Receive buffer to MPI_Reduce

[funceval]

Hi,

Does MPI_Reduce expect us to allocate memory to the receive buffer on
processors other than the root node? Is something like this guaranteed
to work:

int* local_sum = (int*)calloc(sizeof(int), count);
int* total_sum = 0;

//processor 0 is the root node
if (proc_id == 0)
{
total_sum = (int*)calloc(sizeof(int), count);
}

MPI_Reduce(local_sum, total_sum, count, MPI_INT, 0, MPI_COMM_WORLD);

free(local_sum);
if (proc_id == 0)
{
free(total_sum);
}

[Gilles]

Hi,

Post by funceval
Hi,
Does MPI_Reduce expect us to allocate memory to the receive buffer on
processors other than the root node?

No, for processes other than the root one, the receive buffer is ignored.

Is something like this guaranteed

Post by funceval
int* local_sum = (int*)calloc(sizeof(int), count);
int* total_sum = 0;
//processor 0 is the root node
if (proc_id == 0)
{
total_sum = (int*)calloc(sizeof(int), count);
}
MPI_Reduce(local_sum, total_sum, count, MPI_INT, 0, MPI_COMM_WORLD);

It should fail, not due to the receive buffer not allocated for processes other than root which is correct, but because there's no operator here. May be MPI_SUM is missing...

Gilles

Post by funceval
free(local_sum);
if (proc_id == 0)
{
free(total_sum);
}

[funceval]

Post by Gilles

Post by funceval
//processor 0 is the root node
if (proc_id == 0)
{
    total_sum = (int*)calloc(sizeof(int), count);
}
MPI_Reduce(local_sum, total_sum, count, MPI_INT, 0, MPI_COMM_WORLD);

It should fail, not due to the receive buffer not allocated for processes other than root which is correct, but because there's no operator here. May be MPI_SUM is missing...

That was a typo. Thanks for the answer.