# Using Enzyme with LinearSolve, scary errors!

**URL:** <https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250>\
**Category:** General Usage\
**Tags:** sciml\
**Created:** [February 15, 2024, 1:28pm UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250 "2024-02-15T13:28:46Z")\
**Posts on this page:** 11\
**Page:** 1

<div class="post-metadata">

**Author:** ![Larbino1](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/larbino1/32/37831_2.png) [@Larbino1](https://discourse.julialang.org/u/Larbino1)\
**Post date:** [February 15, 2024, 1:28pm UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250/1 "2024-02-15T13:28:46Z")

</div>

I have a MWE of using Enzyme with linearsolve that fails, giving first a warning about using a BLAS fallback (which gets shoved out of my terminal by what follows), then a huge amount of Enzyme `TypeAnalysisDepthLimit` warnings, and the stacktrace below.

It is obvious that work has been done to make Enzyme and LinearSolve compatible, due to the extension that exists, but I haven’t found any examples of it in use. I would really appreciate being pointed to one. Bonus points if it can be done non-allocating!

```julia
using Enzyme
using LinearSolve

function testls(A, b, u)
    oa = OperatorAssumptions(true, condition = LinearSolve.OperatorCondition.WellConditioned)
    prob = LinearProblem{true}(A, b, u0=u)
    linsolve = init(prob, oa)
    solve(linsolve)
    return nothing
end

A = [1. 2.; 3. 4.]
b = [1., 2.]
u = zero(b)
dA = deepcopy(A)
db = deepcopy(b)
du = deepcopy(u)
@time testls(A, b, u)

Enzyme.autodiff(Reverse, testls, Duplicated(A, dA), Duplicated(b, db), Duplicated(u, du))

```

```julia
ERROR: Enzyme execution failed.
Enzyme: Augmented forward pass custom rule Tuple{EnzymeCore.EnzymeRules.ConfigWidth{1, true, false, (false, true)}, Const{typeof(solve!)}, Type{Const{SciMLBase.LinearSolution{Float64, 1, Vector{Float64}, Nothing, LinearSolve.DefaultLinearSolver, LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}, Nothing}}}, Duplicated{LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}}} return type mismatch, expected EnzymeCore.EnzymeRules.AugmentedReturn{SciMLBase.LinearSolution{Float64, 1, Vector{Float64}, Nothing, LinearSolve.DefaultLinearSolver, LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}, Nothing}, Nothing, Any} found EnzymeCore.EnzymeRules.AugmentedReturn{SciMLBase.LinearSolution{Float64, 1, Vector{Float64}, Nothing, LinearSolve.DefaultLinearSolver, LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}, Nothing}, SciMLBase.LinearSolution{Float64, 1, Vector{Float64}, Nothing, LinearSolve.DefaultLinearSolver, LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}, Nothing}, Tuple{Vector{Float64}, Vector{Float64}, LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}, Tuple{Matrix{Float64}}, Tuple{Vector{Float64}}}}
Stacktrace:
 [1] #solve#96
   @ ./deprecated.jl:105

Stacktrace:
  [1] throwerr(cstr::Cstring)
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/jOGYG/src/compiler.jl:1317
  [2] #solve#96
    @ ./deprecated.jl:105
  [3] testls
    @ ~/Dropbox (Cambridge University)/shared_Daniel/code/RobotDynamics.jl/examples/gradients/enzyme_experiments.jl:178 [inlined]
  [4] diffejulia_testls_4370wrap
    @ ~/Dropbox (Cambridge University)/shared_Daniel/code/RobotDynamics.jl/examples/gradients/enzyme_experiments.jl:0
  [5] macro expansion
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/jOGYG/src/compiler.jl:5306 [inlined]
  [6] enzyme_call(::Val{…}, ::Ptr{…}, ::Type{…}, ::Type{…}, ::Val{…}, ::Type{…}, ::Type{…}, ::Const{…}, ::Type{…}, ::Duplicated{…}, ::Duplicated{…}, ::Duplicated{…})
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/jOGYG/src/compiler.jl:4984
  [7] (::Enzyme.Compiler.CombinedAdjointThunk{…})(::Const{…}, ::Duplicated{…}, ::Vararg{…})
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/jOGYG/src/compiler.jl:4926
  [8] autodiff(::ReverseMode{…}, ::Const{…}, ::Type{…}, ::Duplicated{…}, ::Vararg{…})
    @ Enzyme ~/.julia/packages/Enzyme/jOGYG/src/Enzyme.jl:215
  [9] autodiff(::ReverseMode{…}, ::Const{…}, ::Duplicated{…}, ::Duplicated{…}, ::Vararg{…})
    @ Enzyme ~/.julia/packages/Enzyme/jOGYG/src/Enzyme.jl:238
 [10] autodiff(::ReverseMode{…}, ::typeof(testls), ::Duplicated{…}, ::Duplicated{…}, ::Vararg{…})
    @ Enzyme ~/.julia/packages/Enzyme/jOGYG/src/Enzyme.jl:224
 [11] top-level scope
    @ ~/Dropbox (Cambridge University)/shared_Daniel/code/RobotDynamics.jl/examples/gradients/enzyme_experiments.jl:191
Some type information was truncated. Use `show(err)` to see complete types.

