[Mitgcm-support] RE: pgi+mpi linker problems

[mitgcm-support at dev.mitgcm.org]

Chris, 

Great!  -Msecond_underscore makes all the link-time problems disappear
except for one:

/usr/mpi/mpich-gm-1.2..5/lib//libmpich.a(farg.o): In function
`mpir_iargc__': farg.o(.text+0x4): undefined reference to `f__xargc'

f__xargc is apparently in; mpichf, mpichfarg, pmpich, and mpich libs but
explicitly including them doesn't resolve the problem.

Any suggestions?

Thanks, Tom.

On Fri, 7 Jun 2002, Chris Hill wrote:

> Tom,
> 
>  I think the tools/as hack is not needed with the up to date
> RH and PGI. 
> I think the problem you have is becaue you built mpi
> with g77 as your Fortran compiler and now you are using
> pgf77. This means that your mpi_recv routine that gets
> linked from fortran is called mpi_recv__ not mpi_recv_.
> Dumb and annoying I know!
> The pgi option that may fix this is
> -second_underscore. 
> It could be something totally different though!
> 
> Chris
> 
> -----Original Message-----
> From: Tom Haine [mailto:twnh at stommel.eps.jhu.edu]
> Sent: Friday, June 07, 2002 5:22 PM
> To: support at mitgcm.org
> Cc: Thomas Haine
> Subject: pgi+mpi linker problems
> 
> 
> Hi,
> 
> I'm having problems linking with pgif77 and mpi on RH linux. I get:
> 
> undefined reference to `mpi_recv_'
> 
> for example. pgif77 without mpi works OK and compiling with g77+mpi also
> works (so libs are defined OK).
> 
> I know I should send the compiler output through the tools/as wrapper for
> the gnu assembler (with FFLAGS = .... -Ya,../tools in Makefile). But this
> doesn't help. The sed command in tools/as
> 
> sed 's:/:#:' /tmp/pgf772gQPv1 | exec /usr/bin/as -o timestep_tracer.o --
> 
> (for example) translates "/" characters to "#" characters in the temporary
> assembler file (right?). But why is this helpful?  The pgif77 comments,
> apparently the cause of the trouble, are like:
> 
> # PGFTN 3.2 -opt 2 -norecursive
> # lineno: 353
> 
> so I don't see how tools/as should fix the problem. Actually, I don't see
> the symptoms of this should be undefined references to mpi fns.
> 
> Please can you give me some advice about how to tackle this problem?  
> Perhaps my version of pgif77, sed, or as gives different results to yours?
> 
> Thanks, Tom.
> P.S. Great model and documentation!  Can't wait to get going properly!
> --------------------------------------------
>  Thomas W. N. Haine
> 
>  Associate Professor of Physical Oceanography,
>  Department of Earth & Planetary Sciences,
>  329 Olin Hall, The Johns Hopkins University,
>  Baltimore, MD 21218, USA.
>  Tel : 410 516 7048,   Fax : 410 516 7933
>  Thomas.Haine at jhu.edu  
>  http://www.jhu.edu/~eps/faculty/haine
> --------------------------------------------
> 
> 
> On Mon, 16 Jul 2001, Chris Hill wrote:
> 
> > For pgf77 linker problem, we had to do this  ( see tools/as in any
> > mitgcm from 
> > checkpoint30 onwards ).
> > 
> > !/bin/csh -f
> > 
> > # This is a real hack of a wrapper for the linux assembler.
> > # It replaces comment characters generated by pgf77 with
> > # recognized comment characters. Invoke with -Ya,../tools
> > # in the FFLAGS variable in Makefile
> > 
> > set asfile=$argv[$#argv]
> > set argv[$#argv]="--"
> > 
> > echo "sed 's:/:#:' $asfile | exec /usr/bin/as $argv"
> > 
> > sed 's:/:#:' $asfile | exec /usr/bin/as $argv
> > 
> > Chris
> 
>