Collective global computation routines

Global Computation Routines

• The function passed to a global computation routine is either:
  • a predefined MPI function, e.g., mpi_sum.
  • a user supplied function
• A handle to the user supplied function is created using:

  mpi_op_create (function, commute, op, ierr)

• Three versions of reduce that return results to:
  • a single process
  • all processes
  • scatter a vector of results across all processes
• Segmented scans can be done by creating a subgroup for each segment.
• There are 4 global computation routines, plus 2 auxiliary routines (mpi_op_create and mpi_op_free).