[PATCH] D71435: [WIP] [Attributor] Function level undefined behavior attribute

[Stefanos Baziotis via Phabricator via llvm-commits]

baziotis added a comment.

In D71435#1790358 <https://reviews.llvm.org/D71435#1790358>, @jdoerfert wrote:

> Thanks! LGTM. I can commit this for you if you want. Maybe update the commit message first.


Thank you! Of course, you may commit it and update the commit message as you wish.

> We can do follow up patches for other instructions/situations now. Some ideas:
> 
> - Look at stores and other memory accesses.
> - Look at branches and other control flow instructions.
> - Look at attribute violations, e.g., null is passed but the `nonnull` attribute is present.
> - Look not only for null but also for undef, maybe even known dangling pointers (via another attribute).
> - Use AAValueSimplify to get an assumed simplified value that we check (that would force us to look at assumed but not known UB instructions every updateImpl call!)
> - Implement the function UB analysis that checks if an instruction that is in the "must-be-executed-context" of the function entry is assumed to have UB.
> 
>   Let me know if you plan to tackle any/all of them.

I'd totally like to tackle all of them, it's pretty interesting already! But it's better to not get a (implicit) assignment in any of them as time is constrained right now.
Definitely I'll start with which seem more approachable like stores and other memory accesses, branches and attribute violations.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D71435/new/

https://reviews.llvm.org/D71435