[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
John W. Eaton
Wed, 3 Nov 1999 15:05:24 -0600 (CST)
On 14-Oct-1999, Ross A. Lippert <address@hidden> wrote:
| Besides the double prompt, things are working fine and I am
| trying to push the bleeding edge a bit by updating the LAPACK
| routines to LAPACK3.0 routines.
| The difference between 3.0 and 2.0 are bug fixes and a few
| new functions that compute faster than the older functions
| (like divide and conquer SVD). Note: no functions have been
| deleted (though some of the aux functions were consolidated),
| so the interface to LAPACK has been extended, but should
| still be backward compatable.
| So I think the right way to proceed is to copy the corresponding
| LAPACK 3.0 routines into libcruft/lapack (+some minor changes
| due to aux function stuff), thereby updating each 2.0 LAPACK
| subroutine with a 3.0 subroutine, building that. Next, test the
| resulting octave. Next debate about whether to replace any of
| the algorithms (I vote strongly for the new divide and conquer
| Anyhow, phase 1 is complete. I have replaced the libcruft/lapack
| files with new ones and built it and it builds.
OK, I've updated my sources to use the Lapack 3.0 stuff. This change
uncovered a couple of problems in the way the schur function called
dgeesx (zgeesx for the complex version). I've fixed these problems
and checked them in to the CVS archive. There should also be at least
one new snapshot release sometime later this week.
|[Prev in Thread]
||[Next in Thread]|
- LAPACK 3.0,
John W. Eaton <=