Merge branch 'hzhang/fix-getsubmats-reuse'

cleanup

bugfix for memory error as reported by nightly tests

fix warnings reported by nightly tests

fixbug for MPIBAIJ

reorganize/rename MatDestroy_xxx_SubMatrices

bugfix

bugfix for matgetmatrices_

fix memory leak of runex56_2

add MatDestroySubMatrices(); bugfix

bugfix MatGetSubMatricesMPI_MPIXAIJ()

add MATDUMMY for saving matrix reuse data

MatGetSubMatrices_MPIAIJ() works for nis=0 and max_no=0

MatAIJ KLU: remove redefinition of SuiteSparse_long

SuiteSparse_config.h has it's own logic for determining what SuiteSparse_long should be,
and provides a mechanism for overriding it, but that shou

MatAIJ KLU: remove redefinition of SuiteSparse_long

SuiteSparse_config.h has it's own logic for determining what SuiteSparse_long should be,
and provides a mechanism for overriding it, but that should be done at configure time,
not in a source file. Dictating causes compliance errors, even when
is the same size.

show more ...

bugfix: enable nis=0 for MatGetSubMatrices_MPIAIJ

Merge remote-tracking branch 'origin/hzhang/fix-matcreate-null-a'

bugfix for '-mat_view' when matvalue array = null
Reported-by: Pierre Jolivet <pierre.jolivet@enseeiht.fr>

Merge branch 'hzhang/opt-matmatmult-nonscalable'

The developer manual clearly states that MPI_Comm_size() returns size it doesn't return
numProcs or other nonsense. Pretty soon you'll start using mype instead of rank?

Using consistent names for va

The developer manual clearly states that MPI_Comm_size() returns size it doesn't return
numProcs or other nonsense. Pretty soon you'll start using mype instead of rank?

Using consistent names for variables makes code simpler to maintain

show more ...

may switch to scalable only when B->cmap->N > 100000

Merge branch 'hzhang/opt-matgetsubmatrices-mpibaij'

set default algorithm based on B->cmap->N for MatMatMult_MPIAIJ_MPIAIJ()

fix divide by zero exposed by 8534686b5096cd4c91a8f3895408b386c0628431 - and produced by petsc4py testsuite

Always check for missing rows that are being zeroed

Thanks-to: Jan Blechta

optimize nonscalable MatMatMult for mpiaij matrix