MatrixFree: Do not construct hanging nodes structure for non-adapted meshes #13872
Conversation
| bool all_cells_on_same_level = true; | ||
| for (const auto &cell : triangulation.active_cell_iterators()) | ||
| if (cell->level() == static_cast<int>(triangulation.n_levels()) - 1) | ||
| break; |
There was a problem hiding this comment.
I don't understand, do you take advantage of the order the cells are traversed here? Also can you wrap the contents of the for loop in { }?
There was a problem hiding this comment.
Yes, I do: I do not want to loop through all cells, since it is enough to complete the loop through all active cells on a coarser level and see if there is one non-artificial. We maybe have an iterator filter, but after looking for 30 seconds I decided I was quicker to implement it. Do you want me to not assume the order of active cells level by level? I could do so by traversing all cells and removing this if clause here, at a probably negligible impact on performance.
kronbichler
force-pushed
the
reduce_cost_hanging_nodes
branch
from
c70745b
to
bedf652
Compare
| // In case all our cells are on the same level, there is nothing to | ||
| // do. We only need to check for active cells (non-artificial) that are | ||
| // on a level coarser than the finest one. | ||
| bool all_cells_on_same_level = true; |
There was a problem hiding this comment.
Why not use Triangulation::has_hanging_nodes()?
There was a problem hiding this comment.
I don't want to go through the communication involved there, and wanted it to be a local operation.
There was a problem hiding this comment.
Wouldn't it be better to move the information to number_cache?
There was a problem hiding this comment.
At some point yes, but for the current release it feels too unsafe to touch that part. But let me use a similar code as the has_hanging_nodes function, to reduce the variability we have in the code base.
There was a problem hiding this comment.
I now pushed the update.
kronbichler
force-pushed
the
reduce_cost_hanging_nodes
branch
from
bedf652
to
d584927
Compare
This reduces the initialization cost of a data structure that is added with this release, so I would like to tag it with 9.4. Note that this code is well-tested in around a hundred of tests for matrix-free computations.