CookieCat uses an interactive console interface. The program doesn't have much code for xboard or UCI, only stubs.Don wrote:By the way, just out of curiosity, how does cookie work with user interfaces if it cannot even understand long algebraic? I think both xboard and UCI expect moves in that format, no?
Perft(14)
Moderator: Ras
-
sje
- Posts: 4675
- Joined: Mon Mar 13, 2006 7:43 pm
Re: Distributed perft
-
sje
- Posts: 4675
- Joined: Mon Mar 13, 2006 7:43 pm
CookieCat reloaded
Code: Select all
gail:~ sje$ cd tmp
gail:tmp sje$ ppcx64 -O3 CookieCat.pas
Free Pascal Compiler version 2.6.0 [2011/12/30] for x86_64
Copyright (c) 1993-2011 by Florian Klaempfl and others
Target OS: Darwin for x86_64
Compiling CookieCat.pas
Assembling (pipe) CookieCat.s
Linking CookieCat
ld: warning: -macosx_version_min not specified, assuming 10.7
15904 lines compiled, 1.6 sec
gail:tmp sje$ ./CookieCat
CookieCat 2012.01.30 Copyright (c) 2012 by S. J. Edwards
CookieCat ready
[] help
Enter a command, or a sequence of one or more SAN chess moves
Commands:
bench Run the benchmark
d Display everything
dao Display active options
db Display board
dbbdb Display bitboard database
dbcolor Display board (ANSI color)
dbmono Display board (monochrome)
desc Display evaluation score components
dfen Display FEN
dm Display moves
dmsig Display material signature
dobm Display opening book moves
dp Display position
dpgn Display PGN
dpi Display program identification
dps Display pawn structure components
dtbm Display tablebase moves
dtbs Display tablebase score
echo Echo parameters
efnormal Normalize from an EPD <input-file> to an EPD <output-file>
exit Exit program
ffmate Scan FEN <input-file> data, find mate in <number> full moves
ffnormal Normalize from a FEN <input-file> to a FEN <output-file>
ffperft Summing over a FEN <input-file> data, run perft to <depth>
g Search for a move and then play it
gg Play both sides until the end of the game
help Show help
loadfen Load FEN from an <input-file>
loadpgn Load PGN from an <input-file>
mate Search for a checkmate in <number> full moves
mtperft Run multithreaded perft to <depth> with full node visits
new New game
noop No operation
optreset Reset <option> [<option>]*
optset Set <option> [<option>]*
perftbulk Run perft to <depth> with bulk counting
perftfull Run perft to <depth> with each node visited
perfttran Run perft to <depth> with transposition help
pfnormal Normalize from a PGN <input-file> to a PGN <output-file>
rg Generate and display a single random game
rgdump Dump to a PGN <output-file> <number> random games
rgstat Generate a report for <number> random game(s)
rm Retract move
rmp Retract move pair
rmts Retract move(s) to start
s Search for a move but do not play it
savefen Save FEN to an <output-file>
savepgn Save PGN to an <output-file>
selftest Run the self test
sfen Set FEN <mpd> <good> <cavs> <epsq> <hmvc> <fmvn>
slevfd Set level fixed depth <ply>
slevfn Set level fixed nodes <count>
slevft Set level fixed time <seconds>
slevnt Set level nominal time <seconds>
slevut Set level unlimited time
stag Set PGN <tagname> to <tagvalue>
test Run developer test
ttreset Transposition table data reset
Options:
adcc Auto display: chess clock
adcp Auto display: chess position
mono Monochrome only output
trcv Trace: current variation
trea Trace: EPD analysis
trfd Trace: first time node at depth
trfv Trace: final (predicted) variation
trif Trace: input FEN
trit Trace: iteration start/finish
trpv Trace: predicted variation
trst Trace: search termination
trts Trace: timing statistics
[] perfttran 7
Na3 120142144
Nc3 148527161
Nf3 147678554
Nh3 120669525
a3 106743106
a4 137077337
b3 133233975
b4 134087476
c3 144074944
c4 157756443
d3 227598692
d4 269605599
e3 306138410
e4 309478263
f3 102021008
f4 119614841
g3 135987651
g4 130293018
h3 106678423
h4 138495290
[ts] Count: 3195901860 Time: 000.00:00:39.860 Frequency: 80178170
[] exit
CookieCat done
gail:tmp sje$
-
sje
- Posts: 4675
- Joined: Mon Mar 13, 2006 7:43 pm
Re: CookieCat reloaded
Like perfttran, perftbulk is single threaded.
