Re: Fourth Draft "Unsafe fp optimizations" project description.

["Tim Prince" <tprince at computer dot org>]

----- Original Message -----
From: <dewar@gnat.com>
To: <kaih@khms.westfalen.de>;
<moshier@moshier.ne.mediaone.net>
Cc: <gcc@gcc.gnu.org>
Sent: Wednesday, August 15, 2001 4:38 PM
Subject: Re: Fourth Draft "Unsafe fp optimizations" project
description.


> <<Whatever the IEEE 754 committee may say about it, however,
the fact is
> that vendors are going to continue to offer flush to zero
machines
> because there are good engineering reasons to do so.  It
would be most
> helpful if the committee would seek to determine and codify
the
> industry practice.  It will be much less helpful if the
committee
> comes out with unpracticed rules that the industry is going
to ignore
> as it has already done in the past.
> >>
>
> THat's a little fierce. It is true that some companies have
ignored
> denormals (notably Alpha -- to be discontinued in 2004, and
SGI/MIPS --
> with SGI stock at 50 cents, destined to play a smaller role
in the
> future of high end numerical computing I fear :-)
>
> However, most architectures do adequately support denormals.

The hardware designers' idea of adequacy has moved in the
direction of giving the choice of correct denormal (soon to be
officially subnormal) results with inadequate performance, or
abrupt underflow with good performance, usually without
adequate facilities in the compiler to specify a preference.