[PATCH] D96922: [Coroutine] Check indirect uses of alloca when checking lifetime info

[Chuanqi Xu via Phabricator via llvm-commits]

ChuanqiXu added a comment.

In D96922#2586732 <https://reviews.llvm.org/D96922#2586732>, @lxfind wrote:

>> Well, even if we haven't seen any new bugs, a more robust solution would be welcome; it's hard to feel confident in what's happening there.
>
> I agree. Although I am leaning towards a different approach than specializing allocations. I am thinking about introducing another ramp function, whose sole job is to allocate the frame, copy the parameters, and then just to the real ramp function. This will prevent any random instructions from getting moved before coro.begin, because now we have a function boundary.

This idea looks to emit two function (the new ramp and the original one with whose argument is FramePtr) in the frontend. If it is the way, it could achieve the goal I try to discuss in llvm-dev before: Normalize Ramp Function <https://lists.llvm.org/pipermail/llvm-dev/2021-February/148733.html>. So I think it would be a good idea too.

In D96922#2586758 <https://reviews.llvm.org/D96922#2586758>, @rjmccall wrote:

> 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D96922/new/

https://reviews.llvm.org/D96922