I downloaded lapack recently and wanted to put a dent in it by calling one
of its subroutines. I hoped to call one that would determine the
eigenvalues of a real matrix, but that doesn't seem to be on the menu. A
person seems to need to put quite a bit of effort to determine exactly what
the subroutines do, and deciphering the titles is one step on that path.

I found SGEHRD.f, where I think the S is single precision, the GE is a
general matrix, the H is Hessenberg and RD might be reduction. So I'll be
hoping to take a eal matrix and use lapack to give me the Hessenberg form,
which, I can then use for bigger and better things.

With the fortran, so far I've got:

implicit none
integer, parameter ::n = 5
integer, dimension(n,n):: a
integer, dimension(n,n):: b

On Sep 18, 9:37 pm, Ron Ford wrote:

[snip]

gfortran (like g95) will compile both types of source from the same
command line:

On Thu, 18 Sep 2008 22:03:46 -0700 (PDT), e p chandler posted:

I hadn't noticed that there's a half dozen calls to other routines, not all
of which are in the same folder.

EXTERNAL SAXPY, SGEHD2, SGEMM, SLAHR2, SLARFB, STRMM,
$ XERBLA

We'll see if it cascades much farther. I suppose I could put them all in a
module. It would be about 30 megs, 1500 files: yikes. I think I'm finding
out why NAG, for example, makes an f08 version of this.

I'll just keep muddling. Thanks, elliot.

On Thu, 18 Sep 2008 22:03:46 -0700 (PDT), e p chandler posted:

I've gone through the files and collected the ones I think I need. It was
nothing short of tedious:

C:\Program Files\University Of Tennessee\LAPACK 3.1.1\src\depend>dir
Volume in drive C has no label.
Volume Serial Number is 486B-CFF3

Directory of C:\Program Files\University Of Tennessee\LAPACK
3.1.1\src\depend

09/19/2008 12:06 AM .
09/19/2008 12:06 AM ..
07/19/2007 11:02 AM 1,496 saxpy.f
07/19/2007 11:02 AM 18,082 slarfb.f
07/19/2007 11:02 AM ...

I prefer to compile LAPACK and create a library. I then, create a
module with interfaces to LAPACK subroutines. For instance I have
created a function for the eigenvalues and eigenvectors of a real or a
real symmetric matrix. So when I want to calculate, say, the
eigenvectors of a real matrix A I type v=eigVectors(A), or if I want
the eigenvectors of a symmetric matrix A I type
v=eigVectors(A,sym="Y"). In this way the cumbersome argument list of
the LAPACK routines does not appear, or have to be used anywhere.

Michael

On 2008-09-19, Ron Ford wrote:

If you want an easier to use Fortran 95 lapack interface there's always
LAPACK95:

http://www.netlib.org/lapack95/

As others explained, that's no problem. Though LAPACK95 is nice in that
it provides explicit interfaces, that among other things helps protect
the programmer from his own fumbling.

On Sep 19, 3:09 am, Ron Ford wrote:

Den Thu, 18 Sep 2008 19:37:28 -0600 skrev Ron Ford:

Hi,

Lapack has both driver routines and expert routines. If you just want to
get started with for example eigenvalue systems then drivers like DGEEV
or DSYEV are a good choice. They will do the transformations to
triangular form and other stuff for you.

Ron Ford wrote:

On Fri, 19 Sep 2008 15:22:51 -0400, user1 posted:

That's a good link. I bookmarked it.

I think I've got the upper hand on it now. I can paste them into a module
that's written in fixed form by Plato III. The IDE handles all the spacing
madness. I added a contains statement and "end subroutine" where Plato
says I need to elaborate on the end statement in this context.

I chose a routine that has no further dependencies, and it compiles:

module routines_from_lapack
contains
SUBROUTINE SAXPY(N,SA,SX,INCX,SY,INCY)
* .. Scalar Arguments ..
REAL SA
INTEGER INCX,INCY,N
* ..
* .. Array Arguments...