Date: Sun, 11 Jan 2009 11:52:07 -0600 From: Stephen Montgomery-Smith <stephen@math.missouri.edu> To: "Pedro F. Giffuni" <giffunip@tutopia.com> Cc: "O. Hartmann" <ohartman@mail.zedat.fu-berlin.de>, freebsd-current@freebsd.org Subject: Re: Alternatives to gcc (was Re: gcc 4.3: when will it become standard compiler?) Message-ID: <496A31C7.2020107@math.missouri.edu> In-Reply-To: <458984.49823.qm@web32708.mail.mud.yahoo.com> References: <61484.71762.qm@web32708.mail.mud.yahoo.com> <20090111044448.GC5661@server.vk2pj.dyndns.org> <4969CC6D.6030707@mail.zedat.fu-berlin.de> <458984.49823.qm@web32708.mail.mud.yahoo.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Pedro F. Giffuni wrote: > > ----- Original Message ---- > ... >> Well, initially my question was triggered by reading a performance duell >> between FreeBSD 7/8, most recent U(n)buntu and OpenSolaris and someone >> stated the 3% performance gain of U(n)buntu over FreeBSD was due to the >> gcc4.3 compiler, which generates more efficient code. 3% mean >> performance gain could mean (as I made this experience) a better >> advantage in some special cases and having in mind numerical modelling >> running on my lab's FreeBSd box (yet, but I think this is about to >> change and move towards a RH Linux system due to the better support of >> HPC and, a pitty, our admins build the cluster with RH and not FBSD). >> > > Even when it can be measured, performance can be very subjective, performance > depends on many factors: the threading libraries, the options used to build the > packages, the filesystems and maybe even the position of the moon ;-). Most of > my numerical packages don't depend on the system compiler but rather depend on > what the ports system uses as the Fortran compiler so you will be glad to know > that we are indeed using gcc4.3 since last week. I also do quite a bit of numerical work. For me a 3% performance gain is not that much, and really becomes negligible compared to other issues. I have written some software that, a year ago, ran twice as fast under Fedora Linux than it did under FreeBSD. Now FreeBSD has completely caught up! And I didn't change the software itself in any substantial manner. My guess is that FreeBSD has improved its cache management/threading management considerably (because my programs (a) use large amounts of data and (b) are threaded). So, for me, a big difference is 2 to 1. A factor of 3% is definitely something dependent on the "position of the moon" as Pedro put it so eloquently. Stephen
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?496A31C7.2020107>