Version 6.29.7. Previously, if a dataset was created with rotate or average in parallel the initial trajectory processing run would be fine, but if that data was to be used in a subsequent trajectory processing run that would fail. This is because data sets are synced to master for output, but not broadcast back out to non-master ranks. This PR fixes this behavior for rotate and average. Also adds a test to ensure this works properly for rotate. Should address #1096.
Under the hood, a bunch of MPI broadcast routines have been added to almost all of the DataSet classes. This is to eventually enable syncing datasets across all ranks for running commands like filter in parallel (some time in the future).
Version 6.29.7. Previously, if a dataset was created with
rotateoraveragein parallel the initial trajectory processing run would be fine, but if that data was to be used in a subsequent trajectory processing run that would fail. This is because data sets are synced to master for output, but not broadcast back out to non-master ranks. This PR fixes this behavior forrotateandaverage. Also adds a test to ensure this works properly forrotate. Should address #1096.Under the hood, a bunch of MPI broadcast routines have been added to almost all of the DataSet classes. This is to eventually enable syncing datasets across all ranks for running commands like
filterin parallel (some time in the future).