Implemented parallelized quicksort using Openmpi library.

• Split the main vector into a number of smaller parts using mpi_scatter function of mpi.
• If length of vector is |n|, then if n%numprocs == 0 --> we give m = n/numprocs to each process else we zero-pad the array and give m = n/numprocs+1 to each process.
• We then sort each individual subpart/subvectors parrallely in each process.
• We now have m sorted vectors.We need to merge them.
• We can merge by using same in O(m) time and in log(p) steps.
• In the merge step, I have used MPI_Send and MPI_Recv to exchange vectors between processes.
• Testing:
  • 1. sequential-0.50517 parallel-0.243557
  • 2. sequential-0.5076 parallel-0.272
  • 3. sequential-0.482768 parallel-0.2523
  • 4. sequential-0.499613 parallel-0.25343
  • 5. sequential-0.343654 parallel-0.23374
  • 6. sequential-0.365511 parallel-0.185698
  • 7. sequential-0.414262 parallel-0.217249
  • 8. sequential-0.452915 parallel-0.23488
  • 9. sequential-0.4259 parallel-0.2376
  • 10. sequential-0.504616 parallel-0.243047
  • 11. sequential-0.50169 parallel-0.281339
  • 12. sequential-0.50517 parallel-0.243557
  • 13. sequential-0.50924 parallel-0.247054
  • 14. sequential-0.4852 parallel-0.2435
  • 15. sequential-0.513111 parallel-0.24779

• Split the edge_u,edge_v,weight vectors into smaller chunks and distributed among p process.
• Zero-padded the vectors and used the same above mentioned strategy for partitioning.
• Relaxed the edges (n-1) times (Assumption: No negative cycle is present).
• Have used MPI_AllReduce function to sync all the distances vector and the changed variable.
• Testing:
  • 1. sequential-0.000317 parallel-0.000396
  • 2. sequential-0.018164 parallel-0.009953
  • 3. sequential-0.369531 parallel-0.156303
  • 4. sequential-0.369531 parallel-0.156303
  • 5. sequential-0.473369 parallel-0.188599
  • 5. sequential-0.447071 parallel-0.257097