
Example 2.17 "Use of nonblocking communications in Jacobi computation"
from "MPI: The Complete Reference, vol. 1".

There is an fault in this code that is revealed by particular
configurations.  The problem happens when on at least one proc, m==1,
which will happen whenever n<2p.  Then two sends are posted from the
same buffer (the single column of local matrix B on at least one
process).  See the Makefile for the case that reveals the fault.

Otherwise, we are able to verify that the sequential and parallel
versions correspond.  (The sequential version is given as Ex. 2.12
earlier in the book.)

