Discussion of anything and everything relating to chess playing software and machines.
Moderator: Ras
MikeB
Posts: 4889 Joined: Thu Mar 09, 2006 6:34 am
Location: Pen Argyl, Pennsylvania
Post
by MikeB » Sat Jul 16, 2016 5:02 am
includes current source from the official SF site:
https://github.com/official-stockfish/S ... ree/master
Plus two pulls (code not yet merged into master) that passed with flying colors:
Code: Select all
submitted by ajithcj
Gradually relax the NMP staticEval check as we go to higher depths.
Use tuned values.
STC
LLR: 2.95 (-2.94,2.94) [0.00,5.00]
Total: 16745 W: 3371 L: 3168 D: 10206
LTC:
LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 5906 W: 875 L: 736 D: 4295
Code: Select all
submitted by pb00068
ELO: 1.74 +-3.5 (95%) LOS: 83.7%
Total: 10000 W: 1324 L: 1274 D: 7402 tc=10+0.1 th = 31
http://tests.stockfishchess.org/tests/view/5777d7560ebc5972faa1657b
ELO: 5.10 +-3.5 (95%) LOS: 99.8%
Total: 12000 W: 1993 L: 1817 D: 8190 tc=3+0.03 th = 31
http://tests.stockfishchess.org/tests/view/5774bd640ebc5972faa16514
source used to compile, also a Mac exe is also available for download:
https://github.com/MichaelB7/Stockfish
Code: Select all
===========================
Total time (ms) : 4091
Nodes searched : 8548212
Nodes/second : 2089516
yanquis1972
Posts: 1766 Joined: Wed Jun 03, 2009 12:14 am
Post
by yanquis1972 » Sat Jul 16, 2016 8:33 pm
thanks for the notice, michael.
thats 2 builds in 2 weeks with significant elo improvements...sincere best wishes to the komodo guys trying to keep pace, & congrats to everyone involved w/ SF on the continued great work!
MikeB
Posts: 4889 Joined: Thu Mar 09, 2006 6:34 am
Location: Pen Argyl, Pennsylvania
Post
by MikeB » Sat Jul 16, 2016 8:35 pm
yanquis1972 wrote: thanks for the notice, michael.
thats 2 builds in 2 weeks with significant elo improvements...sincere best wishes to the komodo guys trying to keep pace, & congrats to everyone involved w/ SF on the continued great work!
;>) I only build a new version when I see two or more patches adding a nice bump .., its' rare to see it back to back weeks...
S.Taylor
Posts: 8514 Joined: Thu Mar 09, 2006 3:25 am
Location: Jerusalem Israel
Post
by S.Taylor » Sat Jul 16, 2016 10:00 pm
yanquis1972 wrote: thanks for the notice, michael.
thats 2 builds in 2 weeks with significant elo improvements...sincere best wishes to the komodo guys trying to keep pace,
Isn't komodo actually trying to keep AHEAD of sf?
MikeB
Posts: 4889 Joined: Thu Mar 09, 2006 6:34 am
Location: Pen Argyl, Pennsylvania
Post
by MikeB » Tue Jul 19, 2016 5:05 am
Updated Source and Mac exe on Github
https://github.com/MichaelB7/Stockfish
includes all the latest commits from the official SF site plus Ronald's latest syzygy updates:
https://github.com/syzygy1/tb
Code: Select all
this includes:
Two TB initialisation fixes:
- potential memory corruption if SyzygyPath was set to <empty> twice in a row;
- potential undefined behavior by initialising a mutex twice.
-Simplification of probe_dtz().
-more care en passant handling in probe_wdl() and in particular setting of *success = 2
MikeB
Posts: 4889 Joined: Thu Mar 09, 2006 6:34 am
Location: Pen Argyl, Pennsylvania
Post
by MikeB » Tue Jul 19, 2016 5:51 am
MikeB wrote: Updated Source and Mac exe on Github
https://github.com/MichaelB7/Stockfish
includes all the latest commits from the official SF site plus Ronald's latest syzygy updates:
https://github.com/syzygy1/tb
Code: Select all
this includes:
Two TB initialisation fixes:
- potential memory corruption if SyzygyPath was set to <empty> twice in a row;
- potential undefined behavior by initialising a mutex twice.
