Author: F. Huber
Date: 08:16:38 11/03/05
Go up one level in this thread
On November 03, 2005 at 10:49:01, Joachim Rang wrote: >On November 03, 2005 at 10:39:17, Uri Blass wrote: > >>On November 03, 2005 at 09:21:17, Uri Blass wrote: >> >>>On November 03, 2005 at 08:40:02, Ignacio Santos Crespo wrote: >>> >>>>Hi Uri, >>>> >>>>Thank you for your answer. The following position is interesting: Fruit crash >>>>and Movei crash after depth 16. >>>> >>>>Mate in 69 >>>> >>>>[D]1K1n2q1/1p4pR/1prPPpPn/1RB1p1rP/8/1B2p3/2p2b2/1k2N3 w - - 0 1 >>>> >>>>Regards, >>>>Ignacio. >>> >>> >>>Thanks for posting it. >>> >>>I could verify that the public version crash in console mode at depth 16 >>>when latest private version even crash at depth 14. >>> >>>The question is also what is the reason that fruit crashed. >>>If it crashed because the interface cannot get long pv then it may be a bug in >>>the interface but if it crashed because of another reason then it is a bug in >>>the engine. >>> >>>Uri >> >>More details >> >>The version that was compiled by Dan Crobit crash at depth 14 >>slower version that was compiled by me does not crash and can get even depth >>17(maybe the reason is that I added some printf that print the maximal ply every >>time that it get it and for some reason the bug with printf does not lead to >>crash but only to high depth that is not logical(I get branching factor that is >>close to 1.2 or 1.3 at depth 14-17) >> >>the maximal selective depth of movei is 100 so it cannot see mate in 69 >> >>At depth 13 the maximal selective depth that movei search in this position is 97 >>At depth 14 it get selective depth of 100 and it seems that the version that Dan >>Corbit compiled crash near the time that it gets selective depth of 100 plies. >> >>I suspect that the problem may be that my arrays are too short at some point >>I tried to reduce the maximal search depth to 99 and it did not help. >> >>I tried again to reduce the maximal search depth to 97 and it did not crash but >>it also does not get depth 17 fast like the previous version and at the time of >>posting it only started depth 16 that is clearly logical because I do not expect >>branching factor of 1.2 and branching factor that is slightly more than 2 is >>clearly more logical. >> >> >>Uri > > >it has to do with the compiler. Fruit compiled by Bryan Hofmann (official >version ) crashes while other dev-compile from me does not. Probably some kind >of optimization which can cause a crash in rare circumstances. > >Joachim Of course I don´t know the internal details of Fruit, but maybe it has to do with the stacksize allocated in the compiler!? I had exactly the same problem in one of the last WinChest compilations, when the maximum search depth was increased from 62 to 126 - solving a mate longer than about 100 made WinChest crash. After having found out that the problem was the stacksize (the reason was of course needing more memory for the recursive calls), I doubled it from the compiler default of 1MB to 2MB and the problem was solved! :-) Maybe this also helps for Fruit ... Regards, Franz.
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.