I have a large Python method that calculates a series of ndarray quantities of dtype=np.complex128. I have been experimenting with what speed increases are possible using Cython. In converting a Python method to Cython, there is the thought about using NumPy calculations vs using C libraries to do it. Exploring the latter, this is a watered-down Cython method for computing the complex exponential of an input NumPy array, x: import numpy as np cimport numpy as cnp cnp.import_array() cdef extern from "complex.h": double complex cexp(double complex z) DTYPEC = np.complex128 ctypedef cnp.complex128_t DTYPEC_t cpdef cnp.ndarray[DTYPEC_t, ndim=1] stupid_exp(cnp.ndarray[DTYPEC_t, ndim=1] x): cdef int num_pts, j cdef double complex tmp num_pts = len(x) cdef cnp.ndarray[DTYPEC_t, ndim=1] result = np.zeros(num_pts, dtype=DTYPEC) for j in range(num_pts): result[j] = cexp(x[j]) return result On my Windows 10 machine, compiled with Visual Studio 2022, the result[j] = cexp(x[j]) line creates a pair of errors: error C2440: 'function': cannot convert from '__pyx_t_double_complex' to '_Dcomplex' error C2440: '=': cannot convert from '_Dcomplex' to '__pyx_t_double_complex' This says that NumPy complex128 and C's double complex are not being seen as the same data type. A LLM suggested explicit type conversion: tmp = <double complex>x[j] tmp = cexp(tmp) result[j] = <DTYPEC_t>tmp but this doesn't fix the errors. Any thoughts about how I could get this simple code to compile correctly in Cython? Continue reading...
