From chemistry-request:~at~:www.ccl.net Wed Sep 2 10:26:02 1998 Received: from ccl.net (atlantis.ccl.net [192.148.249.4]) by www.ccl.net (8.8.3/8.8.6/OSC/CCL 1.0) with ESMTP id KAA21834 Wed, 2 Sep 1998 10:26:01 -0400 (EDT) Received: from bacchus.pc1.uni-duesseldorf.de (root ":at:" bacchus.pc1.uni-duesseldorf.de [134.99.152.11]) by ccl.net (8.8.6/8.8.6/OSC 1.1) with ESMTP id KAA18213 for ; Wed, 2 Sep 1998 10:25:56 -0400 (EDT) Received: from bacchus.pc1.uni-duesseldorf.de (jochen $#at#$ localhost [127.0.0.1]) by bacchus.pc1.uni-duesseldorf.de (8.8.7/8.8.7) with SMTP id QAA09590 for ; Wed, 2 Sep 1998 16:25:31 +0200 From: Jochen Kuepper Reply-To: jochen -8 at 8- uni-duesseldorf.de To: CCL Subject: Re: Linux on Alpha, math library Date: Sat, 9 Oct 2134 22:52:41 +0200 X-Mailer: KMail [version 0.7.9] Content-Type: text/plain MIME-Version: 1.0 Message-Id: <98090216253100.09585 #at# bacchus.pc1.uni-duesseldorf.de> Content-Transfer-Encoding: 8bit X-MIME-Autoconverted: from quoted-printable to 8bit by www.ccl.net id KAA21835 Sorry if that goes to the list twice, I haven't gotten it back from the list yet. ---------- Forwarded Message, on Tue, 01 Sep 1998 Jochen Kuepper wrote: ---------- On Sat, 29 Aug 1998 Victoria Buch wrote: >> Igor J. Barani writes: >> > I have just recently installed Linux on a DEC Alpha and when I compile >> > several programs in fortran on it, they seem to compile to twice the >> > size and require 9 times as long to run than on DEC UNIX. Do you have >> > any idea why this may be so and how I may be able to speed up the >> > processes? Any suggestions would be greatly appreciated. Thank you. >We have just been there. You have the latest Red Hat Linux, dont you? >So: the latest math library libm version 2.0.7 from gnu is messed up. >Our system menager installed old version 2.0.1 - works faster by a factor 2. >More generally our sexperience suggests that >Alpha / Linux for FP operations is good only >for simple operations + - * / . My simulation >programs are still 40% below >the performance of Alpha-DEC, and only 10% faster than Petium II (400MHz) >which is more than 2 times cheaper. Disappointing. >Good luck. Victoria The slowdown of the math-library is due to bug-fixes. The old glibc based math-libs do have problems with wrong results for sqrt ( and others ) ! So you have to decide wether you use correct code or ( little ) faster code that might produce wrong answers. I would suggest to use the newest possible libm, if you are really for speed you might consider recompiling it with optimization turned high (v.i.). More important, there are replacement libraries for the more important math routines, that are hand-optimized assembler. IIRC, one of them is based on code Cray developed for their Alpha based machines. You can use both of them, since each one only provides a subset of functions in the next slower one (libm <- libfm <- libffm ) ! Look at http://people.frankfurt.netsurf.de/Joachim.Wesner/ Another important fact is the compiler to use. egcs-1.1pre is really stable for me, you might consider using it since there are very usefull new optimization passes. Also check your code and the compiler flags you are using, consider -fno-exceptions, -fno-rtti, ... and all the other -m, -f, -O options and take all these that will speed up your programs ! Hope it helps and you all get more power out of your Alpha-toys. If you need some more info, feel free to ask ( Victoria, aren't you in D"usseldorf right now ? ) Greetings, Jochen ----------------------------------------------------------------------- Jochen K"upper Heinrich-Heine-Universit"at D"usseldorf jochen ^at^ uni-duesseldorf.de Institut f"ur Physikalische Chemie I Universit"atsstr. 1, Geb 26.43 Raum 02.29 phone ++49-211-8113681 40225 D"usseldorf fax ++49-211-8115195 Germany http://www-public.rz.uni-duesseldorf.de/~jochen -----------------------------------------------------------------------