Crafty 25.0 Release

[bob]

John,

I can't reproduce this. My version gets this with 5 piece egtbs enabled:

62 2:26 Mat33 1. Kb1 Ka8 2. Kb2 Ka7 3. Kb3 Ka6 4. Kc2
Kb7 5. Kc3 Kb6 6. Kd2 Kc7 7. Kd3 Kb7
8. Ke3 Kc7 9. Kf2 Kd7 10. Kg3 Ke7 11. Kh4
Kf6 12. Kh5 Kf7 13. Kg5 Kg7 14. Kxf5 Kf7
15. Ke4 Kf6 16. Kd3 Kf5 17. Kc4 Kxf4
18. Kb5 Ke4 19. Kc6 Kf5 20. Kxd6 Kf6
21. Kc6 Kg7 22. d6 Kf8 23. d7 Ke7 24. Kc7
Kf7 25. d8=Q Kg6 26. Qd5 Kf6 27. Qxa5
<EGTB>

and this without any:

53-> 52.83/16:39 23.49 1. Kb1 Ka8 2. Kb2 Ka7 3. Kb3 Kb7 4. Kc3
Ka6 5. Kd3 Kb7 6. Ke3 Kc7 7. Kf2 Kd7
8. Kg3 Ke7 9. Kh4 Kf7 10. Kg5 Kg7 11. Kxf5
Kf7 12. Kg5 Kg7 13. f5 Kf7 14. f6 Kf8
15. Kg4 Ke8 16. Kf4 Kf8 17. Kg5 Kg8
18. Kg6 Kh8 19. Kf7 Kh7 20. Ke6 Kg6 21. f7
Kg5 22. f8=Q Kg4 23. Kxd6 Kg3 24. Ke6 Kg2
25. d6 Kh2 26. d7 Kg3 27. d8=Q

both one thread on my old macbook.

Here's the 20 core box:

56-> 15.91/16:39 Mat38 1. Kb1 Kb7 2. Kc1 Kc7 3. Kd1 Kd7 4. Kc2
Kc7 5. Kd3 Kb6 6. Kd2 Kb7 7. Ke1 Kc8
8. Kd2 Kd8 9. Kc3 Kc7 10. Kd3 Kb7 11. Ke2
Kc8 12. Kf2 Kd8 13. Kg3 Ke7 14. Kh4 Kf6
15. Kh5 Kf7 16. Kg5 Kg7 17. Kxf5 Kf7
18. Kg5 Kg7 19. f5 Kf7 20. f6 Kf8 21. Kg4
Kg8 22. Kf4 Kf8 23. Kg5 Kg8 24. Kg6 Kh8
25. Kf7 Kh7 26. Ke6 Kg6 27. f7 Kg7 28. Ke7
Kg6 29. f8=Q Kg5 30. Ke6 Kg4 31. Kxd6 Kg3
32. Ke6 Kh2 33. d6 Kg2 34. d7 Kg3 35. d8=Q
Kg4 36. Qf5+ Kg3 37. Qg8+ Kh2 38. Qfh7#

and with EGTB's:


62-> 15.36/16:39 Mat34 1. Kb1 Kb7 2. Kc1 Kc7 3. Kd1 Kd7 4. Kc2
Kc8 5. Kd2 Kd7 6. Kc3 Kc7 7. Kd3 Kb6
8. Ke3 Kc7 9. Kf2 Kd8 10. Kg3 Ke7 11. Kh4
Kf6 12. Kh5 Kf7 13. Kg5 Kg7 14. Kxf5 Kf7
15. Ke4 Kf6 16. Kd3 Kf5 17. Kc4 Kg6
18. Kb5 Kf5 19. Kxa5 Kg6 20. f5+ Kg5
21. Kb5 Kf6 22. a5 Kg5 23. a6 Kf6 24. a7
Kg5 25. a8=Q Kxf5 26. Qf8+ Kg5 27. Qxd6
<EGTB>


Is this windows or linux? All of my output above is from Linux machines... (or OS X, still unix).

I notice the analyze output needs fixing as those () characters should not be there. I've not used analyze mode during all the testing, so I will look at that...

[JVMerlino]

Is this windows or linux? All of my output above is from Linux machines... (or OS X, still unix).

