Merge remote-tracking branch 'origin/haplav/petscoptionsbegin-title-obligatory'

Merge branch 'jed/cuda-aware-spectrum-mpi' into 'master'

CUDA: adjust check or GPU awareness to work for Spectrum MPI

See merge request petsc/petsc!2650

Improve documentation of new -info option and related API.

* Cast to PetscObject takes place right in PetscInfo* macros.
* PetscInfo_Private() takes PetscObject instead of void*

docs-only changes:

Improve documentation of new -info option and related API.

* Cast to PetscObject takes place right in PetscInfo* macros.
* PetscInfo_Private() takes PetscObject instead of void*

docs-only changes:
* dev.html:
- Focus mainly on -info relevant for normal users.
- API additions in a sublist.
* Unify -info synopsis.
* manpages
- Remove redundant/confusing details.
- Fix levels.
- Main deal of information if in front-facing PetscInfo().

show more ...

CUDA: adjust check or GPU awareness to work for Spectrum MPI

And any other Open MPI that does not support the query interface,
but that is GPU-aware.

Thanks-to: Junchao Zhang <jczhang@mcs.anl.gov>

PetscOptionsBegin: Title argument made obligatory

to avoid (null) headers in -help output.

Additionally
* PetscOptionsBegin() inarg validation
* fix a few examples which break this
* fix TaoMatGetS

PetscOptionsBegin: Title argument made obligatory

to avoid (null) headers in -help output.

Additionally
* PetscOptionsBegin() inarg validation
* fix a few examples which break this
* fix TaoMatGetSubMat() to use PetscObjectOptionsBegin() instead of PetscOptionsBegin()
* fix weird position of PetscOptionsEnd() in TAO examples

show more ...

Improve -help output.

* remove redundant separators
* separators more clearly separate sections
* improve help for -log_exclude
* separate PetscOptionsGetViewer() help section
* get rid of PetscOpti

Improve -help output.

* remove redundant separators
* separators more clearly separate sections
* improve help for -log_exclude
* separate PetscOptionsGetViewer() help section
* get rid of PetscOptionsCheckInitial_Components()

show more ...

Variable/function renaming for easy reading

set the default algorithm of -build_twosided based on comm size

The Allreduce algorithm performs better than the ibarrier algorithm with small number of ranks.
However, the latter scales better than

set the default algorithm of -build_twosided based on comm size

The Allreduce algorithm performs better than the ibarrier algorithm with small number of ranks.
However, the latter scales better than the former.

See more experimental results at https://gitlab.com/petsc/petsc/-/merge_requests/2611

show more ...