Gradients are dropped when using a sparse cholesky

[ptiede]

I just realized I get zeros in my gradient when using a sparse cholesky. Here is a MWE

using Enzyme
using SparseArrays
using LinearAlgebra
using FiniteDifferences


function f(C, x)
    y = C\x
    return sum(abs2, y)
end

S = sprand(4, 4, 0.5)
C = cholesky(S'S + 0.1I)

x = ones(4)
dx = zero(x)
autodiff(Reverse, f, Active, Const(C), Duplicated(x, dx))

fdm = central_fdm(5,1)
gfd, = grad(fdm, x->f(C, x), x)

@show gfd
gfd = [-28.632697807114134, 119.34259912937537, -11.822709134186741, -0.4882945852142454]

@show dx
dx = [0.0, 0.0, 0.0, 0.0]