mpiBarrier(mpiCommWorld) not working correctly?

General Discussion
1

It seems like mpiBarrier(mpiCommWorld) is not functioning as expected. Here’s the situation:

Given the following FreeFEM++ code:
/////////////////////////////////////////////////////////////////////////
verbosity = 0;
mpiComm comm(mpiCommWorld, 0, 0);
for (int i = 0; i < mpisize; ++i) {
if (i == mpirank) {
cout << "mprocess = " << mpirank << " among = " << mpisize << endl;
}
mpiBarrier(mpiCommWorld);
}
/////////////////////////////////////////////////////////////////////////////
The output I get is:
mprocess = 0 among = 4
mprocess = 1 among = 4
mprocess = 3 among = 4
mprocess = 2 among = 4

However, I was expecting this output:
mprocess = 0 among = 4
mprocess = 1 among = 4
mprocess = 2 among = 4
mprocess = 3 among = 4

Is mpiBarrier(mpiCommWorld) not working correctly? It should ensure that all processes synchronize at each iteration, so only one process outputs at a time. Could you clarify if there’s an issue with my understanding or usage of mpiBarrier?

2

What is not working as expected? Why would you expect the output to be:

mprocess = 0 among = 4
mprocess = 1 among = 4
mprocess = 2 among = 4
mprocess = 3 among = 4

I’m guessing if you want to be sure you get the expected output, you should probably flush cout.

3

Thank you very much for your reply.
The reason I expected the output to be:

mprocess = 0 among = 4
mprocess = 1 among = 4
mprocess = 2 among = 4
mprocess = 3 among = 4

is because in my code, I use a for loop where each process only executes the cout statement when its rank matches the current iteration index i. For example:

  • When i == 0, only process 0 executes cout.
  • When i == 1, only process 1 executes cout, and so on.

After each iteration, I use mpiBarrier(mpiCommWorld); to synchronize all processes. My understanding was that this synchronization would ensure that the output from process 0 is completed before process 1 starts its output, and so on.

However, the actual output is out of order, such as:

mprocess = 0 among = 4
mprocess = 1 among = 4
mprocess = 3 among = 4
mprocess = 2 among = 4

Regarding your suggestion about flushing cout, I was under the impression that using endl in my cout statements already performs a flush. Could you tell me why endl might not be sufficient in this case? Or could the issue be related to the way mpiBarrier is working in this specific context?

4

My bad, endl does actually flush the buffer. I’m not sure why you get this, what is your system? FWIW, on my Linux and macOS, I get your expected output.

5

Sorry for late reply.
I’m using Windows 11, Intel Core i5 8265.

6

OK, not a big surprise then.