LastEmperor - Chess960 perft tool

JohnWoe · Post by **JohnWoe** » Sat Dec 28, 2019 12:25 pm

I notices www.chessprogramming.org has such a nice list of all Chess960 positions. So I stripped a move generator from my engine and made it a (hopefully) useful tool to somebody!

Source code: https://github.com/SamuraiDangyo/LastEmperor
Release: https://github.com/SamuraiDangyo/LastEm ... s/tag/v1.0

Here is a output on the whole suite 512MB hash. No errors. All correct. Took 3 hours on my processor:

https://github.com/SamuraiDangyo/bins/b ... _perft.txt

Code: Select all

lastemperor -suite 6

Hash can be adjusted by -hash.

16 MB hash gets 51.716M NPS

Code: Select all

:~$ lastemperor -hash 16 -perft 7
### Perft ( 16 MB ) ###

[ rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1 ]
depth          nodes           mnps           time
    0              1          0.000          0.000
    1             20          0.000          0.000
    2            400          0.000          0.000
    3           8902          0.000          0.000
    4         197281         16.440          0.012
    5        4865609         24.574          0.198
    6      119060324         37.797          3.150
    7     3195901860         52.531         60.838

===
depth          nodes           mnps           time
    7     3320034397         51.716         64.198

1024MB hash gets 72.675M NPS

Code: Select all

:~$ lastemperor -hash 1024 -perft 7
### Perft ( 1024 MB ) ###

[ rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1 ]
depth          nodes           mnps           time
    0              1          0.000          0.000
    1             20          0.000          0.000
    2            400          0.000          0.000
    3           8902          8.902          0.001
    4         197281         12.330          0.016
    5        4865609         18.430          0.264
    6      119060324         35.194          3.383
    7     3195901860         76.058         42.019

===
depth          nodes           mnps           time
    7     3320034397         72.675         45.683

More

Code: Select all

:~$ lastemperor -hash 1024 -fen "k7/8/2N5/1N6/8/8/8/K6n b - -" -perft 11
### Perft ( 1024 MB ) ###

[ k7/8/2N5/1N6/8/8/8/K6n b - - ]
depth          nodes           mnps           time
    0              1          0.000          0.000
    1              3          0.000          0.000
    2             51          0.000          0.000
    3            345          0.000          0.000
    4           5301          5.301          0.001
    5          38348          7.670          0.005
    6         588695         25.595          0.023
    7        5041119         47.558          0.106
    8       76286462        176.999          0.431
    9      688086116        503.355          1.367
   10    10529852043       1822.089          5.779
   11   100557269878       2684.319         37.461

===
depth          nodes           mnps           time
   11   111857168362       2476.195         45.173

Ajedrecista · Post by **Ajedrecista** » Sat Dec 28, 2019 7:16 pm

Hello John:

Very interesting! Thank you very much for your effort. I hope to see someting like this "suite <depth>" but with the starting positions from #1 to #960 and rank them from more nodes to less nodes or something like that.

JohnWoe wrote: ↑Sat Dec 28, 2019 12:25 pm[...]

Code: Select all

:~$ lastemperor -hash 1024 -fen "k7/8/2N5/1N6/8/8/8/K6n b - -" -perft 11
### Perft ( 1024 MB ) ###

[ k7/8/2N5/1N6/8/8/8/K6n b - - ]
depth          nodes           mnps           time
    0              1          0.000          0.000
    1              3          0.000          0.000
    2             51          0.000          0.000
    3            345          0.000          0.000
    4           5301          5.301          0.001
    5          38348          7.670          0.005
    6         588695         25.595          0.023
    7        5041119         47.558          0.106
    8       76286462        176.999          0.431
    9      688086116        503.355          1.367
   10    10529852043       1822.089          5.779
   11   100557269878       2684.319         37.461

===
depth          nodes           mnps           time
   11   111857168362       2476.195         45.173

In the meantime, I checked your KNNkn example up to perft(11) with JetChess 1.0.0.0 perft counter (test passed!) and computed the following values up to perft(17):

