Julia LateLowerGCFrame Assertion

[nrontsis]

When running

using Enzyme
using ComponentArrays

const x = ComponentVector(; a=1.0)
const u = ComponentVector(; a=1.0, b=1.0)

Enzyme.gradient(Enzyme.ForwardMode(), p -> sum(copy(x)), u)

on: Intel MacOS 12.6, Julia 1.8.2, Enzyme v0.10.12, ComponentArrays v0.13.4

I get:

signal (11): Segmentation fault: 11
in expression starting at ~/mwe.jl:7
_ZN16LateLowerGCFrame13NumberAllBaseER5StatePN4llvm5ValueE at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-codegen.1.8.dylib (unknown line)
_ZN16LateLowerGCFrame10NumberBaseER5StatePN4llvm5ValueE at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-codegen.1.8.dylib (unknown line)
_ZN16LateLowerGCFrame6NumberER5StatePN4llvm5ValueE at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-codegen.1.8.dylib (unknown line)
_ZN16LateLowerGCFrame9LocalScanERN4llvm8FunctionE at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-codegen.1.8.dylib (unknown line)
_ZN16LateLowerGCFrame13runOnFunctionERN4llvm8FunctionE at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-codegen.1.8.dylib (unknown line)
_ZN4llvm13FPPassManager13runOnFunctionERNS_8FunctionE at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libLLVM.dylib (unknown line)
_ZN4llvm13FPPassManager11runOnModuleERNS_6ModuleE at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libLLVM.dylib (unknown line)
_ZN4llvm6legacy15PassManagerImpl3runERNS_6ModuleE at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libLLVM.dylib (unknown line)
LLVMRunPassManager at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libLLVM.dylib (unknown line)
LLVMRunPassManager at ~/.julia/packages/LLVM/WjSQG/lib/13/libLLVM_h.jl:4898 [inlined]
run! at ~/.julia/packages/LLVM/WjSQG/src/passmanager.jl:39 [inlined]
#109 at ~/.julia/packages/Enzyme/7ekWs/src/compiler/optimize.jl:501
#ModulePassManager#64 at ~/.julia/packages/LLVM/WjSQG/src/passmanager.jl:33
unknown function (ip: 0x10a16eafa)
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
ModulePassManager at ~/.julia/packages/LLVM/WjSQG/src/passmanager.jl:30
post_optimze! at ~/.julia/packages/Enzyme/7ekWs/src/compiler/optimize.jl:498 [inlined]
post_optimze! at ~/.julia/packages/Enzyme/7ekWs/src/compiler/optimize.jl:492 [inlined]
_thunk at ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:6339 [inlined]
_thunk at ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:6315
unknown function (ip: 0x108f9c1a9)
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
cached_compilation at ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:6359
unknown function (ip: 0x108f74ea1)
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
#s836#163 at ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:6419 [inlined]
#s836#163 at ./none:0
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
GeneratedFunctionStub at ./boot.jl:582
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
jl_call_staged at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
ijl_code_for_staged at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
get_staged at ./compiler/utilities.jl:114
retrieve_code_info at ./compiler/utilities.jl:126 [inlined]
InferenceState at ./compiler/inferencestate.jl:284
typeinf_edge at ./compiler/typeinfer.jl:868
abstract_call_method at ./compiler/abstractinterpretation.jl:647
abstract_call_gf_by_type at ./compiler/abstractinterpretation.jl:139
abstract_call_known at ./compiler/abstractinterpretation.jl:1716
abstract_call at ./compiler/abstractinterpretation.jl:1786
abstract_call at ./compiler/abstractinterpretation.jl:1753
abstract_eval_statement at ./compiler/abstractinterpretation.jl:1910
typeinf_local at ./compiler/abstractinterpretation.jl:2386
typeinf_nocycle at ./compiler/abstractinterpretation.jl:2482
_typeinf at ./compiler/typeinfer.jl:230
typeinf at ./compiler/typeinfer.jl:213
typeinf_edge at ./compiler/typeinfer.jl:877
abstract_call_method at ./compiler/abstractinterpretation.jl:647
abstract_call_gf_by_type at ./compiler/abstractinterpretation.jl:139
abstract_call_known at ./compiler/abstractinterpretation.jl:1716
abstract_call at ./compiler/abstractinterpretation.jl:1786
abstract_call at ./compiler/abstractinterpretation.jl:1753
abstract_eval_statement at ./compiler/abstractinterpretation.jl:1910
typeinf_local at ./compiler/abstractinterpretation.jl:2360
typeinf_nocycle at ./compiler/abstractinterpretation.jl:2482
_typeinf at ./compiler/typeinfer.jl:230
typeinf at ./compiler/typeinfer.jl:213
typeinf_ext at ./compiler/typeinfer.jl:967
typeinf_ext_toplevel at ./compiler/typeinfer.jl:1000
typeinf_ext_toplevel at ./compiler/typeinfer.jl:996
jfptr_typeinf_ext_toplevel_18218.clone_1 at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/sys.dylib (unknown line)
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
jl_type_infer at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
jl_generate_fptr_impl at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-codegen.1.8.dylib (unknown line)
jl_compile_method_internal at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
#gradient#30 at ~/.julia/packages/Enzyme/7ekWs/src/Enzyme.jl:582
unknown function (ip: 0x108f4f968)
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
gradient at ~/.julia/packages/Enzyme/7ekWs/src/Enzyme.jl:578
unknown function (ip: 0x108f4f20e)
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
do_call at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
eval_body at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
jl_interpret_toplevel_thunk at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
jl_toplevel_eval_flex at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
jl_toplevel_eval_flex at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
ijl_toplevel_eval_in at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
eval at ./boot.jl:368 [inlined]
include_string at ./loading.jl:1428
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
_include at ./loading.jl:1488
include at ./Base.jl:419
jfptr_include_45928.clone_1 at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/sys.dylib (unknown line)
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
exec_options at ./client.jl:303
_start at ./client.jl:522
jfptr__start_36967.clone_1 at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/sys.dylib (unknown line)
ijl_apply_generic at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
true_main at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
jl_repl_entrypoint at /Applications/Julia-1.8.app/Contents/Resources/julia/lib/julia/libjulia-internal.1.8.dylib (unknown line)
Allocations: 33356866 (Pool: 33318358; Big: 38508); GC: 34

