PetscSF: Incorrect check for computed degrees

Merge branch 'barry/fix-dmlocaltoglobal_da'

remove unneed uses of local DM vectors and DMLocalToGlobalBegin/End in examples
removed to prevent cut and paste repeats of their use

VecGetSubVector: for non-contiguous index sets, compose scatter with subvector

This avoids an extra synchronization to check for contiguity and the
overhead of re-creating the VecScatter. The memor

VecGetSubVector: for non-contiguous index sets, compose scatter with subvector

This avoids an extra synchronization to check for contiguity and the
overhead of re-creating the VecScatter. The memory for the VecScatter
is resident for the life of the subvector, but I consider this to be a
minor overhead compared to the extra synchronizations.

show more ...

VecGetSubVector: simplify state handling

My more complicated and incorrectly-implemented state-tracking idea does
not seem necessary, so simplify state tracking to avoid confusion.
Thanks to Patrick

VecGetSubVector: simplify state handling

My more complicated and incorrectly-implemented state-tracking idea does
not seem necessary, so simplify state tracking to avoid confusion.
Thanks to Patrick Lacasse for pointing out the mistake.

show more ...

PetscSF: Fixes for sparse leaves
- When sf->mine exists, the leaves can be numbered non-sequentially. This causes allocation based upon sf->nleaves to underallocate.
Replacing sf->nleaves with the ma

PetscSF: Fixes for sparse leaves
- When sf->mine exists, the leaves can be numbered non-sequentially. This causes allocation based upon sf->nleaves to underallocate.
Replacing sf->nleaves with the maximum leaf number is sufficient in this case.

show more ...

PetscSection: Added PetscSectionCopy()

SF: fix recent changes that switched nroots to nleaves.

This caused lots of tests to fail with sefaults.

Merge branch 'knepley/feature-parallel-partition' into mlange/feature-parallel-distribute

Conflicts:
include/petscdmplex.h

documentation in VecView() and VecLoad() about handling the vector blocksizes with vector prefixes

Reported-by: Ed Bueler <elbueler@alaska.edu>

Merge branch 'maint'

Conflicts:
config/BuildSystem/config/packages/MPI.py

error when VecView_Seq_Binary writes out vector blocksize, it appended an incorrect _ after the prefix
it is up to the prefix to have an underscore or not

PetscSF: Fix ComputeDegree and GetMultiSF for sparse SFs

Merge branch 'tisaac/thplex'

This merge brings in related work in the following areas:
- Added a local constraint matrix to DM (see DMSetDefaultConstraints())
- Added point-to-anchor-point constrain

Merge branch 'tisaac/thplex'

This merge brings in related work in the following areas:
- Added a local constraint matrix to DM (see DMSetDefaultConstraints())
- Added point-to-anchor-point constraints in DMPlex (see
DMPlexSetAnchors()), which build on the DM constraints
- Added hierarchically-constrained points to DMPlex (see
DMPlexSetReferenceTree(), DMPlexSetTree()), which builds on the anchor
constraints
- Tests for these constraints in src/dm/impls/plex/examples/ex3.c (try
runex3_* examples in that directory)
- Added quadrature order (see PetscQuadratureGet/SetOrder()) (work by
Matt Knepley)
- Added default return values for DMGetCoordinateDim() (defaults to
DMGetDimension())

Conflicts:
include/petscdmplex.h
src/dm/dt/interface/dt.c
src/dm/impls/plex/examples/tests/ex3.c
src/dm/impls/plex/makefile
src/dm/impls/plex/plexfem.c

show more ...

Merge branch 'sarich/fix-vecmedian-restore'

Merge branch 'sarich/fix-vecmedian-restore' into maint

fix vecmedian restore parameter

VecCUSPCopyFromGPUSome should not mark host pointer as up to date

because it only fetches part of the vector from the GPU.

Ensure that host array is allocated.

Emit error message if scatter type cannot be determined.

Fix call to VecScatterCUSPIndicesCreate_StoS.

fslots and tslots where in wrong order. Also fixed confusing variable
name that probably lead to wrong argument order.

With this change runex1_cusp1 i

Fix call to VecScatterCUSPIndicesCreate_StoS.

fslots and tslots where in wrong order. Also fixed confusing variable
name that probably lead to wrong argument order.

With this change runex1_cusp1 is valgrind clean.

show more ...

Merge branch 'dmeiser/fix-scatterinitcusp'

dmeiser/petsc/fix-scatterinitcusp: Check whether VecScatter is
sequential in VecScatterInitializeForGPU. This fixes a segfault
which occurs when VecScatter

Merge branch 'dmeiser/fix-scatterinitcusp'

dmeiser/petsc/fix-scatterinitcusp: Check whether VecScatter is
sequential in VecScatterInitializeForGPU. This fixes a segfault
which occurs when VecScatters are used with mpicusp vectors with
1 MPI process.

show more ...

Merge branch 'jose/split-reduction'

Check whether VecScatter is sequential in VecScatterInitializeForGPU.

Merge branch 'fix-scattertype'