> Hugs971031 gives floatDigits to 24 (resp. 53) for 4 (resp 8) byte IEEE
> binary floats; however, the number of bits in the mantissa are 23 (resp
> 52). (There looks to be a errors on p.119 in "A C Reference Manual" by
> Harbison-Steele. For a 4-byte float, they give a 24 bit mantissa, an
> exponent in the range [-125, 128], which consumes 8 bits, plus a sign bit,
> which adds up to 33, not 32.)
>
> So what is the correct value of floatDigits?
I would say that 24 and 53 bits are the correct values for normalized
IEEE numbers. Since a normalized number always has a 1 as the MSB
that bit is not stored, hence there is one more bit of precision than
there are bits in the format.
-- Lennart
