Introduction to Programming with MPI
------------------------------------

Practical Exercise 08 (More on Point-to-Point)
----------------------------------------------

For general instructions, see the introduction to the collective
practicals.  The remarks on buffered, synchronous and ordinary sends
in the introduction to the point-to-point practicals also apply to
these ones.

Note that these examples are of the very simplest ways of using
non-blocking I/O, which is all you are recommended to do initially.



Question 1
----------

1.1 Take a copy of the program you wrote in question 3.1 in practical
exercise 4, that was called 'Daggoo'.  Remove the support for the
buffering, and change the rotation collective to use non-blocking,
ordinary (i.e. not buffering) sends and blocking, synchronous receives.


1.2 Change the rotation collective to use blocking, synchronous sends
and non-blocking, ordinary receives.



Question 2
----------

2.1 Take a copy of the program you wrote in question 5.1 in practical
exercise 3, that was called 'Stubb'.  Write a function that implements
all-to-all using non-blocking sends and receives, and waits on all of
them at once, to provide the same effect as MPI_Alltoall.  Change the
call of MPI_Alltoall to use that.



Question 3
----------

3.1 This will seem a little pointless, but its usefulness will appear
later, under problem decomposition and I/O.  Use process 0 as the root
process, and write a new program.

In each process except the root process, set an integer variable m to
the process id plus one (i.e. integers from 2 upwards) and integer
variable n to 1.  Then write an indefinite loop that calculates
 'n = MOD(m*n,32)' in Fortran and 'n = (m*n)%32' in C/C++, and send each
number to process 0 (using MPI_Send) as it is calculated.  When n
reaches 0 or 1, break out of the loop and finish off.

In the root process, code an indefinite loop that uses one non-blocking
receive for each other process and MPI_Waitany to print out the process
number it receives from and the value of n it was sent, in the order
they are received.  It will need to create a new non-blocking receive
for the sending process every time it gets a new value.  When all other
processes have sent 0 or 1, break out of the loop and finish off.


3.2 Change the program to use a single non-blocking receive and
MPI_Wait, but make the program do the same job.