PATCH: Fix the checking of noduplicate in the inline cost analysis

[Chandler Carruth]

Hello folks,

I was reading the code and spotted bugs and inefficiencies in how the
noduplicate checking was done in the inline cost analysis. The following
patch makes it much more consistent with (for example) the returns_twice
and other correctness-invariant checks in the inliner.

However, I don't work on any systems that actually use noduplicate and
there aren't a lot of test cases and so I wanted to punt this to folks that
actually work in this area to see if someone could come up with appropriate
test cases here. I've CC'ed a couple of folks, but really anyone dealing
with SPMD stuff could probably whip together a nicer test case than me.

Anyways, I've no idea when I'll find time to craft a test case myself, and
wanted to at least make sure folks had the fix in hand in case they're
running into this.

-Chandler
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131226/106b9d5c/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: fix-noduplicate-analysis.patch
Type: application/octet-stream
Size: 3886 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131226/106b9d5c/attachment.obj>