problem building with openmpi

[Jim Edwards]

I haven't tried shared libraries yet, but I spoke to a system support
person for the system in question and he thinks that the -lpbs is due to
the way openmpi was built and
will work on a solution.

On Mon, Nov 6, 2017 at 5:35 PM, Wei-keng Liao <wkliao at eecs.northwestern.edu>
wrote:

> Hi, Jim
>
> Compiling Fortran programs is painfully slow today :(
>
> I have not got a chance to test OpenMPI 3.0.
> I tested against OpenMPI 2.1.2 and it was fine.
> 2.1.2 also put mpi.mod in lib folder instead of include.
> PnetCDF configure.ac does not look for mpi.mod explicitly.
> It just tried compile a short program that uses mpi.mod.
> Please send me the file config.log, if you would like me to take a look.
>
> I have no idea why -lpbs appears there. I assume libtool finds it from
> linker dependency list. Does this only happen when building static-only
> library? Did you try build shared libraries and see the same problem?
>
> Wei-keng
>
> On Nov 6, 2017, at 11:57 AM, Jim Edwards wrote:
>
> > It turns out that openmpi puts it's mpi.mod in the lib instead of
> include directory and the pnetcdf configure wasn't finding it.
> > Once I fixed this and use mpi.mod instead of mpif.h that solved the
> problem, however then I run into another problem at the
> > link step:
> >
> >  CCLD     ncmpivalid
> >
> > ld: cannot find -lpbs
> >
> > I haven't figured out where the requirement on a library pbs comes from
> or how to get rid of it?
> >
> >
> > On Mon, Nov 6, 2017 at 9:36 AM, Jim Edwards <jedwards at ucar.edu> wrote:
> >
> > On Mon, Nov 6, 2017 at 9:16 AM, Jim Edwards <jedwards at ucar.edu> wrote:
> > I am trying to build with openmpi/3.0.0 and intel17 and hanging in the
> build
> > of pnetcdf.f90.
> >
> > The make output at the time is:
> > make[4]: Entering directory '/glade/u/home/jedwards/src/
> parallel-netcdf/src/binding/f90'
> >
> >   FC       pnetcdf.lo
> >
> > examining the process I see:
> >
> >
> > /glade/u/apps/opt/intel/2017u1/compilers_and_
> libraries/linux/bin/intel64/ifort -I. -I. -debug minimal -g -O0 -c
> pnetcdf.f90 -o pnetcdf.o -I/glade/u/apps/ch/os/usr/include
> -L/glade/u/apps/opt/intel/2017u1/compilers_and_
> libraries/linux/lib/intel64_lin -L/glade/u/apps/ch/os/usr/lib64
> -L/glade/u/apps/ch/os/lib64 -L/glade/u/apps/ch/os/usr/lib
> -L/glade/u/apps/ch/os/lib -L/glade/u/apps/ch/opt/openmpi/3.0.0/intel/17.0.1/lib
> -Wl,--disable-new-dtags -Wl,-rpath,/glade/u/apps/opt/
> intel/2017u1/compilers_and_libraries/linux/lib/intel64_lin
> -Wl,-rpath,/glade/u/apps/ch/os/usr/lib64 -Wl,-rpath,/glade/u/apps/ch/os/lib64
> -Wl,-rpath,/glade/u/apps/ch/os/usr/lib -Wl,-rpath,/glade/u/apps/ch/os/lib
> -Wl,-rpath,/glade/u/apps/ch/opt/openmpi/3.0.0/intel/17.0.1/lib -lrt -ldl
> >
> >
> >
> > ​Please ignore the following incorrect statement, I did this in another
> window where I still had the default mpt loaded instead of openmpi, if I do
> this with openmpi, it still hangs. ​
> >
> >
> >
> >
> > If I remove all of the -L and -rpath flags and just do:
> >
> >  /glade/u/apps/opt/intel/2017u1/compilers_and_
> libraries/linux/bin/intel64/ifort -I. -I. -debug minimal -g -O0 -c
> pnetcdf.f90 -o pnetcdf.o
> >
> > That completes right away, so the problem seems to have to do with
> libtool.   Also building with mpt instead of openmpi on the same system
> works.
> >
> > Any ideas?
> >
> > --
> > Jim Edwards
> >
> > CESM Software Engineer
> > National Center for Atmospheric Research
> > Boulder, CO
> >
> >
> >
> > --
> > Jim Edwards
> >
> > CESM Software Engineer
> > National Center for Atmospheric Research
> > Boulder, CO
> >
> >
> >
> > --
> > Jim Edwards
> >
> > CESM Software Engineer
> > National Center for Atmospheric Research
> > Boulder, CO
>
>


-- 
Jim Edwards

CESM Software Engineer
National Center for Atmospheric Research
Boulder, CO
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/parallel-netcdf/attachments/20171106/ed3f4ad7/attachment.html>