In order to compile an MPI application, the mpicc compiler wrapper is used. This executable should be part of whichever MPI implementation has been installed.

Using it is, however, identical to how one would use, for example, GCC:

    $ mpicc -o mpi_hello_world mpi_hello_world.c

This can be compared to:

    $ gcc mpi_hello_world.c -lmsmpi -o mpi_hello_world

This would compile and link our Hello World example into a binary, ready to be executed. Executing this binary is, however, not done by starting it directly, but instead a launcher is used, like this:

    $ mpiexec.exe -n 4 mpi_hello_world.exe
    Hello world from processor Generic_PC, rank 0 out of 4 processors
    Hello world from processor Generic_PC, rank 2 out of 4 processors
    Hello world from processor Generic_PC, rank 1 out of 4 processors
    Hello world from processor Generic_PC, rank 3 out of 4 processors
  

The preceding output is from Open MPI running inside a Bash shell on a Windows system. As we can see, we launch four processes in total (4 ranks). The processor name is reported as the hostname for each process ("PC").

The binary to launch MPI applications with is called mpiexec or mpirun, or orterun. These are synonyms for the same binary, though not all implementations will have all synonyms. For Open MPI, all three are present and one can use any of these.