[MPICH] a problem with MPI

[Matthew Chambers]

MPI_Init and MPI_Finalize are process-scope functions.  They are not 
designed to be called more than once, IIRC.  You call MPI_Init once 
before you make any other MPI calls (and typically right after that you 
get the number of processes and the current process's id and store those 
globally), and you call MPI_Finalize once after you are finished making 
all MPI calls.  For your for loop that needs to be parallelized, there's 
little help we can give you until you are more specific about what 
you're trying to accomplish.  It seems like a read through of some MPI 
references is in order though.

-Matt Chambers


Anna Mereu wrote:
> Hi, i have a problem concerning the implementation of a program in a 
> parallel way. I don't know if this is the  right forum, but i don't 
> know any other forum that can help me.
> Inside my program i have  a procedure that must be parallel. So at the 
> beginning and at the end of it, there are the MPI_Init and 
> MPI_Finalize commands. At this point all seems to go correctly.
> But the fact is that i have to run this procedure a lot of times 
> subsiquently, and i do it with a for cycle.
> The problem that occurs to me is that, while during the first 
> iteration the process ids go from 0 to nproc-1 (so i can simpy 
> identify them and assign different jobs to them), during the next 
> iterations the ids are different and in particular they seem to be 
> random. In this way i cannot identify a job (for instance the master) 
> from another.
> In practice inside the procedure i have something like...
>
> if(myid==0) then......
> else.....
>
> How can i face this problem?
>
> Thank you in advance for the help
>
> Anna
>
>