def test_default_arguments(backend):
branch_true = gtscript.stencil(backend=backend,
definition=a_stencil,
externals={"BRANCH": True},
rebuild=True)
branch_false = gtscript.stencil(backend=backend,
definition=a_stencil,
externals={"BRANCH": False},
rebuild=True)
arg1 = gt_storage.ones(backend=backend,
dtype=np.float64,
shape=(3, 3, 3),
default_origin=(0, 0, 0))
arg2 = gt_storage.zeros(backend=backend,
dtype=np.float64,
shape=(3, 3, 3),
default_origin=(0, 0, 0))
arg3 = gt_storage.ones(backend=backend,
dtype=np.float64,
shape=(3, 3, 3),
default_origin=(0, 0, 0))
tmp = np.asarray(arg3)
tmp *= 2
branch_true(arg1, None, arg3, par1=2.0)
np.testing.assert_equal(arg1, 14 * np.ones((3, 3, 3)))
branch_true(arg1, None, par1=2.0)
np.testing.assert_equal(arg1, 196 * np.ones((3, 3, 3)))
branch_false(arg1, arg2, arg3, par1=2.0, par3=2.0)
np.testing.assert_equal(arg1, 56 * np.ones((3, 3, 3)))
with pytest.raises((ValueError, AssertionError)):
branch_false(arg1, arg2, par1=2.0, par3=2.0)
arg1 = gt_storage.ones(backend=backend,
dtype=np.float64,
shape=(3, 3, 3),
default_origin=(0, 0, 0))
arg2 = gt_storage.zeros(backend=backend,
dtype=np.float64,
shape=(3, 3, 3),
default_origin=(0, 0, 0))
arg3 = gt_storage.ones(backend=backend,
dtype=np.float64,
shape=(3, 3, 3),
default_origin=(0, 0, 0))
tmp = np.asarray(arg3)
tmp *= 2
branch_true(arg1, arg2=None, par1=2.0, par2=5.0, par3=3.0)
np.testing.assert_equal(arg1, 10 * np.ones((3, 3, 3)))
branch_true(arg1, arg2=None, par1=2.0, par2=5.0)
np.testing.assert_equal(arg1, 100 * np.ones((3, 3, 3)))
branch_false(arg1, arg2, arg3, par1=2.0, par2=5.0, par3=3.0)
np.testing.assert_equal(arg1, 60 * np.ones((3, 3, 3)))
with pytest.raises((TypeError, AssertionError)):
branch_false(arg1, arg2, arg3, par1=2.0, par2=5.0)
def test_halo_checks(backend):
stencil = gtscript.stencil(definition=avg_stencil, backend=backend)
# test default works
in_field = gt_storage.ones(backend=backend,
shape=(22, 22, 10),
default_origin=(1, 1, 0),
dtype=np.float64)
out_field = gt_storage.zeros(backend=backend,
shape=(22, 22, 10),
default_origin=(1, 1, 0),
dtype=np.float64)
stencil(in_field=in_field, out_field=out_field)
assert (out_field[1:-1, 1:-1, :] == 1).all()
# test setting arbitrary, small domain works
in_field = gt_storage.ones(backend=backend,
shape=(22, 22, 10),
default_origin=(1, 1, 0),
dtype=np.float64)
out_field = gt_storage.zeros(backend=backend,
shape=(22, 22, 10),
default_origin=(1, 1, 0),
dtype=np.float64)
stencil(in_field=in_field,
out_field=out_field,
origin=(2, 2, 0),
domain=(10, 10, 10))
assert (out_field[2:12, 2:12, :] == 1).all()
# test setting domain+origin too large raises
in_field = gt_storage.ones(backend=backend,
shape=(22, 22, 10),
default_origin=(1, 1, 0),
dtype=np.float64)
out_field = gt_storage.zeros(backend=backend,
shape=(22, 22, 10),
default_origin=(1, 1, 0),
dtype=np.float64)
with pytest.raises(ValueError):
stencil(in_field=in_field,
out_field=out_field,
origin=(2, 2, 0),
domain=(20, 20, 10))
