Author: Dan Newman
Date: 21:04:38 02/02/00
Go up one level in this thread
On February 02, 2000 at 22:23:05, Michael Neish wrote: >On February 02, 2000 at 22:18:29, Andrew Dados wrote: > >> Who told you bitboards are magic? :) >>... Just another chess representation... >>-Andrew- > >Actually I told myself they were magic. Well, since bit operations are >obviously very fast I thought they would dramatically improve performance. It >turns out they actually reduce it (in my case -- and maybe I'm doing something >wrong). However, it should make certain evaluations faster, so as I keep adding >more knowledge to it, I expect it will be superior in the long run. > >Thanks, > >Mike. I first tried bitboards maybe 2 yrs ago and decided they'd be a good deal slower than (say) 0x88. I tried a few experiments in move generation that seemed to indicate bitboard move generation was about half the speed I could get with 0x88, so I concentrated on 0x88 and got what I thought were fair results (aprox. 130 knps on a P6/200 with almost no eval). Then I decided to try bitboards again--just for the heck of it--and now find that I'm now getting much better performance than I thought possible--much better than with 0x88. (I'm running just under 500 knps on a P3/500 with quite a few eval terms thrown in too.) The change is due in part to using the BSF and BSR operations for doing first_one() and last_one() instead of the C++ code I had been using, and also in part to the compiler I'm using now (MSVC 5.0). (I had been using Watcom, and with Watcom the bitboard code is a bit slower than the 0x88. I think some of this is due to the way I'm getting at the BSF and BSR operations in MSVC vs Watcom.) I'm not sure where the biggest boost came from, but the MS compiler made my bitboard program (Shrike) go 50% faster and made the 0x88 program (SkyBlue) go only 17% faster over the Watcom compiles... -Dan.
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.