Re: [RFC PATCH 1/2] media: docs-rst: Add decoder UAPI specification to Codec Interfaces

[Nicolas Dufresne <nicolas@ndufresne.ca>]

Le jeudi 07 juin 2018 à 16:27 +0900, Tomasz Figa a écrit :
> > > I'd say no, but I guess that would mean that the driver never
> > > encounters it, because hardware wouldn't report it.
> > > 
> > > I wonder would happen in such case, though. Obviously decoding of such
> > > stream couldn't continue without support in the driver.
> > 
> > GStreamer supports decoding of variable resolution streams without
> > driver support by just stopping and restarting streaming completely.
> 
> What about userspace that doesn't parse the stream on its own? Do we
> want to impose the requirement of full bitstream parsing even for
> hardware that can just do it itself?

We do it this way in GStreamer because we can and is more reliable with
existing drivers. I do think that the driver driven renegotiation is
superior as it allow a lot more optimization. Full reset is a just the
slowest possible method of renegotiating. It is not visually fantastic
with dynamic streams, like DASH and HLS. Though, we should think of a
way driver can signal that this renegotiation is supported.

Nicolas