MPIPtAP: Always call MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable in MatPtAPNumeric_MPIAIJ_MPIAIJ_scalable

since we care about the memory usage. Some symbolic algorithms such as heap is really good sinc

MPIPtAP: Always call MatMatMultNumeric_SeqAIJ_SeqAIJ_Scalable in MatPtAPNumeric_MPIAIJ_MPIAIJ_scalable

since we care about the memory usage. Some symbolic algorithms such as heap is really good since
it does not take a global size, but it is attached with MatMatMultNumeric_SeqAIJ_SeqAIJ_Sorted

show more ...

Addressed Barry's comments: always use ctable

MPIPtAP: added a few tests using different MatMatMultSymbolic_SeqAIJ_SeqAIJ

Refactored MatMatMult_SeqAIJ_SeqAIJ so that we can choose different

algorithms during runtime for MPIPtAP

Revert the function name where it is referenced

Merge branch 'hzhang/superlu_dist-rm-globalmode' [PR #1438]

* hzhang/superlu_dist-rm-globalmode:
deprecate MatInputMode == GLOBAL in superlu_dist interface as requested by S. Xiaoye Li

MPIPtAP: Enable large-scale simulations

The old code uses a global size of columns when creating the local matrix

This uses a lot of memory when calling MatMatMultSymbolic_SeqAIJ_SeqAIJ

We here co

MPIPtAP: Enable large-scale simulations

The old code uses a global size of columns when creating the local matrix

This uses a lot of memory when calling MatMatMultSymbolic_SeqAIJ_SeqAIJ

We here compact out many extra columns, and construct a new local size.

show more ...

initialized pointers in MatLUFactorNumeric_SuperLU_DIST()

bug fix for uninitialized 'rstart'

Merge remote-tracking branch 'origin/jczhang/vecscat-in-sf' [PR #1396]

Merged in stefano_zampini/feature-pcbddc-enablelor (pull request #1427)

Stefano zampini/feature pcbddc enablelor

rm SuperLU_MatInputMode

Add the missed VecRestoreArray

VecGetArray() must have a matching VecRestoreArray()

bugfix for MatView_MPIAIJ_PtAP() as reported by Stefano Zampini.
When MatView a matrix Amat which is a duplicate PtAP, it should call base MatView_MPIAIJ() because MatDuplicate only creates base matr

bugfix for MatView_MPIAIJ_PtAP() as reported by Stefano Zampini.
When MatView a matrix Amat which is a duplicate PtAP, it should call base MatView_MPIAIJ() because MatDuplicate only creates base matrix (see Note). However, it calls MatGetOperation(mat,MATOP_VIEW,&viewf); MatSetOperation(B,MATOP_VIEW,viewf); i.e., it copies old matview ops. This patch fix the bug.

show more ...

Fix bugs with locks imposed upon vectors in VecScatterBegin/End

Athough otherwise one can potentially overlap computation/communication in MatMultTranspose_MPIAIJ,
MatMultTransposeAdd_MPIAIJ, that i

Fix bugs with locks imposed upon vectors in VecScatterBegin/End

Athough otherwise one can potentially overlap computation/communication in MatMultTranspose_MPIAIJ,
MatMultTransposeAdd_MPIAIJ, that is based on the assumption that the scattered data is not written
to the destination vector until VecScatterEnd(). This assumption is too strict for implementation
of VecScatter and could lead to wrong results. For example, the origin MatMultTransposeAdd_MPIAIJ is
wrong with -vecscatter_merged. In addition, the assumption is not stated in petsc user's manual.

show more ...

Call new VecScatter interfaces instead of accessing vecscatter private data directly

Merge branch 'jczhang/fix-vecscatter-cuda/maint' [PR #1388]

* jczhang/fix-vecscatter-cuda/maint:
CUDA vecscatter needs to take care of the ScatterMode argument

Restore the error checking code

Add a ScatterMode arg in cuda vecscat to select to/from context

The old code VecScatterInitializeForGPU() initializes the pointer (PetscCUDAIndices*)&inctx->spptr) based
on an input ScatterMode befo

Add a ScatterMode arg in cuda vecscat to select to/from context

The old code VecScatterInitializeForGPU() initializes the pointer (PetscCUDAIndices*)&inctx->spptr) based
on an input ScatterMode before VecScatterBegin() is called.

If a vecscatter context is firstly used for a SCATTER_FORWARD, and secondly used
for a SCATTER_REVERSE, there will be an error. Since in the second VecScatter,
it uses out-of-date (PetscCUDAIndices*)&inctx->spptr)

The solution is "do not prematurely consider ScatterMode when building (PetscCUDAIndices*)&inctx->spptr). Instead,
select correct to/from until VecScatterBegin() is called"

show more ...

MatDestroy_SeqAIJ_MatTransMatMult: prevent from segfaulting

rm all codes of MatInputMode == GLOBAL

Merge branch 'denera/tao-brgn-matlab-segfault-fix' [PR #1184]

* denera/tao-brgn-matlab-segfault-fix:
MATLAB Leastsquares examples segfault fix for BRGN.
Trust region initialization was producing a s

Merge branch 'denera/tao-brgn-matlab-segfault-fix' [PR #1184]

* denera/tao-brgn-matlab-segfault-fix:
MATLAB Leastsquares examples segfault fix for BRGN.
Trust region initialization was producing a segfault on MatMult with the Jacobian.

Evaluating the Jacobian in MATLAB using a SEQAIJ PETSc type, but then converting it to SEQDENSE before using it in Tao operations resolved the problem.

show more ...

deprecate MatInputMode == GLOBAL in superlu_dist interface as requested by S. Xiaoye Li

minor fix

MatCreateSeqAIJWithArrays: fix typo in error message