I am not a programmer but i think this is a new step for general GPU computing !
http://developer.amd.com/gpu/ATIStreamSDK/Pages/default.aspx
http://www.hardocp.com/news/2010/05/03/ati_stream_sdk_now_available/
I'm an active member of that forum. I am waiting for the Open64 group to decide to add Windows-64 as a development host. I don't want to have to pay MS $1000 for VC if I can help it. The forum just had a poll among their members and asked what they wanted most for new improvements. Answers were overwhelmingly in favour of adding Windows support. An inside guy told me he would get back with me on the details and whether they also planned to support the SDK you sited above. Currently, Open64 only runs under Linux. Apple has their own support for CUDA (comes with the OS) but has signed on for the OpenCL effort.
For those of you who are interested, OpenCL is for high level programmers who want to make use of their knowledge of C to take advantage of the GPUs in ATI video graphic cards. CUDA is a similar NVidia-only technology, so if you want to be architecture independent, you should use the up and coming OpenCL standard instead of CUDA. To follow suite, NVidia has also decided to support OpenCL so that they will not be left out of the standardization effort. Supposedly, though, CUDA code typically performs better than the same equivalent software written in OpenCL. However, OpenCL gives you more low-level control over the process. The price is more complexity. The number of companies that have signed up to support OpenCL is staggering so, seems like CUDA will go by the wayside. NVidia's documentation is confusing as CUDA is a trademark of theirs so the OpenCL docs are sprinkled with CUDA marketing lingo which is probably a bad idea. Very confusing. Of course, many of you know I get confused easily when documentation is not accurate (sorry) ::)
Anyway, ATI has an SDK for OpenCL developers and they have an ATI Stream SDK for those who want to write compilers and generate lower-level code. The SDK also is open enough that you can program straight to the machine code if you want to. But the Stream level code will isolate your tools better from low-level implementation details and architectural changes so that is the way I think I am going to go. I am writing a compiler for a parallel programming language that has been under development since 1988. I have a working editor and parser but I will start on the code generator when all of my requirements have been met. In the meanwhile...I'm Terragenning!
I would be pleased if you share news in the future on this subject with us here , PabloMack :)
I'd be glad to. Come to think of it, my Avatar is the logo for the programming language. You can see it animationed on YouTube ;D
http://www.youtube.com/watch?v=xHNcxuJF7no
:)