[PATCH] D129448: [CodeGen][Asan] Emit lifetime intrinsic for bypassed label

luxufan via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Sat Jul 23 01:03:03 PDT 2022


StephenFan added a comment.

In D129448#3669296 <https://reviews.llvm.org/D129448#3669296>, @vitalybuka wrote:

> I am not sure what to do with this patch. I really prefer to avoid it to minimize the risk of regressions.
>
> We probably considered/prototyped to insert starts like in this solution then in 2016. But at the time of the solution was a comment by @rsmith https://reviews.llvm.org/D24693#559609
> Even if it works for asan, I still don't know how this patch will affect miss-compiles in other places. Dropping intrinsics completely as is must not miscompile.
>
> If the goal to fix asan, then it's extremely rare case, I measured it on llvm codebase and entire Google code. This is not worth of effort to me.
> Also HWAsan and MTE, which can be considered asan successors, does not even bother to handle allocas with multiple lifetime start.

Ok. I'm just fixing this out of interest, and it doesn't look like it's a rigid demand to fix it. If this patch may cause potential regression, I respect your decision.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D129448/new/

https://reviews.llvm.org/D129448



More information about the cfe-commits mailing list