SROA pass breaks this code for some reason

Beginners
1

(Complete newbie here!) The following code has been generated by a toy BF compiler that passes all tests in LLVMJIT and compilation when no optimizations have been taken. It should print the character ‘0’, compiled with
llc out.ll -o out.asm; clang out.asm -o out; ./out

; ModuleID = 'out.ll'
source_filename = "stdin"

define void @program(i8* %0) {
entry:
  %index = alloca i32, align 4
  %index.0 = load i32, i32* %index, align 4
  %incremented = add i32 %index.0, 1
  store i32 %incremented, i32* %index, align 4
  %index.01 = load i32, i32* %index, align 4
  %sep = getelementptr inbounds i8, i8* %0, i32 %index.01
  %1 = load i8, i8* %sep, align 1
  %2 = add i8 %1, 1
  store i8 %2, i8* %sep, align 1
  %index.02 = load i32, i32* %index, align 4
  %sep3 = getelementptr inbounds i8, i8* %0, i32 %index.02
  %3 = load i8, i8* %sep3, align 1
  %4 = add i8 %3, 1
  store i8 %4, i8* %sep3, align 1
  %index.04 = load i32, i32* %index, align 4
  %sep5 = getelementptr inbounds i8, i8* %0, i32 %index.04
  %5 = load i8, i8* %sep5, align 1
  %6 = add i8 %5, 1
  store i8 %6, i8* %sep5, align 1
  %index.06 = load i32, i32* %index, align 4
  %sep7 = getelementptr inbounds i8, i8* %0, i32 %index.06
  %7 = load i8, i8* %sep7, align 1
  %8 = add i8 %7, 1
  store i8 %8, i8* %sep7, align 1
  %index.08 = load i32, i32* %index, align 4
  %sep9 = getelementptr inbounds i8, i8* %0, i32 %index.08
  %9 = load i8, i8* %sep9, align 1
  %10 = add i8 %9, 1
  store i8 %10, i8* %sep9, align 1
  %index.010 = load i32, i32* %index, align 4
  %sep11 = getelementptr inbounds i8, i8* %0, i32 %index.010
  %11 = load i8, i8* %sep11, align 1
  %12 = add i8 %11, 1
  store i8 %12, i8* %sep11, align 1
  br label %cond0

cond0:                                            ; preds = %block0, %entry
  %offset = load i32, i32* %index, align 4
  %ptr = getelementptr inbounds i8, i8* %0, i32 %offset
  %cell = load i8, i8* %ptr, align 1
  %13 = icmp eq i8 %cell, 0
  br i1 %13, label %after0, label %block0

block0:                                           ; preds = %cond0
  %index.012 = load i32, i32* %index, align 4
  %decremented = sub i32 %index.012, 1
  store i32 %decremented, i32* %index, align 4
  %index.013 = load i32, i32* %index, align 4
  %sep14 = getelementptr inbounds i8, i8* %0, i32 %index.013
  %14 = load i8, i8* %sep14, align 1
  %15 = add i8 %14, 1
  store i8 %15, i8* %sep14, align 1
  %index.015 = load i32, i32* %index, align 4
  %sep16 = getelementptr inbounds i8, i8* %0, i32 %index.015
  %16 = load i8, i8* %sep16, align 1
  %17 = add i8 %16, 1
  store i8 %17, i8* %sep16, align 1
  %index.017 = load i32, i32* %index, align 4
  %sep18 = getelementptr inbounds i8, i8* %0, i32 %index.017
  %18 = load i8, i8* %sep18, align 1
  %19 = add i8 %18, 1
  store i8 %19, i8* %sep18, align 1
  %index.019 = load i32, i32* %index, align 4
  %sep20 = getelementptr inbounds i8, i8* %0, i32 %index.019
  %20 = load i8, i8* %sep20, align 1
  %21 = add i8 %20, 1
  store i8 %21, i8* %sep20, align 1
  %index.021 = load i32, i32* %index, align 4
  %sep22 = getelementptr inbounds i8, i8* %0, i32 %index.021
  %22 = load i8, i8* %sep22, align 1
  %23 = add i8 %22, 1
  store i8 %23, i8* %sep22, align 1
  %index.023 = load i32, i32* %index, align 4
  %sep24 = getelementptr inbounds i8, i8* %0, i32 %index.023
  %24 = load i8, i8* %sep24, align 1
  %25 = add i8 %24, 1
  store i8 %25, i8* %sep24, align 1
  %index.025 = load i32, i32* %index, align 4
  %sep26 = getelementptr inbounds i8, i8* %0, i32 %index.025
  %26 = load i8, i8* %sep26, align 1
  %27 = add i8 %26, 1
  store i8 %27, i8* %sep26, align 1
  %index.027 = load i32, i32* %index, align 4
  %sep28 = getelementptr inbounds i8, i8* %0, i32 %index.027
  %28 = load i8, i8* %sep28, align 1
  %29 = add i8 %28, 1
  store i8 %29, i8* %sep28, align 1
  %index.029 = load i32, i32* %index, align 4
  %incremented30 = add i32 %index.029, 1
  store i32 %incremented30, i32* %index, align 4
  %index.031 = load i32, i32* %index, align 4
  %sep32 = getelementptr inbounds i8, i8* %0, i32 %index.031
  %cell33 = load i8, i8* %sep32, align 1
  %dec_cell = sub i8 %cell33, 1
  store i8 %dec_cell, i8* %sep32, align 1
  br label %cond0

