Crash by Crystal v1.1

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
zullil
Posts: 5606
Joined: Mon Jan 08, 2007 11:31 pm
Location: PA USA
Full name: Louis Zulli

Crash by Crystal v1.1

Post by zullil » Thu Sep 05, 2019 3:23 pm

Very preliminary report. Crystal v1.1 (bench 7410759) segfaulted:

Code: Select all

kernel: [84966.051797] crystal[26855]: segfault at b3a0000017d ip 000000000042095f sp 00007f4342fc6490 error 4 in crystal[400000+3e000]
Using 16 threads, 16384 MB hash, all other settings at default. Linux.

[EDIT] Was actually using MultiPV = 2.

supersharp77
Posts: 746
Joined: Sat Jul 05, 2014 5:54 am
Location: Southwest USA

Re: Crash by Crystal v1.1

Post by supersharp77 » Fri Sep 06, 2019 2:45 am

zullil wrote:
Thu Sep 05, 2019 3:23 pm
Very preliminary report. Crystal v1.1 (bench 7410759) segfaulted:

Code: Select all

kernel: [84966.051797] crystal[26855]: segfault at b3a0000017d ip 000000000042095f sp 00007f4342fc6490 error 4 in crystal[400000+3e000]
Using 16 threads, 16384 MB hash, all other settings at default. Linux.

[EDIT] Was actually using MultiPV = 2.
My version of Crystal (crusty) either crashed or lost on time...dunno yet..still checking...the more books the more time issues
... :lol: :wink:

User avatar
Eelco de Groot
Posts: 4148
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: Crash by Crystal v1.1

Post by Eelco de Groot » Fri Sep 06, 2019 8:08 am

zullil wrote:
Thu Sep 05, 2019 3:23 pm
Very preliminary report. Crystal v1.1 (bench 7410759) segfaulted:

Code: Select all

kernel: [84966.051797] crystal[26855]: segfault at b3a0000017d ip 000000000042095f sp 00007f4342fc6490 error 4 in crystal[400000+3e000]
Using 16 threads, 16384 MB hash, all other settings at default. Linux.

[EDIT] Was actually using MultiPV = 2.
I have seen no crashes here but this in my modified version of Crystal. I hope Joseph does not mind my making a fork of sorts, nothing officicial. It is called Palantir Crystal :P To keep it bit in Lord of the Rings territory and assuming Crystal is something like a crystal ball.


I only used Analysis mode, single thread, no Multi PV.

This part (around line 987 in search.cpp) I think is a bit risky. I don't see how it could produce a segfault by itself, so possibly it is not related to that, but if you put in double extension even only in PV nodes, it is possible that your search never returns. I can see it is tempting though and have not found a better thing even though I do not really know what exactly code is doing... My changed Crystal version most of the times does not go deeper than around depth 25 Then search explosion. I removed the 'else' before going into Singular extension because sometimes Singular extensions can be something of a reduction in my version (so helping against exploding searches) but that means I have to change the && !gamecycle later into something else, or search does not go from the first gameCycle extension possibility into SE.

Code: Select all

      
      // Step 13. Extensions (~70 Elo)
      if (    gameCycle
          && (alpha > VALUE_DRAW || (ss->ply % 2 == 0))
          && (depth < 5 * ONE_PLY || PvNode))
          extension = (2 - (ss->ply % 2 == 0 && !PvNode)) * ONE_PLY;

      // Singular extension search (~60 Elo). If all moves but one fail low on a
      // search of (alpha-s, beta-s), and just one fails high on (alpha, beta),
      // then that move is singular and should be extended. To verify this we do
      // a reduced search on all the other moves but the ttMove and if the
      // result is lower than ttValue minus a margin then we will extend the ttMove.
      if ( // else removed here
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

corres
Posts: 1569
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Crash by Crystal v1.1