Code: Select all
[] perftbulk 7
Na3 120142144
Nc3 148527161
Nf3 147678554
Nh3 120669525
a3 106743106
a4 137077337
b3 133233975
b4 134087476
c3 144074944
c4 157756443
d3 227598692
d4 269605599
e3 306138410
e4 309478263
f3 102021008
f4 119614841
g3 135987651
g4 130293018
h3 106678423
h4 138495290
[ts] Count: 3195901860 Time: 000.00:03:21.540 Frequency: 15857407
-
Ajedrecista
- Posts: 2098
- Joined: Wed Jul 13, 2011 9:04 pm
- Location: Madrid, Spain.
Re: (Perft): CookieCat versus JetChess.
Hello Steven:
CookieCat is quite fast regarding perft! Congratulations. I want to see your multi-threaded perft with hash, bulk counting or whatever tricks to speed up the count (please note that I do not have idea about all this technical stuff).
I would like to take part in this distributed project of Perft(14) but I have slow hardware (a dual core Intel Pentium D930 at 3 GHz) that can not be used 24/7 (I need it for other tasks), and very limited access to an Intel i5-760 at 2.8 GHz (both PCs with Windows XP 32-bit), so I can not participate in Perft(14). I wish you good luck with the project.
Regards from Spain.
Ajedrecista.
Good! Just for comparison, I ran Perft(1) to Perft(10) back in January using JetChess 1.0.0.0 perft counter. These counts were ran in an Intel i5-760 (2.8 GHz); JetChess is single core and 32-bit:sje wrote:Like perfttran, perftbulk is single threaded.Code: Select all
[] perftbulk 7 Na3 120142144 Nc3 148527161 Nf3 147678554 Nh3 120669525 a3 106743106 a4 137077337 b3 133233975 b4 134087476 c3 144074944 c4 157756443 d3 227598692 d4 269605599 e3 306138410 e4 309478263 f3 102021008 f4 119614841 g3 135987651 g4 130293018 h3 106678423 h4 138495290 [ts] Count: 3195901860 Time: 000.00:03:21.540 Frequency: 15857407
Code: Select all
Hash = 64 MB:
Perft(1) ---> Time: 0 ms (0:00:00.000).
Perft(2) ---> Time: 0 ms (0:00:00.000).
Perft(3) ---> Time: 0 ms (0:00:00.000).
Perft(4) ---> Time: 1 ms (0:00:00.001).
Perft(5) ---> Time: 23 ms (0:00:00.023).
Perft(6) ---> Time: 320 ms (0:00:00.320).
Perft(7) ---> Time: 3.864 s (0:00:03.864).Code: Select all
Hash = 256 MB:
Perft(8) ---> Time: 56.260 s (0:00:56.260).Code: Select all
Hash = 1 GB:
Perft(9) ---> Time: 764.290 s (0:12:44.290).
Perft(10) ---> Time: 10607.827 s (2:56:47.827).I would like to take part in this distributed project of Perft(14) but I have slow hardware (a dual core Intel Pentium D930 at 3 GHz) that can not be used 24/7 (I need it for other tasks), and very limited access to an Intel i5-760 at 2.8 GHz (both PCs with Windows XP 32-bit), so I can not participate in Perft(14). I wish you good luck with the project.
Regards from Spain.
Ajedrecista.
-
Don
- Posts: 5106
- Joined: Tue Apr 29, 2008 4:27 pm
Re: CookieCat reloaded
Can you make a binary that will work on 64 bit Linux as well as Windows?
