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

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

re: [[clang::musttail]]: if the C++-level types match, musttail should always succeed.  If it doesn't, that's generally a backend bug.  Note that the [[clang::musttail]] rules imply the sret markings match.  (The old PowerPC ABI has complicated issues with tail calls, though...)



================
Comment at: llvm/test/CodeGen/X86/sibcall.ll:861
 
 define fastcc void @t21_sret_to_non_sret(%struct.foo* noalias sret(%struct.foo) %agg.result) nounwind  {
 ; X86-LABEL: t21_sret_to_non_sret:
----------------
Are you sure it's legal to transform t21_sret_to_non_sret?

>From the x86_64 ABI document: "If the type has class MEMORY [...] On  return `%rax` will contain  the address that has been passed in by the caller in `%rdi`."  t21_f_non_sret won't do that.


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

https://reviews.llvm.org/D105807