pnetcdf with WRF

[Lev Lafayette]

----- "Rob Latham" <robl at mcs.anl.gov> wrote:

> On Mon, Nov 22, 2010 at 04:45:00PM +1100, Lev Lafayette wrote:
> > Hi everyone,
> > 
> > I am attempting to install WRF (Weather Research and Forecasting) on
> a cluster using, as required and expected, pnetcdf.
> > 
> > No matter which openmpi and compilers I use (pgi, intel and gcc) I
> am receiving an error which is preventing a successful compilation of
> WRF 
> > 
> > i.e.,
> > 
> > strerrorf.f:(.text+0x2c): undefined reference to `nfmpi_xstrerror_'
> > /usr/local/pnetcdf/1.2.0-pgi/lib/libpnetcdf.a(inq_libversf.o): In
> function `nfmpi_inq_libvers_':
> > inq_libversf.f:(.text+0x21): undefined reference to
> `nfmpi_xinq_libvers_'
> 


Hi Rob,

> 
> In addition to what Wei-keng suggested, I'd like to see what the
> strerrorf.o and xstrerrorf.o files contain -- it will help us figure
> out the fortran name mangling convention detected.
> 

Well, these are binary files from the source

$parallel-netcdf-1.2.0/src/libf/strerrorf.o
$parallel-netcdf-1.2.0/src/libf/xstrerrorf.o

> There are some key lines printed out amid the mass of configure
> output.  On my system, which detected the fortran compiler and built
> all the symbols correctly, I've got lines like this:
> 
> ...
> checking for pgf90... no
> checking for f77... no
> checking for fort77... no
> checking for g77... no
> checking for gfortran... gfortran
> checking for Fortran .F compiler... 
> checking if Fortran-77 compiler handles *.F files... yes
> checking how to make dependencies... false
> configure: checking for nm utility...
> checking for nm... nm
> checking nm flags... 
> checking for C-equivalent to Fortran routine "SUB"... sub_

These are generating results that I would expect from pgi (10.9)
and openmpi (1.4.3)

checking for pgf90... pgf90
checking for Fortran .F compiler... 
checking if Fortran-77 compiler handles *.F files... yes
checking how to make dependencies... false
configure: checking for nm utility...
checking for nm... nm
checking nm flags... 
checking for C-equivalent to Fortran routine "SUB"... sub_


> What might be happening is that some parts of pnetcdf are built with
> the system compiler and some parts are built with the MPI compiler.
> That approach allows for a lot of flexibility, but maybe it's causing
> you some grief here.  If you configure pnetcdf by explicitly
> selecting
> the C and Fortran compilers, do you still get this problem?
> 
> CC=mpicc FC=mpif90 F77=mpif90 F90=mpif90 MPICC=mpicc MPIF90=mpif90

OK, trying that and running make testing I end up with a similar set 
of errors as with the pgi install :(

-- 
Lev Lafayette, Systems Administrator, Victorian Partnership for Advanced Computing, lev at vpac.org
"If you are sending a reply to a message or a posting be sure you summarize the original at the top of the message, or include just enough text of the original to give a context. This will make sure readers understand when they start to read your response."  RFC 1855
110 Victoria Street                                        phone:   03 9925 4909
Carlton South, Victoria 3053,  Australia          mobile:  0432 255 208