For Jim Ablett, about problems with new Typhoon builds

[Volker Pittlik]

Hi Jim,

...
Suggestion: delete 'makefile', rename 'gnumakefile' to 'makefile'
and edit and use that one.

I think I did exactly that way. However, I have made a copy of Gnumakefile now (Makefile) and have to delete the GNUmakefile before the Makefile is used.

You can have a smaller memory footprint by editing 'chess.h'.
Use these values for 65mb footprint.

Code: Select all

#define EVAL_HASH_TABLE_SIZE (699050) // was  32Mb (per thread) 2097152

#define PAWN_HASH_TABLE_SIZE (43690) // was 5.5Mb (per thread)  131072


That helps a bit but it still uses more than 100 MB. I can live with that, but it's not very nice of course.

...If you don't have a hammer handy add this line to beginning of 'util.c'

Code: Select all

#include <errno.h>


Shame on me...

Thank you very much. Mystically no crash occurred since xboard debug feature is on. Test restarted with the new executable.

Regards

Volker

[Jim Ablett]

New win32 build (revision 234) now on my homepage.

http://homepages.tesco.net/henry.ablett/jims.html

Stackdump errror persists, but doesn't cause any problem with running the engine.

Jim.

[Jim Ablett]

Forgot to adjust pawnhash/evalhash sizes.

This build (revision 234) uses 62mb >

http://www.orbitfiles.com/download/id530969350

Jim.

[Volker Pittlik]

It does not happen so often anymore but I still get some crashes from time to time. I have absolutely no idea why.

Code: Select all

3075246 <first : 49      0   153 339931       Kd2 Kb6 Ke3 Kc6 Kd3 Kd6 Ke3 Ke6 Kd2 Kd6 Kd3 Ke6 Ke3 Kd7 Kd3 Kd6 <=>

3075270 <second: 28 2 211 434636  Ka6-b6 Ke2-e3 Kb6-c6 Ke3-e2 Kc6-d6 Ke2-f2 Kd6-e6 Kf2-e2 Ke6-e7 Ke2-e3 Ke7-f6 Ke3-f2 Kf6-f7 Kf2-e3 Kf7-e7 Ke3-e2 Ke7-e6 Ke2-f2 Ke6-d6 Kf2-e3 Kd6-c7 Ke3-e2 Kc7-c6 Ke2-e3 Kc6-d7 Ke3-e2 Kd7-e8 Ke2-e3 Ke8-f7 Ke3-e2 Kf7-e6

3075908 <first : 50      0   219 465064       Kd2 Kb6 Ke3 Kc6 Kd3 Kd6 Ke3 Ke6 Kd2 Kd6 Kd3 Ke6 Ke3 Kd7 Kd3 Kd6 <=>

xboard: Error: first chess program (./typhoon --egtbpath /media/tbs --hash 2097152) exited unexpectedly

GameEnds(0, (null), 2)

I'll do nothing for some weeks and simply wait .

Volker

[Guenther Simon]

It does not happen so often anymore but I still get some crashes from time to time. I have absolutely no idea why.

Code: Select all

3075246 <first : 49      0   153 339931       Kd2 Kb6 Ke3 Kc6 Kd3 Kd6 Ke3 Ke6 Kd2 Kd6 Kd3 Ke6 Ke3 Kd7 Kd3 Kd6 <=>

3075270 <second: 28 2 211 434636  Ka6-b6 Ke2-e3 Kb6-c6 Ke3-e2 Kc6-d6 Ke2-f2 Kd6-e6 Kf2-e2 Ke6-e7 Ke2-e3 Ke7-f6 Ke3-f2 Kf6-f7 Kf2-e3 Kf7-e7 Ke3-e2 Ke7-e6 Ke2-f2 Ke6-d6 Kf2-e3 Kd6-c7 Ke3-e2 Kc7-c6 Ke2-e3 Kc6-d7 Ke3-e2 Kd7-e8 Ke2-e3 Ke8-f7 Ke3-e2 Kf7-e6

3075908 <first : 50      0   219 465064       Kd2 Kb6 Ke3 Kc6 Kd3 Kd6 Ke3 Ke6 Kd2 Kd6 Kd3 Ke6 Ke3 Kd7 Kd3 Kd6 <=>

xboard: Error: first chess program (./typhoon --egtbpath /media/tbs --hash 2097152) exited unexpectedly

GameEnds(0, (null), 2)

I'll do nothing for some weeks and simply wait .

