[mpich-discuss] undefined symbol: MPIR_Thread

[Gus Correa]

Matthias Demuzere wrote:
> Anthony, Gus,
> 
> I think the problem is solved due to the suggestion of Anthony. Although 
> I rebuilt hdf5 and netcdf4 already a million of times, I realised I 
> probably didn't do that after the discovery of mpich still being present 
> from the ubuntu repository.
> 
> Therefore, I rebuild it again as follows (following guidelines on the 
> unidate webpage: 
> http://www.unidata.ucar.edu/software/netcdf/docs/netcdf-install/parallel.html#parallel) 
> 
> 
> For HDF5:  CC=/usr/local/mpich2-1.3.1/bin/mpicc ./configure 
> --enable-parallel --prefix=/usr/local/hdf5 -- with-zlib-zusr/local/zlib 
> --disable-shared && sudo make check install
> 
> For NETCDF:   CPPFLAGS="-DgFortran" 
> FC=/usr/local/mpich2-1.3.1/bin/mpif90 
> CXX=/usr/local/mpich2-1.3.1/bin/mpicxx 
> CC=/usr/local/mpich2-1.3.1/bin/mpicc ./configure 
> --prefix=/usr/local/netcdf-4.1.1 --with-hdf5=/usr/local/hdf5
> sudo make check install
> 
> I compilled cesm again, and run the ccsm.exe file.
> Although I do get some sort of error message, I think the model ran 
> properly.
> 
> So I guess the problem was lying in the fact that hdf5 still had this 
> hidden unbuntu mpi dependence.
> 
> I would like to talk you for your efforts and patience, I hope I can now 
> fully start to use the model!!
> 

Hi Matthias

Glad to know that you've got CESM to work.
I also had trouble to build CESM with NetCDF4, and HDF5.
Eventually I switched to NetCDF3, and things worked with less hassle.
There is not much that you gain with NetCDF4, unless you are
going to do parallel IO and/or data compression
(not sure if CESM uses the latter).
In any case, you need to make sure that these features work also.
Have you tried parallel IO on CESM?

Gus Correa