Segfault with Polyester.jl

[kiranshila]

Seemingly a different issue from #2208

I tried to reduce this example a bit, apologies for some missing math/context - but this should reproduce the error:

using Polyester, StatsBase, Enzyme, StaticArrays

@inline function lossless_abcd(z, βl)
    sbl, cbl = sincos(βl)
    @SMatrix [cbl z*sbl*im; (1/z)*sbl*im cbl]
end

function cascade(zs, βl)
    acc = lossless_abcd(first(zs), βl)
    for z in zs
        acc *= lossless_abcd(z, βl)
    end
    acc
end

function cascade!(A, zs, L, freqs)
    δ = L / length(freqs)
    # Solve each frequency point in parallel
    @batch for i in eachindex(freqs, A)
        βl = freqs[i]
        A[i] = cascade(zs, βl)
    end
    A
end

function objective!(A, L, zs, freqs)
    cascade!(A, zs, L, freqs)
    # Generate some scalar for the MWE
    sum(abs2.(sum.(A)))
end

function d_objective!(dzs, zs, A, shadow_A, L, freqs)
    Enzyme.make_zero!(shadow_A)
    Enzyme.make_zero!(dzs)
    Enzyme.autodiff(Enzyme.Reverse, objective!, Active,
        Duplicated(A, A_shadow),
        Const(L),
        Duplicated(zs, dzs),
        Const(freqs))
    dzs
end

freqs = ones(100)
zs = ones(100)
dzs = similar(zs)
A = Array{SMatrix{2,2,ComplexF64,4}}(undef, length(freqs))
A_shadow = similar(A)

d_objective!(dzs, zs, A, A_shadow, 1.0, freqs)

The error returned is

[1229359] signal 11 (1): Segmentation fault
in expression starting at /path/to/file.jl:62
_ZN4llvm21SymbolTableListTraitsINS_14GlobalVariableEE18removeNodeFromListEPS1_ at /home/kshila/.julia/juliaup/julia-1.11.2+0.x64.linux.gnu/bin/../lib/julia/libLLVM-16jl.so (unknown line)
unknown function (ip: 0x16ac14f)
Allocations: 23260080 (Pool: 23259464; Big: 616); GC: 19

Things seem to work correctly if I remove that @batch call in cascade!.

Versions, etc

Julia Version 1.11.2
Commit 5e9a32e7af2 (2024-12-01 20:02 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 24 × Intel(R) Xeon(R) Gold 6128 CPU @ 3.40GHz
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, skylake-avx512)
Threads: 24 default, 0 interactive, 12 GC (on 24 virtual cores)
Environment:
  JULIA_EDITOR = code

  [7da242da] Enzyme v0.13.24
  [f517fe37] Polyester v0.7.16

[kiranshila]

Or perhaps this is the same issue? Here's a trace running in Julia 1.10

ERROR: LoadError: Current scope: 
; Function Attrs: mustprogress willreturn
define internal fastcc void @preprocess_julia_cascade__596({} addrspace(10)* noundef nonnull align 16 dereferenceable(40) "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" "enzymejl_parmtype"="123247878569488" "enzymejl_parmtype_ref"="2" %0, {} addrspace(10)* noundef nonnull align 16 dereferenceable(40) "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" "enzymejl_parmtype"="123247705933040" "enzymejl_parmtype_ref"="2" %1, {} addrspace(10)* noundef nonnull align 16 dereferenceable(40) "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" "enzymejl_parmtype"="123247705933040" "enzymejl_parmtype_ref"="2" %2) unnamed_addr #93 !dbg !3590 {
top:
  %3 = call noalias nonnull dereferenceable(48) dereferenceable_or_null(48) i8* @malloc(i64 48), !enzyme_fromstack !142
  %4 = bitcast i8* %3 to { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }*, !enzyme_caststack !0
  %.sub = bitcast { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4 to i8*
  %5 = call noalias nonnull dereferenceable(64) dereferenceable_or_null(64) i8* @malloc(i64 64), !enzymejl_allocart !2322, !enzyme_type !329, !enzyme_fromstack !89
  %6 = bitcast i8* %5 to [1 x [4 x [2 x double]]]*, !enzyme_caststack !0
  %7 = call noalias nonnull dereferenceable(16) dereferenceable_or_null(16) i8* @malloc(i64 16), !enzyme_fromstack !89
  %newstruct71 = bitcast i8* %7 to { i64, [1 x i64] }*, !enzyme_caststack !0
  %8 = call noalias nonnull dereferenceable(64) dereferenceable_or_null(64) i8* @malloc(i64 64), !enzymejl_allocart !2322, !enzyme_type !329, !enzyme_fromstack !89
  %9 = bitcast i8* %8 to [1 x [4 x [2 x double]]]*, !enzyme_caststack !0
  %10 = call noalias nonnull dereferenceable(64) dereferenceable_or_null(64) i8* @malloc(i64 64), !enzymejl_allocart !2322, !enzyme_type !329, !enzyme_fromstack !89
  %11 = bitcast i8* %10 to [1 x [4 x [2 x double]]]*, !enzyme_caststack !0
  %12 = call {}*** @julia.get_pgcstack() #94
  %current_task1156 = getelementptr inbounds {}**, {}*** %12, i64 -14
  %current_task1 = bitcast {}*** %current_task1156 to {}**
  %ptls_field157 = getelementptr inbounds {}**, {}*** %12, i64 2
  %13 = bitcast {}*** %ptls_field157 to i64***
  %ptls_load158159 = load i64**, i64*** %13, align 8, !tbaa !66
  %14 = getelementptr inbounds i64*, i64** %ptls_load158159, i64 2
  %safepoint = load i64*, i64** %14, align 8, !tbaa !70
  fence syncscope("singlethread") seq_cst
  call void @julia.safepoint(i64* %safepoint) #94, !dbg !3591
  fence syncscope("singlethread") seq_cst
  %15 = call i64 @julia_nthreads_790() #95, !dbg !3592
  %.not = icmp eq i64 %15, 1, !dbg !3594
  br i1 %.not, label %L4, label %L51, !dbg !3595

L4:                                               ; preds = %top
  %16 = addrspacecast {} addrspace(10)* %2 to { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)*, !dbg !3596
  %arraylen_ptr = getelementptr inbounds { i8 addrspace(13)*, i64, i16, i16, i32 }, { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)* %16, i64 0, i32 1, !dbg !3596
  %arraylen = load i64, i64 addrspace(11)* %arraylen_ptr, align 8, !dbg !3596, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %17 = addrspacecast {} addrspace(10)* %0 to { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)*, !dbg !3603
  %arraylen_ptr2 = getelementptr inbounds { i8 addrspace(13)*, i64, i16, i16, i32 }, { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)* %17, i64 0, i32 1, !dbg !3603
  %arraylen3 = load i64, i64 addrspace(11)* %arraylen_ptr2, align 8, !dbg !3603, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %.not160 = icmp eq i64 %arraylen, %arraylen3, !dbg !3610
  br i1 %.not160, label %L25, label %L14, !dbg !3609

L14:                                              ; preds = %L4
  %box = call noalias nonnull dereferenceable(8) "enzyme_inactive" "enzyme_type"="{[-1]:Pointer, [-1,-1]:Integer}" {} addrspace(10)* @julia.gc_alloc_obj({}** nonnull %current_task1, i64 noundef 8, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247752959712 to {}*) to {} addrspace(10)*)) #96, !dbg !3609
  %memcpy_refined_dst = bitcast {} addrspace(10)* %box to i64 addrspace(10)*, !dbg !3609, !enzyme_inactive !0
  store i64 %arraylen, i64 addrspace(10)* %memcpy_refined_dst, align 8, !dbg !3609, !tbaa !230, !alias.scope !81, !noalias !3612
  %box31 = call noalias nonnull dereferenceable(8) "enzyme_inactive" "enzyme_type"="{[-1]:Pointer, [-1,-1]:Integer}" {} addrspace(10)* @julia.gc_alloc_obj({}** nonnull %current_task1, i64 noundef 8, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247752959712 to {}*) to {} addrspace(10)*)) #96, !dbg !3609
  %memcpy_refined_dst33 = bitcast {} addrspace(10)* %box31 to i64 addrspace(10)*, !dbg !3609, !enzyme_inactive !0
  store i64 %arraylen3, i64 addrspace(10)* %memcpy_refined_dst33, align 8, !dbg !3609, !tbaa !230, !alias.scope !81, !noalias !3612
  %18 = call nonnull "enzyme_type"="{[-1]:Pointer}" {} addrspace(10)* ({} addrspace(10)* ({} addrspace(10)*, {} addrspace(10)**, i32, {} addrspace(10)*)*, {} addrspace(10)*, {} addrspace(10)*, ...) @julia.call2({} addrspace(10)* ({} addrspace(10)*, {} addrspace(10)**, i32, {} addrspace(10)*)* noundef nonnull @ijl_invoke, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247893132112 to {}*) to {} addrspace(10)*), {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247774899088 to {}*) to {} addrspace(10)*), {} addrspace(10)* addrspacecast ({}* inttoptr (i64 123247744295344 to {}*) to {} addrspace(10)*), {} addrspace(10)* nofree nonnull %box, {} addrspace(10)* nofree nonnull %box31) #95, !dbg !3609
  unreachable, !dbg !3609

L25:                                              ; preds = %L4
  %.not161 = icmp eq i64 %arraylen, 0, !dbg !3615
  br i1 %.not161, label %L392, label %L34.preheader, !dbg !3602

L34.preheader:                                    ; preds = %L25
  %19 = addrspacecast {} addrspace(10)* %2 to double addrspace(13)* addrspace(11)*
  %20 = addrspacecast {} addrspace(10)* %0 to [1 x [4 x [2 x double]]] addrspace(13)* addrspace(11)*
  %21 = bitcast [1 x [4 x [2 x double]]]* %9 to i8*
  br label %L34, !dbg !3619

L34:                                              ; preds = %idxend16, %L34.preheader
  %iv12 = phi i64 [ %iv.next13, %idxend16 ], [ 0, %L34.preheader ]
  %iv.next13 = add nuw nsw i64 %iv12, 1, !dbg !3619
  %22 = add nsw i64 %iv.next13, -1, !dbg !3619
  %arraylen10 = load i64, i64 addrspace(11)* %arraylen_ptr, align 8, !dbg !3619, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %inbounds = icmp ult i64 %22, %arraylen10, !dbg !3619
  br i1 %inbounds, label %idxend, label %oob, !dbg !3619

L51:                                              ; preds = %top
  %23 = addrspacecast {} addrspace(10)* %2 to {} addrspace(11)*, !dbg !3621
  %24 = addrspacecast {} addrspace(10)* %2 to { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)*, !dbg !3621
  %arraylen_ptr34 = getelementptr inbounds { i8 addrspace(13)*, i64, i16, i16, i32 }, { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)* %24, i64 0, i32 1, !dbg !3621
  %arraylen35 = load i64, i64 addrspace(11)* %arraylen_ptr34, align 8, !dbg !3621, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %25 = addrspacecast {} addrspace(10)* %0 to {} addrspace(11)*, !dbg !3628
  %26 = addrspacecast {} addrspace(10)* %0 to { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)*, !dbg !3628
  %arraylen_ptr37 = getelementptr inbounds { i8 addrspace(13)*, i64, i16, i16, i32 }, { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)* %26, i64 0, i32 1, !dbg !3628
  %arraylen38 = load i64, i64 addrspace(11)* %arraylen_ptr37, align 8, !dbg !3628, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %.not165 = icmp eq i64 %arraylen35, %arraylen38, !dbg !3635
  br i1 %.not165, label %L72, label %L61, !dbg !3634

