Introduce isconstrained bitvector in ConstraintHandler

[KnutAM]

This makes it easier and faster to check if a dof is constrained.
Previous check was haskey(ch.dof_mapping, dofnr), now just ch.isconstrained[dofnr].

This speeds up zero_out_rows! and makes creating free_dofs easier (and a tiny bit faster).

Benchmarks

# close!
4.963 ms (59 allocations: 25.78 MiB) # PR
5.075 ms (57 allocations: 25.41 MiB) # master

# apply!
293.966 ms (0 allocations: 0 bytes) # PR
1.333 s (0 allocations: 0 bytes)    # master

benchmark code

using Ferrite, BenchmarkTools
grid = generate_grid(Hexahedron, (100,100,100));
dh = close!(add!(DofHandler(grid), :u, Lagrange{RefHexahedron, 1}()^3));
@btime close!(ch) setup = (ch = add!(ConstraintHandler(dh), Dirichlet(:u, getfacetset(grid,"left"), Returns(zero(Vec{3}))))) evals = 1;

ch = close!(add!(ConstraintHandler(dh), Dirichlet(:u, getfacetset(grid,"left"), Returns(zero(Vec{3})))));
K = allocate_matrix(dh); 
f = zeros(ndofs(dh));
@btime apply!($K, $f, $ch);

Extra runs for close!:

# PR 
  5.007 ms (59 allocations: 25.78 MiB)
  5.025 ms (59 allocations: 25.78 MiB)
  5.052 ms (59 allocations: 25.78 MiB)
  
# master
  5.100 ms (57 allocations: 25.41 MiB)
  5.121 ms (57 allocations: 25.41 MiB)
  5.111 ms (57 allocations: 25.41 MiB)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.31%. Comparing base (8eb6a07) to head (f244f12).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1310   +/-   ##
=======================================
  Coverage   94.30%   94.31%           
=======================================
  Files          40       40           
  Lines        6763     6768    +5     
=======================================
+ Hits         6378     6383    +5     
  Misses        385      385           

☔ 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.

[KristofferC]

There is also BitSet but I haven't used it much so I don't really know the trade off.

[KnutAM]

There is also BitSet but I haven't used it much so I don't really know the trade off.

  BitSet([itr])

  Construct a sorted set of Ints generated by the given iterable object, 
  or an empty set. Implemented as a bit string, and therefore designed 
  for dense integer sets. If the set will be sparse (for example, holding 
  a few very large integers), use Set instead.

I thought of it but discarded when reading this, since for the constrained dofs we have very sparse sets for large grids. But is there a better way to use this than my naive thinking of prescribed = BitSet(ch.prescribed_dofs) and checking haskey(prescribed, dofnr)?

[KristofferC]

Yeah, probably not worthwhile. I don't think it matters in practice though, this should be a lot smaller than many of the other data structures used already.

[KristofferC]

pairs here feels odd. Should it be enumerate?

[KnutAM]

My interpretation here is that colval and nzval have the same indices, hence it should be pairs. Of course, in practice for SparseMatrixCSR both colval and nzval are Vectors, so it doesn't matter.

[KristofferC]

Ok, I think that is fair.

[KristofferC]

Same w.r.t pairs

[KnutAM]

As above, but here the storage types in AbstractSparseMatrixCSC could be offset arrays, and then I assume that it should be pairs since we assume rowval and nzval have the same indexing?

[KristofferC]

👍

[KristofferC]

Should be good to go I think?