```

---

<div class="post-metadata">

**Author:** ![wsmoses](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/wsmoses/32/26497_2.png) [@wsmoses](https://discourse.julialang.org/u/wsmoses)\
**Post date:** [February 15, 2024, 6:57pm UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250/2 "2024-02-15T18:57:40Z")

</div>

That looks to be an error in the LinearSolve.jl EnzymeRule.

Open an issue on LinearSolve.jl and perhaps consider using the builtin solvers in the interim?

---

<div class="post-metadata">

**Author:** ![Larbino1](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/larbino1/32/37831_2.png) [@Larbino1](https://discourse.julialang.org/u/Larbino1)\
**Post date:** [February 29, 2024, 12:20pm UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250/3 "2024-02-29T12:20:58Z")

</div>

Thanks for the reply - I’ve had a go with the builtin solvers and cooked up a minimal example to show what’s going on.

The first case seems to work, but the other 2 fail. this should be fine for now, but can you explain to me why this is the case? Is there an enzyme rule written somewhere for `\` but not for the others?

```julia
function testsolve_backslash!(A, b, u)
    copyto!(u, A\b)
    return u[1]
end

function testsolve_lu!(A, b, u)
    factorization = lu!(A) # This line allocates...
    ldiv!(factorization, copyto!(u, b))
    return u[1]
end

function testsolve_cholesky!(A, b, u)
    factorization = cholesky!(A) # This is non allocating
    ldiv!(factorization, copyto!(u, b))
    return u[1]
end

A = [3. 1.; 1. 2.]
b = [1., 2.]
u = zero(b)

for method in (testsolve_backslash!, testsolve_lu!, testsolve_cholesky!)
    _A, _b, _u = copy(A), copy(b), copy(u)
    @time method(_A, _b, _u)
    @assert _u ≈ A\b
end

A_enz, b_enz, u_enz = Duplicated(copy(A), zero(A)), Duplicated(copy(b), zero(b)), Duplicated(copy(u), zero(u))
# This works
Enzyme.autodiff(Reverse, testsolve_backslash!, A_enz, b_enz, u_enz)
# This errors:
Enzyme.autodiff(Reverse, testsolve_lu!, A_enz, b_enz, u_enz)
# This errors:
Enzyme.autodiff(Reverse, testsolve_cholesky!, A_enz, b_enz, u_enz)

```

---

<div class="post-metadata">

**Author:** ![wsmoses](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/wsmoses/32/26497_2.png) [@wsmoses](https://discourse.julialang.org/u/wsmoses)\
**Post date:** [February 29, 2024, 4:11pm UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250/4 "2024-02-29T16:11:53Z")

</div>

Yeah the first has a rule, but I don’t think the later to do.

@michel_schanen and @simsurace have been looking at linear solve rules, tagging them for visibility. Maybe file an issue for what doesn’t work right now?

---

<div class="post-metadata">

**Author:** ![ChrisRackauckas](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/chrisrackauckas/32/77_2.png) [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)\
**Post date:** [February 29, 2024, 4:33pm UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250/5 "2024-02-29T16:33:10Z")

</div>

I don’t know what that error message is supposed to mean though. Is it that the return type is somehow different?

> [@Larbino1](#):
>
> Thanks for the reply - I’ve had a go with the builtin solvers and cooked up a minimal example to show what’s going on.

That shouldn’t matter because it doesn’t actually differentiate the solve, it boxes that part.

---

<div class="post-metadata">

**Author:** ![wsmoses](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/wsmoses/32/26497_2.png) [@wsmoses](https://discourse.julialang.org/u/wsmoses)\
**Post date:** [February 29, 2024, 8:34pm UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250/6 "2024-02-29T20:34:21Z")

