[FFmpeg-devel] [PATCH v6 1/3] avcodec: add flags for packets with top/bottom field

Sat May 19 00:17:06 EEST 2018

On Fri, 18 May 2018 21:59:13 +0100
Rostislav Pehlivanov <atomnuker at gmail.com> wrote:

> On 18 May 2018 at 21:03, wm4 <nfxjfg at googlemail.com> wrote:
> 
> > On Fri, 18 May 2018 20:09:02 +0100
> > Rostislav Pehlivanov <atomnuker at gmail.com> wrote:
> >  
> > > On 18 May 2018 at 20:05, Patrick Keroulas <  
> > > patrick.keroulas at savoirfairelinux.com> wrote:  
> > >  
> > > >
> > > > ----- Original Message -----  
> > > > > From: "Rostislav Pehlivanov" <atomnuker at gmail.com>
> > > > > To: "FFmpeg development discussions and patches" <  
> > > > ffmpeg-devel at ffmpeg.org>  
> > > > > Sent: Tuesday, May 15, 2018 4:46:02 PM
> > > > > Subject: Re: [FFmpeg-devel] [PATCH v6 1/3] avcodec: add flags for  
> > > > packets with top/bottom field
> > > >  
> > > > > On 15 May 2018 at 18:03, wm4 <nfxjfg at googlemail.com> wrote:
> > > > >  
> > > > >> On Tue, 15 May 2018 17:15:05 +0100
> > > > >> Rostislav Pehlivanov <atomnuker at gmail.com> wrote:
> > > > >>  
> > > > >> > On 15 May 2018 at 15:55, wm4 <nfxjfg at googlemail.com> wrote:
> > > > >> >  
> > > > >> > > On Mon, 14 May 2018 18:26:35 -0400
> > > > >> > > Patrick Keroulas <patrick.keroulas at savoirfairelinux.com> wrote:
> > > > >> > >  
> > > > >> > > > Signed-off-by: Patrick Keroulas <patrick.keroulas@  
> > > > >> savoirfairelinux.com>  
> > > > >> > > > ---
> > > > >> > > > doc/APIchanges | 3 +++
> > > > >> > > > libavcodec/avcodec.h | 8 ++++++++
> > > > >> > > > libavcodec/version.h | 4 ++--
> > > > >> > > > 3 files changed, 13 insertions(+), 2 deletions(-)
> > > > >> > > >
> > > > >> > > > diff --git a/doc/APIchanges b/doc/APIchanges
> > > > >> > > > index bbefc83..d06868e 100644
> > > > >> > > > --- a/doc/APIchanges
> > > > >> > > > +++ b/doc/APIchanges
> > > > >> > > > @@ -15,6 +15,9 @@ libavutil: 2017-10-21
> > > > >> > > >
> > > > >> > > > API changes, most recent first:
> > > > >> > > >
> > > > >> > > > +2018-05-xx - xxxxxxxxxx - lavc 58.20.100 - avcodec.h
> > > > >> > > > + Add AV_PKT_FLAG_TOP_FIELD and AV_PKT_FLAG_BOTTOM_FIELD.
> > > > >> > > > +
> > > > >> > > > 2018-05-xx - xxxxxxxxxx - lavu 56.18.101 - hwcontext_cuda.h
> > > > >> > > > Add AVCUDADeviceContext.stream.
> > > > >> > > >
> > > > >> > > > diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h
> > > > >> > > > index fb0c6fa..14811be 100644
> > > > >> > > > --- a/libavcodec/avcodec.h
> > > > >> > > > +++ b/libavcodec/avcodec.h
> > > > >> > > > @@ -1480,6 +1480,14 @@ typedef struct AVPacket {
> > > > >> > > > */
> > > > >> > > > #define AV_PKT_FLAG_DISPOSABLE 0x0010
> > > > >> > > >
> > > > >> > > > +/**
> > > > >> > > > + * The packet contains a top field.
> > > > >> > > > + */
> > > > >> > > > +#define AV_PKT_FLAG_TOP_FIELD 0x0020
> > > > >> > > > +/**
> > > > >> > > > + * The packet contains a bottom field.
> > > > >> > > > + */
> > > > >> > > > +#define AV_PKT_FLAG_BOTTOM_FIELD 0x0040
> > > > >> > > >
> > > > >> > > > enum AVSideDataParamChangeFlags {
> > > > >> > > > AV_SIDE_DATA_PARAM_CHANGE_CHANNEL_COUNT = 0x0001,
> > > > >> > > > diff --git a/libavcodec/version.h b/libavcodec/version.h
> > > > >> > > > index 3fda743..b9752ce 100644
> > > > >> > > > --- a/libavcodec/version.h
> > > > >> > > > +++ b/libavcodec/version.h
> > > > >> > > > @@ -28,8 +28,8 @@
> > > > >> > > > #include "libavutil/version.h"
> > > > >> > > >
> > > > >> > > > #define LIBAVCODEC_VERSION_MAJOR 58
> > > > >> > > > -#define LIBAVCODEC_VERSION_MINOR 19
> > > > >> > > > -#define LIBAVCODEC_VERSION_MICRO 101
> > > > >> > > > +#define LIBAVCODEC_VERSION_MINOR 20
> > > > >> > > > +#define LIBAVCODEC_VERSION_MICRO 100
> > > > >> > > >
> > > > >> > > > #define LIBAVCODEC_VERSION_INT AV_VERSION_INT(LIBAVCODEC_  
> > > > VERSION_MAJOR,  
> > > > >>  
> > > > >> > > \  
> > > > >> > > >  
> > > > >> > > LIBAVCODEC_VERSION_MINOR, \
> > > > >> > >
> > > > >> > > So far we could avoid codec-specific packet flags, and I think  
> > it  
> > > > >> > > should stay this way. Maybe make it side data, something with  
> > naming  
> > > > >> > > specific to the bitpacked codec. Or alternatively, if this codec
> > > > >> > > is 100% RTP specific and there's no such thing as standard  
> > bitpacked  
> > > > >> > > packets (e.g. muxed in other files etc.), add it to the packet
> > > > >> > > directly. The RTP code "repacks" it already on the libavformat  
> > side  
> > > > >> > > anyway.
> > > > >> > > _______________________________________________
> > > > >> > > ffmpeg-devel mailing list
> > > > >> > > ffmpeg-devel at ffmpeg.org
> > > > >> > > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> > > > >> > >  
> > > > >> >
> > > > >> > This codec isn't RTP specific, its the same as V210. There are  
> > no  
> > > > flags  
> > > > >> in  
> > > > >> > the bitstream, its just a sequence of packed pixels. And just  
> > like  
> > > > v210  
> > > > >> > there's a standard for what packets need to look like (rfc4175,  
> > and  
> > > > >> > unfortunately it does specify the fields need to be separate),  
> > so  
> > > > >> packing 2  
> > > > >> > fields in the muxer isn't really an option.  
> > > > >>
> > > > >> Then why are there separate bitpacked and v210 decoders/codec_ids?
> > > > >>  
> > > > >
> > > > > Its a different type of packing.
> > > > >
> > > > >
> > > > >  
> > > > >> > Side data seems a bit of an overkill for a flag. I'd say the  
> > field  
> > > > flags  
> > > > >> > are not codec specific as some raw codecs and containers can  
> > signal  
> > > > >> fields  
> > > > >> > per packet. So I don't really mind them.  
> > > > >>
> > > > >> You want codec specific flags to accumulate in AVPacket.flags?  
> > Can't way  
> > > > >> until we change the flags field to int128_t.
> > > > >>
> > > > >>  
> > > > > No, but I think 2 more bits won't hurt much, especially considering  
> > > > pretty  
> > > > > much all formats supporting interlaced content split each field into  
> > a  
> > > > > separate packet.  
> > > >
> > > > Recomposing a frame from fields on the demux side would make the  
> > bitpacked  
> > > > decoder
> > > > completely useless. Can we find a consensus? How about reusing
> > > > AVPictureStructure
> > > > as suggested by Derek?
> > > >  
> > > > > _______________________________________________
> > > > > ffmpeg-devel mailing list
> > > > > ffmpeg-devel at ffmpeg.org
> > > > > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel  
> > > > _______________________________________________
> > > > ffmpeg-devel mailing list
> > > > ffmpeg-devel at ffmpeg.org
> > > > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> > > >  
> > >
> > > Reusing that structure would mean adding a field to AVPackets. I'd rather
> > > avoid that, so I'm okay with the packet flags.  
> >
> > We can't add fields to AVPacket (ABI issues). I'm against the flags
> > though. None of the current packet flags are needed for correct
> > decoding, why change that suddenly?
> > _______________________________________________
> > ffmpeg-devel mailing list
> > ffmpeg-devel at ffmpeg.org
> > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> >  
> 
> AV_PKT_FLAG_TRUSTED is needed to decode some packets, so it would not be an
> entirely new change.

