libm: optimize fmod

[quaternic]

This is kind of a retry at #898. One of the problems there was that it would have added overhead and regressed performance for typical inputs.

Unlike that PR, this doesn't aim for sub-linear scaling; the cost of evaluating fmod(x, y) is still roughly proportional to log2(|x/y|). However, the constant factor is much better. Running the random-benchmarks locally, I got walltime reductions of

fmodf16:  -56.9%
fmodf:    -85.0%
fmod:     -95.4%
fmodf128: -98.7%

Needs #1011 and #1012

[tgross35]

Sorry I haven't reviewed yet; I am excited about this but have been a bit short on time.

Any chance you're willing to commits to separate PRs? From a quick glance the first two look good, but I'll need a bit more time for the last one.

fmodf128: -98.7%

Seriously impressive :)

[quaternic]

Yeah, I probably should have split it. I'm not home at this time, but I'll try to do that on Tuesday.

If possible, could you run the walltime benchmarks on something non-x86?