[PATCH] D106000: [NFC] X86: Annotate sibcall test

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

I'd prefer to structure the comments in the following form, so it's clear at a glance which transforms are legal/illegal:

  ; Tailcall legal: caller and callee signatures match
  ; QOI: we don't currently emit a tailcall on i686 because arguments on stack don't match.



================
Comment at: llvm/test/CodeGen/X86/sibcall.ll:68
+;; callee-clobberable?, of course such rewriting can get hairy due
+;; to RAW conflicts
 define dso_local void @t4(void (i32)* nocapture %x) nounwind ssp {
----------------
The callee owns the argument space; we can rewrite it if we need to.  We do in fact rewrite it if you use musttail. (Well, the IR verifier doesn't allow musttail for this specific testcase, but in general.)

Yes, making sure we perform the operations in the right order can get a bit hairy, particularly in cases involving byval.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D106000/new/

https://reviews.llvm.org/D106000