issues: 654678508

This data as json

id node_id number title user state locked assignee milestone comments created_at updated_at closed_at author_association active_lock_reason draft pull_request body reactions performed_via_github_app state_reason repo type
654678508 MDExOlB1bGxSZXF1ZXN0NDQ3MzU2ODAw 4214 Add initial cupy tests 1610850 closed 0     8 2020-07-10T10:20:33Z 2020-07-13T16:32:35Z 2020-07-13T15:07:45Z CONTRIBUTOR   0 pydata/xarray/pulls/4214

Added some initial unit tests for cupy. Mainly to create a place for cupy tests to go and to check some basic functionality.

I've created a fixture which constructs the dataset from the Toy weather data example and converts the underlying arrays to cupy. Then I've added a test which checks that after calling operations such as mean and groupby the resulting DataArray is still backed by a cupy array.

The main penalty with working on GPUs is accidentally shunting data back and forth between the GPU and system memory. Copying data over the PCI bus is slow compared to the rest of the work so should be avoided. So this first test is checking that we are leaving things on the GPU.

Because this data copying is so expensive cupy have intentionally broken the __array__ method and introduced a .get method instead. This means that users have to be explicit in converting back to numpy and copying back to the main memory. Therefore we will need to add some logic to xarray to use .get in appropriate situations such as plotting.

  • [x] Releated to #4212
  • [x] Tests added
  • [x] Passes isort -rc . && black . && mypy . && flake8
 
{
    "url": "https://api.github.com/repos/pydata/xarray/issues/4214/reactions",
    "total_count": 2,
    "+1": 2,
    "-1": 0,
    "laugh": 0,
    "hooray": 0,
    "confused": 0,
    "heart": 0,
    "rocket": 0,
    "eyes": 0
}
     13221727 pull

Links from other tables