Weird inference failure in `promote_u0` #918

fjebaker · 2023-07-28T14:41:08Z

Originally opened in SciML/OrdinaryDiffEq.jl#2001

MWE:

using DiffEqBase

struct Thing
    a::Float64
end
struct Wrapper1{T}
    thing::T
end
struct Wrapper2{T}
    thing::T
end

thing = Thing(1.0)
x = 1.0

DiffEqBase.promote_u0(x, Wrapper1(thing), (0.0, 1.0))
@code_warntype DiffEqBase.promote_u0(x, Wrapper1(thing), (0.0, 1.0))

DiffEqBase.promote_u0(x, Wrapper2(thing), (0.0, 1.0))
@code_warntype DiffEqBase.promote_u0(x, Wrapper2(thing), (0.0, 1.0))

Output:

MethodInstance for DiffEqBase.promote_u0(::Float64, ::Wrapper1{Thing}, ::Tuple{Float64, Float64})
  from promote_u0(u0, p, t0) @ DiffEqBase ~/.julia/packages/DiffEqBase/9rTlH/src/forwarddiff.jl:208
Arguments
  #self#::Core.Const(DiffEqBase.promote_u0)
  u0::Float64
  p::Wrapper1{Thing}
  t0::Tuple{Float64, Float64}
Locals
  T::Any
Body::Any
1 ─       nothing
│         Core.NewvarNode(:(T))
│   %3  = DiffEqBase.eltype(u0)::Core.Const(Float64)
│   %4  = ForwardDiff.Dual::Core.Const(ForwardDiff.Dual)
│   %5  = (%3 <: %4)::Core.Const(false)
│   %6  = !%5::Core.Const(true)
└──       goto #6 if not %6
2 ─       (T = DiffEqBase.anyeltypedual(p))
│   %9  = (T === DiffEqBase.Any)::Bool
└──       goto #4 if not %9
3 ─       return u0
4 ─ %12 = T::Any
│   %13 = ForwardDiff.Dual::Core.Const(ForwardDiff.Dual)
│   %14 = (%12 <: %13)::Bool
└──       goto #6 if not %14
5 ─ %16 = Base.broadcasted(T, u0)::Any
│   %17 = Base.materialize(%16)::Any
└──       return %17
6 ┄       return u0

MethodInstance for DiffEqBase.promote_u0(::Float64, ::Wrapper2{Thing}, ::Tuple{Float64, Float64})
  from promote_u0(u0, p, t0) @ DiffEqBase ~/.julia/packages/DiffEqBase/9rTlH/src/forwarddiff.jl:208
Arguments
  #self#::Core.Const(DiffEqBase.promote_u0)
  u0::Float64
  p::Wrapper2{Thing}
  t0::Tuple{Float64, Float64}
Locals
  T::Type{Any}
Body::Float64
1 ─      nothing
│        Core.NewvarNode(:(T))
│   %3 = DiffEqBase.eltype(u0)::Core.Const(Float64)
│   %4 = ForwardDiff.Dual::Core.Const(ForwardDiff.Dual)
│   %5 = (%3 <: %4)::Core.Const(false)
│   %6 = !%5::Core.Const(true)
└──      goto #5 if not %6
2 ─      (T = DiffEqBase.anyeltypedual(p))
│   %9 = (T::Core.Const(Any) === DiffEqBase.Any)::Core.Const(true)
└──      goto #4 if not %9
3 ─      return u0
4 ─      Core.Const(:(T))
│        Core.Const(:(ForwardDiff.Dual))
│        Core.Const(:(%12 <: %13))
│        Core.Const(:(goto %19 if not %14))
│        Core.Const(:(Base.broadcasted(T, u0)))
│        Core.Const(:(Base.materialize(%16)))
└──      Core.Const(:(return %17))
5 ┄      Core.Const(:(return u0))

The first one always fails to infer (i.e. doing Wrapper2 first makes it fail to infer Wrapper2).

