Author: Gerd Isenberg
Date: 02:32:29 09/13/03
Go up one level in this thread
On September 13, 2003 at 04:48:48, Sune Fischer wrote: >On September 12, 2003 at 17:02:04, Gerd Isenberg wrote: >> >>Yes, but if "key" is whitePawns-blackPawns, bitboards as 64-bit ints, you may >>loose some pawns. That's the advantage with zobrist keys. > >True. >I've never tried the subtraction, interesting idea. >I would guess, that it doesn't spread as good as random zobrist keys, but it may >not be a big issue for a pawn hash. Yes, i believe it spreads good enough considering my hit rates. > >>I guess a 32-bit key, >>incremental updated if pawn structure changed, is good enough to address the >>pawn hash table. > >Why consider 32 bit? >You have the 64 bit zobrist table already in cache. I use the pawn-bitboard signature anyway to make sure that there is no collision. So i only need a 15- or 16-bit key as table index. > >You're a bitboarder, think 64 bit and drop the 32 bit kludges.. ;) > Sure, but if i need less than 33 bits it seems a waste of space and time to use 64 bits - even on opteron. You still have the 32-bit gp-registers with one byte shorter opcode. Note that "int" is still 32-bit in msc for AMD64. It is even possible to pack two 32-bit ints inside one 64-bit word and do simd-like processing. On the other hand one (or compiler) should avoid memory size mismatches. Gerd >-S. >>Gerd
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.