Author: Dieter Buerssner
Date: 10:55:33 05/30/05
Go up one level in this thread
On May 30, 2005 at 07:45:30, Ross Boyd wrote: >so the last line in my makemove was: > >return ++nodes; > >This works fine until the 32 bit nodes counter rolled over.... and then it >disastrously returned FALSE... CRUNCH! I had a very similar bug in Yace. Some user reported a crash after a very long search (I cannot remember, if he did reproduce it). It was a time, when I still had K6-2. The engine did not crash after the first overflow of the node counter. I tried to reproduce the problem, but couldn't. I could not use the same amount of hash as the user (had only 56 M RAM at that time). It turned out, that the overflow to zero would cause a crash, when it happened in normal search (it didn't do anything harmful when it happened during qsearch). So, with different size of hash, it did not happen in that reported position here, due to luck (in my engine, typically over 70% of nodes++ happen in qsearch). Also, it is not a test, you want to do often - let run the engine for several node counter overflows on a slow computer (IIRC it took a day for one overflow). Because one overflow passed without problem in the log I got sent, I did not immediately think of this problem. >It sure felt good to finally fix it. Sure! Regards, Dieter
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.