Code: Select all

FEN: k7/8/2N5/1N6/8/8/8/K6n b - -



perft(12):
  1  nh1-f2           515,853,754,000
  2  nh1-g3           556,617,690,659
  3  ka8-b7           466,628,276,107
-------------------------------------
TOTAL:              1,539,099,720,766



perft(13):
  1  nh1-f2         5,048,588,418,268
  2  nh1-g3         5,445,575,825,983
  3  ka8-b7         4,655,797,637,786
-------------------------------------
TOTAL:             15,149,961,882,037



perft(14):
  1  nh1-f2        77,907,389,191,081
  2  nh1-g3        84,639,794,719,193
  3  ka8-b7        71,271,976,743,195
-------------------------------------
TOTAL:            233,819,160,653,469



perft(15):
  1  nh1-f2       781,035,692,757,588
  2  nh1-g3       848,458,665,705,615
  3  ka8-b7       726,873,258,768,157
-------------------------------------
TOTAL:          2,356,367,617,231,360



perft(16):
  1  nh1-f2    12,085,769,515,531,226
  2  nh1-g3    13,206,217,018,587,597
  3  ka8-b7    11,153,010,997,566,568
-------------------------------------
TOTAL:         36,444,997,531,685,391


perft(17):
  1  nh1-f2   123,457,373,557,212,906
  2  nh1-g3   134,886,385,850,843,993
  3  ka8-b7   115,333,700,166,200,528
-------------------------------------
Total:        373,677,459,574,257,427

perft(18) of that position should not overflow 64-bit integers yet although something could be thought to overcome that when present, like in this example from year 2013:

Re: KBNk ---> perft(20) result.

Regards from Spain.

Ajedrecista.

JohnWoe · Post by **JohnWoe** » Sun Dec 29, 2019 12:17 am

Thanks for the feedback!

Yes it contains all Chess960 positions:
https://github.com/SamuraiDangyo/LastEm ... er/perft.h
I copied these positions from here: https://www.chessprogramming.org/Chess960_Perft_Results

So that's the LastEmperor's suite. All 960 FENs.
With stronger processor depths 7 and 8 could be solved. Mine took 3 hours just for depth 6.

Code: Select all

lastemperor -suite 8

Yes there's a danger in 64bit overflow. Also 64bit hash signature might collide. Might need 128bit. Also support for multiple cores. So these would be the areas to develop, but I keep my software simple.

The reason for this 64bit overflow is because draws are not detected. You would need full history etc.

So this position up to 12. So equal here too.

Code: Select all

./lastemperor -hash 512 -fen "k7/8/2N5/1N6/8/8/8/K6n b - -" -perft 12
### Perft ( 512 MB ) ###

[ k7/8/2N5/1N6/8/8/8/K6n b - - ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1               3           0.000           0.000
    2              51           0.000           0.000
    3             345           0.345           0.001
    4            5301           0.000           0.000
    5           38348           9.587           0.004
    6          588695          25.595           0.023
    7         5041119          49.423           0.102
    8        76286462         190.716           0.400
    9       688086116         549.151           1.253
   10     10529852043        1909.311           5.515
   11    100557269878        2706.718          37.151
   12   1539099720766        4666.399         329.826

===
depth           nodes            mnps            time
   12   1650956889128        4411.080         374.275

Ajedrecista · Post by **Ajedrecista** » Tue Dec 31, 2019 12:06 pm

Hello John:

JohnWoe wrote: ↑Sun Dec 29, 2019 12:17 am Thanks for the feedback!

Yes it contains all Chess960 positions:
https://github.com/SamuraiDangyo/LastEm ... er/perft.h
I copied these positions from here: https://www.chessprogramming.org/Chess960_Perft_Results

So that's the LastEmperor's suite. All 960 FENs.
With stronger processor depths 7 and 8 could be solved. Mine took 3 hours just for depth 6.
Code: Select all
lastemperor -suite 8
[...]

I think I did not make my point clear. Positions from Chess Programming Wiki (as they are at 2019/12/31) are not starting positions themselves, they are positions after 8 moves (maybe plies?) played at random, like the intro of the article suggests. Your verification of that test suite is great, though I was pointing out to other direction: the 960 starting positions. The following threads are courtesy of Mike B:

Strictly for Chess960 Enthusiasts ...

Chess960 Generator and Lookup Tool

Code: Select all

bbqnnrkr/pppppppp/8/8/8/8/PPPPPPPP/BBQNNRKR w KQkq - 0 1
bqnbnrkr/pppppppp/8/8/8/8/PPPPPPPP/BQNBNRKR w KQkq - 0 1
bqnnrbkr/pppppppp/8/8/8/8/PPPPPPPP/BQNNRBKR w KQkq - 0 1
[...]

Castling rights might need a modification, like replacing KQkq by the files where there are rooks in the starting positions, from right to left:

Code: Select all

bbqnnrkr/pppppppp/8/8/8/8/PPPPPPPP/BBQNNRKR w HFhf - 0 1
bqnbnrkr/pppppppp/8/8/8/8/PPPPPPPP/BQNBNRKR w HFhf - 0 1
bqnnrbkr/pppppppp/8/8/8/8/PPPPPPPP/BQNNRBKR w HEhe - 0 1
[...]

