Merge branch 'hzhang/mataxpy-subsetcleanup'

Merge branch 'barry/add-some-matsolveforward-back'

Fix buffer overrun in threaded MatZeroEntries_SeqBAIJ

rmap->trstarts counts rows in the matrix, but a->i only has nrows/bs
entries. Fix start and end indexing by dividing by the block size.

While

Fix buffer overrun in threaded MatZeroEntries_SeqBAIJ

rmap->trstarts counts rows in the matrix, but a->i only has nrows/bs
entries. Fix start and end indexing by dividing by the block size.

While we're here, update the sizeof call to use MatScalar rather than
PetscScalar.

show more ...

Merge branch 'barry/fix-options-default'

Conflicts:
src/ksp/ksp/examples/tutorials/ex52.c

rm xtoy, XtoY, previously used for MatAXPY, case SUBSET; add a test contributed by Jose Roman

bugfix for MatGetRedundantMatrix() with user provided subcomm

rm psubcomm from Mat_Redundant

Added MatForward/BackwardSolve for SeqBAIJ block size 1-3 natural ordering
Added testing for them in already existing test code

update use of PetscOptionsBool() to not rely on default setting automatically

cleanup of use of PetscOptionsInt and friends including use of flag final argument

reduce dependancy on psubcomm for MatGetRedundantMatrix_MPIAIJ()

rename MatCreateMPIBAIJConcatenateSeqBAIJ() to MatCreateMPIMatConcatenateSeqMat()

mv struct 'Mat_Redundant' to '_p_Mat’ for simplify codes

compile: fixes for windows/solaris

add MatGetRedundantMatrix_MPISBAIJ() - via MatConvert & MatGetRedundantMatrix_MPIBAIJ()

add MatGetRedundantMatrix_MPIBAIJ() -- almost same as MatGetRedundantMatrix_MPIAIJ()

MatCreateMPIBAIJConcatenateSeqBAIJ() works for bs>1

add MatCreateMPIBAIJConcatenateSeqBAIJ() to be used by MatGetRedundant for sbaij format; bs>1 is not tested yet

added test to manual pages clarifying what row and column Mat blocksize actually mean
someone added them without any regard for how confusing it would be to users

MatGetVecs() and KSPGetVecs() replaced with MatCreateVecs() and KSPCreateVecs()
since they are creators not getters

Keep the block size when converting from MPIBAIJ matrices to MPIAIJ matrices.

Parallel matrices now behave the same than sequential matrices when converting from BAIJ to AIJ format.

MatZeroRowsColumns_MPI(B)AIJ: avoid nrows*comm_size complexity

Switch from bad "trick" to short-circuited binary search.

MatZeroRows_MPI(B)AIJ: avoid communication when MAT_NO_OFF_PROC_ZERO_ROWS

This functionality existed prior to Matt's refactor to use PetscSF and
was considered useful to PFLOTRAN.

Reported-by: Glen

MatZeroRows_MPI(B)AIJ: avoid communication when MAT_NO_OFF_PROC_ZERO_ROWS

This functionality existed prior to Matt's refactor to use PetscSF and
was considered useful to PFLOTRAN.

Reported-by: Glenn Hammond <gehammo@sandia.gov>

show more ...

MatZeroRows_MPI(B)AIJ: avoid nrows*comm_size complexity

If the rows are in reverse order, the old implementation would linear
search the ownership ranges for each row. Use binary search instead.

reduce code duplication for MatAXPYGetPreallocation_MPIXXX()