LAPACK/ScaLAPACK Development

I have read a message cocerning DGER errors after DGESV
I have a similar problem when I want use DGESDD with jobz 'N' to compute only the singular values.

To do that I issue a call with LWORK = -1 to obtain the correct size of the WORK parameter.

The returned needed size is 1, So I create an array of the requested size and I issue a new call to DGESSD and get an error message from DGEBRD :

** On entry to DGEBRD parameter number 10 had an illegal value

parameter 10 is the size of the workspace required by DGEBRD internally i.e. WORK.

If I issue the second call to DGEBRD with a differnt correct size for WORK for example 64*(N+M), which is quite optimal for DGEBRD on my system,
everything works fine...

It is of course always possible that I have mistaken some way, but I think it can be a bug.

It seems to me that the returned size is incorrectly computed to 1 in the 'N' case not taking account of the internal needs of DGEBRD.

In the 'A' case everything looks to go well.

thanks for answers or hints

Did you apply the modification from 10/5/2000 available at:
http://www.netlib.org/lapack/release_notes.html
?

If no, can you apply the patch. Note that you can get all the modified file in one single tar file at:
http://www.netlib.org/lapack-dev/lapack--3.0--patch--10042002.tgz

If your version is the updated one, can you give the exact calling sequence, i.e. what are JOBZ, M, N ?

Julien

thank you,

I have not used the patches yet , in fact I work under a debian ubuntu breezy, and I wonder why there is no library update for bugs lasting from so long times ?

All the way, thanks a lot and I will try to apply all patches.