after0:                                           ; preds = %cond0
  %index.034 = load i32, i32* %index, align 4
  %decremented35 = sub i32 %index.034, 1
  store i32 %decremented35, i32* %index, align 4
  %index.036 = load i32, i32* %index, align 4
  %sep37 = getelementptr inbounds i8, i8* %0, i32 %index.036
  %cell38 = load i8, i8* %sep37, align 1
  %cast = sext i8 %cell38 to i32
  %print = call i32 @putchar(i32 %cast)
  ret void
}

; Function Attrs: nofree nounwind
declare noundef i32 @putchar(i32 noundef) #0

; Function Attrs: nofree nounwind
declare noundef i32 @getchar() #0

define i32 @main() {
entry:
  %0 = trunc i64 100000 to i32
  %mallocsize = mul i32 %0, ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i32)
  %stack = tail call i8* @malloc(i32 %mallocsize)
  call void @llvm.memset.p0i8.i64(i8* align 1 %stack, i8 0, i64 100000, i1 false)
  call void @program(i8* %stack)
  ret i32 0
}

; Function Attrs: inaccessiblememonly mustprogress nofree nounwind willreturn
declare noalias noundef i8* @malloc(i32 noundef) #1

; Function Attrs: argmemonly mustprogress nofree nounwind willreturn writeonly
declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1 immarg) #2

attributes #0 = { nofree nounwind }
attributes #1 = { inaccessiblememonly mustprogress nofree nounwind willreturn }
attributes #2 = { argmemonly mustprogress nofree nounwind willreturn writeonly }

I then tried to optimize with
opt out.ll --O1 -S -o opt.ll (and eventually -opt-bisect-limit=N)
But the function program breaks from SROAPass. The last valid bisection (5) for me is the following:

; ModuleID = 'out.ll'
source_filename = "stdin"

define void @program(i8* %0) {
entry:
  %index = alloca i32, align 4
  %index.0 = load i32, i32* %index, align 4
  %incremented = add i32 %index.0, 1
  store i32 %incremented, i32* %index, align 4
  %index.01 = load i32, i32* %index, align 4
  %sep = getelementptr inbounds i8, i8* %0, i32 %index.01
  %1 = load i8, i8* %sep, align 1
  %2 = add i8 %1, 1
  store i8 %2, i8* %sep, align 1
  %index.02 = load i32, i32* %index, align 4
  %sep3 = getelementptr inbounds i8, i8* %0, i32 %index.02
  %3 = load i8, i8* %sep3, align 1
  %4 = add i8 %3, 1
  store i8 %4, i8* %sep3, align 1
  %index.04 = load i32, i32* %index, align 4
  %sep5 = getelementptr inbounds i8, i8* %0, i32 %index.04
  %5 = load i8, i8* %sep5, align 1
  %6 = add i8 %5, 1
  store i8 %6, i8* %sep5, align 1
  %index.06 = load i32, i32* %index, align 4
  %sep7 = getelementptr inbounds i8, i8* %0, i32 %index.06
  %7 = load i8, i8* %sep7, align 1
  %8 = add i8 %7, 1
  store i8 %8, i8* %sep7, align 1
  %index.08 = load i32, i32* %index, align 4
  %sep9 = getelementptr inbounds i8, i8* %0, i32 %index.08
  %9 = load i8, i8* %sep9, align 1
  %10 = add i8 %9, 1
  store i8 %10, i8* %sep9, align 1
  %index.010 = load i32, i32* %index, align 4
  %sep11 = getelementptr inbounds i8, i8* %0, i32 %index.010
  %11 = load i8, i8* %sep11, align 1
  %12 = add i8 %11, 1
  store i8 %12, i8* %sep11, align 1
  br label %cond0

