High quality deinterlacing on VA-API

[bodyjarrocks]

Hi all,

I have been patiently for sandy bridge to surface so I can build my new MythTV frontend. Sandy bridge should future proof me for some time.

I'm keen to take advantage of H/W decode that should let me get away with a very cool CPU.

The concern I have relates to deinterlacing. Will the Sandy Bridge drivers expose decent de-interlacing methods? From va_x11.h I can only see Bob and One-field. I was hoping for a high quality spatial-temporal method like NVIDIA support on VDPAU.

Can anyone shed light on this? I suppose it could be done on the CPU using Yadif already built into mythtv. That will require code changes though assuming VA-API can render to ordinary memory. Even better it could be implemented in a shader.

Regards
Mike

[gbeauche]

The concern I have relates to deinterlacing. Will the Sandy Bridge drivers expose decent de-interlacing methods? From va_x11.h I can only see Bob and One-field. I was hoping for a high quality spatial-temporal method like NVIDIA support on VDPAU.

A VA/PP API was proposed but not implemented yet. For Ironlake, there already is spatial or temporal deinterlacing implicitly implemented IIRC. I once saw the driver maintaing an history of presented surfaces for that. Or code was just there but not used yet.