L61:                                              ; preds = %L51
  %box108 = call noalias nonnull dereferenceable(8) "enzyme_inactive" "enzyme_type"="{[-1]:Pointer, [-1,-1]:Integer}" {} addrspace(10)* @julia.gc_alloc_obj({}** nonnull %current_task1, i64 noundef 8, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247752959712 to {}*) to {} addrspace(10)*)) #96, !dbg !3634
  %memcpy_refined_dst110 = bitcast {} addrspace(10)* %box108 to i64 addrspace(10)*, !dbg !3634, !enzyme_inactive !0
  store i64 %arraylen35, i64 addrspace(10)* %memcpy_refined_dst110, align 8, !dbg !3634, !tbaa !230, !alias.scope !81, !noalias !3612
  %box112 = call noalias nonnull dereferenceable(8) "enzyme_inactive" "enzyme_type"="{[-1]:Pointer, [-1,-1]:Integer}" {} addrspace(10)* @julia.gc_alloc_obj({}** nonnull %current_task1, i64 noundef 8, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247752959712 to {}*) to {} addrspace(10)*)) #96, !dbg !3634
  %memcpy_refined_dst114 = bitcast {} addrspace(10)* %box112 to i64 addrspace(10)*, !dbg !3634, !enzyme_inactive !0
  store i64 %arraylen38, i64 addrspace(10)* %memcpy_refined_dst114, align 8, !dbg !3634, !tbaa !230, !alias.scope !81, !noalias !3612
  %27 = call nonnull "enzyme_type"="{[-1]:Pointer}" {} addrspace(10)* ({} addrspace(10)* ({} addrspace(10)*, {} addrspace(10)**, i32, {} addrspace(10)*)*, {} addrspace(10)*, {} addrspace(10)*, ...) @julia.call2({} addrspace(10)* ({} addrspace(10)*, {} addrspace(10)**, i32, {} addrspace(10)*)* noundef nonnull @ijl_invoke, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247893132112 to {}*) to {} addrspace(10)*), {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247774899088 to {}*) to {} addrspace(10)*), {} addrspace(10)* addrspacecast ({}* inttoptr (i64 123247744295344 to {}*) to {} addrspace(10)*), {} addrspace(10)* nofree nonnull %box108, {} addrspace(10)* nofree nonnull %box112) #95, !dbg !3634
  unreachable, !dbg !3634

L72:                                              ; preds = %L51
  %.not167 = icmp eq i64 %arraylen35, 0, !dbg !3637
  br i1 %.not167, label %L392, label %L76, !dbg !3639

L76:                                              ; preds = %L72
  %28 = call i64 @llvm.smin.i64(i64 %15, i64 %arraylen35) #94, !dbg !3641
  %.not168 = icmp eq i64 %28, 0, !dbg !3643
  br i1 %.not168, label %L383.lr.ph, label %L84, !dbg !3644

L84:                                              ; preds = %L76
  %29 = trunc i64 %28 to i32, !dbg !3645
  %30 = add i32 %29, -1, !dbg !3645
  %31 = call nonnull {}* @julia.pointer_from_objref({} addrspace(11)* noundef addrspacecast ({}* inttoptr (i64 123247601534016 to {}*) to {} addrspace(11)*)) #97, !dbg !3649
  %32 = icmp sgt i32 %30, 0, !dbg !3651
  br i1 %32, label %L94, label %L383.lr.ph, !dbg !3652

L94:                                              ; preds = %L84
  %p.i = bitcast {}* %31 to i64*, !dbg !3654
  %v.i = atomicrmw xchg i64* %p.i, i64 0 acq_rel, align 8, !dbg !3654
  %33 = call i64 @llvm.ctpop.i64(i64 %v.i) #94, !dbg !3657, !range !2422
  %34 = trunc i64 %33 to i32, !dbg !3659
  %35 = sub nsw i32 %30, %34, !dbg !3660
  %36 = icmp slt i32 %35, 0, !dbg !3662
  br i1 %36, label %L107, label %L142, !dbg !3665

L107:                                             ; preds = %L94
  %37 = call i64 @llvm.ctlz.i64(i64 %v.i, i1 noundef false) #94, !dbg !3666, !range !2422
  %38 = trunc i64 %37 to i32, !dbg !3668
  br label %L110, !dbg !3669

L110:                                             ; preds = %L110, %L107
  %iv = phi i64 [ %iv.next, %L110 ], [ 0, %L107 ]
  %value_phi95 = phi i32 [ %38, %L107 ], [ %39, %L110 ]
  %value_phi96 = phi i32 [ %35, %L107 ], [ %48, %L110 ]
  %value_phi97 = phi i64 [ %v.i, %L107 ], [ %44, %L110 ]
  %iv.next = add nuw nsw i64 %iv, 1, !dbg !3670
  %39 = sub i32 %value_phi95, %value_phi96, !dbg !3670
  %40 = sub i32 64, %39, !dbg !3672
  %41 = zext i32 %40 to i64, !dbg !3674
  %42 = icmp ugt i32 %40, 63, !dbg !3674
  %notmask = shl nsw i64 -1, %41, !dbg !3672
  %.op = xor i64 %notmask, -1, !dbg !3672
  %43 = select i1 %42, i64 -1, i64 %.op, !dbg !3672
  %44 = and i64 %43, %value_phi97, !dbg !3675
  %45 = xor i64 %44, %value_phi97, !dbg !3677
  %46 = call i64 @llvm.ctpop.i64(i64 %45) #94, !dbg !3678, !range !2422
  %47 = trunc i64 %46 to i32, !dbg !3680
  %48 = add i32 %value_phi96, %47, !dbg !3681
  %.not176 = icmp eq i32 %48, 0, !dbg !3682
  br i1 %.not176, label %L131, label %L110, !dbg !3683

L131:                                             ; preds = %L110
  %49 = xor i64 %44, -1, !dbg !3684
  %50 = and i64 %v.i, %49, !dbg !3686
  store atomic i64 %50, i64* %p.i release, align 16, !dbg !3687, !noalias !3688
  br label %L142, !dbg !3669

L142:                                             ; preds = %L131, %L94
  %value_phi48 = phi i32 [ %30, %L131 ], [ %34, %L94 ]
  %value_phi49 = phi i64 [ %44, %L131 ], [ %v.i, %L94 ]
  %51 = icmp sgt i32 %value_phi48, 0, !dbg !3689
  br i1 %51, label %L198.lr.ph, label %L383.lr.ph, !dbg !3690

L198.lr.ph:                                       ; preds = %L142
  %52 = zext i32 %value_phi48 to i64, !dbg !3691
  %53 = add nuw nsw i64 %52, 1, !dbg !3708
  %54 = udiv i64 %arraylen35, %53, !dbg !3710
  %55 = mul i64 %54, %53, !dbg !3711
  %56 = sub i64 %arraylen35, %55, !dbg !3713
  %57 = call nonnull "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" {}* @julia.pointer_from_objref({} addrspace(11)* noundef %23) #97, !dbg !3714
  %58 = bitcast {}* %57 to i8**, !dbg !3714
  %arrayptr52 = load i8*, i8** %58, align 8, !dbg !3714, !tbaa !245, !alias.scope !183, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BFloat64\7D !0
  %59 = ptrtoint i8* %arrayptr52 to i64, !dbg !3714
  %arraylen54 = load i64, i64 addrspace(11)* %arraylen_ptr34, align 8, !dbg !3724, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %60 = call nonnull "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" {}* @julia.pointer_from_objref({} addrspace(11)* noundef %25) #97, !dbg !3730
  %61 = bitcast {}* %60 to i8**, !dbg !3730
  %arrayptr59 = load i8*, i8** %61, align 8, !dbg !3730, !tbaa !245, !alias.scope !183, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BSMatrix\7B2\2C\202\2C\20ComplexF64\2C\204\7D\7D !0
  %62 = ptrtoint i8* %arrayptr59 to i64, !dbg !3730
  %arraylen61 = load i64, i64 addrspace(11)* %arraylen_ptr37, align 8, !dbg !3736, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %63 = addrspacecast {} addrspace(10)* %1 to {} addrspace(11)*, !dbg !3714
  %64 = call nonnull "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" {}* @julia.pointer_from_objref({} addrspace(11)* noundef %63) #97, !dbg !3714
  %65 = bitcast {}* %64 to i8**, !dbg !3714
  %arrayptr66 = load i8*, i8** %65, align 8, !dbg !3714, !tbaa !245, !alias.scope !183, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BFloat64\7D !0
  %66 = ptrtoint i8* %arrayptr66 to i64, !dbg !3714
  %67 = addrspacecast {} addrspace(10)* %1 to { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)*, !dbg !3724
  %arraylen_ptr67 = getelementptr inbounds { i8 addrspace(13)*, i64, i16, i16, i32 }, { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)* %67, i64 0, i32 1, !dbg !3724
  %arraylen68 = load i64, i64 addrspace(11)* %arraylen_ptr67, align 8, !dbg !3724, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %68 = getelementptr inbounds { i64, [1 x i64] }, { i64, [1 x i64] }* %newstruct71, i64 0, i32 1, i64 0, !dbg !3742
  store i64 %arraylen68, i64* %68, align 8, !dbg !3742, !tbaa !94, !alias.scope !96, !noalias !3743
  %69 = getelementptr inbounds { i64, [1 x i64] }, { i64, [1 x i64] }* %newstruct71, i64 0, i32 0, !dbg !3744
  store i64 %66, i64* %69, align 8, !dbg !3744, !tbaa !94, !alias.scope !96, !noalias !3743
  %newstruct72.sroa.4.32..sroa_cast = bitcast { i64, [1 x i64] }* %newstruct71 to i8*, !dbg !3721
  %newstruct72.sroa.0.sroa.0.0.newstruct72.sroa.0.0..sroa_cast.sroa_cast = getelementptr inbounds { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, i64 0, i32 0, i32 0, !dbg !3747
  store i64 %59, i64* %newstruct72.sroa.0.sroa.0.0.newstruct72.sroa.0.0..sroa_cast.sroa_cast, align 16, !dbg !3747, !tbaa !462, !alias.scope !2556, !noalias !3748
  %newstruct72.sroa.0.sroa.2.0.newstruct72.sroa.0.0..sroa_cast.sroa_cast = getelementptr inbounds { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, i64 0, i32 0, i32 1, i64 0, !dbg !3747
  store i64 %arraylen54, i64* %newstruct72.sroa.0.sroa.2.0.newstruct72.sroa.0.0..sroa_cast.sroa_cast, align 8, !dbg !3747, !tbaa !462, !alias.scope !2556, !noalias !3748
  %newstruct72.sroa.2.0..sroa_idx128 = getelementptr inbounds { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, i64 0, i32 1, i32 0, !dbg !3747
  store i64 %62, i64* %newstruct72.sroa.2.0..sroa_idx128, align 16, !dbg !3747, !tbaa !462, !alias.scope !2556, !noalias !3748
  %newstruct72.sroa.3.0..sroa_idx129 = getelementptr inbounds { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, i64 0, i32 1, i32 1, i64 0, !dbg !3747
  store i64 %arraylen61, i64* %newstruct72.sroa.3.0..sroa_idx129, align 8, !dbg !3747, !tbaa !462, !alias.scope !2556, !noalias !3748
  %newstruct72.sroa.4.0..sroa_idx = getelementptr inbounds { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, i64 0, i32 2, !dbg !3747
  %newstruct72.sroa.4.0..sroa_cast = bitcast { i64, [1 x i64] }* %newstruct72.sroa.4.0..sroa_idx to i8*, !dbg !3747
  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(16) %newstruct72.sroa.4.0..sroa_cast, i8* noundef nonnull align 8 dereferenceable(16) %newstruct72.sroa.4.32..sroa_cast, i64 noundef 16, i1 noundef false) #94, !dbg !3747
  %70 = call token (...) @llvm.julia.gc_preserve_begin({} addrspace(10)* nonnull %2, {} addrspace(10)* nonnull %0, {} addrspace(10)* nonnull %1) #94, !dbg !3721
  %71 = icmp sgt i64 %56, -1
  br label %L198, !dbg !3749

