Hi Ahmet, there remains one problem with the icc compiler in one CINT routine in -O mode. This problem has been localized and reproduced in a small test program and Intel is working on a fix. In the ROOT makefile I've added a protection such that this file is always compiled in -O0 mode while the rest of ROOT is compiled in -O (i.e. -O2) mode. I retract evrything I said about icc in the April 8th mail, since a lot of confusion was based on a misunderstanding of the icc man pages in relation to building shared libs. I use icc as default compiler on my machine. Cheers, Fons. On Mon, 2002-05-20 at 22:47, Ahmet Sedat Ayan wrote: > > Dear Rooters, > > I have been investigating if I should be re-compiling/installing > my ROOT with Intel's "icc". In the roottalk I think there are conflicting > opinions on the performance and/or worthiness of this (see the messages > below by Fonds) > > I have a Redhat7.2 Linux box with a P4 1.8 GHz processor and running my > analysis code (no GUI etc involved); reading data from the disk (I/O) , > calculating same variables and filling and plotting many histograms. > (Whole data run takes about 3-4 hours) > > My question is then if it is worth to compile ROOT with "icc" and if I > did if I would see a big improvement in the analysis code running time?? > > thanks a lot > ahmet > > > > **************************************************************** > May,14,2002 > ROOTers, > > with the current version of ROOT in CVS we now fully support the > latest Intel icc v6 compiler. This compiler provides about 30 to 45% > faster code then gcc 2.95 or gcc 3.x on Intel PIII and P4 systems. The > Linux version of this compiler can be freely downloaded from: > http://www.intel.com/software/products/eval/ (get the "non-commercial > unsupported software" version which is free). With ROOT on my P4 machine > this compiler produces 35% faster code (both in debug and optimized > mode). At the same time it compiles also about 20-30% faster than gcc. > To install ROOT do: ./configure linuxicc; make > > For more on this compiler see also: > http://www.open-mag.com/features/Vol_15/IntelC/intelc.htm > > It is nice to see your good old PC suddenly run 35% faster. > > > > Cheers, Fons. > **************************************************************************** > On April,8th,2002 > Hi Andre, > > your observations are correct. Concerning icc (and ecc, the IA-64 > version of icc), I have several incident reports open with Intel > concerning the compiler failing on ROOT's very conservative C++ code. > Actually the icc v6beta cannot even execute ROOT because it generates in > all cases (-g, -O, etc) bad code. This Intel compiler has its origins as > Intel's SPEC reference compiler, i.e. it has been optimized and tuned > with the sole goal to produce the best binaries for the SPEC program > suite (rumour has it it could even generate SPEC binaries without input > source code, i.e. the binaries were hand coded and included in the > compiler proper). > > Cheers, Fons. > > ---------------------------- > > *************** > > Ahmet Sedat Ayan > > Physics & Astronomy Dept. > Van Allen Hall > University of Iowa > Iowa City, IA, 52242 > > Voice : (++ 1 319) 335-1941 (W) (GMT-6) > Occupation : Ph.D Candidate (But still dreamer!) > e-mail : ayan@cms.physics.uiowa.edu > web : http://jazz.physics.uiowa.edu/~ayan > -- Org: CERN, European Laboratory for Particle Physics. Mail: 1211 Geneve 23, Switzerland E-Mail: Fons.Rademakers@cern.ch Phone: +41 22 7679248 WWW: http://root.cern.ch/~rdm/ Fax: +41 22 7679480
This archive was generated by hypermail 2b29 : Sat Jan 04 2003 - 23:50:53 MET