The code currently use cur_buf as the target of the copy,
but cur_buf can be cleared by the filter if it has given
the reference away or stored it elsewhere as soon as start_frame.
The code still relies on the fact that the reference is not
destroyed until end_frame. All filters currently follow that condition.
An av_assert1() is added to check it; it should at least cause
very visible errors in valgrind.
Refactoring mmx2/mmxext YASM code with cpuflags will force renames.
So switching to a consistent naming scheme beforehand is sensible.
The name "mmxext" is more official and widespread and also the name
of the CPU flag, as reported e.g. by the Linux kernel.
The boilerplate states that the files are under LGPL,
but refer the user to the GPL at one place.
These files were (re)implemented specifically for FFmpeg.
Semantic for the function ff_null_start_frame() was changed in
07bad27810, and it has now the same behavior of
ff_null_start_frame_keep_ref(), thus it makes no sense to keep both of
them.
There is no need to start_frame immediately on the output link
since the rest is only done with the last frame of the tile.
link->out_buf is now automatically dereferenced; since we give
it away it must be cleared.
Fix an assert failure; found by Clément Bœsch.
There is no need to duplicate the input reference, since a filter should
not pass along a reference which is later modified. If this happens the
filter passing the reference should be fixed.
Also improve consistency with start_frame() of settb, allowing a pending
factorization.
Avoid to fill the same picture again and again with the same content.
Optimize computation, and provides an example for the use of the
AV_PERM_REUSE permission flag.
This allows for padding/trimming at the start of stream. By default, no
assumption is made about the first frame's expected pts, so no padding or
trimming is done.
The voice register functions return the same voice structure
upon multiple registration. It causes us two problems:
If we delete a voice without deregistering it, it leaves
a dangling pointer inside the library.
If we delete or unregister a voice at uninit, it may still
be in use by another instance of the filter.
The second problem is solved by keeping an usage counter inside
asrc_flite. This is not thread-safe, but neither is flite itself.
Since the recent changes, movie and amovie are able to deal with more
than one type of stream, so they should be categorized as "multimedia
sources" rather than audio/video sources.
Allow substitution of channel pairs in the input for nearby channel pairs in
the output in order to get a closer match. Also weigh LFE channel mismatch
differently to favor matching the same layout without LFE over one less
channel with LFE.