This directory contains three programs aimed at highlighting the differences between blocking and non-blocking function calls.
A blocking function will not return until it's designated task is completed. In the context of MPI, this may be sending or receiving a message.
A non-blocking function can return immediately, but we cannot
assume that the function's designated task has been completed.
Instead we should think of calling this function as posting a
request for some action.
We must check at some point later
in our program whether the action has been completed.
In the
context of MPI, and non-blocking receive function records the
request to receive a message but may well return before that
message has actually arrived.
MPI_Test() and MPI_Wait() are
two functions used to determine whether a request has been
satisfied.
This is a simple point-to-point message passing program using
the blocking communications functions MPI_Send() and MPI_Recv().
Processes are set to sleep for various periods of time so that
the messages are sent at different times, and so the calls to
MPI_Recv() take some seconds to return.
In this program, we substitute MPI_Send() and MPI_Recv() with
their non-blocking counterparts, MPI_Isend() and MPI_Irecv().
Both these functions can return immediately, and not wait
for the communication transaction to be completed.
Again delays are introduced to the program via the sleep
function and MPI_Wait() is used to check on the receive
requests.
This program omits the call to MPI_Wait() and so subsequent
stages of the program are reached before they should be, i.e.
before key messages are received.
Experiment with the MPI_Test() function:
int MPI_Test(MPI_Request *request, int *flag, MPI_Status *status)