Context: this discourse thread,

[wsmoses]

So this error doesn't actually come from Enzyme, but Julia itself (and perhaps this would be best moved to Julia).

In essence what appears to be happening is that we have a function argument as follows:

[2 x { {} addrspace(10)* } addrspace(11)*] %"'"

The particular error occurs here: https://github.com/JuliaLang/julia/blob/master/src/llvm-late-gc-lowering.cpp#L917
for an extractvalue:

  %3 = extractvalue [2 x { {} addrspace(10)* } addrspace(11)*] %"'", 1, !dbg !33

Tracked.Size() == 0 from auto Tracked = TrackCompositeType(EVI->getAggregateOperand()->getType()); since TrackCompositeType only recurses on addrspace10: https://github.com/JuliaLang/julia/blob/54e68997108d33a6478bc1bfbbc2b7d1ea072202/src/llvm-late-gc-lowering.cpp#L429

However, on the other end of the assertion:

# called now with the argument
auto BaseNumbers = NumberAll(S, EVI->getAggregateOperand());
# becomes 
auto tracked = CountTrackedPointers(CurrentV->getType());

This call is fine with any form of special pointer: https://github.com/JuliaLang/julia/blob/54e68997108d33a6478bc1bfbbc2b7d1ea072202/src/llvm-late-gc-lowering.cpp#L392 including the addrspace11 at the outside, causing it to return 2 -- hence the mismatch.

@Keno @vtjnash which is the correct behavior?

[vtjnash]

Special pointers in FCA is often incorrect IR, since this pass would need to know how to make a shadow object and trace it, and derived pointers are prohibited by the language semantics from appearing as fields. But it should count as a trackable pointer, since it is required to be tracked

[wsmoses]

Yeah the rough semantics we're emitting here is to coalesce several arguments passed by reference as { {} addrspace(10)* } addrspace(11)* into a single register. So in this case the caller would've had a rooted object that it passes here.

[wsmoses]

The fix was merged into Julia proper, closing here. You will need a version of julia with the fix.

