Test latest Enzyme

[ChrisRackauckas]

No description provided.

[ChrisRackauckas]

https://github.com/SciML/SciMLSensitivity.jl/actions/runs/15832710852/job/44629049370?pr=1224#step:6:1369

https://github.com/SciML/SciMLSensitivity.jl/actions/runs/15832710852/job/44629049361?pr=1224#step:6:1124

It's two GC errors, one is a a GC corruption and the other is a segfault. @wsmoses @gbaraldi

[ChrisRackauckas]

Reproducer 1:

using SciMLSensitivity, OrdinaryDiffEq, Enzyme

u0 = [0.0]
p = [2.0, 1.0]
prob = NonlinearProblem((du, u, p) -> du[1] = u[1] - p[1] + p[2], u0, p)

function test_loss(p, prob, alg)
    _prob = remake(prob, p = p)
    sol = sum(solve(_prob, alg,
        sensealg = SteadyStateAdjoint(autojacvec = ReverseDiffVJP())))
    return sol
end

function enzyme_gradient(p, prob, alg)
    dp = Enzyme.make_zero(p)
    dprob = Enzyme.make_zero(prob)
    Enzyme.autodiff(Reverse, test_loss, Active, Duplicated(p, dp),
        Duplicated(prob, dprob), Const(alg))
    return dp
end
dp1_enzyme = enzyme_gradient(p, prob, NewtonRaphson())

Reproducer 2, I think, it's a bit hard to read the GC corruption because it messes with the line numbers, but it seems like it's this:

using OrdinaryDiffEq, SciMLSensitivity, Enzyme
Enzyme.API.typeWarning!(false)

odef(du, u, p, t) = du .= u .* p
const prob = ODEProblem(odef, [2.0], (0.0, 1.0), [3.0])

struct senseloss0{T}
    sense::T
end
function (f::senseloss0)(u0p)
    prob = ODEProblem{true}(odef, u0p[1:1], (0.0, 1.0), u0p[2:2])
    sum(solve(prob, Tsit5(), abstol = 1e-12, reltol = 1e-12, saveat = 0.1))
end
u0p = [2.0, 3.0]
du0p = zeros(2)
Enzyme.autodiff(Reverse, senseloss0(InterpolatingAdjoint()), Active, Duplicated(u0p, du0p))

[ChrisRackauckas]

Both are related to Enzyme on the outside, not the EnzymeVJP. So it's GC corruption related to https://github.com/SciML/DiffEqBase.jl/blob/master/ext/DiffEqBaseEnzymeExt.jl. But I'm not sure what weird stuff is in that rule...