Ned Coltman authored 1 year ago and Mathis Kelm committed 1 year ago

[changelog][staggered][fixes] Update changelog with a description of the changes to the old staggered made in !3572

[pdesolver] Add apply interface that returns converged result

See merge request !3647

[subgrid] Impove error message when buffer size doesn't match

See merge request !3656

[fcstaggered][elemvolvars] Use a local estimate for the size of boundary volvars

See merge request !3655

[cmake] do not duplicate threading backend selector message

See merge request !3653

[codespell] Fix spelling mistakes codespell 2.2.6

See merge request !3650

[experimental] Multistage assembler

See merge request !3608

Timo Koch authored 1 year ago and Dennis Gläser committed 1 year ago

Reference solution is coming from the previous fix branch.
It is slightly different that the current solution, however,
with the standard assembler 20 Newton iterations where necessary
because the approximation of the Jacobian was bad.
The el2p should therefore only be used with the experimental assembler.

Timo Koch authored 1 year ago and Dennis Gläser committed 1 year ago

The steps were
* Add multistage assemblers draft
* Combine residual and jacobian step
* Move computation of split residuals into the localassembler
* Remove isImplicit and cleanup
* Separate local assembler implementations from old impl
* simplify Dirichlet/constraints
* Get rid of implicit template argument
The decision can be made at runtime.

In LocalAssembler: deduce LocalResidual from assembler

See merge request !3643