From owner-freebsd-alpha Tue Mar 18 13:51:51 2003 Delivered-To: freebsd-alpha@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id 6A94037B401 for ; Tue, 18 Mar 2003 13:51:50 -0800 (PST) Received: from wall.polstra.com (wall-gw.polstra.com [206.213.73.130]) by mx1.FreeBSD.org (Postfix) with ESMTP id 349BD43FA3 for ; Tue, 18 Mar 2003 13:51:49 -0800 (PST) (envelope-from jdp@polstra.com) Received: from vashon.polstra.com (vashon.polstra.com [206.213.73.13]) by wall.polstra.com (8.12.3/8.12.3) with ESMTP id h2ILpcJL061907 (version=TLSv1/SSLv3 cipher=EDH-RSA-DES-CBC3-SHA bits=168 verify=NO); Tue, 18 Mar 2003 13:51:39 -0800 (PST) (envelope-from jdp@vashon.polstra.com) Received: (from jdp@localhost) by vashon.polstra.com (8.12.6/8.12.6/Submit) id h2ILpcPP035490; Tue, 18 Mar 2003 13:51:38 -0800 (PST) (envelope-from jdp) Date: Tue, 18 Mar 2003 13:51:38 -0800 (PST) Message-Id: <200303182151.h2ILpcPP035490@vashon.polstra.com> To: alpha@freebsd.org From: John Polstra Cc: peter.jeremy@alcatel.com.au Subject: Re: HEADS UP: Don't upgrade your Alphas! In-Reply-To: <20030318195946.GZ90290@gsmx07.alcatel.com.au> References: <20030317161219.GA1429@sunbay.com> <15991.9122.826678.673067@grasshopper.cs.duke.edu> <200303181645.h2IGjcuG035185@vashon.polstra.com> <20030318195946.GZ90290@gsmx07.alcatel.com.au> Organization: Polstra & Co., Seattle, WA Sender: owner-freebsd-alpha@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.org In article <20030318195946.GZ90290@gsmx07.alcatel.com.au>, Peter Jeremy wrote: > On 2003-Mar-18 08:45:38 -0800, John Polstra wrote: > > > >I agree! The lack of -mieee has nailed me on several occasions. > >"Fast but broken" is a lousy default. > > The Compaq compiler also defaults to not generating IEEE code. Good point. But I think that's mostly because DEC marketing had to play "win the benchmark" games that we don't have to play. Also, DEC had only one platform architecture. Since we have several architectures, we should strive to make them behave as much alike by default as we reasonably can. > For most FP work, the AXP partial IEEE support is "good enough". True, but the trouble is that it's usually the input to the program that determines whether the partial IEEE support will cause a problem or not. A program may be good enough during testing, but fail miserably later with real world inputs. > On older Alpha's, full IEEE compliance is fairly expensive because > of the lack of exact FP exceptions. Granted. But older Alphas are total dogs anyway (even compared to Intel PCs from the same era), so I don't think we should let them influence our decisions much. > In any case, Seymour Cray made quite a good living pandering to > people for whom speed was more important than accuracy :-). I can't argue with that ... :-) John -- John Polstra John D. Polstra & Co., Inc. Seattle, Washington USA "Disappointment is a good sign of basic intelligence." -- Chögyam Trungpa To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-alpha" in the body of the message