PCBDDC: it does not have any sense to require adaptivity just for edges.

less stringent hermitian test in sub_schurs code

PCBDDC: rearrange symmetry, positive-definitess stuffs

Added pc_bddc_symmetric command line options to shift to unsymmetric basis computation

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

PCBDDC: Implement caching strategy for dirichlet dofs in graph

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

PCBDDC: remove unneeded variables from pcbddc data structure

PCBDDC: sorting of connected components is performed only when it is really needed

PCBDDC: avoid unneeded computations for custom primal vertices (if any)

PCBDDC: optimized eigenvalue problem

explicit inversions of sum(S_E) and sum(St_E) blocks have been removed and the eigenvalue problem has been recasted
remove unneded transpositions for the symmetr

PCBDDC: optimized eigenvalue problem

explicit inversions of sum(S_E) and sum(St_E) blocks have been removed and the eigenvalue problem has been recasted
remove unneded transpositions for the symmetric case

show more ...

PCBDDC: get rid of invert_stildas branch

PCBDDC: wrong logic

PCBDDC: reuse interior solver computed by MUMPS when computing the Schur complement

PCBDDC: add option to select faster application of deluxe scaling

QR factorization of constraints resulting in a unitary transformation is mandatory if a change of basis is requested with faster del

PCBDDC: add option to select faster application of deluxe scaling

QR factorization of constraints resulting in a unitary transformation is mandatory if a change of basis is requested with faster deluxe

show more ...

Merge branch 'master' into tisaac/dmforest

If we're going to start developing on this again, the recent changes to the
location of configuration files will make it annoying to switch to and from
thi

Merge branch 'master' into tisaac/dmforest

If we're going to start developing on this again, the recent changes to the
location of configuration files will make it annoying to switch to and from
this branch. Matt's and my fixes in DMPlex will also be relevant.

* master: (1557 commits)
MatLoad_MPIDense() always allocated array space even if it was previously provided
Clarified lumping options for PCFieldSplitSchurPreType SELFP.
fixed up PetscInfo in GAMG
TSErrorWeightedNorm[2|Infinity]: API changes
MatLoad_SeqDense() always allocated new space for the data even if the user already provided it
Fixed DMCompositeGetAccessArray() to support readonly vectors
Fix obviously-wrong aspects of documentation on SNESVINEWTONRSLS and SNESVINEWTONSSLS.
fixed example to compile - demonstrates a bug
var name 'len' was F90 key word
For user provided Jacobian functions they should always assembly into the second matrix, the one from which the preconditioner is built because the first may be used as matrix-free with -snes_mf_operator
Suggest --with-64-bit-indices if users mistakenly use --with-64bit-indices
Fix configure logging of integer size.
missing underscore in #define of vecgetarrayread translation of fortran stub for VECGETARRAYREAD
updated test output files to reflect differences introduced previously in PCView output
-ksp_type preonly does not support -ksp_monitor_short so remove from test line to prevent extranious warning output
missing CHKERRQ; and ugly double declaration of ierr
Plex ex3: For want of a semicolon, the kingdom was lost
MatSetValuesMPISBAIJ will now print correct global indices when error generated due to new nonzero location
completed basic use of MatSetValuesBlocked_SeqBAIJ_Inlined() by MatSetValuesBlocked_MPIBAIJ() to allow proper global indices being printed when error generated by new nonzero locations.
Added inline version of MatSetValuesBlocked_SeqBAIJ() to be used by MatSetValuesBlocked_MPIBAIJ() to print error message for nonzero block with global indices
...

show more ...

Merge branch 'master' into shri/ts-is-for-differential-variables

Conflicts:
include/petsc-private/tsimpl.h
src/ts/impls/eimex/eimex.c
src/ts/interface/ts.c

PCBDDC: get rid of parallel application of deluxe scaling

code has been greatly simplified
introduced subcommunicators in PCBDDCSubSchursSetUp to improve performances in multilevel cases
adaptivity

PCBDDC: get rid of parallel application of deluxe scaling

code has been greatly simplified
introduced subcommunicators in PCBDDCSubSchursSetUp to improve performances in multilevel cases
adaptivity is supported only in combination with deluxe scaling

show more ...

Merge branch 'master' into jed/chem
Note merging a branch with good stuff in it for a year and a half is insane!

Conflicts:
src/ts/examples/tutorials/makefile
src/ts/interface/ts.c

PCBDDC: remove unneeded code

PCBDDC: if rhs is changed, then the starting vector is zero

PCBDDC: bug fixes for multilevel

hermitian flag is now set to true if the matrix is declared symmetric in real case
This should be handled by PETSc!

PCBDDC: optimize PCApply_BDDC and PCApplyTranspose_BDDC

Merge branch 'master' into stefano_zampini/pcbddc-adaptive-selection

Conflicts:
src/ksp/pc/impls/bddc/bddcprivate.c

PCBDDC: fixes for complex builds

PCBDDC: fix PCSetUp_BDDC

PCBDDC: optimized PCBDDCScatterCoarseDataBeing/End

some fixes