Converted Intel intrinsics aligned loads/stores to unaligned in AIJPERM code.

Converted Intel intrinsics aligned loads/stores to unaligned.

The cost of an unaligned load/store on aligned memory is the same as the
cost of an aligned load/store instruction on modern Intel CPUs.

Merge branch 'rmills/feature-aijmkl-add-matmatmultnumeric'

Added configure option, "--with-avx512-kernels", to specify that kernels hand-coded with AXV-512 should be preferred.

Fixes to enable use of PetscSparseDensePlusDot_AVX512_Private().

Moved PetscSparseDensePlusDot_AVX512_Private to aij.h.
Ensured that sbaij/seq/relax.h does not invoke this function in unsigned short

Fixes to enable use of PetscSparseDensePlusDot_AVX512_Private().

Moved PetscSparseDensePlusDot_AVX512_Private to aij.h.
Ensured that sbaij/seq/relax.h does not invoke this function in unsigned short case.

show more ...

Moved AVX-512 intrinsics code into PetscSparseDensePlusDot_AVX512_Private().

Removed AVX and AVX2 kernels for now.

More performance testing needed before merging into master, so removing these
for the time being.

Fix MatGetInertia_SuperLU_DIST()
Reported-by: Jose E. Roman <jroman@dsic.upv.es>

GPUs: Allow ViennaCL and VECCUDA to be used concurrently.

This also gets rid of the weird veccuda.py.

Updated a comment.

Added support for MatMatMultNumeric() for AIJMKL matrices.

Note that this requires MKL version 18, update 2 or later.

Merge branch 'hzhang/superlu_dist-matinertia'

CUSP: Removed VECCUSP and AIJCUSP as well as preconditioners.

SA-AMG from CUSP will be re-added in a follow-up commit,
made to work with AIJCUSPARSE and VECCUDA.

initialize pointers to 'NULL'

Renamed source files from "csrperm" to "aijperm" to be consistent with AIJPERM matrix class name.

add Cholesky factorization when A->symmetric=true

add MatGetInertia_SuperLU_DIST()
Requested-by: Xiaoye S. Li <xsli@lbl.gov>

replace default process grid to nprow <= npcol
Requested-by: Xiaoye S. Li <xsli@lbl.gov>

CUDA: Fixed most test errors due to MatMultTranspose*

Merge branch 'stefano_zampini/feature-factor-sell' [PR #860]

* stefano_zampini/feature-factor-sell:
This branch adds LU factorization support to MATSEQSELL via MUMPS and SuperLU.

Merge branch 'jose/cuda-tests'

* jose/cuda-tests: Fix CUDA-related tests in new test harness. See PR #848.

CUSP: Fixed failure of KSP tutorial ex7 and test ex43 in nightly tests.

Merge branch 'balay/update-superlu_dist-5.3.0'

Merge remote-tracking branch 'origin/maint'

MATMKL_PARDISO: don't use the structurally symmetric option by default