Unblock upcoming nonlinear solves and accelerate MFEM tests

[nmnobre]

Reason

a) Unblock upcoming nonlinear solves using the MFEM backend affected by the changes in hypre 3.0 to PCG and GMRES.
b) Accelerate MFEM test executions.

Design

a) Extended mfem::HyprePCG and mfem::HypreGMRES to reproduce the effect of mfem/mfem#5065.
b) Split LOR tests into a different folder to gain some extra concurrency when running the test suite.

Impact

b) This mostly benefits people working on the MFEM backend, I don't expect us to see much difference on CIVET.
On my workstation, with ./run_tests -C tests/mfem -j64:

Configuration	Before [s]	After [s]	Speedup
--dbg --compute-device cuda	208.5	147.0	42%
--dbg --compute-device cpu	223.1	152.8	46%
--opt --compute-device cuda	83.0	72.8	14%
--opt --compute-device cpu	83.2	71.9	16%

[moosebuild]

Job Documentation, step Docs: sync website on 2e0537d wanted to post the following:

View the site here

This comment will be updated on new commits.

[moosebuild]

Job Coverage, step Generate coverage on 2e0537d wanted to post the following:

Framework coverage

	2acb41	#31798 2e0537
	Total	Total	+/-	New
Rate	85.99%	85.99%	+0.00%	100.00%
Hits	124177	124181	+4	24
Misses	20240	20240	-	0

Diff coverage report

Full coverage report

Modules coverage

Coverage did not change

Full coverage reports

Reports

• framework
• chemical_reactions
• combined
• contact
• electromagnetics
• external_petsc_solver
• fluid_properties
• fsi
• functional_expansion_tools
• geochemistry
• heat_transfer
• level_set
• misc
• navier_stokes
• optimization
• peridynamics
• phase_field
• porous_flow
• ray_tracing
• rdg
• reactor
• richards
• scalar_transport
• solid_mechanics
• solid_properties
• stochastic_tools
• subchannel
• thermal_hydraulics
• xfem

This comment will be updated on new commits.

[lindsayad]

why do these changes lead to a speedup?

[nmnobre]

why do these changes lead to a speedup?

Apologies, I've edited the original post to make it clearer. What makes it faster is the LOR splitting bit.

[nmnobre]

Notable changes since last reviewed: 1) moved the new class definitions to a separate header, so it can be included (and thus reused) in different compilation units, solving the non-unity build problems and 2) disabled multithreading for MFEM problems to avoid issues when adding objects for each thread (in particular in FEProblemBase::addFunction - 003a961), with the (niche?) caveat that libMesh subapps will also be single-threaded.

[moosebuild]

Job Test, step Results summary on 2e0537d wanted to post the following:

Framework test summary

Compared against 2acb415 in job civet.inl.gov/job/3334968.

Removed tests

Test	Time (s)
mfem/kernels.MFEMCurlCurlLOR	10.22
mfem/kernels.MFEMHeatConductionLOR	6.58
mfem/kernels.MFEMGradDivLOR	6.41
mfem/kernels.MFEMDiffusionLOR	2.13

Added tests

Test	Time (s)
mfem/lor.MFEMCurlCurlLOR	9.99
mfem/lor.MFEMHeatConductionLOR	6.81
mfem/lor.MFEMGradDivLOR	5.81
mfem/lor.MFEMDiffusionLOR	2.13

Run time changes

Modules test summary

Compared against 2acb415 in job civet.inl.gov/job/3334968.

Removed tests

Added tests

Run time changes

Test	Base (s)	Head (s)	+/-
stochastic_tools/test:web_server_control.stochastic_control/batch_reset_multi	6.00	2.90	-51.71%