Plex ex3: make separate fe and fv tests

Plex ex3: use MPIU_MAX, not MPI_MAX

DMPlex: added DMPlexReconstructGradientsFVM()

Plex: fix typo in change to cohesive cell creation

Plex: add ex4 and ex5 to makefile so they can be used for regression tests

DMPlex: remove overzealous error condition from FV gradient reconstruction

We only care about whether there are enough faces to reconstruct a
gradient if a cell is local: ghost cell gradients are br

DMPlex: remove overzealous error condition from FV gradient reconstruction

We only care about whether there are enough faces to reconstruct a
gradient if a cell is local: ghost cell gradients are broadcast from
their owners anyway, right?

show more ...

DMLabel+Doc: Adding function documentation

DMPlex: correct pointSF size in new point shift approach

I wasn't update the number of roots, getting valgrind errors in parallel

DMPlexMigrate(): use correct global indices when migrating tree

DMPlexMigrate(): Don't invalidate old DM by overwriting cones

Plex ex3: use DMPlexDistributeOverlap()

Is this broken, though? Will test to see.

DMPlexPartitionLabelClosure: traverse the tree up and down

for FV, it's important for children to be reproduced, even if they
aren't referred to by a local cell, because they are needed for flux
cal

DMPlexPartitionLabelClosure: traverse the tree up and down

for FV, it's important for children to be reproduced, even if they
aren't referred to by a local cell, because they are needed for flux
calculations.

show more ...

PetscPartitionerSimple: work in parallel

It looks like PetscPartitionerSimple assumed all the points were
originally on proc 0: this fixes that.

Plex ex3: try to create overlap for FV computations

This isn't working with the simple partitioner right now: it may not
work the way it's supposed to in parallel.

Plex ex3: don't use limiter in gradient test

We're interested in comparing the exact gradients: the current cell
limiter may have some catastrophic cancellation issues(?) that can cause
a false posi

Plex ex3: don't use limiter in gradient test

We're interested in comparing the exact gradients: the current cell
limiter may have some catastrophic cancellation issues(?) that can cause
a false positive (eliminating a correct gradient).

show more ...

Plex ex3: small fix to printf args

Plex ex3: use SNES computed geometry in TestFVGrad

This includes the computed: the geometry returned by
DMPlexComputeGeometryFVM does not.

Plex ex3: print FV grad max frobenius error

PlexTree: protect against truncated sections when computing anchors

I had only tested on sections that included all points before.
Testing on the gradients generated in FVM exposed this.

Plex ex3: try projecting rigid body modes into FVM discretization

These will be used to test cases where gradient reconstruction is exact

Plex Tree: small fixes (bounds, frees) for Tree+FVM

DMPlex Tree: fix DMPlexTreeExchangeSupports()

Handle the case when supports have already been shared.

Merge branch 'master' into barry/fix-petscviewer-attempt-2

Plex: fix DMPlexShiftTree_Internal()

I wasn't keeping track of the difference in size between the old section
and the new section

Plex ex3: use DMPlexConstructGhostCells in TestFVGrad