Change the IBM Fortran and C wrappers on Summit

[minxu74]

Change the IBM Fortran and C wrappers from the Fortran 77 to
Fortran 90 functionality and change their MPI wrappers to the
common names (i.e. mpif90, mpicc, and mpicxx).

[sarats]

Regarding the generic mpixx wrappers, are they guaranteed to be set properly for the right compiler family? I recall there were issues last year. Anyway if that's recommended by OLCF now, we should switch.

[minxu74]

Hi @sarats Yes, the MPI wrappers are recommended in the user guides of Summit. Based on the same link, the xlf_r mainly includes F77 functionality and I changed it to xlf90_r. This change removed the errors reported in the issues #2687 and #2713

[sarats]

@minxu74 Thanks for the clarification. I'm okay with Fortran compiler change but thought it might be preferable to be explicit about the Fortran variant through a flag.
-qlanglvl=90std/2003std etc.

Btw, if one were to use the MPI wrappers, this is what they invoke. Both mpif90 and mpifort are same.
So we should use xlf2008_r to keep it consistent?

[sarat@login5 ~ ]$ mpif90 -showme
/sw/summit/xl/16.1.1-1/xlf/16.1.1/bin/xlf2008_r -I/autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/xl-16.1.1-1/spectrum-mpi-10.2.0.10-20181214-y75ruim4h6cvqcdymmcnb6mllodtyfuz/include -qthreaded -I/autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/xl-16.1.1-1/spectrum-mpi-10.2.0.10-20181214-y75ruim4h6cvqcdymmcnb6mllodtyfuz/lib/XL -L/autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/xl-16.1.1-1/spectrum-mpi-10.2.0.10-20181214-y75ruim4h6cvqcdymmcnb6mllodtyfuz/lib -lmpiprofilesupport -lmpi_ibm_usempi -lmpi_ibm_mpifh -lmpi_ibm
[sarat@login5 ~ ]$ mpifort -showme
/sw/summit/xl/16.1.1-1/xlf/16.1.1/bin/xlf2008_r -I/autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/xl-16.1.1-1/spectrum-mpi-10.2.0.10-20181214-y75ruim4h6cvqcdymmcnb6mllodtyfuz/include -qthreaded -I/autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/xl-16.1.1-1/spectrum-mpi-10.2.0.10-20181214-y75ruim4h6cvqcdymmcnb6mllodtyfuz/lib/XL -L/autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/xl-16.1.1-1/spectrum-mpi-10.2.0.10-20181214-y75ruim4h6cvqcdymmcnb6mllodtyfuz/lib -lmpiprofilesupport -lmpi_ibm_usempi -lmpi_ibm_mpifh -lmpi_ibm

[minxu74]

@sarats I do not know. The results of mpif77 --showme and mpif90 --showme and mpifort --showme are exactly same. But if you look at the wrappers by vim, they do choose the compilers based on the name of the wrapper.

[sarats]

Hmm...I'm confused as well. All of them ultimately link to script below.

I looked inside the script and it doesn't seem to distinguish between the Fortran variants in change_flags_xlf ()

/autofs/nccs-svm1_sw/summit/.swci/1-compute/opt/spack/20180914/linux-rhel7-ppc64le/xl-16.1.1-1/spectrum-mpi-10.2.0.10-20181214-y75ruim4h6cvqcdymmcnb6mllodtyfuz/bin/mpifort -> opal_wrapper

[minxu74]

@sarat in the mpif90 script:

    elif [ $1 = 'mpif90' ] ; then
        complist="gfortran:ifort:ifc:pgfortran:pgf90:pathf90"
        if [ $ppc -eq 1 ] ; then
            complist="xlf90_r:$complist"
        fi
        evar=OMPI_FC
    else

When it is mpif90, the xlf90_r is included in the compiler list. But it is still unclear why "--showme" gives us the same results. I will ask this question in next week Summit workshop.

[rljacob]

What is the status of this PR?

[minxu74]

@rljacob I will merge it today.