This also better handles in-memory interlaced XFB data placed by the CPU by considering the stride from the VI.