The default behaviour of np.zeros is to create a float64 data, if you want to force the datatype to be float32 you should specify it. You should also put all cdefs in the beginning of the function.