PCBDDC: avoid possible deadlock

PCBDDCGraphGetDirichletDofs is collective whereas the pressure dofs (if any) are local

PCBDDC: raise error if CSR graph is not consistent

Merge remote-tracking branch 'origin/knepley/feature-snes-deflation'

* origin/knepley/feature-snes-deflation: (55 commits)
Plex: Correct complex problem
Plex ex12: Now PCMG monitor creates its o

Merge remote-tracking branch 'origin/knepley/feature-snes-deflation'

* origin/knepley/feature-snes-deflation: (55 commits)
Plex: Correct complex problem
Plex ex12: Now PCMG monitor creates its own coarse solutions
Plex ex12: Put in working setting for PCMG coarsening test
XDMF: Now take a bunch of files as input
SNES ex12: Use monitors instead of brute force to monitor errors
SNES ex12: Renumber meshes to match MG numbering
PC+SNES: Attach a level id to subsolvers in a multilevel solve - We should have done this years ago
Plex ex12: Added error output on each level
Plex: Added DMPlexComputeL2DiffVec() - This puts cellwise error integrals in a Vec
Plex: Fix variable init
Add an option to ex12 to dump the coarsened hierarchy to disk.
Plex: Fixed unused variables
Plex: Fixed PetscScalar conversion problem
Plex: Fixed function header
Profile DMCreateInterpolation (nontrivial cost for nonnested meshes)
Replace the arguments for coarsening in ex12 with correct ones
Plex: Using a refinement function
Plex: Added DMPlexGet/SetRefinementFunction() - Use to do non-uniform refinement
Events for DMCoarsen and DMLocatePoints
Update the pragmatic build script to use master, not my branch
...

show more ...

PCBDDC: extend benign trick to PCApplyTranspose_BDDC

PCBDDC: make code for checking operator a function

PCBDDC: make code more clear

PCBDDC: move check into separate function

PCBDDC: move code to a separate function

PCBDDC: use consistent name for setting or retrieving P0 values

PCBDDC: remove stupid requirement of nonzero near null space on vertices

Fix PCMGSetLevels() to preserve previously set PCMGType and PCMGCycleType

Fix manual page for option -pc_mg_multiplicative_cycles

Add missing __FUNCT__ define for PCMGGetType()

Fixed girestriction

Changed a little bit

PCBDDC: Enable benign trick even with constraints

The user is responsible to either

1) use an iteration matrix whose basis is already changed to make \int_F v*n a primal dof for all F
or
2) attach

PCBDDC: Enable benign trick even with constraints

The user is responsible to either

1) use an iteration matrix whose basis is already changed to make \int_F v*n a primal dof for all F
or
2) attach a proper MatNullSpace object which guaranties that B0_R is null
or
3) provide a proper change of basis

In cases 1) and 3), adaptive selection of constraints could be used.

show more ...

Merge branch 'barry/optimize-xolotl'

PCBDDC: added TODO

PCBDDC: avoid playing with pointers. Use API instead

PCBDDC: unused variable

PCBDDC: Dirichlet dofs on internal nodes should be imported

this is needed by the benign trick because
b(v_I,p_0) should be zero for all v_I = 0 \Partial\Omega (including Essential boundary conditio

PCBDDC: Dirichlet dofs on internal nodes should be imported

this is needed by the benign trick because
b(v_I,p_0) should be zero for all v_I = 0 \Partial\Omega (including Essential boundary conditions)

show more ...

PCBDDC: remove primal vertices from face/edge deluxe scaling

this does not change the behaviour of standard BDDC.
It is required by the benign trick in order to
keep the iterates in the benign space

PCBDDC: remove primal vertices from face/edge deluxe scaling

this does not change the behaviour of standard BDDC.
It is required by the benign trick in order to
keep the iterates in the benign space

remove extra code to check benign-ness inside PCBDDCMatMult_Private

show more ...

PCBDDC: raise error if benign trick and user defined change of basis are both present

PCBDDC: Wrong update during postsolve and forgot to restore the original local matrix

Now the benign trick works
Still need to do something on the deluxe side

PCBDDC: fix memory leak