[nrontsis]

Thanks for the Julia fix, but when running the above example with: Julia version 1.9.0-DEV.1680 (commit 42563aac30) I get:

ERROR: Enzyme compilation failed.
Current scope:
; Function Attrs: mustprogress willreturn
define double @preprocess_julia__3_3595_inner.1({ {} addrspace(10)* } %0) local_unnamed_addr #7 !dbg !194 {
entry:
  %1 = alloca { {} addrspace(10)* }, align 8
  %2 = bitcast { {} addrspace(10)* }* %1 to i8*
  call void @llvm.lifetime.start.p0i8(i64 noundef 8, i8* noundef nonnull align 8 dereferenceable(8) %2) #8
  %3 = call {}*** @julia.get_pgcstack() #8
  %ptls_field.i3 = getelementptr inbounds {}**, {}*** %3, i64 2
  %4 = bitcast {}*** %ptls_field.i3 to i64***
  %ptls_load.i45 = load i64**, i64*** %4, align 8, !tbaa !12
  fence syncscope("singlethread") seq_cst
  %5 = getelementptr inbounds i64*, i64** %ptls_load.i45, i64 2
  %safepoint.i = load i64*, i64** %5, align 8, !tbaa !16, !invariant.load !11
  %6 = load volatile i64, i64* %safepoint.i, align 8
  fence syncscope("singlethread") seq_cst
  %7 = load atomic {} addrspace(10)*, {} addrspace(10)** inttoptr (i64 5612376208 to {} addrspace(10)**) unordered, align 16, !dbg !195, !tbaa !169, !nonnull !11, !dereferenceable !30, !align !31
  %8 = call nonnull {} addrspace(10)* @ijl_array_copy({} addrspace(10)* nonnull %7) #8, !dbg !199
  %.fca.0.gep = getelementptr inbounds { {} addrspace(10)* }, { {} addrspace(10)* }* %1, i64 0, i32 0, !dbg !200
  store {} addrspace(10)* %8, {} addrspace(10)** %.fca.0.gep, align 8, !dbg !200
  %9 = addrspacecast { {} addrspace(10)* }* %1 to { {} addrspace(10)* } addrspace(11)*, !dbg !200
  %10 = call fastcc double @julia__mapreduce_3598({ {} addrspace(10)* } addrspace(11)* nocapture nofree noundef nonnull readonly align 8 dereferenceable(8) %9) #7, !dbg !200
  call void @llvm.lifetime.end.p0i8(i64 noundef 8, i8* noundef nonnull %2) #8, !dbg !209
  ret double %10, !dbg !210
}

; Function Attrs: mustprogress willreturn
define double @preprocess_julia__3_3595_inner.1({ {} addrspace(10)* } %0) local_unnamed_addr #7 !dbg !194 {
entry:
  %1 = alloca { {} addrspace(10)* }, align 8
  %2 = bitcast { {} addrspace(10)* }* %1 to i8*
  call void @llvm.lifetime.start.p0i8(i64 noundef 8, i8* noundef nonnull align 8 dereferenceable(8) %2) #8
  %3 = call {}*** @julia.get_pgcstack() #8
  %ptls_field.i3 = getelementptr inbounds {}**, {}*** %3, i64 2
  %4 = bitcast {}*** %ptls_field.i3 to i64***
  %ptls_load.i45 = load i64**, i64*** %4, align 8, !tbaa !12
  fence syncscope("singlethread") seq_cst
  %5 = getelementptr inbounds i64*, i64** %ptls_load.i45, i64 2
  %safepoint.i = load i64*, i64** %5, align 8, !tbaa !16, !invariant.load !11
  %6 = load volatile i64, i64* %safepoint.i, align 8
  fence syncscope("singlethread") seq_cst
  %7 = load atomic {} addrspace(10)*, {} addrspace(10)** inttoptr (i64 5612376208 to {} addrspace(10)**) unordered, align 16, !dbg !195, !tbaa !169, !nonnull !11, !dereferenceable !30, !align !31
  %8 = call nonnull {} addrspace(10)* @ijl_array_copy({} addrspace(10)* nonnull %7) #8, !dbg !199
  %.fca.0.gep = getelementptr inbounds { {} addrspace(10)* }, { {} addrspace(10)* }* %1, i64 0, i32 0, !dbg !200
  store {} addrspace(10)* %8, {} addrspace(10)** %.fca.0.gep, align 8, !dbg !200
  %9 = addrspacecast { {} addrspace(10)* }* %1 to { {} addrspace(10)* } addrspace(11)*, !dbg !200
  %10 = call fastcc double @julia__mapreduce_3598({ {} addrspace(10)* } addrspace(11)* nocapture nofree noundef nonnull readonly align 8 dereferenceable(8) %9) #7, !dbg !200
  call void @llvm.lifetime.end.p0i8(i64 noundef 8, i8* noundef nonnull %2) #8, !dbg !209
  ret double %10, !dbg !210
}

