Bridgman Is No Longer "The AMD Open-Source Guy"

[twriter]

So maybe you can answer the million dollar question then... UVD open source driver... when?

John already answered that question. We've been doing a lot of work but we don't yet know if/when we can release. That's all I can say for now.

Tim

[darkbasic]

What about h.264/webm shader based decoding? Is there someone still working on it?

[89c51]

What about h.264/webm shader based decoding? Is there someone still working on it?

That doesn't depend on AMD afaik. Anyone with enough skills can do this right now. I think the guy that was writing the h.264 got hired by AMD and works on other stuff now.

[darkbasic]

Do you mean Tom Stellar? As far as I know he started working on video decode *after* being hired by amd...
I also remember someone else working on webm...

[smitty3268]

Do you mean Tom Stellar? As far as I know he started working on video decode *after* being hired by amd...
I also remember someone else working on webm...

The guy working on video decode for AMD was Christian König, i think, and he continued in that role after getting hired for AMD. I'm not sure he ever got beyond MPEG2, though - i don't think there's been much work on h264 from the AMD side, or at least not publicly.

I think lately he's been working on getting the radeonsi driver working.

[HokTar]

The guy working on video decode for AMD was Christian König, i think, and he continued in that role after getting hired for AMD. I'm not sure he ever got beyond MPEG2, though - i don't think there's been much work on h264 from the AMD side, or at least not publicly.

I think lately he's been working on getting the radeonsi driver working.

Yep, it was indeed Christian. However, all that's visible from AMD these days are for either radeonsi or r600/llvm improvements. Not something that is useful right now for most people...

Thanks for Bridgman so far and please keep us informed about HSA!
Welcome Tim!

[darkbasic]

The guy working on video decode for AMD was Christian König

You're right, I'm sorry Christian

[89c51]

LLVM is something we need for OpenCL i think. And yes i was referring to C.Konig.

[Deathsimple]

What about h.264/webm shader based decoding? Is there someone still working on it?

Well, if somebody is still working on it than it's probably me

No really, I started evaluating if shader based h.264 makes sense or not when I started working for AMD, and got a demo working using parts of ffmpeg code and the simple MC and IDCT shaders originally used for MPEG2 acceleration.

Unfortunately the outcome was that even with a really lot of effort (and we are talking about really lot of effort) the improvement that could be archived would never be more than 20%, while full hardware based decoding promised to easily accelerate more than 90% of the whole decoding process.

So what I've done after that is actually looking if, when, how we could ever be able to release UVD and where the pitfalls are. But I can't really say when and or if we are going to release it, it might be tomorrow, it might be next year, but it could also be never.

Christian.

[oliver]

Well, if somebody is still working on it than it's probably me

No really, I started evaluating if shader based h.264 makes sense or not when I started working for AMD, and got a demo working using parts of ffmpeg code and the simple MC and IDCT shaders originally used for MPEG2 acceleration.

Unfortunately the outcome was that even with a really lot of effort (and we are talking about really lot of effort) the improvement that could be archived would never be more than 20%, while full hardware based decoding promised to easily accelerate more than 90% of the whole decoding process.

So what I've done after that is actually looking if, when, how we could ever be able to release UVD and where the pitfalls are. But I can't really say when and or if we are going to release it, it might be tomorrow, it might be next year, but it could also be never.

Christian.

So basically you are saying, that shaders just aren't up for the task? E.g. the workload not parallelizeble enough or the shaders just aren't powerful enough for the amount of work needing to be done etc...

Those are very interesting findings actually basically meaning, "uvd" in any form (xvba, vpdau, cedarx etc etc) is here to stay, and won't ever be replaced by shaders/opencl.