Error in Using DataInterpolations.jl and Enzyme.jl for Real-Time Interpolation and Gradient Computation in ODE Problems

[chooron]

Describe the bug 🐞

Hello, I currently have an ODE problem where I need to use DataInterpolations.jl to interpolate and obtain real-time values, which are then used for computing du. Next, I need to use Enzyme.jl to compute the gradients of parameters during the computation process. According to the suggestion in ChrisRackauckas's reply, I opened an issue here for this bug.

Expected behavior

Successful execution yields results similar to ForwardDiff

Minimal Reproducible Example 👇

Without MRE, we would only be able to help you to a limited extent, and attention to the issue would be limited. to know more about MRE refer to wikipedia and stackoverflow.

using DifferentiationInterface, Enzyme
using DifferentialEquations, DataInterpolations, SciMLSensitivity
using ComponentArrays

t_data = collect(0.0:1.0:11.0)
alpha_values = rand(length(t_data))

function loss_function(params, config)

    interpolator = config.interpolator_type(
        alpha_values,
        config.t_data,
    )

    function my_ode!(du, u, p, t)
        alpha_t = interpolator(t)
        du[1] = -alpha_t * u[1] * p[1] + p[2]
        return nothing
    end

    u0 = [10.0]
    tspan = (1.0, 10.0)

    prob = ODEProblem(my_ode!, u0, tspan, params, sensealg=GaussAdjoint(autojacvec=EnzymeVJP()))
    sol = solve(prob, Tsit5(), save_everystep=false, save_start=false)
    return Array(sol) |> sum
end

config = (
    interpolator_type=ConstantInterpolation,
    t_data=t_data,
)

initial_alpha_values = rand(length(t_data))
params = [1.0, 2.0]
target_func(p) = loss_function(p, config)
target_func(params)
ad_backend = AutoEnzyme(mode=Enzyme.set_runtime_activity(Enzyme.Reverse),)
val, grad = value_and_gradient(target_func, ad_backend, params)

Error & Stacktrace ⚠️
I tried using ConstantInterpolation or LinearInterpolation for interpolation, but both resulted in the following issues:

