Merge branch 'maint'

Merged in stefano_zampini/fix-bddc-p4est (pull request #1625)

PCBDDC: fix uninitialized valgrind error

Approved-by: BarryFSmith <bsmith@mcs.anl.gov>

PCHypre: simplify near null space reset logic; set n_hmnull=0

PCHypre: fix BoomerAMG memory leak for multiple setups with near null spaces

Commit-type: bug-fix
Reported-by: Elias Karabelas <elias.karabelas@medunigraz.at>

Remove dead code. Subclasses of factor simply called parent method; remove unneeded subclass function

Commit-type: style-fix

Development Tools: Vim, Emacs, Eclipse

PCBDDC: fix uninitialized valgrind error

With polynomial orders > 1, some of the local dofs encoded
in the subdomain section were not listed by the subdomain detection, because
they were associated

PCBDDC: fix uninitialized valgrind error

With polynomial orders > 1, some of the local dofs encoded
in the subdomain section were not listed by the subdomain detection, because
they were associated to parent points.


==24703== Conditional jump or move depends on uninitialised value(s)
==24703== at 0x584A034: PCBDDCGraphComputeCC_Private (bddcgraph.c:727)
==24703== by 0x584A034: PCBDDCGraphComputeConnectedComponentsLocal (bddcgraph.c:803)
==24703== by 0x584A344: PCBDDCGraphComputeConnectedComponents (bddcgraph.c:300)
==24703== by 0x581D7D5: PCBDDCAnalyzeInterface (bddcprivate.c:7137)
==24703== by 0x57DB7A9: PCSetUp_BDDC (bddc.c:1702)
==24703== by 0x59670D3: PCSetUp (precon.c:932)
==24703== by 0x5A1B07D: KSPSetUp (itfunc.c:391)
==24703== by 0x5A1B90C: KSPSolve (itfunc.c:725)
==24703== by 0x5ACCB0D: SNESSolve_NEWTONLS (ls.c:225)
==24703== by 0x5A925C3: SNESSolve (snes.c:4563)
==24703== by 0x40AFBF: main (ex12.c:1062)

Reported-by: Satish Balay <balay@mcs.anl.gov>

show more ...

Use MatDenseGetArrayRead() rather than MatDenseGetArray() when possible

Commit-type: style-fix, feature

Deprecate old {Mat|PC|KSP}ComputeExplicitOperator in favor of {Mat|PC|KSP}ComputeOperator

{Mat|PC|KSP}ComputeExplicitOperator: add extra argument for matrix type

Add new function PCFieldSplitGetISByIndex()

move check whether there are any cells into PCPatchPrecomputePatchTensors_Private

Don't attempt to do precomputation of patch tensors if there are no cells to assemble over

MG: If smoothu is not the same as smoothd, give it the DM also

fix missing CHKERRQ

doc: fixes

Merged in mapdes/petsc/wence/fix/snes-patch-zero-matrix (pull request #1459)

Fix semantics of Patch callback functions

Approved-by: BarryFSmith <bsmith@mcs.anl.gov>

PATCH: Also zero residual in ComputeFunction_Internal

The docs say that we do this, but we didn't.

PATCH: Zero patch matrix on entry to ComputeOperator_Internal

Reduces number of places we need to remember to do this.

Revert the function name where it is referenced

Merge branch 'pr1417/wence/feature-patch-all-at-once/master' [PR #1417]

* pr1417/wence/feature-patch-all-at-once/master:
Some new features in PCPatch: interior face terms, and an option for "all at

Merge branch 'pr1417/wence/feature-patch-all-at-once/master' [PR #1417]

* pr1417/wence/feature-patch-all-at-once/master:
Some new features in PCPatch: interior face terms, and an option for "all at once" assembly.

This adds two new features to PCPatch:

You can now also set callbacks for interior face terms. We need these for problems with IP discretisations.
(Optionally, defaulting to false) do an “all at once” computation of all the element tensors. In our benchmarking this is quite a bit faster (problem dependent).

We have, once again, not hooked up the new callbacks with Plex (there is a TODO + SETERRQ in this case): hopefully it does not require much.
As a result, tests are in Firedrake again: https://github.com/firedrakeproject/firedrake/pull/1383

show more ...

Merge remote-tracking branch 'origin/jczhang/vecscat-in-sf' [PR #1396]

Merged in knepley/feature-move-adj-to-dm (pull request #1424)

Knepley/feature move adj to dm

Merged in stefano_zampini/feature-pcbddc-enablelor (pull request #1427)

Stefano zampini/feature pcbddc enablelor

Use new vector read-lock function names

The new one are VecLockReadPush, VecLockReadPop, VecSetErrorIfLocked

PATCH: More memory cleanup