T J wrote:
> Hi,
>
> I want to call x[start:stop:skip].sum() on a NumPy array within cython
> code. The problem with doing this is that the call is a Python call,
> which is handled in C, whose result is converted into a Python float.
> So, I'd have to convert this Python float back into a double. So it
> seems best if I just wrote the summation manually. I give it a NumPy
>
The conversion to float isn't the problem, that's almost free compared
to taking that x[start:stop:skip] call and doing the sum() itself.
I expect you are doing lots and lots of really small summations?
Otherwise there's no real point.
> array, a start index, stop index, and a skip index. I can write the
> loop, but I am not sure how I should define the cdef function
> signature.
>
> Can someone help me out? I need something like:
>
> cdef double sum(np.ndarray[np.float64_t, ndim=1] x, unsigned int
> start, unsigned int stop, int skip):
>
You unfortunately need to do
cdef double sum(object x, unsigned int
start, unsigned int stop, int skip):
cdef np.ndarray[np.float64_t, ndim=1] x_buf = x
then use x_buf. There's a slight speed penalty with that assignment which
Also, please use Py_ssize_t instead of unsigned int. unsigned ints are
dangerous in Cython.
> cdef unsigned int i
> cdef double result
>
You need to initialize result to 0.
Dag Sverre
_______________________________________________
Cython-dev mailing list
[email protected]
http://codespeak.net/mailman/listinfo/cython-dev
