Merged in bugfix/fix_dim_divides_logic (pull request #275)
eceb4bf·Author: David Dickinson·Closed by: David Dickinson·2020-03-05
Description
Add split_nicely calculated from dim_divides
and use this to determine if we are allowed to use sub-communicators. This fixes a bug in the logic used to disable the use of the velocity integration sub-communicators.
Note that length one dimensions will always divide nicely
Try to set properties of dimensions to the left of a dimension that doesn't split nicely
Add a comment detailing potential issue with _local flags.
Provide fix for the potential issue.
Fix for le_lo ?_local variables
This should fix a couple of potential issues with the decomposition logic in g_lo (and le_lo, although this isn’t used). The bug pertains to the variables used to automatically disable the use of sub-communicators in velocity space integrations. The fixed code attempts to be cautious, disabling this if there is any possible doubt.
Add split_nicely calculated from dim_divides
and use this to determine if we are allowed to use sub-communicators. This fixes a bug in the logic used to disable the use of the velocity integration sub-communicators.
Note that length one dimensions will always divide nicely
Try to set properties of dimensions to the left of a dimension that doesn't split nicely
Add a comment detailing potential issue with _local flags.
Provide fix for the potential issue.
Fix for le_lo ?_local variables
This should fix a couple of potential issues with the decomposition logic in g_lo (and le_lo, although this isn’t used). The bug pertains to the variables used to automatically disable the use of sub-communicators in velocity space integrations. The fixed code attempts to be cautious, disabling this if there is any possible doubt.