OpenCL 1.1 Specification Released

[brent]

I'm not sure whether shader-based decode acceleration is very useful. VLD cannot be accelerated, but this is a very CPU-intensive task, especially when CABAC is used and at high bitrates - and that's where decode acceleration is most needed!
As far as I know the shader-based acceleration implementations that exist already need quite a bit of shader horsepower. So, no acceleration on low-end GPUs or IGPs. Plus, much power consumption.

[deanjo]

nvidia just failt to dev a shader based acceleration metod..

They had no reason to build one. If you want to use GLSL acceleration that still can be done on a 8800 GTS. XBMC has had this option available for quite some time now.

[Veerappan]

Well, it's kind of sad that the OS drivers don't have support for it, but does it really matter, at least at the moment? I mean, is there really anything that take advantage of OpenCL? With OS drivers still not supporting OpenGL 3(nor 4), without good acceleration for video playback(at least not for my r700) and with rather poor performaces for those still without a working Gallivm3D driver(in particular after the KMS switch), IMO the OpenCL support should be at the very last position on the OS drivers' TODO list.

Re: video playback acceleration.

If there were a fully working Gallium OpenCL state tracker, it would be feasible for someone to write an OpenCL version of the decoder for any video codec... and I'm about to start writing a thesis proposal to do just that (with VP8). I'm planning on targeting Nvidia+AMD's OpenCL run-times/libraries, but the moment there's a working Gallium OpenCL state tracker, it would work there too.

Of course, I'll need to actually get the proposal written up, approved, and an advisor will need to be found, but it should work.... and hopefully it'll be simpler to code/maintain than a full OpenGL+GLSL implementation (which has been partially done for h.264). Hopefully the overhead won't kill the performance too much.

[jrch2k8]

think that is only partly true... since if for instance openCL requires features in the hardware driver that no other state tracker has needed yet that would still have to be implemented in the driver so drivers still have to be worked on independantly to some degree though probably not nearly as much as before.... thats just how it seems to me.

yeah you are rigth gallium won't provide 100% for the code for the 100% of the hardware outhere, the trackers contains the most common code that is pretty much hardware independant, ofc after the trackers is there you need the driver of that piece of hardware to be aware of that trackers.

is pretty much like mesa(common code hardware independant) and dri/drm/kms(hardware dependant code )

[cb88]

Well, you gotta love Gallium3D because once the state tracker is there; it is there for ever and for every card that has a driver out there.

think that is only partly true... since if for instance openCL requires features in the hardware driver that no other state tracker has needed yet that would still have to be implemented in the driver so drivers still have to be worked on independantly to some degree though probably not nearly as much as before.... thats just how it seems to me.

[V!NCENT]

Well, you gotta love Gallium3D because once the state tracker is there; it is there for ever and for every card that has a driver out there.

[chaos386]

Agreed, but (a) I wasn't aware of any open source OpenCL-on-CPU implementations, and (b) at least this would allow chaos386 to stay with the open source graphics drivers (probably).

Oh, I'm running fglrx (5850 owner); I don't have any philosophical opposition to closed source software. I was just curious if there was a pure OSS way of doing OpenCL, since I know there are people out there who prefer to do things that way.

I did download the Stream SDK and will be looking through the OpenCL docs on the AMD website, though, so thanks!

[madbiologist]

MisterIO wrote: Well, it's kind of sad that the OS drivers don't have support for it, but does it really matter, at least at the moment? I mean, is there really anything that take advantage of OpenCL?

[email protected] will soon be using OpenCL - see http://folding.typepad.com/news/2010...-or-later.html and http://folding.typepad.com/news/2010...st-update.html

[bridgman]

Agreed, but (a) I wasn't aware of any open source OpenCL-on-CPU implementations, and (b) at least this would allow chaos386 to stay with the open source graphics drivers (probably).

[Zhick]

Go to www.amd.com and download our Stream SDK. It provides OpenCL support for both CPU and GPU.

That's closed-source as well though. I'd imagine that someone who refuses to use proprietary graphic-drivers to get OpenCL would also be reluctant to use a proprietary software-implementation.