Please submit a bug report with steps to reproduce this fault, and any error messages that follow (in their entirety). Thanks.
Exception: EXCEPTION_ACCESS_VIOLATION at 0x7ffc3391e40d -- gc_try_setmark_tag at C:/workdir/src\gc.c:826 [inlined]
gc_try_claim_and_push at C:/workdir/src\gc.c:2094 [inlined]
gc_mark_obj16 at C:/workdir/src\gc.c:2146
in expression starting at e:\JlCode\HydroModels\dev\tmp2.jl:39
gc_try_setmark_tag at C:/workdir/src\gc.c:826 [inlined]
gc_try_claim_and_push at C:/workdir/src\gc.c:2094 [inlined]
gc_mark_obj16 at C:/workdir/src\gc.c:2146
gc_mark_outrefs at C:/workdir/src\gc.c:2895 [inlined]
gc_mark_loop_serial_ at C:/workdir/src\gc.c:2938
gc_mark_loop_serial at C:/workdir/src\gc.c:2961
gc_mark_loop at C:/workdir/src\gc.c:3143 [inlined]
_jl_gc_collect at C:/workdir/src\gc.c:3532
ijl_gc_collect at C:/workdir/src\gc.c:3893
maybe_collect at C:/workdir/src\gc.c:926 [inlined]
jl_gc_pool_alloc_inner at C:/workdir/src\gc.c:1319
jl_gc_pool_alloc_noinline at C:/workdir/src\gc.c:1386 [inlined]
jl_gc_alloc_ at C:/workdir/src\julia_internal.h:523 [inlined]
jl_gc_alloc at C:/workdir/src\gc.c:3946
_new_genericmemory_ at C:/workdir/src\genericmemory.c:56 [inlined]
jl_alloc_genericmemory at C:/workdir/src\genericmemory.c:99
GenericMemory at .\boot.jl:516 [inlined]
array_new_memory at .\array.jl:1058 [inlined]
#134 at .\array.jl:1129
_growend! at .\array.jl:1116 [inlined]
push! at .\array.jl:1261 [inlined]
compute_basic_blocks at .\compiler/ssair\ir.jl:144
InferenceState at .\compiler\inferencestate.jl:288
InferenceState at .\compiler\inferencestate.jl:500
typeinf_edge at .\compiler\typeinfer.jl:913
abstract_call_method at .\compiler\abstractinterpretation.jl:660
abstract_call_gf_by_type at .\compiler\abstractinterpretation.jl:102
abstract_call_known at .\compiler\abstractinterpretation.jl:2200
abstract_call at .\compiler\abstractinterpretation.jl:2282
abstract_call at .\compiler\abstractinterpretation.jl:2275
abstract_call at .\compiler\abstractinterpretation.jl:2423
abstract_eval_call at .\compiler\abstractinterpretation.jl:2438
abstract_eval_statement_expr at .\compiler\abstractinterpretation.jl:2454
abstract_eval_statement at .\compiler\abstractinterpretation.jl:2752
abstract_eval_basic_statement at .\compiler\abstractinterpretation.jl:3068
typeinf_local at .\compiler\abstractinterpretation.jl:3331
typeinf_nocycle at .\compiler\abstractinterpretation.jl:3413
_typeinf at .\compiler\typeinfer.jl:244
typeinf at .\compiler\typeinfer.jl:215
typeinf_edge at .\compiler\typeinfer.jl:923
abstract_call_method at .\compiler\abstractinterpretation.jl:660
abstract_call_gf_by_type at .\compiler\abstractinterpretation.jl:102
abstract_call_known at .\compiler\abstractinterpretation.jl:2200
abstract_call at .\compiler\abstractinterpretation.jl:2282
abstract_call at .\compiler\abstractinterpretation.jl:2275
abstract_call at .\compiler\abstractinterpretation.jl:2423
abstract_eval_call at .\compiler\abstractinterpretation.jl:2438
abstract_eval_statement_expr at .\compiler\abstractinterpretation.jl:2454
abstract_eval_statement at .\compiler\abstractinterpretation.jl:2752
abstract_eval_basic_statement at .\compiler\abstractinterpretation.jl:3044
typeinf_local at .\compiler\abstractinterpretation.jl:3331
typeinf_nocycle at .\compiler\abstractinterpretation.jl:3413
_typeinf at .\compiler\typeinfer.jl:244
typeinf at .\compiler\typeinfer.jl:215
typeinf_edge at .\compiler\typeinfer.jl:923
abstract_call_method at .\compiler\abstractinterpretation.jl:660
abstract_call_gf_by_type at .\compiler\abstractinterpretation.jl:102
abstract_call_unknown at .\compiler\abstractinterpretation.jl:2269
abstract_call at .\compiler\abstractinterpretation.jl:2279
abstract_call at .\compiler\abstractinterpretation.jl:2275
abstract_call at .\compiler\abstractinterpretation.jl:2423
abstract_eval_call at .\compiler\abstractinterpretation.jl:2438
abstract_eval_statement_expr at .\compiler\abstractinterpretation.jl:2454
abstract_eval_statement at .\compiler\abstractinterpretation.jl:2752
abstract_eval_basic_statement at .\compiler\abstractinterpretation.jl:3068
typeinf_local at .\compiler\abstractinterpretation.jl:3331
typeinf_nocycle at .\compiler\abstractinterpretation.jl:3413
_typeinf at .\compiler\typeinfer.jl:244
typeinf at .\compiler\typeinfer.jl:215
typeinf_ext at .\compiler\typeinfer.jl:1101
typeinf_ext_toplevel at .\compiler\typeinfer.jl:1139
typeinf_ext_toplevel at .\compiler\typeinfer.jl:1135
jfptr_typeinf_ext_toplevel_41140.1 at C:\Users\OMEN\.julia\juliaup\julia-1.11.5+0.x64.w64.mingw32\lib\julia\sys.dll (unknown line)
_jl_invoke at C:/workdir/src\gf.c:2948 [inlined]
ijl_apply_generic at C:/workdir/src\gf.c:3125 [inlined]
jl_apply at C:/workdir/src\julia.h:2157 [inlined]
jl_type_infer at C:/workdir/src\gf.c:390
