Adding a mate searching thread to my engine

[Steve Maughan]

To be clear, the tactical thread would still use the hash table to make cutoffs e.g. if the hash entry is upperbound at +10 and alpha is +mate, then you'd do a cutoff. It just wouldn't "push" it's search results to the hash table unless a genuine mate had been found. What you want avoid is the hash table being filled with practically useless entries that basically say "this position is less than mate".

Steve

[Uri Blass]

To be clear, the tactical thread would still use the hash table to make cutoffs e.g. if the hash entry is upperbound at +10 and alpha is +mate, then you'd do a cutoff. It just wouldn't "push" it's search results to the hash table unless a genuine mate had been found. What you want avoid is the hash table being filled with practically useless entries that basically say "this position is less than mate".

Steve

You also want to avoid searching the same position to the same depth twice by the mate search when there is no mate.
I do not see how you do it if you do not store in the hash table an information like this position is less than a mate assuming you search to depth 9 plies or smaller.

[Ronald]

You could try to define a separate hashtable for the mate thread, look in both hashtables for the position, store regular positions in the separate table and store mate scores in both tables.

You can create a separate table fi by using one bit of the index as separator. Drawback is that you halved the space for regular search.

[Joerg Oster]

Let me answer this straightforward:
Return 0 at the leafs.
Remove all pruning.

You will end up with pure alphabeta which should be around 7 lines of code and will find forced wins for you.

If that recursion returns a win it is a forced win IE mate in X


You can still do iterative deepening when the number of open paths remain low

So no heuristics that could possibly help?
About pruning, maybe the only pruning I can make safely is mate distance pruning. I'm going to test that.

This https://github.com/joergoster/Stockfish/tree/huntsman simple approach shows quite some success.
Maybe it can provide some hints for you.

[Steve Maughan]

You could try to define a separate hashtable for the mate thread, look in both hashtables for the position, store regular positions in the separate table and store mate scores in both tables.

You can create a separate table fi by using one bit of the index as separator. Drawback is that you halved the space for regular search.

Yes — that should work!!

[Cardoso]

You could try to define a separate hashtable for the mate thread, look in both hashtables for the position, store regular positions in the separate table and store mate scores in both tables.

You can create a separate table fi by using one bit of the index as separator. Drawback is that you halved the space for regular search.

I already had a separate a separate hashtable for the mate searcher, it just didn't occurred me I should store mate scores in both tables. Thanks Ronald.