Don
Don
sje wrote:Code: Select all
gail:~ sje$ cd tmp gail:tmp sje$ ppcx64 -O3 CookieCat.pas Free Pascal Compiler version 2.6.0 [2011/12/30] for x86_64 Copyright (c) 1993-2011 by Florian Klaempfl and others Target OS: Darwin for x86_64 Compiling CookieCat.pas Assembling (pipe) CookieCat.s Linking CookieCat ld: warning: -macosx_version_min not specified, assuming 10.7 15904 lines compiled, 1.6 sec gail:tmp sje$ ./CookieCat CookieCat 2012.01.30 Copyright (c) 2012 by S. J. Edwards CookieCat ready [] help Enter a command, or a sequence of one or more SAN chess moves Commands: bench Run the benchmark d Display everything dao Display active options db Display board dbbdb Display bitboard database dbcolor Display board (ANSI color) dbmono Display board (monochrome) desc Display evaluation score components dfen Display FEN dm Display moves dmsig Display material signature dobm Display opening book moves dp Display position dpgn Display PGN dpi Display program identification dps Display pawn structure components dtbm Display tablebase moves dtbs Display tablebase score echo Echo parameters efnormal Normalize from an EPD <input-file> to an EPD <output-file> exit Exit program ffmate Scan FEN <input-file> data, find mate in <number> full moves ffnormal Normalize from a FEN <input-file> to a FEN <output-file> ffperft Summing over a FEN <input-file> data, run perft to <depth> g Search for a move and then play it gg Play both sides until the end of the game help Show help loadfen Load FEN from an <input-file> loadpgn Load PGN from an <input-file> mate Search for a checkmate in <number> full moves mtperft Run multithreaded perft to <depth> with full node visits new New game noop No operation optreset Reset <option> [<option>]* optset Set <option> [<option>]* perftbulk Run perft to <depth> with bulk counting perftfull Run perft to <depth> with each node visited perfttran Run perft to <depth> with transposition help pfnormal Normalize from a PGN <input-file> to a PGN <output-file> rg Generate and display a single random game rgdump Dump to a PGN <output-file> <number> random games rgstat Generate a report for <number> random game(s) rm Retract move rmp Retract move pair rmts Retract move(s) to start s Search for a move but do not play it savefen Save FEN to an <output-file> savepgn Save PGN to an <output-file> selftest Run the self test sfen Set FEN <mpd> <good> <cavs> <epsq> <hmvc> <fmvn> slevfd Set level fixed depth <ply> slevfn Set level fixed nodes <count> slevft Set level fixed time <seconds> slevnt Set level nominal time <seconds> slevut Set level unlimited time stag Set PGN <tagname> to <tagvalue> test Run developer test ttreset Transposition table data reset Options: adcc Auto display: chess clock adcp Auto display: chess position mono Monochrome only output trcv Trace: current variation trea Trace: EPD analysis trfd Trace: first time node at depth trfv Trace: final (predicted) variation trif Trace: input FEN trit Trace: iteration start/finish trpv Trace: predicted variation trst Trace: search termination trts Trace: timing statistics [] perfttran 7 Na3 120142144 Nc3 148527161 Nf3 147678554 Nh3 120669525 a3 106743106 a4 137077337 b3 133233975 b4 134087476 c3 144074944 c4 157756443 d3 227598692 d4 269605599 e3 306138410 e4 309478263 f3 102021008 f4 119614841 g3 135987651 g4 130293018 h3 106678423 h4 138495290 [ts] Count: 3195901860 Time: 000.00:00:39.860 Frequency: 80178170 [] exit CookieCat done gail:tmp sje$
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
-
Don
- Posts: 5106
- Joined: Tue Apr 29, 2008 4:27 pm
Re: Distributed perft
Steven,sje wrote:CookieCat uses an interactive console interface. The program doesn't have much code for xboard or UCI, only stubs.Don wrote:By the way, just out of curiosity, how does cookie work with user interfaces if it cannot even understand long algebraic? I think both xboard and UCI expect moves in that format, no?
