np_to_af_array copy #172

selimyoussry · 2018-01-18T02:15:08Z

Hi everyone, the source code for "np_to_af_array" seems guaranteed to raise a RuntimeError("Copy can not be False for numpy arrays"), shouldn't it only be present in the "else" statement?

I'm thinking of this:

def np_to_af_array(np_arr, copy=True):
        in_shape = np_arr.shape
        in_ptr = np_arr.ctypes.data_as(c_void_ptr_t)
        in_dtype = _nptype_to_aftype[np_arr.dtype.str[1:]]

        if (np_arr.flags['F_CONTIGUOUS']):
            return _fc_to_af_array(in_ptr, in_shape, in_dtype)
        elif (np_arr.flags['C_CONTIGUOUS']):
            return _cc_to_af_array(in_ptr, np_arr.ndim, in_shape, in_dtype)
        else:
            if not copy:
                raise RuntimeError("Copy cannot be false if the data is not a single, Fortran-style (F_CONTIGUOUS flag) or C-style (C_CONTIGUOUS flag) contiguous segment")
            return np_to_af_array(np_arr.copy())

instead of the current code

What do you think?

Thanks!

pavanky · 2018-01-18T02:20:09Z

@selimyoussry copy should never be false. You can send in a PR to remove that option if you want to.

arrayfire / arrayfire-python

np_to_af_array copy #172

np_to_af_array copy #172

selimyoussry commented Jan 18, 2018

pavanky commented Jan 18, 2018

arrayfire / arrayfire-python

np_to_af_array copy #172

np_to_af_array copy #172

Comments

selimyoussry commented Jan 18, 2018

pavanky commented Jan 18, 2018