4 MPI Primer

MPI is a series of subroutines which are used to facilitate parallel processing. These routines allow data transfer between processing nodes and also support a variety of mathematical functions. They can be invoked using the C or FORTRAN languages; however, since the programs are written in FORTRAN90, only the FORTRAN bindings will be discussed. This section will focus on the bare essentials of MPI needed to understand the programs.

All MPI programs have a few things in common. They need to use a MPI include file, they must be initialized, finalized and the communicating processors must belong to the same communicating group. Here is a list of the generic MPI calls the user will encounter as they read the code. Any variables used in these calls will be defined in turn. Italicized font will be used when there are multiple options available.

• MPI_INIT(ierr)
• MPI_FINALIZE(ierr)
• MPI_COMM SIZE( MPI_Comm, nprocs, ierr )
•  
• MPI_COMM RANK( MPI_Comm, myrank, ierr )
• MPI_SEND(X,count,MPI_Data_Type,dest,itag,MPI_Comm,ierr)
•  
• MPI_RECV(Y,count,MPI_Data_Type,source,itag,MPI_Comm,status,ierr)
• MPI_IRECV(Y,count,MPI_Data_Type,source,itag,MPI_Comm,irequest,ierr)
• MPI_WAIT(irequest,status,ierr)
• MPI_BARRIER(MPI_Comm,ierr)
• MPI_ALLREDUCE(dA,A,count,MPI_Data_Type,MPI_Op,MPI_Comm,ierr)

• 4.1 Initialization
• 4.2 Sending and Receiving Data
• 4.3 Built-in Mathematical Functions
• 4.4 Summary