[PATCH] Teach DeadArgElimination not to eliminate return values of functions with 'returned' arguments

[Evan Cheng]

On Jun 26, 2013, at 3:56 PM, Chris Lattner <clattner at apple.com> wrote:

> 
> On Jun 26, 2013, at 12:58 PM, Evan Cheng <evan.cheng at apple.com> wrote:
> 
>>> The 'returned' attribute is interprocedural register allocation done at the IR level, encoded with parameter attributes.
>>> 
>>> This is a major architectural change to llvm. I don't think anybody realized the implications of the 'returned' patch when it was under review.
>> 
>> FYI, this was discussed inside the Apple team and Chris has approved this approach. 
> 
> Uh, what?  While I'm a big fan of the "returned" attribute (in that it can model some cool optimizations) I didn't "ok" charging ahead against the feeling of the community.

I'm specifically talking about the "returned" attribute. The current thread is on dead argument elimination. Nick seems to be objecting to the 'returned' patch which went through the proposal, discussion, and has long been committed to llvm.

Evan

> 
> -Chris
>