Thomas Christensen

I tested the performance impact of this just now: as far as I can see, there is a speed-up for `Hermitian{ComplexF64, ...}` matrices but a slow-down for `Hermitian{Float64, ...}` matrices...

Shouldn't this issue have remained opened? It's not a terribly urgent thing but might as well track that `(\)(::StaticArray{3,3}, ::StaticVector{3}) is unstable for ill-conditioned matrices and could potentially be improved.

The reason for not calling into LAPACK is speed. The whole point of the package is to exploit the knowledge about the size of the matrices (and their storage on...

I don't think calling out to LAPACK is a good idea; it'd be a huge performance gotcha for the vast majority of cases. Potentially, hand-rolled Gauss-Jordan elimination or similar could...

Ah, the fallback for larger `StaticMatrix` in fact already [goes via an LU approach](https://github.com/JuliaArrays/StaticArrays.jl/blame/ff8ed9b64f41833fe22af81bd2859d5a0b9c3d18/src/inv.jl#L74-L85), using StaticArrays-specific LU factorization (since https://github.com/JuliaArrays/StaticArrays.jl/pull/424) - so the following is pretty much as fast as...

Disregarding that it would not be a good idea to work with such large matrices, a simple implementation of this could e.g. be: ```jl function repeat(v::SVector{N,T}, vM::Val{M}) where {N,M,T} SMatrix{N,M,T,N*M}(ntuple(i...

I tested various permutations of this quite extensively yesterday (pretty sure I also tested the `A*U` shuffle). Puzzlingly, I found that almost all the compilation time comes from the presence...

It would be really great to port the BlossomV dependency since it has an [awful licence](https://github.com/JuliaGraphs/LightGraphsMatching.jl/pull/12#issuecomment-434390768): in practice, the GraphsMatching.jl package only works as long as the BlossomV code is...

I looked at this before with the same idea: unfortunately, the StaticArrays.jl dependency of ArnoldiMethod.jl does not just appear to be a simple extension (and so cannot just be converted...

> almost all of the rest were unnecessary I don't completely understand this: doesn't this imply that *some* of them were necessary in some capacity? In the diff, unless I...