cond0:                                            ; preds = %block0, %entry
  %offset = load i32, i32* %index, align 4
  %ptr = getelementptr inbounds i8, i8* %0, i32 %offset
  %cell = load i8, i8* %ptr, align 1
  %13 = icmp eq i8 %cell, 0
  br i1 %13, label %after0, label %block0

block0:                                           ; preds = %cond0
  %index.012 = load i32, i32* %index, align 4
  %decremented = sub i32 %index.012, 1
  store i32 %decremented, i32* %index, align 4
  %index.013 = load i32, i32* %index, align 4
  %sep14 = getelementptr inbounds i8, i8* %0, i32 %index.013
  %14 = load i8, i8* %sep14, align 1
  %15 = add i8 %14, 1
  store i8 %15, i8* %sep14, align 1
  %index.015 = load i32, i32* %index, align 4
  %sep16 = getelementptr inbounds i8, i8* %0, i32 %index.015
  %16 = load i8, i8* %sep16, align 1
  %17 = add i8 %16, 1
  store i8 %17, i8* %sep16, align 1
  %index.017 = load i32, i32* %index, align 4
  %sep18 = getelementptr inbounds i8, i8* %0, i32 %index.017
  %18 = load i8, i8* %sep18, align 1
  %19 = add i8 %18, 1
  store i8 %19, i8* %sep18, align 1
  %index.019 = load i32, i32* %index, align 4
  %sep20 = getelementptr inbounds i8, i8* %0, i32 %index.019
  %20 = load i8, i8* %sep20, align 1
  %21 = add i8 %20, 1
  store i8 %21, i8* %sep20, align 1
  %index.021 = load i32, i32* %index, align 4
  %sep22 = getelementptr inbounds i8, i8* %0, i32 %index.021
  %22 = load i8, i8* %sep22, align 1
  %23 = add i8 %22, 1
  store i8 %23, i8* %sep22, align 1
  %index.023 = load i32, i32* %index, align 4
  %sep24 = getelementptr inbounds i8, i8* %0, i32 %index.023
  %24 = load i8, i8* %sep24, align 1
  %25 = add i8 %24, 1
  store i8 %25, i8* %sep24, align 1
  %index.025 = load i32, i32* %index, align 4
  %sep26 = getelementptr inbounds i8, i8* %0, i32 %index.025
  %26 = load i8, i8* %sep26, align 1
  %27 = add i8 %26, 1
  store i8 %27, i8* %sep26, align 1
  %index.027 = load i32, i32* %index, align 4
  %sep28 = getelementptr inbounds i8, i8* %0, i32 %index.027
  %28 = load i8, i8* %sep28, align 1
  %29 = add i8 %28, 1
  store i8 %29, i8* %sep28, align 1
  %index.029 = load i32, i32* %index, align 4
  %incremented30 = add i32 %index.029, 1
  store i32 %incremented30, i32* %index, align 4
  %index.031 = load i32, i32* %index, align 4
  %sep32 = getelementptr inbounds i8, i8* %0, i32 %index.031
  %cell33 = load i8, i8* %sep32, align 1
  %dec_cell = sub i8 %cell33, 1
  store i8 %dec_cell, i8* %sep32, align 1
  br label %cond0

after0:                                           ; preds = %cond0
  %index.034 = load i32, i32* %index, align 4
  %decremented35 = sub i32 %index.034, 1
  store i32 %decremented35, i32* %index, align 4
  %index.036 = load i32, i32* %index, align 4
  %sep37 = getelementptr inbounds i8, i8* %0, i32 %index.036
  %cell38 = load i8, i8* %sep37, align 1
  %cast = sext i8 %cell38 to i32
  %print = call i32 @putchar(i32 %cast)
  ret void
}

