Macintosh Glaurung users: Please help

[Tord Romstad]

Hello all,

The Mac OS X universal binary of Glaurung included in the Glaurung 2.0.1 download works fine on both my Macs, a Core 2 Duo iMac and a Core Duo MacBook, both running OS X 10.5.1. However, one user with a Core Duo iMac running OS X 10.4.11 reported that the binary does not work for him. He gets the following error message:

Code: Select all

dyld: lazy symbol binding failed: Symbol not found: _pthread_cond_init$UNIX2003
 Referenced from: /Users/louis/Desktop/glaurung2/MacOSX/./glaurung
 Expected in: /usr/lib/libSystem.B.dylib

I have no idea what is the cause of this error message. It might be related to the fact that he is running 10.4 while I am running 10.5, but I am not sure. It would therefore be very useful for me if as many Mac users as possible could try running Glaurung 2.0.1, let me know whether it works or not, and give me your system specification (the CPU and the OS X version would be enough). The discussion probably wouldn't be of interest to the public, so it is probably better to drop me a personal message or an e-mail than to post replies in this thread.

Tord

[Guetti]

Hello all,

The Mac OS X universal binary of Glaurung included in the Glaurung 2.0.1 download works fine on both my Macs, a Core 2 Duo iMac and a Core Duo MacBook, both running OS X 10.5.1. However, one user with a Core Duo iMac running OS X 10.4.11 reported that the binary does not work for him. He gets the following error message:

Code: Select all

dyld: lazy symbol binding failed: Symbol not found: _pthread_cond_init$UNIX2003
 Referenced from: /Users/louis/Desktop/glaurung2/MacOSX/./glaurung
 Expected in: /usr/lib/libSystem.B.dylib

I have no idea what is the cause of this error message. It might be related to the fact that he is running 10.4 while I am running 10.5, but I am not sure. It would therefore be very useful for me if as many Mac users as possible could try running Glaurung 2.0.1, let me know whether it works or not, and give me your system specification (the CPU and the OS X version would be enough). The discussion probably wouldn't be of interest to the public, so it is probably better to drop me a personal message or an e-mail than to post replies in this thread.

Tord

Lazy symbol binding? Wasn't this the same problem with the fruit executables? Do you remember, the ones you should have sent to Brian.
Did you try to manually strip the exe instead?

Ok, the provided binary for OSX does not work on my system, but I don't even know if it is an Universal binary. UCI initialization works, but when I force it to calculate it crashes immediately.
Anyway, I didn't even try this at first as I always compile glaurung myself form the included sources and the selfcompiled version works.

Mac OS X 10.5.1
4 x 2.5 Ghz PowerPC G5
gcc version 4.0.1 (Apple Inc. build 5465)

[Tord Romstad]

Hello Andreas,

Thanks for your input!

Lazy symbol binding? Wasn't this the same problem with the fruit executables? Do you remember, the ones you should have sent to Brian.
Did you try to manually strip the exe instead?

The binaries are either manually stripped, or not stripped at all. I don't remember. I use the exact same compiler and Makefile as for Glaurung Lodz 2007, which appeared to be running fine on all Macs (or at least on all Intel Macs).

Ok, the provided binary for OSX does not work on my system, but I don't even know if it is an Universal binary. UCI initialization works, but when I force it to calculate it crashes immediately.

That's bad news. I suppose I should simply not try to include support for 64-bit PowerPC CPUs in my universal binaries, but just use a 32-bit PowerPC binary (I can test the 32-bit PowerPC binary myself, by using Rosetta). There are probably not a lot of users with G5 Macs left anyway.

Anyway, I didn't even try this at first as I always compile glaurung myself form the included sources and the selfcompiled version works.

That's an option for PowerPC users, but unfortunately not for Intel Mac users, unless they accept a huge performance penalty. Most users would have to use the Apple's ancient GCC version (why are they still using 4.0.1, even in Leopard?), which means that Glaurung would run at less than half the speed of my own ICC compiled binary (1.33 MN/s vs 3.1 MN/s from the opening position on my 2.8 GHz Core Duo).