# test 2*origin+domain does not raise if still fits (c.f. previous bug in c++ check.)
in_field = gt_storage.ones(backend=backend,
shape=(23, 23, 10),
default_origin=(1, 1, 0),
dtype=np.float64)
out_field = gt_storage.zeros(backend=backend,
shape=(23, 23, 10),
default_origin=(1, 1, 0),
dtype=np.float64)
stencil(in_field=in_field,
out_field=out_field,
origin=(2, 2, 0),
domain=(20, 20, 10))
def test_lower_dimensional_inputs(backend):
@gtscript.stencil(backend=backend)
def stencil(
field_3d: gtscript.Field[gtscript.IJK, np.float_],
field_2d: gtscript.Field[gtscript.IJ, np.float_],
field_1d: gtscript.Field[gtscript.K, np.float_],
):
with computation(FORWARD):
with interval(0, 1):
field_2d = field_1d[1]
with computation(PARALLEL):
with interval(0, -1):
tmp = field_2d[0, 1] + field_1d[1]
with interval(-1, None):
tmp = field_2d[0, 1] + field_1d[0]
with computation(PARALLEL):
with interval(0, 1):
field_3d = tmp[1, 0, 0] + field_1d[1]
with interval(1, None):
field_3d = tmp[-1, 0, 0]
full_shape = (6, 6, 6)
default_origin = (1, 1, 0)
dtype = float
field_3d = gt_storage.zeros(backend,
default_origin,
full_shape,
dtype,
mask=None)
assert field_3d.shape == full_shape[:]
field_2d = gt_storage.zeros(backend,
default_origin[:-1],
full_shape[:-1],
dtype,
mask=[True, True, False])
assert field_2d.shape == full_shape[:-1]
field_1d = gt_storage.ones(backend, (default_origin[-1], ),
(full_shape[-1], ),
dtype,
mask=[False, False, True])
assert field_1d.shape == (full_shape[-1], )
stencil(field_3d, field_2d, field_1d, origin=(1, 1, 0), domain=(4, 3, 6))
field_3d.device_to_host()
np.testing.assert_allclose(field_3d.view(np.ndarray)[1:-1, 1:-2, :1], 3)
np.testing.assert_allclose(field_3d.view(np.ndarray)[1:-1, 1:-2, 1:], 2)
stencil(field_3d, field_2d, field_1d)
def test_optional_arg_provide(backend):
@gtscript.stencil(backend=backend)
def stencil(
inp: gtscript.Field[np.float64],
unused_field: gtscript.Field[np.float64],
outp: gtscript.Field[np.float64],
unused_par: float,
):
with computation(PARALLEL), interval(...):
outp = inp # noqa F841: local variable 'outp' is assigned to but never used
frozen_stencil = stencil.freeze(
domain=(3, 3, 10),
origin={
"inp": (2, 2, 0),
"outp": (2, 2, 0),
"unused_field": (0, 0, 0)
},
)
inp = gt_storage.from_array(data=7.0,
dtype=np.float64,
shape=(10, 10, 10),
default_origin=(0, 0, 0),
backend=backend)
outp = gt_storage.zeros(dtype=np.float64,
shape=(10, 10, 10),
default_origin=(0, 0, 0),
backend=backend)
unused_field = gt_storage.zeros(dtype=np.float64,
shape=(10, 10, 10),
default_origin=(0, 0, 0),
backend=backend)
inp.host_to_device()
outp.host_to_device()
@dace.program(
device=dace.DeviceType.GPU if "gpu" in backend else dace.DeviceType.CPU
)
def call_frozen_stencil():
frozen_stencil(inp=inp,
unused_field=unused_field,
outp=outp,
unused_par=7.0)
call_frozen_stencil()
outp.device_to_host(force=True)
assert np.allclose(inp, 7.0)
assert np.allclose(np.asarray(outp)[2:5, 2:5, :], 7.0)
assert np.sum(np.asarray(outp), axis=(0, 1, 2)) == 90 * 7.0
def test_ndarray_warning():
"""test that proper warnings are raised depending on field type."""
