[PATCH] D79454: [IR] `byval` arguments cause reads at call sites

[Hal Finkel via Phabricator via llvm-commits]

hfinkel added a comment.

> I don't think it is preferable to say the copy happens in the callee because the user could reasonably write something like this: https://godbolt.org/z/8vxz7P and we would need to resolve the discrepancy between the pure attribute and the byval passing in Clang.

I think that's a good point. It does introduce a discrepancy where the callsite attributes/properties and those of the callee. This isn't novel, I suppose, because we have that situation with operand bundles too. That all having been said, if we have a situation where a byval argument is never accessed, perhaps we really want the optimizer to just remove the byval attribute?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D79454/new/

https://reviews.llvm.org/D79454