[PYTHON MATRIX-SIG] linear algebra

[Konrad HINSEN]

> Transposition costs almost nothing.  Transposition followed by a copy is
> O(n).  These FORTRAN routines expect contiguous arrays, which requires
> the transposed array to be copied before calling them.  Therefore this
> is a non-trivial amount of computation.

Fortran routines rarely require contiguous arrays, and LAPACK routines
definitely don't. It's just a matter of correct interfacing, but
this would have to be done at the C level, since on the Python level
the necessary data about the memory layout is no longer available.

The current low-level LAPACK interface expects Fortran-layout
arrays, but one could as well have one that handles the "transposition"
itself, although that would be a more difficult task. So maybe we
should leave everything as it is for now and improve the interface
later.

> I agree that python users should not have to know anything about
> implementation details.  I'm just wondering why the transposed order is
> more natural than the native one.

No more natural, but conventional.

Konrad.

-------------------------------------------------------------------------------
Konrad Hinsen                     | E-Mail: hinsenk@ere.umontreal.ca
Departement de chimie             | Tel.: +1-514-343-6111 ext. 3953
Universite de Montreal            | Fax:  +1-514-343-7586
C.P. 6128, succ. Centre-Ville     | Deutsch/Esperanto/English/Nederlands/
Montreal (QC) H3C 3J7             | Francais (phase experimentale)
-------------------------------------------------------------------------------

=================
MATRIX-SIG  - SIG on Matrix Math for Python

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