[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

binary8



In the same vein...

Two models seem appropriate:

(a) p=1, emax=63, emin=-62:  No subnormals (like my decimal8).

    Positive values are just a set of powers of 2.  All positive
    integers fit in a 64-bit unsigned integer, and all but one
    (namely +2**63) fit in a 64-bit signed integer.

(b) p=3, emax=15, emin=-14:  Matches exponent range of binary16.

    Minimal model that exploits all rounding possibilities, i.e.
    exact, half-exact, inexact above and inexact below, for both
    normal and subnormal.  Should be great for test cases!

MIchel.
Sent: 2007-12-30 17:57:21 UTC


754 | revision | FAQ | references | list archive