This reverts commit fd4cb488.

We don't need this commit if we don't use posix_memalign().

This is significantly slower than malloc().

Main loop now burns 3 cycles per 8 values, if I compute right.
This is not quite main bottleneck, but it is pretty much always used
(since we output 16-bits PCM).

AVI demuxer now also check the number of frames from the index and the
avi header.

Thanks MSJ for the report on the forum.

audio_format handles all conversion between integer and float formats,
but none of the conversions from/to fixed-point.

This partially reverts commit 7cde22ed.

 * It did not fix the issue with UDP packet loos that brought it in the
   first place (#1919, it was a socket buffer issue),
 * It does not play nice with posix_memalign(),
 * It is slow.

In particular, this fixes invalid memory use when downsampling
with the "trivial" plugin.

It also fixed a race condition with mouse events.

It does not change anything because it was already at the right
value.

T.140 support will be included in the next release of live555

Please improve

Modified the way DDS (display definition segment - HD res dvb subs) is handled and added comments/explanation.

Applied by jb

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

For some reason, the aea probe needs to be > 2048 + 212. I would thought that >= would be enough, but it doesn't seem so.
I hope it won't increase too much in the future.

audio filter plugins are now useless.
This breaks visual (needs to be ported to filter_t) and will likely
expose bugs in under-used audio filter2 infrastructure.

TODO: use filter chain instead of custom filter handling in core.

Big filter changes, and worse, moved audio_format plugin.