Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!mips!pacbell.com!pacbell!rtech!ingres!Ingres.COM!jpk
From: jpk@Ingres.COM (Jon Krueger)
Newsgroups: comp.arch
Subject: Re: IEEE arithmetic (Goldberg paper)
Message-ID: <1991Jun29.021450.18391@ingres.Ingres.COM>
Date: 29 Jun 91 02:14:49 GMT
References: <9106210542.AA21933@ucbvax.Berkeley.EDU> <1991Jun24.211407.17389@ingres.Ingres.COM> <64885@bbn.BBN.COM>
Reply-To: jpk@Ingres.COM (Jon Krueger)
Organization: Ingres Division, ASK Computer Systems.
Lines: 24

I wrote:
>>It is a good thing for compilers to expose broken programs.

Stan wrote:
>Many vector machines do summations in different than program order; thus
>optimization on vs. off can and does produce different answers.  They
>can be very different depending upon the actual numbers.

Not what I was referring to.  Such compilers are broken.

The meaning of some programs depends on the ordering of some of
their steps.  To change the ordering is to change the meaning --
an impermissible optimization.

A program in a sequential language may reasonably depend on the
sequential ordering of its steps.  Such programs are not broken.

A program in a language where order of expression evaluation is
undefined may not reasonably depend on left-to-right evaluation.
Such programs are broken.

-- Jon

Jon Krueger, jpk@ingres.com