[SciPy-user] Performance Python examples

David Powell davidanthonypowell@gmail....
Thu Jun 25 20:06:44 CDT 2009


Hi all,

I was looking at the code on the PerformancePython wiki page and I had
a few questions/comments:

Firstly, the f2py based example code does not build under windows.
For me it gives an error and asks me to use the -c mingw option, and
even if I do run setup.py with this option it still doesn't work.  Has
anyone else had this problem?

Secondly, I noticed that on my computer the "fast inline" option runs
a little bit slower than the "inline" version (not by much but it
seems consistent when re-running).

Finally, I have recently become aware of the package "numexpr" - it
seems to me that this would be a good candidate to add to this page.
Is anyone familiar with this package willing to add it, or should I
just go ahead and do it myself? (disclaimer: since I am unfamiliar
with this package I may not know how to use it properly so I might not
give a fair comparison)

The reason I am looking into this is because I am considering writing
some more computationally intensive code in python, and would like to
get a handle on the best way to speed up the key pieces of code.
Currently I am leaning towards weave.inline because it doesn't seem to
need much extra wrapping code, loops can often be parallelised on an
SMP machine quite easily using openmp directives and it is fast.  On
the other hand I seem to remember reading somewhere that it is not
really maintained at the moment - certainly its documentation in the
scipy online manual is almost nonexistant.  But I am happy to listen
to any advice on this matter.

cheers

David


More information about the SciPy-user mailing list