ENH: Add complex number support for fromfile

[zjpoh]

As discussed in #14227, we want to support complex numbers for fromfile too.

[zjpoh]

I got this error on Azure Windows Python 3.7 32 bit build but I don't understand the failure.

self = <numpy.core.tests.test_multiarray.TestMethods object at 0x1256D090>
func = <ufunc 'matmul'>, dtype = 'F'

    @pytest.mark.parametrize('func', (np.dot, np.matmul))
    @pytest.mark.parametrize('dtype', 'ifdFD')
    def test_no_dgemv(self, func, dtype):
        # check vector arg for contiguous before gemv
        # gh-12156
        a = np.arange(8.0, dtype=dtype).reshape(2, 4)
        b = np.broadcast_to(1., (4, 1))
        ret1 = func(a, b)
        ret2 = func(a, b.copy())
        assert_equal(ret1, ret2)
    
        ret1 = func(b.T, a.T)
        ret2 = func(b.T.copy(), a.T)
        assert_equal(ret1, ret2)
    
        # check for unaligned data
        dt = np.dtype(dtype)
        a = np.zeros(8 * dt.itemsize // 2 + 1, dtype='int16')[1:].view(dtype)
        a = a.reshape(2, 4)
        b = a[0]
        # make sure it is not aligned
        assert_(a.__array_interface__['data'][0] % dt.itemsize != 0)
        ret1 = func(a, b)
>       ret2 = func(a.copy(), b.copy())
E       RuntimeWarning: invalid value encountered in matmul

a          = array([[0.+0.j, 0.+0.j, 0.+0.j, 0.+0.j],
       [0.+0.j, 0.+0.j, 0.+0.j, 0.+0.j]], dtype=complex64)
b          = array([0.+0.j, 0.+0.j, 0.+0.j, 0.+0.j], dtype=complex64)
dt         = dtype('complex64')
dtype      = 'F'
func       = <ufunc 'matmul'>
ret1       = array([0.+0.j, 0.+0.j], dtype=complex64)
ret2       = array([[ 6.+0.j, 22.+0.j]])
self       = <numpy.core.tests.test_multiarray.TestMethods object at 0x1256D090>

C:\hostedtoolcache\windows\Python\3.7.4\x86\lib\site-packages\numpy\core\tests\test_multiarray.py:2899: RuntimeWarning

However, I found #12494 (comment), which is referring to the same error

[seberg]

Do not worry about it, it is unrelated and random (and additionally very certainly an issue with the linear algebra backend and not numpy itself)

[mattip]

Thanks @zjpoh