PetscFE determines basis for examples/fluids/navierstokes

[KennethEJansen]

As we prepare to bring simplex meshes into the examples/fluids/navierstokes application within libCEED, it is important to let PETSc define the basis functions. This code revision followed similar work done in Ratel to address this issue.

This has been completed at this point for the primary operators but not for the Data/Learning extensions but hopefully those will be resolved within this PR.

As all changes are within examples/fluids and off of a current main, all tests of libCEED should still pass.

As the work mimicked Ratel development (and work from a branch of Rezgar Shekeri) to acquire the basis information from PETSc, credit should go those efforts while responsibility for miss-interpretations of that effort belong to Kenneth Jansen with co-author help from James Wright.

At this time the fluids example has been verified correct without loss of computational efficiency on Polaris in both serial and parallel including STG with Q1, Q2, and Q3 elements. Further tests with simplex elements are ongoing and updates will be provided as they become available.

[jrwrigh]

TODO:

• Remove Q_dim from GetRestrictionForDomain arguments (just use Q for quadrature points per element, aligned with CeedBasis nomenclature)
• Style/rearrange
• Fix all calls to GetRestrictionForDomain
• Fix test errors.

[KennethEJansen]

In the failing tests I see -q_extra 2 I guess it is obvious that PestscFE does not understand/honor this request.

@jedbrown how do you propose we resolve this? Is there a way to communicate this request to PETSc?

[jrwrigh]

In the failing tests I see -q_extra 2 I guess it is obvious that PestscFE does not understand/honor this request.

@jedbrown how do you propose we resolve this? Is there a way to communicate this request to PETSc?

I'm guessing we probably just need to change:

libCEED/examples/fluids/src/setupdm.c

Line 53 in 51f8a93

PetscCall(PetscFECreateLagrange(PETSC_COMM_SELF, problem->dim, num_comp_q, is_simplex, degree, PETSC_DECIDE, &fe));

from PETSC_DECIDE to the actual quadrature order we want. See PETSc docs

[jedbrown]

Yeah, see how q_order is defined and used here.
https://gitlab.com/micromorph/ratel/-/blob/main/src/materials/ratel-fem.c?ref_type=heads#L463-478