</div>

> [@Larbino1](#):
>
> ```julia
> Enzyme: Augmented forward pass custom rule Tuple{EnzymeCore.EnzymeRules.ConfigWidth{1, true, false, (false, true)}, Const{typeof(solve!)}, Type{Const{SciMLBase.LinearSolution{Float64, 1, Vector{Float64}, Nothing, LinearSolve.DefaultLinearSolver, LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}, Nothing}}}, Duplicated{LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}}} return type mismatch, expected EnzymeCore.EnzymeRules.AugmentedReturn{SciMLBase.LinearSolution{Float64, 1, Vector{Float64}, Nothing, LinearSolve.DefaultLinearSolver, LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}, Nothing}, Nothing, Any} found EnzymeCore.EnzymeRules.AugmentedReturn{SciMLBase.LinearSolution{Float64, 1, Vector{Float64}, Nothing, LinearSolve.DefaultLinearSolver, LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}, Nothing}, SciMLBase.LinearSolution{Float64, 1, Vector{Float64}, Nothing, LinearSolve.DefaultLinearSolver, LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}, Nothing}, Tuple{Vector{Float64}, Vector{Float64}, LinearSolve.LinearCache{Matrix{Float64}, Vector{Float64}, Vector{Float64}, SciMLBase.NullParameters, LinearSolve.DefaultLinearSolver, LinearSolve.DefaultLinearSolverInit{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, LinearAlgebra.QRCompactWY{Float64, Matrix{Float64}, Matrix{Float64}}, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Vector{Int64}}, Nothing, Nothing, Nothing, LinearAlgebra.SVD{Float64, Float64, Matrix{Float64}, Vector{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int32}}, Base.RefValue{Int32}}, Tuple{LinearAlgebra.LU{Float64, Matrix{Float64}, Vector{Int64}}, Base.RefValue{Int64}}, LinearAlgebra.QRPivoted{Float64, Matrix{Float64}, Vector{Float64}, Vector{Int64}}, Krylov.CraigmrSolver{Float64, Float64, Vector{Float64}}, Krylov.LsmrSolver{Float64, Float64, Vector{Float64}}}, IdentityOperator, IdentityOperator, Float64, Bool}, Tuple{Matrix{Float64}}, Tuple{Vector{Float64}}}}
> 
> ```

Removing the giant sciml types to make it easier to read

```julia
Enzyme: Augmented forward pass custom rule [input types for the function], return type mismatch, expected EnzymeCore.EnzymeRules.AugmentedReturn{..., Nothing, Any} found EnzymeCore.EnzymeRules.AugmentedReturn{..., ..., ...}

```

Looks like the rule returned a shadow return, where it was not requested. You can do `EnzymeCore.needs_shadow(config)` to determine whether you should return a shadow. [Enzyme.jl/lib/EnzymeCore/src/rules.jl at e665d895b7eb0cdb7c3bb230d9272425a1918219 · EnzymeAD/Enzyme.jl · GitHub](https://github.com/EnzymeAD/Enzyme.jl/blob/e665d895b7eb0cdb7c3bb230d9272425a1918219/lib/EnzymeCore/src/rules.jl#L37)

---

<div class="post-metadata">

**Author:** ![ChrisRackauckas](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/chrisrackauckas/32/77_2.png) [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)\
**Post date:** [March 1, 2024, 6:57am UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250/7 "2024-03-01T06:57:41Z")

</div>

Can someone open an issue for this? I don’t want to forget/lose it

---

<div class="post-metadata">

**Author:** ![simsurace](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/simsurace/32/30216_2.png) [@simsurace](https://discourse.julialang.org/u/simsurace)\
**Post date:** [March 2, 2024, 7:17am UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250/8 "2024-03-02T07:17:57Z")

</div>

Does the third example work with [Fix rules for `cholesky` and `ldiv!` on `Cholesky` by simsurace · Pull Request #1307 · EnzymeAD/Enzyme.jl · GitHub](https://github.com/EnzymeAD/Enzyme.jl/pull/1307)?

---

<div class="post-metadata">

**Author:** ![simsurace](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/simsurace/32/30216_2.png) [@simsurace](https://discourse.julialang.org/u/simsurace)\
**Post date:** [March 2, 2024, 11:14pm UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250/9 "2024-03-02T23:14:01Z")

</div>

Indeed it doesn’t, because the rule is only for `cholesky`, but not `cholesky!`. Apart from that, there are still errors in that PR, need to revisit once more.

There is currently no rule for `lu!`.

---

<div class="post-metadata">

**Author:** ![Larbino1](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/larbino1/32/37831_2.png) [@Larbino1](https://discourse.julialang.org/u/Larbino1)\
**Post date:** [March 4, 2024, 1:32pm UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250/10 "2024-03-04T13:32:38Z")

</div>

[I’ve made an issue here](https://github.com/SciML/LinearSolve.jl/issues/479) - it’s my first every on an open project, let me know if I’ve messed up anything.

---

<div class="post-metadata">

**Author:** ![ChrisRackauckas](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/chrisrackauckas/32/77_2.png) [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)\
**Post date:** [March 4, 2024, 3:13pm UTC](https://discourse.julialang.org/t/using-enzyme-with-linearsolve-scary-errors/110250/11 "2024-03-04T15:13:26Z")

</div>

That looks good, thanks