FPTaylor
FPTaylor copied to clipboard
Remove the subnormal term in the improved rounding model
The improved rounding model can be rewritten as
rnd_b(x) = p_2(x, b) * eps(b)
where p_2(x, b) = p_2(x) if |x| = 0 or |x| is in the normal range and p_2(x, b) = min_pos_float(b) / eps(b) when |x| is subnormal (and not 0).