speed up adding sparsity pattern entries with coupling

[KristofferC]

Ref #993

As written then in check for j was performed on every iteration over i.

Before:
julia> @time allocate_matrix(dh; topology=top, interface_coupling = interface_coupling)
 25.123126 seconds (217.41 k allocations: 1.820 GiB, 0.87% gc time)


After:
julia> @time allocate_matrix(dh; topology=top, interface_coupling = interface_coupling)
 12.786278 seconds (217.84 k allocations: 1.821 GiB, 1.79% gc time)

I think the approach here of doing this in check like this has to be rethought do.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.61%. Comparing base (98e2b67) to head (18db56d).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1028      +/-   ##
==========================================
+ Coverage   93.60%   93.61%   +0.01%     
==========================================
  Files          39       39              
  Lines        5883     5893      +10     
==========================================
+ Hits         5507     5517      +10     
  Misses        376      376              

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

[KristofferC]

I see that the check for the coupling being non-zero is done inside _add_interface_entry. Lifting it out to avoid all these in for the zero case is probably worth while.

[fredrikekre]

There is already a duplicate-check here

Ferrite.jl/src/Dofs/sparsity_pattern.jl

Line 170 in 98e2b67

if k == length(x) + 1 || @inbounds(x[k]) != item

so possibly there shouldn't be one in the function modified here at all?

[KristofferC]

I now have

julia> @time allocate_matrix(dh);
  2.430063 seconds (1.21 k allocations: 1.767 GiB, 14.28% gc time)

julia> @time allocate_matrix(dh; topology, interface_coupling)
  3.207378 seconds (1.04 M allocations: 1.880 GiB, 13.06% gc time)

In #993, the first only took 0.7 seconds. I'm on a pretty fast CPU so did something regress @fredrikekre?

[KristofferC]

so possibly there shouldn't be one in the function modified here at all?

If I just flat out comment out the in checks the tests fail at least.

[KristofferC]

@AbdAlazezAhmed maybe you could look over this.

[AbdAlazezAhmed]

LGTM :D