I have a quick question concering the strategy, that the Cilk Plus runtime system uses to divide up cilk_for iterations. I saw a few examples that showed how this worked when you have a number of iterations that is a power of 2, but how is this done in a general case? Does the grain size play a role here? Or is the number of iterations upped to the next power of 2 and then some kind of work stealing kicks in by those workers, who have nothing to do?
If anyone could give me some info on that, or direct me to a file/documentation that explains how this works, I would be very grateful!