[Dune] Timings for the changes in dune-common r4695

[Oliver Sander]

Hi Markus!
From looking at the gcc documentation

http://gcc.gnu.org/onlinedocs/gcc-4.1.1/gcc/Optimize-Options.html#Optimize-Options

it seems to me that even -O3 doesn't turn on loop-unrolling.  Quote:

-O2: [...] The compiler does not perform loop unrolling [...]

-O3: turns on all optimizations specified by -O2 and also turns on the
      -finline-functions, -funswitch-loops and -fgcse-after-reload options.

Could you try your example with -funroll-loops and see what happens?

Thanks,
Oliver

************************************************************************
* Oliver Sander                ** email: sander at mi.fu-berlin.de        *
* Freie Universität Berlin     ** phone: + 49 (30) 838 75217           *
* Institut für Mathematik II   ** URL  : page.mi.fu-berlin.de/~sander  *
* Arnimallee 6                 ** -------------------------------------*
* 14195 Berlin, Germany        ** Member of MATHEON (www.matheon.de)   *
************************************************************************

On Tue, 15 Aug 2006, Markus Blatt wrote:

> Hi,
>
> it seems to me, that the template metaprogramms were useful. At least
> for small blocksizes.
>
> I have attached timings for solving a laplacian on a structured 2d
> 250x250 grid with AMG (with 2 steps jacobi pre and postsmoothing) as a
> preconditioner to CG. (BS is the blocksize, T the time for solving,
> TIT the time for one iteration). The code was compiled with g++-4.0
> and -O3 -DNDEBUG options.
>
> It seems to be that at least for small blocksizes there is a
> difference. Probably one should test also bigger block sizes, but
> right now my time is limited.
>
> Cheers,
>
> Markus
> -- 
> DUNE -- The Distributed And Unified Numerics Environment
> <http://www.dune-project.org>
>