The numbering scheme might be a little different between Chess Programming Wiki (CPW) positions and what I posted (Mike's tool, who followed Scharnagl's methods described at Wikipedia). CPW seems to use a different scheme than Wikipedia (Scharnagl's methods):

Code: Select all

In ascent order:

  #       CPW:      Scharnagl:
------------------------------
   0      N/A        BBQNNRKR
   1    BQNBNRKR     BQNBNRKR
   2    BQNNRBKR     BQNNRBKR
[...]    [...]        [...]
 959    RKRNNQBB     RKRNNQBB
 960    BBQNNRKR       N/A

The only difference I was able to see is that CPW #960 is called #0 under Scharnagl's methods.

Anyway, I was thinking in Perft(1), Perft(2), ... of the 960 starting positions:

Code: Select all

Perft(1) = 18 for the positions with four knights on the corners.
Perft(1) = 19 for the positions with two knights (one by side) on the corners.
Perft(1) = 20 for the positions without knights on the corners.

Perft(2) = [Perft(1)]²

And following Perft(n) values should have lower branching factors Perft(n)/Perft(n-1) than CPW test suite (like in normal chess, where branching factor has a raising tendency from the first plies), so this could speed up a little the calculations since there should be fewer nodes to count.

Other than that, thanks again for your effort.

Regards from Spain.

Ajedrecista.

JohnWoe · Post by **JohnWoe** » Tue Dec 31, 2019 6:30 pm

Hello.
Thanks for the feedback!

I added all Chess960 starting positions. With nodes: depth 0 -> depth 6.

I wrote a Chess960 generator myself and generated all 960 positions and put them in here: https://github.com/SamuraiDangyo/LastEm ... erft.h#L10

To me seems legit.

Here are the other verify Chess960 positions: https://github.com/SamuraiDangyo/LastEm ... rft.h#L979

Done 0 -> 6 depths: https://github.com/SamuraiDangyo/bins/b ... uite_6.txt

Id test is just those chessprogramming.org 960 positions up to depth 6.

lastemperor -id

Code: Select all

### Chess960 id test ( 256 MB ) ###
[ 1: bqnb1rkr/pp3ppp/3ppn2/2p5/5P2/P2P4/NPP1P1PP/BQ1BNRKR w HFhf - 2 9 ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              21           0.000           0.000
    2             528           0.000           0.000
    3           12189           0.000           0.000
    4          326672          18.148           0.018
    5         8146062          23.889           0.341
    6       227689589          43.560           5.227

[ 2: 2nnrbkr/p1qppppp/8/1ppb4/6PP/3PP3/PPP2P2/BQNNRBKR w HEhe - 1 9 ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              21           0.000           0.000
    2             807           0.000           0.000
    3           18002          18.002           0.001
    4          667366          19.068           0.035
    5        16253601          23.153           0.702
    6       590751109          47.139          12.532

[ 3: b1q1rrkb/pppppppp/3nn3/8/P7/1PPP4/4PPPP/BQNNRKRB w GE - 1 9 ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              20           0.000           0.000
    2             479           0.000           0.000
    3           10471           0.000           0.000
    4          273318          18.221           0.015
    5         6417013          24.681           0.260
    6       177654692          41.850           4.245

[ 4: qbbnnrkr/2pp2pp/p7/1p2pp2/8/P3PP2/1PPP1KPP/QBBNNR1R w hf - 0 9 ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              22           0.000           0.000
    2             593           0.000           0.000
    3           13440          13.440           0.001
    4          382958          17.407           0.022
    5         9183776          22.676           0.405
    6       274103539          44.132           6.211

[ 5: 1nbbnrkr/p1p1ppp1/3p4/1p3P1p/3Pq2P/8/PPP1P1P1/QNBBNRKR w HFhf - 0 9 ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              28           0.000           0.000
    2            1120           1.120           0.001
    3           31058          31.058           0.001
    4         1171749          18.599           0.063
    5        34030312          25.226           1.349
    6      1250970898          48.811          25.629

....

lastemperor -suite 6

Code: Select all

### Chess960 test suite ( 256 MB ) ###
[ 1: rkqrbbnn/pppppppp/8/8/8/8/PPPPPPPP/RKQRBBNN w DAda - 0 1 ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              19           0.000           0.000
    2             361           0.000           0.000
    3            7678           0.000           0.000
    4          162755          18.084           0.009
    5         3871700          21.874           0.177
    6        91540671          34.182           2.678

[ 2: rkqrbnnb/pppppppp/8/8/8/8/PPPPPPPP/RKQRBNNB w DAda - 0 1 ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              20           0.000           0.000
    2             400           0.000           0.000
    3            8838           0.000           0.000
    4          194525          17.684           0.011
    5         4738058          22.244           0.213
    6       114586864          35.096           3.265

