Sum of an array using MPI

Prerequisite: MPI – Distributed Computing made easy

Message Passing Interface(MPI) is a library of routines that can be used to create parallel programs in C or Fortran77. It allows users to build parallel applications by creating parallel processes and exchange information among these processes.
MPI uses two basic communication routines:

  • MPI_Send, to send a message to another process.
  • MPI_Recv, to receive a message from another process.

The syntax of MPI_Send and MPI_Recv is:



int MPI_Send(void *data_to_send, 
             int send_count, 
             MPI_Datatype send_type, 
             int destination_ID, 
             int tag, 
             MPI_Comm comm); 

int MPI_Recv(void *received_data, 
             int receive_count, 
             MPI_Datatype receive_type,
             int sender_ID, 
             int tag, 
             MPI_Comm comm, 
             MPI_Status *status);

To reduce the time complexity of the program, parallel execution of sub-arrays is done by parallel processes running to calculate their partial sums and then finally, the master process(root process) calculates the sum of these partial sums to return the total sum of the array.

Examples:

Input : {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
Output : Sum of array is 55

Input : {1, 3, 5, 10, 12, 20, 4, 50, 100, 1000}
Output : Sum of array is 1205

Note – You must have MPI installed on your Linux based system for executing the following program. For details to do so, please refer MPI – Distributed Computing made easy

Compile and run the program using following code:

mpicc program_name.c -o object_file
mpirun -np [number of processes] ./object_file

Below is the implementation of the above topic:

filter_none

edit
close

play_arrow

link
brightness_4
code

#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
  
// size of array
#define n 10
  
int a[] = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 };
  
// Temporary array for slave process
int a2[1000];
  
int main(int argc, char* argv[])
{
  
    int pid, np,
        elements_per_process,
        n_elements_recieved;
    // np -> no. of processes
    // pid -> process id
  
    MPI_Status status;
  
    // Creation of parallel processes
    MPI_Init(&argc, &argv);
  
    // find out process ID,
    // and how many processes were started
    MPI_Comm_rank(MPI_COMM_WORLD, &pid);
    MPI_Comm_size(MPI_COMM_WORLD, &np);
  
    // master process
    if (pid == 0) {
        int index, i;
        elements_per_process = n / np;
  
        // check if more than 1 processes are run
        if (np > 1) {
            // distributes the portion of array
            // to child processes to calculate
            // their partial sums
            for (i = 1; i < np - 1; i++) {
                index = i * elements_per_process;
  
                MPI_Send(&elements_per_process,
                         1, MPI_INT, i, 0,
                         MPI_COMM_WORLD);
                MPI_Send(&a[index],
                         elements_per_process,
                         MPI_INT, i, 0,
                         MPI_COMM_WORLD);
            }
  
            // last process adds remaining elements
            index = i * elements_per_process;
            int elements_left = n - index;
  
            MPI_Send(&elements_left,
                     1, MPI_INT,
                     i, 0,
                     MPI_COMM_WORLD);
            MPI_Send(&a[index],
                     elements_left,
                     MPI_INT, i, 0,
                     MPI_COMM_WORLD);
        }
  
        // master process add its own sub array
        int sum = 0;
        for (i = 0; i < elements_per_process; i++)
            sum += a[i];
  
        // collects partial sums from other processes
        int tmp;
        for (i = 1; i < np; i++) {
            MPI_Recv(&tmp, 1, MPI_INT,
                     MPI_ANY_SOURCE, 0,
                     MPI_COMM_WORLD,
                     &status);
            int sender = status.MPI_SOURCE;
  
            sum += tmp;
        }
  
        // prints the final sum of array
        printf("Sum of array is : %d\n", sum);
    }
    // slave processes
    else {
        MPI_Recv(&n_elements_recieved,
                 1, MPI_INT, 0, 0,
                 MPI_COMM_WORLD,
                 &status);
  
        // stores the received array segment
        // in local array a2
        MPI_Recv(&a2, n_elements_recieved,
                 MPI_INT, 0, 0,
                 MPI_COMM_WORLD,
                 &status);
  
        // calculates its partial sum
        int partial_sum = 0;
        for (int i = 0; i < n_elements_recieved; i++)
            partial_sum += a2[i];
  
        // sends the partial sum to the root process
        MPI_Send(&partial_sum, 1, MPI_INT,
                 0, 0, MPI_COMM_WORLD);
    }
  
    // cleans up all MPI state before exit of process
    MPI_Finalize();
  
    return 0;
}

chevron_right


Output:

Sum of array is 55

Below is the snapshot of the processes calculating their partial sums:



My Personal Notes arrow_drop_up

Check out this Author's contributed articles.

If you like GeeksforGeeks and would like to contribute, you can also write an article using contribute.geeksforgeeks.org or mail your article to contribute@geeksforgeeks.org. See your article appearing on the GeeksforGeeks main page and help other Geeks.

Please Improve this article if you find anything incorrect by clicking on the "Improve Article" button below.




Article Tags :
Practice Tags :


3


Please write to us at contribute@geeksforgeeks.org to report any issue with the above content.