Split from #800.
As per NvEncodeAPI v.6.0, we would have to keep both input and output buffers around until the frames come out (stops returning
NV_ENC_ERR_NEED_MORE_INPUT), asynchronously of course.
So we would need multiple output "bitstreamBuffer"s and multiple CUDA output buffers (used with
unmapInputResource would have to wait until the frames come out..)
The documentation is poor and conflicting: it says you have to do the picture type decision yourself in synchronous mode, but the pseudo-code does the opposite. Zero documentation on what
displayPOCSyntax is. etc..
asking for B frames doesn't seem to do anything either, frames still come out..
this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/1235