MPI  benchmark on Cray XD1 (2.2 GHz AMD Opteron 248)
Remarks:
General  : The programs are compiled with mpif90 which in fact is a shell
           around the PathScale f90 compiler, version 2.1 and the
	   MPI version mpich-1.2.6-path20. As the name of the latter
	   indicates, it is not a full MPI-2 implementation which shows
	   in program mod1k (see below).
- mod1k  : This program could not be run because the MPI/Pathscale
           implementation does not (yet) support one-sided communication,
	   like MPI_Put/MPI_Get.
- mod2ci : The last entry on 16 processors is nonsense because the MPI
           implementation does not support (yet) the MPI_Integer8 type.
	   Therefore the flop count had to be typed as Integer, resulting
	   in overflow of the flop count and a corresponding bad value
	   for the Mflop rate.
- mod2i  : For as yet untraceable reasons the parallel sort on 16 processors
           does not run. There is no usable information as to what caused
	   the failure but, again, the MPI implementation is the most
	   probable suspect.	    	   