Volker

Volker was this a pawn endgame or an endgame K vs. K only?
(from the PV I deduct those 2 possibilities, could there be a problem
also to max depth? What type of tablebases is used 3/4 or other?)

Regards,
Guenther

[Volker Pittlik]

...
Volker was this a pawn endgame or an endgame K vs. K only?

It happened in this position: 8/8/k6p/p1p2p1P/1pP2Pp1/1P1K2P1/1P6/8 w - - 0 65

[diag]8/8/k6p/p1p2p1P/1pP2Pp1/1P1K2P1/1P6/8 w - - 0 65[/diag]

... What type of tablebases is used 3/4 or other?)...

All 4 piece TBs in a ramdisk.

Volker

[Guenther Simon]

...
Volker was this a pawn endgame or an endgame K vs. K only?

It happened in this position: 8/8/k6p/p1p2p1P/1pP2Pp1/1P1K2P1/1P6/8 w - - 0 65

[diag]8/8/k6p/p1p2p1P/1pP2Pp1/1P1K2P1/1P6/8 w - - 0 65[/diag]

... What type of tablebases is used 3/4 or other?)...

All 4 piece TBs in a ramdisk.

Volker

Interesting, I tried the FEN with the latest version 234(small
memory build from Jim) and it immediately reaches depth 31
and never shows PVs after it, but still makes the move after
the time limit is up.
Can you post the full pgn too? May be it is again hash related,
(I doubt it has to do with the egtbs now), or it happens only in
non fixed hash versions?

Code: Select all

; PositionToFen(pos): 8/8/k6p/p1p2p1P/1pP2Pp1/1P1K2P1/1P6/8 w  - - 1 0
BookMove: Signature not in book.
SetMoveTimer: 40 moves left to do this in 45:00.000 sec.
SetMoveTimer: Soft time limit 01:07.500 seconds.
SetMoveTimer: Hard time limit 04:29.500 seconds.
SetMoveTimer: Checking input and timer every 32767 nodes.
24      0    43 90810        Kd2 Kb6 Ke3 Kc6 Ke2 Kd6 Kd3 Ke6 Ke3 Kd6 Kd3 <=>
25      0    87 165071       Kd2 Kb6 Ke3 Kc6 Ke2 Kd6 Kd3 Ke6 Ke3 Kd6 Kd3 <=>
26      0    91 173857       Kd2 Kb6 Ke3 Kc6 Ke2 Kd6 Kd3 Ke6 Ke3 Kd6 Kd3 <=>
27      0    96 184092       Kd2 Kb6 Ke3 Kc6 Ke2 Kd6 Kd3 Ke6 Ke3 Kd6 Kd3 <=>
28      0   103 198108       Kd2 Kb6 Ke3 Kc6 Ke2 Kd6 Kd3 Ke6 Ke3 Kd6 Kd3 <=>
29      0   112 216419       Kd2 Kb6 Ke3 Kc6 Ke2 Kd6 Kd3 Ke6 Ke3 Kd6 Kd3 <=>
30      0   121 239434       Kd2 Kb6 Ke3 Kc6 Ke2 Kd6 Kd3 Ke6 Ke3 Kd6 Kd3 <=>
31      0   139 278448       Kd2 Kb6 Ke3 Kc6 Ke2 Kd6 Kd3 Ke6 Ke3 Kd6 Kd3 <=>
SOFT TIMER (67.6 sec) --> stop searching now
move Kd2
---------------------------------------------
Searched for  67.6 seconds, saw 9994351 nodes (7401439 qnodes) (147743 nps).
tellothers d31, +0.00,  67.6s, 147743 nps, PV=
Hashing percentages: (51.276 total, 37.607 useful)
Pawn hash hitrate: 99.484 percent.
Null move cutoff rate: 99.699 percent.
Never used the avoid null move heuristic.
Never used the quick null move heuristic.
First move beta cutoff rate was 94.226 percent.
Eval percentages: (12.00 hash, 56.20 lazy, 31.79 full)
Extensions: (954576 check, 58334 qcheck, 18210 1mv, 170 !kmvs, 0 multi,
             56841 pawn, 8 threat, 0 zug, 1188 sing, 50984 endg, 8781 recap)
Avg. cpu cycles in eval:   1595.4.