Komodo already has a perft calculation in it, so I'm building a binary for both platforms to make it a standalone perft calculator and I'm removing all the cruft from the chess programs. I changed the key type to __uint128_t also to avoid hash table collisions.
I also added memoization to speed up the perft calculation but I'm not trying to do anything particularly sophisticated. Also, the move generator was not designed for perft and neither was the move make routine so I'm positive this is far from optimal. I'm pretty sure I'm losing a lot at the leaf nodes because I have to generate and then test each move for legality. In the chess program there is no point in making this optimization because the first move is often a cutoff.
So we can use this as a reference point - I will set this up so that if you follow the interface you can use any perft you choose, but it will have to follow a simple protocol which I will publish shortly.
Bottom line is that my perft will probably be pretty slow - but it will get us started.
What is a good time for a 7 and 8 ply perft? I'm on an i5 modern desktop at 3.00 GHz and here is what I'm getting. It ok if it's way out of line with the better perft generators, as we can plug and play. The time is in seconds of course:
Code: Select all
.................... 1 0.0 20
.................... 2 0.0 400
.................... 3 0.0 8902
.................... 4 0.0 197281
.................... 5 0.2 4865609
.................... 6 2.7 119060324
.................... 7 48.8 3195901860
.................... 8 1257.2 84998978956
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
-
Don
- Posts: 5106
- Joined: Tue Apr 29, 2008 4:27 pm
Re: Distributed perft
Sorry, I just noticed that you published perft numbers. Your perft is obviously very fast so we should try to use that if we can.
Are you using any special hash table magic? I'm simply overwriting any entry in the table if the depth is less than the depth of the new entry that I wish to place in the table.
Don
Are you using any special hash table magic? I'm simply overwriting any entry in the table if the depth is less than the depth of the new entry that I wish to place in the table.
Don
Don wrote:Steven,sje wrote:CookieCat uses an interactive console interface. The program doesn't have much code for xboard or UCI, only stubs.Don wrote:By the way, just out of curiosity, how does cookie work with user interfaces if it cannot even understand long algebraic? I think both xboard and UCI expect moves in that format, no?
Komodo already has a perft calculation in it, so I'm building a binary for both platforms to make it a standalone perft calculator and I'm removing all the cruft from the chess programs. I changed the key type to __uint128_t also to avoid hash table collisions.
I also added memoization to speed up the perft calculation but I'm not trying to do anything particularly sophisticated. Also, the move generator was not designed for perft and neither was the move make routine so I'm positive this is far from optimal. I'm pretty sure I'm losing a lot at the leaf nodes because I have to generate and then test each move for legality. In the chess program there is no point in making this optimization because the first move is often a cutoff.
So we can use this as a reference point - I will set this up so that if you follow the interface you can use any perft you choose, but it will have to follow a simple protocol which I will publish shortly.
Bottom line is that my perft will probably be pretty slow - but it will get us started.
What is a good time for a 7 and 8 ply perft? I'm on an i5 modern desktop at 3.00 GHz and here is what I'm getting. It ok if it's way out of line with the better perft generators, as we can plug and play. The time is in seconds of course:
Code: Select all
.................... 1 0.0 20 .................... 2 0.0 400 .................... 3 0.0 8902 .................... 4 0.0 197281 .................... 5 0.2 4865609 .................... 6 2.7 119060324 .................... 7 48.8 3195901860 .................... 8 1257.2 84998978956
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
-
Don
- Posts: 5106
- Joined: Tue Apr 29, 2008 4:27 pm
Re: (Perft): CookieCat versus JetChess.
