Re: [netcdfgroup] netcdf-mpi build

• To: Jim Edwards <jedwards@xxxxxxxx>
• Subject: Re: [netcdfgroup] netcdf-mpi build
• From: Constantine Khroulev <c.khroulev@xxxxxxxxx>
• Date: Wed, 15 Jan 2014 14:24:17 -0900

Jim,

In the PISM build system (which also uses CMake) I check for the
presence of netcdf_par.h:

  # Try to find netcdf_par.h. We assume that NetCDF was compiled with
  # parallel I/O if this header is present.
  find_file(NETCDF_PAR_H netcdf_par.h HINTS ${NETCDF_INCLUDES} NO_DEFAULT_PATH)

  if (NOT NETCDF_PAR_H)
    set (Pism_USE_PARALLEL_NETCDF4 OFF CACHE BOOL "Enables parallel
NetCDF-4 I/O." FORCE)
    message(STATUS "Selected NetCDF library does not support parallel I/O.")
  endif()

I don't know how reliable this is and would love to hear from NetCDF
developers, but it works so far. (Should work in a cross-compiling
setup, too, as far as I can tell.)

-- 
Constantine

On Wed, Jan 15, 2014 at 2:13 PM, Jim Edwards <jedwards@xxxxxxxx> wrote:
> Hi,
>
> I'm trying to figure out with cmake whether netcdf was built with hdf5 mpi
> parallel IO support, I can compile a test program but I cannot expect it to
> run as many of the systems I run on use cross compiling.
>
> Variables like NC_MPIIO and NC_COLLECTIVE  appear to be defined even if this
> mode is not available as is the function nc_var_par_access.
>
> On some systems I can try linking without mpi and look for the link errors,
> but the compiler wrappers on other systems (eg cray) don't make it easy to
> compile without mpi.
>
> Any suggestions?
>
>
> --
> Jim Edwards
>
> CESM Software Engineering Group
> National Center for Atmospheric Research
> Boulder, CO
> 303-497-1842
>
> _______________________________________________
> netcdfgroup mailing list
> netcdfgroup@xxxxxxxxxxxxxxxx
> For list information or to unsubscribe,  visit:
> http://www.unidata.ucar.edu/mailing_lists/

• Follow-Ups:
  • Re: [netcdfgroup] netcdf-mpi build
    • From: Jim Edwards

• References:
  • [netcdfgroup] netcdf-mpi build
    • From: Jim Edwards