[Pkg-octave-devel] Bug#570231: Bug#570231: octave3.2: Errors in working with complex matrices on i386

[Thomas Weber]

On Thu, Feb 18, 2010 at 12:55:21AM +0100, Steven De Herdt wrote:
> On Wed, Feb 17, 2010 at 8:50 PM, Thomas Weber
> <thomas.weber.mail at gmail.com>wrote:
> 
> > Could you use the example from the above bug report and see if
> > uninstalling the optimized libraries helps?
> >
>  No problem:
> 
> **with libatlas3gf-sse2:
> 
> octave:1> a = [1 2 3; 4 5 6; 7 8 9]*J;
> octave:2> [u,s,v]=svd(a)
> u =
> 
>   -0.032510 - 0.122510i   0.334536 - 0.011954i   0.112290 + 0.000145i
>    0.045563 - 0.535097i  -0.458023 + 0.414578i   0.736504 + 0.051393i
>    0.070008 - 0.854667i  -0.724855 - 0.113233i  -0.608726 + 0.014930i
> 
> s =
> 
> Diagonal Matrix
> 
>    17.224974           0           0
>            0    0.922528           0
>            0           0    0.043549
> 
> v =
> 
>   -0.46941 - 0.00000i  -0.85522 - 0.00000i   0.21966 + 0.00000i
>   -0.56108 - 0.11783i   0.19894 + 0.22349i  -0.42448 + 0.61832i
>   -0.64827 - 0.17467i   0.39474 - 0.06283i   0.15152 - 0.61790i
> 
> octave:3> u*s*v'
> ans =
> 
>   -0.00000 + 1.00000i   0.61970 + 1.04384i   0.85486 + 1.28824i
>    0.00000 + 4.00000i   0.63485 + 5.41375i   0.91378 + 6.25676i
>    0.00000 + 7.00000i   0.91329 + 8.54694i   1.52782 + 9.65463i
> 
> 
> **without libatlas3gf-sse2:
> 
> octave:1> a = [1 2 3; 4 5 6; 7 8 9]*J;
> octave:2> [u,s,v]=svd(a)
> u =
> 
>   -0.000000 - 0.214837i  -0.000000 - 0.887231i  -0.406669 - 0.035873i
>    0.000000 - 0.520587i   0.000000 - 0.249644i   0.813338 + 0.071746i
>    0.000000 - 0.826338i   0.000000 + 0.387943i  -0.406669 - 0.035873i
> 
> s =
> 
> Diagonal Matrix
> 
>    1.6848e+01            0            0
>             0   1.0684e+00            0
>             0            0   9.1627e-16
> 
> v =
> 
>   -0.47967 - 0.00000i   0.77669 + 0.00000i   0.40825 + 0.00000i
>   -0.57237 - 0.00000i   0.07569 - 0.00000i  -0.81650 + 0.00000i
>   -0.66506 + 0.00000i  -0.62532 + 0.00000i   0.40825 - 0.00000i
> 
> octave:3> u*s*v'
> ans =
> 
>   -0.0000 + 1.0000i   0.0000 + 2.0000i   0.0000 + 3.0000i
>    0.0000 + 4.0000i  -0.0000 + 5.0000i  -0.0000 + 6.0000i
>    0.0000 + 7.0000i  -0.0000 + 8.0000i  -0.0000 + 9.0000i
> 
> (note that U(:,3) is not quite relevant for this rank-2 matrix)

Okay, so the fault is in the sse2 packages.

> Package libatlas3gf-sse2 has
> http://bugs.debian.org/<http://bugs.debian.org/569436>517826
> and the test mentioned fails with and succeeds without it.  Should this bug
> be moved there?

No, there is already a bug there and I want to know when there are
problems in Octave or libraries used by it.

Thanks
	Thomas