[PYTHON MATRIX-SIG] Poll - Lapack and FFTPack

Jim Hugunin hugunin@mit.edu
Thu, 15 Aug 1996 14:45:43 -0400

I've discovered once again that building Lapack on a system without a
FORTRAN compiler is a lot of ugly work, and you get something
significantly slower than the "real" version in the bargain.

Nonetheless, I've managed to convince myself that LAPACK is clearly the
right way to go as far a standard linear algebra library is concerned
for NumPy.  Doug Heisterkamp's (and Konrad's) lapack interface module
covers most of what people have been asking for in a standard library.

Possible solution:

1) Everybody buys a FORTRAN compiler ;)

2) Everyone goes through a hellish day of porting to get lapack compiled
on their system

3) I distribute precompilied lapack and fftpack libraries (with the
blas) for ALL the common architectures.

So, here's the poll:

If you have a way to produce lapack/fftpack library binaries for a given
platform and are able and willing to allow those binaries to be freely
distributed, please send me a piece of mail listing your computer/os and
what libraries you could provide.

I'm assuming here that if you give me a statically linked lapack
library, linked against libblas and libf77, then I can link this into my
C module with no problem.  If this isn't true, please help me understand
this better.

A little bit at my wits end over the whole standard library business...

- Jim

MATRIX-SIG  - SIG on Matrix Math for Python

send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org