mpich mpif08

[hkershaw-brown]

I think the mpi_utilities errors are mpich using f77 bindings when you use mpi. Hello world mpi (no dart code) will not compile if you have a mpi_send(integer) mpi_send(real) in the same file.

hello.f90.txt
hello_f08.f90.txt

The f08 mpi module use mpi_f08 follows the fortran standard.
There is a branch mpi_f08 which has a version of mpi_utilites_mod.f90 with mpi_f08 bindings. This will work with mpich and gfortran 10+ (only done a quick test).
The cray window module (cray pointers) could get replaced with iso_c_bindings.

or short term solution: port select --set mpi openmpi-mp-fortran

Originally posted by @hkershaw-brown in #158 (comment)

[nancycollins]

i think that changing our code to 'use mpi_f08' instead of just 'use mpi' is the right solution (also the iso_c_bindings). but i don't how to verify how many of our users have access to mpi with the 2008 bindings compared to older versions only. if this breaks our build for a lot of people then it's bad. if most installations now have 2008 bindings then this allows us to let the compilers do better error checking by removing compile-time flags from our mkmf.templates which is a good thing.

it would be nice if there was an easy way to know whether an mpi had a 2008 module or not. a simple kind of 'configure' command that runs first and checks things out? or just make the change and tell users to edit mpi_utilities_mod.f90 if it doesn't build?

just thinking out loud here. not sure what the right solution is yet.

[hkershaw-brown]

I love these foundation shaking questions from Nancy. Have you played with autoconf to configure/make DART vs. mkmf?

[hkershaw-brown]

other notes from #288
review MPI_INTEGER_KIND, COUNT MPI3 as part of this update.
and fixsystem.

Also see #245 namelist standard

mpi_f08 https://github.com/NCAR/DART/tree/mpi_f08

[hkershaw-brown]

note the mpi_f08 branch is a ways behind main so does not have the NAG bug-fix #328.

[hkershaw-brown]

up-to-date now. I think this should go in a f2008(3) update (see #288). Hanging out on the mpi_f08 branch now

[hkershaw-brown]

moved mpi_f08 https://github.com/hkershaw-brown/DART/tree/mpi_f08
someone can grab it or not as needed if they are updating to f08.

[hkershaw-brown]

Gust gfortran

hkershaw@gust01:~/DART/models/lorenz_96/work.gfortran> module -t list
ncarenv/23.04
craype/2.7.20
gcc/12.2.0
ncarcompilers/0.8.0
hdf5/1.12.2
netcdf/4.9.1
cray-mpich/8.1.25

[hkershaw-brown]

To reproduce this on gust you can use the following modules:

hkershaw@gust01:~/DART/models/lorenz_96/work> module load gcc
hkershaw@gust01:~/DART/models/lorenz_96/work> module -t list
ncarenv/23.04
craype/2.7.20
gcc/12.2.0
ncarcompilers/0.8.0
hdf5/1.12.2
netcdf/4.9.1
cray-mpich/8.1.25

./quickbuild.sh gets you:

Error: Type mismatch between actual argument at (1) and actual argument at (2) (REAL(8)/INTEGER(4)).
/glade/u/home/hkershaw/DART/assimilation_code/modules/utilities/mpi_utilities_mod.f90:649:18:

  595 |       call MPI_Ssend(srcarray, int(itemcount,i4), datasize, dest_id, tag, &
      |                     2
......
  649 |    call MPI_Ssend(itime, 2, MPI_INTEGER, dest_id, tag*2, my_local_comm, errcode)
      |                  1
Error: Type mismatch between actual argument at (1) and actual argument at (2) (INTEGER(4)/REAL(8)).

[mjs2369]

Updated the old mpi_f08 branch to main. Updated branch located at https://github.com/NCAR/DART/tree/mpif08