FS#293 Get rid of "self-written" parallel linear solvers
Metadata
Property | Value |
---|---|
Project | dumux |
Category | General |
Reported by | Bernd Flemisch (bernd@iws.uni-stuttgart.de) |
Reported at | Sep 14, 2015 10:28 |
Type | Feature Request |
Version | Git |
Last edited by | Bernd Flemisch (bernd@iws.uni-stuttgart.de) |
Last edited at | Dec 9, 2015 15:50 |
Closed by | Bernd Flemisch (bernd@iws.uni-stuttgart.de) |
Closed at | Dec 9, 2015 15:50 |
Closed in version | 2.9 |
Resolution | Implemented |
Description
In dumux/linear/boxlinearsolver.hh, there are several linear solver backends which use "self-written" infrastructure for parallel runs from the files borderindex.hh, domesticoverlapfrombcrsmatrix.hh, elementborderlistfromgrid.hh, foreignoverlapfrombcrsmatrix.hh, globalindices.hh, overlapping...hh, vectorexchange.hh, vertexborderlistfromgrid.hh.
Since the parallelism is restricted to box, the iteration counts deteriorate with growing process numbers, the code is hard to maintain or to generalize, and we have the general / rather robust / mostly upstream-maintained AMGBackend, I propose to get rid of boxlinearsolver.hh and the infrastructure files.