You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I'm nearly done fixing #156 and that work has highlighted a shortcoming in how the stand-alone driver reports timing results: the branch allows multiple columns to be split among multiple marbl instances, but the timer summary in the driver was designed with a single instance in mind. I think the proper solution would be to report the sum of each timer over all the instances.
Also, the current MPI implementation duplicates instances on the different tasks. Ideally, the different instances of this new test would be distributed (in which case the timer summary needs to know which instances belong to which task when summarizing the results).
It may make sense to split this issue in two, and handle the MPI work separately from the timers.
The text was updated successfully, but these errors were encountered:
I'm nearly done fixing #156 and that work has highlighted a shortcoming in how the stand-alone driver reports timing results: the branch allows multiple columns to be split among multiple marbl instances, but the timer summary in the driver was designed with a single instance in mind. I think the proper solution would be to report the sum of each timer over all the instances.
Also, the current MPI implementation duplicates instances on the different tasks. Ideally, the different instances of this new test would be distributed (in which case the timer summary needs to know which instances belong to which task when summarizing the results).
It may make sense to split this issue in two, and handle the MPI work separately from the timers.
The text was updated successfully, but these errors were encountered: