Holy Crap! You Can Use XvMC With ATI Gallium3D!

[agd5f]

Will it be usefull for R300 hardware (like mine RV350 and RV370), too? Or does it lack of power/shader units needed for such acceleration?

Yes, it's still useful on r3xx-r5xx hardware. There's no UVD hw on those asics so they always used shaders for MC in the proprietary driver.

[bridgman]

No, on the VIA Unichrome it also supports Motion Compensation, IDCT and Variable-Length Decoding for MPEG-4 ASP.

I never quite understood why NVIDIA and AMD had to come up with their own XvMC replacements. XvMC can be extended to provide playback acceleration for most video codecs, and if the limitations are that bad a common open-source successor should have been developed. If you think nearly every player had to be patched to support VPDAU, just so it can accelerate playback for a codec which could have been easily added to XvMC...

My understanding was that the XvMC API would have needed such drastic changes to handle anything past ASP (eg H.264/MPEG-4 AVC) at MC or IDCT level that the resulting interface would have been nothing like XvMC. Don't know if that is the case but that seems to be the conventional wisdom. I believe that only VLD-level support was added for H.264 since the API differences are much less at VLD level.

The other point AFAIK was simply that XvMC had not been broadly adopted so it probably seemed like a good time to introduce a newer API. Unfortunately VA-API, VDPAU and XvBA were independently developed at more or less the same time, so we went from zero new APIs to three new APIs fairly quickly.

It's like when you see a part of town with no hotels... everyone recognizes the opportunity at the same time so you end up with three hotels and only enough customers to fill one of them

[bridgman]

Cooper is still at AMD, but he moved from software to one of the hardware teams.

[Nille]

It should be possible it get it working on r300g as well. The new code looks like a port of Cooper's original r300g pipe-video work:
http://cgit.freedesktop.org/mesa/mes...1fb9c17d2c2ecc

By the way what is with Cooper Yuan? Since an long time he has nothing committed to mesa or the radeon Stuff.
Is he still at AMD?

[tball]

Correct me if I'm wrong but doesn't XvMC only accelerate mpeg2?

As in one of the formats least needing acceleration?

I understand, that it can be done and it was fairly quickly implemented, but I don't get why people are rejoicing over this. Vdpau or Va-api yes, but this?

Thats right. But having the video pipe up and running on a simpler api as xvmc makes the creating of more advanced api's easier.

[sturmflut]

Correct me if I'm wrong but doesn't XvMC only accelerate mpeg2?

No, on the VIA Unichrome it also supports Motion Compensation, IDCT and Variable-Length Decoding for MPEG-4 ASP.

I never quite understood why NVIDIA and AMD had to come up with their own XvMC replacements. XvMC can be extended to provide playback acceleration for most video codecs, and if the limitations are that bad a common open-source successor should have been developed. If you think nearly every player had to be patched to support VPDAU, just so it can accelerate playback for a codec which could have been easily added to XvMC...

[xeros]

It should be possible it get it working on r300g as well. The new code looks like a port of Cooper's original r300g pipe-video work:
http://cgit.freedesktop.org/mesa/mes...1fb9c17d2c2ecc

Will it be usefull for R300 hardware (like mine RV350 and RV370), too? Or does it lack of power/shader units needed for such acceleration?

[agd5f]

This is great news indeed!

I own a X1600pro (AGP) that for sure can handle XvMC and accelerate H.264 (720p) content.

I'm wondering if there is (or will ever be) this kind support for R500 cards?

It should be possible it get it working on r300g as well. The new code looks like a port of Cooper's original r300g pipe-video work:

http://cgit.freedesktop.org/mesa/mesa/commit/?h=pipe-video&id=6783672a974d73b9900f596fc71fb9c17d2c2ecc

[ethana2]

How does this affect folks like me running Compiz on Nouveau?

[jrch2k8]

I think the point is that once you know how to implement shader-assisted decode for one API and format then implementing it for another API or format is much less daunting.

can't be said better than that + the shaders code shouldn't change that much anyway