Minor optimizations on MooseVariableData

[GiudGiud]

locally (M1 max) I observed 8% speedup on simple_diffusion.i with -r 7.
I am hoping someone can reproduce so we can merge this.
I doubt the test suite will show a consistent improvement but let's see with this PR's run

[moosebuild]

Job Documentation, step Docs: sync website on d7b226f wanted to post the following:

View the site here

This comment will be updated on new commits.

[moosebuild]

Job Coverage, step Generate coverage on d7b226f wanted to post the following:

Framework coverage

	70e113	#29697 d7b226
	Total	Total	+/-	New
Rate	85.25%	85.29%	+0.05%	79.28%
Hits	108182	107972	-210	176
Misses	18725	18621	-104	46

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
• thermal_hydraulics
• xfem

Warnings

• framework new line coverage rate 79.28% is less than the suggested 90.0%

This comment will be updated on new commits.

[moosebuild]

Job Controlled app tests on dd74a6c : invalidated by @loganharbour

[moosebuild]

Job Disable HDF5 on dd74a6c : invalidated by @loganharbour

[MengnanLi91]

I'll test this on my M1 and see how it goes

[GiudGiud]

looking at the test suite, the Framework 1 and 2 recipe and Module
https://civet.inl.gov/branch/6695/

it seems we are running:

• Framework 1 5 minutes slower
• Framework 2 25 minutes faster
• Modules 12 minutes faster
  but maybe it's just build machine differences

[MengnanLi91]

looking at the test suite, the Framework 1 and 2 recipe and Module

https://civet.inl.gov/branch/6695/

it seems we are running:

• Framework 1 5 minutes slower

• Framework 2 25 minutes faster

• Modules 12 minutes faster

but maybe it's just build machine differences

I tried on my Mac M1 with mpiexec -n 10 -i simple_diffusion.i -r 7. Not much difference in total execution time

[MengnanLi91]

Not sure if the communication time matters here. I can try a serial run tomorrow

[GiudGiud]

looking at more recipes, OpenMPI and ARM Mac, seems 0 change in runtime.

[MengnanLi91]

I ran it with perf_graph, the memory seems to change with new commit. I think it may have a direct effect from the changes

[GiudGiud]

for the better or worse?

[loganharbour]

Profiling reveals a factor of 2 improvement in MooseVariableData::computeValues() with simple diffusion, 7 refinements. This problem is the worst case for testing this as all we do is fetch vector tags. With problems that are more complex, the improvement will likely be better. Testing a transient problem now.

[loganharbour]

I did some verification about the inlining of the lambdas with our current GCC and clang with -O2: https://godbolt.org/z/94ra47jnK

GCC is smart enough to see that the version with the lambda is exactly the same as the one without, and actually just produces a jump. Cool cool cool