Add assembly into rectangular sparse matrices

[termi-official]

Closes #1133 (and #968 ?)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.13%. Comparing base (511e715) to head (3eb5097).
⚠️ Report is 7 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1279      +/-   ##
==========================================
- Coverage   94.23%   94.13%   -0.11%     
==========================================
  Files          40       40              
  Lines        6751     6766      +15     
==========================================
+ Hits         6362     6369       +7     
- Misses        389      397       +8     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[KnutAM]

Just some minor comments regarding docs and tests, but looks good to me.

Could you post a benchmark in the PR comparing master and PR for assembling a constant Ke and fe with some grid, just so we document that there are no performance regressions (I assume there isn't?)

[KnutAM]

Another question, what about BlockAssembler (I don't know this code so well, but looking at it I guess this should be extended to that case as well?)

And side-note, but supertype(BlockAssembler) = AbstractAssembler, can this be both CSR and CSC? Or should that subtype AbstractCSCAssembler?

[termi-official]

Another question, what about BlockAssembler (I don't know this code so well, but looking at it I guess this should be extended to that case as well?)

Potentially, yes, but right now I do not have a real use-case for it and also cannot think about one. So it could become a good first issue.

And side-note, but supertype(BlockAssembler) = AbstractAssembler, can this be both CSR and CSC? Or should that subtype AbstractCSCAssembler?

This looks correct to me. The block format is neither CSR nor CSC.

[KnutAM]

This looks correct to me. The block format is neither CSR nor CSC.

But the underlying matrix is one of them? For me it was just not clear if both are supported (perhaps I should just try but it is late 😄)

[termi-official]

using Ferrite, SparseArrays, BenchmarkTools

Ke = zeros(4,4)
K = sprand(4,4,1.0)

assembler = start_assemble(K)
dofs = collect(1:4)
@btime assemble!($assembler, $dofs, $Ke)

This benchmark show no statistically significant difference between the PR and master.

[KnutAM]

Sorry, missed the aliasing in the previous round and had some questions about that. I didn't notice that and that was the reason for the performance question since I figured with double the vectors there could be some additional costs.

Otherwise just some minor tweaks.

[KnutAM]

LGTM!