[FFmpeg-devel] Q: QSV filters (state of implementation)

Mark Thompson sw at jkqxz.net
Tue Feb 21 16:01:10 EET 2017

On 21/02/17 13:51, Haris Zukanovic wrote:
> Hi everyone,
> The last record of an attempt to introduce QSV filters in the mailing list would be the
> thread titled lavf/vpp: Enable vpp filter, an Intel GPU accelerated scaler from 8/25/16.
> ...but I understand the hwaccell sharing context was not at all ready at that point...
> Since now ffmpeg does implement api for hwaccell to enable zero-copy pipeline between decoder and encoder, I wonder how far away it is from being ready to accept implementation of some of the hw-accellerated filters as well, like deinterlace and resize?
> Would you say the QSV zero-copy filters implementation will be able to support a scenario for multiple quality encoding where, for example, when producing 3 qualities I would:
> 1. decode + deinterlace once + split into 3
> 2. resize 3 times (each split)
> 3. encode 3 times (once for each resized output)

This all works in libav/avconv - if you want a working implementation now then I suggest you use it there.

Merging it to ffmpeg is currently blocked by <http://ffmpeg.org/pipermail/ffmpeg-devel/2017-February/207178.html>.

- Mark

More information about the ffmpeg-devel mailing list