; Function Attrs: nofree nounwind
declare noundef i32 @putchar(i32 noundef) #0

; Function Attrs: nofree nounwind
declare noundef i32 @getchar() #0

define i32 @main() {
entry:
  %0 = trunc i64 100000 to i32
  %mallocsize = mul i32 %0, ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i32)
  %stack = tail call i8* @malloc(i32 %mallocsize)
  call void @llvm.memset.p0i8.i64(i8* align 1 %stack, i8 0, i64 100000, i1 false)
  call void @program(i8* %stack)
  ret i32 0
}

; Function Attrs: inaccessiblememonly mustprogress nofree nounwind willreturn
declare noalias noundef i8* @malloc(i32 noundef) #1

; Function Attrs: argmemonly mustprogress nofree nounwind willreturn writeonly
declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1 immarg) #2

attributes #0 = { nofree nounwind }
attributes #1 = { inaccessiblememonly mustprogress nofree nounwind willreturn }
attributes #2 = { argmemonly mustprogress nofree nounwind willreturn writeonly }

after which SROA (6) removes the alloca generating this module:

; ModuleID = 'out.ll'
source_filename = "stdin"

define void @program(i8* %0) {
entry:
  %incremented = add i32 undef, 1
  %sep = getelementptr inbounds i8, i8* %0, i32 %incremented
  %1 = load i8, i8* %sep, align 1
  %2 = add i8 %1, 1
  store i8 %2, i8* %sep, align 1
  %sep3 = getelementptr inbounds i8, i8* %0, i32 %incremented
  %3 = load i8, i8* %sep3, align 1
  %4 = add i8 %3, 1
  store i8 %4, i8* %sep3, align 1
  %sep5 = getelementptr inbounds i8, i8* %0, i32 %incremented
  %5 = load i8, i8* %sep5, align 1
  %6 = add i8 %5, 1
  store i8 %6, i8* %sep5, align 1
  %sep7 = getelementptr inbounds i8, i8* %0, i32 %incremented
  %7 = load i8, i8* %sep7, align 1
  %8 = add i8 %7, 1
  store i8 %8, i8* %sep7, align 1
  %sep9 = getelementptr inbounds i8, i8* %0, i32 %incremented
  %9 = load i8, i8* %sep9, align 1
  %10 = add i8 %9, 1
  store i8 %10, i8* %sep9, align 1
  %sep11 = getelementptr inbounds i8, i8* %0, i32 %incremented
  %11 = load i8, i8* %sep11, align 1
  %12 = add i8 %11, 1
  store i8 %12, i8* %sep11, align 1
  br label %cond0

cond0:                                            ; preds = %block0, %entry
  %index.0 = phi i32 [ %incremented, %entry ], [ %incremented30, %block0 ]
  %ptr = getelementptr inbounds i8, i8* %0, i32 %index.0
  %cell = load i8, i8* %ptr, align 1
  %13 = icmp eq i8 %cell, 0
  br i1 %13, label %after0, label %block0

