Hi,
In the distance vectors one finds the information about how often a
data reference is reused (how many iterations we have to wait until
the data referenced is used again). This patch uses a very dumb
heuristic, maybe not very useful in the way it is in this patch, but
maybe with the help of experiments could be improved. Basically the
function that computes the profitability is sum_distances_on_loop.
Using this function could be non profitable in some strange cases,
when some of the distances are huge, thus if somebody wants to play
with it and propose better heuristics, welcome!
Dan, could you verify that the comments that I've added or fixed are
correct? Thanks.