This is 64-bit Windows 7, using Daniel's compiles (I tried both popcnt and non-popcnt). I also made sure that Crafty 24.1 did not exhibit the same behavior, just in case my eyes were playing tricks on me.

jm

[rcmaddox]

Congrats, Bob, for your work!!!
It is stimulating to see a man of science that keeps working and improving his works beyond 60 years of age.
I am 66 and feel as you I still can give a lot.
Looks like this new Crafty will kick lot of asses...

Fern

Yes. What Fern said. (I'm 61)

[cdani]

Is this windows or linux? All of my output above is from Linux machines... (or OS X, still unix).

This is 64-bit Windows 7, using Daniel's compiles (I tried both popcnt and non-popcnt). I also made sure that Crafty 24.1 did not exhibit the same behavior, just in case my eyes were playing tricks on me.

jm

Can you try with the first executables posted here from Jose Ma. Velasco? The version is previous to the last patch from Bob, but maybe it will give us an idea.

[JVMerlino]

Is this windows or linux? All of my output above is from Linux machines... (or OS X, still unix).

This is 64-bit Windows 7, using Daniel's compiles (I tried both popcnt and non-popcnt). I also made sure that Crafty 24.1 did not exhibit the same behavior, just in case my eyes were playing tricks on me.

jm

Can you try with the first executables posted here from Jose Ma. Velasco? The version is previous to the last patch from Bob, but maybe it will give us an idea.

Even more weirdness. All of the below is with Jose's compiles.

The popcnt compile does work properly, getting a +5 score with Kb1 instantly and increasing from there.

The non-popcnt compile does not work, but in a very different way from your compile. Without posting a long logfile, what happens is that the chosen move is Kb2 (which is wrong), with a score of only about +0.6 for more than 10 seconds, and then a huge fail high at depth 52 resulting in a score of more than +11, but STILL CHOOSING KB2!? This score and KB2 move combination remains relatively steady up through depth 62, when I stopped it.

I then tried the UCI version in console mode, typing the following commands:
uci
ucinewgame
position fen 8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - - 0 1
go infinite

...and the only output I got was:
info depth 1 score cp 1

weird.....

jm

[Werner]

I get here on a Windows 10 pc with crafty.rc, 5men tbs:
FEN: 8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - - 0 1