L264.preheader:                                   ; preds = %L252
  %value_phi84189 = add i64 %83, 1, !dbg !3750
  %.not173190 = icmp sgt i64 %value_phi84189, %arraylen35, !dbg !3751
  br i1 %.not173190, label %L331.preheader, label %L281.lr.ph, !dbg !3752

L281.lr.ph:                                       ; preds = %L264.preheader
  %72 = addrspacecast { i64, [1 x i64] }* %newstruct71 to { i64, [1 x i64] } addrspace(11)*
  %73 = bitcast [1 x [4 x [2 x double]]]* %11 to i8*
  %74 = add i64 %54, %value_phi76194, !dbg !3754
  %umin = call i1 @llvm.umin.i1(i1 %80, i1 %71), !dbg !3752
  %75 = zext i1 %umin to i64, !dbg !3752
  %76 = add i64 %74, %75, !dbg !3754
  br label %L281, !dbg !3752

L198:                                             ; preds = %L252, %L198.lr.ph
  %iv2 = phi i64 [ %iv.next3, %L252 ], [ 0, %L198.lr.ph ]
  %value_phi78196 = phi i64 [ %value_phi49, %L198.lr.ph ], [ %89, %L252 ]
  %value_phi76194 = phi i64 [ 0, %L198.lr.ph ], [ %83, %L252 ]
  %value_phi75193 = phi i32 [ 0, %L198.lr.ph ], [ %85, %L252 ]
  %iv.next3 = add nuw nsw i64 %iv2, 1, !dbg !3760
  %77 = icmp ne i64 %value_phi78196, 0, !dbg !3760
  call void @llvm.assume(i1 noundef %77) #94, !dbg !3763
  %78 = call i64 @llvm.cttz.i64(i64 %value_phi78196, i1 noundef true) #94, !dbg !3764, !range !2422
  %79 = trunc i64 %78 to i32, !dbg !3766
  %80 = icmp ugt i64 %56, %iv2, !dbg !3767
  %not.ifelse_cond79 = and i1 %71, %80, !dbg !3771
  %81 = zext i1 %not.ifelse_cond79 to i64, !dbg !3771
  %82 = add i64 %value_phi76194, %54, !dbg !3771
  %83 = add i64 %82, %81, !dbg !3772
  %84 = add nuw nsw i32 %79, 1, !dbg !3773
  %85 = add i32 %84, %value_phi75193, !dbg !3775
  %86 = zext i32 %84 to i64, !dbg !3777
  %87 = lshr i64 %value_phi78196, %86, !dbg !3777
  %88 = icmp eq i32 %79, 63, !dbg !3777
  %89 = select i1 %88, i64 0, i64 %87, !dbg !3777
  %90 = load i64, i64* inttoptr (i64 123246368836896 to i64*), align 32, !dbg !3779, !tbaa !77, !alias.scope !81, !noalias !84
  %91 = shl i32 %85, 9, !dbg !3785
  %92 = zext i32 %91 to i64, !dbg !3786
  %93 = inttoptr i64 %90 to i8*, !dbg !3790
  %94 = getelementptr i8, i8* %93, i64 %92, !dbg !3790
  %95 = getelementptr i8, i8* %94, i64 8, !dbg !3791
  %coercion = bitcast i8* %95 to i64*, !dbg !3797
  store i64 ptrtoint (void (i64)* @jlcapi_BatchClosure_600 to i64), i64* %coercion, align 1, !dbg !3797, !tbaa !134, !alias.scope !81, !noalias !3612
  %96 = getelementptr i8, i8* %94, i64 16, !dbg !3801
  %97 = bitcast i8* %96 to { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }**, !dbg !3805
  store { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }** %97, align 1, !dbg !3805, !tbaa !134, !alias.scope !81, !noalias !3612
  %98 = getelementptr i8, i8* %94, i64 24, !dbg !3809
  %coercion81 = bitcast i8* %98 to i64*, !dbg !3813
  store i64 %value_phi76194, i64* %coercion81, align 1, !dbg !3813, !tbaa !134, !alias.scope !81, !noalias !3612
  %99 = getelementptr i8, i8* %94, i64 32, !dbg !3817
  %coercion82 = bitcast i8* %99 to i64*, !dbg !3821
  store i64 %83, i64* %coercion82, align 1, !dbg !3821, !tbaa !134, !alias.scope !81, !noalias !3612
  %p.i119 = bitcast i8* %94 to i32*, !dbg !3825
  %v.i120 = atomicrmw xchg i32* %p.i119, i32 0 acq_rel, align 4, !dbg !3825
  %.not172 = icmp eq i32 %v.i120, 1, !dbg !3828
  br i1 %.not172, label %L249, label %L252, !dbg !3829

L249:                                             ; preds = %L198
  call fastcc void @julia_wake_thread__782(i32 zeroext %85) #94, !dbg !3829
  br label %L252, !dbg !3829

L252:                                             ; preds = %L249, %L198
  %100 = icmp eq i64 %iv.next3, %52, !dbg !3830
  br i1 %100, label %L264.preheader, label %L198, !dbg !3749

L331.preheader.loopexit:                          ; preds = %L281
  br label %L331.preheader, !dbg !3832

L331.preheader:                                   ; preds = %L331.preheader.loopexit, %L264.preheader
  %101 = icmp eq i64 %value_phi49, 0, !dbg !3832
  br i1 %101, label %L368, label %L336.preheader, !dbg !3834

L336.preheader:                                   ; preds = %L331.preheader
  br label %L336, !dbg !3835

L281:                                             ; preds = %L281, %L281.lr.ph
  %iv4 = phi i64 [ %iv.next5, %L281 ], [ 0, %L281.lr.ph ]
  %102 = add i64 %76, %iv4, !dbg !3754
  %iv.next5 = add nuw nsw i64 %iv4, 1, !dbg !3754
  %103 = add i64 %value_phi84189, %iv4, !dbg !3754
  %104 = shl i64 %102, 3, !dbg !3754
  %105 = getelementptr i8, i8* %arrayptr52, i64 %104, !dbg !3838
  %coercion86 = bitcast i8* %105 to double*, !dbg !3839
  %pointerref = load double, double* %coercion86, align 1, !dbg !3839, !tbaa !134, !alias.scope !81, !noalias !84
  call void @llvm.lifetime.end.p0i8(i64 noundef 48, i8* noundef nonnull %.sub) #94
  call fastcc void @julia_cascade_770([1 x [4 x [2 x double]]]* noalias nocapture nofree noundef nonnull writeonly sret([1 x [4 x [2 x double]]]) align 8 dereferenceable(64) %11, { i64, [1 x i64] } addrspace(11)* nocapture nofree noundef nonnull readonly align 8 dereferenceable(16) %72, double %pointerref) #94, !dbg !3843
  %106 = shl i64 %102, 6, !dbg !3844
  %coercion88 = getelementptr i8, i8* %arrayptr59, i64 %106, !dbg !3848
  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 1 dereferenceable(64) %coercion88, i8* noundef nonnull align 8 dereferenceable(64) %73, i64 noundef 64, i1 noundef false) #94, !dbg !3849, !noalias !3688
  %value_phi84 = add i64 %103, 1, !dbg !3750
  %exitcond200 = icmp eq i64 %103, %arraylen35, !dbg !3751
  br i1 %exitcond200, label %L331.preheader.loopexit, label %L281, !dbg !3752

L336:                                             ; preds = %L336.preheader, %L366
  %iv6 = phi i64 [ 0, %L336.preheader ], [ %iv.next7, %L366 ]
  %value_phi92188 = phi i64 [ %111, %L366 ], [ %value_phi49, %L336.preheader ]
  %value_phi91187 = phi i32 [ %113, %L366 ], [ 0, %L336.preheader ]
  %iv.next7 = add nuw nsw i64 %iv6, 1, !dbg !3853
  %107 = call i64 @llvm.cttz.i64(i64 %value_phi92188, i1 noundef true) #94, !dbg !3853, !range !2422
  %108 = trunc i64 %107 to i32, !dbg !3855
  %109 = add nuw nsw i32 %108, 1, !dbg !3856
  %110 = zext i32 %109 to i64, !dbg !3858
  %111 = lshr i64 %value_phi92188, %110, !dbg !3858
  %112 = icmp eq i32 %108, 63, !dbg !3858
  %113 = add i32 %109, %value_phi91187, !dbg !3860
  %114 = load i64, i64* inttoptr (i64 123246368836896 to i64*), align 32, !dbg !3862, !tbaa !77, !alias.scope !81, !noalias !84
  %115 = shl i32 %113, 9, !dbg !3865
  %116 = zext i32 %115 to i64, !dbg !3866
  %117 = inttoptr i64 %114 to i8*, !dbg !3870
  %118 = getelementptr i8, i8* %117, i64 %116, !dbg !3870
  %p.i121 = bitcast i8* %118 to i32*, !dbg !3871
  %v.i122184 = load atomic i32, i32* %p.i121 acquire, align 16, !dbg !3871
  %.not174185 = icmp eq i32 %v.i122184, 0, !dbg !3873
  br i1 %.not174185, label %L356.preheader, label %L366, !dbg !3835

L356.preheader:                                   ; preds = %L336
  br label %L356, !dbg !3874

L356:                                             ; preds = %L356.preheader, %L363
  %iv8 = phi i64 [ 0, %L356.preheader ], [ %iv.next9, %L363 ]
  %119 = trunc i64 %iv8 to i32
  %iv.next9 = add nuw nsw i64 %iv8, 1
  call void @llvm.lifetime.end.p0i8(i64 noundef 48, i8* noundef nonnull %.sub) #94
  call void asm sideeffect "pause", "~{memory}"() #98, !dbg !3875
  %120 = add i32 %119, 1, !dbg !3877
  %121 = icmp ult i32 %120, 65537, !dbg !3878
  br i1 %121, label %L363, label %L360, !dbg !3874