[Volker Pittlik]

...
Can you post the full pgn too?

Here it is

Code: Select all

[Event "Computer chess game"]
[Site "vlk-ubuntu"]
[Date "2006.07.27"]
[Round "3"]
[White "typhoon 1.00"]
[Black "Scorpio 1.8"]
[Result "*"]
[TimeControl "300+5"]

1. e4 e5 2. Nf3 Nf6 3. Nxe5 d6 4. Nf3 Nxe4 5. d4 d5 6. Bd3 Bd6 7. O-O O-O
8. c4 c6 9. Qc2 Na6 10. Bxe4 dxe4 11. Qxe4 Re8 12. Qd3 Bg4 13. Ng5 g6 14.
Qb3 Qc7 15. g3 h6 16. Nf3 Qb6 17. Ne5 Qxb3 18. axb3 Bxe5 19. dxe5 Nc5 20.
Be3 Rxe5 21. Bxc5 Rxc5 22. f3 Bf5 23. Re1 Bxb1 24. Rexb1 Re5 25. Re1 Rxe1+
26. Rxe1 Rd8 27. Re7 Rd1+ 28. Kf2 a5 29. Re1 Rxe1 30. Kxe1 f5 31. Kf2 Kf7
32. Ke3 c5 33. Kf4 Ke6 34. h3 b6 35. Ke3 Ke5 36. f4+ Kf6 37. Kd3 Ke7 38.
Ke2 Ke6 39. Kf2 Kf7 40. Ke1 Ke8 41. Ke2 Kf8 42. Ke3 Kf7 43. Kd2 Ke6 44. Kd1
g5 45. h4 Kf6 46. h5 g4 47. Ke2 Ke6 48. Ke3 Kf7 49. Kd2 Ke7 50. Ke3 Ke6 51.
Kd3 Kd6 52. Ke3 Kd7 53. Kd3 Kc7 54. Ke3 Kd8 55. Kf2 Kc8 56. Ke2 Kb7 57. Ke3
Kc6 58. Kd3 Kd7 59. Ke3 Kc7 60. Kd3 Kb8 61. Kc2 Kb7 62. Kc3 Ka6 63. Kd2 b5
64. Kd3 b4
*

...May be it is again hash related,
(I doubt it has to do with the egtbs now), or it happens only in
non fixed hash versions?

I made the same changes to the source as Jim did.


Volker

[Guenther Simon]

...
Can you post the full pgn too?

Here it is

Code: Select all

[Event "Computer chess game"]
[Site "vlk-ubuntu"]
[Date "2006.07.27"]
[Round "3"]
[White "typhoon 1.00"]
[Black "Scorpio 1.8"]
[Result "*"]
[TimeControl "300+5"]

