Extended the jacobian to functionals involving Skeleton terms

[kishore-nori]

• jacobian for functionals involving inetgration over Skeleton terms can be handled now
• Added corresponding tests compared with ForwardDiff
• The high-level API remains the same

[amartinhuertas]

Hi @kishore-nori ! thanks for the PR! Only minor comments.

[amartinhuertas]

Edit the NEWS.md file, please

[codecov-commenter]

Codecov Report

Merging #803 (8b7d3ba) into master (26a1886) will increase coverage by 0.21%.
The diff coverage is 95.65%.

@@            Coverage Diff             @@
##           master     #803      +/-   ##
==========================================
+ Coverage   88.23%   88.45%   +0.21%     
==========================================
  Files         159      159              
  Lines       17763    17865     +102     
==========================================
+ Hits        15673    15802     +129     
+ Misses       2090     2063      -27     

Impacted Files	Coverage Δ
src/FESpaces/FEAutodiff.jl	92.72% <95.65%> (+3.71%)	⬆️
src/Polynomials/QGradMonomialBases.jl	99.25% <0.00%> (+0.16%)	⬆️
src/Visualization/Vtk.jl	87.93% <0.00%> (+1.00%)	⬆️
src/CellData/CellFields.jl	91.20% <0.00%> (+1.62%)	⬆️
src/Fields/ApplyOptimizations.jl	57.05% <0.00%> (+2.35%)	⬆️
src/ReferenceFEs/NedelecRefFEs.jl	91.17% <0.00%> (+2.37%)	⬆️
src/CellData/SkeletonCellFieldPair.jl	88.37% <0.00%> (+9.30%)	⬆️
src/Fields/AffineMaps.jl	70.17% <0.00%> (+14.03%)	⬆️

📣 Codecov can now indicate which changes are the most critical in Pull Requests. Learn more

[kishore-nori]

Hi Alberto, thank you very much for the help, inputs and discussion for the PR, and for reviewing. I fixed the suggested changes and added further details and pushed them.

[amartinhuertas]

Hi @kishore-nori ... I am afraid that we are increasing too much testing times in Github Actions with this PR.

Looking at the logs, we have increased total time from 46 min to 1h 5min approx (by 20 mins approx).

See attached picture below to identify where the hot spot is.

Can we try to do a better job here and curate the current set of tests such that we significantly reduce this testing times while still testing the same amount of code base within Gridap kernel? Perhaps we can take a look at code coverage reports (https://app.codecov.io/gh/gridap/Gridap.jl/branch/AD-for-SkeletonIntegration/) to understand which parts of the codes we are going through with each of the tests.

[kishore-nori]

Hi Alberto, thank you for the logs. Yeah I realised that the testing times have increased due to the new jacobian for Skeleton functions in FEAutodiffTests.jl. The thing is I put in 5 functionals which reflect most of the complications which are possible in such cases. These cover usual mean and jump terms of uh and vh, terms with combinations of ∇(uh), ∇(vh), Δ(u) and Δ(vh). Additionally, there are test where jump and mean have operations of CellFields inside. One way that I can reduce the run-time is by having 2x2 grid or even 2x1 grid instead of current 3x3 grid. I can change this based on your suggestions. I kept reffe order at 2 to test even laplacians.

[amartinhuertas]

One way that I can reduce the run-time is by having 2x2 grid or even 2x1 grid instead of current 3x3 grid. I can change this based on your suggestions.

I am not sure this is going to help much. Note in the logs that 99% of the time is compilation time. So I would say it is more about the amount of specialized code for which you are triggering compilation from your tests, not the size of the test cases.

[amartinhuertas]

CI tests now take 45 mins. Well done @kishore-nori !

[kishore-nori]

That's great! I think the wrapping of assemble(jacobian(...)) into a function might have also helped.