Coloring benchmark problem

[ChrisRackauckas]

using SparseDiffTools, LinearAlgebra, SparseArrays
N = 256
Mx = Tridiagonal([1.0 for i in 1:N-1],[-2.0 for i in 1:N],[1.0 for i in 1:N-1])
My = copy(Mx)
Iy = SparseMatrixCSC(I,N,N)
Ix = SparseMatrixCSC(I,N,N)
fJ = ones(3,3)
Dz = [1 0 0
      0 0 0
      0 0 0]
jacsparsity = kron(Dz,Iy,sparse(Mx)) + kron(Dz,sparse(My),Ix) + kron(fJ,Iy,Ix)
@time colorvec = matrix_colors(jacsparsity)

[matbesancon]

what are Mx and My here?

[ChrisRackauckas]

Oh, I cut SciML/SciMLBenchmarks.jl#91 a little too aggressively.

Mx = Tridiagonal([1.0 for i in 1:N-1],[-2.0 for i in 1:N],[1.0 for i in 1:N-1])
My = copy(Mx)

[matbesancon]

For the record, recorded 600 seconds initially (otherwise I'll forget and have to recompute it)

julia> @time colorvec = matrix_colors(jacsparsity)
683.859562 seconds (1.10 M allocations: 121.140 MiB, 0.01% gc time)

[matbesancon]

It's the whole matrix to graph function which is slow, the rest can wait:

julia> @time matrix2graph(jacsparsity, false)
475.198843 seconds (719.90 k allocations: 100.197 MiB, 0.01% gc time)

[matbesancon]

two potential speedups already: dropzeros should be the inplace version, the non-inplace currently used returns a value that is ignored

[matbesancon]

since the indices are known, @inbounds should be used throughout the function

[matbesancon]

#94 is a first take on that, I think there may be a way to improve this even more