Cinnamon 2.3

[Graham Banks]

http://cinnamonchess.altervista.org/

Chess 960
Killer heuristics
Improved move ordering
Improved Gaviota Tablebase search
Syzygy Tablebases
BMI2 Instructions
Bug fix in enpassant
Bug fix parallel perft

Thanks to the engine author, Giuseppe Cannella.

[Gabor Szots]

Very good news, thanks for sharing.

[Gerd Isenberg]

Thanks! Some details on BMI2 instructions would be nice.

[Gerd Isenberg]

Thanks! Some details on BMI2 instructions would be nice.

PEXT - for linewise (rank, files, diagonal, antidiagonal) sliding piece attacks, arrays indexed by square and 8-bit occupied index per line, thus the author can safe some memory space considering inner six squares only.

Code: Select all

//Bitboard.cpp
u64 Bitboard::BITBOARD_DIAGONAL[64][256];
u64 Bitboard::BITBOARD_ANTIDIAGONAL[64][256];
u64 Bitboard::BITBOARD_FILE[64][256];
u64 Bitboard::BITBOARD_RANK[64][256];

//Bitboard.h
BITBOARD_DIAGONAL[position][diagonalIdx(position, allpieces)] ...

#ifdef USE_BMI2
...
   static uchar diagonalIdx(const int position, const u64 allpieces) {
        return _pext_u64(allpieces, DIAGONAL[position]);
    }
#endif

[CMCanavessi]

Any idea of improvement in terms of elo?

[geko]

Hi all I'm the author, the gain in terms of elo is almost zero at least for Cinnamon, the advantage is that the code is shorter.

[Gerd Isenberg]

Hi all I'm the author, the gain in terms of elo is almost zero at least for Cinnamon, the advantage is that the code is shorter.

Yep, and/mul/shr is not so much slower than pext. However, you may divide the table space by 4 using only 128 Kib lookup rather than 512 without more computation. Because of the outer squares are not relevant for the attack set.

Code: Select all

u64 Bitboard::BITBOARD_DIAGONAL[64][64 /* rather than 256 */];

In terms of Elo it might be almost zero as well, but sometimes using less memory pays off.

[Roland Chastain]

Chess 960

Good new! Glad to know that there is one more engine playing FRC.

I have made a quick test, a tournament with CuteChess. It seems that there are things to fix. When you switch to FRC, there are always things to fix.

I didn't have time to investigate, just copied the games in a PGN file. Hope this can help.

games.zip

[Roland Chastain]

I made another test, running CuteChess with -debug option.

Here is a command on which Cinnamon disconnects:

Code: Select all

25689 >Cinnamon 2.3(0): position fen qnrnbkrb/pppppppp/8/8/8/8/PPPPPPPP/QNRNBKRB w GCgc - 0 1 moves g2g3 f8g8 d1e3 e7e6 d2d4 f7f5 d4d5 b8a6 e1c3 c7c6 d5e6 d8e6 e3f5 f8f5
25689 >Cinnamon 2.3(0): isready
25689 <Cinnamon 2.3(0): fenStr: f8f5 from: 58
25689 <Cinnamon 2.3(0): Tue Oct 20 07:19:32 2020 ********************************** assert error in GenMoves.cpp:539  **********************************
Terminating process of engine Cinnamon 2.3(0)

Here is the PGN file:

cinnamon23-disconnects-4.zip

[geko]

ops there is a bug in chess960 I've moved the web page here http://cinnamonchess.altervista.org/index_2.3.php until the fix.
Gerd I'll check into 'The Outer Squares'

thanks
Giuseppe