MPI_Bcast, broadcasting array buffer to specific locations in the receive buffer - c++

Question

Using MPI, I want to do a broadcast operation by all processes in the communication group such that at the end of boradcast by all processes, the buffer in all processes have the same data.
Here is a fragment of code depicting what I want to do :
//assume there are 10 processes
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &size);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
double globalArray[100];
for(int i=0; i<100; ++i) {
A[i] = (double)i + 1.0;
}
double buffer[10]; //assume all entries in buffer is zero
//only one array location in each rank is initialized and rest remain zero
buffer[rank] = globalArray[(rank + 1)*10]
MPI_Bcast(&buffer, 1, MPI_DOUBLE, rank, MPI_COMM_WORLD);
MPI_Barrier(MPI_COMM_WORLD);
for(int i=0; i<10; ++i) {
std::cout << buffer[i] << "\t";
}
//expect to get [10 20 ... 100] in all the processes
MPI_Finalize();
I know there is MPI_Scatterv which on calling on each processor can do the work but this means I have to create two additional arrays, one for send_counts and displacements, which will always be the same for each scatterv operation.
Is there an easier way to do this ?

Answer 1

If i understand correctly your question, what you really need is MPI_Allgather()
double myvalue = (rank + 1) * 10;
MPI_Allgather(&myvalue, 1, MPI_DOUBLE, buffer, 1, MPI_DOUBLE, MPI_COMM_WORLD);
MWE:
#include <mpi.h>
#include <iostream>
int main(int argc, char* argv[])
{
int size, rank;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &size);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
double buffer[10];
double myvalue = (rank + 1) * 10;
MPI_Allgather(&myvalue, 1, MPI_DOUBLE, buffer, 1, MPI_DOUBLE, MPI_COMM_WORLD);
if (rank == 0) {
for(int i=0; i<10; ++i) {
std::cout << buffer[i] << "\t";
}
}
// answer for all processes: 10 20 ... 100
MPI_Finalize();
return 0;
}