PGI compilation problem when MPI enabled

[jmrosinski]

I was confused before about sometimes PGI failing and sometimes succeeding. I now see the issue is "with MPI" fails and "without MPI" succeeds. From theia supercomputer, see attached configout.txt and makeout.txt, where mpicc runs pgcc and the MPI library is openmpi. As I recall, I get similar (maybe identical) on my home machine. I do not understand what is being complained about in make.output. Note that PGI+openmpi compiles and runs to successful completion in the old (pre-autotools) stuff.

configout.txt
makeout.txt

[jmrosinski]

Not shown is the fact that "make clean" was done in between "configure" and "make".

[jmrosinski]

Just checked and behavior on home Linux machine is the same: CC=pgcc works fine, but CC=mpicc (runs pgcc) fails with the same complaint.

[jmrosinski]

Per request in issue #25 attached are config.log and output from "make" which is failing in a way I don't understand. NOTE: This strange failure occurs ONLY with when enabling MPI (mpicc) and ONLY when attempting a shared library build with PGI.
config.log
out.make.mpicc_pgi.txt

[edhartnett]

This worked pre-autotools because you were only ever building static libraries. Now you are attempting shared libraries.

On the PGI compiler you are using, are shared libraries supported? Lots of HPC places and software automatically assume static-only. (Not a good idea, IMO, but common.)

If the mpicc is not built for shared, then only static libraries will work with that compiler.