Working with moves or with positions.
Guillem Barnolas
I was looking to the source of KCChess, a pascal chess program, and I
see that they have two procedures that make a move and unmake it, so
when looking to a position they generate all the moves, and then start
making, evaluating and then unmaking.
In the computer chess programming web page, in xs4all or something
like that, in the pseudo-c search algorithms it seems it works with
positions instead of with movements, I mean, when generating the
moves, it stores the positions reached, so it's faster, because you
only have to evaluate the position, without making-unmaking.
Can anyone comment this, thanks..
Guillem.
Bo Persson
As usual, it depends :-)
Depending on the amount of data you store for a position and
how fast the memory in your computer is, the save-restore
(or copy-modify-discard) can take even more time than unmaking
the move.
Bo Persson
bo.pe...@mbox3.swipnet.se
Robert Hyatt
Guillem Barnolas <guill...@redestb.es> wrote:
:
: I was looking to the source of KCChess, a pascal chess program, and I
: see that they have two procedures that make a move and unmake it, so
: when looking to a position they generate all the moves, and then start
: making, evaluating and then unmaking.
: In the computer chess programming web page, in xs4all or something
: like that, in the pseudo-c search algorithms it seems it works with
: positions instead of with movements, I mean, when generating the
: moves, it stores the positions reached, so it's faster, because you
: only have to evaluate the position, without making-unmaking.
: Can anyone comment this, thanks..
: Guillem.
It's an idea that is sub-optimal for the PC, which has a very low memory
bandwidth. Early versions of Crafty were based on copy-update (copy old
position to new position, then make the move, so that the unmake is not
needed, just back up to the old position and copy again). This was not
so hot on the PC. Make/UnMake executes more instructions, but instructions
are far less important than memory loads/stores, since a P6/200 can execute
40 instructions (two at a time or more) while waiting on one cache miss to
be handled...
--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170
Guillem Barnolas
Robert Hyatt <hy...@cis.uab.edu> wrote:
>Guillem Barnolas <guill...@redestb.es> wrote:
>:
>: I was looking to the source of KCChess, a pascal chess program, and I
>: see that they have two procedures that make a move and unmake it, so
>: when looking to a position they generate all the moves, and then start
>: making, evaluating and then unmaking.
>: In the computer chess programming web page, in xs4all or something
>: like that, in the pseudo-c search algorithms it seems it works with
>: positions instead of with movements, I mean, when generating the
>: moves, it stores the positions reached, so it's faster, because you
>: only have to evaluate the position, without making-unmaking.
>: Can anyone comment this, thanks..
>: Guillem.
>It's an idea that is sub-optimal for the PC, which has a very low memory
>bandwidth. Early versions of Crafty were based on copy-update (copy old
>position to new position, then make the move, so that the unmake is not
>needed, just back up to the old position and copy again). This was not
>so hot on the PC. Make/UnMake executes more instructions, but instructions
>are far less important than memory loads/stores, since a P6/200 can execute
>40 instructions (two at a time or more) while waiting on one cache miss to
>be handled...
Thanks, very clear again...
Greetings, Guillem.
Greg Teets
> In the computer chess programming web page, in xs4all or something
>
Please give me the address for this page.
Thanks
Greg Teets
Cincinnati, OH
Komputer Korner
http://www.xs4all.nl/~verhelst/chess/programming.html
--
--
Komputer Korner
The inkompetent komputer
To send email take the 1 out of my address. My email address is
kor...@netcom.ca but take the 1 out before sending the email.
Greg Teets wrote in message <3569f79...@news.ececs.uc.edu>...