Differences between revisions 13 and 14
|Deletions are marked like this.||Additions are marked like this.|
|Line 6:||Line 6:|
|svn co http://svn.scipy.org/svn/scipy/trunk/Lib/weave weave||svn co http://svn.scipy.org/svn/scipy/trunk/scipy/weave weave|
The weave package allows the inclusion of C/C++ within Python code and is useful in accelerating Python code.
Weave is a subpackage of scipy. (I.e. you have it already if you installed SciPy)
- Alternatively, you can check-out and install weave separately using
svn co http://svn.scipy.org/svn/scipy/trunk/scipy/weave weave cd weave sudo python setup.py install
PerformancePython: A comparison of various ways to improve the performance of Python code using Numeric, weave, Pyrex, Psyco and Fortran (f2py) for solving Laplace's equation. These are compared with code written in C++.
Cookbook/Weave Some cookbook examples of using low level Numpy C-API
If you have scipy installed, weave includes several examples here:
And the above tutorial is on your installation also:
To find where your site-packages directory holding scipy is, run this python command in a terminal:
python -c "from scipy import weave; print weave.__path__"
some random notes
when is code compiled
a) Is it possible to distribute modules using weave to other people who might NOT have a C compiler installed ? b) when I (or someone who does not have a C compiler !) change parts of that module that should not require a recompiling of the C part - is weave smart enough to recognise this ?
> It's my understanding that a re-compile is triggered by a mismatch to a > MD5 generated on the C string that is to be compiled and the cached MD5 > for the expression. This would mean that only changes to that string > would force a re-compile. However, even formatting changes (even to > whitespace) in the C string force a recompile. > The types of the inputs are also taken into account. > And (to be pedantic :) ) the version the numpy API.
how to distribute code to people w/o a C compiler
You can make weave generate an extension module of your choosing. Look at examples/fibonacci.py which builds a fibonacci_ext.cpp/fibonacci_ext.so pair in the current directory.
Weave and Numpy array arguments
>> if I pass a numpy array 'arr' as argument >> a) how does the C code get arr.ndim ? >> b) how does the C code get arr.shape,... ? >> c) if the C code changes elements of arr, are those changes *on the >> original data* ? Yes. >> In other words, is weave.inline making a copy of arr ? No. >> I searched through >> http://projects.scipy.org/scipy/scipy/browser/trunk/scipy/weave/doc/tutorial. >> html?format=raw but did not find a definite answer. From >> the 'array3d.py' example in weave in looks like Narr would contain the >> shape !? Yes. Specifically: arr_array is the actual PyArrayObject* corresponding to the Python object. Narr = arr_array->dimensions Sarr = arr_array->strides Darr = arr_array->nd arr = arr_array->data > > Oh, and I forgot: How about non-contiguous arrays !? Passed straight on through, just like contiguous arrays. >> In the case that these are handled - does that slow things down for proper >> aligned arrays, too !? You will have to take the strides into account in your code.