[PATCH] D105807: [X86] pr51000 struct return tailcalling

[Josh Haberman via Phabricator via llvm-commits]

haberman added a comment.

> a) middle-end says no -- eg, a local stack-resident variable is live across the call.

What is an example of this? The `musttail` attribute requires that it is immediately followed by a `ret`, so I can't think of any way to trigger this case that would pass IR verification.

The Clang code for `musttail` already detects and rejects the case where a variable with a non-trivial destructor is in scope at the call site: https://github.com/llvm/llvm-project/blob/072669521456a369409cf9db30739a3fac740173/clang/test/SemaCXX/attr-musttail.cpp#L58-L62

> b) the back-end says no -- eg, we have a cross-SO call here (and the got-pointer is callee-save).

This is the case I am worried about. I don't know how we detect this in the front-end, since it's arch-dependent whether cross-SO calls will work.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D105807/new/

https://reviews.llvm.org/D105807