I have no idea why, since anyeltypedual seems to be const folded:

@code_warntype DiffEqBase.anyeltypedual(Wrapper1(thing))
@code_warntype DiffEqBase.anyeltypedual(Wrapper2(thing))
MethodInstance for DiffEqBase.anyeltypedual(::Wrapper1{Thing})
  from anyeltypedual(x) @ DiffEqBase ~/.julia/packages/DiffEqBase/9rTlH/src/forwarddiff.jl:101
Arguments
  #self#::Core.Const(DiffEqBase.anyeltypedual)
  x::Wrapper1{Thing}
Body::Type{Any}
1 ─ %1 = (#self#)(x, 0)::Core.Const(Any)
└──      return %1

MethodInstance for DiffEqBase.anyeltypedual(::Wrapper2{Thing})
  from anyeltypedual(x) @ DiffEqBase ~/.julia/packages/DiffEqBase/9rTlH/src/forwarddiff.jl:101
Arguments
  #self#::Core.Const(DiffEqBase.anyeltypedual)
  x::Wrapper2{Thing}
Body::Type{Any}
1 ─ %1 = (#self#)(x, 0)::Core.Const(Any)
└──      return %1

The text was updated successfully, but these errors were encountered:

fjebaker · 2023-07-28T14:44:06Z

I'm not really sure what I'm doing with these tools, but I'm having a look anyway. SnoopCompile finds the same:

julia> inv1 = @snoopi_deep DiffEqBase.promote_u0(x, Wrapper1(thing), (0.0, 1.0)) 
InferenceTimingNode: 0.005841/0.056050 on Core.Compiler.Timings.ROOT() with 3 direct children

julia> inv2 = @snoopi_deep DiffEqBase.promote_u0(x, Wrapper2(thing), (0.0, 1.0)) 
InferenceTimingNode: 0.004530/0.005901 on Core.Compiler.Timings.ROOT() with 2 direct children

julia> inference_triggers(inv1)
1-element Vector{InferenceTrigger}:
 Inference triggered to call DiffEqBase.diffeqmapreduce(::DiffEqBase.DualEltypeChecker{Wrapper1{Thing}}, ::typeof(DiffEqBase.promote_dual), ::Tuple{Val{:thing}}) from anyeltypedual (/Users/lx21966/.julia/packages/DiffEqBase/9rTlH/src/forwarddiff.jl:105) inlined into DiffEqBase.promote_u0(::Float64, ::Wrapper1{Thing}, ::Tuple{Float64, Float64}) (/Users/lx21966/.julia/packages/DiffEqBase/9rTlH/src/forwarddiff.jl:210)

julia> inference_triggers(inv2)
InferenceTrigger[]

julia> summary(accumulate_by_source(Method, inference_triggers(inv1)))
promote_u0(u0, p, t0) @ DiffEqBase ~/.julia/packages/DiffEqBase/9rTlH/src/forwarddiff.jl:208 had 1 specializations
Triggering calls:
Inlined anyeltypedual at /Users/lx21966/.julia/packages/DiffEqBase/9rTlH/src/forwarddiff.jl:105: calling diffeqmapreduce (1 instances)

It actually even narrows it down to anyeltypedual. Looking deeper

Update: pulling the code apart a bit boils the problem down to

1-element Vector{InferenceTrigger}:
 Inference triggered to call map(::DiffEqBase.DualEltypeChecker{Wrapper1{Thing}}, ::Tuple{Val{:thing}}) from anyeltypedual (/Users/lx21966/Developer/jl/test-inf3.jl:38) with specialization DiffEqBase.anyeltypedual(::Wrapper1{Thing}, ::Int64)

ChrisRackauckas · 2023-07-31T06:03:00Z

Moved to Julia's Base: JuliaLang/julia#50735. Indeed, that's weird.

thomvet · 2024-02-28T09:23:00Z

Not sure why, but when I am running the above code on my computer, the return type now seems correctly inferred. Code warntype output (after a direct copy paste of the above code sample in a fresh Julia session):

