Forwarded Re: optimizing floating point (fwd) <9205152310.AA15642akitzbuhl.iag.hp.com>

[sun!Eng!Keith.Bierman]

> (2) that the meaning of the arithmetic operations of either the C or Fortran
> language is defined in terms of a model of floating-point contained in or
> referenced by the language standard,

There is, effectively, no model of floating point in Fortran (any
version, including "90"). This was not an accident; it was done
intentionally. X3J3 neither wanted to ban any existing machine
implementation (since all machines have FORTRAN compilers, and it is
hoped that all with have Fortran compilers, making a distinction
between machine and compilation system is resisted by the committee.
Also, performance is very important to the X3J3 consititioncy (sp)).

It is, for example, perfectly in conformance to X3.9-1978 or ISO
1539:1991 to have

	2.0+2.0=5.0

This is well known to X3J3 and is not considered a bug in the
standard. 

Some of us hope that IEEE 754 will become so important that the *next*
round of language standards will *mandate*it*. However, the efforts of
projects like LCAS, and new machines with interesting features (fused
muladd, very deferred exceptions, etc.)  make this somewhat unlikely.