1. e4 e5 2. Nf3 Nf6 3. Nxe5 d6 4. Nf3 Nxe4 5. d4 d5 6. Bd3 Bd6 7. O-O O-O
8. c4 c6 9. Qc2 Na6 10. Bxe4 dxe4 11. Qxe4 Re8 12. Qd3 Bg4 13. Ng5 g6 14.
Qb3 Qc7 15. g3 h6 16. Nf3 Qb6 17. Ne5 Qxb3 18. axb3 Bxe5 19. dxe5 Nc5 20.
Be3 Rxe5 21. Bxc5 Rxc5 22. f3 Bf5 23. Re1 Bxb1 24. Rexb1 Re5 25. Re1 Rxe1+
26. Rxe1 Rd8 27. Re7 Rd1+ 28. Kf2 a5 29. Re1 Rxe1 30. Kxe1 f5 31. Kf2 Kf7
32. Ke3 c5 33. Kf4 Ke6 34. h3 b6 35. Ke3 Ke5 36. f4+ Kf6 37. Kd3 Ke7 38.
Ke2 Ke6 39. Kf2 Kf7 40. Ke1 Ke8 41. Ke2 Kf8 42. Ke3 Kf7 43. Kd2 Ke6 44. Kd1
g5 45. h4 Kf6 46. h5 g4 47. Ke2 Ke6 48. Ke3 Kf7 49. Kd2 Ke7 50. Ke3 Ke6 51.
Kd3 Kd6 52. Ke3 Kd7 53. Kd3 Kc7 54. Ke3 Kd8 55. Kf2 Kc8 56. Ke2 Kb7 57. Ke3
Kc6 58. Kd3 Kd7 59. Ke3 Kc7 60. Kd3 Kb8 61. Kc2 Kb7 62. Kc3 Ka6 63. Kd2 b5
64. Kd3 b4
*

...May be it is again hash related,
(I doubt it has to do with the egtbs now), or it happens only in
non fixed hash versions?

I made the same changes to the source as Jim did.


Volker

Thanks Volker,

Feeding the full PGN now I see that Typhoon exceeds depth
31 on this slow machine. It needs over one minute though
to produce ply 32 and from there slowly crawls on. I stopped
it for now at depth 36 and will repeat it on my much faster machine
until I reach your depth 50.

Best regards,
Guenther

[Volker Pittlik]

...

Feeding the full PGN now I see that Typhoon exceeds depth
31 on this slow machine. It needs over one minute though
to produce ply 32 and from there slowly crawls on. I stopped
it for now at depth 36 and will repeat it on my much faster machine
until I reach your depth 50.

Best regards,
Guenther

Now it is getting mystical. I use an old XP2000+ here. Setup the position depth 31 is reached in less than a second but depth 38 will not exceeded after ~5 minutes. The time control for the game was 5+5!

Volker

[Jim Ablett]

Here's the actual make+options I use.

Code: Select all

make SEE_HEAPS=1 TEST_NULL=1 EVAL_TIME=1 PERF_COUNTERS=1


Jim.

[Scott Gasch]

I reproduced this latest crash here; it's my first priority as soon as I have some free time to work on the engine. This message thread seems to have become the "typhoon bug report" line. Thanks for finding these.

Also, whomever built your binary defined TEST_NULL which is useless (and a waste of speed) in the current engine.

I don't get any sort of error on process exit but last night I cleaned up the engine shutdown code and checked in revision 235 this morning. I'd appreciate it if someone who is experiencing these stack traces would let me know if that fixes the problem.

[Jim Ablett]

New revision 236 build on my homepage

http://homepages.tesco.net/henry.ablett/jims.html

No more stackdump error! Great work Scott

Also, whomever built your binary defined TEST_NULL which is useless (and a waste of speed) in the current engine.

Sorry, I'm to blame for that. I'm one of those sad people who is
just fascinated and enthrawled by statistics - I couldn't resist it !

I think I need to get out more

Jim.

[Jim Ablett]

Forgot to mention:

My Gcc-Cygwin builds are tested working ok under Windows 98
as well (not the smp version - Win98 doesn't have dual processor support)

Jim.

[Guenther Simon]

Hi Scott(and Jim),

Since all versions run properly now since revision 236
(at least with ponder off - it seems there are still problems
with pondering according to Leos test tourneys though?),
I wish a cosmetical 'flaw' in Typhoons output would be fixed too.

Typhoon never displays any score, if it has reached a tablebase position.
If this happens vs. an opponent which cannot resign
on its own, GUI ajudication will fail.
Please let Typhoon show some mate scores in tbs positions.

Best regards,
Guenther

[Jim Ablett]

Hi Guenther,

Timing problems addressed in latest update (revision 246).
Windows build on my homepage later.

No fix for pv output yet, but I'm sure fix will come soon.

Regards,
Jim.