Merge branch 'barry/propagate-pcsetup-failures'

removed KSPSetNullSpace() since mat already carries that information
and having two locations for it is confusing, unnecessary and prone to error
added MatSetTransposeNullSpace() for providing the nu

removed KSPSetNullSpace() since mat already carries that information
and having two locations for it is confusing, unnecessary and prone to error
added MatSetTransposeNullSpace() for providing the null space of the transpose of the matrix and added support in KSPSolve() to remove this null space from the right handside of the linear system thus making the system consistent.

show more ...

PCBDDC: add output for psi basis functions

PCBDDC: rearrange symmetry, positive-definitess stuffs

Added pc_bddc_symmetric command line options to shift to unsymmetric basis computation

PCBDDC: no need for eigensolver is S == St

PCBDDC: complete support for MUMPS Schur solver reusage

is_R and is_B should be created during SubSchursSetUp
Added preprocessor guards

PCBDDC: added support for transpose application of the BDDC when reusing MUMPS solver

PCBDDC: reuse Schur solver provided by MUMPS when possible

MATIS: n_sends should stay there!!!!

if n_sends is not part of the logic, we cannot overlap levels!

PCBDDC: optimize for reusage of Schur solver (if any)

R_to_B and (optionally) R_to_D scatters are different
reusage is not possible when change of basis has been requested

Infrastructure that allows failures in PCSetUp(), PCApply(), MatMult() etc due to, for example a zero pivot or function evaluation outside its domain, to propagate up and become a KSP_DIVERGED inste

Infrastructure that allows failures in PCSetUp(), PCApply(), MatMult() etc due to, for example a zero pivot or function evaluation outside its domain, to propagate up and become a KSP_DIVERGED instead of generating an error that stops the program. In response to Issue 96. This includes failures in MatCreateSNESMF() applications due to domain errors. The mechanism to propagate some errors is by setting Info or Nan into the output vector and using the norm or inner product reductions in SNES or KSP to propagate the error condition to all processes and then handling them immediately after the norm or inner product.

This allows, for example, ODE integrators to try again with a smaller time-step if the PCSetUp failed instead of requiring a complete restart of the run with other options.

Currently some error conditions, such as function domain error in a line search may not get progated up using the correct SNESConvergedReason

See src/snes/examples/tests/ex69.c for the handling of several conditions

show more ...

Merge branch 'master' into knepley/feature-composable-callbacks

* master: (151 commits)
PetscMalloc: Fix handling of zero-sized allocations
add note to DMGetLocal/GlobalVector() about intended s

Merge branch 'master' into knepley/feature-composable-callbacks

* master: (151 commits)
PetscMalloc: Fix handling of zero-sized allocations
add note to DMGetLocal/GlobalVector() about intended short-time usage of the vectors.
sf: use patched openmpi tarball that has a potential fix for sf example hang
SNES problems that set VI bounds must use a VI solver since the other solvers ignore the bounds
pounders viewer did not cast PetscReal to double in printf nor use %D instead of %d
fix ts ex20adj
fix ts theta
fixed another variable declaration in the middle of a code block
example main() had PetscInt return type and argument that should be int
DMDAVTKWriteAll_VTS: Fix handling of higher-dimension coordinates
fixed example that declared variables in the middle of a function
C programs do not return PetscInt nor do they take an argument that is PetscInt, fixed example that did
You cannot declare variables in the middle of functions in PETSc.
add ierr checking to PetscOptionsBegin/End() in example
configure: add error check if user specifies PETSC_ARCH=-dbg [i.e starts with a '-']
scale norm of vector difference by norm of vector to insure it is small enough to consider the two vectors equal
missing PetscViewerDestroy() added
MatNestGetISs: use custom fortranstub to handle PETSC_NULL_OBJECT
update location of open-mx website on FAQ page
If the user sets bounds and calls a solver that does not support them, throw an error.
...

show more ...

PCBDDC: reuse as much as possible the MUMPS Schur solver

optimizations still need to be applied

PCBDDC: bugfixes

- change of basis falg should be set for each cc (not primal constraint)
- work should be set to zero when importing boundary values of A_RRmA_RV to avoid undefined values
- A_VV sh

PCBDDC: bugfixes

- change of basis falg should be set for each cc (not primal constraint)
- work should be set to zero when importing boundary values of A_RRmA_RV to avoid undefined values
- A_VV should be converted to BAIJ when working with SeqBAIJ

show more ...

PCBDDC: reduce memory footprint of BDDC application

The previous code was designed without adaptivity in the picture
pcbddc->local_auxmat1/2 should act on boundary dofs only
except when the static s

PCBDDC: reduce memory footprint of BDDC application

The previous code was designed without adaptivity in the picture
pcbddc->local_auxmat1/2 should act on boundary dofs only
except when the static switch has been selected: in this case pcbddc->local_auxmat2 should act on the whole set of R dofs

tested on linear spd systems and with NKS with bs 3

show more ...

PCBDDC: scale a smaller matrix

PCBDDC; improved numbering of coarse dofs when many constraints are presents per connected component

PCBDDC: merge together two loops

PCBDDC: rewrite code for constraints setup

the code is now simpler and more general. It will also allow to renumber coarse dofs clustering them together with respect to connected components

PCBDDC: remove unneeded variables from pcbddc data structure

PCBDDC: nullspace adaptation of change of basis enabled only in debug mode

PCBDDC: aux_primal_numbering is no longer needed

PCBDDC: remove unneeded code

PCBDDC: get rid of no longer needed sorts

PCBDDC: get rid of PCBDDCGetPrimalVerticesLocalIdxs and PCBDDCGetPrimalConstraintsLocalIdxs

these functions are no longer needed