[PATCH] XCore fix for SingleSource/Benchmarks/Misc-C++/Large/sphereflake

[Daniel Dunbar]

  Hi Robert,

  It seems to me if the sin function only differs in the last two bits, that
  rounding the LSB should be sufficient, but I haven't really looked at this.

  If you find that rounding the last two bits suffices for making Xcore match
  I would be ok with just checking that change in (not under ifdefs), and
  updating the reference outputs if necessary. I'd find that preferable to
  making the code conditional for a particular platform.

   - Daniel

http://llvm-reviews.chandlerc.com/D3022