I've done a couple of AC rms measurement projects, and have *never* used floating point. Integer arithmetic is fine if you watch what you're doing (overflow and the like).

One of my projects used a lookup table to handle the square root function. I had the advantage that I know the range of values I was dealing with, so my table only covered those values. On the next project I had a different compiler which had an integer sqrt function, so I used that.

By taking my samples every 277.77us I was able to take 60 samples per cycle (with 60Hz power). I had a little state machine to watch for zero crossings, start sampling immediately after the first crossing, stop after the second. Integer arithmetic provided the rms value I was looking for, and the number of samples gave me the frequency.

Dennis