54 00:25 197.413.084 7.818.339 +28,01 1. Kb1 Kb7 2. Kc1 Kb8 3. Kc2 Kc8 4. Kd2 Kc7 5. Kd3 Kb6 6. Ke2 Kc7 7. Kf3 Kd7 8. Kg3 Ke7 9. Kh4 Kf7 10. Kg5 Kg7 11. Kxf5 Kf7 12. Ke4 Kf6 13. Kd3 Ke7 14. Kc4 Kd8 15. Kb5 Kd7 16. Kxa5 Ke7 17. Kb6 Kf6 18. a5 Kf5 19. a6 Kg6 20. a7 Kf5 21. a8=Q Ke4 22. Qe8+ Kxf4 23. Qf8+ Kg4 24. Qxd6 Kf3 25. Qf6+ Ke2 26. d6 Kd3 27. d7 Ke3 28. d8=Q Kd2 29. Qc8
55 00:26 211.887.727 7.847.693 +28,17 1. Kb1!
55 00:27 216.626.398 7.877.323 +28,33 1. Kb1!
55 00:30 238.759.882 7.892.888 +38,25 1. Kb1!
55 02:37 1.223.661.994 7.756.969 +M46 1. Kb1 Ka8 2. Kb2 Ka7 3. Kb3 Kb6 4. Kc2 Kb7 5. Kc1 Kc7 6. Kd1 Kd8 7. Kc2 Kc8 8. Kd2 Kd7 9. Kc3 Kc7 10. Kd3 Kb6 11. Ke2 Kc7 12. Kf3 Kd7 13. Kg3 Ke7 14. Kh4 Kf6 15. Kh5 Kf7 16. Kg5 Kg7 17. Kxf5 Kf7 18. Ke4 Kf6 19. Kd3 Ke7 20. Kc4 Kd8 21. Kb5 Kd7 22. f5 Kc7 23. Ka6 Kd8 24. Kb6 Kd7 25. Kb7 Kd8 26. Kc6 Ke7 27. Kc7 Kf7 28. Kxd6 Kf6 29. Kc7 Kxf5 30. d6 Ke4 31. d7 Kd3 32. Kb6 Ke3 33. Kxa5 Ke4 34. d8=B
55 02:37 1.223.662.125 7.756.970 +M46 1. Kb1 Ka8 2. Kb2 Ka7 3. Kb3 Kb6 4. Kc2 Kb7 5. Kc1 Kc7 6. Kd1 Kd8 7. Kc2 Kc8 8. Kd2 Kd7 9. Kc3 Kc7 10. Kd3 Kb6 11. Ke2 Kc7 12. Kf3 Kd7 13. Kg3 Ke7 14. Kh4 Kf6 15. Kh5 Kf7 16. Kg5 Kg7 17. Kxf5 Kf7 18. Ke4 Kf6 19. Kd3 Ke7 20. Kc4 Kd8 21. Kb5 Kd7 22. f5 Kc7 23. Ka6 Kd8 24. Kb6 Kd7 25. Kb7 Kd8 26. Kc6 Ke7 27. Kc7 Kf7 28. Kxd6 Kf6 29. Kc7 Kxf5 30. d6 Ke4 31. d7 Kd3 32. Kb6 Ke3 33. Kxa5 Ke4 34. d8=B
56 02:48 1.323.436.430 7.842.121 +M46 1. Kb1 Ka8 2. Kb2 Ka7 3. Kb3 Kb6 4. Kc2 Kb7 5. Kc1 Kc7 6. Kd1 Kd8 7. Kc2 Kc8 8. Kd2 Kd7 9. Kc3 Kc7 10. Kd3 Kb6 11. Ke2 Kc7 12. Kf3 Kd7 13. Kg3 Ke7 14. Kh4 Kf6 15. Kh5 Kf7 16. Kg5 Kg7 17. Kxf5 Kf7 18. Ke4 Kf6 19. Kd3 Ke7 20. Kc4 Kd8 21. Kb5 Kd7 22. f5 Kc7 23. Ka6 Kd8 24. Kb6 Kd7 25. Kb7 Kd8 26. Kc6 Ke7 27. Kc7 Kf7 28. Kxd6 Kf6 29. Kc7 Kxf5 30. d6 Ke4 31. d7 Kd3 32. Kb6 Ke3 33. Kxa5 Ke4 34. d8=B
56 02:48 1.323.436.532 7.842.122 +M46 1. Kb1 Ka8 2. Kb2 Ka7 3. Kb3 Kb6 4. Kc2 Kb7 5. Kc1 Kc7 6. Kd1 Kd8 7. Kc2 Kc8 8. Kd2 Kd7 9. Kc3 Kc7 10. Kd3 Kb6 11. Ke2 Kc7 12. Kf3 Kd7 13. Kg3 Ke7 14. Kh4 Kf6 15. Kh5 Kf7 16. Kg5 Kg7 17. Kxf5 Kf7 18. Ke4 Kf6 19. Kd3 Ke7 20. Kc4 Kd8 21. Kb5 Kd7 22. f5 Kc7 23. Ka6 Kd8 24. Kb6 Kd7 25. Kb7 Kd8 26. Kc6 Ke7 27. Kc7 Kf7 28. Kxd6 Kf6 29. Kc7 Kxf5 30. d6 Ke4 31. d7 Kd3 32. Kb6 Ke3 33. Kxa5 Ke4 34. d8=B

and without crafty.rc and no path to my tbs:

FEN: 8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - - 0 1

Crafty-25.0:

65 00:35 173.452.017 4.818.111 0,00 0. ... () 1. Kb1 Kb7 2. Kc2 Kc8 3. Kb3 Kc7 4. Kc4 Kb6 5. Kc3 Kb7 6. Kd3 Kc7 7. Ke2 Kd7 8. Kf3 Ke7 9. Kf2 Ke8 10. Ke1 Kd8 11. Kd2 Kc8 12. Kc2 Kb7 13. Kc3 Kc7 14. Kc4 <3-fold>

[bob]

Is this windows or linux? All of my output above is from Linux machines... (or OS X, still unix).

This is 64-bit Windows 7, using Daniel's compiles (I tried both popcnt and non-popcnt). I also made sure that Crafty 24.1 did not exhibit the same behavior, just in case my eyes were playing tricks on me.

jm

Is it happening consistently? If so that points to the windows version as I ran SMP/non-SMP and EGTB/non-EGTB multiple times just now and while there is a bit of variability in the SMP stuff, no draw scores are showing up for me, ever...

