For Now Intel's Beignet Seems Better Off Than Radeon Gallium3D Clover

[tstellar]

For me only Luxball works, but Room and Sala GPU lockup ... not that i need opencl much on Kabini just tested it it has only 2 compute cores but it is good to see something goes from "not working" to "it works" status

Sala works for me on Bonaire, which is a Sea Islands part, so you might want to update your LLVM, though it's possible some of the WIP patches on my local branch is what makes it work. Room also lockups up for me, and I haven't actually been able to test luxball, because I get an unsupported image format error message from Luxmark when it tries to load one of its resource files. My guess is I'm missing a USE flag on
some kind of image library Luxmark is using, but it's not an OpenCL issue. It's cool to see other people have it working.

As Veerappan already pointed out, when apps fail with the ' "static" storage class specifier not supported' error, it is a bug in the application. static storage class specifiers are a 1.2 feature and clover is currently reporting 1.1.

Not mentioned in this article, is the ViennaCL benchmark, which works for me too on Bonaire.

[smitty3268]

...

Any plans to add image support sometime soon? It seems like a pretty major piece of missing functionality at this point.

[blackiwid]

hmm I think its logical that intel would be faster if they have a vendor-specific driver vs abstracting some code for every gpu.

And of course Intel invests more money. But I dont like that, its not even majorly a freedom thing, but a classical opensource thing, u want to have companies work together not concurent each other inventing the wheel a 100 times again.

On the other hand I dont care so much about opencl stuff, dont think I get much of it, as a i3wm and emacs user. Maybe it could fix browsers at least on the speed side, but I even doubt that.

[dungeon]

Sala works for me on Bonaire, which is a Sea Islands part, so you might want to update your LLVM, though it's possible some of the WIP patches on my local branch is what makes it work. Room also lockups up for me, and I haven't actually been able to test luxball, because I get an unsupported image format error message from Luxmark when it tries to load one of its resource files. My guess is I'm missing a USE flag on
some kind of image library Luxmark is using, but it's not an OpenCL issue. It's cool to see other people have it working.

It must be i clicked two times on Room or something so yeah doublechecked it Sala works too... practicaly everything works for me there: Luxball, Luxball Sky, Luxball HDR and Sala, besides Room scene but that is known thing to you .

[tstellar]

Any plans to add image support sometime soon? It seems like a pretty major piece of missing functionality at this point.

I don't currently have any plans to work on this, but this could always change. Image support is a lot of work, and I only know of one application which uses it, x264, but I'm not sure how much value there is in OpenCL x264, since we now have VCE support. If anyone has an OpenCL program they like which requires image support, please file a bug at bugs.freedesktop.org. This will help me get an idea how important this feature is relative to other things.

[dungeon]

Someone posted here DarkTable wants image support

http://www.phoronix.com/forums/showthread.php?106181-For-Now-Intel-s-Beignet-Seems-Better-Off-Than-Radeon-Gallium3D-Clover&p=439955#post439955

Tried now ViennaCL benchmarks you mentioned so those also works on Kabini, just one i think complained about double precision but that is not supported on Kabini AFAIK

[darkbasic]

I agree that x264 OpenCL is useless, however darktable's isn't and I remember there was at least another app making use of image support.

[Redi44]

+1 for Darktable

[System25]

AFAIK DarkTable requires OpenCL image support, which is not yet implemented in the open source OpenCL implementation.

And that's a big omission from AMD's side. This app is really computation-heavy and it takes about 10 seconds to do computations for 10Mpix photo on 8-core AMD FX-8120, keeping all 8 cores busy. Needless to say it could be better, OpenCL is here due to mentioned reason. It's kinda sad that AMD's GPU can't help with it.

You'll need to use the AMD Catalyst Linux driver to get OpenCL image support for now.

Nope, thanks. Use this bugged crap yourself. It would hang my system in a matter of 3-4 days under OpenGL loads. Needless to say, it is annoying. And then it seems nobody cares about these bugs. I even do not have idea how to report it for Catalyst. Linux kernel devs will tell me GTFO with tainted kernel, Catalyst lacks bug tracker. This condition persists for years and I really gave up on Catalyst. Open driver can withstand weeks to months of uptime if I do not mess with OpenCL. Also, opensource driver does not fails on distro version upgrades, unline Catalyst which can easily leave me in system with black screen due to kernel module issues.

If you guys are up for all this accelerated computing buzz, it is really strange to see OpenCL in such a poor state in opensource driver. OpenCL has finally got some traction around various Linux software. And guess what? We can see your driver fails most workloads. Dammit. AMD's GPUs hardware seems to be way better than Intel's crap. However shitty software literally kills this adantage.

[bridgman]

I realize you don't need it now, but the Catalyst bug tracker the devs look at is here :

http://ati.cchtml.com/