some comments

use a preallocator to build the adjacency matrix

undo some empty lines

printf to a viewer

updated test output

use PetscMemcpy instead of manually doing the loop

get rid of old comments, undo some debugging changes

added test for point rebalancing

specify which type of entity to balance (e.g. 0 => vertices, 1 => edges in 2D)

no more warnings or valgrind complaints

trying to be a good citizen by following the style guide

clean up and comments

add ability to perform the rebalancing in serial, add timings

cleanup, debug output

WIP: build full connectivity between vertices on parallel interface

try assembling the graph via buiding the adjacency matrix, takes care of
some of the communication for us

move things to plexpartition and use ptscotch

add option to use current parallel distribution as initial guess

Balance seems to improve significantly if we sort the adjacency list

a few comments

first go at C implementation of SF rebalancing

Revert "Merge branch 'pr1451/hmorgan/fix-GLL-quadrature/master' [PR #1451]"

This reverts commit 4d195b5eb92504439ee3a75f081f195aff73b70e, reversing
changes made to 57cf99f718c63d6d4dd898d489c9b5204d

Revert "Merge branch 'pr1451/hmorgan/fix-GLL-quadrature/master' [PR #1451]"

This reverts commit 4d195b5eb92504439ee3a75f081f195aff73b70e, reversing
changes made to 57cf99f718c63d6d4dd898d489c9b5204d7f7338.

Accidentally merged to `master` instead of `next`.
If tests turn out okay, this revert will be reverted.

show more ...

DM: Expose creation functions for subdm and superdm from a Section

DMPlex: Speedup DMPlexPartitionLabelInvert()

Setting up a process SF with dense all-to-all ran graph is not scalable,
so try to use plain MPI All-to-All.

DMPlex: Faster construction of the partition label