L360:                                             ; preds = %L356
  %122 = call fastcc i8 @julia_checktask_612(i32 zeroext %113) #94, !dbg !3880
  %123 = and i8 %122, 1, !dbg !3880
  %.not175 = icmp eq i8 %123, 0, !dbg !3880
  br i1 %.not175, label %L363, label %L366.loopexit, !dbg !3880

L363:                                             ; preds = %L360, %L356
  %v.i122 = load atomic i32, i32* %p.i121 acquire, align 16, !dbg !3871
  %.not174 = icmp eq i32 %v.i122, 0, !dbg !3873
  br i1 %.not174, label %L356, label %L366.loopexit, !dbg !3835

L366.loopexit:                                    ; preds = %L360, %L363
  br label %L366, !dbg !3832

L366:                                             ; preds = %L366.loopexit, %L336
  %124 = icmp eq i64 %111, 0, !dbg !3832
  %125 = select i1 %112, i1 true, i1 %124, !dbg !3832
  br i1 %125, label %L368.loopexit, label %L336, !dbg !3834

L368.loopexit:                                    ; preds = %L366
  br label %L368, !dbg !3881

L368:                                             ; preds = %L368.loopexit, %L331.preheader
  %v.i118 = atomicrmw or i64* %p.i, i64 %value_phi49 acq_rel, align 8, !dbg !3881
  call void @llvm.julia.gc_preserve_end(token %70) #94, !dbg !3884
  br label %L392, !dbg !3884

L383.lr.ph:                                       ; preds = %L142, %L84, %L76
  %126 = addrspacecast {} addrspace(10)* %2 to double addrspace(13)* addrspace(11)*
  %127 = addrspacecast {} addrspace(10)* %0 to [1 x [4 x [2 x double]]] addrspace(13)* addrspace(11)*
  %128 = bitcast [1 x [4 x [2 x double]]]* %6 to i8*
  %umax = call i64 @llvm.umax.i64(i64 %arraylen35, i64 noundef 1) #94, !dbg !3885
  br label %L383, !dbg !3885

L383:                                             ; preds = %L383, %L383.lr.ph
  %iv10 = phi i64 [ %iv.next11, %L383 ], [ 0, %L383.lr.ph ]
  %iv.next11 = add nuw nsw i64 %iv10, 1, !dbg !3887
  %129 = add nsw i64 %iv.next11, -1, !dbg !3887
  %arrayptr44170 = load double addrspace(13)*, double addrspace(13)* addrspace(11)* %126, align 16, !dbg !3887, !tbaa !245, !alias.scope !3890, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BFloat64\7D !0
  %130 = getelementptr inbounds double, double addrspace(13)* %arrayptr44170, i64 %129, !dbg !3887
  %arrayref45 = load double, double addrspace(13)* %130, align 8, !dbg !3887, !tbaa !325, !alias.scope !81, !noalias !84
  call fastcc void @julia_cascade_787([1 x [4 x [2 x double]]]* noalias nocapture nofree noundef nonnull writeonly sret([1 x [4 x [2 x double]]]) align 8 dereferenceable(64) %6, {} addrspace(10)* noundef nonnull align 16 dereferenceable(40) %1, double %arrayref45) #94, !dbg !3891
  %arrayptr47171 = load [1 x [4 x [2 x double]]] addrspace(13)*, [1 x [4 x [2 x double]]] addrspace(13)* addrspace(11)* %127, align 16, !dbg !3892, !tbaa !245, !alias.scope !3890, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BSMatrix\7B2\2C\202\2C\20ComplexF64\2C\204\7D\7D !0
  %131 = getelementptr inbounds [1 x [4 x [2 x double]]], [1 x [4 x [2 x double]]] addrspace(13)* %arrayptr47171, i64 %129, i64 0, !dbg !3892
  %132 = bitcast [4 x [2 x double]] addrspace(13)* %131 to i8 addrspace(13)*, !dbg !3892
  call void @llvm.memcpy.p13i8.p0i8.i64(i8 addrspace(13)* noundef align 8 dereferenceable(64) %132, i8* noundef nonnull align 8 dereferenceable(64) %128, i64 noundef 64, i1 noundef false) #94, !dbg !3892, !tbaa !462, !alias.scope !2556, !noalias !3748
  %133 = add nuw nsw i64 %iv.next11, 1, !dbg !3893
  %exitcond.not = icmp eq i64 %iv.next11, %umax, !dbg !3896
  br i1 %exitcond.not, label %L392.loopexit1, label %L383, !dbg !3885

L392.loopexit:                                    ; preds = %idxend16
  br label %L392

L392.loopexit1:                                   ; preds = %L383
  br label %L392

L392:                                             ; preds = %L392.loopexit1, %L392.loopexit, %L368, %L72, %L25
  call void @llvm.lifetime.end.p0i8(i64 noundef 48, i8* noundef nonnull %.sub) #94
  ret void, !dbg !3602

oob:                                              ; preds = %L34
  %errorbox = alloca i64, align 8, !dbg !3619
  store i64 %iv.next13, i64* %errorbox, align 8, !dbg !3619, !noalias !3688
  %134 = addrspacecast {} addrspace(10)* %2 to {} addrspace(12)*, !dbg !3619
  call void @ijl_bounds_error_ints({} addrspace(12)* noundef %134, i64* noundef nonnull align 8 %errorbox, i64 noundef 1) #99, !dbg !3619
  unreachable, !dbg !3619

idxend:                                           ; preds = %L34
  %arrayptr162 = load double addrspace(13)*, double addrspace(13)* addrspace(11)* %19, align 16, !dbg !3619, !tbaa !245, !alias.scope !3890, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BFloat64\7D !0
  %135 = getelementptr inbounds double, double addrspace(13)* %arrayptr162, i64 %22, !dbg !3619
  %arrayref = load double, double addrspace(13)* %135, align 8, !dbg !3619, !tbaa !325, !alias.scope !81, !noalias !84
  call fastcc void @julia_cascade_787([1 x [4 x [2 x double]]]* noalias nocapture nofree noundef nonnull writeonly sret([1 x [4 x [2 x double]]]) align 8 dereferenceable(64) %9, {} addrspace(10)* noundef nonnull align 16 dereferenceable(40) %1, double %arrayref) #94, !dbg !3897
  %arraylen12 = load i64, i64 addrspace(11)* %arraylen_ptr2, align 8, !dbg !3898, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %inbounds13 = icmp ult i64 %22, %arraylen12, !dbg !3898
  br i1 %inbounds13, label %idxend16, label %oob14, !dbg !3898

oob14:                                            ; preds = %idxend
  %errorbox15 = alloca i64, align 8, !dbg !3898
  store i64 %iv.next13, i64* %errorbox15, align 8, !dbg !3898, !noalias !3688
  %136 = addrspacecast {} addrspace(10)* %0 to {} addrspace(12)*, !dbg !3898
  call void @ijl_bounds_error_ints({} addrspace(12)* noundef %136, i64* noundef nonnull align 8 %errorbox15, i64 noundef 1) #99, !dbg !3898
  unreachable, !dbg !3898

idxend16:                                         ; preds = %idxend
  %arrayptr18163 = load [1 x [4 x [2 x double]]] addrspace(13)*, [1 x [4 x [2 x double]]] addrspace(13)* addrspace(11)* %20, align 16, !dbg !3898, !tbaa !245, !alias.scope !3890, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BSMatrix\7B2\2C\202\2C\20ComplexF64\2C\204\7D\7D !0
  %137 = getelementptr inbounds [1 x [4 x [2 x double]]], [1 x [4 x [2 x double]]] addrspace(13)* %arrayptr18163, i64 %22, i64 0, !dbg !3898
  %138 = bitcast [4 x [2 x double]] addrspace(13)* %137 to i8 addrspace(13)*, !dbg !3898
  call void @llvm.memcpy.p13i8.p0i8.i64(i8 addrspace(13)* noundef align 8 dereferenceable(64) %138, i8* noundef nonnull align 8 dereferenceable(64) %21, i64 noundef 64, i1 noundef false) #94, !dbg !3898, !tbaa !462, !alias.scope !2556, !noalias !3748
  %.not164 = icmp eq i64 %iv.next13, %arraylen, !dbg !3899
  %139 = add nuw nsw i64 %iv.next13, 1, !dbg !3669
  br i1 %.not164, label %L392.loopexit, label %L34, !dbg !3602
}

; Function Attrs: mustprogress willreturn
define internal fastcc void @preprocess_julia_cascade__596({} addrspace(10)* noundef nonnull align 16 dereferenceable(40) "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" "enzymejl_parmtype"="123247878569488" "enzymejl_parmtype_ref"="2" %0, {} addrspace(10)* noundef nonnull align 16 dereferenceable(40) "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" "enzymejl_parmtype"="123247705933040" "enzymejl_parmtype_ref"="2" %1, {} addrspace(10)* noundef nonnull align 16 dereferenceable(40) "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" "enzymejl_parmtype"="123247705933040" "enzymejl_parmtype_ref"="2" %2) unnamed_addr #93 !dbg !3590 {
top:
  %3 = call noalias nonnull dereferenceable(48) dereferenceable_or_null(48) i8* @malloc(i64 48), !enzyme_fromstack !142
  %4 = bitcast i8* %3 to { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }*, !enzyme_caststack !0
  %.sub = bitcast { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4 to i8*
  %5 = call noalias nonnull dereferenceable(64) dereferenceable_or_null(64) i8* @malloc(i64 64), !enzymejl_allocart !2322, !enzyme_type !329, !enzyme_fromstack !89
  %6 = bitcast i8* %5 to [1 x [4 x [2 x double]]]*, !enzyme_caststack !0
  %7 = call noalias nonnull dereferenceable(16) dereferenceable_or_null(16) i8* @malloc(i64 16), !enzyme_fromstack !89
  %newstruct71 = bitcast i8* %7 to { i64, [1 x i64] }*, !enzyme_caststack !0
  %8 = call noalias nonnull dereferenceable(64) dereferenceable_or_null(64) i8* @malloc(i64 64), !enzymejl_allocart !2322, !enzyme_type !329, !enzyme_fromstack !89
  %9 = bitcast i8* %8 to [1 x [4 x [2 x double]]]*, !enzyme_caststack !0
  %10 = call noalias nonnull dereferenceable(64) dereferenceable_or_null(64) i8* @malloc(i64 64), !enzymejl_allocart !2322, !enzyme_type !329, !enzyme_fromstack !89
  %11 = bitcast i8* %10 to [1 x [4 x [2 x double]]]*, !enzyme_caststack !0
  %12 = call {}*** @julia.get_pgcstack() #94
  %current_task1156 = getelementptr inbounds {}**, {}*** %12, i64 -14
  %current_task1 = bitcast {}*** %current_task1156 to {}**
  %ptls_field157 = getelementptr inbounds {}**, {}*** %12, i64 2
  %13 = bitcast {}*** %ptls_field157 to i64***
  %ptls_load158159 = load i64**, i64*** %13, align 8, !tbaa !66
  %14 = getelementptr inbounds i64*, i64** %ptls_load158159, i64 2
  %safepoint = load i64*, i64** %14, align 8, !tbaa !70
  fence syncscope("singlethread") seq_cst
  call void @julia.safepoint(i64* %safepoint) #94, !dbg !3591
  fence syncscope("singlethread") seq_cst
  %15 = call i64 @julia_nthreads_790() #95, !dbg !3592
  %.not = icmp eq i64 %15, 1, !dbg !3594
  br i1 %.not, label %L4, label %L51, !dbg !3595

