# exp(x) = 2^hi + 2^hi (2^lo - 1)
# where hi+lo = log2e*x with 128bit precision
# exact log2e*x calculation depends on nearest rounding mode
# exp(x) = 2^hi + 2^hi (2^lo - 1)
# where hi+lo = log2e*x with 128bit precision
# exact log2e*x calculation depends on nearest rounding mode