Parallel solver dummies for preCICE

This repository currently contains four solver dummies for the coupling library preCICE that can be executed in parallel. The solver dummies can also be mixed with each other, i.e., each solver can play any role in the coupling!

Current solver dummies included

• C solver dummy: c/solverdummy-c-parallel.cpp
• C++ solver dummy: cpp/solverdummy-cpp-parallel.cpp
• Fortran solver dummy: fortran/solverdummy-fortran-parallel.f90
• Julia solver dummy: julia/solverdummy-julia-parallel.jl
• Python solver dummy: python/solverdummy-python-parallel.py

Requirements

All dummies

• preCICE v2 (only tested with v2.2.0)

C dummy

• CMake
• A C compiler
• MPI (only tested with OpenMPI)

C++ dummy

• CMake
• A C++ compiler
• MPI (only tested with OpenMPI)

Fortran dummy

• CMake
• A Fortran compiler
• MPI (only tested with OpenMPI)

Julia dummy

• Julia v1.6 or above
• Julia bindings for preCICE

Python dummy additionally needs

• Python 3
• mpi4py
• Python bindings for preCICE

If you have preCICE and the Python bindings installed, you should have all requirements installed.

Compilation

C

Compile the C solver dummy and copy it back to the c directory:

cd c
mkdir build
cd build
cmake ..
make
cp solverdummy-c-parallel ..
cd ..

C++

Compile the C++ solver dummy and copy it back to the cpp directory:

cd cpp
mkdir build
cd build
cmake ..
make
cp solverdummy-cpp-parallel ..
cd ..

Fortran

Compile the Fortran solver dummy and copy it back to the fortran directory:

cd fortran
mkdir build
cd build
cmake ..
make
cp solverdummy-fortran-parallel ..
cd ..

Test solvers

N and M refer to the number of ranks one want to use. Start with 1 in order to run the serial computation and increase N and M afterwards to have truly parallel computations.

Further information:

• N and M may be different numbers.
• The solver have to run in different shells or you have to send the first solver to the background.
• You pass the participant's name and the corresponding mesh name on the terminal. The name and mesh have to be one of the following
  • SolverOne and MeshOne
  • SolverTwo and MeshTwo
• Replace mpirun by the suitable MPI wrapper of your machine.
• Step into the directory of the solver you want to use, i.e. cd cpp/ for the C++ solver.
• If you copy the solvers into other directories, please adapt the file precice-config-parallel.xml accordingly.

Note You can mix the different dummies arbitrarily with each other. Some examples are given below:

Setup 1: C++ and C++

mpirun -n N ./solverdummy-cpp-parallel ../precice-config-parallel.xml SolverOne
mpirun -n M ./solverdummy-cpp-parallel ../precice-config-parallel.xml SolverTwo

Example:

mpirun -n 4 ./solverdummy-cpp-parallel ../precice-config-parallel.xml SolverOne
mpirun -n 2 ./solverdummy-cpp-parallel ../precice-config-parallel.xml SolverTwo

Setup 2: Python and Python

mpirun -n N python3 solverdummy-python-parallel.py ../precice-config-parallel.xml SolverOne
mpirun -n M python3 solverdummy-python-parallel.py ../precice-config-parallel.xml SolverTwo

Example:

mpirun -n 4 python3 solverdummy-python-parallel.py ../precice-config-parallel.xml SolverOne
mpirun -n 2 python3 solverdummy-python-parallel.py ../precice-config-parallel.xml SolverTwo

Setup 3: C++ and Python

mpirun -n N ./solverdummy-cpp-parallel ../precice-config-parallel.xml SolverOne
mpirun -n M python3 solverdummy-python-parallel.py ../precice-config-parallel.xml SolverTwo

Example:

mpirun -n 4 ./solverdummy-cpp-parallel ../precice-config-parallel.xml SolverOne
mpirun -n 2 python3 solverdummy-python-parallel.py ../precice-config-parallel.xml SolverTwo

Setup 4: Fortran and C++

mpirun -n N ./solverdummy-cpp-parallel ../precice-config-parallel.xml SolverOne
mpirun -n M ./solverdummy-fortran-parallel ../precice-config-parallel.xml SolverTwo

Example:

mpirun -n 4 ./solverdummy-cpp-parallel ../precice-config-parallel.xml SolverOne
mpirun -n 2 ./solverdummy-fortran-parallel ../precice-config-parallel.xml SolverTwo

Setup 4: C++ and C

mpirun -n N ./solverdummy-cpp-parallel ../precice-config-parallel.xml SolverOne
mpirun -n N ./solverdummy-c-parallel ../precice-config-parallel.xml SolverOne

Example:

mpirun -n 4 ./solverdummy-cpp-parallel ../precice-config-parallel.xml SolverOne
mpirun -n 2 ./solverdummy-c-parallel ../precice-config-parallel.xml  SolverTwo

Setup 5: Julia and Julia

The Julia solver dummy uses Julia's native parallelization and not the MPI wrapper.

You need to add <master:sockets/> to the precice-config-parallel.xml file in both participants.

Example:

<participant name="SolverOne">
    <master:sockets/>
    <use-mesh name="MeshOne" provide="yes"/>
    <write-data name="dataOne" mesh="MeshOne" />
    <read-data  name="dataTwo" mesh="MeshOne" />
</participant>

For convenience, the repository provides a precice-config-parallel-julia.xml configuration that can be used with the Julia solver dummy.

Run the dummies with

julia solverdummy-julia-parallel.jl N ../precice-config-parallel-julia.xml SolverOne
julia solverdummy-julia-parallel.jl M ../precice-config-parallel-julia.xml SolverTwo

Example:

julia solverdummy-julia-parallel.jl 4 ../precice-config-parallel-julia.xml SolverOne
julia solverdummy-julia-parallel.jl 2 ../precice-config-parallel-julia.xml SolverTwo

Setup 6: Julia and C++

As with Setup 5 you need to add <master:sockets/> to the precice-config-parallel.xml file in both participants or you can used the provided precice-config-parallel-julia.xml configuration.

mpirun -n N ./solverdummy-cpp-parallel ../precice-config-parallel-julia.xml SolverOne MeshOne
julia solverdummy-julia-parallel.jl M ../precice-config-parallel-julia.xml SolverTwo

Example:

mpirun -n 4 ./solverdummy-cpp-parallel ../precice-config-parallel-julia.xml SolverOne MeshOne
julia solverdummy-julia-parallel.jl 2 ../precice-config-parallel-julia.xml SolverTwo

Remarks

• The codes are heavily based on the solver dummies of preCICE and the corresponding Python bindings.