Save some memory by allocating fewer buffers. This is still not optimal, but
close to where we can get with the current implementation.
Signed-off-by: Julian Scheel <julian@jusst.de>
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>