How to improve further ?

[bob]

There are things that are fast, but you really must want to do those thngs and do a lot of them to earn back the overhead. On a simple engine like the one we are talking about I don't think there would be enough such things to make it pay off.

I think magic move generation would make it a win, by itself. The downside is that it takes time to become competent at this kind of programming, it is initially a more obscure way of programming until you get into "bitmap mode".

[Daniel Shawul]

Bitboards are supposedly better for mobility calculation. One thing that I found beneficail was to construct a bitboard of pawns right in eval (this takes nothing at all because pawn tt have over 95% hit rate). I used that along with pradu's majics for pawn evaluation, mobility calculation, and also crude king safety calculation. Piece blockade is ignored for mobility here. That took me only a day to implement and i didn't need a major over-haul of code to test what they are good at. And with every popcnt call I did performance drops , that is on 32 bits. Right now it is still weaker than previous version, but the speed up I got from removing the old attack table generation code is making me hopeful to research this further..