[PATCH] D45653: [X86] Enable sibling-call optimization for functions returning structs

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

In https://reviews.llvm.org/D45653#1215433, @aleksandr.urakov wrote:

> Yes, I meant 32-bit of course.
>
> I'm not sure about this fix, now it doesn't seem so simple to me... What do you think about the next check?
>
>   if (IsCalleeWin64 != IsCallerWin64)
>     return false;
>   
>
> Do you see any pitfalls here?


I would rather not mess with this... mixing win64 and non-win64 calling conventions is going to be extremely rare in practice, and I don't really want to spend time combing through the ABI documents trying to figure out if anything can go wrong.

> As for different sret arguments, I think that it doesn't matter, because they will be processed in the same way for the same calling convention, and it shouldn't break the stack. So we can replace a `call` with a `jump` in this case, right?

The problem would just be that EAX/RAX is set to the wrong value.


https://reviews.llvm.org/D45653