Stockfish v2.1 now available

[Jim Ablett]

This is Stockfish 2.1, available from Stockfish site:

Windows/Linux/Android compiles.

http://www.stockfishchess.com/

& Jim's site:

http://jimablett.hostwebs.com/


Diff stats from 2.0.1 says:

54 files changed, 4138 insertions(+), 5656 deletions(-)

So more then 1500 lines of code removed for this release.

In this release a new skill level facility has been introduced.
Stockfish can be set to skill level 20 (default) to play at maximum
strength or, through setting "Skill Level" UCI option, can be lowered
until 0 when should be beatable even by weak players. Implementation
of skill levels has been developed around an idea from Heinz van
Saanen and is time control and CPU speed independent: you can set
skill level at 4 and SF will play with the same strength both on a
cell phone at 1' per game or on a super QUAD machine at 120'+30" TC.

Also of note, "seldepth" UCI info is now correctly implemented and
maximum number of threads has been raised from 16 to 32.

For people compiling themselves, stockfish 2.1 signature is 6487630.
It means that running from command line 'stockfish bench' you should
get at the end 6487630 searched nodes. If this not happens please
report back, it means SF is miscompiled.

We would like to thank Heinz, Onno, Justin, BB+ and Fruity for their
contributions in ideas and useful suggestions, and of course, a big
thank you to Jim for its usual kind help.

Have fun
Stockfish Team

[mhalstern]

Tord, Marco, Joona, Jim


Thanks as always.

Does this version have an expected elo release over 2.0, or is it just to add the new strength limiting features?

[gerold]

Thanks for posting it Jim. Thanks to SF team.

Best,
Gerold.

[wims]

Awesome, thanks!

[Graham Banks]

Tord, Marco, Joona, Jim


Thanks as always.

+1

[Joerg Oster]

+2

[Gian-Carlo Pascutto]

Is it possible to add LTO support in the Linux Makefiles? This should give a good speed boost with GCC 4.5+ on Linux.

When I attempted to add this myself, I ran into an issue that it didn't work combined with profiling. If this is a GCC issue and not me messing up, it should be reported to the GCC devs.

[rbarreira]

54 files changed, 4138 insertions(+), 5656 deletions(-)

So more then 1500 lines of code removed for this release.

Amazing... Is this mostly code simplification, or is it a lot of outright removal of features which don't increase ELO?

[wims]

What about expected rating changes? Is this version any stronger than 2.01, or is it non performance related changes mostly?

[mcostalba]

Is it possible to add LTO support in the Linux Makefiles? This should give a good speed boost with GCC 4.5+ on Linux.

When I attempted to add this myself, I ran into an issue that it didn't work combined with profiling. If this is a GCC issue and not me messing up, it should be reported to the GCC devs.

It is known that LTO support in gcc is still unstable until GCC 4.6 (http://gcc.gnu.org/gcc-4.6/changes.html) :

"A large number of bugs were fixed. GCC itself, Mozilla Firefox and other large applications can be built with LTO enabled."

That's the reason is not supported in the Makefile, but admittely I have not tested it. I have just read that is still somewhat experimental until 4.6 and I have just abandoned the idea to add support for LTO in the Makefile.