TODO:
----
Elliminate floa128 use, Intel uses only 67-bit precision calculations
when float128 has at least 112-bit. Replacement of float128 with for
example 96-bit precision number could significantly speed up
calculations.