Computer Chess Club Archives


Search

Terms

Messages

Subject: Bitmap Type Re: Tinker 81 secs Re: Testing speed of "position visiting"

Author: Brian Richardson

Date: 04:24:11 04/24/00

Go up one level in this thread


Tinker uses "regular" MSVC __int64 non-rotated bitmaps, but without separate
queen maps--just rooks/queens and bishops/queens.  This helps move generation by
about 10% as I recall (saves all the ORs when looking for moves), but hurts a
little in eval() to separate just queens with wq(only) = wrq & wbq.  Bob's
*move++ method would save an additional 5-10% but I just left modified
gen_push() once I realized move gen less than 20% of overall time.

BTW raw move gen speed without recursion or in-check testing (but still MVV/LVA)
on PIII 733MHz:
perf
generated 48000000 moves in 4.47 seconds
10743062 per second
make/take 48000000 moves in 19.22 seconds
2497658 per second

PS  My hardware is pretty fast.  If you would like to email me a binary of your
routine I could run your test to see exactly how much of the difference is h/w
(or you could adjust from TSCP benchmarks if you have them handy).



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.