[ 3: rkqrnbbn/pppppppp/8/8/8/8/PPPPPPPP/RKQRNBBN w DAda - 0 1 ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              19           0.000           0.000
    2             361           0.000           0.000
    3            7782           0.000           0.000
    4          166301          16.630           0.010
    5         3956472          21.620           0.183
    6        93289372          34.022           2.742

[ 4: rkqrnnbb/pppppppp/8/8/8/8/PPPPPPPP/RKQRNNBB w DAda - 0 1 ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              20           0.000           0.000
    2             400           0.000           0.000
    3            8948           0.000           0.000
    4          198454          18.041           0.011
    5         4844522          22.021           0.220
    6       117181441          34.577           3.389

[ 5: rkqbrnbn/pppppppp/8/8/8/8/PPPPPPPP/RKQBRNBN w EAea - 0 1 ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              19           0.000           0.000
    2             361           0.000           0.000
    3            7727           7.727           0.001
    4          163964          18.218           0.009
    5         3883643          20.993           0.185
    6        91239486          32.796           2.782

...

JohnWoe · Post by **JohnWoe** » Tue Dec 31, 2019 7:15 pm

Here is hash size effects. I made minimum hash size to 1 MB. If there is collisions then 1 MB should be used. Otherwise half of the available RAM?

All done w/ profile built LastEmperor.

1 MB hash.

Code: Select all

lastemperor -fen "8/2p5/3p4/KP5r/1R3p1k/8/4P1P1/8 w" -hash 1 -perft 8
### Perft ( 1 MB ) ###

[ 8/2p5/3p4/KP5r/1R3p1k/8/4P1P1/8 w ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              14           0.000           0.000
    2             191           0.191           0.001
    3            2812           0.000           0.000
    4           43238          21.619           0.002
    5          674624          29.331           0.023
    6        11030083          44.298           0.249
    7       178633661          52.355           3.412
    8      3009794393          57.333          52.497

===
total      3200179017          56.959          56.184

16 MB hash.

Code: Select all

lastemperor -fen "8/2p5/3p4/KP5r/1R3p1k/8/4P1P1/8 w" -hash 16 -perft 8
### Perft ( 16 MB ) ###

[ 8/2p5/3p4/KP5r/1R3p1k/8/4P1P1/8 w ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              14           0.000           0.000
    2             191           0.000           0.000
    3            2812           2.812           0.001
    4           43238          14.413           0.003
    5          674624          23.263           0.029
    6        11030083          48.591           0.227
    7       178633661          96.925           1.843
    8      3009794393         175.008          17.198

===
total      3200179017         165.804          19.301

512 MB hash.

Code: Select all

lastemperor -fen "8/2p5/3p4/KP5r/1R3p1k/8/4P1P1/8 w" -hash 555 -perft 8
### Perft ( 512 MB ) ###

[ 8/2p5/3p4/KP5r/1R3p1k/8/4P1P1/8 w ]
depth           nodes            mnps            time
    0               1           0.000           0.000
    1              14           0.000           0.000
    2             191           0.000           0.000
    3            2812           0.000           0.000
    4           43238           8.648           0.005
    5          674624          16.454           0.041
    6        11030083          34.905           0.316
    7       178633661          87.011           2.053
    8      3009794393         219.341          13.722

===
total      3200179017         198.313          16.137

JohnWoe · Post by **JohnWoe** » Mon Oct 12, 2020 8:26 pm

New version of LastEmperor (C -> C++):
https://github.com/SamuraiDangyo/LastEmperor

Code: Select all

lastemperor -hash=1024 -fen="r3k2r/p1ppqpb1/bn2pnp1/3PN3/1p2P3/2N2Q1p/PPPBBPPP/R3K2R w KQkq -" -perft=7
[ r3k2r/p1ppqpb1/bn2pnp1/3PN3/1p2P3/2N2Q1p/PPPBBPPP/R3K2R w KQkq - ]
Depth         Nodes          Mnps        Time
0                 1             0           0
1                48             0           0
2             2,039        1.0195       0.001
3            97,862       10.8736       0.008
4         4,085,603       18.8277       0.216
5       193,690,690       33.2859       5.818
6     8,031,647,685       57.9033     138.707
7   374,190,009,323       109.198      3426.7
============================================= 
=   382,419,533,251       107.077     3571.45

LastEmperor - Chess960 perft tool

LastEmperor - Chess960 perft tool

Re: LastEmperor - Chess960 perft tool.

Re: LastEmperor - Chess960 perft tool

Re: LastEmperor - Chess960 perft tool.

Re: LastEmperor - Chess960 perft tool

Re: LastEmperor - Chess960 perft tool

Re: LastEmperor - Chess960 perft tool