Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sun-barr!olivea!uunet!stanford.edu!leland.Stanford.EDU!elaine18.Stanford.EDU!dhinds
From: dhinds@elaine18.Stanford.EDU (David Hinds)
Newsgroups: comp.arch
Subject: Re: Implementing Interval Arithmetic with IEEE rounding modes
Message-ID: <1991Jun18.193953.12264@leland.Stanford.EDU>
Date: 18 Jun 91 19:39:53 GMT
References: <399@validgh.com>
Sender: news@leland.Stanford.EDU (Mr News)
Organization: Stanford University - AIR
Lines: 18

In article <399@validgh.com> dgh@validgh.com (David G. Hough on validgh) writes:
>Dik Winter observes that dynamic rounding modes (that can be varied
>at run time) don't have to be quite so expensive on pipelined systems
>if the control modes can be written without emptying the pipeline.
>
>If you don't have such a feature then the cost of interval addition
>is dominated by the cost of emptying the pipe and the cost of the two
>additions is minor.  
>
Really, it seems to me that if support for interval arithmetic is built
into a compiler, it should be able to optimize away most of the rounding
mode changes without too much trouble.  This should be trivial for the
cases where the mode changes would otherwise do the most damage, such as
in tight inner loops with no dependencies, i.e., vector operations.  Just
split and do two loops, one for the lower and one for the upper vectors.

 -David Hinds
  dhinds@cb-iris.stanford.edu