1. create cuda sparse tensor with cpu indices tensor
   master:

   >>> torch.sparse.FloatTensor(torch.zeros(1,4).long().cuda(), torch.randn(4,4,4), [3,4,4])
   Traceback (most recent call last):
     File "<stdin>", line 1, in <module>
   RuntimeError

   0.4:

   >>> torch.sparse.FloatTensor(torch.zeros(1,4).long().cuda(), torch.randn(4,4,4), [3,4,4])
   Traceback (most recent call last):
     File "<stdin>", line 1, in <module>
   RuntimeError: Expected object of type torch.LongTensor but found type torch.cuda.LongTensor for argument #1 'indices'

2. add cuda sparse tensor to cpu dense tensor
   master:

   >>> x = torch.zeros(3,4)
   >>> sparse_y = torch.cuda.sparse.FloatTensor(torch.zeros(1,4).long().cuda(), torch.randn(4,4,4).cuda(), [3,4,4])
   >>> x + sparse_y
   [1]    2065657 segmentation fault (core dumped)  python -i ~/env.py

   0.4:

   >>> x + sparse_y
   Traceback (most recent call last):
     File "<stdin>", line 1, in <module>
   RuntimeError: The size of tensor a (3) must match the size of tensor b (4) at non-singleton dimension 1

3. dense + sparse with wrong number of dimdensions
   master:

   >>> x = torch.zeros(3,4)
   >>> sparse_y = torch.cuda.sparse.FloatTensor(torch.zeros(1,4).long().cuda(), torch.randn(4,4,4).cuda(), [3,4,4])
   >>> x.cuda()+sparse_y
   tensor([[ 2.1478, -0.7767, -0.8796, -2.2104],
           [ 0.0000,  0.0000,  0.0000,  0.0000],
           [ 0.0000,  0.0000,  0.0000,  0.0000]], device='cuda:0')

   0.4:

   >>> x = torch.zeros(3,4)
   >>> sparse_y = torch.cuda.sparse.FloatTensor(torch.zeros(1,4).long().cuda(), torch.randn(4,4,4).cuda(), [3,4,4])
   >>> x.cuda() + sparse_y
   Traceback (most recent call last):
     File "<stdin>", line 1, in <module>
   RuntimeError: The size of tensor a (3) must match the size of tensor b (4) at non-singleton dimension 1

cc @ezyang