Draft: Slice reference output for unit tests to immitate stencil action #400
Conversation
muellch
changed the title
| ) | ||
|
|
||
| @pytest.fixture | ||
| def bounds(self, grid): |
There was a problem hiding this comment.
Do you foresee the bounds being largely the same across most stencils? In that case it would be worth exploring whether you could define a "default" fixture in common, and then let the user overwrite it in each StencilTest
| @@ -151,44 +152,49 @@ def allocate_data(backend, input_data): | |||
| } | |||
| return input_data | |||
|
|
|||
| Bounds = namedtuple("Bounds", ["horizontal_start", "horizontal_end", "vertical_start", "vertical_end"]) | |||
|
|
|||
| @dataclass(frozen=True) | |||
| class Output: | |||
There was a problem hiding this comment.
The Output class can now be deleted
| offset_provider=grid.offset_providers, | ||
| ) | ||
|
|
||
| reference_data = allocate_data(backend, input_data) |
There was a problem hiding this comment.
It may not be necessary to use allocate_data on the reference data, this is to construct gt4py fields and allocate them on the GPU. In this case a simple copy should probably be enough.
There was a problem hiding this comment.
I have no clue how this works, if the gt4py stuff is allocated on the GPU, then you have cupy fields instead of numpy fields?
Then let's say the computations happen on GPU, when we do the np.allclose() is the result field copied back to CPU before we do the comparison?
There was a problem hiding this comment.
Ok I think you can actually leave it as it is with the allocate_data, I missed that the data is being passed as numpy arrays here after all:
| self.PROGRAM.with_backend(backend)( | ||
| **input_data, | ||
| offset_provider=grid.offset_providers, | ||
| ) | ||
| for out in self.OUTPUTS: |
There was a problem hiding this comment.
Could you rename out to output_field_name or something like that for clarity?
| }, | ||
| ) | ||
|
|
||
| input_data = allocate_data(backend, input_data) | ||
| # To imitate the action of the gt4py stencil, we slice the reference output to only apply it in the given domain |
There was a problem hiding this comment.
Could you add another sentence as to why we choose to imitate the action of the gt4py stencil?
|
Mandatory Tests Please make sure you run these tests via comment before you merge!
Optional Tests To run benchmarks you can use:
To run tests and benchmarks with the DaCe backend you can use:
In case your change might affect downstream icon-exclaim, please consider running
For more detailed information please look at CI in the EXCLAIM universe. |
No description provided.