Package: openblas Version: 0.1.1-6 Severity: serious when using python-numpy with openblas set als libblas.so.3 provider it gives wrong results in some situations on i386. amd64 seems to work. It also works fine with libblas3 and libatlas. It seems to only occur when you change between double and single precision in a single script.
$ apt-get install python-numpy libopenblas-base $ update-alternatives --set libblas.so.3 /usr/lib/openblas-base/libopenblas.so.0 $ cat test.py import numpy.linalg as linalg from numpy import array, single, double, dot # removing any lines here makes it work again a = array([[1.,2.], [3.,4.]], dtype=double) b = array([2., 1.], dtype=double) x = linalg.solve(a, b) r = dot(a, x) a = array([[1.,2.], [3.,4.]], dtype=single) b = array([2., 1.], dtype=single) print dot(a, b) $ python ./test.py [ nan nan] $ update-alternatives --set libblas.so.3 /usr/lib/libblas/libblas.so.3 $ python ./test.py [ 4. 10.] $ update-alternatives --set libblas.so.3 /usr/lib/atlas-base/atlas/libblas.so.3 $ python ./test.py [ 4. 10.] -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