L4:                                               ; preds = %top
  %16 = addrspacecast {} addrspace(10)* %2 to { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)*, !dbg !3596
  %arraylen_ptr = getelementptr inbounds { i8 addrspace(13)*, i64, i16, i16, i32 }, { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)* %16, i64 0, i32 1, !dbg !3596
  %arraylen = load i64, i64 addrspace(11)* %arraylen_ptr, align 8, !dbg !3596, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %17 = addrspacecast {} addrspace(10)* %0 to { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)*, !dbg !3603
  %arraylen_ptr2 = getelementptr inbounds { i8 addrspace(13)*, i64, i16, i16, i32 }, { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)* %17, i64 0, i32 1, !dbg !3603
  %arraylen3 = load i64, i64 addrspace(11)* %arraylen_ptr2, align 8, !dbg !3603, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %.not160 = icmp eq i64 %arraylen, %arraylen3, !dbg !3610
  br i1 %.not160, label %L25, label %L14, !dbg !3609

L14:                                              ; preds = %L4
  %box = call noalias nonnull dereferenceable(8) "enzyme_inactive" "enzyme_type"="{[-1]:Pointer, [-1,-1]:Integer}" {} addrspace(10)* @julia.gc_alloc_obj({}** nonnull %current_task1, i64 noundef 8, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247752959712 to {}*) to {} addrspace(10)*)) #96, !dbg !3609
  %memcpy_refined_dst = bitcast {} addrspace(10)* %box to i64 addrspace(10)*, !dbg !3609, !enzyme_inactive !0
  store i64 %arraylen, i64 addrspace(10)* %memcpy_refined_dst, align 8, !dbg !3609, !tbaa !230, !alias.scope !81, !noalias !3612
  %box31 = call noalias nonnull dereferenceable(8) "enzyme_inactive" "enzyme_type"="{[-1]:Pointer, [-1,-1]:Integer}" {} addrspace(10)* @julia.gc_alloc_obj({}** nonnull %current_task1, i64 noundef 8, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247752959712 to {}*) to {} addrspace(10)*)) #96, !dbg !3609
  %memcpy_refined_dst33 = bitcast {} addrspace(10)* %box31 to i64 addrspace(10)*, !dbg !3609, !enzyme_inactive !0
  store i64 %arraylen3, i64 addrspace(10)* %memcpy_refined_dst33, align 8, !dbg !3609, !tbaa !230, !alias.scope !81, !noalias !3612
  %18 = call nonnull "enzyme_type"="{[-1]:Pointer}" {} addrspace(10)* ({} addrspace(10)* ({} addrspace(10)*, {} addrspace(10)**, i32, {} addrspace(10)*)*, {} addrspace(10)*, {} addrspace(10)*, ...) @julia.call2({} addrspace(10)* ({} addrspace(10)*, {} addrspace(10)**, i32, {} addrspace(10)*)* noundef nonnull @ijl_invoke, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247893132112 to {}*) to {} addrspace(10)*), {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247774899088 to {}*) to {} addrspace(10)*), {} addrspace(10)* addrspacecast ({}* inttoptr (i64 123247744295344 to {}*) to {} addrspace(10)*), {} addrspace(10)* nofree nonnull %box, {} addrspace(10)* nofree nonnull %box31) #95, !dbg !3609
  unreachable, !dbg !3609

L25:                                              ; preds = %L4
  %.not161 = icmp eq i64 %arraylen, 0, !dbg !3615
  br i1 %.not161, label %L392, label %L34.preheader, !dbg !3602

L34.preheader:                                    ; preds = %L25
  %19 = addrspacecast {} addrspace(10)* %2 to double addrspace(13)* addrspace(11)*
  %20 = addrspacecast {} addrspace(10)* %0 to [1 x [4 x [2 x double]]] addrspace(13)* addrspace(11)*
  %21 = bitcast [1 x [4 x [2 x double]]]* %9 to i8*
  br label %L34, !dbg !3619

L34:                                              ; preds = %idxend16, %L34.preheader
  %iv12 = phi i64 [ %iv.next13, %idxend16 ], [ 0, %L34.preheader ]
  %iv.next13 = add nuw nsw i64 %iv12, 1, !dbg !3619
  %22 = add nsw i64 %iv.next13, -1, !dbg !3619
  %arraylen10 = load i64, i64 addrspace(11)* %arraylen_ptr, align 8, !dbg !3619, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %inbounds = icmp ult i64 %22, %arraylen10, !dbg !3619
  br i1 %inbounds, label %idxend, label %oob, !dbg !3619

L51:                                              ; preds = %top
  %23 = addrspacecast {} addrspace(10)* %2 to {} addrspace(11)*, !dbg !3621
  %24 = addrspacecast {} addrspace(10)* %2 to { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)*, !dbg !3621
  %arraylen_ptr34 = getelementptr inbounds { i8 addrspace(13)*, i64, i16, i16, i32 }, { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)* %24, i64 0, i32 1, !dbg !3621
  %arraylen35 = load i64, i64 addrspace(11)* %arraylen_ptr34, align 8, !dbg !3621, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %25 = addrspacecast {} addrspace(10)* %0 to {} addrspace(11)*, !dbg !3628
  %26 = addrspacecast {} addrspace(10)* %0 to { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)*, !dbg !3628
  %arraylen_ptr37 = getelementptr inbounds { i8 addrspace(13)*, i64, i16, i16, i32 }, { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)* %26, i64 0, i32 1, !dbg !3628
  %arraylen38 = load i64, i64 addrspace(11)* %arraylen_ptr37, align 8, !dbg !3628, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %.not165 = icmp eq i64 %arraylen35, %arraylen38, !dbg !3635
  br i1 %.not165, label %L72, label %L61, !dbg !3634

L61:                                              ; preds = %L51
  %box108 = call noalias nonnull dereferenceable(8) "enzyme_inactive" "enzyme_type"="{[-1]:Pointer, [-1,-1]:Integer}" {} addrspace(10)* @julia.gc_alloc_obj({}** nonnull %current_task1, i64 noundef 8, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247752959712 to {}*) to {} addrspace(10)*)) #96, !dbg !3634
  %memcpy_refined_dst110 = bitcast {} addrspace(10)* %box108 to i64 addrspace(10)*, !dbg !3634, !enzyme_inactive !0
  store i64 %arraylen35, i64 addrspace(10)* %memcpy_refined_dst110, align 8, !dbg !3634, !tbaa !230, !alias.scope !81, !noalias !3612
  %box112 = call noalias nonnull dereferenceable(8) "enzyme_inactive" "enzyme_type"="{[-1]:Pointer, [-1,-1]:Integer}" {} addrspace(10)* @julia.gc_alloc_obj({}** nonnull %current_task1, i64 noundef 8, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247752959712 to {}*) to {} addrspace(10)*)) #96, !dbg !3634
  %memcpy_refined_dst114 = bitcast {} addrspace(10)* %box112 to i64 addrspace(10)*, !dbg !3634, !enzyme_inactive !0
  store i64 %arraylen38, i64 addrspace(10)* %memcpy_refined_dst114, align 8, !dbg !3634, !tbaa !230, !alias.scope !81, !noalias !3612
  %27 = call nonnull "enzyme_type"="{[-1]:Pointer}" {} addrspace(10)* ({} addrspace(10)* ({} addrspace(10)*, {} addrspace(10)**, i32, {} addrspace(10)*)*, {} addrspace(10)*, {} addrspace(10)*, ...) @julia.call2({} addrspace(10)* ({} addrspace(10)*, {} addrspace(10)**, i32, {} addrspace(10)*)* noundef nonnull @ijl_invoke, {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247893132112 to {}*) to {} addrspace(10)*), {} addrspace(10)* noundef addrspacecast ({}* inttoptr (i64 123247774899088 to {}*) to {} addrspace(10)*), {} addrspace(10)* addrspacecast ({}* inttoptr (i64 123247744295344 to {}*) to {} addrspace(10)*), {} addrspace(10)* nofree nonnull %box108, {} addrspace(10)* nofree nonnull %box112) #95, !dbg !3634
  unreachable, !dbg !3634

L72:                                              ; preds = %L51
  %.not167 = icmp eq i64 %arraylen35, 0, !dbg !3637
  br i1 %.not167, label %L392, label %L76, !dbg !3639

L76:                                              ; preds = %L72
  %28 = call i64 @llvm.smin.i64(i64 %15, i64 %arraylen35) #94, !dbg !3641
  %.not168 = icmp eq i64 %28, 0, !dbg !3643
  br i1 %.not168, label %L383.lr.ph, label %L84, !dbg !3644

L84:                                              ; preds = %L76
  %29 = trunc i64 %28 to i32, !dbg !3645
  %30 = add i32 %29, -1, !dbg !3645
  %31 = call nonnull {}* @julia.pointer_from_objref({} addrspace(11)* noundef addrspacecast ({}* inttoptr (i64 123247601534016 to {}*) to {} addrspace(11)*)) #97, !dbg !3649
  %32 = icmp sgt i32 %30, 0, !dbg !3651
  br i1 %32, label %L94, label %L383.lr.ph, !dbg !3652

L94:                                              ; preds = %L84
  %p.i = bitcast {}* %31 to i64*, !dbg !3654
  %v.i = atomicrmw xchg i64* %p.i, i64 0 acq_rel, align 8, !dbg !3654
  %33 = call i64 @llvm.ctpop.i64(i64 %v.i) #94, !dbg !3657, !range !2422
  %34 = trunc i64 %33 to i32, !dbg !3659
  %35 = sub nsw i32 %30, %34, !dbg !3660
  %36 = icmp slt i32 %35, 0, !dbg !3662
  br i1 %36, label %L107, label %L142, !dbg !3665

L107:                                             ; preds = %L94
  %37 = call i64 @llvm.ctlz.i64(i64 %v.i, i1 noundef false) #94, !dbg !3666, !range !2422
  %38 = trunc i64 %37 to i32, !dbg !3668
  br label %L110, !dbg !3669

