On the one hand, when decoding in hardware, the video surface
dimensions will match the coded dimensions, rather than the visible
ones. Otherwise the hardware decoder would not have enough space for
its output. In this the libavcodec plugin does not add padding (since
libavcodec SIMD optimizations are not involved).

On the other hand, when decoding in software, the video surface height
must not include the padding added by libavcodec SIMD optimizations. So
then the visible height needs to be used to avoid bleeding green lines.