sergey-v-kuznetsov

accuracy problems with the new LAPACK 3.10 *LARTG implementations

17

**Description** Under certain conditions the new *LARTG implementations can essentially increase the 2-norm of initial vector. It can lead to significant loss of accuracy of singular vectors or eigenvectors in...

I'm attaching a Fortran code which demonstrates the problem. The code reads two diagonal matrix from a file in the binary format and call DBDSDC: CALL DBDSDC( 'U', 'I', N,...