PCBDDC: fix from nightly tests

PCBDDC: use the original local discrete gradient to compute extrow, extcols and check for endpoints

PCBDDC: remove unneeded variables from PCSetUp_BDDC

PCBDDC: added field specification to the API

PCBDDC: bugfix for the nonconforming case

PCBDDC: added support when the Nedelec dofs are a subset of all the dofs

PCBDDC: no need to set freecsr to false during graph init

PCBDDC: added special case for graph analysis

PCBDDC: change of basis for Nedelec should be computed when recompute_topography is true

PCBDDC: return if size of edge is zero

PCBDDC: fix man page

PCBDDC: test for out-of-memory

PCBDDC: remove unused variable

PCBDDC: it may happen that GetRowIJ is called on an empty matrix

PCBDDC: added support for variable order edge elements

PCBDDC: raise error if the factorization fails

Microsoft compilers cannot caste from double to enum directly

Time: 0 hours
Reported-by: nightly tests

Windows compiler won't allow cast from double to enum

Time: 0 hours
Reported-by: nightly tests

refactor common setter/getter methods for PCFACTOR class into common shared code

Time: 2 hours

Extract the submatrices

Extract the submatrices from the preconditioning matrix instead of the Jacobian matrix.
The reason for this is that the MatMults for the submatrices (at line 1004, 2035 and 1

Extract the submatrices

Extract the submatrices from the preconditioning matrix instead of the Jacobian matrix.
The reason for this is that the MatMults for the submatrices (at line 1004, 2035 and 1054 of fieldsplit.c)
involves function evaluations (which is not cheap) when we are using the matrix-free method for inexact Newton.
I have some tests in MOOSE, and this change halves the compute time.

show more ...

Merge branch 'maint'

Added PCFactorGetZeroPivot(), PCFactorGetShiftType(), and PCFactorGetShiftAmount()

Time: .25 hours
Reported-by: Glenn E Hammond <gehammo@sandia.gov>

Merge branch 'barry/add-matshell-matgetdiagonalblock-fortran'

PCBDDC: merge two zerorows steps and add runtime check for the number of coarse edges' endpoints

fix issue with mkl 2016.3.210 in PCBDDCConstraintsSetUp

Removed the use of PetscObjectQueryFunction() with with MatGetDiagonalBlock()

that I missed when I first converted MatGetDiagonalBlock() from a query function
to a table function

Time: .3 hours
Re

Removed the use of PetscObjectQueryFunction() with with MatGetDiagonalBlock()

that I missed when I first converted MatGetDiagonalBlock() from a query function
to a table function

Time: .3 hours
Reported-by: nightly tests

show more ...