Segmentation faults occur when I run a parallel program with Open MPI

Question

on my previous post I needed to distribute data of pgm files among 10 computers. With help from Jonathan Dursi and Shawn Chin, I have integrate the code. I can compile my progra

Answer 1

Congratulations; the code almost ran completely perfectly, it died on almost the final lines of code.

The issue would have been a little clearer with valgrind, but you have to be trickier running valgrind with MPI -- or anything that involves a program launcher. Instead of:

valgrind mpirun -np 10 ./exmpi_2 balloons.pgm output.pgm

which does a valgrind of mpirun, which you don't really care about, you want to do

mpirun -np 10 valgrind ./exmpi_2 balloons.pgm output.pgm

-- that is, you want to launch 10 valgrinds, each running one process' worth of exmpi_2. If you do that (and you've compiled with -g), you'll find towards the end, valgrind output like the following:

==6303==  Access not within mapped region at address 0x1
==6303==    at 0x387FA60C17: fclose@@GLIBC_2.2.5 (in /lib64/libc-2.5.so)
==6303==    by 0x401222: main (pgm.c:124)

.. and that's all there is to it; you have all processes doing the fclose()s, when only one process has a handle to a fopen()ed file in the first place. Simply replacing

fclose(FR);
fclose(FW);

with

if (rank == IONODE) {
    fclose(FR);
    fclose(FW);
}

seems to work for me.