Accessing MPI and MPE libraries

To enable use of the parallel processing code included with GRASP, one needs to link the code with an MPI function call library. (If you do not intend to use any of the multiprocessing code, we'll tell you what to do.) For performance monitoring purposes, we also make calls to the Message Passing Environment (MPE) library, which is included with mpich [3]. If these function libraries are not currently available on your system, you should obtain the public domain implementation mpich from the URL
http://www.mcs.anl.gov/mpi/mpich/ http://www.mcs.anl.gov/mpi/mpich/ and follow the instructions required to build the MPI/MPE libraries for your system. After the installation process is complete, the necessary libraries will be contained in a library archive, for example /usr/local/mpi/lib/libmpi.a and /usr/local/mpe/lib/libmpe.a. The path to these libraries is set in the file SiteSpecific by means of the variable MPI_LIBS. A typical line in SiteSpecific might then read:
MPI_LIBS=-L/usr/local/mpi/lib -lmpi -lmpe.
You must also set BUILD_MPI= true in SiteSpecific. Finally, in order to include appropriate header files in any MPI programs, you will need to include a path to these header files in the file SiteSpecific. You can do this by setting MPI_INCLUDES in the file SiteSpecific. A typical installation might have
MPI_INCLUDES = -I/usr/local/mpi/include.
NOTE: If you don't want to use any of the MPI code, just set:
BUILD_MPI= false
in SiteSpecific. All the other MPI-specific defines are then ignored.