That only indicates how the packet was created. It's necessary to
"decode" some packets, but it doesn't really say anything about the
contents. Besides, it could get removed when cleaning up the side data
API.

Other flags are also generic. For a long time, AV_PKT_FLAG_KEY was the
only flag. It applies to every codec, and is in theory not necessary
for decoding (although some decoders might read it anyway, but that's
a separate story).

If we add flags about fields now, this will cause only confusion,
because:

1. Unlike someone would expect, they're not always set, but only for
   the very obscure situation of RTP uncompressed data transport.
2. They're suddenly needed for correct decoding.

> On the other hand, using side data would mean having to use
> AV_CODEC_CAP_PARAM_CHANGE, adding a AV_SIDE_DATA_PARAM_CHANGE_FIELD and
> adding a new AVCodecContext field to indicate the current field of a
> packet. Or adding a new 1-byte large side data type to indicate packet
> field.

I don't know why it can't be written to the packet data itself (and I
don't understand the patch author's comments about this - why does
adding a new byte to the packet data require "recomposing" data?).

But I think a new side data type would be much saner. We could even
just make it something generic, like AV_PKT_DATA_ANCILLARY or
something. It's apparently just packet data which somehow couldn't go
into the packet data.

There's no need to stuff that into packet flags instead.

> I think the packet flag solution is much saner than that.