RFC: min/max/abs IR intrinsics

[Owen Anderson]

> On Apr 24, 2015, at 2:19 PM, Gerolf Hoflehner <ghoflehner at apple.com> wrote:
> 
> I was wondering if there were a performance advantage to have a higher level representation of min/max. My hunch is that recognizing the idiom at the low-level is ‘good enough’.

I suspect (aside from vectorization) the impact to be minimal for integer min/max.  For floating point, I expect there to be greater upside because of ISA level support, and because the SW emulation expansion is typically complex.

—Owen
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150424/b6c512d4/attachment.html>