Author: Sune Fischer
Date: 14:12:16 08/19/03
Go up one level in this thread
On August 19, 2003 at 17:04:44, Dieter Buerssner wrote: >On August 19, 2003 at 16:52:43, Sune Fischer wrote: > >First I thought: don't clear the bits. With every new search, toggle the >"current_age_bit". Add little logic to Probe/Store functions (it should not >decrease efficiency much, because we need a slow memory look-up anyway), that >just assumes the entry is empty, when the bit does not fit. > >But at second thought, this would not really help. Because the new search might >not touch one entry, and the second new search would assume, that the entry is >current. So, there is again an influence of the history of the game due to TTs, >something that Johan wanted to avoid from the start. > >Another approach would be, to have the age bits in one seperate small table. >Assuming you have 64 Mb hash, with 16 bytes each entry, this would be 4 million >bits, or 1/2 Mb of RAM. Something that probably would stay in second level cache >most of the time on a modern computer. Clearing would use practically no time, >but accessing the second table would mean some overhead. > >Regards, >Dieter Lucky me I use 4 bits:) I figure that if 16 move old entries still exists in the hash when we get to that point, then the hash is not being terribly overfilled and there really is no problem. Besides I do always replace as part of the scheme, so I'll just usev one of the other bins. -S.
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.