Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!zephyr.ens.tek.com!tektronix!reed!intelhf!ichips!ichips!glew From: glew@pdx007.intel.com (Andy Glew) Newsgroups: comp.arch Subject: Re: IEEE arithmetic (Goldberg paper) Message-ID: Date: 9 Jun 91 18:08:17 GMT References: <9106052113.AA14439@ucbvax.Berkeley.EDU> <3652@charon.cwi.nl> Sender: news@ichips.intel.com (News Account) Organization: Intel Corp., Hillsboro, Oregon Lines: 30 In-Reply-To: dik@cwi.nl's message of 8 Jun 91 00:48:27 GMT > > > Also as has already been pointed out there is a software cost > > >due to such features as 0*x no longer being 0. In some cases this will > > >have a performance impact in that compilers will generate inferior code. >I still have a question here: do you know realistic cases where inferior >code will be generated? I do not think that 'a = 0.0 * b' is realistic. Urghh.... I used to be of this opinion - until I started making heavy use of preprocessors so that a single "generic" code can serve multiple purposes. Basically, I do poor man's partial evaluation - where I say "these parameters are fixed" - now give me an optimized program that evaluates the rest of the numbers I want. In this programming model expressions like 0.0 * b arise frequently. And it sure would be nice if they could be optimized away. For an introduction to partial evaluation for scientific codes, read "Compiling scientific Code Using Partial Evaluation", Andrew Berlin and Daniel Weise, IEEE Computer, December 1990. -- Andy Glew, glew@ichips.intel.com Intel Corp., M/S JF1-19, 5200 NE Elam Young Parkway, Hillsboro, Oregon 97124-6497 This is a private posting; it does not indicate opinions or positions of Intel Corp.