Check this code

[Warren Weckesser]

I'm learning about the array syntax and fused types.  The goal of the following code is to provide a function that sums a 1-d numpy array of type int, float or complex (numpy's np.int32, np.float64 and np.complex128, resp.).  Is there a better or more idiomatic way to do this with cython 0.16?

-----
import numpy as _np
cimport cython


ctypedef fused my_type:
     cython.int
     cython.double
     cython.complex


cdef my_type _sum(my_type[:] a):
    cdef int n = a.shape[0]
    cdef unsigned i
    cdef my_type total

    total = 0
    for i in range(n):
        total += a[i]

    return total


def sum(a):
    """a must be a 1-d numpy array of type int32, float64 or complex128."""

    if a.dtype == _np.float64:
        result = _sum[cython.double](a)
    elif a.dtype == _np.int32:
        result = _sum[cython.int](a)
    elif a.dtype == _np.complex128:
        result = _sum[cython.complex](a)
    else:
        raise ValueError('data type not implemented')

    return result
-----


Warren

[mark florisson]

I think the topic of the thread could be improved :) Anyway, currently
with 0.16 it does not automatically dispatch buffers or numpy arrays.
You can try out this branch though:
https://github.com/markflorisson88/cython/tree/_fused_dispatch_rebased
, which does it automatically for you.

[Warren Weckesser]

Sorry, I guess that was a bit terse.

 

Anyway, currently
with 0.16 it does not automatically dispatch buffers or numpy arrays.
You can try out this branch though:
https://github.com/markflorisson88/cython/tree/_fused_dispatch_rebased
, which does it automatically for you.

OK, thanks.  For 0.16, is the dispatch function in the example a reasonable approach?  It works for me, and it is far better than writing three almost identical functions or using a separate templating technique, so I'm happy with it.  I'm just trying to make sure that I'm not missing any additional techniques or tricks that could be used (while sticking to the released version).

Warren

[mark florisson]

Yeah, there's not really any better way to do it. The only other way
would be to pass in a scalar of the same type as the dtype, but that
will likely fail in the face of promotion (i.e. it won't know whether
you want the 'float' or 'double' C type given a Python float object).

[Warren Weckesser]

OK, thanks Mark.

Warren