backend = "numpy"
stencil = gtscript.stencil(definition=avg_stencil, backend=backend)
# test numpy int types are accepted
in_field = gt_storage.ones(
backend=backend,
shape=np.asarray((23, 23, 10), dtype=np.int64),
default_origin=np.asarray((1, 1, 0), dtype=np.int64),
dtype=np.float64,
)
out_field = gt_storage.zeros(
backend=backend,
shape=np.asarray((23, 23, 10), dtype=np.int64),
default_origin=np.asarray((1, 1, 0), dtype=np.int64),
dtype=np.float64,
)
with pytest.warns(RuntimeWarning):
stencil(
in_field=in_field.view(np.ndarray),
out_field=out_field.view(np.ndarray),
origin=np.asarray((2, 2, 0), dtype=np.int64),
domain=np.asarray((20, 20, 10), dtype=np.int64),
)
with pytest.warns(None) as record:
stencil(
in_field=in_field,
out_field=out_field,
origin=np.asarray((2, 2, 0), dtype=np.int64),
domain=np.asarray((20, 20, 10), dtype=np.int64),
)
assert len(record) == 0
def test_stage_merger_induced_interval_block_reordering(backend):
field_in = gt_storage.ones(
dtype=np.float_, backend=backend, shape=(23, 23, 23), default_origin=(0, 0, 0)
)
field_out = gt_storage.zeros(
dtype=np.float_, backend=backend, shape=(23, 23, 23), default_origin=(0, 0, 0)
)
@gtscript.stencil(backend=backend)
def stencil(field_in: gtscript.Field[np.float_], field_out: gtscript.Field[np.float_]):
with computation(BACKWARD):
with interval(-2, -1): # block 1
field_out = field_in
with interval(0, -2): # block 2
field_out = field_in
with computation(BACKWARD):
with interval(-1, None): # block 3
field_out = 2 * field_in
with interval(0, -1): # block 4
field_out = 3 * field_in
stencil(field_in, field_out)
np.testing.assert_allclose(field_out.view(np.ndarray)[:, :, 0:-1], 3)
np.testing.assert_allclose(field_out.view(np.ndarray)[:, :, -1], 2)
def run_rank_adder_test(backend: str, rebuild: bool):
comm = MPI.COMM_WORLD
size = comm.Get_size()
set_backend(backend)
StencilTable.clear()
origin = (0, 0, 0)
domain = (1, 1, size)
out_field = gt_storage.zeros(shape=domain,
default_origin=origin,
dtype=np.int64,
backend=backend)
ref_field = np.arange(1, size + 1, dtype=np.int64)
for rank in range(0, size):
stencil_object = future_stencil(
definition=add_rank,
backend=backend,
rebuild=rebuild,
externals={"rank": rank},
)
stencil_object(out_field, domain=domain, origin=origin)
assert np.array_equal(out_field[0, 0, :], ref_field)
def test_exec_info(backend):
"""test that proper warnings are raised depending on field type."""
stencil = gtscript.stencil(definition=avg_stencil, backend=backend)
exec_info = {}
# test numpy int types are accepted
in_field = gt_storage.ones(
backend=backend,
shape=(np.int8(23), np.int16(23), np.int32(10)),
default_origin=(1, 1, 0),
dtype=np.float64,
)
out_field = gt_storage.zeros(
backend=backend, shape=(23, 23, 10), default_origin=(1, 1, 0), dtype=np.float64
)
stencil(
in_field=in_field,
out_field=out_field,
origin=(2, 2, 0),
domain=(20, 20, 10),
exec_info=exec_info,
)
timings = ["call", "call_run", "run"]
assert all([k + "_start_time" in exec_info for k in timings])
assert all([k + "_end_time" in exec_info for k in timings])
assert all([exec_info[k + "_end_time"] > exec_info[k + "_start_time"] for k in timings])
if backend.startswith("gt:"):
assert "run_cpp_start_time" in exec_info
assert "run_cpp_end_time" in exec_info
assert exec_info["run_cpp_end_time"] > exec_info["run_cpp_start_time"]
def test_negative_origin(backend):
def stencil_i(
input_field: gtscript.Field[gtscript.IJK, np.int32],
output_field: gtscript.Field[gtscript.IJK, np.int32],
):
with computation(PARALLEL), interval(...):
output_field = input_field[1, 0, 0]
def stencil_k(
input_field: gtscript.Field[gtscript.IJK, np.int32],
output_field: gtscript.Field[gtscript.IJK, np.int32],
):
with computation(PARALLEL), interval(...):
output_field = input_field[0, 0, 1]
input_field = gt_storage.ones(
backend, default_origin=(0, 0, 0), shape=(1, 1, 1), dtype=np.int32
)
output_field = gt_storage.zeros(