-Simplification of probe_dtz().
-more care en passant handling in probe_wdl() and in particular setting of *success = 2
there was another update made just a few minutes ago, source and exe updated
MikeB
Posts: 4889 Joined: Thu Mar 09, 2006 6:34 am
Location: Pen Argyl, Pennsylvania
Post
by MikeB » Sun Jul 31, 2016 1:57 am
The latest
unoffical developmental Stockfish...
- the main change since the 071816 - it now plays endgame positions about 10x faster when it is hitting the tb's (no expected ELO impact) also a few other changes pulled from the official SF site that may be worth a few ELO
the dash"-y" signifies that it incorporates Ronald de Man, updated tb files for SF
source and Mac exe available @ the Github link below
https://github.com/MichaelB7/Stockfish
not expecting a big change here - just making sure that i didn't break anything
Code: Select all
240 of 1200 games completed...
time control: 1500+25
Date: 07/30/16 : 19:54:17
240 game(s) loaded
Rank Name Rating Δ + - # Σ Σ% W L D W% =% OppR
------------------------------------------------------------------------------------------------------
1 Stockfish 073016-y 3126 0.0 30 30 120 66.0 55.0 33 21 66 27.5 55.0 3091
2 Stockfish 072516-y 3118 8.1 29 29 119 63.5 53.4 26 18 75 21.8 63.0 3094
3 Stockfish 072516-x 3110 7.6 30 30 121 62.5 51.7 29 25 67 24.0 55.4 3096
4 Komodo 10.1 64-bit 3046 64.5 31 31 120 48.0 40.0 23 47 50 19.2 41.7 3118
------------------------------------------------------------------------------------------------------
MikeB
Posts: 4889 Joined: Thu Mar 09, 2006 6:34 am
Location: Pen Argyl, Pennsylvania
Post
by MikeB » Sun Jul 31, 2016 2:06 am
MikeB wrote: The latest
unoffical developmental Stockfish...
- the main change since the 071816 - it now plays endgame positions about 10x faster when it is hitting the tb's (no expected ELO impact) also a few other changes pulled from the official SF site that may be worth a few ELO
the dash"-y" signifies that it incorporates Ronald de Man, updated tb files for SF
source and Mac exe available @ the Github link below
https://github.com/MichaelB7/Stockfish
not expecting a big change here - just making sure that i didn't break anything
Code: Select all
240 of 1200 games completed...
time control: 1500+25
Date: 07/30/16 : 19:54:17
240 game(s) loaded
Rank Name Rating Δ + - # Σ Σ% W L D W% =% OppR
------------------------------------------------------------------------------------------------------
1 Stockfish 073016-y 3126 0.0 30 30 120 66.0 55.0 33 21 66 27.5 55.0 3091
2 Stockfish 072516-y 3118 8.1 29 29 119 63.5 53.4 26 18 75 21.8 63.0 3094
3 Stockfish 072516-x 3110 7.6 30 30 121 62.5 51.7 29 25 67 24.0 55.4 3096
4 Komodo 10.1 64-bit 3046 64.5 31 31 120 48.0 40.0 23 47 50 19.2 41.7 3118
------------------------------------------------------------------------------------------------------
also updated the makefile so that it automatically include the date of the build in the name of the exe
Code: Select all
NOW := $(shell date +"%m%d%y" | tr ' :' '__')
EXE = Stockfish-$(NOW)-y
and for those that have a Mac and uses gcc with PGO with modern settings, that is is now the default - simply type "make" - makefile also points SF to where the syzygy tb files are located - modify as you see fit
Code: Select all
Mac-Pro:Stockfish-073016-y.d michaelbyrne$ make
