Winboard Forum

I'm having troubles find a way to efficiently do a 3-move repetetion detection in a parallel tree-splitting search. Anyone here have any ideas on how to do this efficiently. Assume the search will split at shallow depths as well.

Efficient - not using extensive loops or large amounts of memory copying.

Hi Pradu
you shouldn't worry much about memory copying much.
splitting is done in very few places like (say 1% of of total nodes searched).
I just copy the repetition detection table when i split.
compared to board copying (piece list in my case) this is nothing at all.
At the tip of the tree splitting is not done usually because the cost of spliiting outweighs its gain, so there is really no need for you to split there.
In scorpio i copy the whole of repetion stack. It also includes moves that have already been played. Small optimization for me is to copy only those in the tree . position that are already been played can be copied at search startup and left untouched as they do not change during search.
regards
daniel