Introducing Omniperft

[Dann Corbit]

With PGO optimization:

Code: Select all

OmniPerft 1.0 by Ilari Pihlajisto

r n b q k b n r
p p p p p p p p
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
P P P P P P P P
R N B Q K B N R
FEN: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
perft 4
Perft with 4 thread(s)
197281
Elapsed time: 0 second(s)
perft 5
Perft with 4 thread(s)
4865609
Elapsed time: 1 second(s)
perft 6
Perft with 4 thread(s)
119060324
Elapsed time: 7 second(s)
perft 7
Perft with 4 thread(s)
3195901860
Elapsed time: 90 second(s)

[Denis P. Mendoza]

AMD64 X2 4600+ (2 threads) + windows xp 32 bits

- ICC compile of Denis.

Perft 7 --> 278 s

- MSVC 2005 (no pgo)

Perft 7 --> 310 s

- MSVC 2008 (pgo)

Perft 7 --> 288 s

Denis, your compilation is PGO?

Pedro

No Pedro. I just compiled using the normal optimization switches without PGO. I just tried a quick build pal! Thanks.

[Denis P. Mendoza]

Ilari,

May I know what version of GCC/MingW are you using?

I'm still using GCC 4.1.2/ MingW 3.11 and i'm surprised that ICC and MSVC were beaten by my GCC and MingW (non-PGO) builds too. I'll have a look too.

BTW. It's nice that Sloppy is not 'pthreads' dependent anymore. It improved a bit in speed and search. Congrats.

Denis

[pedrox]

AMD64 X2 4600+ (2 threads) + windows xp 32 bits

- ICC compile of Denis.

Perft 7 --> 278 s

- MSVC 2005 (no pgo)

Perft 7 --> 310 s

- MSVC 2008 (pgo)

Perft 7 --> 288 s

Denis, your compilation is PGO?

Pedro

No Pedro. I just compiled using the normal optimization switches without PGO. I just tried a quick build pal! Thanks.

Thanks Denis,

I've tried to do a compilation with icl + PGO, the result for perft 7 was 291 s, higher than the 278 s of yours without PGO.

I must be doing something wrong or maybe is a bit complicated thing to have an AMD rather than Intel and use a virtual machine?.

Pedro

[ilari]

Ilari,

May I know what version of GCC/MingW are you using?

I'm still using GCC 4.1.2/ MingW 3.11 and i'm surprised that ICC and MSVC were beaten by my GCC and MingW (non-PGO) builds too. I'll have a look too.

I think it's GCC 4.4.2. It was in the Qt bundle, so it may have optimizations which I'm unaware of. Link: http://trolltech.com/downloads/opensour ... indows-cpp

BTW. It's nice that Sloppy is not 'pthreads' dependent anymore. It improved a bit in speed and search. Congrats.

Denis

Thanks.
Currently it affects only perft, but it had to be done. Pthreads for Windows isn't exactly bug-free. I'm going to miss condition variables when I start writing a parallel search though.

[Richard Allbert]

Hi,

Thanks a lot for Omniperft.

I've been using it today to debug implementing FRC in Jabba.

When I put the following FEN into Omniperft, it crashes:

qrk1b1rn/pppp1ppp/4pbn1/8/8/4PBN1/PPPP1PPP/Q1KRB1RN b gb -

I also tried replacing the gb with kq, but same result.

Have you seen this before? The crash says "assertion failed type > 0 file src\zorbist.cpp"

Thanks

Richard

[Richard Allbert]

Sorry to be a pain.... but

Code: Select all

setboard 1rk3r1/8/8/8/8/8/8/1RK1R3 w EBgb -
. r k . . . r .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. R K . R . . .
FEN: 1rk3r1/8/8/8/8/8/8/1RK1R3 w KQkq - 0 1
divide 2
Perft with 2 thread(s)
b1b2: 23
b1b3: 22
b1b4: 21
b1b5: 20
b1b6: 19
b1b7: 18
b1b8: 1
b1a1: 24
c1c2: 24
c1d1: 24
O-O-O: 24
e1e2: 24
e1e3: 24
e1e4: 24
e1e5: 24
e1e6: 24
e1e7: 24
e1e8: 4
e1d1: 24
e1f1: 24
e1g1: 24
e1h1: 25
c1d2: 24
489
Elapsed time: 0 second(s)
1rk3r1/8/8/8/8/8/8/1RK1R3 w EBgb -

I think it's incorrect... there are 3 legal moves after b1b8. Also, e1d1 has 21 moves not 24.

I don't mean to be a pain! You did ask for feedback

Regards

Richard

PS Thanks also for Cutechess...

[Roman Hartmann]

Below is the output of roce.

Roman

Code: Select all

roce: chess960

FRC/Chess960 enabled

roce: setboard 1rk3r1/8/8/8/8/8/8/1RK1R3 w EBgb - 

roce: divide 2
c1b1 22
c1d1 23
c1c2 23
c1d2 23
b1a1 24
b1b2 22
b1b3 21
b1b4 20
b1b5 19
b1b6 18
b1b7 16
b1b8 3
e1f1 23
e1g1 23
e1h1 24
e1d1 21
e1e2 23
e1e3 23
e1e4 23
e1e5 23
e1e6 23
e1e7 21
e1e8 3
Moves: 23
Nodes: 464

roce: 

[Richard Allbert]

Hi Roman,

Thanks, mine was the same also.

I've played a few games with debug on, and all seems ok

Richard

[Sven]

Also, e1d1 has 21 moves not 24.

I see these 20 non-castling moves for black after Re1-d1:
Rb8 has 1 horizontal move
Rb8 has 7 vertical moves
Rg8 has 4 horizontal moves
Rg8 has 7 vertical moves
Kc8 has 1 legal move (Kc8-c7)
so my question is which of the black castles is possible? Is black allowed to castle on queen side by moving Rb8 to d8 and leaving Kc8 where it is? That would explain the number of 21. King side castling is prevented by Rd1.

Sven