http://bugzilla.novell.com/show_bug.cgi?id=569978 http://bugzilla.novell.com/show_bug.cgi?id=569978#c0 Summary: The 64 bit version of the BLAS package returns wrong results (seems to be compiled with bad compiler options) Classification: openSUSE Product: openSUSE 11.2 Version: Final Platform: x86-64 OS/Version: openSUSE 11.2 Status: NEW Severity: Normal Priority: P5 - None Component: Other AssignedTo: bnc-team-screening@forge.provo.novell.com ReportedBy: bergeest@googlemail.com QAContact: qa@suse.de Found By: --- Blocker: --- Created an attachment (id=336157) --> (http://bugzilla.novell.com/attachment.cgi?id=336157) Octave script to reproduce the error. User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.1.7) Gecko/20091222 SUSE/3.5.7-2.4 Firefox/3.5.7 The svd command in octave returns wrong results. Also the library ITPP which depends on lapack and uses the svd (which depends on blas) returns the same wrong result. I compiled lapack by hand (without any optimizations) using the suse blas package and the error still remained. Afterwards I compiled lapack and blas (without any optimizations) and now octave (installed without installing the suse blas and lapack packages) and the ITPP lib return the right svd results. This could be reproduced on another Suse 11.2 x86_64 system but not on for example a Fedora 12 64 bit system that uses the same lapack and blas versions (see: http://sourceforge.net/projects/itpp/forums/forum/115656/topic/3510175 and http://www-old.cae.wisc.edu/pipermail/bug-octave/2010-January/010140.html). Reproducible: Always Steps to Reproduce: 1. enter the directory where the attached file is saved 2. start octave 3. svdTest Actual Results: The result is: norm(A - U*S*V') with [U S V] = svd(A) ans = 3.1069 Expected Results: The right solution is something close to zero: norm(A - U*S*V') with [U S V] = svd(A) ans = 2.2026e-15 -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.