[Pythonmac-SIG] Speed up Python on the Mac?

[Christopher Barker]

Louis Pecora wrote:

>>Although this will not help you know, I have been told, Pycho 
>><http://psyco.sourceforge.net/> will run on the Intel Chip.

> Thanks for the tips, Brian.  Wish I had that Intel Chip Mac Powerbook 
> now.  :-)

Even so, Psyco doesn't understand NumPy arrays, which I hope you're 
already using.

Essentially, you are right that you need to write some of your code in 
C, and that it's quite tedious to do it be hand. There are a number of 
tools that make it easier however. Here's my assessment, but I haven't 
actually used all of these:

SWIG -- Automatically creates wrapper for existing C/C++ code. A good 
choice if you are working with a n existing library, particularly if 
that library is under development (i.e. wxPython)

f2py -- similar to SWIG, but for Fortran code. Understands NumPy arrays. 
A good option if you have Fortran code you want to use, or like writing 
Fortran

SIP: similar to SWIG, used for PyQT, that's all I know.

Boost::Python -- C++ wrappers for the basic Python data types, handles 
reference counting etc. for you. I think there are some wrappers for 
NumPy arrays as well. A good choice if you want to write custom 
extension code in C++

Pyrex -- A python-like language for writing C extensions. It makes it 
very easy to write what is essentially a mixture of Python and C, with 
Python syntax. Easy to call C functions. Probably the easiest way to 
write custom extensions. Can work with NumPy arrays, but only at the C 
level -- working with a pointer to the data.

Psyco -- automatically translates pieces of your runtime code into 
machine language. Understands standard python data types, including the 
array.array type. Does not know about NumPy arrays (darn!). Can make 
major speed improvements for numeric code, but only if you use 
array.array or regular old lists and tuples. It's extremely easy to use, 
so why not try it on your code? Only available for x86 processors.

Weave.Blitz -- From the SciPy project. Automatically, and on the fly, 
translates and compiles individual NumPy expressions into C++ using the 
Blitz++ library. Probably the easiest way to speed up small sections of 
code.

Cxx -- kind of like Boost::Python. Is it still being maintained?

Have I got them all? I hope this helps.

Make sure to check out the SciPy docs, there is discussion of optimizing 
numerical code in various places there.

-Chris











-- 
Christopher Barker, Ph.D.
Oceanographer
                                      		
NOAA/OR&R/HAZMAT         (206) 526-6959   voice
7600 Sand Point Way NE   (206) 526-6329   fax
Seattle, WA  98115       (206) 526-6317   main reception

Chris.Barker at noaa.gov