Hi,
I think that OpenCL must have a scan/reduce/sort library like it exists for CUDA. By example there are :
http://code.google.com/p/thrust/
http://code.google.com/p/cudpp/
But I have see nothing for OpenCL !!!
So, I think that together we can create a project on googlecode (by example) and provide this kind of library.
What do you think ? Does someone is interested to collaborate ?