Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!cmcl2!rutgers!iuvax!iucs!bobmon
From: bobmon@iucs.UUCP (RAMontante [condition that I not be identified])
Newsgroups: comp.sys.ibm.pc
Subject: Re: IEEE floating point
Message-ID: <4714@iucs.UUCP>
Date: Sun, 11-Oct-87 21:44:18 EDT
Article-I.D.: iucs.4714
Posted: Sun Oct 11 21:44:18 1987
Date-Received: Tue, 13-Oct-87 00:56:35 EDT
References: <871@cup.portal.com> <1118@omepd>
Reply-To: bobmon@iucs.UUCP (RAMontante [condition that I not be identified])
Distribution: na
Organization: Indiana University, Bloomington
Lines: 25
Keywords: IEEE F.P. standard

Steven_N_Jones@cup.portal.com writes:
>    In its announcements of new language products, Microsoft has said
>that it is standardizing on the IEEE floating point format.  Could
>somebody point me in the right direction; I'm not sure where this
>format would be explained.

John Wakerly (in his book Microcomputer Architecture and Programming) describes
a proposed (as of 1981 or so) IEEE standard as follows:

    The IEEE Standards Committee has proposed a 32-bit floating-point format
    similar to the PDP-11's.  In the proposed standard, the binary point lies
    to the _right_ of the hidden bit; thus mantissas are between 1 and 2.  The
    exponent uses excess-127 notation for normalized numbers, with the range of
    valid exponents being -126 (00000001 base2) through +127 (11111110 base2).
    The maximum and minimum exponents (00000000 base2) and (11111111 base2) are
    used to indicate exceptional conditions.  The standard also specifies a
    64-bit ("double precision") format and gives guidelines for extended
    formats (e.g., 128-bit "quad precision" format).

The "hidden bit" refers to the most significant bit, which is guaranteed to
be one for a normalized (non-zero) number.  Since it's known to be one, there's
no need to actually store it, and it is simply assumed to exist.  Thus 23 bits
can be used to store the mantissa to 24 bits of precision.  I'm pretty sure the
"minimum exponent" (00000000 base2) is used to indicate a value of exactly zero.
Hope this helps.