@code_warntype DiffEqBase.promote_u0(x, Wrapper1(thing), (0.0, 1.0))
MethodInstance for DiffEqBase.promote_u0(::Float64, ::Wrapper1{Thing}, ::Tuple{Float64, Float64})
  from promote_u0(u0, p, t0) @ DiffEqBase C:\Users\thvt\.julia\packages\DiffEqBase\4084R\src\forwarddiff.jl:250
Arguments
  #self#::Core.Const(DiffEqBase.promote_u0)
  u0::Float64
  p::Wrapper1{Thing}
  t0::Tuple{Float64, Float64}
Locals
  T::Type{Any}
Body::Float64
1 ─      nothing
│        Core.NewvarNode(:(T))
│   %3 = DiffEqBase.eltype(u0)::Core.Const(Float64)
│   %4 = ForwardDiff.Dual::Core.Const(ForwardDiff.Dual)
│   %5 = (%3 <: %4)::Core.Const(false)
│   %6 = !%5::Core.Const(true)
└──      goto #6 if not %6
2 ─      (T = DiffEqBase.anyeltypedual(p))
│   %9 = (T::Core.Const(Any) === DiffEqBase.Any)::Core.Const(true)
└──      goto #5 if not %9
3 ─      return u0
4 ─      Core.Const(:(goto %13))
5 ┄      Core.Const(:(T))
│        Core.Const(:(ForwardDiff.Dual))
│        Core.Const(:(%13 <: %14))
│        Core.Const(:(goto %20 if not %15))
│        Core.Const(:(Base.broadcasted(T, u0)))
│        Core.Const(:(Base.materialize(%17)))
└──      Core.Const(:(return %18))
6 ┄      Core.Const(:(return u0))

And the second example provides the same output as given above, i.e., inference still works.

Pkg status:
[2b5f629d] DiffEqBase v6.147.1

Manifest:

  [47edcb42] ADTypes v0.2.6
  [79e6a3ab] Adapt v4.0.1
  [4fba245c] ArrayInterface v7.7.1
  [62783981] BitTwiddlingConvenienceFunctions v0.1.5
  [2a0fbf3d] CPUSummary v0.2.4
  [fb6a15b2] CloseOpenIntervals v0.1.12
  [38540f10] CommonSolve v0.2.4
  [bbf7d656] CommonSubexpressions v0.3.0
  [34da2185] Compat v4.14.0
  [187b0558] ConstructionBase v1.5.4
  [adafc99b] CpuId v0.3.1
  [9a962f9c] DataAPI v1.16.0
  [864edb3b] DataStructures v0.18.17
  [e2d170a0] DataValueInterfaces v1.0.0
  [2b5f629d] DiffEqBase v6.147.1
  [163ba53b] DiffResults v1.1.0
  [b552c78f] DiffRules v1.15.1
  [ffbed154] DocStringExtensions v0.9.3
  [4e289a0a] EnumX v1.0.4
  [f151be2c] EnzymeCore v0.6.5
  [e2ba6199] ExprTools v0.1.10
  [7034ab61] FastBroadcast v0.2.8
  [f6369f11] ForwardDiff v0.10.36
  [069b7b12] FunctionWrappers v1.1.3
  [77dc65aa] FunctionWrappersWrappers v0.1.3
  [46192b85] GPUArraysCore v0.1.6
  [615f187c] IfElse v0.1.1
  [92d709cd] IrrationalConstants v0.2.2
  [82899510] IteratorInterfaceExtensions v1.0.0
  [692b3bcd] JLLWrappers v1.5.0
  [10f19ff3] LayoutPointers v0.1.15
  [2ab3a3ac] LogExpFunctions v0.3.27
  [1914dd2f] MacroTools v0.5.13
  [d125e4d3] ManualMemory v0.1.8
  [46d2c3a1] MuladdMacro v0.2.4
  [77ba4419] NaNMath v1.0.2
  [bac558e1] OrderedCollections v1.6.3
  [d96e819e] Parameters v0.12.3
  [f517fe37] Polyester v0.7.9
  [1d0040c9] PolyesterWeave v0.2.1
  [d236fae5] PreallocationTools v0.4.20
  [aea7be01] PrecompileTools v1.2.0
  [21216c6a] Preferences v1.4.1
  [3cdcf5f2] RecipesBase v1.3.4
  [731186ca] RecursiveArrayTools v3.10.1
  [189a3867] Reexport v1.2.2
  [ae029012] Requires v1.3.0
  [7e49a35a] RuntimeGeneratedFunctions v0.5.12
  [94e857df] SIMDTypes v0.1.0
  [0bca4576] SciMLBase v2.29.0
  [c0aeaf25] SciMLOperators v0.3.8
  [efcf1570] Setfield v1.1.1
  [276daf66] SpecialFunctions v2.3.1
  [aedffcd0] Static v0.8.10
  [0d7ed370] StaticArrayInterface v1.5.0
  [1e83bf80] StaticArraysCore v1.4.2
  [7792a7ef] StrideArraysCore v0.5.2
  [2efcf032] SymbolicIndexingInterface v0.3.8
  [3783bdb8] TableTraits v1.0.1
  [bd369af6] Tables v1.11.1
  [8290d209] ThreadingUtilities v0.5.2
  [410a4b4d] Tricks v0.1.8
  [781d530d] TruncatedStacktraces v1.4.0
  [3a884ed6] UnPack v1.0.2
  [efe28fd5] OpenSpecFun_jll v0.5.5+0
  [0dad84c5] ArgTools v1.1.1
  [56f22d72] Artifacts
  [2a0f44e3] Base64
  [ade2ca70] Dates
  [8ba89e20] Distributed
  [f43a241f] Downloads v1.6.0
  [7b1f6079] FileWatching
  [9fa8497b] Future
  [b77e0a4c] InteractiveUtils
  [b27032c2] LibCURL v0.6.4
  [76f85450] LibGit2
  [8f399da3] Libdl
  [37e2e46d] LinearAlgebra
  [56ddb016] Logging
  [d6f4376e] Markdown
  [ca575930] NetworkOptions v1.2.0
  [44cfe95a] Pkg v1.10.0
  [de0858da] Printf
  [3fa0cd96] REPL
  [9a3f8284] Random
  [ea8e919c] SHA v0.7.0
  [9e88b42a] Serialization
  [6462fe0b] Sockets
  [2f01184e] SparseArrays v1.10.0
  [10745b16] Statistics v1.10.0
  [4607b0f0] SuiteSparse
  [fa267f1f] TOML v1.0.3
  [a4e569a6] Tar v1.10.0
  [8dfed614] Test
  [cf7118a7] UUIDs
  [4ec0a83e] Unicode
  [e66e0078] CompilerSupportLibraries_jll v1.0.5+1
  [deac9b47] LibCURL_jll v8.4.0+0
  [e37daf67] LibGit2_jll v1.6.4+0
  [29816b5a] LibSSH2_jll v1.11.0+1
  [c8ffd9c3] MbedTLS_jll v2.28.2+1
  [14a3606d] MozillaCACerts_jll v2023.1.10
  [4536629a] OpenBLAS_jll v0.3.23+2
  [05823500] OpenLibm_jll v0.8.1+2
  [bea87d4a] SuiteSparse_jll v7.2.1+1
  [83775a58] Zlib_jll v1.2.13+1
  [8e850b90] libblastrampoline_jll v5.8.0+1
  [8e850ede] nghttp2_jll v1.52.0+1
  [3f19e933] p7zip_jll v17.4.0+2

