I never understood your post, but from Michael's translation I understand you are using Scala and wish to run similar code on a GPU. I wrote code in Java and wished to run it on multiple GPU's. Not to scare anyone, but am just now putting the finishing touches on something that I started in October 2007, originally in GLSL. It is not the only thing I have done since then, but this used almost 1.5 years of that. My process is very difficult to begin with. The OpenCL version needed to be very different from the Java version to make the data parallelization work well.
Unlike games, where the color of a "pixel" could off by a slight amount, and no one would ever know, in my case it is just wrong. I have over 2k worth of Java source just to compare the output of the OpenCL version with that of the Java "Reference" version. True, it is un-natural / awkward to have multiple versions, but I would never been able to do this without my Java version to tell me if it was right.
You might have lucked out though, because there is the beginnings of an OpenCL implementation of Scala. It does not use JOCL bindings, and I do not know how good it is. The potential payoff might be hugh if it works for you though. You would just need to compare the outputs from this and a conventional implementation, using only 1 source. See the Scala links from:
I still kind of doubt the performance increase would be near as good as the multiple source method, but if this is not commercial, then it might be good enough.