Regression when using ForwardDiff.gradient and ForwardDiff.Hessian with 3.19

[sebapersson]

Describe the example

Starting from what I believe is v3.20, in PEtab.jl I noticed a drastic performance decrease (factor 10x) when computing either the gradient or Hessian of a cost-function where an ODEProblem is solved with Rodas5P

Minimal Reproducible Example 👇

To fully reproduce the code, you need to run with different versions of LinearSolve (I used 3.18.2 and 3.25.1).

using OrdinaryDiffEqRosenbrock, ForwardDiff, LinearSolve, BenchmarkTools

function model(du, u, p, t)
    du[1] = p[1] * u[1] - p[2]
end

function my_loss(_p)
    oprob_new = remake(oprob; p = _p)
    sol = solve(oprob_new, Rodas5P(), abstol = 1e-8, reltol = 1e-8)
    return sum(sol)
end

p = [1.0, 2.0]
u0 = [1.0]
oprob = ODEProblem(model, u0, [0.0, 100.0], p)

my_loss(p)
b1 = @btime ForwardDiff.hessian($my_loss, $p)
b2 = @btime ForwardDiff.gradient($my_loss, $p)

Error & Stacktrace ⚠️

# LinearSolve v3.25 grad: around 60-61ms
# LinearSolve v3.25 hess: around 230-260ms

# LinearSolve v3.18 grad: around 600µs
# LinearSolve v3.18 hess: around 1ms

• Output of using Pkg; Pkg.status()

  [6e4b80f9] BenchmarkTools v1.6.0
  [f6369f11] ForwardDiff v1.0.1
⌃ [7ed4a6bd] LinearSolve v3.18.2
  [43230ef6] OrdinaryDiffEqRosenbrock v1.12.0

• Output of using Pkg; Pkg.status(; mode = PKGMODE_MANIFEST)

