I am playing around the reduction sample code from SDK and also reading the white paper written by Harris. Here is one question, in the “Complete Unrolling” version (Reduction #6, I believe). The SDK source uses function template to bring in the block size. I don’t understand this actually. I am sure there is some reason to do this, but would that be easier to just use blockDim.x instead of using function template? Could some one point out this for me?