[FFmpeg-devel] [PATCH 2/2] avfilter/vf_scale2ref: switch to FFFrameSync

Michael Niedermayer michael at niedermayer.cc
Tue Mar 19 23:55:56 EET 2024


On Wed, Mar 13, 2024 at 01:24:25PM +0100, Niklas Haas wrote:
> From: Niklas Haas <git at haasn.dev>
> 
> This filter's existing design has a number of issues:
> 
> - There is no guarantee whatsoever about the order in which frames are
>   pushed onto the main and ref link, due to this being entirely
>   dependent on the order in which downstream filters decide to request
>   frames from their various inputs. As such, there is absolutely no
>   synchronization for ref streams with dynamically changing resolutions
>   (see e.g. fate/h264/reinit-*).
> 
> - For some (likely historical) reason, this filter outputs its ref
>   stream as a second ref output, which is in principle completely
>   unnecessary (complex filter graph users can just duplicate the input
>   pin), but in practice only required to allow this filter to
>   "eventually" see changes to the ref stream (see first point). In
>   particular, this means that if the user uses the "wrong" pin, this
>   filter may break completely.
> 
> - The default filter activation function is fundamentally incapable of
>   handling filters with multiple inputs cleanly, because doing so
>   requires both knowledge of how these inputs should be internally
>   ordered, but also how to handle EOF conditions on either input (or
>   downstream). Both of these are best left to the filter specific
>   options. (See #10795 for the consequences of using the default
>   activate with multiple inputs).
> 
> Switching this filter to framesync fixes all three points:
> 
> - ff_framesync_activate() correctly handles multiple inputs and EOF
>   conditions (and is configurable with the framesync-specific options)
> - framesync only supports a single output, so we can (indeed must) drop
>   the redundant ref output stream
> 
> Update documentation, changelog and tests to correspond to the new usage
> pattern.
> 
> Fixes: https://trac.ffmpeg.org/ticket/10795
> ---
>  Changelog                                |   2 +
>  doc/filters.texi                         |  10 +-
>  libavfilter/vf_scale.c                   | 130 ++++++++++++-----------
>  tests/filtergraphs/scale2ref_keep_aspect |   3 +-
>  4 files changed, 76 insertions(+), 69 deletions(-)
> 
> diff --git a/Changelog b/Changelog
> index 069b8274489..bacda2524ea 100644
> --- a/Changelog
> +++ b/Changelog
> @@ -32,6 +32,8 @@ version <next>:
>  - DVD-Video demuxer, powered by libdvdnav and libdvdread
>  - ffprobe -show_stream_groups option
>  - ffprobe (with -export_side_data film_grain) now prints film grain metadata
> +- scale2ref now only has a single output stream, and supports the framesync
> +  options

is it neccessary to drop compatibility to the old syntax ?

thx

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

If you drop bombs on a foreign country and kill a hundred thousand
innocent people, expect your government to call the consequence
"unprovoked inhuman terrorist attacks" and use it to justify dropping
more bombs and killing more people. The technology changed, the idea is old.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20240319/e026156e/attachment.sig>


More information about the ffmpeg-devel mailing list