Versioninfo:

Julia Version 1.10.0
Commit 3120989f39 (2023-12-25 18:01 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 24 × 12th Gen Intel(R) Core(TM) i7-12850HX
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, alderlake)
  Threads: 35 on 24 virtual cores

ChrisRackauckas · 2024-03-03T20:47:12Z

It recently got some tweaks.

#1017

That would have fixed these cases.

ChrisRackauckas mentioned this issue Jul 31, 2023

Non-deterministic inference results? JuliaLang/julia#50735

Open

ChrisRackauckas closed this as completed Mar 3, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Weird inference failure in `promote_u0` #918

Weird inference failure in `promote_u0` #918

fjebaker commented Jul 28, 2023 •

edited

Loading

fjebaker commented Jul 28, 2023 •

edited

Loading

ChrisRackauckas commented Jul 31, 2023

thomvet commented Feb 28, 2024

ChrisRackauckas commented Mar 3, 2024

Weird inference failure in promote_u0 #918

Weird inference failure in promote_u0 #918

Comments

fjebaker commented Jul 28, 2023 • edited Loading

fjebaker commented Jul 28, 2023 • edited Loading

ChrisRackauckas commented Jul 31, 2023

thomvet commented Feb 28, 2024

ChrisRackauckas commented Mar 3, 2024

Weird inference failure in `promote_u0` #918

Weird inference failure in `promote_u0` #918

fjebaker commented Jul 28, 2023 •

edited

Loading

fjebaker commented Jul 28, 2023 •

edited

Loading