Hi LAPACK developers,
I have encountered a test matrix for which dgesdd fail to converge. The returned INFO is 1, not 0. The test matrix has size 1497x1564. See attached program for reproduction.
Intel MKL 10.3.11 converges for this matrix. However, their latest version 11.0 also fails to converge.
Duncan Po
MathWorks
dgesdd fails to converge
2 posts
• Page 1 of 1
dgesdd fails to converge
by duncanpo » Tue Mar 12, 2013 3:41 pm
- Attachments
-
inteldgesdd.tar.gz- (182.36 KiB) Downloaded 198 times
- duncanpo
- Posts: 4
- Joined: Fri Apr 01, 2011 5:09 pm
Re: dgesdd fails to converge
by rodney » Tue Mar 12, 2013 7:43 pm
Duncan,
Thanks for reporting this. I am not exactly sure what is causing the problem, but it goes all the way back to version 3.2.2 (the routine converges in 3.2.1). Also, it seems very sensitive to the particular BLAS library used; I was able to get convergence in later versions of LAPACK using a third party BLAS library some of the time. As soon as we have a fix for this I will post it to the forum and commit the patch to the svn repository.
Regards,
Rodney James
Thanks for reporting this. I am not exactly sure what is causing the problem, but it goes all the way back to version 3.2.2 (the routine converges in 3.2.1). Also, it seems very sensitive to the particular BLAS library used; I was able to get convergence in later versions of LAPACK using a third party BLAS library some of the time. As soon as we have a fix for this I will post it to the forum and commit the patch to the svn repository.
Regards,
Rodney James
- rodney
- Posts: 49
- Joined: Thu Feb 10, 2011 8:20 pm
- Location: Colorado College
2 posts
• Page 1 of 1
Who is online
Users browsing this forum: No registered users and 1 guest
