apply/findTransformation

hinsen@cnrs-orleans.fr hinsen@cnrs-orleans.fr
Tue, 9 Jun 1998 12:12:55 -0500 (CDT)


> profiling my MMTK-based code, I found that for the particular thing I'm
> doing, I used up most of the runtime in the findTransformation; even more
> for the applyTransformation call. I'm using this on a protein of ~3000
> atoms. As this is just the beginning, I wondered whether anybody had
> written a C version of the respective Collection.py methods
> (find/applyTransformation) ? With things like LAPACK++, it shouldn't be a
> big deal (BTW: Has anybody already linked C++ as a Python extensions?
> Where's the gotchas?). If nobody has, I might give it a try...

There is a whole SIG dealing with C++ integration; and LLNL has
some supporting software.

As for findTransformation, it already uses LAPACK, but that part
of the calculation (diagonalization of a 4x4 matrix) is trivial.
All the time is spent in looping over the atoms and constructing
that matrix.

Moving that code completely to C is certainly possible, but I am not
sure how much of a speedup it will provide. Part of the work involves
accessing Python data structures, which is not a lot faster in C.
But it's definitely worth a try.

Konrad.
-- 
-------------------------------------------------------------------------------
Konrad Hinsen                            | E-Mail: hinsen@cnrs-orleans.fr
Centre de Biophysique Moleculaire (CNRS) | Tel.: +33-2.38.25.55.69
Rue Charles Sadron                       | Fax:  +33-2.38.63.15.17
45071 Orleans Cedex 2                    | Deutsch/Esperanto/English/
France                                   | Nederlands/Francais
-------------------------------------------------------------------------------