Stockfish-2.2.1 Bench Signature Oddity

[zullil]

I compile 2.2.1 from source and get 5457475 from ./stockfish bench

I then start from clean source and change

Code: Select all

 o["Use Search Log"]              = UCIOption(false);

to

Code: Select all

o["Use Search Log"]              = UCIOption(true);

in ucioption.cpp and compile again using the same Makefile. When I run ./stockfish bench, I now get 5460725.

This seems very odd to me. Marco?

[UncombedCoconut]

I think this is OK. When Stockfish writes a search log, it plays through the PV so it can write it in standard algebraic notation. The do_move operations add to the node count, but SF isn't actually searching more positions.

[zullil]

I think this is OK. When Stockfish writes a search log, it plays through the PV so it can write it in standard algebraic notation. The do_move operations add to the node count, but SF isn't actually searching more positions.

Thanks, Justin. I made a number of small changes to the source and Makefile before compiling, so it took a while to determine which was responsible for altering the node count.

This infelicity in the Stockfish code should be fixed I suppose, since search statistics should not be altered by logging the search.

[mcostalba]

I think this is OK. When Stockfish writes a search log, it plays through the PV so it can write it in standard algebraic notation. The do_move operations add to the node count, but SF isn't actually searching more positions.

Yes, I confirm what Justin said. The test for verifying correct signature shall be done with the default version.

BTW I have reports of crashes on big-endian machines, Louis do you have a PowerPC CPU on your Mac ?

[zullil]

I think this is OK. When Stockfish writes a search log, it plays through the PV so it can write it in standard algebraic notation. The do_move operations add to the node count, but SF isn't actually searching more positions.

Yes, I confirm what Justin said. The test for verifying correct signature shall be done with the default version.

BTW I have reports of crashes on big-endian machines, Louis do you have a PowerPC CPU on your Mac ?

Thanks for the confirmation, Marco.

No. All my Macs have Intel processors now.

[UncombedCoconut]

I might be able to test endian issues (and I've fixed plenty of them in the past).
I have an emulated Linux/PowerPC setup for testing. (It runs under QEMU). The bigendian-32 build does not crash during a bench command, and in the latest git version has a bench signature of 5447426, same as x86-64-modern. What did the crash reports say?

[mcostalba]

What did the crash reports say?

I will send you an email. Thanks for your help.