Skip to content

Implement MPI all-to-all at start and end of each iteration

Max Holicki requested to merge ImplementAll2AllAtStartAndEndOfIteration into master

Goals:

  1. Finish implementing MPI all-to-all. The implementation now runs an all-to-all test at the beginning of each iteration and another at the end of each iteration. Bandwindths are calculated based on the total equivalent point-to-point messages sent, i.e. message_length*number_of_tasks*(number_of_tasks-1)/all-to-all_time.
  2. Remove serial retest output if the number of serial retests was set to zero. Previously setting the number of serial retests to zero resulted in NAN outputs in the log file.

Merge request reports