We won't have an F90 binding for MPI until MPI-2, but that doesn't stop
people from combining the two standards. Ian Foster et al. proposed
an HPF binding for MPI [72,71]. A very nice
on-line document by David LaFrance-Linden [118] describes
how to use HPF to code up message passing between computational kernels
(e.g. how to write ``gasketware'').
Several of the HPF tutorials at Cornell show how to use MPI,
e.g.,
http://www.tc.cornell.edu/Edu/Tutor/HPF/DimensionReducingGasket/
.
In fact, many HPF compilers use MPI as the communications layer.
So MPI tools work (to varying degrees) for HPF programs.
For example, VAMPIR could be used to determine communications overhead
in an HPF code.