[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
NaNs for storage formats?
Are we assuming that infinities and NaNs are
supported for the storage as well as basic formats?
Definitely yes -- otherwise it would not be an IEEE format.
Even my fictitious binary8 format used to illustrate underflow
for cos(x) supported Inf, Nan and subnormals.
Note that this only requires giving up one reserved exponent.
Even IBM's Hexadecimal XEXP (1980) did that (reserving both
extreme exponents of -8192 and 8191 (power of 16)), even though
the format did not exploit it (it had neither Inf nor denormals).
IBM's regular HFP supports a wrapped exponent for overflow and
underflow by discarding exponent carry, so it could not reserve
an exponent for special values -- so there are situations where
this is an issue -- but not for IEEE-style formats.
Michel.
Sent: 2007-06-05 20:44:41 UTC