pub(super) const LN2_SPLIT: [f64; 3];
ln 2 in three parts: A rounded to 42 bits (so that e·A is exact for |e| < 2¹¹), then RN(ln 2 − A), RN(ln 2 − A − B) (Algorithm L, step L6).
ln 2
A
e·A
|e| < 2¹¹
RN(ln 2 − A)
RN(ln 2 − A − B)