Yes, the design of Chest is THAT old, that no one thought about 64 bitbnst wrote:From my tests there is almost no speed gain, but with the -M <megatbytes> option it is possible to use more memory for the hash table with the 64-bit compile I think.
Chest uses 64-bit entities in its calculation which could have been speeded up a lot, but because the program is designed for 32-bit it doesn't help just to recompile. I think Heiner Marxen himself or somebody with time to understand the inner workings of the program has to change it. If you understand the program I don't think it takes much more than some change in the way some elements are declared get significant speed gains from 64-bit execution. I have ported similar program from 32-bit to 64-bit and gotten a doubling of the speed.
Regards
Andreas
quantities at that time (not me at least). And the most time consuming
parts of Chest are move generation and move execution, both of which are
not exactly bound by memory throughput. They are quite loaded with
branches (conditionl execution) and handle mostly 1-byte and 2-byte
quantities... and attack data:
Attack boards in Chest are not the standard bit boards, coding 64 squares,
but rather 32 bit quantities, with 1 bit for each piece.
Empty squares cannot be part of such a "bit board".
Even with a careful rework I would not expect much speed from
a 64-bit compile for Chest. Whether a change of the attack board design
from 32-bit for pieces to 64 bit for squares would yield any speed,
I don't know. May be yes, but may be not at all. Chest is quite sensitive
to the L1-Cache hit rate, and that is hard to predict.
And: such a change would be a LOT of work.
Cheers,
Heiner