benchmarking Poly/ML & floating point

[Stephen Weeks]

> O.k.  It's a very simple peephole optimization anyways, so I think I'll
> write it up and run it through the regressions and benchmarks and see if
> it is ever activated.  From the CPS code, I'm guessing that if it is ever
> used, it will be in some extreme case.  If it doesn't appear to hit
> anything, then I'll just leave it turned off.

If it's not too slow (I guess it's not), I'd leave it in anyways.