make -j profile-build ARCH=x86-64-modern COMP=gcc COMPCXX=g++-mp-4.8 SYZYGY-PATH="/Applications/Tablebases"
/Applications/Xcode.app/Contents/Developer/usr/bin/make ARCH=x86-64-modern COMP=gcc config-sanity
Config:
debug: 'no'
optimize: 'yes'
arch: 'x86_64'
bits: '64'
prefetch: 'yes'
bsfq: 'yes'
popcnt: 'yes'
sse: 'yes'
pext: 'no'
Flags:
CXX: g++-mp-4.8
CXXFLAGS: -std=c++11 -pedantic -Wextra -Wshadow -ftree-loop-distribution -ftree-vectorize -arch x86_64 -mmacosx-version-min=10.9 -DNDEBUG -flto -O2 --param max-inline-insns-auto=100 --param inline-min-speedup=25 -mdynamic-no-pic -DIS_64BIT -msse -DUSE_BSFQ --param l1-cache-size=32 --param l2-cache-size=256 -mtune=native -mssse3 -mpopcnt -march=native -mtune=generic -funroll-loops -DUSE_POPCNT -DPREFETCH
LDFLAGS: -arch x86_64 -mmacosx-version-min=10.9 -static-libstdc++ -lpthread -std=c++11 -pedantic -Wextra -Wshadow -ftree-loop-distribution -ftree-vectorize -arch x86_64 -mmacosx-version-min=10.9 -DNDEBUG -flto -O2 --param max-inline-insns-auto=100 --param inline-min-speedup=25 -mdynamic-no-pic -DIS_64BIT -msse -DUSE_BSFQ --param l1-cache-size=32 --param l2-cache-size=256 -mtune=native -mssse3 -mpopcnt -march=native -mtune=generic -funroll-loops -DUSE_POPCNT -DPREFETCH
Testing config sanity. If this fails, try 'make help' ...
Step 0/4. Preparing for profile build.
/Applications/Xcode.app/Contents/Developer/usr/bin/make ARCH=x86-64-modern COMP=gcc gcc-profile-prepare
/Applications/Xcode.app/Contents/Developer/usr/bin/make ARCH=x86-64-modern COMP=gcc gcc-profile-clean
Step 1/4. Building executable for benchmark ...
/Applications/Xcode.app/Contents/Developer/usr/bin/make -j ARCH=x86-64-modern COMP=gcc gcc-profile-make
/Applications/Xcode.app/Contents/Developer/usr/bin/make ARCH=x86-64-modern COMP=gcc \
EXTRACXXFLAGS='-fprofile-generate ' \
EXTRALDFLAGS='-lgcov' \
all
g++-mp-4.8 -std=c++11 -fprofile-generate -pedantic -Wextra -Wshadow -ftree-loop-distribution -ftree-vectorize -arch x86_64 -mmacosx-version-min=10.9 -DNDEBUG -flto -O2 --param max-inline-insns-auto=100 --param inline-min-speedup=25 -mdynamic-no-pic -DIS_64BIT -msse -DUSE_BSFQ --param l1-cache-size=32 --param l2-cache-size=256 -mtune=native -mssse3 -mpopcnt -march=native -mtune=generic -funroll-loops -DUSE_POPCNT -DPREFETCH -c -o src/stockfish.o src/stockfish.cpp
g++-mp-4.8 -std=c++11 -fprofile-generate -pedantic -Wextra -Wshadow -ftree-loop-distribution -ftree-vectorize -arch x86_64 -mmacosx-version-min=10.9 -DNDEBUG -flto -O2 --param max-inline-insns-auto=100 --param inline-min-speedup=25 -mdynamic-no-pic -DIS_64BIT -msse -DUSE_BSFQ --param l1-cache-size=32 --param l2-cache-size=256 -mtune=native -mssse3 -mpopcnt -march=native -mtune=generic -funroll-loops -DUSE_POPCNT -DPREFETCH -c -o src/endgame.o src/endgame.cpp
g++-mp-4.8 -o Stockfish-073016-y src/stockfish.o src/endgame.o -lgcov -arch x86_64 -mmacosx-version-min=10.9 -static-libstdc++ -lpthread -std=c++11 -fprofile-generate -pedantic -Wextra -Wshadow -ftree-loop-distribution -ftree-vectorize -arch x86_64 -mmacosx-version-min=10.9 -DNDEBUG -flto -O2 --param max-inline-insns-auto=100 --param inline-min-speedup=25 -mdynamic-no-pic -DIS_64BIT -msse -DUSE_BSFQ --param l1-cache-size=32 --param l2-cache-size=256 -mtune=native -mssse3 -mpopcnt -march=native -mtune=generic -funroll-loops -DUSE_POPCNT -DPREFETCH
Step 2/4. Running benchmark for pgo-build ...
