[PATCH] D28484: SimplifyLibCalls: Replace more unary libcalls with intrinsics

[Davide Italiano via Phabricator via llvm-commits]

davide added a comment.

In https://reviews.llvm.org/D28484#640523, @efriedma wrote:

> For anyone else following along with a copy of IEEE 754, this is the roundToIntegral* family of functions.
>
> Aren't llvm.rint() and llvm.nearbyint() precisely equivalent?  We should probably get rid of llvm.rint()... (we can add it back once we actually start modeling FP exceptions).


I think so.



================
Comment at: test/Transforms/InstCombine/win-math.ll:273
     %3 = fptrunc double %2 to float
     ret float %3
 }
----------------
efriedma wrote:
> Windows doesn't have round()?  Weird.
I'm surprised too https://msdn.microsoft.com/en-us/library/dn353646.aspx
This also compiles http://rextester.com/MUCW42464 (unless I'm misunderstanding).


https://reviews.llvm.org/D28484