Take strides into account with numpy arrays

Numpy arrays are not necessarily contiguous... We need to take this into account when dealing with Numpy arrays if we use their internal C-arrays data.

Changed the signature of

cdef cnp.ndarray[cnp.double_t, ndim=1] multiply_ll_mat_with_numpy_vector(LLSparseMatrix A, cnp.ndarray[cnp.double_t, ndim=1] b):

to

cdef cnp.ndarray[cnp.double_t, ndim=1] multiply_ll_mat_with_numpy_vector(LLSparseMatrix A, cnp.ndarray[cnp.double_t, ndim=1, mode="c"] b):

The mode="c" test if numpy array is C-contiguous or not:

ValueError: ndarray is not C-contiguous

Added both cases:

• the numpy vector is non C-contiguous and
• the LLSparseMatrix is symmetrical

[dpo]

We should be able to multiply a matrix with a non-contiguous array. With the LL_Mat format, I don't think we can be more efficient if that numpy array is contiguous (with the CSR format, we can).

Done