Fwd: Re: Compiler performance for MTL application: intel C++ vs gcc

In my experience and on our finite element code, icc is also slower compared
to gcc 3.1. gcc3.1 and gcc 2.95.3 show about the same performance.

You missed some optimisation flags though. For the icc, you can also use
-tpp6 and -axK. For the gcc compiler you can set options to set the
inlining-level etc which also affect performance.

toon

On Thursday 15 August 2002 03:50, König, Oliver wrote:
> Hi,
>
> We just got some strange results, comparing the performance of different
> compilers for an application that heavily uses MTL. We tested our Finite
> Element library FELyX (http://felyx.sourceforge.net
> <http://felyx.sourceforge.net/> ) that uses MTL for all matrix and linear
> algebra operations. We use the new version of MTL 2.1.2-21 and did include
> the matrix specializations offered from Trevor Blackwell
> (http://tlb.org/software.html) that boost matrix multiplications in the
> assembly processes of FELyX quite a bit....
>
> Resulting runtimes for the analysis of a midsized FEM-structure resulting
> in an envelope matrix with app. 7000 rows:
>
> g++ 2.95.3:                   3.5s
> g++ 3.1.1:                     3.9s
> intel C++ V6.0:              8.2s !
>
> Tests were run on an AMD XP2000+ (1700 Mhz) under Suse Linux V8.0 .
> Compiler options for all three tests:  -O3
>
> Does anybody have some explanations for this results? Looking at different
> benchmark tests we did expect to see a very fast intel C++ followd by a
> fast g++ 3.1.1 and an outperformed g++ 2.95.3.  Looking closer at the
> different runs, one can see that Intel's C++ linearly looses time during
> the whole process. Maybe we missed some important points how to use this
> newer compilers ?
>
> Sincerely
> ok
>
> ----------------------------------------------------
> Oliver Koenig
> Center of Structure Technologies
> Institute of Mechanical Systems
> LEO C12
> ETH Zurich
> CH-8092 Zurich
> Phone : +41 1 632 23 28
> Fax :   +41 1 632 17 02
> Mail :  [email protected]
> WWW :  http://www.imes.ethz.ch/st/
> ----------------------------------------------------

-------------------------------------------------------
_______________________________________________
This list is archived at http://www.osl.iu.edu/MailArchives/mtl-devel/