Winboard Forum

A couple of months ago Kirill Kryukov of CCRL found an EGTB bug in certain 64-bit engines (Thor's Hammer, Pepito). Could you track down the problem? I assume that your latest 64-bit compiles of other engines are free of the problem, can you confirm that? Thanks you.

Cheers,

Gábor Szots wrote:A couple of months ago Kirill Kryukov of CCRL found an EGTB bug in certain 64-bit engines (Thor's Hammer, Pepito). Could you track down the problem? I assume that your latest 64-bit compiles of other engines are free of the problem, can you confirm that? Thanks you.

Hi Gabor,

I did have a look a few months ago. I suspect these engines are using old version Nalimov EGTB code which isn't fully 64-bit compatible.
Easy way to fix it - disable the EGTB's.
Seriously, I will try to fix this once I get a few other projects out of the way.

regards,
Jim.

Pepito 1.59.2 JA by Carlos del Cacho

Fixed EGTBs in 64 bit.

I tested Pepito using some of Kirill's longest mate positions and
seems to working ok now. Pepito's egtbs only accessed in winboard
mode. In uci mode it doesn't use them, which it correct I suppose as
the chess gui handles it instead.

Download:
http://www.mediafire.com/?km52tmzzzny

Jim.

If there were source code changes, I would like to get a copy

Jim Ablett wrote:Pepito 1.59.2 JA by Carlos del Cacho

Fixed EGTBs in 64 bit.

I tested Pepito using some of Kirill's longest mate positions and
seems to working ok now. Pepito's egtbs only accessed in winboard
mode. In uci mode it doesn't use them, which it correct I suppose as
the chess gui handles it instead.

Download:
http://www.mediafire.com/?km52tmzzzny

Jim.

Thank you very much, Jim!

Would it be possible for you to enable Pepito's access to EGTB in UCI mode too? Rationale:
1. This is how the original author's build of Pepito 1.59 works.
2. This is how your previous build of Pepito 1.59 works.
3. In my tourney (KCEC) I run engines via UCI, and I don't let GUI handle the endgame. I would hate it if I have to run Pepito through WB2UCI adapter just to make it access the EGTB.

Thanks,
Kirill

Dann Corbit wrote:If there were source code changes, I would like to get a copy

Hi Dann,

No code change. Just compiler change Intel > Msvc.
I did remove some WIN32 specific code from the TB probe src
just to be sure that it wasn't being used, but it shouldn't have
been anyway.

I forgot to include src so I've re-upped it with src included>
http://www.mediafire.com/?zmy1w5jyynn

Jim.

Kirill Kryukov wrote:

Jim Ablett wrote:Pepito 1.59.2 JA by Carlos del Cacho

Fixed EGTBs in 64 bit.

I tested Pepito using some of Kirill's longest mate positions and
seems to working ok now. Pepito's egtbs only accessed in winboard
mode. In uci mode it doesn't use them, which it correct I suppose as
the chess gui handles it instead.

Download:
http://www.mediafire.com/?km52tmzzzny

Jim.

Thank you very much, Jim!

Would it be possible for you to enable Pepito's access to EGTB in UCI mode too? Rationale:
1. This is how the original author's build of Pepito 1.59 works.
2. This is how your previous build of Pepito 1.59 works.
3. In my tourney (KCEC) I run engines via UCI, and I don't let GUI handle the endgame. I would hate it if I have to run Pepito through WB2UCI adapter just to make it access the EGTB.

Thanks,
Kirill

Hi Kirill,

I didn't know Pepito used tablebases in UCI mode by design.
Maybe it does work in a game, but not in analysis (which I tested in)
I'll take a look to see if I can fix it if broken.

Jim.

Jim Ablett wrote:Hi Kirill,

I didn't know Pepito used tablebases in UCI mode by design.
Maybe it does work in a game, but not in analysis (which I tested in)
I'll take a look to see if I can fix it if broken.

Jim.

Hi Jim,

Great, I'm looking forward to it!

Best,
Kirill

Hi Jim, here is another engine with the same bug: Sunsetter C10. Your 64-bit compile of it is losing drawn endings as soon as a 5-men table is used.

So for Sunsetter and other affected engines I have to choose between faster search and EGTB usage. Probably EGTB is less important than speed increase, still I'd much rather use the full-featured and a full-speed version.

The list of Jim's 64-bit compiles affected by a 64-bit EGTB bug: Thor's Hammer 2.28, Pepito 1.59 (in UCI mode), Sunsetter C10. I hope it is not too hard to locate a 64-bit-compatible version of egtb code and recompile. Really looking forward to have fixed 64-bit compiles of these engines.

I was just starting to test Sunsetter C10 for my tourney, now I am really wondering what should I do - switch to a 32-bit version, or switch off EGTB and stay with 64-bit, or use some other compile, or may be switch to another engine for now and come back to Sunsetter later.

