Hi --
Could somebody explain why the following program hangs? I am running
LAM 7.0.3 on a dual-processor xeon box with RedHat 9. If I run it
simulating parallelism with more than 2 processes, it hangs. I know
there are more efficient ways to implement what the program is doing,
but it distills the nature of something that is messing up a much more
complex application. I have been coding in MPI since 1996, although
mostly in a more asynchronous mode with Isends and Irecvs. It seems to
me this program should work. I tried putting a barrier between the
"gather" and "scatter" phases, but that did *not* help! Perhaps my
brain is just locked -- anything to clear the fog would be appreciated!
-- (Professor) Jonathan Eckstein
#include <mpi.h>
#include <stdio.h>
int main(int argc, char** argv)
{
int r,s,t,u,v;
MPI_Status status;
MPI_Init(&argc,&argv);
MPI_Comm_rank(MPI_COMM_WORLD,&r);
MPI_Comm_size(MPI_COMM_WORLD,&s);
t = 0;
if (r==0)
{
for(u=1; u<s; u++)
{
MPI_Recv(&v,1,MPI_INT,MPI_ANY_SOURCE,1,MPI_COMM_WORLD,&status);
printf("Got message from %d\n",status.MPI_SOURCE);
t += v;
}
printf("Total=%d\n",t);
for(u=1; u<s; u++)
{
printf("Root sending to %d\n",u);
MPI_Send(&t,1,MPI_INT,1,2,MPI_COMM_WORLD);
}
}
else
{
printf("Processor %d sending...\n",r);
MPI_Send(&r,1,MPI_INT,0,1,MPI_COMM_WORLD);
MPI_Recv(&t,1,MPI_INT,0,2,MPI_COMM_WORLD,&status);
printf("Processor %d got %d\n",r,t);
}
MPI_Finalize();
return 0;
}
|
