LAPACK Archives

[Lapack] possible error in lapack 3.4.0's gesvd

Thank you Clint for your bug report.
We actually had it already on our list as bug0085: but it was not fixed. Now it is...
I committed a quick fix (rev 1207), but we may look at a different way to 
compute workspace requirement within LAPACK. It will be mention on the next 
LAPACK meeting.

Regarding the comments, those are Doxygen comments. I know this does not 
enhance readability of the comments but I believe this remains readable by the 
Doxygen allows us to generate html documentation and nicer manpages.

On Dec 29, 2011, at 9:04 AM, Clint Whaley wrote:


I believe there is an error in 3.4.0's SVD, though perhaps you have
purposely changed the calling requirements?  I attach a test file that
an ATLAS user sent in; this test works if linked against lapack 3.3.1,
but fails if linked with lapack 3.4.0 with the reference BLAS (i.e.,
no ATLAS code at all).  See the full ATLAS tracker item at:

The actual error message 3.4.0 generates is:
  ** On entry to DORGLQ parameter number 5 had an illegal value

When I scope 3.4.0, I see this line that didn't exist in 3.3.1:
  CALL DORGLQ( N, N, M, VT, LDVT, DUM(1), DUM(1), -1, IERR )

However, from what a quick read tells me, the user is allowed to call
this routine with LDVT < N (1 in fact) if the JOB params are set to
'No", which appears to cause this error.

Finally, I suppose you are purposely uglying-up the Fortran comments
with all these latex commands?


** R. Clint Whaley, PhD ** Assist Prof, UTSA ** **
Lapack mailing list

<Prev in Thread] Current Thread [Next in Thread>
  • [Lapack] possible error in lapack 3.4.0's gesvd, julie langou <=

For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or