From owner-freebsd-current Tue Mar 17 08:46:50 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id IAA22295 for freebsd-current-outgoing; Tue, 17 Mar 1998 08:46:50 -0800 (PST) (envelope-from owner-freebsd-current@FreeBSD.ORG) Received: from symbion.srrc.usda.gov ([199.78.118.118]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id IAA22170 for ; Tue, 17 Mar 1998 08:46:37 -0800 (PST) (envelope-from glenn@nola.srrc.usda.gov) Received: from symbion.srrc.usda.gov (localhost [127.0.0.1]) by symbion.srrc.usda.gov (8.8.8/8.8.8) with ESMTP id KAA29281; Tue, 17 Mar 1998 10:40:51 -0600 (CST) (envelope-from glenn@symbion.srrc.usda.gov) Message-Id: <199803171640.KAA29281@symbion.srrc.usda.gov> X-Mailer: exmh version 2.0.2 2/24/98 To: Steve Kargl cc: Terry Lambert , freebsd-current@FreeBSD.ORG, ckempf@enigami.com From: Glenn Johnson Subject: Re: Compilers: 2.8.1 v 2.7.2.1? In-reply-to: Your message of "Tue, 17 Mar 1998 07:53:40 PST." Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Date: Tue, 17 Mar 1998 10:40:51 -0600 Sender: owner-freebsd-current@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG > >> FWIW, I've been using egcs-1.0.x for testing g77. On my simple, > >> real-world benchmark (my code :-), g77 is giving about a 22% increase > >> in execution speed over f77 (f2c+gcc). > > > > > >The moral to this story? "I can make it run as fast as you want, as > >long as it doesn't have to actually work". > > > > > You're moral seems to be a non-sequitur with respect to my g77 observation. > Most people would assume a statement of "real-world benchmark (my code)" > means not only was g77 22% faster but it also gives the right answer. > In fact, I can cook up a Fortran program that runs 16 times faster when > complied with g77 than with f2c+gcc. I hardly would call this a > real-world benchmark. (Oh yeah, both give the expected results). > Just to put my 0.02 cents in, there seems to be problems with g77 when compiling "real-world" applications. I am talking about applications such as Gaussian94 and GAMESS (computational chemistry programs). Both of these programs are written in portable Fortan and compile and run on many UNIX platforms. They both compile and run fine when using f2c/cc but g77 chokes on them. On programs that do compile, I have noticed about a 20-30% improvement in execution time, but have also observed some run-time problems, particularly with free(): warning: modified (page-) pointer errors. Interestingly, I only see this with FreeBSD, Linux does not give this error for the equivalent program. Another interesting point is that I can get significant run time speed-ups in Fortran code by compiling the code as a Linux binary using Linux f2c/cc with the -malign-double cc option. I installed the Linux_devel port and added Linux f2c to it to achieve this. Of course, I am working with a small subset ! of Fortran applications but those are my observations, FWIW. -- Glenn Johnson Technician USDA-ARS-SRRC Phone: (504) 286-4252 1100 Robert E. Lee Boulevard FAX: (504) 286-4217 New Orleans, LA 70124 e-mail: gjohnson@nola.srrc.usda.gov To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message