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
Alberto F. Martin edited this page Jul 2, 2023
·
1 revision
To be curated, linked against the actual software versions, etc.
Notes GridapDistributed
GLOBAL identifiers are such that we first enumerate the identifiers corresponding to entities owned by P0, then the identifiers corresponding to entities owned by P1, and so on, till Pn-1.
The partition corresponding to FE spaces is such that we may have own and ghost LOCAL identifiers interleaved.
The partition corresponding to linear algebra structures, i.e., (row,cols) for a PSparseMatrix, and rows or cols for a PVector, are such that we have FIRST LOCAL owned identifiers followed by ghost identifiers.
The local portion of PSparseMatrix is locally indexed using the local identifiers corresponding to the partition of its rows and cols, i.e., see previous point. As a consequence, during numerical assembly of a PSparseMatrix for which we already have computed its sparsity pattern, we need to transform among local identifiers of the partition corresponding to the trial and test FE spaces, and the linear algebra data structures.