jl_generate_fptr_impl at C:/workdir/src\jitlayers.cpp:519
jl_compile_method_internal at C:/workdir/src\gf.c:2536
jl_compile_method_internal at C:/workdir/src\gf.c:2946 [inlined]
_jl_invoke at C:/workdir/src\gf.c:2940 [inlined]
ijl_invoke at C:/workdir/src\gf.c:2955
active_reg_inner at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:447
unknown function (ip: 00000204b273f9b0)
unknown function (ip: 00000204b273f34d)
unknown function (ip: 00000204b273efb1)
Merger at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:35 [inlined]
macro expansion at .\ntuple.jl:72 [inlined]
ntuple at .\ntuple.jl:69 [inlined]
active_reg_inner at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:447
unknown function (ip: 00000204b273eee0)
unknown function (ip: 00000204b273e60d)
unknown function (ip: 00000204b273e5a1)
Merger at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:35 [inlined]
ntuple at .\ntuple.jl:48 [inlined]
active_reg_inner at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:447
unknown function (ip: 00000204b273e4d0)
unknown function (ip: 00000204b273dc7d)
unknown function (ip: 00000204b273db01)
Merger at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:35 [inlined]
macro expansion at .\ntuple.jl:72 [inlined]
ntuple at .\ntuple.jl:69 [inlined]
active_reg_inner at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:447
unknown function (ip: 00000204b273da30)
unknown function (ip: 00000204b273b15d)
unknown function (ip: 00000204b273aed1)
Merger at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:35 [inlined]
macro expansion at .\ntuple.jl:72 [inlined]
ntuple at .\ntuple.jl:69 [inlined]
active_reg_inner at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:447
unknown function (ip: 00000204b273ae00)
unknown function (ip: 00000204b27397ed)
unknown function (ip: 00000204b2739561)
Merger at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:35 [inlined]
macro expansion at .\ntuple.jl:72 [inlined]
ntuple at .\ntuple.jl:69 [inlined]
active_reg_inner at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:447
unknown function (ip: 00000204b27391f0)
active_reg_inner at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:206 [inlined]
active_reg_inner at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:206 [inlined]
active_reg_inner at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:206 [inlined]
guaranteed_const_nongen at D:\Julia\packages\packages\Enzyme\lmg7D\src\analyses\activity.jl:487 [inlined]
thunkbase at D:\Julia\packages\packages\Enzyme\lmg7D\src\compiler.jl:5885
thunk_generator at D:\Julia\packages\packages\Enzyme\lmg7D\src\compiler.jl:6122
jfptr_thunk_generator_36804 at D:\Julia\packages\compiled\v1.11\Enzyme\G1p5n_WkAWM.dll (unknown line)
jl_call_staged at C:/workdir/src\method.c:601
ijl_code_for_staged at C:/workdir/src\method.c:656
get_staged at .\compiler\utilities.jl:123
retrieve_code_info at .\compiler\utilities.jl:135 [inlined]
InferenceState at .\compiler\inferencestate.jl:497
typeinf_edge at .\compiler\typeinfer.jl:913
abstract_call_method at .\compiler\abstractinterpretation.jl:660
abstract_call_gf_by_type at .\compiler\abstractinterpretation.jl:102
abstract_call_known at .\compiler\abstractinterpretation.jl:2200
abstract_call at .\compiler\abstractinterpretation.jl:2282
abstract_call at .\compiler\abstractinterpretation.jl:2275
abstract_call at .\compiler\abstractinterpretation.jl:2423
abstract_eval_call at .\compiler\abstractinterpretation.jl:2438
abstract_eval_statement_expr at .\compiler\abstractinterpretation.jl:2454
abstract_eval_statement at .\compiler\abstractinterpretation.jl:2752
abstract_eval_basic_statement at .\compiler\abstractinterpretation.jl:3068
typeinf_local at .\compiler\abstractinterpretation.jl:3331
typeinf_nocycle at .\compiler\abstractinterpretation.jl:3413
_typeinf at .\compiler\typeinfer.jl:244
typeinf at .\compiler\typeinfer.jl:215
typeinf_ext at .\compiler\typeinfer.jl:1101
typeinf_ext_toplevel at .\compiler\typeinfer.jl:1139
typeinf_ext_toplevel at .\compiler\typeinfer.jl:1135
jfptr_typeinf_ext_toplevel_41140.1 at C:\Users\OMEN\.julia\juliaup\julia-1.11.5+0.x64.w64.mingw32\lib\julia\sys.dll (unknown line)
_jl_invoke at C:/workdir/src\gf.c:2948 [inlined]
ijl_apply_generic at C:/workdir/src\gf.c:3125 [inlined]
jl_apply at C:/workdir/src\julia.h:2157 [inlined]
jl_type_infer at C:/workdir/src\gf.c:390
jl_generate_fptr_impl at C:/workdir/src\jitlayers.cpp:519
jl_compile_method_internal at C:/workdir/src\gf.c:2536
jl_compile_method_internal at C:/workdir/src\gf.c:2946 [inlined]
_jl_invoke at C:/workdir/src\gf.c:2940 [inlined]
ijl_apply_generic at C:/workdir/src\gf.c:3125
#solve#43 at D:\Julia\packages\packages\DiffEqBase\AfBQT\src\solve.jl:1083
Allocations: 225954686 (Pool: 225949721; Big: 4965); GC: 72