L110:                                             ; preds = %L110, %L107
  %iv = phi i64 [ %iv.next, %L110 ], [ 0, %L107 ]
  %value_phi95 = phi i32 [ %38, %L107 ], [ %39, %L110 ]
  %value_phi96 = phi i32 [ %35, %L107 ], [ %48, %L110 ]
  %value_phi97 = phi i64 [ %v.i, %L107 ], [ %44, %L110 ]
  %iv.next = add nuw nsw i64 %iv, 1, !dbg !3670
  %39 = sub i32 %value_phi95, %value_phi96, !dbg !3670
  %40 = sub i32 64, %39, !dbg !3672
  %41 = zext i32 %40 to i64, !dbg !3674
  %42 = icmp ugt i32 %40, 63, !dbg !3674
  %notmask = shl nsw i64 -1, %41, !dbg !3672
  %.op = xor i64 %notmask, -1, !dbg !3672
  %43 = select i1 %42, i64 -1, i64 %.op, !dbg !3672
  %44 = and i64 %43, %value_phi97, !dbg !3675
  %45 = xor i64 %44, %value_phi97, !dbg !3677
  %46 = call i64 @llvm.ctpop.i64(i64 %45) #94, !dbg !3678, !range !2422
  %47 = trunc i64 %46 to i32, !dbg !3680
  %48 = add i32 %value_phi96, %47, !dbg !3681
  %.not176 = icmp eq i32 %48, 0, !dbg !3682
  br i1 %.not176, label %L131, label %L110, !dbg !3683

L131:                                             ; preds = %L110
  %49 = xor i64 %44, -1, !dbg !3684
  %50 = and i64 %v.i, %49, !dbg !3686
  store atomic i64 %50, i64* %p.i release, align 16, !dbg !3687, !noalias !3688
  br label %L142, !dbg !3669

L142:                                             ; preds = %L131, %L94
  %value_phi48 = phi i32 [ %30, %L131 ], [ %34, %L94 ]
  %value_phi49 = phi i64 [ %44, %L131 ], [ %v.i, %L94 ]
  %51 = icmp sgt i32 %value_phi48, 0, !dbg !3689
  br i1 %51, label %L198.lr.ph, label %L383.lr.ph, !dbg !3690

L198.lr.ph:                                       ; preds = %L142
  %52 = zext i32 %value_phi48 to i64, !dbg !3691
  %53 = add nuw nsw i64 %52, 1, !dbg !3708
  %54 = udiv i64 %arraylen35, %53, !dbg !3710
  %55 = mul i64 %54, %53, !dbg !3711
  %56 = sub i64 %arraylen35, %55, !dbg !3713
  %57 = call nonnull "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" {}* @julia.pointer_from_objref({} addrspace(11)* noundef %23) #97, !dbg !3714
  %58 = bitcast {}* %57 to i8**, !dbg !3714
  %arrayptr52 = load i8*, i8** %58, align 8, !dbg !3714, !tbaa !245, !alias.scope !183, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BFloat64\7D !0
  %59 = ptrtoint i8* %arrayptr52 to i64, !dbg !3714
  %arraylen54 = load i64, i64 addrspace(11)* %arraylen_ptr34, align 8, !dbg !3724, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %60 = call nonnull "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" {}* @julia.pointer_from_objref({} addrspace(11)* noundef %25) #97, !dbg !3730
  %61 = bitcast {}* %60 to i8**, !dbg !3730
  %arrayptr59 = load i8*, i8** %61, align 8, !dbg !3730, !tbaa !245, !alias.scope !183, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BSMatrix\7B2\2C\202\2C\20ComplexF64\2C\204\7D\7D !0
  %62 = ptrtoint i8* %arrayptr59 to i64, !dbg !3730
  %arraylen61 = load i64, i64 addrspace(11)* %arraylen_ptr37, align 8, !dbg !3736, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %63 = addrspacecast {} addrspace(10)* %1 to {} addrspace(11)*, !dbg !3714
  %64 = call nonnull "enzyme_type"="{[-1]:Pointer, [-1,0]:Pointer, [-1,0,-1]:Float@double, [-1,8]:Integer, [-1,9]:Integer, [-1,10]:Integer, [-1,11]:Integer, [-1,12]:Integer, [-1,13]:Integer, [-1,14]:Integer, [-1,15]:Integer, [-1,16]:Integer, [-1,17]:Integer, [-1,18]:Integer, [-1,19]:Integer, [-1,20]:Integer, [-1,21]:Integer, [-1,22]:Integer, [-1,23]:Integer, [-1,24]:Integer, [-1,25]:Integer, [-1,26]:Integer, [-1,27]:Integer, [-1,28]:Integer, [-1,29]:Integer, [-1,30]:Integer, [-1,31]:Integer, [-1,32]:Integer, [-1,33]:Integer, [-1,34]:Integer, [-1,35]:Integer, [-1,36]:Integer, [-1,37]:Integer, [-1,38]:Integer, [-1,39]:Integer}" {}* @julia.pointer_from_objref({} addrspace(11)* noundef %63) #97, !dbg !3714
  %65 = bitcast {}* %64 to i8**, !dbg !3714
  %arrayptr66 = load i8*, i8** %65, align 8, !dbg !3714, !tbaa !245, !alias.scope !183, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BFloat64\7D !0
  %66 = ptrtoint i8* %arrayptr66 to i64, !dbg !3714
  %67 = addrspacecast {} addrspace(10)* %1 to { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)*, !dbg !3724
  %arraylen_ptr67 = getelementptr inbounds { i8 addrspace(13)*, i64, i16, i16, i32 }, { i8 addrspace(13)*, i64, i16, i16, i32 } addrspace(11)* %67, i64 0, i32 1, !dbg !3724
  %arraylen68 = load i64, i64 addrspace(11)* %arraylen_ptr67, align 8, !dbg !3724, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %68 = getelementptr inbounds { i64, [1 x i64] }, { i64, [1 x i64] }* %newstruct71, i64 0, i32 1, i64 0, !dbg !3742
  store i64 %arraylen68, i64* %68, align 8, !dbg !3742, !tbaa !94, !alias.scope !96, !noalias !3743
  %69 = getelementptr inbounds { i64, [1 x i64] }, { i64, [1 x i64] }* %newstruct71, i64 0, i32 0, !dbg !3744
  store i64 %66, i64* %69, align 8, !dbg !3744, !tbaa !94, !alias.scope !96, !noalias !3743
  %newstruct72.sroa.4.32..sroa_cast = bitcast { i64, [1 x i64] }* %newstruct71 to i8*, !dbg !3721
  %newstruct72.sroa.0.sroa.0.0.newstruct72.sroa.0.0..sroa_cast.sroa_cast = getelementptr inbounds { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, i64 0, i32 0, i32 0, !dbg !3747
  store i64 %59, i64* %newstruct72.sroa.0.sroa.0.0.newstruct72.sroa.0.0..sroa_cast.sroa_cast, align 16, !dbg !3747, !tbaa !462, !alias.scope !2556, !noalias !3748
  %newstruct72.sroa.0.sroa.2.0.newstruct72.sroa.0.0..sroa_cast.sroa_cast = getelementptr inbounds { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, i64 0, i32 0, i32 1, i64 0, !dbg !3747
  store i64 %arraylen54, i64* %newstruct72.sroa.0.sroa.2.0.newstruct72.sroa.0.0..sroa_cast.sroa_cast, align 8, !dbg !3747, !tbaa !462, !alias.scope !2556, !noalias !3748
  %newstruct72.sroa.2.0..sroa_idx128 = getelementptr inbounds { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, i64 0, i32 1, i32 0, !dbg !3747
  store i64 %62, i64* %newstruct72.sroa.2.0..sroa_idx128, align 16, !dbg !3747, !tbaa !462, !alias.scope !2556, !noalias !3748
  %newstruct72.sroa.3.0..sroa_idx129 = getelementptr inbounds { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, i64 0, i32 1, i32 1, i64 0, !dbg !3747
  store i64 %arraylen61, i64* %newstruct72.sroa.3.0..sroa_idx129, align 8, !dbg !3747, !tbaa !462, !alias.scope !2556, !noalias !3748
  %newstruct72.sroa.4.0..sroa_idx = getelementptr inbounds { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, i64 0, i32 2, !dbg !3747
  %newstruct72.sroa.4.0..sroa_cast = bitcast { i64, [1 x i64] }* %newstruct72.sroa.4.0..sroa_idx to i8*, !dbg !3747
  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(16) %newstruct72.sroa.4.0..sroa_cast, i8* noundef nonnull align 8 dereferenceable(16) %newstruct72.sroa.4.32..sroa_cast, i64 noundef 16, i1 noundef false) #94, !dbg !3747
  %70 = call token (...) @llvm.julia.gc_preserve_begin({} addrspace(10)* nonnull %2, {} addrspace(10)* nonnull %0, {} addrspace(10)* nonnull %1) #94, !dbg !3721
  %71 = icmp sgt i64 %56, -1
  br label %L198, !dbg !3749

L264.preheader:                                   ; preds = %L252
  %value_phi84189 = add i64 %83, 1, !dbg !3750
  %.not173190 = icmp sgt i64 %value_phi84189, %arraylen35, !dbg !3751
  br i1 %.not173190, label %L331.preheader, label %L281.lr.ph, !dbg !3752

L281.lr.ph:                                       ; preds = %L264.preheader
  %72 = addrspacecast { i64, [1 x i64] }* %newstruct71 to { i64, [1 x i64] } addrspace(11)*
  %73 = bitcast [1 x [4 x [2 x double]]]* %11 to i8*
  %74 = add i64 %54, %value_phi76194, !dbg !3754
  %umin = call i1 @llvm.umin.i1(i1 %80, i1 %71), !dbg !3752
  %75 = zext i1 %umin to i64, !dbg !3752
  %76 = add i64 %74, %75, !dbg !3754
  br label %L281, !dbg !3752