Much appreciate your work,
With best regards,
Kirill

Kirill Kryukov wrote:Hi Jim, here is another engine with the same bug: Sunsetter C10. Your 64-bit compile of it is losing drawn endings as soon as a 5-men table is used.

So for Sunsetter and other affected engines I have to choose between faster search and EGTB usage. Probably EGTB is less important than speed increase, still I'd much rather use the full-featured and a full-speed version.

The list of Jim's 64-bit compiles affected by a 64-bit EGTB bug: Thor's Hammer 2.28, Pepito 1.59 (in UCI mode), Sunsetter C10. I hope it is not too hard to locate a 64-bit-compatible version of egtb code and recompile. Really looking forward to have fixed 64-bit compiles of these engines.

I was just starting to test Sunsetter C10 for my tourney, now I am really wondering what should I do - switch to a 32-bit version, or switch off EGTB and stay with 64-bit, or use some other compile, or may be switch to another engine for now and come back to Sunsetter later.

Much appreciate your work,
With best regards,
Kirill

Hi Kirill,
My advice to friend....use Sunsetter C9d,a 32 bit version but it works as charm
It's playing in one of my tournaments now,so....if you're interested I'll send it to you....
Dr.D

Dr.Wael Deeb wrote:Hi Kirill,
My advice to friend....use Sunsetter C9d,a 32 bit version but it works as charm
It's playing in one of my tournaments now,so....if you're interested I'll send it to you....
Dr.D

Hi Doctor,
I don't have that version. My email adress is mgemuh[@]yahoo[.]com
Thanks,
Matthias.

Kirill Kryukov wrote:Hi Jim, here is another engine with the same bug: Sunsetter C10. Your 64-bit compile of it is losing drawn endings as soon as a 5-men table is used.

So for Sunsetter and other affected engines I have to choose between faster search and EGTB usage. Probably EGTB is less important than speed increase, still I'd much rather use the full-featured and a full-speed version.

The list of Jim's 64-bit compiles affected by a 64-bit EGTB bug: Thor's Hammer 2.28, Pepito 1.59 (in UCI mode), Sunsetter C10. I hope it is not too hard to locate a 64-bit-compatible version of egtb code and recompile. Really looking forward to have fixed 64-bit compiles of these engines.

I was just starting to test Sunsetter C10 for my tourney, now I am really wondering what should I do - switch to a 32-bit version, or switch off EGTB and stay with 64-bit, or use some other compile, or may be switch to another engine for now and come back to Sunsetter later.

Much appreciate your work,
With best regards,
Kirill

Hi Kirill,

I will take a look at these 64 bit compiles very soon and fix (if possible for me)

best ,
Jim.

Jim Ablett wrote:Hi Kirill,

I will take a look at these 64 bit compiles very soon and fix (if possible for me)

best ,
Jim.

Thanks, I'll be really happy if you fix those compiles!

Dr.Wael Deeb wrote:Hi Kirill,
My advice to friend....use Sunsetter C9d,a 32 bit version but it works as charm
It's playing in one of my tournaments now,so....if you're interested I'll send it to you....
Dr.D

Thanks Wael, I still prefer to test C10 - the latest one. If 64-bit compile can't be fixed then I'll have to choose between slower 32-bit compile, or 64-bit one without EGTB.

Or do you think C9d is stronger than C10?

Kirill Kryukov wrote:

Dr.Wael Deeb wrote:Hi Kirill,
My advice to friend....use Sunsetter C9d,a 32 bit version but it works as charm
It's playing in one of my tournaments now,so....if you're interested I'll send it to you....
Dr.D

Thanks Wael, I still prefer to test C10 - the latest one. If 64-bit compile can't be fixed then I'll have to choose between slower 32-bit compile, or 64-bit one without EGTB.

Or do you think C9d is stronger than C10?

Oh,this I can't confirm....what can assure you is that Sunsitter C9d is pretty stable and it's playing around 2050 Elo in my private rating list....
Dr.D

Ok, x64 version fixed now.

I tested against Kirill's ETGB test suite - old version couldn't find the solutions, this one can.
Built with Intel compiler 11.

Sunsetter C10.2 JA Update

Download:
http://www.mediafire.com/?00nmv5nzw2m

Jim.

Terrific! Thanks Jim!!

Many thanks Jim !

Let's update Sunsetter in current ChessWar E

Olivier

Jim, as already reported, your compiles of Sunsetter don't seem to use the opening book... or possibly I am missing something !

Olivier

Olivier Deville wrote:Jim, as already reported, your compiles of Sunsetter don't seem to use the opening book... or possibly I am missing something !

Olivier

Hi Oliver,

I think I found the error.

These files must be renamed back to original: (don't know why I renamed these)

ss-C10_x64_JA.bin > ss-C10.bin

ss-C10_x64_JA.ini > ss-C10.ini

Jim.