From chemistry-request@server.ccl.net Sun Mar 31 23:49:02 2002 Received: from mallard.duc.auburn.edu ([131.204.2.23]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g314n1p31164 for ; Sun, 31 Mar 2002 23:49:01 -0500 Received: from localhost (kwonohy@localhost) by mallard.duc.auburn.edu (8.9.3/8.9.3) with SMTP id WAA08931 for ; Sun, 31 Mar 2002 22:48:56 -0600 (CST) X-Authentication-Warning: mallard.duc.auburn.edu: kwonohy owned process doing -bs Date: Sun, 31 Mar 2002 22:48:56 -0600 (CST) From: Ohyun Kwon X-Sender: kwonohy@mallard To: chemistry@ccl.net Subject: ecp for gold Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Dear CCLers; I am looking for effective core potentials for gold atom. In particular, large core ECPs(larger than 60 core electrons) for gold atom is needed. I would appreciate it if anybody would kindly give me some advice on this matter. Thank you in advance. Best wishes, Tommy Kwon, Ph.D Department of Chemistry University of Arizona Tucson, AZ email : okwon@u.arizona.edu homepage : http://www.auburn.edu/~kwonohy From chemistry-request@server.ccl.net Mon Apr 1 09:55:09 2002 Received: from sina.iums.ac.ir ([194.225.184.20]) by server.ccl.net (8.11.6/8.11.0) with ESMTP id g31Eswp09966 for ; Mon, 1 Apr 2002 09:55:02 -0500 Received: from amsh0me.iums.ac.ir ([194.225.184.228]) by sina.iums.ac.ir (8.9.2/8.9.2) with ESMTP id TAA19866; Mon, 1 Apr 2002 19:26:26 +0330 (IST) Message-Id: <5.0.2.1.2.20020331143030.00b5f0e8@sina.iums.ac.ir> X-Sender: armin@sina.iums.ac.ir X-Mailer: QUALCOMM Windows Eudora Version 5.0.2 Date: Mon, 01 Apr 2002 19:23:45 +0430 To: CHEMISTRY@ccl.net From: "Armin M. Sobhani" Subject: Re: CCL:Linux/x86 compiler question Cc: Joe M Leonard In-Reply-To: <200203291442.JAA17215@world.std.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; format=flowed On Friday, 29 March 2002, Joe Leonard wrote: >I've been trying to develop numerically-intensive code on >an x86/Linux box (RedHat) after years of working on SGI's. >To date, I've found that the double-precision floating-point >performance of gcc isn't quite as good as IRIX C, and I >wonder whether it's the compiler, the chipset or both... > >Has anybody tried different compilers with things like GAMESS >or Amber? Has there been noticeable improvement with these >or similar codes? > >Are there alternate chipset/compiler Linux's which I should >be looking at? The matter that you have raised about the performance of double-precision math on x86 platform is not only a compiler related thing but also architecture dependent. Traditionally, floating point performance of RISC chips (as you have mentioned about SGI's Iris) is better than CISC counterparts. Nevertheless, there are some workarounds to increase floating point performance of x86 platform. But some history lessons (or better to say terminology) first: When Intel planned to introduce a floating point coprocessor for their new 8086 microprocessor, named 8087 FPU, they actually introduced three floating point formats: single precision (32 bits: 24 bits significand, 8 bits exponent), double precision (64 bits: 53 sig., 11 exp.), and extended precision (80 bits: 64 sig., 15 exp, 1 sign). The single and double precision formats corresponded to C's float and double types or FORTRAN's real and double precision types. But the important thing is that on the x87 FPUs and the 80486 CPU, all computations are done using the extended precision form which is rather slow. Whenever you load a single or double precision value, the FPU automatically converts it to an extended precision value. Likewise, when you store a single or double precision value to memory, the FPU automatically rounds the value down to the appropriate size before storing it. However, when designing the Pentium CPU, Intel redesigned the built-in floating point unit to better compete with RISC chips. Most RISC chips support a native 64 bit double precision format which is faster than Intel's extended precision format. Therefore, Intel provided native 64 bit operations on the Pentium to better compete against the RISC chips. So by changing precision control field of x87 FPU Control Word on the Pentium and latter chips (PII, PIII, P4 ...), you can choose between 24, 53 or 64 bits of precision for floating point operations. That's exactly where the compiler can play an important role: while default precision of FPU is 64 bits (extended double-precision), optimizing compilers on Wintel platform like MS Visual C++ and Intel C/C++ and Fortran compilers use 53 bits (double-precision) for floating point calculation in order to increase the speed. I am not sure about this but I think gcc doesn't switch to 53 bits mode and that's why its double-precision floating point performance is rather low. Considering above mentioned matters, it's clear that using a right compiler can do the trick, but which compiler? Intel C/C++ and Fortran line of optimizing compilers for x86 and Itanium architectures are definitely worth a try. You can download the latest non-commercial version of both compilers (i.e. V5.0) for RedHat Linux 6.2 and 7.1 from Intel's web site (http://developer.intel.com/software/products/compilers/) free of charge but without any support. I have successfully installed them on both RedHat 7.2 and Mandrake 8.0 but with a minor change to installation script for the later case. Both compilers can produce optimized code for PIII, P4 and also MMX, SSE and SSE2 technologies on x86 platform. You can use the -Qpcn compiler option to enable floating-point significand precision control as follows: -Qpc32 24 bits (single precision) -Qpc64 53 bits (double precision) <------- default -Qpc80 64 bits (extended precision) Good luck, Armin p-----__---_________----____----------------------------------------q | / \ / _ _ \ / ____\ Armin M. Sobhani, PharmD, PhD | | / \\ \\ \\ \/\ \___/_ Assitant Prof., Pharmacology Dept. | | / /\ \\ \\ \\ \ \_____ \ Iran University of Medical Sciences| | ( ___ \\ \\ \\ \/____/\ \ mailto:armin@iums.ac.ir | |( \__\_/\__\\__\\__\\__\/\_____/ Tel: +98 (21) 805 2265, Fax: ~64 | | \/__/ \/__//__//__//__/\/____/ P.O.Box: 14155-6183, Tehran, IRAN | |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~| | Once you understand the universe at the atomic level everything | | else is easy. -Richard P. Feynman | b-------------------------------------------------------------------d