Hi Dave,
You're looking for an LAPACK quad prec. ?
There have been some progress on this made by Jim Demmel's group but
nothing has been released yet. Contact Jim directly, it might be able to
tell you more. There are also working on an LAPACK with arbitrary
precision.
If your user is just using a few of the LAPACK routines, it's pretty quick
and safe if he converts the routines himself. Basically, changing DOUBLE
PRECISION with REAL*16. I did it several times myself and that works fine.
The tricky part in Jim Demmel's work is not only to convert the whole
LAPACK package but also to make all the TESTING work. That's a completely
different story.
Best wishes,
Julien.
On Fri, 8 Feb 2008, David Cronk wrote:
We have a user that is interested in using an extended precision
version of LAPACK to run a code. The code uses REAL*16 and COMPLEX*32.
Is anyone aware if these is a version of LAPACK available which can take
advantage of extended precision REALs (i.e. REAL*16).
We are seeing mixed information in google, so I thought you folks might
be able to answe this question.
Thanks for any help you may be able to provide.
Dave Cronk
|