( echo setoption value SyzygyPath value "/Applications/Tablebases" ; echo bench 16 1 300 default time ) | ./Stockfish-073016-y > /dev/null
Position: 1/37
Position: 2/37
Position: 3/37
Position: 4/37
Position: 5/37
Position: 6/37
Position: 7/37
Position: 8/37
Position: 9/37
Position: 10/37
Position: 11/37
Position: 12/37
Position: 13/37
Position: 14/37
Position: 15/37
Position: 16/37
Position: 17/37
Position: 18/37
Position: 19/37
Position: 20/37
Position: 21/37
Position: 22/37
Position: 23/37
Position: 24/37
Position: 25/37
Position: 26/37
Position: 27/37
Position: 28/37
Position: 29/37
Position: 30/37
Position: 31/37
Position: 32/37
Position: 33/37
Position: 34/37
Position: 35/37
Position: 36/37
Position: 37/37
===========================
Total time (ms) : 11237
Nodes searched : 13209396
Nodes/second : 1175526
Step 3/4. Building final executable ...
/Applications/Xcode.app/Contents/Developer/usr/bin/make -j ARCH=x86-64-modern COMP=gcc gcc-profile-use
/Applications/Xcode.app/Contents/Developer/usr/bin/make ARCH=x86-64-modern COMP=gcc \
EXTRACXXFLAGS='-fprofile-use -fno-peel-loops -fno-tracer' \
EXTRALDFLAGS='-lgcov' \
all
g++-mp-4.8 -std=c++11 -fprofile-use -fno-peel-loops -fno-tracer -pedantic -Wextra -Wshadow -ftree-loop-distribution -ftree-vectorize -arch x86_64 -mmacosx-version-min=10.9 -DNDEBUG -flto -O2 --param max-inline-insns-auto=100 --param inline-min-speedup=25 -mdynamic-no-pic -DIS_64BIT -msse -DUSE_BSFQ --param l1-cache-size=32 --param l2-cache-size=256 -mtune=native -mssse3 -mpopcnt -march=native -mtune=generic -funroll-loops -DUSE_POPCNT -DPREFETCH -c -o src/stockfish.o src/stockfish.cpp
g++-mp-4.8 -std=c++11 -fprofile-use -fno-peel-loops -fno-tracer -pedantic -Wextra -Wshadow -ftree-loop-distribution -ftree-vectorize -arch x86_64 -mmacosx-version-min=10.9 -DNDEBUG -flto -O2 --param max-inline-insns-auto=100 --param inline-min-speedup=25 -mdynamic-no-pic -DIS_64BIT -msse -DUSE_BSFQ --param l1-cache-size=32 --param l2-cache-size=256 -mtune=native -mssse3 -mpopcnt -march=native -mtune=generic -funroll-loops -DUSE_POPCNT -DPREFETCH -c -o src/endgame.o src/endgame.cpp
g++-mp-4.8 -o Stockfish-073016-y src/stockfish.o src/endgame.o -lgcov -arch x86_64 -mmacosx-version-min=10.9 -static-libstdc++ -lpthread -std=c++11 -fprofile-use -fno-peel-loops -fno-tracer -pedantic -Wextra -Wshadow -ftree-loop-distribution -ftree-vectorize -arch x86_64 -mmacosx-version-min=10.9 -DNDEBUG -flto -O2 --param max-inline-insns-auto=100 --param inline-min-speedup=25 -mdynamic-no-pic -DIS_64BIT -msse -DUSE_BSFQ --param l1-cache-size=32 --param l2-cache-size=256 -mtune=native -mssse3 -mpopcnt -march=native -mtune=generic -funroll-loops -DUSE_POPCNT -DPREFETCH
Step 4/4. Deleting profile data ...
/Applications/Xcode.app/Contents/Developer/usr/bin/make ARCH=x86-64-modern COMP=gcc gcc-profile-clean
Mac-Pro:Stockfish-073016-y.d michaelbyrne$
MikeB
Posts: 4889 Joined: Thu Mar 09, 2006 6:34 am
Location: Pen Argyl, Pennsylvania
Post
by MikeB » Sun Jul 31, 2016 3:25 am
MikeB wrote: The latest
unoffical developmental Stockfish...