Post by corres » Fri Sep 06, 2019 1:27 pm

I run Crystal v1.1 on my machine (Ryzen 7 1800x 8x4000Mhz Windows 7 64bits) and I did not experienced any running issue. But I was disappointed by its power: Its result was against Stockfish (190825) in a blitz match
(1min + 2sec/move) 0:6 with 24 draw (30 games with opening book and reversed color). I planned the match of 100 games but seeing the result I stopped it. I started a new match with 1min + 15 sec/move but I also stopped after four games: Stockfish190825 : Crystal1.1 = 3.5 : 0.5...
I repeated the endgame puzzle test (solving time was 90 sec/position) and I got:
1. Crystal 300819 (v1.1) - 271 solved out of 286 (average solving time = 6.4sec)
2. Stockfish 190825 - 259 solved out of 286 (average solving time = 9.3sec)
So Crystal is a good puzzle solver but it is weak for playing games and analyzing normal chess positions,
the endgame positions may be the exceptions.

Paloma
Posts: 829
Joined: Thu Dec 25, 2008 8:07 pm

Re: Crash by Crystal v1.1

Post by Paloma » Fri Sep 06, 2019 10:11 pm

corres wrote:
Fri Sep 06, 2019 1:27 pm
...
I repeated the endgame puzzle test (solving time was 90 sec/position) and I got:
1. Crystal 300819 (v1.1) - 271 solved out of 286 (average solving time = 6.4sec)
2. Stockfish 190825 - 259 solved out of 286 (average solving time = 9.3sec)
So Crystal is a good puzzle solver but it is weak for playing games and analyzing normal chess positions,
the endgame positions may be the exceptions.
Which puzzle test was it? ACT from Rybka-Forum? But ACT has 287 pos.

corres
Posts: 1569
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Crash by Crystal v1.1

Post by corres » Sat Sep 07, 2019 8:29 am

Paloma wrote:
Fri Sep 06, 2019 10:11 pm
corres wrote:
Fri Sep 06, 2019 1:27 pm
...
I repeated the endgame puzzle test (solving time was 90 sec/position) and I got:
1. Crystal 300819 (v1.1) - 271 solved out of 286 (average solving time = 6.4sec)
2. Stockfish 190825 - 259 solved out of 286 (average solving time = 9.3sec)
So Crystal is a good puzzle solver but it is weak for playing games and analyzing normal chess positions,
the endgame positions may be the exceptions.
Which puzzle test was it? ACT from Rybka-Forum? But ACT has 287 pos.
http://www.mediafire.com/file/8yxjk8dq9 ... 6.2019.zip
It contains 286 positions.

Paloma
Posts: 829
Joined: Thu Dec 25, 2008 8:07 pm

Re: Crash by Crystal v1.1

Post by Paloma » Sat Sep 07, 2019 10:13 pm

Hmm, i downloaded this again and it contains 287 pos.
At least the ACT.epd (23.529 Bytes)

corres
Posts: 1569
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Crash by Crystal v1.1

Post by corres » Sat Sep 07, 2019 10:33 pm

Paloma wrote:
Sat Sep 07, 2019 10:13 pm
Hmm, i downloaded this again and it contains 287 pos.
At least the ACT.epd (23.529 Bytes)
I used the .pgn file, as you could guess it.
What is the issue? The Alternative Chess Test.pgn also contains positions.
I use .pgn or .cbf files only and no .epd.
If you need in .pgn the 287.position use a tool to transform it or ask it from the author.

Paloma
Posts: 829
Joined: Thu Dec 25, 2008 8:07 pm

Re: Crash by Crystal v1.1

Post by Paloma » Sun Sep 08, 2019 2:22 am

I don't want to bother you, but the pgn in my download (89.126 Bytes) also has 287 pos and not 286
Anyway, I just noticed the difference.

corres
Posts: 1569
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Crash by Crystal v1.1

Post by corres » Sun Sep 08, 2019 8:47 am

Paloma wrote:
Sun Sep 08, 2019 2:22 am
I don't want to bother you, but the pgn in my download (89.126 Bytes) also has 287 pos and not 286
Anyway, I just noticed the difference.
O.K.
What kind of chess GUI you use?
I use Chessbase (Fritz) GUI.
Maybe one of position in the .pgn file is defective and Chessbase GUI can not read it.
You may guess I did not leave it intentionally.
In your opinion which position lack?

Post Reply