Function direct_plane 

fn direct_plane(
    frame: &CUeglFrame,
    width: u32,
    height: u32,
) -> Option<DirectPlane>

Whether NVENC can read a mapped CUeglFrame in place, and how.

Either frame kind has to be usable as the session’s width × height input: a first plane present and non-null, and a geometry of at least the session’s. A pitch-linear plane also needs a row pitch covering width * 4 bytes at the 4-byte alignment NV_ENC_REGISTER_RESOURCE requires; a CUDA-array plane has to be four 8-bit channels, the layout NVENC’s packed formats describe. None sends the frame down the per-frame copy into the session’s own input surface.