block0:                                           ; preds = %cond0
  %decremented = sub i32 %index.0, 1
  %sep14 = getelementptr inbounds i8, i8* %0, i32 %decremented
  %14 = load i8, i8* %sep14, align 1
  %15 = add i8 %14, 1
  store i8 %15, i8* %sep14, align 1
  %sep16 = getelementptr inbounds i8, i8* %0, i32 %decremented
  %16 = load i8, i8* %sep16, align 1
  %17 = add i8 %16, 1
  store i8 %17, i8* %sep16, align 1
  %sep18 = getelementptr inbounds i8, i8* %0, i32 %decremented
  %18 = load i8, i8* %sep18, align 1
  %19 = add i8 %18, 1
  store i8 %19, i8* %sep18, align 1
  %sep20 = getelementptr inbounds i8, i8* %0, i32 %decremented
  %20 = load i8, i8* %sep20, align 1
  %21 = add i8 %20, 1
  store i8 %21, i8* %sep20, align 1
  %sep22 = getelementptr inbounds i8, i8* %0, i32 %decremented
  %22 = load i8, i8* %sep22, align 1
  %23 = add i8 %22, 1
  store i8 %23, i8* %sep22, align 1
  %sep24 = getelementptr inbounds i8, i8* %0, i32 %decremented
  %24 = load i8, i8* %sep24, align 1
  %25 = add i8 %24, 1
  store i8 %25, i8* %sep24, align 1
  %sep26 = getelementptr inbounds i8, i8* %0, i32 %decremented
  %26 = load i8, i8* %sep26, align 1
  %27 = add i8 %26, 1
  store i8 %27, i8* %sep26, align 1
  %sep28 = getelementptr inbounds i8, i8* %0, i32 %decremented
  %28 = load i8, i8* %sep28, align 1
  %29 = add i8 %28, 1
  store i8 %29, i8* %sep28, align 1
  %incremented30 = add i32 %decremented, 1
  %sep32 = getelementptr inbounds i8, i8* %0, i32 %incremented30
  %cell33 = load i8, i8* %sep32, align 1
  %dec_cell = sub i8 %cell33, 1
  store i8 %dec_cell, i8* %sep32, align 1
  br label %cond0

after0:                                           ; preds = %cond0
  %decremented35 = sub i32 %index.0, 1
  %sep37 = getelementptr inbounds i8, i8* %0, i32 %decremented35
  %cell38 = load i8, i8* %sep37, align 1
  %cast = sext i8 %cell38 to i32
  %print = call i32 @putchar(i32 %cast)
  ret void
}

; Function Attrs: nofree nounwind
declare noundef i32 @putchar(i32 noundef) #0

; Function Attrs: nofree nounwind
declare noundef i32 @getchar() #0

define i32 @main() {
entry:
  %0 = trunc i64 100000 to i32
  %mallocsize = mul i32 %0, ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i32)
  %stack = tail call i8* @malloc(i32 %mallocsize)
  call void @llvm.memset.p0i8.i64(i8* align 1 %stack, i8 0, i64 100000, i1 false)
  call void @program(i8* %stack)
  ret i32 0
}

; Function Attrs: inaccessiblememonly mustprogress nofree nounwind willreturn
declare noalias noundef i8* @malloc(i32 noundef) #1

; Function Attrs: argmemonly mustprogress nofree nounwind willreturn writeonly
declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1 immarg) #2

attributes #0 = { nofree nounwind }
attributes #1 = { inaccessiblememonly mustprogress nofree nounwind willreturn }
attributes #2 = { argmemonly mustprogress nofree nounwind willreturn writeonly }

which segfaults when compiled.
Successive “optimizations” then reduce the whole program to this nonsense:

; ModuleID = 'out.ll'
source_filename = "stdin"

; Function Attrs: nofree nounwind
define void @program(i8* nocapture %0) local_unnamed_addr #0 {
entry:
  store i8 0, i8* %0, align 1
  %print = call i32 @putchar(i32 0)
  ret void
}

; Function Attrs: nofree nounwind
declare noundef i32 @putchar(i32 noundef) local_unnamed_addr #0

; Function Attrs: nofree nounwind
define i32 @main() local_unnamed_addr #0 {
entry:
  %print.i = call i32 @putchar(i32 0) #1
  ret i32 0
}

attributes #0 = { nofree nounwind }
attributes #1 = { nounwind }

Could somebody explain to me what I did wrong here?
(EDIT: Apropos I’m using Rust + Inkwell to generate the code. The base algorithm should be correct in theory, as even large Brain**** programs (such as rendering a Mandelbrot set) work, so it shouldn’t be a logic error. Ergo I’m assuming I violated some SSA invariant?)

2

You are missing a store to initialize the %index alloca. As such, the loads just result in an undef value.

3

Are you serious? It works now… There went my last five hours… Thank you though.

I will go cry now.