; Function Attrs: mustprogress willreturn
define internal [2 x double] @fwddiffe2julia__3_3595_inner.1({ {} addrspace(10)* } %0, [2 x { {} addrspace(10)* }] %"'") local_unnamed_addr #7 !dbg !211 {
entry:
  %"'ipa" = alloca { {} addrspace(10)* }, align 8
  %1 = insertvalue [2 x { {} addrspace(10)* }*] undef, { {} addrspace(10)* }* %"'ipa", 0
  %"'ipa5" = alloca { {} addrspace(10)* }, align 8
  %2 = insertvalue [2 x { {} addrspace(10)* }*] %1, { {} addrspace(10)* }* %"'ipa5", 1
  store { {} addrspace(10)* } zeroinitializer, { {} addrspace(10)* }* %"'ipa", align 8
  store { {} addrspace(10)* } zeroinitializer, { {} addrspace(10)* }* %"'ipa5", align 8
  %3 = alloca { {} addrspace(10)* }, align 8
  %_replacementA = phi i8*
  %4 = call {}*** @julia.get_pgcstack() #8
  %ptls_field.i3 = getelementptr inbounds {}**, {}*** %4, i64 2
  %5 = bitcast {}*** %ptls_field.i3 to i64***
  %ptls_load.i45 = load i64**, i64*** %5, align 8, !tbaa !12
  fence syncscope("singlethread") seq_cst
  %6 = getelementptr inbounds i64*, i64** %ptls_load.i45, i64 2
  %safepoint.i = load i64*, i64** %6, align 8, !tbaa !16, !invariant.load !11
  %7 = load volatile i64, i64* %safepoint.i, align 8
  fence syncscope("singlethread") seq_cst
  %8 = load atomic {} addrspace(10)*, {} addrspace(10)** inttoptr (i64 5612376208 to {} addrspace(10)**) unordered, align 16, !dbg !212, !tbaa !169, !nonnull !11, !dereferenceable !30, !align !31
  %9 = call nonnull {} addrspace(10)* @ijl_array_copy({} addrspace(10)* nonnull %8) #8, !dbg !216
  %"'dual_phi2" = phi [2 x {} addrspace(10)*] , !dbg !217
  %.fca.0.gep = getelementptr inbounds { {} addrspace(10)* }, { {} addrspace(10)* }* %3, i64 0, i32 0, !dbg !217
  %".fca.0.gep'dual_phi" = phi [2 x {} addrspace(10)**] , !dbg !217
  store {} addrspace(10)* %9, {} addrspace(10)** %.fca.0.gep, align 8, !dbg !217
  %10 = addrspacecast { {} addrspace(10)* }* %3 to { {} addrspace(10)* } addrspace(11)*, !dbg !217
  %"'dual_phi3" = phi [2 x { {} addrspace(10)* } addrspace(11)*] , !dbg !217
  %11 = call fastcc double @julia__mapreduce_3598({ {} addrspace(10)* } addrspace(11)* nocapture nofree noundef nonnull readonly align 8 dereferenceable(8) %10) #7, !dbg !217
  %"'dual_phi4" = phi fast [2 x double] , !dbg !226
  call void @llvm.lifetime.end.p0i8(i64 noundef 8, i8* noundef nonnull %_replacementA) #8, !dbg !226
  ret double %11, !dbg !227

allocsForInversion:                               ; No predecessors!
}

in Mode: ForwardMode
cannot handle unknown instruction
  fence syncscope("singlethread") seq_cst


