[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Idea for an IDL numerical toolbox
- Subject: Idea for an IDL numerical toolbox
- From: Liam Gumley <Liam.Gumley(at)ssec.wisc.edu>
- Date: Mon, 12 Apr 1999 14:35:55 -0500
- Newsgroups: comp.lang.idl-pvwave
- Organization: Space Science and Engineering Center, University of Wisconsin-Madison
- Xref: news.doit.wisc.edu comp.lang.idl-pvwave:14310
In light of some recent discussions about calling FORTRAN routines from
within IDL, I started thinking about how one might go about putting
together an IDL toolbox based on a well-known freely available FORTRAN
library such as LAPACK (http://www.netlib.org/lapack).
One reason to do this would be to extend the numeric capabilities of IDL
by making use of a very well tested FORTRAN library that is available in
optimized form for many architectures, e.g.
and is also freely available in source form (see LAPACK link above).
Another reason would be to help combat a view I've heard from some
Matlab users, which is that IDL is not well suited to intensive
Some items of information that I think could be relevant:
(1) It seems like there would be a fair amount of work involved in
writing the C wrappers and IDL interface routines for a large FORTRAN
library. Could this be automated somehow? NAG markets something called
the NAGWare Gateway Generator, which is described at
http://www.nag.co.uk/nagware/NN.html. Here's a quote:
"The NAGWare Gateway Generator (NGG) produces an interface that
seamlessly integrates Fortran subprograms with MATLAB. It analyses the
source code of a Fortran routine and automatically generates a MATLAB
gateway. This allows the user to access the Fortran subprogram as if it
were any other MATLAB command. Both forward and reverse gateways can be
generated. Reverse gateways allow the Fortran program to access MATLAB
commands and scripts."
There is a very nice freely available FORTRAN analyzer named ftnchek,
available at http://www.dsm.fordham.edu/~ftnchek which will
automatically extract information about the arguments of a FORTRAN
subroutine. Using this tool it shouldn't be too difficult to build a
script that automatically creates a C wrapper and IDL interface routine
for a FORTRAN subroutine.
(2) There are a number of well documented and tested freely available
FORTRAN numerical libraries built around LAPACK, one of which is "The
Control and Systems Library SLICOT", available at
(3) Some work would be needed to make the compiling options work
seamlessly on all IDL platforms (Unix at least). Right now it seems like
the first time someone tries to link IDL with FORTRAN, there's a steep
learning curve. I'm thinking of a package that can be downloaded,
installed once, and then used directly from IDL without the user having
to worry about how it was compiled or linked. The package would be smart
enough to check (at installation time?) for the presence of optimized
LAPACK on the host architecture, and if it was not present, LAPACK would
be compiled from source.
Sounds like a nice cooperative project for the readers of
Any comments? Volunteers?
Liam E. Gumley
Space Science and Engineering Center, UW-Madison