[SciPy-user] installation on Suse8.1 (blas problem)

[Pearu Peterson]

On Mon, 17 Mar 2003 baecker at physik.tu-dresden.de wrote:

> I just did a fresh checkout I get the following
> 
> ======================================================================
> FAIL: check_nrm2 (test_blas.test_fblas1_simple)
> ----------------------------------------------------------------------
> Traceback (most recent call last):
>   File
> "/home/baecker/PYTHON/lib/python2.2/site-packages/scipy/linalg/tests/test_blas.py",
> line 90, in check_nrm2
>     assert_almost_equal(f([3,-4,5]),math.sqrt(50))
>   File
> "/home/baecker/PYTHON/lib/python2.2/site-packages/scipy_test/testing.py",
> line 349, in assert_almost_equal
>     assert round(abs(desired - actual),decimal) == 0, msg
> AssertionError:
> Items are not equal:
> DESIRED: 7.0710678118654755
> ACTUAL: 7.0710676908493042
> 
> ----------------------------------------------------------------------
> Hmm, the numbers are not that unequal, but still -
> what do you think ?
> (I haven't seen this error before...)

Here I get

>>> scipy.linalg.fblas.dnrm2([3,-4,5])
7.0710678118654755
>>> scipy.linalg.fblas.snrm2([3,-4,5])
7.0710678100585938

What Fortran compiler flags were used to compile the blas library? 
Note that using 3dnow instructions may cause differences in floating
point results but I never thought it would be so noticable.

> In any case, I should emphasize, that I just took the example
> which was there before and it very well might be that
> some of the "ssymv strmv ctrmv" and some others could be included,
> i.e. excluded from the above list.

If any of those symbols would be missing in the exclude list, importing
linalg would just fail. So, I think only these 'rot' routines are
missing in your blas library.

Pearu