Arbitrary Precision LAPACK

[kdallen]

Is there any plan to release an arbitrary precision version (Fortran) of LAPACK?

[yurii_]

And what precision is necessary to you?

[kdallen]

I found the following dated from 2008 and was wondering if the project ever completed.

We are currently working on a multiple precision version of LAPACK,
and hope to release it within a year. You are welcome to take LAPACK
and modify it for your own purposes, under the guidelines of the
mBSD software license.

Jim Demmel

[Julien Langou]

Hello Ken,

I think the effort Jim was speaking about is now stopped unfortunately.

Two things:
1) we still have this in our mind and are seriously thinking about it, but there is nothing concrete yet.
2) A relevant software package is: http://sourceforge.net/projects/mplapack/

Julien.

[kdallen]

Julien,

Thanks for your reply. I found MPACK in searches but would prefer a 100% Fortran solution. I've actually modified the routines for DGELS to be compatible with

http://myweb.lmu.edu/dmsmith/fmlib.html

and seem to have gotten it to work, but it's quite a bit of effort.

[yurii_]

Is there any plan to release an arbitrary precision version (Fortran) of LAPACK?

It is possible to receive fourfold accuracy easily. The compiler of Intel supports it. But it is necessary to change code LAPACK a little. The program with fourfold accuracy will slowly work.

[CyLith]

I have (had) been working on translating LAPACK into templated C++ code, templating on the number type. I actually converted the entire computation chain for unblocked ZGEEV (but not entirely templated), and you can get it here: https://github.com/victorliu/RNP. I would be incredibly interested in having the entirety of LAPACK translated this way since we could then plug in whatever number types we want, as well as being able to more easily interface it with existing C/C++ code.