L198:                                             ; preds = %L252, %L198.lr.ph
  %iv2 = phi i64 [ %iv.next3, %L252 ], [ 0, %L198.lr.ph ]
  %value_phi78196 = phi i64 [ %value_phi49, %L198.lr.ph ], [ %89, %L252 ]
  %value_phi76194 = phi i64 [ 0, %L198.lr.ph ], [ %83, %L252 ]
  %value_phi75193 = phi i32 [ 0, %L198.lr.ph ], [ %85, %L252 ]
  %iv.next3 = add nuw nsw i64 %iv2, 1, !dbg !3760
  %77 = icmp ne i64 %value_phi78196, 0, !dbg !3760
  call void @llvm.assume(i1 noundef %77) #94, !dbg !3763
  %78 = call i64 @llvm.cttz.i64(i64 %value_phi78196, i1 noundef true) #94, !dbg !3764, !range !2422
  %79 = trunc i64 %78 to i32, !dbg !3766
  %80 = icmp ugt i64 %56, %iv2, !dbg !3767
  %not.ifelse_cond79 = and i1 %71, %80, !dbg !3771
  %81 = zext i1 %not.ifelse_cond79 to i64, !dbg !3771
  %82 = add i64 %value_phi76194, %54, !dbg !3771
  %83 = add i64 %82, %81, !dbg !3772
  %84 = add nuw nsw i32 %79, 1, !dbg !3773
  %85 = add i32 %84, %value_phi75193, !dbg !3775
  %86 = zext i32 %84 to i64, !dbg !3777
  %87 = lshr i64 %value_phi78196, %86, !dbg !3777
  %88 = icmp eq i32 %79, 63, !dbg !3777
  %89 = select i1 %88, i64 0, i64 %87, !dbg !3777
  %90 = load i64, i64* inttoptr (i64 123246368836896 to i64*), align 32, !dbg !3779, !tbaa !77, !alias.scope !81, !noalias !84
  %91 = shl i32 %85, 9, !dbg !3785
  %92 = zext i32 %91 to i64, !dbg !3786
  %93 = inttoptr i64 %90 to i8*, !dbg !3790
  %94 = getelementptr i8, i8* %93, i64 %92, !dbg !3790
  %95 = getelementptr i8, i8* %94, i64 8, !dbg !3791
  %coercion = bitcast i8* %95 to i64*, !dbg !3797
  store i64 ptrtoint (void (i64)* @jlcapi_BatchClosure_600 to i64), i64* %coercion, align 1, !dbg !3797, !tbaa !134, !alias.scope !81, !noalias !3612
  %96 = getelementptr i8, i8* %94, i64 16, !dbg !3801
  %97 = bitcast i8* %96 to { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }**, !dbg !3805
  store { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }* %4, { { i64, [1 x i64] }, { i64, [1 x i64] }, { i64, [1 x i64] } }** %97, align 1, !dbg !3805, !tbaa !134, !alias.scope !81, !noalias !3612
  %98 = getelementptr i8, i8* %94, i64 24, !dbg !3809
  %coercion81 = bitcast i8* %98 to i64*, !dbg !3813
  store i64 %value_phi76194, i64* %coercion81, align 1, !dbg !3813, !tbaa !134, !alias.scope !81, !noalias !3612
  %99 = getelementptr i8, i8* %94, i64 32, !dbg !3817
  %coercion82 = bitcast i8* %99 to i64*, !dbg !3821
  store i64 %83, i64* %coercion82, align 1, !dbg !3821, !tbaa !134, !alias.scope !81, !noalias !3612
  %p.i119 = bitcast i8* %94 to i32*, !dbg !3825
  %v.i120 = atomicrmw xchg i32* %p.i119, i32 0 acq_rel, align 4, !dbg !3825
  %.not172 = icmp eq i32 %v.i120, 1, !dbg !3828
  br i1 %.not172, label %L249, label %L252, !dbg !3829

L249:                                             ; preds = %L198
  call fastcc void @julia_wake_thread__782(i32 zeroext %85) #94, !dbg !3829
  br label %L252, !dbg !3829

L252:                                             ; preds = %L249, %L198
  %100 = icmp eq i64 %iv.next3, %52, !dbg !3830
  br i1 %100, label %L264.preheader, label %L198, !dbg !3749

L331.preheader.loopexit:                          ; preds = %L281
  br label %L331.preheader, !dbg !3832

L331.preheader:                                   ; preds = %L331.preheader.loopexit, %L264.preheader
  %101 = icmp eq i64 %value_phi49, 0, !dbg !3832
  br i1 %101, label %L368, label %L336.preheader, !dbg !3834

L336.preheader:                                   ; preds = %L331.preheader
  br label %L336, !dbg !3835

L281:                                             ; preds = %L281, %L281.lr.ph
  %iv4 = phi i64 [ %iv.next5, %L281 ], [ 0, %L281.lr.ph ]
  %102 = add i64 %76, %iv4, !dbg !3754
  %iv.next5 = add nuw nsw i64 %iv4, 1, !dbg !3754
  %103 = add i64 %value_phi84189, %iv4, !dbg !3754
  %104 = shl i64 %102, 3, !dbg !3754
  %105 = getelementptr i8, i8* %arrayptr52, i64 %104, !dbg !3838
  %coercion86 = bitcast i8* %105 to double*, !dbg !3839
  %pointerref = load double, double* %coercion86, align 1, !dbg !3839, !tbaa !134, !alias.scope !81, !noalias !84
  call void @llvm.lifetime.end.p0i8(i64 noundef 48, i8* noundef nonnull %.sub) #94
  call fastcc void @julia_cascade_770([1 x [4 x [2 x double]]]* noalias nocapture nofree noundef nonnull writeonly sret([1 x [4 x [2 x double]]]) align 8 dereferenceable(64) %11, { i64, [1 x i64] } addrspace(11)* nocapture nofree noundef nonnull readonly align 8 dereferenceable(16) %72, double %pointerref) #94, !dbg !3843
  %106 = shl i64 %102, 6, !dbg !3844
  %coercion88 = getelementptr i8, i8* %arrayptr59, i64 %106, !dbg !3848
  call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 1 dereferenceable(64) %coercion88, i8* noundef nonnull align 8 dereferenceable(64) %73, i64 noundef 64, i1 noundef false) #94, !dbg !3849, !noalias !3688
  %value_phi84 = add i64 %103, 1, !dbg !3750
  %exitcond200 = icmp eq i64 %103, %arraylen35, !dbg !3751
  br i1 %exitcond200, label %L331.preheader.loopexit, label %L281, !dbg !3752

L336:                                             ; preds = %L336.preheader, %L366
  %iv6 = phi i64 [ 0, %L336.preheader ], [ %iv.next7, %L366 ]
  %value_phi92188 = phi i64 [ %111, %L366 ], [ %value_phi49, %L336.preheader ]
  %value_phi91187 = phi i32 [ %113, %L366 ], [ 0, %L336.preheader ]
  %iv.next7 = add nuw nsw i64 %iv6, 1, !dbg !3853
  %107 = call i64 @llvm.cttz.i64(i64 %value_phi92188, i1 noundef true) #94, !dbg !3853, !range !2422
  %108 = trunc i64 %107 to i32, !dbg !3855
  %109 = add nuw nsw i32 %108, 1, !dbg !3856
  %110 = zext i32 %109 to i64, !dbg !3858
  %111 = lshr i64 %value_phi92188, %110, !dbg !3858
  %112 = icmp eq i32 %108, 63, !dbg !3858
  %113 = add i32 %109, %value_phi91187, !dbg !3860
  %114 = load i64, i64* inttoptr (i64 123246368836896 to i64*), align 32, !dbg !3862, !tbaa !77, !alias.scope !81, !noalias !84
  %115 = shl i32 %113, 9, !dbg !3865
  %116 = zext i32 %115 to i64, !dbg !3866
  %117 = inttoptr i64 %114 to i8*, !dbg !3870
  %118 = getelementptr i8, i8* %117, i64 %116, !dbg !3870
  %p.i121 = bitcast i8* %118 to i32*, !dbg !3871
  %v.i122184 = load atomic i32, i32* %p.i121 acquire, align 16, !dbg !3871
  %.not174185 = icmp eq i32 %v.i122184, 0, !dbg !3873
  br i1 %.not174185, label %L356.preheader, label %L366, !dbg !3835

L356.preheader:                                   ; preds = %L336
  br label %L356, !dbg !3874

L356:                                             ; preds = %L356.preheader, %L363
  %iv8 = phi i64 [ 0, %L356.preheader ], [ %iv.next9, %L363 ]
  %119 = trunc i64 %iv8 to i32
  %iv.next9 = add nuw nsw i64 %iv8, 1
  call void @llvm.lifetime.end.p0i8(i64 noundef 48, i8* noundef nonnull %.sub) #94
  call void asm sideeffect "pause", "~{memory}"() #98, !dbg !3875
  %120 = add i32 %119, 1, !dbg !3877
  %121 = icmp ult i32 %120, 65537, !dbg !3878
  br i1 %121, label %L363, label %L360, !dbg !3874

L360:                                             ; preds = %L356
  %122 = call fastcc i8 @julia_checktask_612(i32 zeroext %113) #94, !dbg !3880
  %123 = and i8 %122, 1, !dbg !3880
  %.not175 = icmp eq i8 %123, 0, !dbg !3880
  br i1 %.not175, label %L363, label %L366.loopexit, !dbg !3880

L363:                                             ; preds = %L360, %L356
  %v.i122 = load atomic i32, i32* %p.i121 acquire, align 16, !dbg !3871
  %.not174 = icmp eq i32 %v.i122, 0, !dbg !3873
  br i1 %.not174, label %L356, label %L366.loopexit, !dbg !3835

L366.loopexit:                                    ; preds = %L360, %L363
  br label %L366, !dbg !3832

L366:                                             ; preds = %L366.loopexit, %L336
  %124 = icmp eq i64 %111, 0, !dbg !3832
  %125 = select i1 %112, i1 true, i1 %124, !dbg !3832
  br i1 %125, label %L368.loopexit, label %L336, !dbg !3834

L368.loopexit:                                    ; preds = %L366
  br label %L368, !dbg !3881

L368:                                             ; preds = %L368.loopexit, %L331.preheader
  %v.i118 = atomicrmw or i64* %p.i, i64 %value_phi49 acq_rel, align 8, !dbg !3881
  call void @llvm.julia.gc_preserve_end(token %70) #94, !dbg !3884
  br label %L392, !dbg !3884

L383.lr.ph:                                       ; preds = %L142, %L84, %L76
  %126 = addrspacecast {} addrspace(10)* %2 to double addrspace(13)* addrspace(11)*
  %127 = addrspacecast {} addrspace(10)* %0 to [1 x [4 x [2 x double]]] addrspace(13)* addrspace(11)*
  %128 = bitcast [1 x [4 x [2 x double]]]* %6 to i8*
  %umax = call i64 @llvm.umax.i64(i64 %arraylen35, i64 noundef 1) #94, !dbg !3885
  br label %L383, !dbg !3885

L383:                                             ; preds = %L383, %L383.lr.ph
  %iv10 = phi i64 [ %iv.next11, %L383 ], [ 0, %L383.lr.ph ]
  %iv.next11 = add nuw nsw i64 %iv10, 1, !dbg !3887
  %129 = add nsw i64 %iv.next11, -1, !dbg !3887
  %arrayptr44170 = load double addrspace(13)*, double addrspace(13)* addrspace(11)* %126, align 16, !dbg !3887, !tbaa !245, !alias.scope !3890, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BFloat64\7D !0
  %130 = getelementptr inbounds double, double addrspace(13)* %arrayptr44170, i64 %129, !dbg !3887
  %arrayref45 = load double, double addrspace(13)* %130, align 8, !dbg !3887, !tbaa !325, !alias.scope !81, !noalias !84
  call fastcc void @julia_cascade_787([1 x [4 x [2 x double]]]* noalias nocapture nofree noundef nonnull writeonly sret([1 x [4 x [2 x double]]]) align 8 dereferenceable(64) %6, {} addrspace(10)* noundef nonnull align 16 dereferenceable(40) %1, double %arrayref45) #94, !dbg !3891
  %arrayptr47171 = load [1 x [4 x [2 x double]]] addrspace(13)*, [1 x [4 x [2 x double]]] addrspace(13)* addrspace(11)* %127, align 16, !dbg !3892, !tbaa !245, !alias.scope !3890, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BSMatrix\7B2\2C\202\2C\20ComplexF64\2C\204\7D\7D !0
  %131 = getelementptr inbounds [1 x [4 x [2 x double]]], [1 x [4 x [2 x double]]] addrspace(13)* %arrayptr47171, i64 %129, i64 0, !dbg !3892
  %132 = bitcast [4 x [2 x double]] addrspace(13)* %131 to i8 addrspace(13)*, !dbg !3892
  call void @llvm.memcpy.p13i8.p0i8.i64(i8 addrspace(13)* noundef align 8 dereferenceable(64) %132, i8* noundef nonnull align 8 dereferenceable(64) %128, i64 noundef 64, i1 noundef false) #94, !dbg !3892, !tbaa !462, !alias.scope !2556, !noalias !3748
  %133 = add nuw nsw i64 %iv.next11, 1, !dbg !3893
  %exitcond.not = icmp eq i64 %iv.next11, %umax, !dbg !3896
  br i1 %exitcond.not, label %L392.loopexit1, label %L383, !dbg !3885

