Thanks for the insightful information. I do have a better perspective on your offering now. Although, in response, I do have no doubt that the set of tools you mention reduces the amount of effort required to implement an application on a multi-core system such as the Cell, but I am not yet convinced that using these tools that you could significantly outperform (or even outperform) a hand-coded optimized implementation. We can continue to discuss this point over and over, but until an example is openly and readily available then I am still in the camp that for the performance intensive parts of your application where every cycle matters, you're still not going to get any better than a hand-coded assembly optimized implementation. Thanks again for your insightful article! As a member of the multi-core development community, I look forward to hearing more from you in the near future. »

Straight 'C' and native assembly will always outperform C++ in terms of speed and efficiency. I do like C++, but I don't use it for high performance critical code. As somebody who has had a great deal of experience working with the Cell Broadband Engine, to get the best performance out of your application, there is no replacement for directly using the SPU intrinsics and/or writing assembly language. Optimization of your data access patterns for optimal DMA transactions is equally if not more important to achieve the best performance and efficiency. And of course, arranging your algorithm such that it can be made parallel with minimal synchronization/data dependencies across processors is important too. »