Re: [ublas] [ublas-dev] poor vector/matrix-proxy performance

From: Gunter Winkler (guwi17_at_[hidden])
Date: 2005-08-25 01:11:21

Christophe Prud'homme wrote:

>g++  -o cholesky_test cholesky_test.cpp -DNDEBUG -O2  
>
>for i in 100 200 400 800 1600; do ./cholesky_test $i ; done 
>0: 0 (deco: 0 sec) (prod: 0 sec) 100
>0: 0 (deco: 0.01 sec) (prod: 0 sec) 100
>0: 0 (deco: 0 sec) (prod: 0 sec) 100
>0: 0 (deco: 0.01 sec) (prod: 0.02 sec) 200
>0: 0 (deco: 0.01 sec) (prod: 0.02 sec) 200
>0: 0 (deco: 0.01 sec) (prod: 0.01 sec) 200
>0: 0 (deco: 0.04 sec) (prod: 0.38 sec) 400
>0: 0 (deco: 0.12 sec) (prod: 0.12 sec) 400
>0: 0 (deco: 0.02 sec) (prod: 0.02 sec) 400
>0: 0 (deco: 0.52 sec) (prod: 3.38 sec) 800
>0: 0 (deco: 1 sec) (prod: 0.94 sec) 800
>0: 0 (deco: 0.05 sec) (prod: 0.04 sec) 800
>1570: 1.41535e+13 (deco: 5.08 sec) (prod: 24.27 sec) 1600
>1570: 1.41535e+13 (deco: 8.07 sec) (prod: 7.54 sec) 1600
>0: 0 (deco: 0.22 sec) (prod: 0.07 sec) 1600
>  
>
I should try gcc-4, too.
Your times show the same behaviour: using a triangular matrix slows down 
the deco. I expected the opposite.

mfg
Gunter