[PATCH] D97735: [Globals] Treat nobuiltin fns as maybe-derefined.

[Yuanfang Chen via Phabricator via llvm-commits]

ychen added a comment.

I think we should move forward with this approach. The reason being that: the derefinement *only* happens for replaceable builtins. For library provided version of these, the derefinement should not happen (the implementation must provide expected semantics) otherwise it is should be a UB. For replacement definition, we already decorate the function with "nobuiltin" (and it seems this is the use case "nobuiltin" is designed for (https://groups.google.com/g/llvm-dev/c/0HA_Td1y5Po).

PS: I think implementing what @rsmith has suggested should also work: use TargetLibararyInfo to query if a function definition is builtin. But it seems to do more than we needed (i.e, for library definitions where derefinement cannot happen, we don't need this?), and querying TargetLibararyInfo has some compile-time impact.

thoughts?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D97735/new/

https://reviews.llvm.org/D97735