Mercury Race

Hughes Aircraft Co. has implemented MPI for Mercury RACE systems [18]. RACE MPI is derived from MPICH. There are a few interesting features of this implementation that are worth noting.

• On SHARC systems, where a ``byte'' (defined by ANSI C to be the size of a char) is 32 bits, not 8 bits, this implementation exposed a portability problem for MPI codes.
• The MPICH library has been modified to conserve as much space as possible. Only needed routines are linked, argument checking and strings are omitted.
• Several collective operations have been optimized to use shared memory.