v1.1.1... again

Rob Latham robl at mcs.anl.gov
Wed Apr 21 09:19:56 CDT 2010


On Wed, Apr 21, 2010 at 07:32:36AM -0500, Gerry Creager wrote:
> I'm trying to compile v 1.1.1 on a CentOS cluster, with a newly
> added gluster file system. We're running OpenMPI 1.4.1 and PGI 10.3.
> Here's what I've done, so far:
> 
> module load pgi/10
> module load openmpi/1.4.1/pgi/10.3/64
> export cc=mpicc
> export CC=mpiCC

it looks like you are using a C++ compiler to build C tests.  I'm a
little surprised that does not work, but we also never test things
that way.

> export FC=mpif90
> export F77=mpif90
> export F90=mpif90
> export CFLAGS="-O2"
> export FFLAGS="-O2 -Msignextend"
> export CPPFLAGS="-DpgiFortran"

While you need -DpgiFortran for serial netcdf, you don't need that for
parallel-netcdf

> ./configure --prefix=/apps/pNetCDF/pgi-10-64 \
>        --with-mpi=/apps/openmpi-1.4.1-pgi/ --enable-fortran

Since you specified --with-mpi, you do not need to specify all the
environment variables. No harm in doing both, but as you can see it
allows you to specify weird combinations.

> The fatal error is:
> checking if Fortran "byte" is C "signed char"... configure: error:
> Could not compile conftest.c

...
> configure:11493: checking if Fortran "byte" is C "signed char"
> configure:11503: mpiCC -c -DpgiFortran -O2 conftest.c
> "conftest.c", line 1: error: incomplete type is not allowed
>   	    void sub_(values)
>   	         ^
> 
> "conftest.c", line 1: error: identifier "values" is undefined
>   	    void sub_(values)
>   	              ^
> 
> "conftest.c", line 2: error: expected a ";"
>   		signed char values[4];
>   		^
> 
> "conftest.c", line 3: error: expected a declaration
>   	    {
>   	    ^
> 
> 4 errors detected in the compilation of "conftest.c".

Let me know if the suggestions above make these errors go away.  

==rob

-- 
Rob Latham
Mathematics and Computer Science Division
Argonne National Lab, IL USA


More information about the parallel-netcdf mailing list