Bug 91241

Also, the right border when zooming in is single-pixel band stretched to that area width. More noticeable on this image: http://i.imgur.com/u3drnmk.jpg

mpv author provided some details, here is what he said:

Specifically, it happens with vaPutSurface if you pass VA_FILTER_SCALING_HQ to it, and the source coordinates don't use the full image (i.e. cropping).

As it came out, the issue appears only if BOTH hardware decoding and hardware video output (with HQ scaling) used. If I use only HW output with HQ scaling, while using SW decoding, panscan works well. The same is true for HW decoding and SW output/scaling.

(In reply to Igor Tarasov from comment #1)
> Also, the right border when zooming in is single-pixel band stretched to
> that area width. More noticeable on this image:
> http://i.imgur.com/u3drnmk.jpg

I didn't see this issue, could you provide the command line ?

For example, something like this:

mpv some16:9video --hwdec=vaapi --vo=vaapi:scaling=hq --geometry=1280x640 --panscan=1

You can also use keys w and e to change panscan during playback.

Or to make things even more clear, you could use square window:

mpv some16:9video --hwdec=vaapi --vo=vaapi:scaling=hq --geometry=640x640 --panscan=1

You change scaling=hq in commandline to scaling=fast, in order to see how it is supposed to work.

Now it has became more interesting. I've found a file (http://download.jw.org/files/media_video/6e/ivpro_E_1_r720P.mp4) that is being played as with fish-eye camera if hq scaling is enabled. 

Here are screenshots of what it looks like compared to fast scaling: http://imgur.com/a/EUEgl.

The source ratio (height/weight) and target ratio are different, so hq stretches the surface in a different way than fast.