I found a lot of performance bugs in my version and I made some general improvements. It is now within a factor of about 2.5 of Stevens and slightly slower than 2x yours. But this is without bulk counting. I think with bulk counting I could make this a lot faster:Ajedrecista wrote:Hello Steven:
Good! Just for comparison, I ran Perft(1) to Perft(10) back in January using JetChess 1.0.0.0 perft counter. These counts were ran in an Intel i5-760 (2.8 GHz); JetChess is single core and 32-bit:sje wrote:Like perfttran, perftbulk is single threaded.Code: Select all
[] perftbulk 7 Na3 120142144 Nc3 148527161 Nf3 147678554 Nh3 120669525 a3 106743106 a4 137077337 b3 133233975 b4 134087476 c3 144074944 c4 157756443 d3 227598692 d4 269605599 e3 306138410 e4 309478263 f3 102021008 f4 119614841 g3 135987651 g4 130293018 h3 106678423 h4 138495290 [ts] Count: 3195901860 Time: 000.00:03:21.540 Frequency: 15857407
Code: Select all
Hash = 64 MB: Perft(1) ---> Time: 0 ms (0:00:00.000). Perft(2) ---> Time: 0 ms (0:00:00.000). Perft(3) ---> Time: 0 ms (0:00:00.000). Perft(4) ---> Time: 1 ms (0:00:00.001). Perft(5) ---> Time: 23 ms (0:00:00.023). Perft(6) ---> Time: 320 ms (0:00:00.320). Perft(7) ---> Time: 3.864 s (0:00:03.864).Code: Select all
Hash = 256 MB: Perft(8) ---> Time: 56.260 s (0:00:56.260).CookieCat is quite fast regarding perft! Congratulations. I want to see your multi-threaded perft with hash, bulk counting or whatever tricks to speed up the count (please note that I do not have idea about all this technical stuff).Code: Select all
Hash = 1 GB: Perft(9) ---> Time: 764.290 s (0:12:44.290). Perft(10) ---> Time: 10607.827 s (2:56:47.827).
I would like to take part in this distributed project of Perft(14) but I have slow hardware (a dual core Intel Pentium D930 at 3 GHz) that can not be used 24/7 (I need it for other tasks), and very limited access to an Intel i5-760 at 2.8 GHz (both PCs with Windows XP 32-bit), so I can not participate in Perft(14). I wish you good luck with the project.
Regards from Spain.
Ajedrecista.
.................... 1 0.0 20
.................... 2 0.0 400
.................... 3 0.0 8902
.................... 4 0.0 197281
.................... 5 0.0 4865609
.................... 6 0.6 119060324
.................... 7 8.2 3195901860
.................... 8 174.4 84998978956
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
-
Don
- Posts: 5106
- Joined: Tue Apr 29, 2008 4:27 pm
Re: CookieCat reloaded
Ok, I have a version that does not do bulk counting but it's pretty fast considering that fact.sje wrote:Like perfttran, perftbulk is single threaded.Code: Select all
[] perftbulk 7 Na3 120142144 Nc3 148527161 Nf3 147678554 Nh3 120669525 a3 106743106 a4 137077337 b3 133233975 b4 134087476 c3 144074944 c4 157756443 d3 227598692 d4 269605599 e3 306138410 e4 309478263 f3 102021008 f4 119614841 g3 135987651 g4 130293018 h3 106678423 h4 138495290 [ts] Count: 3195901860 Time: 000.00:03:21.540 Frequency: 15857407
Code: Select all
.................... 1 0.0 20
.................... 2 0.0 400
.................... 3 0.0 8902
.................... 4 0.0 197281
.................... 5 0.0 4865609
.................... 6 0.5 119060324
.................... 7 6.5 3195901860
.................... 8 100.6 84998978956 So I am only a factor of about 1.88 slower. I think I could more than double the speed if I generated only the legal moves and didn't have to test them but I can tackle that later.
Don
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
-
sje
- Posts: 4675
- Joined: Mon Mar 13, 2006 7:43 pm
Re: CookieCat reloaded
Note that "000.00:03:21.540" means "0 days 0 hours 3 minutes 21 seconds 540 milliseconds" or about 202 seconds to do perft(7) bulk mode which gives frequency of about 15.9 MHz.Don wrote:I don't think I can get depth 7 down to your 3.2 number without doing bulk counting.
This is for a single thread on a 2.67 GHz Intel Xeon from six years ago.