sergey-v-kuznetsov

I'm not convinced at all. This is certainly a bug. The norm of initial vector could be any especially in general eigenvalue problems. If everything is ok for random data...

Do you suggest to say the users that their matrix is not good if he gets wrong results??

I didn’t also like both lartg implementations and I just stated that our tests are passed with the Lapack 3.9 lartg. We had two types of failures and unfortunately the...

I also forgot to mention about another minor problem which also causes test failures. Some our old test uses a precomputed singular vectors or precomputed eigenvectors and this kind of...

> "Some our old test uses a precomputed singular vectors or precomputed eigenvectors and this kind of tests is considered as passed if and only if || computed eigen/singular vector...

Please find below my answers to your questions > For the 600x600 matrix, These dimensions (e.g. 600 by 600) are for the DGESDD test and DGESDD is a LAPACK routine....