Stacktrace:
  [1] julia_error(cstr::Cstring, val::Ptr{LLVM.API.LLVMOpaqueValue}, errtype::Enzyme.API.ErrorType, data::Ptr{Nothing})
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:3498
  [2] EnzymeCreateForwardDiff(logic::Enzyme.Logic, todiff::LLVM.Function, retType::Enzyme.API.CDIFFE_TYPE, constant_args::Vector{Enzyme.API.CDIFFE_TYPE}, TA::Enzyme.TypeAnalysis, returnValue::Bool, mode::Enzyme.API.CDerivativeMode, width::Int64, additionalArg::Ptr{Nothing}, typeInfo::Enzyme.FnTypeInfo, uncacheable_args::Vector{Bool})
    @ Enzyme.API ~/.julia/packages/Enzyme/7ekWs/src/api.jl:137
  [3] enzyme!(job::GPUCompiler.CompilerJob{Enzyme.Compiler.EnzymeTarget, Enzyme.Compiler.EnzymeCompilerParams, GPUCompiler.FunctionSpec{var"#3#4", Tuple{ComponentVector{Float64, Vector{Float64}, Tuple{Axis{(a = 1, b = 2)}}}}}}, mod::LLVM.Module, primalf::LLVM.Function, adjoint::GPUCompiler.FunctionSpec{var"#3#4", Tuple{BatchDuplicated{ComponentVector{Float64, Vector{Float64}, Tuple{Axis{(a = 1, b = 2)}}}, 2}}}, mode::Enzyme.API.CDerivativeMode, width::Int64, parallel::Bool, actualRetType::Type, dupClosure::Bool, wrap::Bool, modifiedBetween::Bool, returnPrimal::Bool, jlrules::Vector{String})
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:4774
  [4] codegen(output::Symbol, job::GPUCompiler.CompilerJob{Enzyme.Compiler.EnzymeTarget, Enzyme.Compiler.EnzymeCompilerParams, GPUCompiler.FunctionSpec{var"#3#4", Tuple{ComponentVector{Float64, Vector{Float64}, Tuple{Axis{(a = 1, b = 2)}}}}}}; libraries::Bool, deferred_codegen::Bool, optimize::Bool, ctx::LLVM.ThreadSafeContext, strip::Bool, validate::Bool, only_entry::Bool, parent_job::Nothing)
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:5854
  [5] codegen
    @ ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:5524 [inlined]
  [6] _thunk
    @ ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:6321 [inlined]
  [7] _thunk(job::GPUCompiler.CompilerJob{Enzyme.Compiler.EnzymeTarget, Enzyme.Compiler.EnzymeCompilerParams, GPUCompiler.FunctionSpec{var"#3#4", Tuple{ComponentVector{Float64, Vector{Float64}, Tuple{Axis{(a = 1, b = 2)}}}}}})
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:6315
  [8] cached_compilation(job::GPUCompiler.CompilerJob, key::UInt64, specid::UInt64)
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:6359
  [9] #s859#163
    @ ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:6419 [inlined]
 [10] var"#s859#163"(F::Any, Fn::Any, DF::Any, A::Any, TT::Any, Mode::Any, ModifiedBetween::Any, width::Any, specid::Any, ReturnPrimal::Any, ::Any, #unused#::Type, f::Any, df::Any, #unused#::Type, tt::Any, #unused#::Type, #unused#::Type, #unused#::Type, #unused#::Type, #unused#::Any)
    @ Enzyme.Compiler ./none:0
 [11] (::Core.GeneratedFunctionStub)(::Any, ::Vararg{Any})
    @ Core ./boot.jl:599
 [12] thunk(f::var"#3#4", df::Nothing, ::Type{BatchDuplicatedNoNeed}, tt::Type{Tuple{BatchDuplicated{ComponentVector{Float64, Vector{Float64}, Tuple{Axis{(a = 1, b = 2)}}}, 2}}}, ::Val{Enzyme.API.DEM_ForwardMode}, ::Val{2}, ::Val{false}, ::Val{false})
    @ Enzyme.Compiler ~/.julia/packages/Enzyme/7ekWs/src/compiler.jl:6452
 [13] autodiff
    @ ~/.julia/packages/Enzyme/7ekWs/src/Enzyme.jl:312 [inlined]
 [14] gradient(::EnzymeCore.ForwardMode, f::Function, x::ComponentVector{Float64, Vector{Float64}, Tuple{Axis{(a = 1, b = 2)}}}; shadow::Tuple{ComponentVector{Float64, Vector{Float64}, Tuple{Axis{(a = 1, b = 2)}}}, ComponentVector{Float64, Vector{Float64}, Tuple{Axis{(a = 1, b = 2)}}}})
    @ Enzyme ~/.julia/packages/Enzyme/7ekWs/src/Enzyme.jl:582
 [15] gradient(::EnzymeCore.ForwardMode, f::Function, x::ComponentVector{Float64, Vector{Float64}, Tuple{Axis{(a = 1, b = 2)}}})
    @ Enzyme ~/.julia/packages/Enzyme/7ekWs/src/Enzyme.jl:578
 [16] top-level scope

[wsmoses]

That's a separate issue, it looks like 1.9 is now introducing new instructions which we don't handle. Open a new issue (@vchuravy for what that is doing here).