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

For some reason, libavcodec deverlopers didn't update LIBAVCODEC_VERSION_MICRO when introducing the new CODEC_ID_FRWU. We will then have to wait for the next bump to have it enabled.

I'm not going to suffer another round of VLC bug reports due to glibc.

It shouldn't matter much when we get to such a sorry code path, but
better safe than sorry.

Revert "video_filter_puzzle: remove unnedeed structure (it does not change anything to the race conditions already present)."

This reverts commit a114dc51.

It allows the module to trigger some vout callback while loading.

video_filter_puzzle: remove unnedeed structure (it does not change anything to the race conditions already present).

The library delivers us pixel width/height relative to display size
so we must compute it accordingly, otherwise we have conflicting data
in i_sar_num/i_sar_den and i_aspect.

There is only one button on a touch screen.

Those are highly incomplete, especially as they aren't unrolled.
But it already decreases libmpeg2 CPU usage by 20%.

400 Mb buffer would kill almost any such (existing) system.

(This kinda sucks, but liba52 does not export is build mode through
 any public header)

It's not implemented for fixed-point. And it's so slow that it would
probably not make sense anyway.