Candidate fix for stockfish problems under gcc

[mcostalba]

This is a workaround that fixes the miscompile on gcc

You can download sources from:

http://www.mediafire.com/?ch0tmm3zoj5

Unfortunatly I had to set optimization to -O1 instead of -O2 because many versions of gcc miscompile the file value.h, I will post on the technical forum the details.

Note that with some versions also -O2 works and even the optimal -O3, but you should test yourself.

Please test with gcc and report any problem.

IMPORTANT: There is NO change for Windows version, so the currently JA compiles do not need to be updated.

[jpqy]

Hi Marco,

So there come no 1.6.1 ,and the problems are all related to gcc version?!
I was waiting for the 1.6.1 before i begin to test it!

JP.

[mcostalba]

Hi Marco,

So there come no 1.6.1 ,and the problems are all related to gcc version?!
I was waiting for the 1.6.1 before i begin to test it!

JP.

well, if you follow the download link you will see that the sources are tagged 1.6.1 so that's the 1.6.1 but, as I said, the problem was _only_ related to gcc, so if you have in mind to test, for instance, the windows binaries you can already do this with the 1.6 JA version.

[Graham Banks]

Hi Marco,

So there come no 1.6.1 ,and the problems are all related to gcc version?!
I was waiting for the 1.6.1 before i begin to test it!

JP.

The windows compile is okay. It was just this issue as far as I was aware.

Cheers,
Graham.

[Werner]

Congrats Marco,
to you and your team for the 3rd place in our single list behind Rybka 3 x64 1CPU and Rybka 3 w32 1CPU!
The improvement is around 80+ Elos!!
Our update follows on Sunday!

Best wishes
Werner

[pgeorges]

I reported that Stockfish scored 0.0 for a lost position with -O3 (see http://talkchess.com/forum/viewtopic.ph ... highlight=), and was ok with option -O2.
But now it crashes whatever option I choose (-O1, -O2, -O3). Are you sure it is a compiler bug ?
GCC used : gcc 4.3.2 [gcc-4_3-branch revision 141291]

Pascal Georges

[Martin Thoresen]

Marco,

Thanks for this new update.

I am finding something weird to happen in DS 11 GUI.

I am running Stockfish 1.6 64-bit 4CPU (JA) with 1024 MB Hash. No problem. Then 2048 MB Hash, also no problem. But when I set 4096 MB Hash, the hash "counter" goes to 100% immediately. With the other two options, it starts on 0.0% and goes up after some searching in a position, only to go to 100% after something like 5-8 minutes.

Any explanation for this?

PS: I have plenty of RAM to go around, so that is not the problem.

Best Regards,
Martin

[mcostalba]

I reported that Stockfish scored 0.0 for a lost position with -O3 (see http://talkchess.com/forum/viewtopic.ph ... highlight=), and was ok with option -O2.
But now it crashes whatever option I choose (-O1, -O2, -O3). Are you sure it is a compiler bug ?
GCC used : gcc 4.3.2 [gcc-4_3-branch revision 141291]

Pascal Georges

Could you please try with the correct fen ?

8/6p1/2p4p/pp2k3/P7/2PK2P1/7P/8 w - - 0 1

[mcostalba]

Marco,

Thanks for this new update.

I am finding something weird to happen in DS 11 GUI.

I am running Stockfish 1.6 64-bit 4CPU (JA) with 1024 MB Hash. No problem. Then 2048 MB Hash, also no problem. But when I set 4096 MB Hash, the hash "counter" goes to 100% immediately. With the other two options, it starts on 0.0% and goes up after some searching in a position, only to go to 100% after something like 5-8 minutes.

Any explanation for this?

PS: I have plenty of RAM to go around, so that is not the problem.

Best Regards,
Martin

Currently we cannot allocate more then 2GB (2048MB) of hash table size. The bug here is that UCI option "Hash" allows to set up to 4096 MB while instead should be 2048 MB.

Thanks for pointing out that, I will fix this for next release becasue it is just a wrong maximum limit.

[pgeorges]

Could you please try with the correct fen ?

8/6p1/2p4p/pp2k3/P7/2PK2P1/7P/8 w - - 0 1

Yes, you are right. In that case with -O2 -> correct score, with -O3 -> draw.
But at least no more crash.

Pascal Georges