Re: [Mplapack-devel] converted subroutines - complete / testing / in development?

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

Hello Novak,

I just came back from US. Time has passed, if something is missing
please do not hesitate to ask me again.

From: Novak Elliott <n.s...@cu...>
Subject: [Mplapack-devel] converted subroutines - complete / testing / in development?
Date: Thu, 25 Feb 2010 18:44:41 +0800

> I've started having a go at converting some LAPACK routines to MPACK and
> then I discovered that most of the routines I require already exist in the
> mlapack/library/ directory, even though
right. just converted from FORTRAN by f2c.

> - they are not listed on the web page "LIST OF MLAPACK ROUTINES and LAPACK
> CORRESPONDENCE" (http://mplapack.sourceforge.net/mlapack_routines.html)
> - their prototypes are listed as "/* not yet done ... auxiliary routines */"
> in include/mlapack_gmp.h

> dlacpy.f => Rlacpy.cpp Exists
> dlanhs.f => Rlanhs.cpp Exists
> dlange.f => Rlange.cpp Exists
> dgeqr2.f => Rgeqr2.cpp Exists
> dorm2r.f => Rorm2r.cpp Exists
> dgeqrf.f => Rgeqrf.cpp Exists
> dggbak.f => Rggbak.cpp Exists
> dlasv2.f => Rlasv2.cpp Exists
> dormqr.f => Rormqr.cpp Exists
> dgghrd.f => Rgghrd.cpp Exists
> dlag2.f  => Rlag2.cpp Exists
> dggbal.f => Rggbal.cpp Exists
> dggev.f  => Rggev.cpp Exists
> dlaln2.f => Rlaln2.cpp Exists
> dtgevc.f => Rtgevc.cpp Exists
> dhgeqz.f => Rhgeqz.cpp Exists
> What is the status of these routines - complete / still need testing / under
> development?
> And their complex equivalents?

Status of them; just compiles. so under development is correct.

> Also, are these routines simply yet to be converted:
> dlabad.f
> ieeeck.f
> iparmq.f
> ilaenv.f
> ?

ieeeck.f is for exception handling can be made so 
it requies more efforts. just ignore it at the moment.
dlabad.f is overflow check for the large exponent.
I think we need it for qd/dd, but if you are using GMP, 
exponent range is extraordinary large, so we can simply ignore
this. also, for GMP, size of exponent range varies by architecture;
if you use a 64 bit machine, then 2^64. on a 32bit machine, it is 2^32.

ilaenv may be the same as lapack's. since lapack's ilaenv does not
return an optimized value, so we can simply set it as the same as the lapack's.
In any case, we should carefully optimize... 

I forgot about iparmq...

Best regards,
-- 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