-
Notifications
You must be signed in to change notification settings - Fork 1
Home
Zach Byerly edited this page May 20, 2017
·
5 revisions
Our current goal is to troubleshoot performance problems with the dgswem-hpx branch of the code.
Max and Zach have demonstrated that, when using Intel compilers, the Fortran sections of both codes perform the the same.
Zach is currently:
- Attempting to get dgswem-hpx running on Rostam (Ste||ar group's research cluster) with VTune support enabled.
- Determining the distribution of submeshes on localities by LibGeoDecomp's PTScotch partitioner
- Refactoring his c++ based hpx dataflow code (runs only on single locality, does not use LGD)
- Making sure there isn't anything special about the artificial square meshes that could cause performance problems