[Beowulf] mpich on OS X

[Glen Beane]

I got this fix from Bill Gropp at ANL:

In the OS X environment, some compilers and loaders make global
variables that are not initialized at compile time into a special kind
of symbol that the loader ignores when it is defined in a library.  This
without doubt violates the spirit and probably the letter of the C
standard, but it is the way OS X works.  There is a work around which is
to execute 

ranlib -c libraryname

e.g.,

cd /Users/admin/mpich-1.2.6/lib && ranlib -c libmpich.a 
(and all other libraries in that directory).


On Wed, 2004-08-25 at 16:08, Glen Beane wrote:
> I'll look through my logs more closely, but at quick glance it doesn't 
> look like any problems occur until it tries to link.  I have no problems 
> using 1.2.5, but this is a different machine (in fact I don't have 
> access to the other OS X machine I previously compiled 1.2.5 on), so I 
> figured I'd start with the latest source, since I have to compile anyway 
> (plus 1.2.6 was supposed to have better OS X build support).
> 
> I probably will try MPICH2  as well.
> 
> Joachim Worringen wrote:
> 
> > Glen Beane wrote:
> >
> >> I've compiled mpich-1.2.5 on a different OS X 10.3 system with xlf and
> >> gcc without this problem. I tried using xlf and gcc now, with no change
> >> in the behavior. I'm using very generic ./configure arguments
> >> (--with-device=ch_p4, -prefix=~/mpich, -opt=O2, --disable-c++), and i'm
> >> not passing any extra flags to the compilers or linker (which I'm
> >> assuming is my problem).  Anyone know the linker/compiler flags I need
> >> to get this to link properly? 
> >
> >
> > - Take a look on the output of the compilers (make >&make.log; less 
> > make.log). Probably there's a problem with some files in 
> > mpid/ch_p4(/p4) to compile correctly. Knowing the actual problem makes 
> > it easier to fix.
> >
> > - Why do you need mpich-1.2.6 - it's mostly a bugfix release of 1.2.5. 
> > Do you see any bugs with 1.2.5?
> >
> > - What about MPICH2? Performs significantly better with socket 
> > communication.
> >
> >  Joachim
> >
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf