[Cin] HEVC hw encoding on AMD Polaris cards

Phyllis Smith phylsmith2017 at gmail.com
Wed May 15 05:49:27 CEST 2019

> > We tried a couple of examples of using scale_vaapi  from the ffmpeg
> command
> > line and it always failed for us on a Intel laptop with Broadwell
> graphics
> > board.  For example, something like the following:
> >
> > ffmpeg -hwaccel vaapi -hwaccel_output_format vaapi -i input.mp4 -vf
> > 'scale_vaapi=1280:720'.. out.mp4
> It seems correct line (at least for intel) include a bit more than just
> scale_vaapi:
> https://xtechbiz.com/intel-vaapi-encoding-ubuntu-17-10/
> Using the vaapi_scaler in the video filters : It is possible to use
> Intel’s QuickSync hardware via VAAPI for resize and scaling (when up-or
> downscaling the input source to a higher or lower resolution), using a
> filter snippet such as the one shown below:
> -vf 'format=nv12,hwupload,scale_vaapi=w=1920:h=1080'
> You may specify a different resolution by changing the dimensions in =w=
> and :h= to suit your needs.
>  Thanks for the URL, it was illustrative ... and now GG is typing ...

this strategy uses a "complex filter graph" for the video data path .  cin5
only does simple filters on the "input pads" via the video_filter= keyword
in the decode opts file.  This seems to be a "complex graph" on the output
path.  When you use ffmpeg filters in cin5, the normal filter stack is done
using the ffmpeg "plugins".  They are really ffmpeg filters that have been
glued into the cin5 plugin stack, not into a filter graph.  There has been
no demand (until now) for a output filter chain.  It is probably not
impossible to write code for it, but it is not clear why you would use this
and not just drop a plugin on the stack to scale the data.  There may be a
speed gain, but it would have to be pretty attractive before it would be
worth adding an output filter graph interface to the ffmpeg encode data
path.  This method of  scaling is not very flexible.  The only way I can
see that it would be worth it, is that you intend to encode months of video
data, and the hardware would improve the rate of encoding enough to offset
the software development  time.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cinelerra-gg.org/pipermail/cin/attachments/20190514/236125ed/attachment-0001.html>

More information about the Cin mailing list