
This is a work-around to identify some main profile streams having
wrong profile_idc. There are some wrongly encoded main profile streams
which doesn't have any of the profile_idc values mentioned in Annex-A,
instead, general_profile_idc has been set as zero and the
general_profile_compatibility_flag[general_profile_idc] is TRUE.
Assuming them as MAIN profile for now.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2322>
The segmentation is stateful, its information may depend on the previous
segmentation setting. For example, if loop_filter_delta_enabled is TRUE,
the filter_level[GST_VP9_REF_FRAME_INTRA][1] should inherit the previous
frame's value and can not be calculated by the current frame's segmentation
data only. So we need to maintain the segmentation state inside the vp9
decoder and update it when the new frame header comes.
We also fix the CLAMP issue of lvl_seg and intra_lvl because of their wrong
uint type here.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2369>
The current way of DMA buffer mapping is simply forwarding the job
to parent's map function, which is a mmap(). That can not handle the
non-linear buffers, such as tiling, compressed, etc. The incorrect
mapping of such buffers causes broken images, which are recognized
as bugs. We should directly block this kind of mapping to avoid the
misunderstanding.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2353>
We can only convert from non-cdp to cdp within the confines of valid cdp
framerates. The existing caps negotiation code was allowing any
framerate to convert to a cdp output which is incorrect and would hit an
assertion later.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2372>
We are dropping RASL (Random Access Skipped Leading picture) which
is associated with an IRAP (Intra Random Access Picture) that has
NoRaslOutputFlag equal to 1, since the RASL picture will not be
outputted and also it should not be used for reference picture.
So, corresponding GstVideoCodecFrame should be released immediately.
Otherwise GstVideoDecoder baseclass will hold the unused frame.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2330>
The build fails on macos with the following error:
/usr/local/Cellar/opencv/4.5.0_5/include/opencv4/opencv2/core/mat.hpp:2226:15: error: no template named 'initializer_list' in namespace 'std'
Mat_(std::initializer_list<_Tp> values);
fatal error: too many errors emitted, stopping now [-ferror-limit=]
35 warnings and 20 errors generated.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2368>
If they're not dropped, they can be blocked in the queue even if it is
leaky in the case where there is a buffer being pushed downstream. Since
in webrtc, it's unlikely that there will be a special allocator to
receive RTP packets, there is almost no downside to just ignoring the
queries.
Also drop queries if they get caught in the pad probe after the queue.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2363>
By using the clocksync inside the dtlssrtpenc, all streams inside a
bundled are synchronized together. This will cause problems if their
buffers are not already arriving synchronized: clocksync would wait for
a buffer on one stream and then buffers from the other stream(s) with
lower timestamps would all be sent out too late.
Placing the clocksync before the rtpbin and rtpfunnel synchronizes each
stream individually and they will be send out more smoothly as a result.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2355>
If the decoder has crop_top/left value > 0(e.g. the conformance
window in the H265). Which means that the real output picture
locates in the middle of the decoded buffer. If the downstream can
support VideoCropMeta, a VideoCropMeta is added to notify the
real picture's coordinate and size. But if not, we need to copy
it manually and the other_pool is needed. We always assume that
decoded picture starts from top-left corner, and so there is no
need to do this if crop_bottom/right value > 0.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2298>
The base va decoder's video_align is just used for calculation the
real decoded buffer's width and height. It does not have meaning
for the VideoMeta, because it does not align to the real picture
in the output buffer. We will use VideoCropMeta to replace it later.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2298>
The base va decoder's video_align is just used for calculation the
real decoded buffer's width and height. While the gst_video_info_align
just calculate the offset and stride based on the video_align. But
all the offsets and strides are overwritten in gst_va_dmabuf_allocator_try
or gst_va_allocator_try, which make that calculation useless.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2298>
