Description of changes:https://github.com/awslabs/palace/pull/166 introduced an issue where coefficients with matching attribute lists, might have different size K, if all the attributes are -1. This is a consequence of these lines where a coefficient goes from size zero to size 1, even if there were no local attributes associated. This triggers these lines if attr_mat is all -1, but the incoming coefficient is of size zero. This can occur if one rank does not have any piece of the relevant geometry. The fix is if attr_list is entry in adding a coefficient, to do nothing.
Description of changes: https://github.com/awslabs/palace/pull/166 introduced an issue where coefficients with matching attribute lists, might have different size K, if all the attributes are
-1
. This is a consequence of these lines where a coefficient goes from size zero to size 1, even if there were no local attributes associated. This triggers these lines ifattr_mat
is all -1, but the incoming coefficient is of size zero. This can occur if one rank does not have any piece of the relevant geometry. The fix is ifattr_list
is entry in adding a coefficient, to do nothing.