From chemistry-request@server.ccl.net Sun May 19 04:27:35 2002 Received: from mail.kemi.dtu.dk (IDENT:root@[130.225.67.150]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g4J8RY130035 for ; Sun, 19 May 2002 04:27:35 -0400 Received: from ouabain.ok.dtu.dk (tr4.ok.dtu.dk [192.38.70.74]) by mail.kemi.dtu.dk (8.11.6/8.11.4) with ESMTP id g4J8RQJ29884; Sun, 19 May 2002 10:27:32 +0200 Received: from malcolm by ouabain.ok.dtu.dk with local (Exim 3.12 #1 (Debian)) id 179M2G-0002jG-00; Sun, 19 May 2002 10:27:24 +0200 Date: Sun, 19 May 2002 10:27:24 +0200 To: chemistry@ccl.net Cc: Matt Zwier Subject: Re: CCL:G98: 64-bit file pointers on ia32 architectures Message-ID: <20020519102724.A10459@tr4.ok.dtu.dk> Mail-Followup-To: chemistry@ccl.net, Matt Zwier References: <3CE4F01B.19471.1D1F8816@localhost> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.2.5i In-Reply-To: <3CE4F01B.19471.1D1F8816@localhost>; from matthew.zwier@hope.edu on Fri, May 17, 2002 at 11:57:15AM -0400 From: Malcolm Gillies On Fri, May 17, 2002 at 11:57:15AM -0400, Matt Zwier wrote: > Earlier in the week we ran into the infamous 2GB file limit with G98 > on Linux (i686). I've searched the G98 site and the CCL archive, > and while it's clear that the kernel and libc both support 64-bit file > positions, it's much less clear how to get G98 to use them. Does > this mean writing a replacement FORTRAN IO library? Or does it > just mean setting some compile-time flag and rebuilding G98? Has > anybody conquered this problem? Is there *any* way other than > just splitting our rwf files, something that hasn't worked too well for > us? It is possible to modify the G98 makefile so it will compile using the Linux Large File Support on Intel x86 architecture 32-bit machines. I have successfully done this using the Intel Fortran Compiler, following the advice of Nakata Maho who has looked into this quite extensively. This circumvents the 2GB file limit (assuming you are using a recent Linux kernel and C library) giving you a 16GB scratch file limit instead. Compilation in full 64-bit mode gets around this limit too, but apparently results in an extreme drop in performance on x86 processors, which cannot deal with 64-bit integers very efficiently. You could try contacing Nakata Maho directly, but he is keen not to upset Gaussian Inc. who apparently don't like this kind of information to circulate too freely. cheers, Malcolm -- Malcolm Gillies Post-doctoral researcher Organic Chemistry, Dept Chemistry, Technical University of Denmark