Hi10P Support Proposed For VDPAU

[phoronix]

Phoronix: Hi10P Support Proposed For VDPAU

Open-source developers are hoping to be able to introduce Hi10P support to NVIDIA's Video Decode and Presentation API for Unix (VDPAU) library...

http://www.phoronix.com/vr.php?view=MTQ5NjU

[fritsch]

Phoronix: Hi10P Support Proposed For VDPAU

Open-source developers are hoping to be able to introduce Hi10P support to NVIDIA's Video Decode and Presentation API for Unix (VDPAU) library...

http://www.phoronix.com/vr.php?view=MTQ5NjU

A bit funny to read the own name on phoronix. All cudos to the hard working amd oss developers, those guys made these patches so easy!

We will do our part at xbmc to use that technique, when all parts are ready.

[RussianNeuroMancer]

A bit funny to read the own name on phoronix. All cudos to the hard working amd oss developers, those guys made these patches so easy!

Big thanks to you and Christian for your work!
Hopefully this feature (Hi10P playback with hardware decoder help) that currently not available on other platforms (Windows and OS X) become available on Linux with AMD hardware.

[AdamW]

A bit funny to read the own name on phoronix. All cudos to the hard working amd oss developers, those guys made these patches so easy!

We will do our part at xbmc to use that technique, when all parts are ready.

So help me out here - gotta admit I'm not really au fait with all this 'state tracker' stuff. Executive summary: does this mean we'll get trouble-free HW-accelerated decoding of hi10p in XBMC running on top of...open drivers? Proprietary drivers? In the fairly near term? What would be the oldest HW that would support this decoding? Thanks!

[johnc]

I don't know much about UVD... Is this a dedicated DSP for decoding and if not, is it any better than just using software?

[smitty3268]

I don't know much about UVD... Is this a dedicated DSP for decoding and if not, is it any better than just using software?

Yes, it's custom decoding hardware. Compare it to PureVideo on NVidia gpus.

[johnc]

Yes, it's custom decoding hardware. Compare it to PureVideo on NVidia gpus.

So Radeon GPUs can decode high 10 profile? Which GPUs have this capability?

[przemoli]

So help me out here - gotta admit I'm not really au fait with all this 'state tracker' stuff. Executive summary: does this mean we'll get trouble-free HW-accelerated decoding of hi10p in XBMC running on top of...open drivers? Proprietary drivers? In the fairly near term? What would be the oldest HW that would support this decoding? Thanks!

1) Mesa is about providing unified FRONT END for gpu drivers.
2) This is acheaved by common API entry points (classic + gallium) and some common helper code (classic + gallium) and with State Trackers (gallium).
3) State trackers try to implement given API (OpenGL, OpenGL ES, OpenMAX, etc.) in general terms. Leaving blank spaces for dealing with actuall hardware.
4) Drivers fill in that blanks with actuall code.
5) This way, repeatable code IS NOT repeated in all those drivers.


(And obviously, if someone advance or create State tracker, other gallium drivers can build on that code -> Nouveau theam is usually fast to add any needed code if AMD crew add something to OpenGL State tracker. Which benefit them greatly due to small HR pool)


So:
State tracker means possibility of supporting something. GPU drivers must support it too thoug to make it work.

[fritsch]

For those playing with it, i missed to squash a commit before posting, which resulted in a build error. Should work now: http://sprunge.us/WacV

[AnonymousCoward]

Some features I'd like to see in the VDPAU state tracker:

High quality deinterlacing: DEINTERLACE_TEMPORAL and DEINTERLACE_TEMPORAL_SPATIAL

High quality scaling: HIGH QUALITY SCALING - L1 (even this would be enough which is all my nvidia card supports)

Not sure if these are strictly nvidia hardware based or if it can be done on Radeon hardware as well.

On a somewhat related note, I'd like to see triple buffered vsync for all of the Mesa drivers. Yeah, it takes more memory but it's superior to regular vsync when performance is needed (and tends to lower input lag a bit as well)