Hi.
Speaking of the frame pool system in libavfilter
<https://lists.ffmpeg.org/lore/ffmpeg-devel/[email protected]/T/>,
I notice that libavfilter is no longer capable of direct rendering at
all.
Direct rendering is when you have a filter graph like that:
scale, drawtext, pad, showinfo, buffersink
and scale obtains from pad a frame with a buffer with enough padding for
pad to work in-place without copy, and pad obtains from the application
through buffersink a frame backed by a buffer in a specially allocated
memory area, even though drawtext and showinfo are in between.
It saves a lot of memory copy.
lavfi had the infrastructure for direct-rendering at its beginning. The
moment we lost it is probably when we switched to using AVFrame instead
of a custom structure.
We could re-introduce it. That would not be easy, but not monstrous
either I think. (But I have more interesting things to do right now.)
Regards,
--
Nicolas George
_______________________________________________
ffmpeg-devel mailing list -- [email protected]
To unsubscribe send an email to [email protected]
