simple MPI program fails

Here's the program:

Code:

#include <stdio.h>
#include <mpi.h>

main(int argc, char **argv)
{
  int node;

  MPI_Init(&argc,&argv);
  MPI_Comm_rank(MPI_COMM_WORLD, &node);

  printf("Hello World from Node %d\n",node);

  MPI_Finalize();
}

Here's how I compile it

Code:

mpiCC -o simple simple.cpp -lstdc++

Here's how I run it:

Code:

mpirun -np 2 simple

Here's the error I get:

Code:

[macbook:24160] *** Process received signal ***
[macbook:24160] Signal: Segmentation fault (11)
[macbook:24160] Signal code: Address not mapped (1)
[macbook:24160] Failing at address: 0x440000b0
[macbook:24160] [ 0] 2   libSystem.B.dylib                   0x00007fff871371ba _sigtramp + 26
[macbook:24160] [ 1] 3   libSystem.B.dylib                   0x00007fff70c545e0 __stack_chk_guard + 0
[macbook:24160] [ 2] 4   simple                              0x0000000100000eb6 main + 42
[macbook:24160] [ 3] 5   simple                              0x0000000100000e84 start + 52
[macbook:24160] *** End of error message ***
[macbook:24161] *** Process received signal ***
[macbook:24161] Signal: Segmentation fault (11)
[macbook:24161] Signal code: Address not mapped (1)
[macbook:24161] Failing at address: 0x440000b0
[macbook:24161] [ 0] 2   libSystem.B.dylib                   0x00007fff871371ba _sigtramp + 26
[macbook:24161] [ 1] 3   libSystem.B.dylib                   0x00007fff70c545e0 __stack_chk_guard + 0
[macbook:24161] [ 2] 4   simple                              0x0000000100000eb6 main + 42
[-macbook:24161] [ 3] 5   simple                              0x0000000100000e84 start + 52
[macbook:24161] *** End of error message ***
mpirun noticed that job rank 0 with PID 24160 on node macbook.local exited on signal 11 (Segmentation fault). 
1 additional process aborted (not shown)

Anyone have ANY suggestions? I honestly have no idea.

Does it still fail if you give it command line arguments?

Originally Posted by tabstop

Does it still fail if you give it command line arguments?

I just tried it, and it still fails with the same error stuff.

I did

Code:

mpirun -np 4 testmpi sdf

I don't know enough about MPI to know if this is a problem: you have C code. You have specified the C++ compiler and the C++ standard library. Does it work if you change the compiler/libraries to the C versions? Does it work if you change the code to match the C++ bindings (MPI namespace, etc.)?

Originally Posted by tabstop

I don't know enough about MPI to know if this is a problem: you have C code. You have specified the C++ compiler and the C++ standard library. Does it work if you change the compiler/libraries to the C versions? Does it work if you change the code to match the C++ bindings (MPI namespace, etc.)?

Unfortunately, no. This was just a simple example I got from online and tried to run it. In the code I was actually trying to test, it's C++, and I get the exact same errors.

I was able to log in to a cluster and run this simply on the log-in nodes, and the hello world example does indeed work. There must be something goofed up with my configuration.

I'd like to know what got goofed up.. i've ran sample MPI apps on my macbook before, with no problem.

__stack_chk_guard sounds like some kind of stack corruption protector. What happens if you compile with -fno-stack-protector? Does it crash somewhere else then?
Also try compiling with -g to see the line number of the crash instead of just "main + 42"

Originally Posted by _Mike

__stack_chk_guard sounds like some kind of stack corruption protector. What happens if you compile with -fno-stack-protector? Does it crash somewhere else then?
Also try compiling with -g to see the line number of the crash instead of just "main + 42"

I compiled it with -g and -fno-stack-protector, and I get the exact same error (main +42). Perhaps mpiCC doesn't understand those options?

Edit: i compiled with -g once, and ran it, then with -fno-stack-protector once and ran it. Same error both runs.

Originally Posted by dayalsoap

I compiled it with -g and -fno-stack-protector, and I get the exact same error (main +42). Perhaps mpiCC doesn't understand those options?.

It should. As far as I know mpiCC is just a wrapper for g++. But I have never written any code on mac os before so I don't know if -g is supported there or if you have to use something else.
I don't have any experience with openmpi but I did try your code under Mandriva 2010.2 x64 with openmpi 1.4.1 though, and it works fine for me so the problem is probably not in your code.
Do macs have valgrind? Maybe that will give some more information about the crash.

Yes, I can get it to work on Linux. I'm going to reinstall some things and see what happens.

Ok, I installed the latest version of OpenMPI, and it works. The only difference is, instead of mpiCC, i had to do mpic++. That's it.

Congrats on getting that to work. Were you experimenting with this or did you have a particular application in mind? I've got the same example you posted working in C++ but I'm looking to use IMSL/ScaLapack fortran MPI to parallelize a panel code...if I ever get around to it, which might take exactly a billion years.

Anyway, perhaps you'd like to keep in touch.

Cheers.