Unnecessary scalar indexing in logdet(C::Cholesky)

[sharanry]

Related: JuliaGaussianProcesses/AbstractGPs.jl#329
https://github.com/JuliaLang/julia/blob/master/stdlib/LinearAlgebra/src/cholesky.jl#L672

[sharanry]

Can't we do sum(log, real.(diag(C.factors))) ?

[dkarrasch]

Have you tried it with allowscalar(false) (or whatever the correct syntax is)? I don't think diag(C.factors) does anything else than scalar indexing and collecting the resulting vector. That would be worse in general, because it adds an allocation.

[stevengj]

Seems like a minimal change is:

 @inbounds for i in diagind(C.factors)
    dd += log(real(C.factors[i]))
end

Probably that doesn't help with OP's case, however?