Environment (please complete the following information):

• Output of using Pkg; Pkg.status()

  [82cc6244] DataInterpolations v8.6.0
  [0c46a032] DifferentialEquations v7.16.1
  [a0c0ee7d] DifferentiationInterface v0.7.7
  [7da242da] Enzyme v0.13.73
  [1ed8b502] SciMLSensitivity v7.89.0

• Output of versioninfo()

Julia Version 1.11.5
Commit 760b2e5b73 (2025-04-14 06:53 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 24 × 12th Gen Intel(R) Core(TM) i9-12900HX
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, alderlake)
Threads: 1 default, 0 interactive, 1 GC (on 24 virtual cores)
Environment:
  JULIA_DEPOT_PATH = D:\Julia\packages
  JULIA_EDITOR = code
  JULIA_VSCODE_REPL = 1
  JULIA_NUM_THREADS = 0

Additional context

Add any other context about the problem here.

[ChrisRackauckas]

First step here is to make an MWE. It doesn't look like SciMLSensitivity is related, so see if you can recreate it with differentiating:

    interpolator = config.interpolator_type(
        alpha_values,
        config.t_data,
    )

    function my_ode!(du, u, p, t)
        alpha_t = interpolator(t)
        du[1] = -alpha_t * u[1] * p[1] + p[2]
        return nothing
    end

[chooron]

Okay, I have rewritten an MWE based on your suggestions to test whether DataInterpolations.jl supports differentiation with Enzyme, as shown below:

using DifferentialEquations
using DataInterpolations
using DifferentiationInterface
using Enzyme

alpha_values = rand(100)
config = (interpolator_type=LinearInterpolation, t_data=0.1:0.1:10)

interpolator = config.interpolator_type(
    alpha_values,
    config.t_data,
)

function my_ode!(du, u, p, t)
    alpha_t = interpolator(t)
    du[1] = -alpha_t * u[1] * p[1] + p[2]
    return nothing
end

function target_func(p)
    du = zeros(1)
    u = zeros(1)
    for t in 0.1:0.1:10
        my_ode!(du, u, p, t)
    end
    return sum(du)
end

ad_backend = AutoEnzyme(mode=Enzyme.set_runtime_activity(Enzyme.Reverse),)
val, grad = value_and_gradient(target_func, ad_backend, [0.2, 0.3])

Currently, Enzyme can obtain the gradient calculation results:

(0.3, [0.0, 1.0])

It seems that DataInterpolations.jl can support gradient computation when used alone, but issues arise when it is used together with DifferentialEquations.jl.

