[MITgcm-support] Make depend error

[Don Stark]

> On Tue, 2005-04-26 at 13:20 -0700, Don Stark wrote:
>> Ed,
>> I can get the cde to compile and run without MPI using opton (2), but when
>> I switch on the MP, I get a run time error - "STOP ABNORMAL END: PROGRAM
>> MAIN statement executed." I've captued what the code displays to the screen
>> below.
>> 
>>>  2) if (1) doesn't work, please try delting (or commenting out)
>>>     the line:
>>> 
>>>       MAKEDEPEND=makedepend
>>> 
>>>     from your optfile and see if that woks.
>>> 
>>> Ed
>> anacapa (48)% mpirun -np 2 ./mitgcmuv > dat
>> STOP ABNORMAL END: PROGRAM MAIN statement executedSTOP ABNORMAL END: PROGRAM
>> MAIN statement executed
> 
> Hi Don,
> 
> I'm glad you've got MITgcm running in non-MPI-mode.  Thats a good first
> step.  And, if you would, please send us the optfile that you used since
> it would be nice to share!

#!/bin/bash
#
#  $Header: /u/gcmpack/MITgcm/tools/build_options/irix64_ip30_f77,v 1.1
2004/02/21 14:41:09 dimitri Exp $
#

S64='$(TOOLSDIR)/set64bitConst.sh'
%MAKEDEPEND=makedepend
CPP='/lib/cpp -P -traditional'

DEFINES='-DTARGET_SGI -DWORDLENGTH=4'
FC='f77'
FC_NAMEMANGLE='#define FC_NAMEMANGLE(X) X ## _'
FFLAGS='-n32 -extend_source -bytereclen'
FOPTIM='-O3 -OPT:Olimit=0:roundoff=3:div_split=ON:alias=typed'
LIBS='-lmpi'
INCLUDEDIRS='/usr/include'

The MPI version has the additional tags to the DEFINES of -DALLOW_USE_MPI
and -DALWAYS_USE_MPI
> 
> So, whats happening with your MPI runs?  I really don't know and its
> very hard to understand (or even guess) what the problems are without
> access to your machine.  So, since I cannot access your machine, please
> try the following:
> 
>  1) Can you create a small "hello world"-type MPI programs that
>     compile and run?  This is a very important first step!
>     Most MPI problems have nothing to do with MITgcm per se.
>     They are usually problems with the setup or use of MPI.  And
>     most MPI implementations (eg. LAM and mpich) come with simple
>     example programs that you can use to test your setup.
Ok,
I the MPI optfile is based on the non-MPI file, so I'll look at the include
and libs lines. I have another large gcm that I run on this machine. Looking
at the other code's makefile I see that

    CPP /usr/lib/cpp
    FC f90 -I/usr/local/include
    LDLIBS -L/usr/lib32 -lcomplib.sgimath -lfastm -lmpi

So I can see replacing these three things in the optfile. Thoughts?
> 
>  2) Once you have proven to yourself that MPI is properly setup
>     (by successfully building and running some simple codes),
>     then try MITgcm.  And start with your non-MPI optfile (which
>     you know to be working) and add the few MPI-specific changes
>     (which will usually be the MPI header and library information).
> 
> Its much easier to debug things if you can break them into more
> manageable parts!
> 
> Ed