[PATCH] powerpc: Optimized version of powf()

[Szabolcs Nagy]

* Paul Clarke <pc@us.ibm.com> [2017-06-08 18:37:24 -0500]:
> Most significant changes are code simplification and use of doubles for
> intermediate values.  Also, some rearrangement to move early
> non-dependent code later, out of the faster paths.
> 

note that i plan to post my powf implementation that i think
is better, but i cant promise it for this release cycle.

i have a version that handles error cases inline instead of
via a wrapper (same for exp2f, log2f, expf, logf) and that's
harder to integrate.

this version seems to keep the original float coefficeints,
(and various other constructs from the original code)
i think that's suboptimal.