0.foo.c

struct foo {
    long long x;
};

struct foo foo_factory (void);
struct foo div (struct foo, struct foo);

long long bar (void) {
    return div(div(foo_factory(), foo_factory()), div(foo_factory(), foo_factory())).x;
}

1.foo.ll

; current codegen (well, llvm-14)
%struct.foo = type { i64 }

define dso_local i64 @bar() local_unnamed_addr #0 {
  %1 = alloca %struct.foo, align 8
  %2 = alloca %struct.foo, align 8
  %3 = alloca %struct.foo, align 8
  %4 = alloca %struct.foo, align 8
  %5 = alloca %struct.foo, align 8
  %6 = alloca %struct.foo, align 8
  %7 = alloca %struct.foo, align 8
  call void @foo_factory(ptr nonnull sret(%struct.foo) align 8 %2) #2
  call void @foo_factory(ptr nonnull sret(%struct.foo) align 8 %3) #2
  %8 = load i64, ptr %2, align 8
  %9 = insertvalue [1 x i64] poison, i64 %8, 0
  %10 = load i64, ptr %3, align 8
  %11 = insertvalue [1 x i64] poison, i64 %10, 0
  call void @div(ptr nonnull sret(%struct.foo) align 8 %1, [1 x i64] %9, [1 x i64] %11) #2
  call void @foo_factory(ptr nonnull sret(%struct.foo) align 8 %5) #2
  call void @foo_factory(ptr nonnull sret(%struct.foo) align 8 %6) #2
  %12 = load i64, ptr %5, align 8
  %13 = insertvalue [1 x i64] poison, i64 %12, 0
  %14 = load i64, ptr %6, align 8
  %15 = insertvalue [1 x i64] poison, i64 %14, 0
  call void @div(ptr nonnull sret(%struct.foo) align 8 %4, [1 x i64] %13, [1 x i64] %15) #2
  %16 = load i64, ptr %1, align 8
  %17 = insertvalue [1 x i64] poison, i64 %16, 0
  %18 = load i64, ptr %4, align 8
  %19 = insertvalue [1 x i64] poison, i64 %18, 0
  call void @div(ptr nonnull sret(%struct.foo) align 8 %7, [1 x i64] %17, [1 x i64] %19) #2
  %20 = load i64, ptr %7, align 8
  ret i64 %20
}

declare void @div(ptr sret(%struct.foo) align 8, [1 x i64], [1 x i64]) local_unnamed_addr #1

declare void @foo_factory(ptr sret(%struct.foo) align 8) local_unnamed_addr #1

attributes #0 = { nounwind "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="arm7tdmi" "target-features"="+armv4t,+strict-align,-aes,-bf16,-d32,-dotprod,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-mve.fp,-neon,-sha2,-thumb-mode,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
attributes #1 = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="arm7tdmi" "target-features"="+armv4t,+strict-align,-aes,-bf16,-d32,-dotprod,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-mve.fp,-neon,-sha2,-thumb-mode,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
attributes #2 = { nounwind }

2.foo.ll

; optimal codegen
%struct.foo = type { i64 }

define dso_local i64 @bar() local_unnamed_addr #0 {
  %1 = alloca %struct.foo, align 8
  %2 = alloca %struct.foo, align 8
  %3 = alloca %struct.foo, align 8
  %4 = alloca %struct.foo, align 8
  %5 = alloca %struct.foo, align 8
  %6 = alloca %struct.foo, align 8
  %7 = alloca %struct.foo, align 8
  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %2)
  call void @foo_factory(ptr nonnull sret(%struct.foo) align 8 %2) #2
  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %3)
  call void @foo_factory(ptr nonnull sret(%struct.foo) align 8 %3) #2
  %8 = load i64, ptr %2, align 8
  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %2)
  %9 = insertvalue [1 x i64] poison, i64 %8, 0
  %10 = load i64, ptr %3, align 8
  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %3)
  %11 = insertvalue [1 x i64] poison, i64 %10, 0
  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %1)
  call void @div(ptr nonnull sret(%struct.foo) align 8 %1, [1 x i64] %9, [1 x i64] %11) #2
  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %5)
  call void @foo_factory(ptr nonnull sret(%struct.foo) align 8 %5) #2
  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %6)
  call void @foo_factory(ptr nonnull sret(%struct.foo) align 8 %6) #2
  %12 = load i64, ptr %5, align 8
  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %5)
  %13 = insertvalue [1 x i64] poison, i64 %12, 0
  %14 = load i64, ptr %6, align 8
  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %6)
  %15 = insertvalue [1 x i64] poison, i64 %14, 0
  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %4)
  call void @div(ptr nonnull sret(%struct.foo) align 8 %4, [1 x i64] %13, [1 x i64] %15) #2
  %16 = load i64, ptr %1, align 8
  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %1)
  %17 = insertvalue [1 x i64] poison, i64 %16, 0
  %18 = load i64, ptr %4, align 8
  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %4)
  %19 = insertvalue [1 x i64] poison, i64 %18, 0
  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %7)
  call void @div(ptr nonnull sret(%struct.foo) align 8 %7, [1 x i64] %17, [1 x i64] %19) #2
  %20 = load i64, ptr %7, align 8
  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %7)
  ret i64 %20
}

declare void @div(ptr sret(%struct.foo) align 8, [1 x i64], [1 x i64]) local_unnamed_addr #1

declare void @foo_factory(ptr sret(%struct.foo) align 8) local_unnamed_addr #1

declare void @llvm.lifetime.start.p0(i64 %size, ptr nocapture %ptr)
declare void @llvm.lifetime.end.p0(i64 %size, ptr nocapture %ptr)

attributes #0 = { nounwind "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="arm7tdmi" "target-features"="+armv4t,+strict-align,-aes,-bf16,-d32,-dotprod,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-mve.fp,-neon,-sha2,-thumb-mode,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
attributes #1 = { "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="arm7tdmi" "target-features"="+armv4t,+strict-align,-aes,-bf16,-d32,-dotprod,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-mve.fp,-neon,-sha2,-thumb-mode,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
attributes #2 = { nounwind }