Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Conformance sample 50fps, interlaced, MVCICT-1.264

corner case when frame contains only AU delimiter

Does reinject xps NAL on I Frame. Still not optimal
and duplicate prone, but that's what current
AVC muxer does.

Content for layer >= 63 is unspecified

This fixes a regression from d3db4b7f

Signed-off-by: Thomas Guillem <thomas@gllm.fr>

Signed-off-by: Thomas Guillem <thomas@gllm.fr>

when seeking during stream output we need to be able to flush internal buffers
Signed-off-by: Thomas Guillem <thomas@gllm.fr>

similar to vout_IsEmpty() for stream output
Signed-off-by: Thomas Guillem <thomas@gllm.fr>

Converts any AnnexB prefix to any xVC prefix.
Adds zero copy optimizations for single NAL.

It was mistakenly added in 315f8e47.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

If there are no interrupt contexts in the whole process, then the
thread variable is uninitialized. This adds a check.

Using thread_local would be much simpler and faster (but C11 threads).

When draining (pp_block == NULL), queue a dummy input buffer with a EOS flag,
and wait for the output thread to process all output frames.

This patch also add extra check when processing output buffers. Indeed, in some
case, the last output buffer with a EOS flag can be invalid.

Drain when pp_block is NULL.

The packetizer set the *pp_block to NULL, therefore p_block was NULL too.

This fixes a regression from 7ada6eae

Closes #16128

Audio, video, and sub decoders were not drained. Indeed, pf_decode_* functions
were called with *pp_block == NULL instead of pp_block == NULL. There was no
way to distinguish a drain from an other call from the decoder loop.

There was also an issue with the packetizer. Indeed, a successful call to
pf_packetize leaded to a drain.

This fixes a deadlock when seeking paused. It happened when avcodec needed more
pictures than VLC. It was easily reproducible with h264 in a TS file.