TalkChess.com

[Daniel Shawul]

Fortunately I didn't encounter such a high number of pieces, so other than the need to generate random number at runtime, I didn't need to reduce table size. I am sure the idea of hasing bitboard files/rank & diagonal states is used in some of the bb techniques to reduce table size. Your method that hashes the piece type and square should work fine in general I think but quality may not be good as multiplication can reduce number of set bits in the lower parts of the hash key. I think if the formula is modified as key[piece] * magic_number + key[square] it would look like an LCG generator. But I am not saying that this has advantages over what you are using. Ofcourse another way is to generate a unique set of random numbers every time the hashkey is required, but for efficiency nothing more than an LCG I think. That will have only one multiplication as the method you use now.

Edit
First I wrote about using one hashkey : key[piece] * magic + square but this will probably fail very badly. I am not sure if you do an "add" or "xor" to get your keys so an operation like multiplication should not have a problem I guess.
_________________
https://sites.google.com/site/dshawul/
https://github.com/dshawul