Re: Octave for Mac with Reference Lapack

[Lukas Reichlin]

On 29.05.2012, at 18:26, c. wrote:

> 
> On 29 May 2012, at 17:24, Lukas Reichlin wrote:
> 
>> 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.
> 
> I am running Octave 3.7.0+ linked to the Accelerate framework on OSX 10.6.8.
> This is what I get, which ones are the faiures you call "no go" and which are 
> "not a problem"?
> 
>>> pkg install -forge control
> ar: creating archive slicotlibrary.a
> warning: gen_doc_cache: unusable help text found in file 
> 'control_slicot_functions'
> For information about changes from previous versions of the control package, 
> run 'news ("control")'.
>>> test_control 
> PASSES 34 out of 34 tests
> PASSES 5 out of 5 tests
> PASSES 3 out of 3 tests
> PASSES 5 out of 5 tests
> PASSES 1 out of 1 tests
> PASSES 1 out of 1 tests
> PASSES 9 out of 9 tests
> PASSES 1 out of 1 tests
> PASSES 4 out of 4 tests
> PASSES 1 out of 1 tests
> PASSES 22 out of 22 tests
> PASSES 12 out of 12 tests
> PASSES 8 out of 8 tests
> PASSES 2 out of 2 tests
> PASSES 2 out of 2 tests
> PASSES 27 out of 27 tests
> PASSES 7 out of 7 tests
> PASSES 3 out of 3 tests
> PASSES 3 out of 3 tests
> PASSES 4 out of 4 tests
> PASSES 2 out of 2 tests
> PASSES 5 out of 5 tests
> PASSES 3 out of 3 tests
> PASSES 3 out of 3 tests
> PASSES 2 out of 2 tests
> PASSES 2 out of 2 tests
>  ***** assert (Mo, Me, 1e-4);
> !!!!! test failed
> assert (Mo,Me,1e-4) expected
>  -0.23910   0.30720   1.16300   1.19670  -1.04970
>  -2.97090  -0.23910   2.62700   3.10270  -3.70520
>   0.00000   0.00000  -0.51370  -1.28420   0.82230
>   0.00000   0.00000   0.15190  -0.51370   0.74350
>  -0.44660   0.01430  -0.47800  -0.20130   0.02190
> but got
>  -0.23915   0.30723   1.16297  -1.19671   1.04965
>  -2.97091  -0.23915   2.62702  -3.10273   3.70515
>   0.00000   0.00000  -0.51368   1.28421  -0.82227
>   0.00000   0.00000  -0.15189  -0.51368   0.74348
>   0.44660  -0.01427   0.47803  -0.20129   0.02190
> maximum absolute error 7.41035 exceeds tolerance 0.0001
> shared variables 
>  scalar structure containing the fields:
> 
>    Mo: 5x5 matrix
>    Me: 5x5 matrix
>    Info: 1x1 scalar struct
>    HSVe: 6x1 matrix
> PASSES 2 out of 2 tests
>  ***** assert (Mo, Me, 1e-4);
> !!!!! test failed
> assert (Mo,Me,1e-4) expected
>   0.59460  -0.73360   0.19140  -0.33680   0.00150
>   0.59600  -0.01840  -0.10880   0.02070  -0.02020
>   1.22530   0.20430   0.10090  -1.49480   0.01590
>  -0.03300  -0.02430   1.34400   0.00350  -0.05440
>   0.35340   0.02740   0.03370  -0.03200   0.00000
> but got
>   0.59461  -0.73360  -0.19139   0.33685   0.00146
>   0.59599  -0.01839   0.10883  -0.02070  -0.02016
>  -1.22535  -0.20431   0.10090  -1.49478  -0.01592
>   0.03300   0.02426   1.34404   0.00350   0.05444
>   0.35342   0.02743  -0.03372   0.03204   0.00000
> maximum absolute error 2.45065 exceeds tolerance 0.0001
> shared variables 
>  scalar structure containing the fields:
> 
>    Mo: 5x5 matrix
>    Me: 5x5 matrix
>    Info: 1x1 scalar struct
>    HSVe: 8x1 matrix
> PASSES 2 out of 2 tests
> PASSES 1 out of 1 tests
> PASSES 4 out of 4 tests
> PASSES 1 out of 1 tests
> PASSES 4 out of 4 tests
> PASSES 1 out of 1 tests
> PASSES 2 out of 2 tests
> PASSES 30 out of 30 tests
> PASSES 2 out of 2 tests
> PASSES 1 out of 1 tests
> PASSES 2 out of 2 tests
> 
> 
>> 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
> 
> c.

These are "not a problem". I used Accelerate from Mac OS X 10.6.8 as well until 
a few weeks ago. It works OK for the released control functions but it doesn't 
work correctly for the (upcoming) SLICOT system identification functions any 
more. 
Accelerate from 10.7 is unusable even for control-2.3.50, some algorithms 
simply fail, IIRC tests from "hinfsyn" are amongst them.

Best regards,
Lukas