SMPI is robust, but it still has some limitations, as listed (and further explained) below:
SMPI runs your MPI processes as threads within a single process. So when you declare a global variable in your
program
(which is rarely a good idea anyway), all the threads will share it, as threads do. That's likely not what you
want
in terms of your MPI processes. Use local variables instead.
Because SMPI already uses multithreading to simulate your MPI program's execution, if your MPI program uses
threads
as well, mayhem can occur. In practice, i.e., with real MPI, you typically want to use threads to exploit
multi-core
machines. In these pedagogic modules, simply pretend that each host is single-core and that you don't need to use
threads.
This has no impact on your learning MPI, and if/when you transition to real (as opposed to simulated) MPI
executions, you
can then simply multi-thread computational sections of your code.
Yes, this is vague. smpicc compiles your code, and is fairly robust. But if you go crazy with
tons of macros and C oddness, smpicc may get confused. It should happen only in rather extreme
cases, but you are now warned.
