Distributed `einsum` calls from python

[obackhouse]

Are einsum calls from Python using some distributed layout not supported? The following code segfaults for me on >1 process

import numpy as np
import tiledarray as ta

world = ta.get_default_world()

ref = np.random.rand(4, 4)

x = ta.TArray([4, 4], 2, world=world)
for i, tile in enumerate(x):
    if i % world.size == world.rank:
        slices = tuple(slice(a, b) for a, b in zip(tile.range.start, tile.range.stop))
        tile.data = ref[slices].copy()
world.fence()

y = ta.TArray()
ta.einsum("ij,jk->ik", x, x, y)

as does something even simpler like ta.einsum("ij->ij", x, y).