Kokkos Function System

[NamjaeChoi]

Refs #30655.

Functors are not dispatched but plugged inside the user's parallel code, so I can't think of a way to achieve static polymorphism here. I think I will have to introduce virtual functions minimally in the wrapper class, but any idea to avoid it would be welcome, if you have any. Of course, if the user specifies the exact object type instead of using the abstract type, they will be able to evaluate functors statically.

Concerns are performance and SYCL, but regarding SYCL, Intel is adding an extension to allow virtual functions on GPU. We may be able to leverage it when the time comes to support SYCL. I don't think we are interested in supporting SYCL for backends other than Intel GPU.

[NamjaeChoi]

@lindsayad My next thing to look into would be this. I will start from functions. Do you have any thoughts on this?

[lindsayad]

If we use virtuals, should we consider just adapting the current functor system?

[NamjaeChoi]

The problem with that is we can't construct MOOSE objects directly on device. If an object that has virtual methods is constructed on host, the vtable will be populated with host function pointers. What I'm trying to do here is to define a thin wrapper that has virtual methods and is light enough to be built on device, and let the wrapper call the wrapped functor methods statically.

[lindsayad]

I think this looks ok to me

[NamjaeChoi]

Still working... just want to run test suite once

[moosebuild]

Job Documentation, step Docs: sync website on f9d7ff6 wanted to post the following:

View the site here

This comment will be updated on new commits.

[moosebuild]

Job Coverage, step Generate coverage on f9d7ff6 wanted to post the following:

Framework coverage

	4e5abd	#31653 f9d7ff
	Total	Total	+/-	New
Rate	85.99%	85.93%	-0.05%	100.00%
Hits	124384	124624	+240	510
Misses	20272	20400	+128	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]

going to unsubscribe for now. Mention me when you want more input or it's ready for review

[NamjaeChoi]

@lindsayad I might have to hold this. Virtual functions on device seem to require RDC to be fully functional.

[NamjaeChoi]

The code is currently in a state where CUDA will work when we have RDC.

[NamjaeChoi]

I don't want to indefinitely hold this PR for RDC. I will temporarily disable using virtual dispatch for GPU.

[moosebuild]

Job Test, step Results summary on f9d7ff6 wanted to post the following:

Framework test summary

Compared against 4e5abd1 in job civet.inl.gov/job/3352081.

Removed tests

Added tests

Test	Time (s)
kokkos/functions/piecewise_constant.piecewise_constant/json	1.91
kokkos/functions/piecewise_constant.piecewise_constant/xy_data	1.45
kokkos/functions/piecewise_constant.piecewise_constant/csv	1.44
kokkos/functions/concrete_type.invalid_type	1.42
kokkos/functions/piecewise_constant.piecewise_constant/x_y	1.37
kokkos/functions/constant_function.test	1.37
kokkos/functions/default_function.test	1.33
kokkos/functions/concrete_type.concrete_type	1.12

Run time changes

Modules test summary

Compared against 4e5abd1 in job civet.inl.gov/job/3352081.

Removed tests

Added tests

Run time changes