[PATCH] Added transform for ABS(NABS(X)) and NABS(ABS(X))

[Chandler Carruth]

LGTM

I wouldn't bother re-testing all the permutations of the 'icmp' representation -- that code should already be well tested by existing checks? Still, not a big deal, just trying to minimize the redundancy in the testing.

http://reviews.llvm.org/D4040