Stockfish 3 PA_GTB

[Jeremy Bernstein]

both versions you made cause an exception in fritz when accessing tbs. the sse4.2 completed one game but did not need the tablebases. when I get stockfish from ebrok.eu/stockfish it is the middle compile that works on my computer using fritz.

the sse4.2 version form ebrok.eu/stockfish also causes an exception in fritz on my computer..

I don't think it's the SSE support, but rather the new Gaviota TB support in recent versions of Fritz and Chessbase, added to support Houdini. I'll need to do a little probing to see what's going on, and will have it fixed for version 003.

Thanks,
Jeremy

Correction. It's a MultiPV bug as relates to the tablebase code. Fixing for 003.

jb

[Jeremy Bernstein]

Version 003 is now live. Includes:

- fix for tablebase exceptions with MultiPV > 1
- runtime resolution of POPCNT availability -- only one binary required for 64-bit systems now
- update to tip of Stockfish upstream master

Enjoy, please let me know how you get along...
Jeremy

[endings]

works good with no exceptions... the only thing I did notice is that stockfish only accesses one tablebase at a time and then moves on it, where houdini scales through many bases in one scan then moves. Also stockfish doesn't seem to use all the tablebases only the first few. I only seen up to tablebase 9.

thanks for all your hard work

[endings]

ok stockfish does access more than one tablebases at a time but the most i seen was two tablebases and it went from 1 to 2 or 5 to 6 but not throught all the bases like houdini does. I am upload a you-tube video right now should be ready in 15 mis my time is 19:45 now http://youtu.be/cb7qsjABRJM

[endings]

http://youtu.be/cb7qsjABRJM here is a video I made so you can see. Houdini accesses tablesbases sooner than stockfish and I thought because stockfish has higher search depth that it would access the bases sooner than houdini.

anyway stocfish does access more than one base at a time but I have seen only two so far. Hope the video helps. I try to make a better one

[Jeremy Bernstein]

I am looking for some knowledgable testers with some experience using "learning" or "persistent hash" engine analysis features. If this is you, and you have a little time to spend playing with something that's potentially buggy, please drop me a PM.

Thanks
Jeremy

[Jeremy Bernstein]

http://youtu.be/cb7qsjABRJM here is a video I made so you can see. Houdini accesses tablesbases sooner than stockfish and I thought because stockfish has higher search depth that it would access the bases sooner than houdini.

anyway stocfish does access more than one base at a time but I have seen only two so far. Hope the video helps. I try to make a better one

Thanks for the video. A couple of things:

- I don't know when Houdini probes the tablebases, but it looks like it may happen first. Stockfish probes them AFTER probing the hash table. This is because TB hits are cached in the hash -- so if there's a cached TB hit, we'd rather know about it from the hash table and avoid the costs of probing the tablebases. Or maybe Houdini sets a flag indicating that the hashed value originated from the TBs and uses that information to update tbhits. Hard to say.
- I'm not really sure what you mean by "accessing more than one base at a time". Could you be more precise, maybe provide a test position where I can see what you mean?

Thanks
Jeremy

[endings]

4r3/2k1r3/p1N5/1ppP4/8/2P5/PP6/K2R3R w - - 0 45 try this position on both engines. stockfish was white and was winning like no tomorrow but became a draw. I edited game data and made it a win. anyway when houdini accesses the tablebases you might see tb24, tb128, tb268, tb954, tb1026 and so on but stockfish you will only see tb1, tb2 or tb1, tb6 or tb2, tb4 but does not scale a wide range of tablebases like houdini.


dxc5
{18.60/22 0} 45. Nxe7 {90.76/26 2} Rd8 {18.77/23 15 (Kb7)} 46. d6+ {95.92/25
10 (Ng6)} Rxd6 {12.83/19 1 (Kd7)} 47. Rxd6 {102.09/24 3 (Nd5+)} Kxd6 {11.17/17
1} 48. Rh6+ {102.09/23 6 (Ng6)} Kxe7 {23.60/24 4 (Kc7)} 49. Rxa6 {#24/22 4} Kd7
{#15/28 3} 50. b4 {#14/22 23 (Rb6)} cxb4 {#23/27 3 (c4)} 51. cxb4 {#13/25 6}
Ke7 {#12/31 2 (Kc7)} 52. a4 {#12/34 5 (Kb2)} bxa4 {#11/32 0} 53. Kb2 {#11/1 0}
Kd7 {#10/99 0 (a3+)} 54. Ka3 {#10/1 0} Kc7 {#9/99 0} 55. Rg6 {#9/1 0} Kb7 {#8/
99 0 (Kd8)} 56. Rh6 {#8/1 0 (Kxa4)} Kc7 {#8/99 0 (Ka7)} 57. Rg6 {#8/1 0} Kb7 {
#8/99 0} 58. Rh6 {#8/1 0 (Kxa4)} Kc7 {#8/99 0} 59. Rg6 {#8/1 0 Draw accepted}

[Jeremy Bernstein]

4r3/2k1r3/p1N5/1ppP4/8/2P5/PP6/K2R3R w - - 0 45 try this position on both engines. stockfish was white and was winning like no tomorrow but became a draw. I edited game data and made it a win. anyway when houdini accesses the tablebases you might see tb24, tb128, tb268, tb954, tb1026 and so on but stockfish you will only see tb1, tb2 or tb1, tb6 or tb2, tb4 but does not scale a wide range of tablebases like houdini.


dxc5
{18.60/22 0} 45. Nxe7 {90.76/26 2} Rd8 {18.77/23 15 (Kb7)} 46. d6+ {95.92/25
10 (Ng6)} Rxd6 {12.83/19 1 (Kd7)} 47. Rxd6 {102.09/24 3 (Nd5+)} Kxd6 {11.17/17
1} 48. Rh6+ {102.09/23 6 (Ng6)} Kxe7 {23.60/24 4 (Kc7)} 49. Rxa6 {#24/22 4} Kd7
{#15/28 3} 50. b4 {#14/22 23 (Rb6)} cxb4 {#23/27 3 (c4)} 51. cxb4 {#13/25 6}
Ke7 {#12/31 2 (Kc7)} 52. a4 {#12/34 5 (Kb2)} bxa4 {#11/32 0} 53. Kb2 {#11/1 0}
Kd7 {#10/99 0 (a3+)} 54. Ka3 {#10/1 0} Kc7 {#9/99 0} 55. Rg6 {#9/1 0} Kb7 {#8/
99 0 (Kd8)} 56. Rh6 {#8/1 0 (Kxa4)} Kc7 {#8/99 0 (Ka7)} 57. Rg6 {#8/1 0} Kb7 {
#8/99 0} 58. Rh6 {#8/1 0 (Kxa4)} Kc7 {#8/99 0} 59. Rg6 {#8/1 0 Draw accepted}

Thanks, there was a bug in the way tablebase hits were being counted in multithreaded operation. Fixed for 004.

jb

[CarstenL]

4r3/2k1r3/p1N5/1ppP4/8/2P5/PP6/K2R3R w - - 0 45 try this position on both engines. stockfish was white and was winning like no tomorrow but became a draw. ....

I can confirm that. (3-fold repetition)