[PATCH] D126277: [Debug] [Coroutines] Add deref operator for non complex expression

[Chuanqi Xu via Phabricator via llvm-commits]

ChuanqiXu added a comment.

In D126277#3540094 <https://reviews.llvm.org/D126277#3540094>, @jmorse wrote:

> what scenarios is salvageDebugInfo called in? I'm trying to reason about when the isComplex test might be necessary.

Background: compiler would construct new structs called coroutine frames for the coroutines. It is necessary since a coroutine would be split into multiple functions in the middle end so we need a state to record the information. So the following one:

  coro f(int a) {
     a++;
     co_await something();
     a++;
     co_return a;
  }

would be translated to:

  coro f(int a) {
      frame_ty *frame = malloc(size);
      store `a` to someplace of frame
      %a.tmp = load `a` from frame
      %a.tmp += 1
      store %a.tmp to frame
      ...
  }

then it would be split into:

  coro f(int a) {
      frame_ty *frame = malloc(size);
      store `a` to someplace of frame
      return frame
  }
  
  void f.resume(frame_ty* frame) {
      %a.tmp = load `a` from frame
      %a.tmp += 1
      store %a.tmp to frame
  
  unreachable: // which would be removed finally
      frame_ty *frame = malloc(size);
      store `a` to someplace of frame
      ...
  }

then if we add dbg intrinsics, it would look like:

  coro f(int a) {
      frame_ty *frame = malloc(size);
      call @dbg.declare(frame, "frame_var", DIExpression())
      store `a` to someplace of frame
      call @dbg.declare(the address of `a` in frame, "a", DIExpression())
      return frame
  }
  
  void f.resume(frame_ty* frame) {
      %a.tmp = load `a` from frame
      %a.tmp += 1
      store %a.tmp to frame
  
  unreachable:
      frame_ty *d_frame = malloc(size);
      call @dbg.declare(d_frame, "frame_var", DIExpression())
      store `a` to someplace of d_frame
      call @dbg.declare(the address of `a` in d_frame, "a", DIExpression())
      ...
  }

Then we call `salvageDebugInfo ` here to salvage the debug information from unreachable bb to the entry of the `f.resume` function. Then the main work it tries to do is to replace `d_frame ` to the argument. And I remember the reason why we would create an additional alloca is that the argument couldn't be the address argument of @dbg.declares


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D126277/new/

https://reviews.llvm.org/D126277