PETSc - adjust bpsraw test parameters to avoid too many threads for cuda backends

Merge pull request #560 from CEED/jeremy/diag-vec

LinearAssemble Improvements

Merge pull request #559 from CEED/jeremy/test-cuda-bps

PETSc - fix PETSc test parameters for CUDA backends

Op - change LinearAssemble* to accept CeedVector instead of pointer to CeedVector, allows for less memory movement and interfaces with parent code better

PETSc - fix PETSc test parameters for CUDA backends

style - rename AssembleLinear* to LinearAssemble*

examples/petsc/bps: reuse the same DMPlex for all desired BPs and CEED resources

DM partitioning and redistribution for low-order discretizations are
relatively much more expensive than a solve. Th

examples/petsc/bps: reuse the same DMPlex for all desired BPs and CEED resources

DM partitioning and redistribution for low-order discretizations are
relatively much more expensive than a solve. This way we get more
reuse from the DMPlex.

show more ...

PETSc bps & bpsraw: Remove benchmark_mode (not needed anymore) and always disregard first run performance

This is preferred for benchmarking purposes, so that we are preloading the cache;

PETSc bpsraw: always print DoFs/Sec in CG info (except when in test mode)

PETSc bps: always print DoFs/Sec in CG info (except when in test mode)

petsc/bps/README: Use code style to avoid bold type and specify FEM nodes just to avoid confusion with compute nodes

examples/petsc/bps: allow looping over problems and CEED resources.

examples/petsc/bps: add -ksp_max_it_clip and apply DRY

examples/petsc/bps: support degree lists and local_nodes ranges

* print hostname, total number of ranks, and number of ranks per node
* Allow number of ranks per node to be specified (mostly in case

examples/petsc/bps: support degree lists and local_nodes ranges

* print hostname, total number of ranks, and number of ranks per node
* Allow number of ranks per node to be specified (mostly in case one has an
old MPI that doesn't support MPI_Comm_split_type) and hostname to be
specified at the command line.

show more ...

PETSc BPs: Refactor benchmark mode to run in loop

Petsc BPs: Update comment

PETSc BPs: Make style

PETSc BPs: print Local Elements

examples/petsc/bps: make -local_nodes actually local

examples/petsc: move utility functions to bps.c; make static

PETSc BPs: Add -local_nodes CL option

examples: support for PETSc-3.14 DMPlexGetClosureIndices

Area: Refactoring - Allows the use of the command line option -degree in place of -petscspace_degree

Separate setup of libCEED objects and problem specific data in setuparea.h separate file

BPssphere: remove some dead code, unused variables

Multigrid - fix memtype in CeedVecSetArray call