Author: Robert Hyatt
Date: 19:02:53 07/10/02
Go up one level in this thread
On July 10, 2002 at 14:31:14, Georg v. Zimmermann wrote: > >>Look at what I store in the table... the _entire_ 64 bit hash signature. >>I use the rightmost N bits (N=log2(tablesize)) to compute a table address, >>but then I compare _all_ 64 bits of the current signature to the 64 bits >>of the stored signature before I call it a "match". >> >>I don't follow your "8 bits unchecked" at all... > >Yes you are right of course. >I dont know why I missed that you xor the rightmost 32 bits into word2 before >comparing, sorry. > >But maybe that isnt really needed. If we have a table with at least 2^21 >entries, which sounds reasonable, even if you dont stuff the rightmost 32 bits >into the hash entry itself, you have 21+32 bits, almost as good as 64. > >Georg In Cray Blitz I actually did that, as we used 12 bytes for a position rather than 16. In Crafty I was a bit more lazy and just stored the entire signature. I think either is ok, probability-wise, at today's search speeds. A few years might change that.
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.