LAPACK/ScaLAPACK Development

I am a sysadmin, but haven't written any fortran code since the mid-80's. In fact, up until this morning, I had never heard for LAPACK... until one of the profs in my department asked me to install it on our "general use Linux machine".

Background: running Debian 5 w/ gfortran already installed.

So anyway, I downloaded the lapack-3.2.1 file, gunziped & un-tar'd it. It is currently sitting in /root and I am doing all this logged in as root. I copied the make.inc.example to make.inc, then looked at it to see if there were any mods I needed to make (I didn't see any). Not seeing any mods that jumped out at me (admittedly, not knowing what I was looking at), I went ahead and ran

make


Q1: Am I done? Does this put these libraries in a location where everyone can get to them?

Q2: How does a user call them?

Q1: Am I done? Does this put these libraries in a location where everyone can get to them?

Yes. LAPACK is also distributed in the form of Pakages by Debian.
You may want to rename the library to liblapack.a so that users can link with only -lllapack

Q2: How does a user call them?

He will call a LAPACK routine from his Fortran or C program.
He will link his program with the -llapack.
You user needs also an optimized BLAS not provided in the package. For a Linux Machine, I would recommend ATLAS or GOTO BLAS.
See http://en.wikipedia.org/wiki/Basic_Line ... ubprograms

I would think the original poster would be better off installing lapack
(with ATLAS BLAS, if preferred) via Debian's package management system
(apt-get).

On whether the reference BLAS should or should not be used: I'd say that
depends on the typical size of the matrix problems that the users will be
throwing at lapack. ATLAS offers a very impressive speed improvement for
large problems, but its overhead can slow computation down for moderate-
sized data(assuming the reference BLAS is compiled with a good modern
compiler with reasonable optimization flags).

There's a slightly dated discussion of this topic at
http://ricardo.ecn.wfu.edu/~cottrell/tm ... speed.html

Allin Cottrell