Lifetime markers and coroutine intrinsics

IR & Optimizations
1

I’m using the LLVM coroutine intrinsics in a front-end project and noticed that lifetime markers are only used by the switched-resume lowering to optimize the coroutine frame size (disclaimer: I don’t know much about LLVM coroutines). For instance, I have the following IR:

define void @spilled_alloca_af(ptr %0) {
  %2 = call token @llvm.coro.id.async(i32 32, i32 8, i32 0, ptr @spilled_alloca_afp)
  %3 = call ptr @llvm.coro.begin(token %2, ptr null)
  %4 = getelementptr inbounds i8, ptr %3, i32 -32
  %5 = call ptr @llvm.coro.async.resume()
  %6 = getelementptr inbounds i8, ptr %3, i32 -32
  %7 = getelementptr inbounds { ptr, { ptr, ptr }, i32 }, ptr %6, i32 0, i32 0
  store ptr %5, ptr %7, align 8
  %8 = call { ptr, ptr, ptr } (i32, ptr, ptr, ...) @llvm.coro.suspend.async.sl_p0p0p0s(i32 0, ptr %5, ptr @__kgen_coro_ctxt_proj_fn, ptr @spilled_alloca_af.suspend)
  %9 = alloca i32, align 4
  call void @llvm.lifetime.start.p0(i64 4, ptr %9)
  call void @bar(ptr %9)
  call void @llvm.lifetime.end.p0(i64 4, ptr %9)
  %10 = getelementptr inbounds i8, ptr %3, i32 -32
  %11 = call i1 (ptr, i1, ...) @llvm.coro.end.async(ptr %3, i1 false, ptr @__kgen_coro_end_fn, ptr %10)
  ret void
}

Where the alloca and its lifetime markers are all after the suspend point. Ideally, the first resume function would look like

define internal void @spilled_alloca_af.resume.0(ptr %0) {
entryresume.0:
  %3 = alloca i32, align 4
  call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %3)
  call void @bar(ptr nonnull %3)
  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %3)
  // ...
  ret void
}

But I’m seeing the alloca being put into the coroutine frame instead. In CoroFrame.cpp I see

  // The code that uses lifetime.start intrinsic does not work for functions
  // with loops without exit. Disable it on ABIs we know to generate such
  // code.
  bool ShouldUseLifetimeStartInfo =
      (Shape.ABI != coro::ABI::Async && Shape.ABI != coro::ABI::Retcon &&
       Shape.ABI != coro::ABI::RetconOnce);

My question is: why? Does the Async ABI always generate code like this, and if it can be avoided, can this be made configurable?