[ChrisRackauckas]

Does this require runtime activity? In the solver it's not setting that.

[chooron]

In fact, I'm not sure if I need to set mode=Enzyme.set_runtime_activity(Enzyme.Reverse). When I directly set mode=Enzyme.Reverse, I added this code based on the error message:

ERROR: 
If you are using Enzyme by selecting the `AutoEnzyme` object from ADTypes, you may want to try setting the `mode` option as follows:

        AutoEnzyme(; mode=Enzyme.set_runtime_activity(Enzyme.Forward))
        AutoEnzyme(; mode=Enzyme.set_runtime_activity(Enzyme.Reverse))

This hint appears because DifferentiationInterface and Enzyme are both loaded. It does not necessarily imply that Enzyme is being called through DifferentiationInterface.

Constant memory is stored (or returned) to a differentiable variable.
As a result, Enzyme cannot provably ensure correctness and throws this error.
This might be due to the use of a constant variable as temporary storage for active memory (https://enzyme.mit.edu/julia/stable/faq/#Runtime-Activity).
If Enzyme should be able to prove this use non-differentable, open an issue!
To work around this issue, either:
 a) rewrite this variable to not be conditionally active (fastest, but requires a code change), or
 b) set the Enzyme mode to turn on runtime activity (e.g. autodiff(set_runtime_activity(Reverse), ...) ). This will maintain correctness, but may slightly reduce performance.
Mismatched activity for:   store {} addrspace(10)* %2, {} addrspace(10)** %sret_return.repack, align 8, !dbg !163, !noalias !164 const val: {} addrspace(10)* %2
 value=Unknown object of type Vector{Float64}
Stacktrace:
 [1] #munge_data#41
   @ D:\Julia\packages\packages\DataInterpolations\LFagy\src\interpolation_utils.jl:109

Stacktrace:
 [1] #munge_data#41
   @ D:\Julia\packages\packages\DataInterpolations\LFagy\src\interpolation_utils.jl:109

My only thought was to use Enzyme's Reverse mode.

[chooron]

It appears this error occurs because the interpolator model in my code is constructed within the loss function. This leads to the belief that Constant memory is stored (or returned) to a differentiable variable. Therefore, I have moved the interpolator model outside of the function and then set the mode to Enzyme.Reverse, as shown in the following code:

using DifferentiationInterface, Enzyme
using DifferentialEquations, DataInterpolations, SciMLSensitivity

t_data = collect(0.0:1.0:11.0)
alpha_values = rand(length(t_data))

config = (
    interpolator_type=ConstantInterpolation,
    t_data=t_data,
)

interpolator = config.interpolator_type(
    alpha_values,
    config.t_data,
)

function loss_function(params)

    function my_ode!(du, u, p, t)
        alpha_t = interpolator(t)
        du[1] = -alpha_t * u[1] * p[1] + p[2]
        return nothing
    end

    u0 = [10.0]
    tspan = (1.0, 10.0)

    prob = ODEProblem(my_ode!, u0, tspan, params, sensealg=GaussAdjoint(autojacvec=EnzymeVJP()))
    sol = solve(prob, Tsit5(), save_everystep=false, save_start=false)
    return Array(sol) |> sum
end


params = [1.0, 2.0]

target_func(p) = loss_function(p)
ad_backend = AutoEnzyme(mode=Enzyme.Reverse)
val, grad = value_and_gradient(target_func, ad_backend, params)

However, when I executed the code, the same initial error still occurred.

[ChrisRackauckas]

Interesting maybe @wsmoses can give a pointer to help out.