gcc 2.95.3 vs gcc 3.1.1 vs intel c++ 6.0 with BGL and MTL

From: Wintermantel , Marc (marc.wintermantel_at_[hidden])
Date: 2002-08-15 03:00:43

Hi, 

 

We are working on a Finite Element Library (FELyX -
http://felyx.sourceforge.net <http://felyx.sourceforge.net/>  ) that
makes intensive use of the Boost Graph Library (BGL) as well as the
Matrix Template Library (MTL). Up to now we worked with gcc 2.95.3 on
Linux (AMD XP2000+ and P4 2.2GHz). Due to several sources on the
internet we heard, that gcc 3.1.1 should be at least as fast as gcc
2.95.3 and the Intel c++ 6.0 should be even quite a bit faster then the
other two. We used with all three of them -O3 Optimization, but our
results up to now are rather disappointing. For a typical calculation we
received the following results:

 

Gcc 2.95.3        3.5 sec

Gcc 3.1.1          3.9 sec

Intel C++ 6.0     8.2 sec

 

The results were quasi identical on both machines. We also tried a Intel
PIII 850MHz in a notebook, and the times just scaled up. Further we
looked at BGL, MTL and our own code separately and it came out that
every part of the code loses almost equally compared to 2.95.3. Has
anybody similar experiences or a hint what we could do wrong?

 

Yours truly,

 

Marc Wintermantel 

 

Marc Wintermantel

IMES - Structures Technology

ETH Zurich, LEO C4

Leonhardstrasse 27

CH- 8092 Zurich

Tel. ++41 1 633 35 32

Fax. ++41 1 632 17 02

 

[Non-text portions of this message have been removed]