- the main change since the 071816 - it now plays endgame positions about 10x faster when it is hitting the tb's (no expected ELO impact) also a few other changes pulled from the official SF site that may be worth a few ELO
the dash"-y" signifies that it incorporates Ronald de Man, updated tb files for SF
source and Mac exe available @ the Github link below
https://github.com/MichaelB7/Stockfish
not expecting a big change here - just making sure that i didn't break anything
Code: Select all
240 of 1200 games completed...
time control: 1500+25
Date: 07/30/16 : 19:54:17
240 game(s) loaded
Rank Name Rating Δ + - # Σ Σ% W L D W% =% OppR
------------------------------------------------------------------------------------------------------
1 Stockfish 073016-y 3126 0.0 30 30 120 66.0 55.0 33 21 66 27.5 55.0 3091
2 Stockfish 072516-y 3118 8.1 29 29 119 63.5 53.4 26 18 75 21.8 63.0 3094
3 Stockfish 072516-x 3110 7.6 30 30 121 62.5 51.7 29 25 67 24.0 55.4 3096
4 Komodo 10.1 64-bit 3046 64.5 31 31 120 48.0 40.0 23 47 50 19.2 41.7 3118
------------------------------------------------------------------------------------------------------
wrap-up on this little test run to make sure nothing was broken:
Code: Select all
200 rounds and 1200 games completed...
time control: 1500+25
Date: 07/30/16 : 21:00:42
1200 game(s) loaded
Rank Name Rating Δ + - # Σ Σ% W L D W% =% OppR
------------------------------------------------------------------------------------------------------
1 Stockfish 073016-y 3124 0.0 13 13 600 327.0 54.5 149 95 356 24.8 59.3 3092
2 Stockfish 072516-x 3117 6.9 13 13 600 318.0 53.0 135 99 366 22.5 61.0 3094
3 Stockfish 072516-y 3112 5.3 13 13 600 312.5 52.1 133 108 359 22.2 59.8 3096
4 Komodo 10.1 64-bit 3047 64.2 13 13 600 242.5 40.4 112 227 261 18.7 43.5 3118
------------------------------------------------------------------------------------------------------
MikeB
Posts: 4889 Joined: Thu Mar 09, 2006 6:34 am
Location: Pen Argyl, Pennsylvania
Post
by MikeB » Sun Jul 31, 2016 5:44 am
Another update - this change doubles the number of skill levels from 20 to 40.
As an example, if you were to move from playing level 4 vs level 5 prior to this change, depending upon your hardware, that could be a 130 point ELO jump ..a pretty nasty jump for a human to adjust to..
Code: Select all
100 rounds and 100 games completed...
time control: 600+10
Date: 07/30/16 : 23:23:41
100 game(s) loaded
Rank Name Rating Δ + - # Σ Σ% W L D W% =% OppR
------------------------------------------------------------------------------------------------------
1 Stockfish 073016-y 3164 0.0 34 34 100 67.0 67.0 62 28 10 62.0 10.0 3036
2 Stockfish 073016-2y 3036 129.0 34 34 100 33.0 33.0 28 62 10 28.0 10.0 3164
------------------------------------------------------------------------------------------------------
Code: Select all
with the extra levels, the jump will only be ~half as much ELO:
100 rounds and 100 games completed...
time control: 600+10
Date: 07/30/16 : 23:27:08
100 game(s) loaded
Rank Name Rating Δ + - # Σ Σ% W L D W% =% OppR
------------------------------------------------------------------------------------------------------
1 Stockfish 073016-y 3125 0.0 32 32 100 57.5 57.5 52 37 11 52.0 11.0 3075
2 Stockfish 073016-2y 3075 50.9 32 32 100 42.5 42.5 37 52 11 37.0 11.0 3125
------------------------------------------------------------------------------------------------------
Code: Select all
note:
old level 20 == new level 40
old level 19 == new level 38 -> level 39 did not exist under the level command
...
old level 5 == new level 10
new level 9 ( did not exist prior)
old level 4 == new level 8
..etc etc
the first set of games above were played at levels 8 & 10, the second set of games were played at levels 9 & 10...
source and Mac exe can be found here
https://github.com/MichaelB7/Stockfish