Author: Robert Hyatt
Date: 07:00:29 09/16/99
Go up one level in this thread
On September 16, 1999 at 01:31:14, David Eppstein wrote: >On September 15, 1999 at 23:08:38, Robert Hyatt wrote: >>>Aha. While we are willing to accept very rare false hits, this technique >>>will work. If false hits are not at all acceptable, this will not work. >>>Heiner >> >>Note that if false hits are not at all acceptable, hashing can not be used at >>all... because there is _definitely_ a non-zero probability of collisions. > >You could still use hashing, if you stored the whole position instead of just >your key. Not very practical, though, even after the long threads here about >how to store a whole position in few bits. Much better to take the risk of a >very rare false hit. > >>Tim's scheme does solve the issue by basically destroying entries where a race >>to store occurs... which is ok. And the performance improvement is non-trivial >>to boot... > >Tim's scheme is very cute. SMP programming takes a whole new mind-set. And it is _full_ of such cute things... _once_ you are experienced enough to recognize the right circumstances and then use the right trick. The bitboard stuff is the same, it is full of 'tricks' (most of which are obvious to the crypto guys but not the rest of us that aren't twiddling bits all day long).
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.