On Tue, Dec 15, 2020 at 6:12 AM Nicolas Dufresne <nicolas@xxxxxxxxxxxx> wrote: > > Le vendredi 11 décembre 2020 à 12:18 +0900, Tomasz Figa a écrit : > > A lot of the hardware encoders implement only the basic codec > > features, e.g. they would only encode in-order (no B-frames), have at > > most 2 reference frames, etc. In that case, it's as simple as waiting > > for dequeuing a CAPTURE buffer with a timestamp matching the OUTPUT > > buffer being waited on. > > To me this is not a justification to allow not implementing that drain process > with CMD_STOP. Having a single drain path ensure the userspace will work for > both non-featured and full featured encoder. In the end, recommending to not > implement CMD_STOP seems like giving vendors a short cut, which I don't think is > beneficial to the users. > Fair enough. I think we've been looking for some definitive opinions like this. :) Best regards, Tomasz
