[Insight-users] ITK and netlab

Kaveh Kohan kaveh.kohan at yahoo.com
Wed Apr 1 12:02:11 EDT 2009


Hi Bill,

I am using gcc 4.3.2, Ubuntu8.10, 64bit, 4Gig RAM, ITK vesion 3.10.

I think that I just found the issue. I found out that my program did not link with LAPACK, blas, at Atlas! gmm++ has a flag (#define GMM_USES_LAPACK) which invokes LAPACK libraries but this flag should be set inside of executable source file not the class source file. Perhaps it is very obvious but I did not know that; and I am not sure this is always the case or not but I removed the flag from class source file and it did not complain. When I put it inside of the executable source file, it linked to the libraires and it is fast.


I still have issue with sparse-dense matrix multiplication but not with dense-dense matrix multiplication anymore. Any idea?

Regards,
K



________________________________
From: Bill Lorensen <bill.lorensen at gmail.com>
To: Kaveh Kohan <kaveh.kohan at yahoo.com>
Cc: Luis Ibanez <luis.ibanez at kitware.com>; insight-users at itk.org
Sent: Tuesday, March 31, 2009 10:31:51 PM
Subject: Re: [Insight-users] ITK and netlab

What compiler are you using and what optimization? A debug build can
be very slow compared to a release build.

On Tue, Mar 31, 2009 at 5:23 PM, Kaveh Kohan <kaveh.kohan at yahoo.com> wrote:
> Hi Luis,
>
> Thank you for the answer.
>
> I am looking for linear algebra functionally because part of my algorithm
> has massive iterative matrix-matrix product and I am looking for an
> efficient way of performing matrix-matrix product.
>
> I did not find VNL sparse matrix strong enough thus I ended up using gmm++.
> I thought (also they mentioned in their website) that gmm++ is an interface
> to blas/lapck/atlas but unfortunately I see a huge gap between effciency of
> matrix-matrix prodcut between MATLAB and gmm (for matrix size I have: 17sec
> vs. 0.5sec). Hence, I am tryig to close the gap by using various libraries
> including vnl (only for dense matrix-matrix operation). I do not expect to
> be as fast as MATLAB but I feel that I am doing something wrong since the
> gap is too big.
>
> Any help or idea is highly appreciated.
>
> Regards,
> Kaveh
>
> ________________________________
> From: Luis Ibanez <luis.ibanez at kitware.com>
> To: Kaveh Kohan <kaveh.kohan at yahoo.com>
> Cc: insight-users at itk.org
> Sent: Tuesday, March 31, 2009 7:42:58 PM
> Subject: Re: [Insight-users] ITK and netlab
>
> Hi Kaveh,
>
>
......


      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20090401/386b5223/attachment.htm>


More information about the Insight-users mailing list