User Tools

Site Tools


c:mpi:rank-sort-parallel

MPI Rank Sort in Parallel

We will follow the specific steps of the rank-based sorting algorithm, parallelizing the calculation of ranks. Each process will compute the rank for a subset of the vector and contribute to determining the final position of elements in the sorted array.

The specific steps for the MPI implementation are:

Element Distribution: Each process will receive a subset of the vector to work on. The master process (typically rank 0) will distribute parts of the vector to all processes, including itself.

Rank Calculation: Each process calculates the ranks for its elements. This involves comparing each element with all elements of the vector and counting how many are smaller.

Gathering Ranks and Sorting: After calculating the ranks, each process sends its results back to the master process, which then assembles the sorted vector based on these ranks.

mpi-rank-sort-parallel.c
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
 
#define MASTER 0
 
void distribute_elements(int *vector, int *sub_vector, int n, int sub_n, int world_rank, int world_size) {
    // Distribute parts of the vector to all processes
    MPI_Scatter(vector, sub_n, MPI_INT, sub_vector, sub_n, MPI_INT, MASTER, MPI_COMM_WORLD);
}
 
void calculate_ranks(int *sub_vector, int *vector, int *ranks, int n, int sub_n, int world_rank) {
    // Calculate ranks for each element in the sub-vector
    for (int i = 0; i < sub_n; i++) {
        ranks[i] = 0;
        for (int j = 0; j < n; j++) {
            if (sub_vector[i] > vector[j] || (sub_vector[i] == vector[j] && (world_rank * sub_n + i) > j)) {
                ranks[i]++;
            }
        }
    }
}
 
void gather_and_sort(int *sorted_vector, int *sub_vector, int *ranks, int sub_n, int world_rank, int world_size) {
    // Gather the ranks from all processes
    int *all_ranks = NULL;
    if (world_rank == MASTER) {
        all_ranks = (int*) malloc(world_size * sub_n * sizeof(int));
    }
    MPI_Gather(ranks, sub_n, MPI_INT, all_ranks, sub_n, MPI_INT, MASTER, MPI_COMM_WORLD);
 
    // Master process sorts the array based on the ranks
    if (world_rank == MASTER) {
        for (int i = 0; i < world_size * sub_n; i++) {
            sorted_vector[all_ranks[i]] = sub_vector[i];
        }
    }
 
    free(all_ranks);
}
 
int main(int argc, char** argv) {
    MPI_Init(&argc, &argv);
 
    int world_rank, world_size;
    MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);
    MPI_Comm_size(MPI_COMM_WORLD, &world_size);
 
    int n = 10; // Size of the vector
    int *vector, *sorted_vector, *sub_vector, *ranks;
    int sub_n = n / world_size; // Size of the sub-vector for each process
 
    if (world_rank == MASTER) {
        vector = (int*) malloc(n * sizeof(int));
        sorted_vector = (int*) malloc(n * sizeof(int));
        // Initialize the vector with random numbers
        for (int i = 0; i < n; i++) {
            vector[i] = rand() % 100; // Random numbers between 0 and 99
        }
    }
 
    sub_vector = (int*) malloc(sub_n * sizeof(int));
    ranks = (int*) malloc(sub_n * sizeof(int));
 
    distribute_elements(vector, sub_vector, n, sub_n, world_rank, world_size);
    calculate_ranks(sub_vector, vector, ranks, n, sub_n, world_rank);
    gather_and_sort(sorted_vector, sub_vector, ranks, sub_n, world_rank, world_size);
 
    if (world_rank == MASTER) {
        printf("Sorted array: ");
        for (int i = 0; i < n; i++) {
            printf("%d ", sorted_vector[i]);
        }
        printf("\n");
        free(vector);
        free(sorted_vector);
    }
 
    free(sub_vector);
    free(ranks);
 
    MPI_Finalize();
    return 0;
}

Each process, including the master, gets a subset of the vector to work on.

They calculate the ranks for their elements by comparing each element with all other elements.

After calculating the ranks, the master process gathers these ranks and uses them to assemble the sorted array.

Finally, the master process prints the sorted array.

The size of the array (n) should be a multiple of the number of processes for even distribution. The output will be the sorted array.

c/mpi/rank-sort-parallel.txt · Last modified: 2024/02/21 13:57 by odefta