search

conservation-laws / ryujin

High-performance high-order finite element solver for hyperbolic conservation equations
https://conservation-laws.org
Other
103 stars 25 forks source link

`SparsityPatternSIMD`: Bugfix: make MPI rank matching more robust #138

Closed tamiko closed 2 months ago

tamiko commented 2 months ago

With locally refined meshes and hanging node constraints we might end up in a situation where an MPI rank in the import_targets set does not show up in the ghost_targets set. This is not a problem as we "minimize" the sparsity pattern in (ghosted) locally owned rows to only contain entries that we have in the locally relevant ranges of the tansposed locally owned rows.

BUT, for the next MPI rank we have to try to match up from index 0 again.

@kronbichler ping

Closes #136

kronbichler commented 2 months ago

Great that you found this bug!