[SciPy-dev] Reason for failures inscipy.test() ?

[Pearu Peterson]

On Mon, 24 Nov 2003, Nils Wagner wrote:

> Dear experts, 
>  
> Can someone reproduce these failures with latest cvs ? 
>  
> ====================================================================== 
> FAIL: check_random_complex 
> (scipy.linalg.decomp.test_decomp.test_cholesky) 
> ---------------------------------------------------------------------- 
> ====================================================================== 
> FAIL: check_simple_complex 
> (scipy.linalg.decomp.test_decomp.test_cholesky) 
> ---------------------------------------------------------------------- 

No, I cannot reproduce these errors under Python 2.1.3, 2.2.3, 2.3.2
when using the recent cvs (only under 2.1 I get few 
fitpack,stats,special related errors that you have already reported).
I am currently using
  atlas2-sse2 3.2.1ln-11 from sid
  Numeric 23.1
  gcc 3.3.2

> I guess these failures are related to ATLAS or gcc, since 
> I was able to run scipy.test() on another system without 
> any failures. They differ in ATLAS and gcc. 

This could be the case. What gcc are you using?

> Any idea or hint ? 
>  
> Nils 
>  
> ATLAS version 3.4.2 built by root on Fre Sep 12 22:56:10 UTC 2003: 
>    UNAME    : Linux linux 2.4.20-4GB-athlon #1 Mon Mar 17 17:56:47 
> UTC 2003 i686 unknown unknown GNU/Linux 
>    INSTFLG  : 
>    MMDEF    : /home/nwagner/mysoftware/ATLAS/CONFIG/ARCHS/
> ATHLONSSE1/gcc/gemm 
>    ARCHDEF  : /home/nwagner/mysoftware/ATLAS/CONFIG/ARCHS/
> ATHLONSSE1/gcc/misc 

I am not sure but using sse1 might cause less accurate results.
Make sure that your are using the current CVS version of scipy_distutils
that disables sse1 flag when sse2 is available (do your machine have sse2
Run `python scipy_distutils/command/cpuinfo.py`).

I'll later try the other versions of atlas as well.

Pearu