Tord

[Eelco de Groot]

Anyway, I didn't even try this at first as I always compile glaurung myself form the included sources and the selfcompiled version works.

That's an option for PowerPC users, but unfortunately not for Intel Mac users, unless they accept a huge performance penalty. Most users would have to use the Apple's ancient GCC version (why are they still using 4.0.1, even in Leopard?), which means that Glaurung would run at less than half the speed of my own ICC compiled binary (1.33 MN/s vs 3.1 MN/s from the opening position on my 2.8 GHz Core Duo).

Tord

Hi Tord,

Thanks for Glaurung 2.01! I was wondering, 3.1 MN/s is very fast, is that a 64 bit compiled Glaurung with two threads on a Core Duo, not a Core 2 Duo? Is the Core 2 Duo Mac, I seemed to remember you had one, even faster? On my Athlon with 2009 MHz I got about 560000 nodes per second with Jim's first compile, does that mean that the Core Duo (MacOS) is now a better chip for Glaurung than before compared with the Athlons (Windows)? Or do you think it is primarily a 64 bit performance benefit for Glaurung 2.01? Just curious!

Regards, Eelco

[Tord Romstad]

Hi Eelco!

That's an option for PowerPC users, but unfortunately not for Intel Mac users, unless they accept a huge performance penalty. Most users would have to use the Apple's ancient GCC version (why are they still using 4.0.1, even in Leopard?), which means that Glaurung would run at less than half the speed of my own ICC compiled binary (1.33 MN/s vs 3.1 MN/s from the opening position on my 2.8 GHz Core Duo).

Thanks for Glaurung 2.01! I was wondering, 3.1 MN/s is very fast, is that a 64 bit compiled Glaurung with two threads on a Core Duo, not a Core 2 Duo?

My mistake -- of course it is a Core 2 Duo (am I the only one who thinks these names look quite silly?), not a Core Duo. The 3.1 MN/s is for 64-bit Glaurung 2.0.1 compiled with Intel C++ 10.0, using two threads on the 2.8 GHz Core 2 Duo.

As a comparison, on my 2 GHz Core Duo MacBook, Glaurung 2.0.1 searches about 1.4 MN/s from the opening position.

Is the Core 2 Duo Mac, I seemed to remember you had one, even faster? On my Athlon with 2009 MHz I got about 560000 nodes per second with Jim's first compile, does that mean that the Core Duo (MacOS) is now a better chip for Glaurung than before compared with the Athlons (Windows)? Or do you think it is primarily a 64 bit performance benefit for Glaurung 2.01? Just curious!

I haven't tested my program on an Athlon any time recently, but my guess is that the modern Intel CPUs are probably faster for my program (and for most other chess programs).

Tord

[Guetti]

Hello Andreas,

... snip

That's an option for PowerPC users, but unfortunately not for Intel Mac users, unless they accept a huge performance penalty. Most users would have to use the Apple's ancient GCC version (why are they still using 4.0.1, even in Leopard?), which means that Glaurung would run at less than half the speed of my own ICC compiled binary (1.33 MN/s vs 3.1 MN/s from the opening position on my 2.8 GHz Core Duo).

Tord

3.1MN/s is a lot more than I get with 4 PowerPCs. :drool:
1.3MN/s on the contrary is pretty bad. I get 1.2MN/s with 2 cores (2 x 2.5 Ghz PowerPC) and around 2.1 MN/s with 4 Cores (so the speedup has definitely improved). Of course the Core 2 Duo is highly superior for chess, so gcc does a pretty bad job in your case. I hope Apple will help to improve the efficiency of gcc on Intel like they did for PPC.

Bye the way, any news about the Glaurung Gui? The old one still has a problem to add engines and I would like to see a analysis window that shows a sort of history over time of the score, pv etc of the previous plys.
Something that is copy&pastable.