[FFmpeg-devel] [PATCH] Optimization: support for libx264's mb_info

[Stefano Sabatini]

On date Monday 2023-06-12 17:38:43 +0000, Carotti, Elias wrote:
> On Mon, 2023-06-12 at 18:23 +1000, Kieran Kunhya wrote:
> > CAUTION: This email originated from outside of the organization. Do
> > not click links or open attachments unless you can confirm the sender
> > and know the content is safe.
> > 
> > > 
> > > Looks good to me otherwise, maybe Michael/Anton or someone else
> > > want
> > > to have a look?
> > > 
> > 
> > I don't think we should be adding what is essentially libx264
> > specific code
> > to the public libavutil API.
> > 
> > Kieran
> 
> Hi Kieran,
> I disagree: this is not more libx264-specific than most other code in
> libavutil/libx264.c, like, say, the region of interest code.
> 
> P_SKIPs are in the standard and this API is designed to provide a
> generic way to generate them when the changing portion of a frame is
> known in advance.
> 
> Should other encoders (and I mean it in the broadest sense, not
> specifically h.264 ones) support a method for feeding in hints about
> unchanged portions of a frame, it shouldn't be hard to write the
> encoder-specific code for them in the respective encoder-specific file.
> 
> In fact, the public api provided only allows to specify as side data a
> list of rectangles (in pixels' space) on the frame (just like it's done
> for the region of interest) which is then translated into the specific
> libx264 hinting mechanism (in macroblock space.) 

+1, I don't think there is anything specific for x264 here.

In fact, we could even add a filter to compute the hinting data and
export it in the side data. If you think this is a good idea, we could
split the patches in two parts to make this more apparent.