Fix a few warnings.

Reported-by: Jose E. Roman <jroman@dsic.upv.es>

Merge remote-tracking branch 'origin/maint'

Fix bugs and simplify MatMultTransposeAdd_SeqAIJCUSPARSE

MatT always uses non-compressed row format on GPU. We should not use cusparsestruct->workVector, which is created for Mat instead of MatT, an

Fix bugs and simplify MatMultTransposeAdd_SeqAIJCUSPARSE

MatT always uses non-compressed row format on GPU. We should not use cusparsestruct->workVector, which is created for Mat instead of MatT, and might be less than A->cmap->n (when Mat has compressed rows)

show more ...

Save rowoffsets_gpu and use A->was_assembled to easily update Mat and MatT when matrix was assembled

Old code used if (A->assembled ..), which is wrong and could not take advantage of the fact that

Save rowoffsets_gpu and use A->was_assembled to easily update Mat and MatT when matrix was assembled

Old code used if (A->assembled ..), which is wrong and could not take advantage of the fact that A's non-zero pattern is not changed

show more ...

Fix memory leak: free matTranspose and workVector when rebuilding cusparsestruct

Reported-by: jordic <jordic@cttc.upc.edu>

Add MatBindToCPU_SeqAIJCUSPARSE().

Fix a typo

Revise CHKERRCUDA/CUBLAS/CUSPARSE

It fixes problems in MR 2228

Map a cuda error code to its name and description

Rename valid_GPU_array/matrix to offloadmask

ad hoc fix for cuda bug in mat-transpose-mult