fpu/e_expl.c for i686

[Stephen L Moshier]

> expl (1000) is .19e435 and can only be represented with long doubles.
> How much would the error be for expl (100)?

Instead of exp(x) the program computes  exp(x(1 + dx)) where dx
is the relative input error introduced, in this case by the range
reduction method.
    exp(x(1 + dx)) = exp(x) (1 + x dx + ...)
So if x is 100 and dx is 1/2 ulp, the relative error would be
about 50 ulps (long double ulps).