  [47edcb42] ADTypes v1.16.0
  [7d9f7c33] Accessors v0.1.42
  [79e6a3ab] Adapt v4.3.0
  [4fba245c] ArrayInterface v7.19.0
  [4c555306] ArrayLayouts v1.11.2
  [6e4b80f9] BenchmarkTools v1.6.0
  [62783981] BitTwiddlingConvenienceFunctions v0.1.6
  [2a0fbf3d] CPUSummary v0.2.7
  [d360d2e6] ChainRulesCore v1.25.2
  [fb6a15b2] CloseOpenIntervals v0.1.13
  [38540f10] CommonSolve v0.2.4
  [bbf7d656] CommonSubexpressions v0.3.1
  [f70d9fcc] CommonWorldInvalidations v1.0.0
  [34da2185] Compat v4.18.0
  [a33af91c] CompositionsBase v0.1.2
  [2569d6c7] ConcreteStructs v0.2.3
  [187b0558] ConstructionBase v1.6.0
  [adafc99b] CpuId v0.3.1
  [a8cc5b0e] Crayons v4.1.1
  [9a962f9c] DataAPI v1.16.0
⌅ [864edb3b] DataStructures v0.18.22
  [e2d170a0] DataValueInterfaces v1.0.0
  [2b5f629d] DiffEqBase v6.182.0
  [163ba53b] DiffResults v1.1.0
  [b552c78f] DiffRules v1.15.1
  [a0c0ee7d] DifferentiationInterface v0.7.4
  [ffbed154] DocStringExtensions v0.9.5
  [4e289a0a] EnumX v1.0.5
  [f151be2c] EnzymeCore v0.8.12
  [e2ba6199] ExprTools v0.1.10
  [55351af7] ExproniconLite v0.10.14
  [7034ab61] FastBroadcast v0.3.5
  [9aa1b823] FastClosures v0.3.2
  [a4df4552] FastPower v1.1.3
  [1a297f60] FillArrays v1.13.0
  [6a86dc24] FiniteDiff v2.27.0
  [f6369f11] ForwardDiff v1.0.1
  [069b7b12] FunctionWrappers v1.1.3
  [77dc65aa] FunctionWrappersWrappers v0.1.3
  [46192b85] GPUArraysCore v0.2.0
  [615f187c] IfElse v0.1.1
  [3587e190] InverseFunctions v0.1.17
  [92d709cd] IrrationalConstants v0.2.4
  [82899510] IteratorInterfaceExtensions v1.0.0
  [692b3bcd] JLLWrappers v1.7.1
  [682c06a0] JSON v0.21.4
  [ae98c720] Jieko v0.2.1
  [ba0b0d4f] Krylov v0.10.1
  [b964fa9f] LaTeXStrings v1.4.0
  [10f19ff3] LayoutPointers v0.1.17
  [5078a376] LazyArrays v2.6.2
⌃ [7ed4a6bd] LinearSolve v3.18.2
  [2ab3a3ac] LogExpFunctions v0.3.29
  [1914dd2f] MacroTools v0.5.16
  [d125e4d3] ManualMemory v0.1.8
  [2e0e35c7] Moshi v0.3.7
  [46d2c3a1] MuladdMacro v0.2.4
  [77ba4419] NaNMath v1.1.3
  [bac558e1] OrderedCollections v1.8.1
  [bbf590c4] OrdinaryDiffEqCore v1.26.2
  [4302a76b] OrdinaryDiffEqDifferentiation v1.10.1
  [43230ef6] OrdinaryDiffEqRosenbrock v1.12.0
  [d96e819e] Parameters v0.12.3
  [69de0a69] Parsers v2.8.3
  [f517fe37] Polyester v0.7.18
  [1d0040c9] PolyesterWeave v0.2.2
⌅ [aea7be01] PrecompileTools v1.2.1
  [21216c6a] Preferences v1.4.3
  [08abe8d2] PrettyTables v2.4.0
  [3cdcf5f2] RecipesBase v1.3.4
  [731186ca] RecursiveArrayTools v3.36.0
  [189a3867] Reexport v1.2.2
  [ae029012] Requires v1.3.1
  [7e49a35a] RuntimeGeneratedFunctions v0.5.15
  [94e857df] SIMDTypes v0.1.0
  [0bca4576] SciMLBase v2.108.0
  [c0aeaf25] SciMLOperators v1.4.0
  [53ae85a6] SciMLStructures v1.7.0
  [efcf1570] Setfield v1.1.2
  [ce78b400] SimpleUnPack v1.1.0
  [0a514795] SparseMatrixColorings v0.4.21
  [276daf66] SpecialFunctions v2.5.1
  [aedffcd0] Static v1.2.0
  [0d7ed370] StaticArrayInterface v1.8.0
  [90137ffa] StaticArrays v1.9.14
  [1e83bf80] StaticArraysCore v1.4.3
  [10745b16] Statistics v1.11.1
  [7792a7ef] StrideArraysCore v0.5.7
  [892a3eda] StringManipulation v0.4.1
  [2efcf032] SymbolicIndexingInterface v0.3.42
  [3783bdb8] TableTraits v1.0.1
  [bd369af6] Tables v1.12.1
  [8290d209] ThreadingUtilities v0.5.5
  [781d530d] TruncatedStacktraces v1.4.0
  [3a884ed6] UnPack v1.0.2
  [1d5cc7b8] IntelOpenMP_jll v2025.2.0+0
  [856f044c] MKL_jll v2025.2.0+0
  [efe28fd5] OpenSpecFun_jll v0.5.6+0
  [1317d2d5] oneTBB_jll v2022.0.0+0
  [0dad84c5] ArgTools v1.1.2
  [56f22d72] Artifacts v1.11.0
  [2a0f44e3] Base64 v1.11.0
  [ade2ca70] Dates v1.11.0
  [8ba89e20] Distributed v1.11.0
  [f43a241f] Downloads v1.6.0
  [7b1f6079] FileWatching v1.11.0
  [9fa8497b] Future v1.11.0
  [b77e0a4c] InteractiveUtils v1.11.0
  [4af54fe1] LazyArtifacts v1.11.0
  [b27032c2] LibCURL v0.6.4
  [76f85450] LibGit2 v1.11.0
  [8f399da3] Libdl v1.11.0
  [37e2e46d] LinearAlgebra v1.11.0
  [56ddb016] Logging v1.11.0
  [d6f4376e] Markdown v1.11.0
  [a63ad114] Mmap v1.11.0
  [ca575930] NetworkOptions v1.2.0
  [44cfe95a] Pkg v1.11.0
  [de0858da] Printf v1.11.0
  [9abbd945] Profile v1.11.0
  [9a3f8284] Random v1.11.0
  [ea8e919c] SHA v0.7.0
  [9e88b42a] Serialization v1.11.0
  [6462fe0b] Sockets v1.11.0
  [2f01184e] SparseArrays v1.11.0
  [fa267f1f] TOML v1.0.3
  [a4e569a6] Tar v1.10.0
  [cf7118a7] UUIDs v1.11.0
  [4ec0a83e] Unicode v1.11.0
  [e66e0078] CompilerSupportLibraries_jll v1.1.1+0
  [deac9b47] LibCURL_jll v8.6.0+0
  [e37daf67] LibGit2_jll v1.7.2+0
  [29816b5a] LibSSH2_jll v1.11.0+1
  [c8ffd9c3] MbedTLS_jll v2.28.6+0
  [14a3606d] MozillaCACerts_jll v2023.12.12
  [4536629a] OpenBLAS_jll v0.3.27+1
  [05823500] OpenLibm_jll v0.8.5+0
  [bea87d4a] SuiteSparse_jll v7.7.0+0
  [83775a58] Zlib_jll v1.2.13+1
  [8e850b90] libblastrampoline_jll v5.11.0+0
  [8e850ede] nghttp2_jll v1.59.0+0
  [3f19e933] p7zip_jll v17.4.0+2
Info Packages marked with ⌃ and ⌅ have new versions available. Those with ⌃ may be upgradable, but those with ⌅ are restricted by compatibility constraints from upgrading. To see why use `status --outdated -m`

• Output of versioninfo()

Julia Version 1.11.6
Commit 9615af0f269 (2025-07-09 12:58 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 14 × Intel(R) Core(TM) Ultra 7 165U
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, alderlake)
Threads: 1 default, 0 interactive, 1 GC (on 14 virtual cores)
Environment:
  JULIA_EDITOR = code
  JULIA_VSCODE_REPL = 1

Additional context

Might be related to #627

[ChrisRackauckas]

I think it's the deepcopy in there. We've identified that operation seems to be hilariously bad for many reasons. SciML/RecursiveArrayTools.jl#468 was one solution we were working on, we're going to try a different one soon that is just directly defining copy operations on the set of problems and solvers.

[ChrisRackauckas]

GenericLUFactorization should opt-out of the system, since at that size reducing allocations is more valuable than reordering, which is why it opts out of BLAS. @jClugstor can you set that up?

[jClugstor]

So basically if the alg is GenericalLUFactorization it should just go through the normal init, i.e. it shouldn't make a DualLinearCache etc. ?

[ChrisRackauckas]

Yes. Since the dual cache necessarily has to allocate. For O(n^3) hitting BLAS, that's fine. You have to allocate anyways since you have to reorder memory for optimal A vs Adual. It can probably use a directly malloc/free to avoid the GC, but that's a minor optimization. But for small solves, it should be fully non-allocating, and you're in the size of L1 cache anyways so you don't need to worry about memory ordering.

[ChrisRackauckas]

#685