Re: Octave for Mac with Reference Lapack

[Lukas Reichlin]

On 29.05.2012, at 16:35, Alexander Hansen wrote:

> On 5/29/12 4:22 AM, Ben Abbott wrote:
>> 
>> On May 29, 2012, at 3:29 AM, Lukas Reichlin wrote:
>> 
>>> Hi Alexander
>>> 
>>> Would you help me creating an Octave variant for Fink with reference BLAS 
>>> and LAPACK libraries (both included in lapack 3.4.1 from 
>>> http://www.netlib.org/lapack/)?
>>> 
>>> The reason for such a variant is that I'm experiencing problems with stuff 
>>> like wrong Hankel singular values and non-converging SVDs with both ATLAS 
>>> and Apple's Accelerate/Veclib framework. The SLICOT authors recommend that 
>>> I use the reference LAPACK implementation from Netlib. It solved the 
>>> problems, and now I would like to use reference LAPACK also for Octave 
>>> itself and not only my control package.
>>> 
>>> Best regards,
>>> Lukas
>> 
>> Lukas,
>> 
>> Do you have a simple example that demonstrates the problem ? ... or maybe a 
>> link to the details ?
>> 
>> Ben
>> 
> 
> I _can_ do it, yeah.  We found under testing that it's best to use the
> same LAPACK/BLAS provider for any dependencies of Octave as for Octave
> itself, so that's going to entail making variants for Octave, arpack-ng,
> and qrupdate.  It'll probably be best to create a package based on
> lapack-3.4.1 that all of these can build against.
> 
> I'd like to see a test case, too.
> -- 
> Alexander Hansen, Ph.D.
> Fink User Liaison
> http://finkakh.wordpress.com/

The really bad examples from system identification are not ready for a release 
yet. But it is sufficient to install the official control-2.3.50 package on a 
Mac and run
        test_control
All tests should pass. A few sign changes are not a problem, but wrong Hankel 
singular values are a no-go. Framework Accelerate on Mac OS X 10.7 is the worst 
of all, not to say FUBAR, followed by various ATLAS 3.9.* releases (MacPorts 
doesn't have 3.8.x). Accelerate from 10.6 is slightly better, but still 
unusable for SLICOT system identification purposes. I lost several weeks of my 
thesis because of that …

Best regards,
Lukas