Texel 1.02

[tmokonen]

There was your original release of 1.01, and the JA compile. Maybe it was the JA compile that Chan had problems with.

[petero2]

There was your original release of 1.01, and the JA compile. Maybe it was the JA compile that Chan had problems with.

Possibly, or it could have been a derivative version that someone made as an experiment. Anyway, I don't think it is worth worrying about. If it was my 1.01 version or the JA 1.01 version, they are both obsoleted now by the 1.02 version. If it was a derivative version, whoever created it can worry about fixing regressions in that version.

[petero2]

Jim Ablett has already compiled a JA version which includes a 32-bit windows executable.

I have not done any speed comparisons yet between the JA executables and my executables. However the general rule to use the fastest version that works on your computer applies in this case too.

[IWB]

Intel i5 2500K OS W7

1.02 x64 SEE original = 1
1.02 x64 old original = 0.93
1.02 JA 64bit pop = 0.88

AMD Phenom 2 OS XP-64:

1.02 x64 SEE original = loads, crashing at analysis start!
1.02. old original = 1
1.02 JA 64bit pop = 0.81

AMD 8350 (Piledriver) OS XP-64

1.02 x64 SEE original = 1
1.02. old original = 0.92
1.02 JA 64bit pop = 0.86



A pitty that the original is not running on Phenom2s ...

Bye
Ingo

[lucasart]

Congratulations Peter for this new version of Texel.

I downloaded it and compiled it on Linux with GCC 4.7, using the makefile provided. Works like a bliss:

Code: Select all

$ make
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/bitBoard.o src/bitBoard.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/book.o src/book.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/computerPlayer.o src/computerPlayer.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/enginecontrol.o src/enginecontrol.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/evaluate.o src/evaluate.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/game.o src/game.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/history.o src/history.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/humanPlayer.o src/humanPlayer.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/killerTable.o src/killerTable.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/kpkTable.o src/kpkTable.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/krkpTable.o src/krkpTable.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/krpkrTable.o src/krpkrTable.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/material.o src/material.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/move.o src/move.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/moveGen.o src/moveGen.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/parameters.o src/parameters.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/piece.o src/piece.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/position.o src/position.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/random.o src/random.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/search.o src/search.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/texel.o src/texel.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/textio.o src/textio.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/transpositionTable.o src/transpositionTable.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/treeLogger.o src/treeLogger.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/tuigame.o src/tuigame.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/uciprotocol.o src/uciprotocol.cpp
g++ -O3 -Wall -std=c++11 -pthread -m64 -c -o objdef/util.o src/util.cpp
g++ -pthread -m64 -o texel objdef/bitBoard.o objdef/book.o objdef/computerPlayer.o objdef/enginecontrol.o objdef/evaluate.o objdef/game.o objdef/history.o objdef/humanPlayer.o objdef/killerTable.o objdef/kpkTable.o objdef/krkpTable.o objdef/krpkrTable.o objdef/material.o objdef/move.o objdef/moveGen.o objdef/parameters.o objdef/piece.o objdef/position.o objdef/random.o objdef/search.o objdef/texel.o objdef/textio.o objdef/transpositionTable.o objdef/treeLogger.o objdef/tuigame.o objdef/uciprotocol.o objdef/util.o

I cannot help notice that there are two little optimizations you should include in your makefile:
(i) use '-flto' in the compiling and linking flags. should make your compile a little faster.
(ii) strip the executables, to remove the debug info produced by gcc. should make your compiles smaller.

[petero2]

AMD Phenom 2 OS XP-64:

1.02 x64 SEE original = loads, crashing at analysis start!
1.02. old original = 1
1.02 JA 64bit pop = 0.81

A pitty that the original is not running on Phenom2s ...

Can you please try this version: http://dl.dropboxusercontent.com/u/8968 ... e3-ctz.exe. It is compiled with "-march=athlon64-sse3 -DHAVE_CTZ", which I think should be good for this CPU.

[IWB]

Intel i5 2500K OS W7

1.02 x64 SEE original = 1
1.02 x64 old original = 0.93
1.02 x64 athlon = 0.93
1.02 JA 64bit pop = 0.88

AMD Phenom 2 OS XP-64:

1.02 x64 SEE original = loads, crashing at analysis start!
1.02. old original = 1
1.02 x64 athlon = 1.02
1.02 JA 64bit pop = 0.81

AMD 8350 (Piledriver) OS XP-64

1.02 x64 SEE original = 1
1.02. old original = 0.92
1.02 x64 athlon = 0.93
1.02 JA 64bit pop = 0.86


The special Athlon version is not nessesary. The differences to the "old" version are so small and you limit even more the hardware basis ...

BYe
Ingo

[petero2]

The special Athlon version is not nessesary. The differences to the "old" version are so small and you limit even more the hardware basis ...

OK, I see, thanks for testing. I think it is the popcount instruction that makes the most speed difference and since the Phenom II doesn't have such an instruction your results make sense.

[petero2]

Congratulations Peter for this new version of Texel.

Thanks!

I downloaded it and compiled it on Linux with GCC 4.7, using the makefile provided. Works like a bliss:

Code: Select all

$ make

If you have an Intel CPU with popcount support, you may want to use "make texel64" instead, which adds the compiler flags "-march=corei7 -DHAVE_CTZ -DHAVE_POPCNT".

I cannot help notice that there are two little optimizations you should include in your makefile:
(i) use '-flto' in the compiling and linking flags. should make your compile a little faster.
(ii) strip the executables, to remove the debug info produced by gcc. should make your compiles smaller.

I have never been able to get any speed improvement from "-flto". It does make the program significantly smaller (418KiB vs 483KiB), but not faster in my tests. I don't know why this happens, but I suspect it is because I have many performance critical functions declared inline in header files.

Regarding stripping the executables, yes that makes sense.

[Modern Times]

The special Athlon version is not nessesary. The differences to the "old" version are so small and you limit even more the hardware basis ...

OK, I see, thanks for testing. I think it is the popcount instruction that makes the most speed difference and since the Phenom II doesn't have such an instruction your results make sense.

The Phenom II does have the popcount instruction.