3.1. Parallel for Loop Program Structure: equal chunks

Navigate to: ../02.parallelLoop-equalChunks/

Make and run the code:

make

mpirun -hostfile ~/hostfile -np N ./parallelLoopEqualChunks

Here the N signifies the number of processes to start up in MPI.

Exercises:

• Run, using these numbers of processes, N: 1, 2, 4, and 8.

• Change REPS to 16, save, rerun, varying N again.

• Explain how this pattern divides the iterations of the loop among the processes.

Q-1: Which of the following is the correct assignment of loop iterations to processes for this code, when REPS is 8 and numProcesses is 4?


• The processes are assigned chunks in order of their process number.
• Each process gets assigned to an equal-sized chunk of the repetitions in order.
• In this code, blocks, or chunks of more than one repetition are assigned consecutively.

3.1.1. Explore the code

In the code below, notice the use of the variable called REPS. This is designed to be the total amount or work, or repetitions, that the for loop is accomplishing. This particular code is designed so that if those repetitions do not divide equally by the number of processes, then the program will stop with a warning message printed by the conductor process.

Remember that because this is still also a SPMD program, all processes execute the code in the part of the if statement that evaluates to True. Each process has its own id, and we can determine how many processes there are, so we can choose where in the overall number of REPs of the loop each process will execute.

#include <stdio.h> // printf()
#include <mpi.h>   // MPI

int main(int argc, char** argv) {
    const int REPS = 8;                      // repetitions in a loop
    int id = -1, numProcesses = -1;

    MPI_Init(&argc, &argv);
    MPI_Comm_rank(MPI_COMM_WORLD, &id);
    MPI_Comm_size(MPI_COMM_WORLD, &numProcesses);

    // In this example, ensure that the REPS can ben evenly divided by the
    // number of processors and that the number of processes doesn't exceed REPS.
    // If either is the case, stop.
    if ((REPS % numProcesses) == 0 && numProcesses <= REPS) {

      int chunkSize = REPS / numProcesses;      // find chunk size
      int start = id * chunkSize;               // find starting index
      int stop = start + chunkSize;             // find stopping index

      for (int i = start; i < stop; i++) {      // iterate through our range
          printf("Process %d is performing iteration %d\n", id, i);
      }

    } else {
      if (id == 0) {
          printf("Please run with -np divisible by and less than or equal to %d\n.", REPS);
      }
    }

    MPI_Finalize();
    return 0;
}

You have attempted of activities on this page