[JVMerlino]

Is this windows or linux? All of my output above is from Linux machines... (or OS X, still unix).

This is 64-bit Windows 7, using Daniel's compiles (I tried both popcnt and non-popcnt). I also made sure that Crafty 24.1 did not exhibit the same behavior, just in case my eyes were playing tricks on me.

jm

Is it happening consistently? If so that points to the windows version as I ran SMP/non-SMP and EGTB/non-EGTB multiple times just now and while there is a bit of variability in the SMP stuff, no draw scores are showing up for me, ever...

It is happening consistently, regardless of whether I use the default of 1 thread or use "cores 4" (which is also a little weird, because it says that the option is disabled but it DOES set the maximum threads). Of course, the output is different because of time-to-depth, but the fundamental results are the same.

jm

[JVMerlino]

I get here on a Windows 10 pc with crafty.rc, 5men tbs:
FEN: 8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - - 0 1


54 00:25 197.413.084 7.818.339 +28,01 1. Kb1 Kb7 2. Kc1 Kb8 3. Kc2 Kc8 4. Kd2 Kc7 5. Kd3 Kb6 6. Ke2 Kc7 7. Kf3 Kd7 8. Kg3 Ke7 9. Kh4 Kf7 10. Kg5 Kg7 11. Kxf5 Kf7 12. Ke4 Kf6 13. Kd3 Ke7 14. Kc4 Kd8 15. Kb5 Kd7 16. Kxa5 Ke7 17. Kb6 Kf6 18. a5 Kf5 19. a6 Kg6 20. a7 Kf5 21. a8=Q Ke4 22. Qe8+ Kxf4 23. Qf8+ Kg4 24. Qxd6 Kf3 25. Qf6+ Ke2 26. d6 Kd3 27. d7 Ke3 28. d8=Q Kd2 29. Qc8
55 00:26 211.887.727 7.847.693 +28,17 1. Kb1!
55 00:27 216.626.398 7.877.323 +28,33 1. Kb1!
55 00:30 238.759.882 7.892.888 +38,25 1. Kb1!
55 02:37 1.223.661.994 7.756.969 +M46 1. Kb1 Ka8 2. Kb2 Ka7 3. Kb3 Kb6 4. Kc2 Kb7 5. Kc1 Kc7 6. Kd1 Kd8 7. Kc2 Kc8 8. Kd2 Kd7 9. Kc3 Kc7 10. Kd3 Kb6 11. Ke2 Kc7 12. Kf3 Kd7 13. Kg3 Ke7 14. Kh4 Kf6 15. Kh5 Kf7 16. Kg5 Kg7 17. Kxf5 Kf7 18. Ke4 Kf6 19. Kd3 Ke7 20. Kc4 Kd8 21. Kb5 Kd7 22. f5 Kc7 23. Ka6 Kd8 24. Kb6 Kd7 25. Kb7 Kd8 26. Kc6 Ke7 27. Kc7 Kf7 28. Kxd6 Kf6 29. Kc7 Kxf5 30. d6 Ke4 31. d7 Kd3 32. Kb6 Ke3 33. Kxa5 Ke4 34. d8=B
55 02:37 1.223.662.125 7.756.970 +M46 1. Kb1 Ka8 2. Kb2 Ka7 3. Kb3 Kb6 4. Kc2 Kb7 5. Kc1 Kc7 6. Kd1 Kd8 7. Kc2 Kc8 8. Kd2 Kd7 9. Kc3 Kc7 10. Kd3 Kb6 11. Ke2 Kc7 12. Kf3 Kd7 13. Kg3 Ke7 14. Kh4 Kf6 15. Kh5 Kf7 16. Kg5 Kg7 17. Kxf5 Kf7 18. Ke4 Kf6 19. Kd3 Ke7 20. Kc4 Kd8 21. Kb5 Kd7 22. f5 Kc7 23. Ka6 Kd8 24. Kb6 Kd7 25. Kb7 Kd8 26. Kc6 Ke7 27. Kc7 Kf7 28. Kxd6 Kf6 29. Kc7 Kxf5 30. d6 Ke4 31. d7 Kd3 32. Kb6 Ke3 33. Kxa5 Ke4 34. d8=B
56 02:48 1.323.436.430 7.842.121 +M46 1. Kb1 Ka8 2. Kb2 Ka7 3. Kb3 Kb6 4. Kc2 Kb7 5. Kc1 Kc7 6. Kd1 Kd8 7. Kc2 Kc8 8. Kd2 Kd7 9. Kc3 Kc7 10. Kd3 Kb6 11. Ke2 Kc7 12. Kf3 Kd7 13. Kg3 Ke7 14. Kh4 Kf6 15. Kh5 Kf7 16. Kg5 Kg7 17. Kxf5 Kf7 18. Ke4 Kf6 19. Kd3 Ke7 20. Kc4 Kd8 21. Kb5 Kd7 22. f5 Kc7 23. Ka6 Kd8 24. Kb6 Kd7 25. Kb7 Kd8 26. Kc6 Ke7 27. Kc7 Kf7 28. Kxd6 Kf6 29. Kc7 Kxf5 30. d6 Ke4 31. d7 Kd3 32. Kb6 Ke3 33. Kxa5 Ke4 34. d8=B
56 02:48 1.323.436.532 7.842.122 +M46 1. Kb1 Ka8 2. Kb2 Ka7 3. Kb3 Kb6 4. Kc2 Kb7 5. Kc1 Kc7 6. Kd1 Kd8 7. Kc2 Kc8 8. Kd2 Kd7 9. Kc3 Kc7 10. Kd3 Kb6 11. Ke2 Kc7 12. Kf3 Kd7 13. Kg3 Ke7 14. Kh4 Kf6 15. Kh5 Kf7 16. Kg5 Kg7 17. Kxf5 Kf7 18. Ke4 Kf6 19. Kd3 Ke7 20. Kc4 Kd8 21. Kb5 Kd7 22. f5 Kc7 23. Ka6 Kd8 24. Kb6 Kd7 25. Kb7 Kd8 26. Kc6 Ke7 27. Kc7 Kf7 28. Kxd6 Kf6 29. Kc7 Kxf5 30. d6 Ke4 31. d7 Kd3 32. Kb6 Ke3 33. Kxa5 Ke4 34. d8=B