backend, default_origin=(0, 0, 0), shape=(1, 1, 1), dtype=np.int32
)
for origin, stencil in (((-1, 0, 0), stencil_i), ((0, 0, -1), stencil_k)):
gtscript.stencil(definition=stencil, backend=backend)(
input_field, output_field, origin={"input_field": origin}
)
assert output_field[0, 0, 0] == 1
def test_origin_k_fields(backend):
@gtscript.stencil(backend=backend, rebuild=True)
def k_to_ijk(outp: Field[np.float64], inp: Field[gtscript.K, np.float64]):
with computation(PARALLEL), interval(...):
outp = inp
origin = {"outp": (0, 0, 1), "inp": (2,)}
domain = (2, 2, 8)
data = np.arange(10, dtype=np.float64)
inp = gt_storage.from_array(
data=data,
shape=(10,),
default_origin=(0,),
dtype=np.float64,
mask=[False, False, True],
backend=backend,
)
outp = gt_storage.zeros(
shape=(2, 2, 10), default_origin=(0, 0, 0), dtype=np.float64, backend=backend
)
k_to_ijk(outp, inp, origin=origin, domain=domain)
inp.device_to_host()
outp.device_to_host()
np.testing.assert_allclose(data, np.asarray(inp))
np.testing.assert_allclose(
np.broadcast_to(data[2:], shape=(2, 2, 8)), np.asarray(outp)[:, :, 1:-1]
)
np.testing.assert_allclose(0.0, np.asarray(outp)[:, :, 0])
np.testing.assert_allclose(0.0, np.asarray(outp)[:, :, -1])
def test_origin_offsetting_nofrozen(dace_stencil, domain, outp_origin):
backend = dace_stencil.backend
inp = gt_storage.from_array(data=7.0,
dtype=np.float64,
shape=(10, 10, 10),
default_origin=(0, 0, 0),
backend=backend)
outp = gt_storage.zeros(dtype=np.float64,
shape=(10, 10, 10),
default_origin=(0, 0, 0),
backend=backend)
origin = {"inp": (0, 0, 0), "outp": outp_origin}
inp.host_to_device()
outp.host_to_device()
@dace.program(
device=dace.DeviceType.GPU if "gpu" in backend else dace.DeviceType.CPU
)
def call_stencil_object():
dace_stencil(inp=inp, outp=outp, domain=domain, origin=origin)
call_stencil_object()
outp.device_to_host(force=True)
assert np.allclose(inp, 7.0)
assert np.allclose(
np.asarray(outp)[outp_origin[0]:outp_origin[0] + domain[0],
outp_origin[1]:outp_origin[1] + domain[1],
outp_origin[2]:outp_origin[2] + domain[2], ],
7.0,
)
assert np.sum(np.asarray(outp), axis=(0, 1, 2)) == np.prod(domain) * 7.0
def setup_and_run(backend, **kwargs):
field_a = gt_storage.zeros(
dtype=np.float_, backend=backend, shape=(3, 3, 1), default_origin=(0, 0, 0)
)
@gtscript.stencil(backend=backend, **kwargs)
def divide_by_zero(field_a: gtscript.Field[np.float_]):
with computation(PARALLEL), interval(...):
field_a = 1.0 / field_a
divide_by_zero(field_a)
def test_read_data_dim_indirect_addressing(backend):
INT32_VEC2 = (np.int32, (2,))
def stencil(
input_field: gtscript.Field[gtscript.IJK, INT32_VEC2],
output_field: gtscript.Field[gtscript.IJK, np.int32],
index: int,
):
with computation(PARALLEL), interval(...):
output_field = input_field[0, 0, 0][index]
default_origin = (0, 0, 0)
full_shape = (1, 1, 2)
input_field = gt_storage.ones(backend, default_origin, full_shape, dtype=INT32_VEC2)
output_field = gt_storage.zeros(backend, default_origin, full_shape, dtype=np.int32)
gtscript.stencil(definition=stencil, backend=backend)(input_field, output_field, 1)
assert output_field[0, 0, 0] == 1
def test_lower_dimensional_inputs(backend):
@gtscript.stencil(backend=backend)
def stencil(
field_3d: gtscript.Field[np.float_, gtscript.IJK],
field_2d: gtscript.Field[np.float_, gtscript.IJ],
field_1d: gtscript.Field[np.float_, gtscript.K],
):
with computation(FORWARD):
with interval(0, 1):
field_d = field_1d[1] + field_3d[0, 1, 0]
with computation(PARALLEL):
with interval(0, 1):
tmp = field_2d[0, 1] + field_1d[1]
field_3d = tmp[1, 0, 0] + field_1d[1]
with interval(1, None):
field_3d = tmp[-1, 0, 0]
full_shape = (6, 6, 6)
default_origin = (1, 1, 0)
dtype = float
field_3d = gt_storage.ones(backend,
default_origin,
full_shape,
dtype,
mask=None)
assert field_3d.shape == full_shape[:]
field_2d = gt_storage.ones(backend,
default_origin[:-1],
full_shape[:-1],
dtype,
