Supercomputer Applications
MPI Projects Part 1, Spring 2006

Message Passing Interface Forum, MPI Standard (1995), MPI routines, Constants for C and Fortran

MPI Routines

1. Example programs:
   • Hello world
     • helloWorldMPI.c (C), helloWorldF.f90 (Fortran) - Prints hello from all processors
     • helloWorld2.c (C), helloWorld2F.f90 (Fortran) - Same as above, but includes printing how many processors are involved.
   • Sending values between processors
     • passOneMPI.c (C), passOneF.f90 (Fortran) - sends PI from processor 1 to processor 0 (the root process)
     • arrayMPI.c (C), arrayMPIF.f90 (Fortran) - creates an array in processor 1; sends the entire array to processor 0
   • "Collective" operations - BCAST and REDUCE
     • bcast.c (C), bcastF.f90 (Fortran) - "Broadcasts" pi from the root process to all other processors
     • reduce.c (C), reduceF.f90 (Fortran) - Collects pi from all processors; sums up all the pi values in the root process.

2. MPI Program Lab05, "HelloCluster.c"
   MPI_Init(), MPI_Comm_size() , MPI_Comm_rank(), MPI_Finalize() MPI_COMM_WORLD (scroll down to "Communicators")
   Lab 05 Report form, .doc

3. MPI Program Lab06, share.c
   --Input a number with scanf, broadcast this value to other "nodes"/processes
   uses:
   MPI_Send(), and MPI_Recv() (Vers. 1)
   MPI_Bcast() (Vers. 2)
   also see MPI Data types
   NOTE: Place the call to MPI_Bcast outside of the "if (rank == 0)" block:

            if (rank == 0) {
   	      scanf( ... );
   	  }
   	  MPI_Bcast( ... );  //MPI_Bcast is like a combination send/receive,
   	                     //so each process needs it
   	

4. MPI Program Lab07, DinnerParty.c
   Use MPI_Send(), and MPI_Recv(), INSTEAD OF MPI_Bcast()
   Sample Output (with six processes):
   mpirun -np 6 lab02
   Process 0, Sending out random int 406999 to process 1
   Process 2 received: 1252750, mysalary=380245, sending out 1632995 to process 3
   Process 3 received: 1632995, mysalary=600007, sending out 2233002 to process 4
   Process 1 received: 406999, mysalary=845751, sending out 1252750 to process 2
   Process 0 received 3014004, mySalary=157838, average salary=$460807.17
   Process 4 received: 2233002, mysalary=618007, sending out 2851009 to process 5
   Process 5 received: 2851009, mysalary=162995, sending out 3014004 to process 0

5. MPI Program Lab08, IntervalSlicing.c - Calculate Pi,
   --parallelize the program "gregory.c"
   For an example of MPI_Reduce(), see reduce.c (C), reduceF.f90 (Fortran) - Collects pi from all processors; sums up all the pi values in the root process.
   Use MPI_Reduce() to sum data that is local to each process

6. MPI Program Lab09, Parallel Programming Application - Array Search
   --a parallelized search of a large array for a target value

7. MPI Resources and Sample Programs
   • Introductory Slides on MPI, by William Gropp
   • A Short Overview of MPI Concepts, by William Gropp
   • Intermediate Level MPI programming