Method: Let x = 2 * (1+f)
1. Compute and return log2(x) in two pieces:
log2(x) = w1 + w2,
where w1 has 53-24 = 29 bit trailing zeros.
2. Perform y*log2(x) = n+y' by simulating muti-precision
arithmetic, where |y'|<=0.5.
3. Return x**y = 2**n*exp(y'*log2)