Hi Anyi,
you are using reduce incorrectly: you cannot use the same buffer as input
and output.
If you want to do operation in place, you must specify "MPI_IN_PLACE"
as send buffer at the root process.
Thus, your code should look something like:
--------
int* ttt = (int*)malloc(2 * sizeof(int));
ttt[0] = myworldrank + 1;
ttt[1] = myworldrank * 2;
if (root == myworldrank) {
MPI_Reduce(MPI_IN_PLACE, ttt, 2, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
} else {
MPI_Reduce(ttt, NULL, 2, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
}
FOR_WORLDNODE0 printf("%d, %d\n" , ttt[0],ttt[1]);
--------
hope this helps,
Jelena
PS. If I remember correctly the standard, you must specify send buffer on
non-root nodes - it cannot be MPI_IN_PLACE (if you try it - you'll get
segfault).
On Wed, 11 Jul 2007 any...@pa.uky.edu wrote:
Hi,
I have a code which have a identical vector on each node, I am going to do
the vector sum and return result to root. Such like this,
int* ttt = (int*)malloc(2 * sizeof(int));
ttt[0] = myworldrank + 1;
ttt[1] = myworldrank * 2;
MPI_Allreduce(ttt, ttt, 2, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
FOR_WORLDNODE0 printf("%d, %d\n" , ttt[0],ttt[1]);
myworldrank is the rank of local node, if I run this code under 4 nodes, what
I expect return is 10,12. But what I got is 18,24. So I'm confused here on
MPI_Reduce, is that supposed to do the vector sum ?
I tried MPI_Allreduce, it gave me the correct answer 10, 12.
Is there someone met the same problems or I am wrong on calling MPI_Reduce()
Thanks.
Anyi
_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users
--
Jelena Pjesivac-Grbovic, Pjesa
Graduate Research Assistant
Innovative Computing Laboratory
Computer Science Department, UTK
Claxton Complex 350
(865) 974 - 6722
(865) 974 - 6321
jpjes...@utk.edu
Murphy's Law of Research:
Enough research will tend to support your theory.
