Remove test fixture for OTF backends (#403)

Remove test fixture for OTF backends. The reason for test failure is not in the backend but in the numpy validation code, which does not work with local-area grids.

ci/benchmark.yml

@@ -8,7 +8,8 @@ benchmark_model_icon_grid:
   extends: .test_template
   stage: benchmark
   script:
-    - tox -r -e run_benchmarks -c model/ -- --backend=$BACKEND --grid=icon_grid_global --verbose
+    - tox -r -e run_benchmarks -c model/ -- --backend=$BACKEND --grid=$GRID --verbose
   parallel:
     matrix:
       - BACKEND: [gtfn_cpu, gtfn_gpu]
+        GRID: [icon_grid, icon_grid_global]

ci/dace.yml

@@ -22,7 +22,8 @@ benchmark_model_icon_grid:
     stage: benchmark
     script:
       - pip install dace==$DACE_VERSION
-      - tox -r -e run_benchmarks -c model/ -- --backend=$BACKEND --grid=icon_grid_global --verbose
+      - tox -r -e run_benchmarks -c model/ -- --backend=$BACKEND --grid=$GRID --verbose
     parallel:
       matrix:
         - BACKEND: [dace_cpu, dace_gpu]
+          GRID: [icon_grid, icon_grid_global]

model/atmosphere/diffusion/tests/conftest.py

@@ -16,5 +16,4 @@
 )
 from icon4py.model.common.test_utils.helpers import (  # noqa : F401  # fixtures from test_utils
     backend,
-    uses_local_area_icon_grid_with_otf,
 )

model/atmosphere/diffusion/tests/diffusion_stencil_tests/test_apply_diffusion_to_theta_and_exner.py

@@ -81,12 +81,7 @@ def reference(
         return dict(theta_v=theta_v, exner=exner)
 
     @pytest.fixture
-    def input_data(self, grid, uses_local_area_icon_grid_with_otf):
-        if uses_local_area_icon_grid_with_otf:
-            pytest.skip(
-                "Execution domain needs to be restricted or boundary taken into account in stencil."
-            )
-
+    def input_data(self, grid):
         kh_smag_e = random_field(grid, EdgeDim, KDim)
         inv_dual_edge_length = random_field(grid, EdgeDim)
         theta_v_in = random_field(grid, CellDim, KDim)

model/atmosphere/diffusion/tests/diffusion_stencil_tests/test_apply_diffusion_to_vn.py

@@ -97,12 +97,7 @@ def reference(
         return dict(vn=vn)
 
     @pytest.fixture
-    def input_data(self, grid, uses_local_area_icon_grid_with_otf):
-        if uses_local_area_icon_grid_with_otf:
-            pytest.skip(
-                "Execution domain needs to be restricted or boundary taken into account in stencil."
-            )
-
+    def input_data(self, grid):
         edge = indices_field(EdgeDim, grid, is_halfdim=False, dtype=int32)
 
         u_vert = random_field(grid, VertexDim, KDim)

model/atmosphere/diffusion/tests/diffusion_stencil_tests/test_calculate_nabla2_for_z.py

@@ -54,12 +54,7 @@ def reference(
         return dict(z_nabla2_e=z_nabla2_e)
 
     @pytest.fixture
-    def input_data(self, grid, uses_local_area_icon_grid_with_otf):
-        if uses_local_area_icon_grid_with_otf:
-            pytest.skip(
-                "Execution domain needs to be restricted or boundary taken into account in stencil."
-            )
-
+    def input_data(self, grid):
         kh_smag_e = random_field(grid, EdgeDim, KDim, dtype=vpfloat)
         inv_dual_edge_length = random_field(grid, EdgeDim, dtype=wpfloat)
         theta_v = random_field(grid, CellDim, KDim, dtype=wpfloat)

model/atmosphere/dycore/tests/conftest.py

@@ -16,5 +16,4 @@
 )
 from icon4py.model.common.test_utils.helpers import (  # noqa : F401  # fixtures from test_utils
     backend,
-    uses_local_area_icon_grid_with_otf,
 )

model/atmosphere/dycore/tests/dycore_stencil_tests/test_fused_velocity_advection_stencil_19_to_20.py

@@ -120,12 +120,7 @@ def reference(
         return dict(ddt_vn_apc=ddt_vn_apc)
 
     @pytest.fixture
-    def input_data(self, grid, uses_local_area_icon_grid_with_otf):
-        if uses_local_area_icon_grid_with_otf:
-            pytest.skip(
-                "Execution domain needs to be restricted or boundary taken into account in stencil."
-            )
-
+    def input_data(self, grid):
         z_kin_hor_e = random_field(grid, EdgeDim, KDim)
         coeff_gradekin = random_field(grid, EdgeDim, E2CDim)
         coeff_gradekin_new = as_1D_sparse_field(coeff_gradekin, ECDim)

model/atmosphere/dycore/tests/dycore_stencil_tests/test_fused_velocity_advection_stencil_1_to_7.py

@@ -203,14 +203,10 @@ def reference(
         )
 
     @pytest.fixture
-    def input_data(self, grid, uses_local_area_icon_grid_with_otf):
+    def input_data(self, grid):
         pytest.skip(
             "Verification of z_v_grad_w currently not working, because numpy version incorrect."
         )
-        if uses_local_area_icon_grid_with_otf:
-            pytest.skip(
-                "Execution domain needs to be restricted or boundary taken into account in stencil."
-            )
 
         c_intp = random_field(grid, VertexDim, V2CDim)
         vn = random_field(grid, EdgeDim, KDim)

model/common/src/icon4py/model/common/test_utils/helpers.py

@@ -23,7 +23,6 @@
 from gt4py.next.ffront.decorator import Program
 
 from ..grid.base import BaseGrid
-from ..grid.icon import IconGrid
 from ..type_alias import wpfloat
 
 
@@ -246,27 +245,5 @@ def __init_subclass__(cls, **kwargs):
         setattr(cls, f"test_{cls.__name__}_benchmark", _test_execution_benchmark)
 
 
-@pytest.fixture
-def uses_local_area_icon_grid_with_otf(backend, grid):
-    """Check whether we are using a compiled backend with a limited_area icon_grid.
-
-    Is needed to skip certain stencils where the execution domain needs to be restricted or boundary taken into account.
-    """
-    if hasattr(backend, "executor") and isinstance(grid, IconGrid):
-        if grid.limited_area:
-            if isinstance(
-                backend.executor, gt4py.next.program_processors.modular_executor.ModularExecutor
-            ):
-                return True
-            try:
-                from gt4py.next.program_processors.runners import dace_iterator
-
-                if backend in {dace_iterator.run_dace_cpu, dace_iterator.run_dace_gpu}:
-                    return True
-            except ImportError:
-                pass
-    return False
-
-
 def reshape(arr: np.array, shape: tuple[int, ...]):
     return np.reshape(arr, shape)