Re: [RFC] Stateful codecs and requirements for compressed formats

[Dave Stevenson <dave.stevenson@raspberrypi.org>]

Hi Nicolas

On Fri, 28 Jun 2019 at 16:48, Nicolas Dufresne <nicolas@ndufresne.ca> wrote:
>
> Le vendredi 28 juin 2019 à 16:21 +0100, Dave Stevenson a écrit :
> > Hi Hans
> >
> > On Fri, 28 Jun 2019 at 15:34, Hans Verkuil <hverkuil@xs4all.nl> wrote:
> > > Hi all,
> > >
> > > I hope I Cc-ed everyone with a stake in this issue.
> > >
> > > One recurring question is how a stateful encoder fills buffers and how a stateful
> > > decoder consumes buffers.
> > >
> > > The most generic case is that an encoder produces a bitstream and just fills each
> > > CAPTURE buffer to the brim before continuing with the next buffer.
> > >
> > > I don't think there are drivers that do this, I believe that all drivers just
> > > output a single compressed frame. For interlaced formats I understand it is either
> > > one compressed field per buffer, or two compressed fields per buffer (this is
> > > what I heard, I don't know if this is true).
> >
> > From the discussion that started this thread, with H264 and similar,
> > does the V4L2 buffer contain just the frame data, or the SPS/PPS
> > headers as well.
>
> In existing mainline encoder driver the SPS/PPS is included in the
> first frame produced. Decoders expect them to be in the first frame
> queued. For decoder, this is being relaxed now that we have a mechanism
> to notify the state change after the header has been processed.

So it sounds like the one bit missing is everyone's "friend" -
documentation. I'm eternally grateful to those who are making the
efforts in updating it. It's a thankless task, but absolutely
necessary.

For those outside the core linux-media circles there is a choice to be
made, and different APIs do adopt different approaches.
OpenMAX IL for one explicitly documents exactly the opposite approach
to V4L2, although admittedly through an optional flag. 1.1.2 spec [1],
section 3.1.2.7.1 (page 70)

The OMX_BUFFERFLAG_CODECCONFIG is an optional flag that is
set by an output port when all bytes in the buffer form part or all of a set of
codec specific configuration data. Examples include SPS/PPS nal units
for OMX_VIDEO_CodingAVC or AudioSpecificConfig data for
OMX_AUDIO_CodingAAC. Any component that for a given stream sets
OMX_BUFFERFLAG_CODECCONFIG shall not mix codec
configuration bytes with frame data in the same buffer, and shall send all
buffers containing codec configuration bytes before any buffers containing
frame data that those configurations bytes describe.

  Dave

[1] https://www.khronos.org/registry/OpenMAX-IL/specs/OpenMAX_IL_1_1_2_Specification.pdf

> > > In any case, I don't think this is specified anywhere. Please correct me if I am
> > > wrong.
> > >
> > > The latest stateful codec spec is here:
> > >
> > > https://hverkuil.home.xs4all.nl/codec-api/uapi/v4l/dev-mem2mem.html
> > >
> > > Assuming what I described above is indeed the case, then I think this should
> > > be documented. I don't know enough if a flag is needed somewhere to describe
> > > the behavior for interlaced formats, or can we leave this open and have userspace
> > > detect this?
> > >
> > >
> > > For decoders it is more complicated. The stateful decoder spec is written with
> > > the assumption that userspace can just fill each OUTPUT buffer to the brim with
> > > the compressed bitstream. I.e., no need to split at frame or other boundaries.
> > >
> > > See section 4.5.1.7 in the spec.
> > >
> > > But I understand that various HW decoders *do* have limitations. I would really
> > > like to know about those, since that needs to be exposed to userspace somehow.
> > >
> > > Specifically, the venus decoder needs to know the resolution of the coded video
> > > beforehand and it expects a single frame per buffer (how does that work for
> > > interlaced formats?).
> > >
> > > Such requirements mean that some userspace parsing is still required, so these
> > > decoders are not completely stateful.
> > >
> > > Can every codec author give information about their decoder/encoder?
> > >
> > > I'll start off with my virtual codec driver:
> > >
> > > vicodec: the decoder fully parses the bitstream. The encoder produces a single
> > > compressed frame per buffer. This driver doesn't yet support interlaced formats,
> > > but when that is added it will encode one field per buffer.
> >
> > On BCM283x:
> >
> > The underlying decoder will accept anything, but giving it a single
> > frame per buffer reduces latency as the bitstream parser gets kicked
> > earlier. Based on previous discussions I am setting the flag so that
> > it expects one compressed frame per buffer, but I don't believe it
> > goes wrong should that not be the case (it'll just waste a bit of
> > processing effort).
> > It'll parse the headers and produce a V4L2_EVENT_SOURCE_CHANGE event
> > should the capture queue format not match the stream parameters.
> > Interlacing isn't supported yet (it's on the list), but I believe the
> > hardware produces the equivalent to V4L2_FIELD_INTERLACED_[TB|BT].
> >
> > The encoder currently spits out the H264 SPS/PPS headers as a separate
> > V4L2 buffer, and then one compressed frame per V4L2 buffer (provided
> > the buffer is big enough). Should
> > V4L2_CID_MPEG_VIDEO_REPEAT_SEQ_HEADER be set, then it will repeat the
> > headers in an independent V4L2 buffer before each I frame.
> > I'm quite happy to amend this should we have a decent spec of what is
> > required. As I've never found a spec it's been trial and error until
> > now.
> > There is no interlaced support available.
> >
> >   Dave