[Getfem-users] Benchmark

[Michael Lehn]

Hi,

I am currently preparing a benchmark for comparing FLENS with Gmm++.

As a first example I would like to compare
1) initializing a sparse matrix
2) computing the residual (where vectors are dense)

I already started to compare FLENS with uBLAS.  In a first test the  
matrix was initialized row-wise:

        http://flens.sourceforge.net/session2/tut6.html

in a second test row and column indices were chosen randomly:

        http://flens.sourceforge.net/session2/tut7.html

I attached a first version for a benchmark test case (were the matrix  
gets initialized row-wise).  In a first
test the results were not so good.  So maybe you can give me a hint  
how to improve the code in order to
get a fair comparison.  As compiler-flags I used

g++ -DNDEBUG -ftemplate-depth-50 -funroll-loops -O3 -Wall -I/cluster/ 
include -I . -o ex_residual ex_residual.cc

I will perform the benchmarks on a Linux System with Opteron  
processors (see above links).

Thanks,

Michael

ex_residual_gmm.cc
Description: Binary data