Cannot use 0D tensor as int

[roastduck]

Please look at this line.

https://github.com/ptillet/torch-blocksparse/blame/master/torch_blocksparse/matmul.py#L239

The error below occurs because width is used as an int.

TypeError: empty() received an invalid combination of arguments - got (Tensor, dtype=torch.dtype), but expected one of:
 * (tuple of ints size, *, tuple of names names, torch.memory_format memory_format, torch.dtype dtype, torch.layout layout, torch.device device, bool pin_memory, bool requires_grad)
 * (tuple of ints size, *, torch.memory_format memory_format, Tensor out, torch.dtype dtype, torch.layout layout, torch.device device, bool pin_memory, bool requires_grad)

This error is triggered when using "dsd" mode.

Testing with PyTorch 1.6.0.

[roastduck]

I fixed an irrelevant bug in my code and this bug disappeared. But it would be still be better if you can test the code above more thoroughly. Thanks.