There May Still Be Hope For R600g Supporting XvMC, VDPAU

[phoronix]

Phoronix: There May Still Be Hope For R600g Supporting XvMC, VDPAU

When looking at our 2010 Linux Graphics Survey results, the second most popular technology sought after by Linux desktop users was video playback acceleration. This isn't surprising considering only with NVIDIA's proprietary driver using VDPAU on modern GeForce hardware can you get a decent experience or with select Intel chipsets supporting VA-API. With everything else, you're pretty much limited to nothing or the not-too-useful X-Video. There's also cases like with the ATI Catalyst driver providing XvBA support, but that's often buggy and rubbish, the same goes with the Intel Poulsbo blobs and their VA-API support...

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

[gbeauche]

Phoronix: There May Still Be Hope For R600g Supporting XvMC, VDPAU

When looking at our 2010 Linux Graphics Survey results, the second most popular technology sought after by Linux desktop users was video playback acceleration. This isn't surprising considering only with NVIDIA's proprietary driver using VDPAU on modern GeForce hardware can you get a decent experience or with select Intel chipsets supporting VA-API. With everything else, you're pretty much limited to nothing or the not-too-useful X-Video. There's also cases like with the ATI Catalyst driver providing XvBA support, but that's often buggy and rubbish, the same goes with the Intel Poulsbo blobs and their VA-API support...

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

It's strange they want to implement VDPAU if they only have iDCT or MoComp. In that case, VA-API is better suited. Otherwise, they would have to rewrite a new codec parser, thus duplicating code from the upper library...

BTW, VA-API on Poulsbo HW works very good, since this matches the Windows driver almost exactly.

[HokTar]

Yes!!!

I've been waiting for this (or something similar) for ages!
I do hope this work gets done in the near future! Kudos for Christian even in advance!

[HokTar]

After reading the actual letter it seems to me that xvmc is mostly working already! Cool.
So, when xvmc/vdpau will be implemented I will only need to choose one of them as the output driver in smplayer and I'm already good to go? Assuming that I run the proper r600g driver of course.

That would be just fantastic!

(Sorry for the double, but you know the deal about editing...)

[69acid69]

Well, and what about VDPAU for r300g?

[Chewi]

I've been close to telling Michael off for misusing British slang before but he really made me laugh this time. "Buggy and rubbish." Hahaha.

Excellent news though. MythTV on my 4560 is passable but it could be a lot better.

[tball]

It's strange they want to implement VDPAU if they only have iDCT or MoComp. In that case, VA-API is better suited. Otherwise, they would have to rewrite a new codec parser, thus duplicating code from the upper library...

BTW, VA-API on Poulsbo HW works very good, since this matches the Windows driver almost exactly.

So with VA-API you don't need to write a new bitstream parser? Does it work in similar ways as xvmc, where you receive the decoded bitstream in macroblocks?

[gbeauche]

So with VA-API you don't need to write a new bitstream parser? Does it work in similar ways as xvmc, where you receive the decoded bitstream in macroblocks?

Exactly, for MoComp or iDCT. e.g. VAMacroblockParameterBufferMPEG2 and XvMCMacroBlock structures pretty look the same. I have a test program from Intel that does MPEG-2 MC if someone is interested. Actually, I initially wanted to write an xvmc-video wrapper but I still didn't have time. There are also provisions for H.264 & VC-1 MC since some of their other drivers implement that.

[Kano]

@gbeauche

Can you enable your xvba-video hack for all opengl 4 cards?

[gbeauche]

Can you enable your xvba-video hack for all opengl 4 cards?

Which ones are missing? Are you talking about the Radeon HD 6000 series? If so, I would need the PCI ids.