[Libav-user] h264,h265: maximum referenced frames vs. GOP size

Heiner Mueller ffmpeg at mxwendler.net
Fri Sep 11 13:15:45 CEST 2015

>>> i found this in the DXVA code, and i would like
>>> to optimize it a bit:
>> Then please send your tested patch (made with git
>> format-patch) to the development mailing list
>> where it can be reviewed.
> I can save you the trouble though. First of all, gop_size is an
> encoding parameter, and not useful here at all.
> Secondly, there is unfortunately no parameter that conveys this
> information that is both (1) accurate, and (2) known early enough to
> be useful.

Yes, i was asking for the correctness of this assumption.
So the assumption is not correct =/

If we decode 4K video, saving four surfaces is a considerable
amount of memory. Is there a later point where we can re-set the
amount of surfaces? We can check for I-frame distances, but
this does not guarantee that we find the maximum - this may change
at any time in the stream i think.

Aside: quick check across our testing footage reported a
gop_size of 12 for all h264. where is this set when decoding, i
could not find it in the source code?

> So, this cannot be "improved" really.
> _______________________________________________
> Libav-user mailing list
> Libav-user at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/libav-user

Hendrik Wendler
Schwanseestrasse 92 99427 Weimar
Reichenberger Strasse 107 10999 Berlin

More information about the Libav-user mailing list