(Free42) accuracy of LN1+X

[Paul Dale]

I've not checked the Intel code for this function, but the library does make use of binary floating point functions in several places.

This function is actually fairly easy to implement once its problems are realised. The code on the 34S is:

Code:

decNumber *decNumberLn1p(decNumber *r, const decNumber *x) {
    decNumber u, v, w;

    if (decNumberIsSpecial(x) || dn_eq0(x)) {
        return decNumberCopy(r, x);
    }
    dn_p1(&u, x);
    dn_m1(&v, &u);
    if (dn_eq0(&v)) {
        return decNumberCopy(r, x);
    }
    dn_divide(&w, x, &v);
    dn_ln(&v, &u);
    return dn_multiply(r, &v, &w);
}

Pauli