You need only Brook+ 1.4 SDK, no need Cg SDK for Stream computing.
Current problem is to report signal back to CPU w/o copyi÷ whole array back.
Pephaps I found solution for most common case when there is only single signal per array.
In coding now..
(reduction kernels have some limitations that I don't quite understand. Maybe other ways possible too, will see what AMD forum guys answer...)