Re: [Mplapack-devel] mpack - dggev?

[Maho N. <ch...@ma...>]

Hi Novak, thanks for your e-mail.

From: Novak Elliott <n.s...@cu...>
Subject: [Mplapack-devel] mpack - dggev?
Date: Tue, 23 Feb 2010 17:22:20 +0800

> I've played around with some code over the last week to see what the
> benefits of higher precision would be for my specific problem (wave
> dispersion in fluid/structures).  I've got a quad precision version of dggev
> working (crudely converted all 50 necessary subroutines to real*16 using a
> shell script) and for a progressively more ill-conditioned problem the extra
> precision does help, I'm able to get more accurate eigenvalues (compared to
> theoretical solution).

Great. 
I found fragments of codes by HIDA yozo.
http://www.eecs.berkeley.edu/~yozo/software/lapack-quad-30.tar.gz
http://www.eecs.berkeley.edu/~yozo/software/
He uses QD library and convert LAPACK routines.  

> However, although quad precision is better than double precision it appears
> that I could greatly benefit from much higher precision.

Yes.

> For reference, these are the subroutines required for *dggev*:
> *- blas:
> *daxpy.f
> dcopy.f
> ddot.f
> dgemm.f
> dgemv.f
> dger.f
> dnrm2.f
> drot.f
> dscal.f
> dswap.f
> dtrmm.f
> dtrmv.f
> idamax.f
> *- lapack:*
> ***dgeqr2.f
> ***dgeqrf.f
> ***dggbak.f
> ***dggbal.f
> ***dggev.f
> ***dgghrd.f
> ***dhgeqz.f
> ***dlabad.f
> ***dlacpy.f
> dladiv.f
> ***dlag2.f
> ***dlaln2.f
> dlamch.f
> ***dlange.f
> ***dlanhs.f
> dlapy2.f
> dlapy3.f
> dlarf.f
> dlarfb.f
> dlarfg.f
> dlarft.f
> dlartg.f
> dlascl.f
> dlaset.f
> dlassq.f
> ***dlasv2.f
> dorg2r.f
> dorgqr.f
> ***dorm2r.f
> ***dormqr.f
> ***dtgevc.f
> ***ieeeck.f
> ***ilaenv.f
> ***iparmq.f
> ***lsame.f
> ***xerbla.f
> 
> ***indicates a subroutine that has not yet been converted in MPACK (22
> routines)
> 
> Since your code is clearly laid out I'm thinking of having a go at this
> myself - if you have any comments/suggestions on which routines might be the
> more difficult then I'd be most appreciative.

Code matter: I'm very happy to accept your codes! 

Political matter: You can submit under two clause BSD style license.
Now I have been changing license to this one. Please refer following file.
http://mplapack.cvs.sourceforge.net/viewvc/*checkout*/mplapack/mpack/mblas/reference/Cher2k.cpp

> p.s.  I understand that Jim Demmel
> <http://www.eecs.berkeley.edu/%7Edemmel/>is also working on a
> multiprecision version of lapack but has not released
> any code.  Do you know anything about this?

No, I don't know. Only some rumors.

Many thanks!
-- Nakata Maho http://accc.riken.jp/maho/ , http://ja.openoffice.org/ 
   Nakata Maho's PGP public keys: http://accc.riken.jp/maho/maho.pgp.txt