L392.loopexit:                                    ; preds = %idxend16
  br label %L392

L392.loopexit1:                                   ; preds = %L383
  br label %L392

L392:                                             ; preds = %L392.loopexit1, %L392.loopexit, %L368, %L72, %L25
  call void @llvm.lifetime.end.p0i8(i64 noundef 48, i8* noundef nonnull %.sub) #94
  ret void, !dbg !3602

oob:                                              ; preds = %L34
  %errorbox = alloca i64, align 8, !dbg !3619
  store i64 %iv.next13, i64* %errorbox, align 8, !dbg !3619, !noalias !3688
  %134 = addrspacecast {} addrspace(10)* %2 to {} addrspace(12)*, !dbg !3619
  call void @ijl_bounds_error_ints({} addrspace(12)* noundef %134, i64* noundef nonnull align 8 %errorbox, i64 noundef 1) #99, !dbg !3619
  unreachable, !dbg !3619

idxend:                                           ; preds = %L34
  %arrayptr162 = load double addrspace(13)*, double addrspace(13)* addrspace(11)* %19, align 16, !dbg !3619, !tbaa !245, !alias.scope !3890, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BFloat64\7D !0
  %135 = getelementptr inbounds double, double addrspace(13)* %arrayptr162, i64 %22, !dbg !3619
  %arrayref = load double, double addrspace(13)* %135, align 8, !dbg !3619, !tbaa !325, !alias.scope !81, !noalias !84
  call fastcc void @julia_cascade_787([1 x [4 x [2 x double]]]* noalias nocapture nofree noundef nonnull writeonly sret([1 x [4 x [2 x double]]]) align 8 dereferenceable(64) %9, {} addrspace(10)* noundef nonnull align 16 dereferenceable(40) %1, double %arrayref) #94, !dbg !3897
  %arraylen12 = load i64, i64 addrspace(11)* %arraylen_ptr2, align 8, !dbg !3898, !tbaa !179, !range !182, !alias.scope !183, !noalias !184, !enzyme_type !271, !enzyme_inactive !0, !enzymejl_source_type_UInt64 !0, !enzymejl_byref_BITS_VALUE !0
  %inbounds13 = icmp ult i64 %22, %arraylen12, !dbg !3898
  br i1 %inbounds13, label %idxend16, label %oob14, !dbg !3898

oob14:                                            ; preds = %idxend
  %errorbox15 = alloca i64, align 8, !dbg !3898
  store i64 %iv.next13, i64* %errorbox15, align 8, !dbg !3898, !noalias !3688
  %136 = addrspacecast {} addrspace(10)* %0 to {} addrspace(12)*, !dbg !3898
  call void @ijl_bounds_error_ints({} addrspace(12)* noundef %136, i64* noundef nonnull align 8 %errorbox15, i64 noundef 1) #99, !dbg !3898
  unreachable, !dbg !3898

idxend16:                                         ; preds = %idxend
  %arrayptr18163 = load [1 x [4 x [2 x double]]] addrspace(13)*, [1 x [4 x [2 x double]]] addrspace(13)* addrspace(11)* %20, align 16, !dbg !3898, !tbaa !245, !alias.scope !3890, !noalias !184, !nonnull !0, !enzyme_type !329, !enzymejl_byref_BITS_VALUE !0, !enzymejl_source_type_Ptr\7BSMatrix\7B2\2C\202\2C\20ComplexF64\2C\204\7D\7D !0
  %137 = getelementptr inbounds [1 x [4 x [2 x double]]], [1 x [4 x [2 x double]]] addrspace(13)* %arrayptr18163, i64 %22, i64 0, !dbg !3898
  %138 = bitcast [4 x [2 x double]] addrspace(13)* %137 to i8 addrspace(13)*, !dbg !3898
  call void @llvm.memcpy.p13i8.p0i8.i64(i8 addrspace(13)* noundef align 8 dereferenceable(64) %138, i8* noundef nonnull align 8 dereferenceable(64) %21, i64 noundef 64, i1 noundef false) #94, !dbg !3898, !tbaa !462, !alias.scope !2556, !noalias !3748
  %.not164 = icmp eq i64 %iv.next13, %arraylen, !dbg !3899
  %139 = add nuw nsw i64 %iv.next13, 1, !dbg !3669
  br i1 %.not164, label %L392.loopexit, label %L34, !dbg !3602
}

  %v.i = atomicrmw xchg i64* %p.i, i64 0 acq_rel, align 8, !dbg !193
 Active atomic inst not yet handled

Stacktrace:
 [1] _atomic_xchg!
   @ ~/.local/share/julia/packages/ThreadingUtilities/3z3g0/src/atomics.jl:33
 [2] _exchange_mask!
   @ ~/.local/share/julia/packages/PolyesterWeave/E9Wdf/src/request.jl:67
 [3] __request_threads
   @ ~/.local/share/julia/packages/PolyesterWeave/E9Wdf/src/request.jl:89
 [4] _request_threads
   @ ~/.local/share/julia/packages/PolyesterWeave/E9Wdf/src/request.jl:61
 [5] request_threads
   @ ~/.local/share/julia/packages/PolyesterWeave/E9Wdf/src/request.jl:121
 [6] request_threads
   @ ~/.local/share/julia/packages/PolyesterWeave/E9Wdf/src/request.jl:128
 [7] batch
   @ ~/.local/share/julia/packages/Polyester/eqrC9/src/batch.jl:308
 [8] macro expansion
   @ ~/.local/share/julia/packages/Polyester/eqrC9/src/closure.jl:456
 [9] cascade!
   @ ~/Dropbox/Projects/Julia/wblna_optim_code_for_paper/src/enzyme_mwe.jl:19


Stacktrace:
  [1] count_ones
    @ ./int.jl:415 [inlined]
  [2] __request_threads
    @ ~/.local/share/julia/packages/PolyesterWeave/E9Wdf/src/request.jl:90 [inlined]
  [3] _request_threads
    @ ~/.local/share/julia/packages/PolyesterWeave/E9Wdf/src/request.jl:61 [inlined]
  [4] request_threads
    @ ~/.local/share/julia/packages/PolyesterWeave/E9Wdf/src/request.jl:121 [inlined]
  [5] request_threads
    @ ~/.local/share/julia/packages/PolyesterWeave/E9Wdf/src/request.jl:128 [inlined]
  [6] batch
    @ ~/.local/share/julia/packages/Polyester/eqrC9/src/batch.jl:308 [inlined]
  [7] macro expansion
    @ ~/.local/share/julia/packages/Polyester/eqrC9/src/closure.jl:456 [inlined]
  [8] cascade!
    @ ~/Dropbox/Projects/Julia/wblna_optim_code_for_paper/src/enzyme_mwe.jl:19
  [9] objective!
    @ ~/Dropbox/Projects/Julia/wblna_optim_code_for_paper/src/enzyme_mwe.jl:27 [inlined]
 [10] diffejulia_objective__578wrap
    @ ~/Dropbox/Projects/Julia/wblna_optim_code_for_paper/src/enzyme_mwe.jl:0
 [11] macro expansion
    @ ~/.local/share/julia/packages/Enzyme/ydGh2/src/compiler.jl:5218 [inlined]
 [12] enzyme_call
    @ ~/.local/share/julia/packages/Enzyme/ydGh2/src/compiler.jl:4764 [inlined]
 [13] CombinedAdjointThunk
    @ ~/.local/share/julia/packages/Enzyme/ydGh2/src/compiler.jl:4636 [inlined]
 [14] autodiff
    @ ~/.local/share/julia/packages/Enzyme/ydGh2/src/Enzyme.jl:503 [inlined]
 [15] autodiff
    @ ~/.local/share/julia/packages/Enzyme/ydGh2/src/Enzyme.jl:524 [inlined]
 [16] d_objective!(dzs::Vector{Float64}, zs::Vector{Float64}, A::Vector{SMatrix{2, 2, ComplexF64, 4}}, shadow_A::Vector{SMatrix{2, 2, ComplexF64, 4}}, L::Float64, freqs::Vector{Float64})
    @ Main ~/Dropbox/Projects/Julia/wblna_optim_code_for_paper/src/enzyme_mwe.jl:35
 [17] var"##core#233"()
    @ Main ~/.local/share/julia/packages/BenchmarkTools/QNsku/src/execution.jl:561
 [18] var"##sample#234"(::Tuple{}, __params::BenchmarkTools.Parameters)
    @ Main ~/.local/share/julia/packages/BenchmarkTools/QNsku/src/execution.jl:570
 [19] _lineartrial(b::BenchmarkTools.Benchmark, p::BenchmarkTools.Parameters; maxevals::Int64, kwargs::@Kwargs{})
    @ BenchmarkTools ~/.local/share/julia/packages/BenchmarkTools/QNsku/src/execution.jl:187
 [20] _lineartrial(b::BenchmarkTools.Benchmark, p::BenchmarkTools.Parameters)
    @ BenchmarkTools ~/.local/share/julia/packages/BenchmarkTools/QNsku/src/execution.jl:182
 [21] #invokelatest#2
    @ ./essentials.jl:892 [inlined]
 [22] invokelatest
    @ ./essentials.jl:889 [inlined]
 [23] #lineartrial#46
    @ ~/.local/share/julia/packages/BenchmarkTools/QNsku/src/execution.jl:51 [inlined]
 [24] lineartrial
    @ ~/.local/share/julia/packages/BenchmarkTools/QNsku/src/execution.jl:50 [inlined]
 [25] tune!(b::BenchmarkTools.Benchmark, p::BenchmarkTools.Parameters; progressid::Nothing, nleaves::Float64, ndone::Float64, verbose::Bool, pad::String, kwargs::@Kwargs{})
    @ BenchmarkTools ~/.local/share/julia/packages/BenchmarkTools/QNsku/src/execution.jl:300
 [26] tune!
    @ ~/.local/share/julia/packages/BenchmarkTools/QNsku/src/execution.jl:289 [inlined]
 [27] tune!(b::BenchmarkTools.Benchmark)
    @ BenchmarkTools ~/.local/share/julia/packages/BenchmarkTools/QNsku/src/execution.jl:289
 [28] top-level scope
    @ ~/.local/share/julia/packages/BenchmarkTools/QNsku/src/execution.jl:447
in expression starting at /home/kiran/Dropbox/Projects/Julia/wblna_optim_code_for_paper/src/enzyme_mwe.jl:49

[wsmoses]

The first error we definitely should fix to be nicer, but re the 1.10 error, I think that's equivalent to #2208

[wsmoses]

I believe this should now be resolved on main by the jll bump, please reopen if not!