Sparse tensor creation ignores indices placement #28500
Description
🐛 Bug
Attempting to create a sparse tensor from a tensor of indices and values placed on another GPU than cuda:0 (for instance cuda:1) fails with error device of indices (0) must match device of values (1).
Placing the values on cuda:0 while leaving the indices on cuda:1 works and returns a sparse tensor on cuda:0.
The following demo should show the bug:
In [1]: import torch
In [2]: ind = torch.arange(10).unsqueeze(0).to('cuda:1')
In [3]: sp_ind = torch.cat((ind, ind), 0)
In [4]: sp_val = torch.rand(10).to('cuda:1')
In [5]: sp_shape = (10, 10)
In [6]: x = torch.sparse_coo_tensor(sp_ind, sp_val, sp_shape)
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
<ipython-input-6-776f23c64083> in <module>
----> 1 x = torch.sparse_coo_tensor(sp_ind, sp_val, sp_shape)
RuntimeError: device of indices (0) must match device of values (1)
In [7]: # Now this
In [8]: x = torch.sparse_coo_tensor(sp_ind, sp_val.cuda(0), sp_shape)
In [9]: x
Out[9]:
tensor(indices=tensor([[0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]]),
values=tensor([0.5324, 0.1738, 0.4186, 0.2776, 0.4665, 0.9525, 0.6555,
0.2805, 0.9244, 0.4847]),
device='cuda:0', size=(10, 10), nnz=10, layout=torch.sparse_coo)
In [10]: sp_ind
Out[10]:
tensor([[0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]], device='cuda:1')
In [11]: sp_val
Out[11]:
tensor([0.5324, 0.1738, 0.4186, 0.2776, 0.4665, 0.9525, 0.6555, 0.2805, 0.9244,
0.4847], device='cuda:1')
To Reproduce
Steps to reproduce the behavior:
See code snipped above.
Expected behavior
Provided both indices and values are placed on the same device, torch.sparse_coo_tensor should return a sparse tensor on that device.
Environment
Please copy and paste the output from our
environment collection script
(or fill out the checklist below manually).
You can get the script and run it with:
wget https://raw.githubusercontent.com/pytorch/pytorch/master/torch/utils/collect_env.py
# For security purposes, please check the contents of collect_env.py before running it.
python collect_env.py
- PyTorch Version (e.g., 1.0): 1.3.0
- OS (e.g., Linux): Ubuntu 18.04 LTS
- How you installed PyTorch (
conda,pip, source): conda - Build command you used (if compiling from source): N/A
- Python version: 3.7.4
- CUDA/cuDNN version: 10.2/7.5
- GPU models and configuration: 2x2080 Ti with Nvlink
- Any other relevant information: