Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: [OMPI users] Problems with MPI_Init_Thread(...)
From: devendra rai (rai.devendra_at_[hidden])
Date: 2011-09-19 04:15:42


Hello Community, I am building an application which uses MPI_Ssend(...) and MPI_Recv(...) in threads. So, there is more than one thread which invokes MPI functions. Based on Jeff's inputs, I rebuilt open-mpi with threads support: ./configure --enable-mpi-threads=yes --with-threads=posix ... I believe the above line would do the magic. Now, I learnt that I need to request thread support when initializing MPI application using MPI_Init_thread(...). This is my code fragment: int main(int argc, char *argv[]) {         int rank;         int desired_thread_support = MPI_THREAD_MULTIPLE;         int provided_thread_support;             MPI_Init_thread(&argc, &argv, desired_thread_support, &provided_thread_support);            MPI_Comm_set_errhandler(MPI_COMM_WORLD, MPI_ERRORS_RETURN); I build my application, and run it on 3 nodes. However, I get this error: Sun Sep 18 22:25:11 2011[1,1]<stderr>:*** The MPI_Init_thread() function was called before MPI_INIT was invoked. Sun Sep 18 22:25:11 2011[1,1]<stderr>:*** This is disallowed by the MPI standard. Sun Sep 18 22:25:11 2011[1,1]<stderr>:*** Your MPI job will now abort. Sun Sep 18 22:25:11 2011[1,1]<stderr>:[Morpheus.OTZ:9910] Abort before MPI_INIT completed successfully; not able to guarantee that all other processes were killed! I did read some online examples, and API documentations, and nothing says that I will need to call MPI_Init(...) *followed by* MPI_Init_thread(...). I have open-mpi 1.4.3 from open-mpi official site, running on Linux kernel 2.6.37.6-0.7. I do not understand why this error is happening. Again, the next question is, Am I required to MPI_Init_thread everytime I create a new thread? If yes, will this call be the first line in the thread? Would be nice if someone can throw some pointers. Thanks a lot. Best, Devendra Rai ________________________________