[Zaphod-Users] mpich/p4

[Jimmy Raeder]

Balasz,

this is not a Microway issue since our hardware or software is  
nothing special.

I'd suggest 2 things to solve your problem:  talk to the model  
developers.  They
usually know the pitfalls best because they have run the model on many
different machines or they may have heard similar stories from other  
users.

Second, you might want to peruse the MPICH release notes and mailing  
lists
carefully.  There might be a caveat that prevents codes from  
compiling.  If there own test
routines fail then there is probably something broken or incompatible  
in MPICH.

--  Jimmy



On Nov 16, 2005, at 10:00 PM, Fekete, Balazs M. wrote:

> Hi,
>
> As you may know, I am struggling to compile the GFDL model to use
> mpich/p4 (aka ethernet) as opposed to mpich/gm (aka Myrinet), so I can
> use the remaining ethernet nodes. The issue at hand are as follows:
> - the GFDL code needs to be compiled with ifort (Intel FORTRAN)
> - mpich/p4 only works with codes compiled with Portland Group  
> Compilers
>
> After spending my whole day compiling mpich, some test programs in  
> C and
> the GFDL model with different compiler options, I came to the  
> following
> conclusions.
>
> I can compile the mpich library with (gcc+ifort) or (pgcc+pgcf77/f90).
> The pgcc and ifort combination fails to link some of the  
> executables in
> the mpich distribution.
>
> Regardless of compiling mpich with gcc or pgcc, the test C program  
> works
> only if it is compiled and linked with pgcc (even when the mpich
> libraries were compiled with gcc, quite bizarre). I have to note that
> the test programs in the mpich library also fail using gcc only. The
> make in mpich has a testing mode, which compiles and links a whole  
> bunch
> of test programs and runs them as well. Some of the test programs give
> exactly the same error during mpi_send and mpi_recieve what I am  
> getting
> with my test programs.
>
> I tried to compile the GFDL code with ifort and link against the (gcc
> +ifort) compiled mpich library. The resulting code hangs in  
> mpi_send and
> mpi_recieve calls. I also tried to compile with ifort but link with
> pgf90, but it fails.
>
> I think, the solution lies in making sure that gcc can compile and  
> link
> mpich codes. I wonder, if someone from Microway can be contacted to
> resolve this problem (unless someone in the zaphod user group knows  
> the
> solution).
>
> Balazs
>
> P.S. I tried both the patched and unpatched mpich-1.2.7 library but it
> did not make any difference (as Kai expected since the ethernet  
> version
> should be fairly standard).
>
>
> -- 
> Balazs M. Fekete                E-mail: balazs.fekete at unh.edu
> Water Systems Analysis Group
> Complex Systems Research Center
> Institute for the Study of Earth, Oceans, and Space
> University of New Hampshire     Phone: (603) 8620270
> Morse Hall, Durham NH 03824     Fax: (603) 8620587
>
> _______________________________________________
> Zaphod-Users mailing list
> Zaphod-Users at lists.sr.unh.edu
> http://lists.sr.unh.edu/mailman/listinfo/zaphod-users

------------------------------------------------------------------------ 
---------------------
Joachim (Jimmy) Raeder
Associate Professor, Department of Physics & Space Science Center
University of New Hampshire,
245G Morse Hall, 39 College Rd, Durham, NH 03824-3525
voice: 603-862-3412  e-mail: J.Raeder at unh.edu
WWW: http://terra.sr.unh.edu/jraeder/
------------------------------------------------------------------------ 
---------------------


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.sr.unh.edu/pipermail/zaphod-users/attachments/20051117/3b06b9b0/attachment.html