and without crafty.rc and no path to my tbs:

FEN: 8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - - 0 1

Crafty-25.0:

65 00:35 173.452.017 4.818.111 0,00 0. ... () 1. Kb1 Kb7 2. Kc2 Kc8 3. Kb3 Kc7 4. Kc4 Kb6 5. Kc3 Kb7 6. Kd3 Kc7 7. Ke2 Kd7 8. Kf3 Ke7 9. Kf2 Ke8 10. Ke1 Kd8 11. Kd2 Kc8 12. Kc2 Kb7 13. Kc3 Kc7 14. Kc4 <3-fold>

Well, that made me think that a crafty.rc file was required. So I tried it with the one that comes with Jose's package, and I get the same results.

But your output points to another possibility. Could you try it with a crafty.rc file, but without TBs? I do not have any TBs that Crafty supports.

jm

[bob]

Is this windows or linux? All of my output above is from Linux machines... (or OS X, still unix).

This is 64-bit Windows 7, using Daniel's compiles (I tried both popcnt and non-popcnt). I also made sure that Crafty 24.1 did not exhibit the same behavior, just in case my eyes were playing tricks on me.

jm

Is it happening consistently? If so that points to the windows version as I ran SMP/non-SMP and EGTB/non-EGTB multiple times just now and while there is a bit of variability in the SMP stuff, no draw scores are showing up for me, ever...

It is happening consistently, regardless of whether I use the default of 1 thread or use "cores 4" (which is also a little weird, because it says that the option is disabled but it DOES set the maximum threads). Of course, the output is different because of time-to-depth, but the fundamental results are the same.

jm

When you see "something" is disabled from crafty, it means that either you had something like something=xx on the command line, or something=xx in the crafty.rc/.craftyrc file. If you set something explicitly via crafty.rc or command line, it assumes you mean it and it won't let xboard/winboard override something you explicitly specified...

But it is really looking like the windows version is simply broken. I don't know if microsoft has added an update to their MSVC compiler or what. Would be interesting for someone to compile 24.1 on the same compiler that the broken 25.0 executable comes from, that would reveal whether it is something compiler-related or something 25.0 related...

Those odd draw scores point toward something broken with hashing. What I have no idea, since that code is the same for windows or unix, except for allocation. I'll take a quick look at the windows allocation code to see if there is something different there...