Adjoint Solver

[ChristianSchuler]

ToDo:

• find a way to only initialize parameter matrices that are needed
• find a general way to fill only the parameter matrices that are requested (in assemble_parameter_matrices!())
• add sensitivities related with compute_viscosity function
• boundary treatment
• think about Adjoint struct
• add CUDA module
• pack all internal parameters in the solve function, so that the adjoint part could be separated from the forward solve

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 0% with 539 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/adjoint/AdjointStressKernels.jl	0.00%	211 Missing ⚠️
src/adjoint/Adjoint_Stokes2D.jl	0.00%	97 Missing ⚠️
src/adjoint/AdjointSolve.jl	0.00%	84 Missing ⚠️
src/adjoint/AdjointSensitivities.jl	0.00%	54 Missing ⚠️
src/adjoint/Adjoint_VelocityKernels.jl	0.00%	34 Missing ⚠️
src/types/constructors/stokes.jl	0.00%	28 Missing ⚠️
src/adjoint/Adjoint_PressureKernel.jl	0.00%	19 Missing ⚠️
src/types/stokes.jl	0.00%	8 Missing ⚠️
src/ext/AMDGPU/2D.jl	0.00%	2 Missing ⚠️
src/ext/CUDA/2D.jl	0.00%	2 Missing ⚠️

❗ There is a different number of reports uploaded between BASE (e6a5103) and HEAD (febbbaf). Click for more details.

HEAD has 24 uploads less than BASE

Flag	BASE (e6a5103)	HEAD (febbbaf)
	28	4

Additional details and impacted files

📢 Thoughts on this report? Let us know!

[albert-de-montserrat]

For the sensitivities (can useful in other cases as well), maybe its a good idea to make some struct such that

struct FieldArray{T}
    center::T
    vertex::T
    xy::Union{Nothing, T} # 3D only
    yz::Union{Nothing, T} # 3D only
    xz::Union{Nothing, T} # 3D only
end

and bundle them all into an args::NamedTuple

[albert-de-montserrat]

Great! Does the last commit fix the scaling issues?