MatLoad_AIJ_HDF5: load indices as IS and values as Vec

MatLoad_AIJ_HDF: better sizes handling, account for user-defined

MatLoad_AIJ_HDF5: support for block size

MatLoad_AIJ_HDF5: add rectangular matrix support

fix Marek's MatLoad_AIJ_HDF5

MatLoad_AIJ_HDF5 adapted from PERMON

add MatLoad_AIJ_HDF5 skeleton; called from MatLoad_{MPI,Seq}AIJ for HDF5 load

MatLoad_*: check viewer type at impl level

Write clearly what "row indices" mean for compressed sparse row representation

Commit-type: documentation
Reported-by: Martin J Kuehn <martin.kuehn@uni-koeln.de>

Merged petsc/petsc into master

housekeeping: Silent some warnings from static analyzers

Do some housekeeping and add missing CHKERRQ calls

Minor style fixes

Merge branch 'stefano_zampini/fix-pcksp-fetidp-maint' [PR #1145]

Merge branch 'pr1144/a_selinger/feature-rowmerge-matmatmult' [PR #1144]

* pr1144/a_selinger/feature-rowmerge-matmatmult:
Added another sequential MatMatMult algorithm
This is an extension of the Row

Merge branch 'pr1144/a_selinger/feature-rowmerge-matmatmult' [PR #1144]

* pr1144/a_selinger/feature-rowmerge-matmatmult:
Added another sequential MatMatMult algorithm
This is an extension of the RowMerge algorithm by @Karl Rupp , which enables the multiplication of matrices with an arbitrary number of nonzeros in the left matrix.
It is faster than the default implementation ("sorted") for very sparse matrices.
MatMatMult: Fixed a bug with mem size overflow and reformatted code
Fixed a bug in the "combined"-algorithm, where the variable c_maxmem for memory requiremments could overflow for relatively dense matrices.

show more ...

ex226: Fixed missing reinitialization of output nonzero counter.

Reported-by: Andreas Selinger <andreasselinger@gmail.com>
Fixed-by: Andreas Selinger <andreasselinger@gmail.com>

MatMatMult: Added MatSetType and minor changes

MatMatMult: simplified rowmerge algorithm

MatISTranspose_MPIAIJ: fix bug

the boolean should be reduced on the communicator

Remove last of HAVE_VECCUDA that did not get removed due to new branch merging

Commit-type: bug-fix
Reported-by: nightly tests

Merge branch 'barry/have-veccuda-to-have-cuda'

MatMatMult: Fixed a bug with mem size overflow and reformatted code

Fixed a bug in the "combined"-algorithm, where c_maxmem could overflow
for relatively dense matrices.

MatMatMult: Added rowmerge algorithm

Change outdated HAVE_VECCUDA macro to simply HAVE_CUDA for clarity/ease of maintenance.

Commit-type: style-fix

MatType was set on the wrong matrix in the routine

Reported-by: Jenkins