[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Help-gsl] SVD with GSL outperformed by MatLab by a factor of 10
|
From: |
Brian Gough |
|
Subject: |
Re: [Help-gsl] SVD with GSL outperformed by MatLab by a factor of 10 |
|
Date: |
Thu, 16 Dec 2010 20:48:40 +0000 |
|
User-agent: |
Wanderlust/2.15.6 (Almost Unreal) Emacs/23.2 Mule/6.0 (HANACHIRUSATO) |
At Tue, 14 Dec 2010 13:49:56 +0100,
Kalb Max wrote:
> We could live with less precise singular values, since we throw away the
> smallest ones. Additionally, we do not need the according vectors of the
> U-matrix.
>
> Any idead how to improve the GSL version?
As mentioned, you could try ATLAS for the BLAS. Otherwise I'd suggest
profiling the code. On GNU/Linux I would use valgrind for that.
The algorithm implemented in GSL is not as sophisticated as the one in
LAPACK (which is what MATLAB actually uses I believe) so there's a
limit to what can be done without implementing that algorithm.
--
Brian Gough