From chemistry-request- at -server.ccl.net Mon Dec 3 01:49:45 2001 Received: from post.tau.ac.il ([132.66.16.11]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id fB36nii01688 for ; Mon, 3 Dec 2001 01:49:44 -0500 Received: from post.tau.ac.il (pico.tau.ac.il [132.66.145.20]) by post.tau.ac.il (8.11.6/8.11.6) with ESMTP id fB36nYP02038 for ; Mon, 3 Dec 2001 08:49:34 +0200 (IST) Sender: silviu %-% at %-% post.tau.ac.il Message-ID: <3C0B207E.43677C31%!at!%post.tau.ac.il> Date: Mon, 03 Dec 2001 06:49:34 +0000 From: Zilberman Silviu Organization: TAU X-Mailer: Mozilla 4.7 [en] (X11; I; Linux 2.2.10 i686) X-Accept-Language: en MIME-Version: 1.0 To: ccl Subject: automatic parallelization - summery Content-Type: multipart/mixed; boundary="------------2FAEBD0B9CA48985A17436BF" This is a multi-part message in MIME format. --------------2FAEBD0B9CA48985A17436BF Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Hi everyone, It seems that my question is of interest to many people, though I got only two serious replys. The bottom line from those answers is not to expect too much. It might work for a shared memory SMP machine, but it is not particulary good for distributed memory machines. Attached is my original question and the replys. Silviu. -------- Original Message -------- Hi everyone, Does anyone has experience with using auotmatic parallelization tools, for old Fortran and C codes? We are buying a PC cluster, and I look for a quick and dirty way of getting parallel codes out of seriel one. Thanks, Silviu Zilberman. -- Subject: Re: CCL:automatic parallelization Date: 29 Nov 2001 17:36:42 +0100 From: Konrad Hinsen To: Zilberman Silviu References: 1 Zilberman Silviu writes: > Does anyone has experience with using auotmatic parallelization tools, > for old Fortran and C codes? > We are buying a PC cluster, and I look for a quick and dirty way of > getting parallel codes out of seriel one. Don't expect much from them. Automatic parallelization for SMP machines works well for some codes, but certainly not for all. And I haven't even seen a production compiler that pretends to do automatic parallelization for distributed memory machines such as a PC cluster. If you get biprocessor nodes, you might have some success. Just try the new Intel compilers for Linux with the free evaluation licence! -- Subject: Re: CCL:automatic parallelization Date: Thu, 29 Nov 2001 17:34:39 +0000 From: Maurice Cafiero To: Zilberman Silviu CC: ccl References: 1 hello: I have used the automatic parallelization tools that come with SGI Fortan, and they are pretty terrible. They created a 2/3% speedup and were about 5% parallel. I suggest 2 things: 1) Use parallel numerical libraries. These are pre-existing and can probably be downloaded from netlib. 2) Learning basic MPI is not difficult at all for an experienced FORTRAN programmer. I learned it in about 3 weeks and my code is 80/90% parallel. I get excellent speedup (approximately linear with # procs). What type of program are you trying to parallelize? -- Mauricio Cafiero Doctoral Candidate : Theoretical and Computational Quantum Chemistry Department of Chemistry University of Arizona --------------2FAEBD0B9CA48985A17436BF Content-Type: text/x-vcard; charset=us-ascii; name="silviu.vcf" Content-Transfer-Encoding: 7bit Content-Description: Card for Zilberman Silviu Content-Disposition: attachment; filename="silviu.vcf" begin:vcard n:Zilberman;Silviu tel;work:972-3-6407634 x-mozilla-html:TRUE org:Tel Aviv University;School of Chemistry version:2.1 email;internet:silviu%!at!%post.tau.ac.il adr;quoted-printable:;;Tel Aviv University=0D=0ASchool of Chemistry;Tel Aviv;;69978;Israel x-mozilla-cpt:;-17216 fn:Silviu Zilberman end:vcard --------------2FAEBD0B9CA48985A17436BF--