FEFaceEvaluation: Fix ECL for continuous elements

[bergbauer]

@peterrum @kronbichler

For special snowflakes like myself who need element-centric face integrals for continuous elements. ❄️

[kronbichler]

Looks good to me, apart from the initial setting. I have two questions, though:

• Does the code now work as you expected?
• Do you thin k we can extract a test case from your application? It would be great to find something to ensure that these complicated checks remain correct.

[kronbichler]

/rebuild

[bergbauer]

Does the code now work as you expected?

Yes, I get the same results for ECL and FCL.

Do you think we can extract a test case from your application? It would be great to find something to ensure that these complicated checks remain correct.

I am thinking of a check that compares results from ECL with FCL for FE_Q for a normal gradient jump term.

[bergbauer]

I had problems with MatrixFreeTools::categorize_by_boundary_ids(tria, additional_data);. Do you have an idea why this does not work? @peterrum

The test uses a mask instead of categorization by boundary ids.

[peterrum]

I had problems with MatrixFreeTools::categorize_by_boundary_ids(tria, additional_data);. Do you have an idea why this does not work?

What is the problem?

[bergbauer]

What is the problem?

The computed dof_indices are wrong.

dealii/include/deal.II/matrix_free/fe_evaluation.h

Lines 3535 to 3536 in c4e9688

	dof_indices[v] =
	dof_info.dof_indices.data() + my_index_start[0].first;

[kronbichler]

Thank you for the test.

[bergbauer]

In reference to #15589 use EvaluationFlags interface 6681f2b

[kronbichler]

Please squash to a smaller number of commits, then this is ready.

[bergbauer]

Squashed! @kronbichler