[SciPy-user] Disable ATLAS

Pearu Peterson pearu at scipy.org
Sat Feb 21 14:09:09 CST 2004



On Sat, 21 Feb 2004, Nils Wagner wrote:

> Please can you run lu.py  

When I set

  force_clapack=0

then the output of lu.py is

---------------
X

[[ 1.28518872-0.09428482j -0.46708202-0.37104511j]
 [-0.31203795+0.19175932j  0.73496379-0.19732829j]]

inv(A)

[[ 0.91414361+0.3727972j  -0.46708202-0.37104511j]
 [-0.50936624-0.54320447j  0.73496379-0.19732829j]]

Identity ???

[[  1.00000000e+00 +2.90295132e-17j  -1.99086624e-17 -3.05202912e-17j]
 [  7.09610322e-17 +1.95427442e-16j   1.00000000e-00 +2.33103467e-18j]]

Identity ???

[[  1.00000000e+00 +1.72794721e-17j  -1.99086624e-17 -3.05202912e-17j]
 [ -2.36085023e-17 +1.00000000e+00j   1.00000000e-00 +2.33103467e-18j]]
--------------

Everything is correct up to round-off errors. So, check again that ATLAS 
is not being used. For the record, with force_lapack=1 the results are 
incorrect.

As for a possible Atlas bug, I haven't created a simple C program that
would demonstrate the bug and is necessary to file a bug report to Atlas
people. Any takers?

Pearu



More information about the SciPy-user mailing list