mask=[True, True, False])
assert field_2d.shape == full_shape[:-1]
field_1d = gt_storage.zeros(backend, (default_origin[-1], ),
(full_shape[-1], ),
dtype,
mask=[False, False, True])
assert field_1d.shape == (full_shape[-1], )
stencil(field_3d, field_2d, field_1d, origin=(1, 1, 0))
stencil(field_3d, field_2d, field_1d)
def test_mask_with_offset_written_in_conditional(backend):
@gtscript.stencil(backend, externals={"mord": 5})
def stencil(
outp: gtscript.Field[np.float_],
):
with computation(PARALLEL), interval(...):
cond = True
if cond[0, -1, 0] or cond[0, 0, 0]:
outp = 1.0
else:
outp = 0.0
outp = gt_storage.zeros(
shape=(10, 10, 10), backend=backend, default_origin=(0, 0, 0), dtype=float
)
stencil(outp)
outp.device_to_host()
assert np.allclose(1.0, np.asarray(outp))
def test_read_data_dim_indirect_addressing(backend):
INT32_VEC2 = (np.int32, (2,))
def stencil(
input_field: gtscript.Field[gtscript.IJK, INT32_VEC2],
output_field: gtscript.Field[gtscript.IJK, np.int32],
index: int,
):
with computation(PARALLEL), interval(...):
output_field = input_field[0, 0, 0][index]
default_origin = (0, 0, 0)
full_shape = (1, 1, 2)
input_field = gt_storage.ones(backend, default_origin, full_shape, dtype=INT32_VEC2)
output_field = gt_storage.zeros(backend, default_origin, full_shape, dtype=np.int32)
if backend in (backend.values[0] for backend in LEGACY_GRIDTOOLS_BACKENDS):
with pytest.raises(ValueError):
gtscript.stencil(definition=stencil, backend=backend)
else:
gtscript.stencil(definition=stencil, backend=backend)(input_field, output_field, 1)
assert output_field[0, 0, 0] == 1
def test_nondace_raises():
@gtscript.stencil(backend="numpy")
def numpy_stencil(inp: gtscript.Field[np.float64],
outp: gtscript.Field[np.float64]):
with computation(PARALLEL), interval(...):
outp = inp # noqa F841: local variable 'outp' is assigned to but never used
frozen_stencil = numpy_stencil.freeze(domain=(3, 3, 3),
origin={
"inp": (0, 0, 0),
"outp": (0, 0, 0)
})
inp = gt_storage.from_array(
data=7.0,
dtype=np.float64,
shape=(10, 10, 10),
default_origin=(0, 0, 0),
backend="numpy",
)
outp = gt_storage.zeros(
dtype=np.float64,
shape=(10, 10, 10),
default_origin=(0, 0, 0),
backend="numpy",
)
@dace.program
def call_frozen_stencil():
frozen_stencil(inp=inp, outp=outp)
with pytest.raises(
TypeError,
match=re.escape(
"Only dace backends are supported in DaCe-orchestrated programs."
' (found "numpy")'),
):
call_frozen_stencil()
def test_np_int_types():
backend = "numpy"
stencil = gtscript.stencil(definition=avg_stencil, backend=backend)
# test numpy int types are accepted
in_field = gt_storage.ones(
backend=backend,
shape=(np.int8(23), np.int16(23), np.int32(10)),
default_origin=(np.int64(1), int(1), 0),
dtype=np.float64,
)
out_field = gt_storage.zeros(
backend=backend,
shape=(np.int8(23), np.int16(23), np.int32(10)),
default_origin=(np.int64(1), int(1), 0),
dtype=np.float64,
)
stencil(
in_field=in_field,
out_field=out_field,
origin=(np.int8(2), np.int16(2), np.int32(0)),
domain=(np.int64(20), int(20), 10),
)
def test_np_array_int_types():
backend = "numpy"
stencil = gtscript.stencil(definition=avg_stencil, backend=backend)
# test numpy int types are accepted
in_field = gt_storage.ones(
backend=backend,
shape=np.asarray((23, 23, 10), dtype=np.int64),
default_origin=np.asarray((1, 1, 0), dtype=np.int64),
dtype=np.float64,
)
out_field = gt_storage.zeros(
backend=backend,
shape=np.asarray((23, 23, 10), dtype=np.int64),
default_origin=np.asarray((1, 1, 0), dtype=np.int64),
dtype=np.float64,
)
stencil(
in_field=in_field,
out_field=out_field,
origin=np.asarray((2, 2, 0), dtype=np.int64),
domain=np.asarray((20, 20, 10), dtype=np.int64),
)
def test_basic(backend):
@gtscript.stencil(backend=backend)
def defn(outp: gtscript.Field[np.float64], par: np.float64):
with computation(PARALLEL), interval(...):
outp = par # noqa F841: local variable 'outp' is assigned to but never used
outp = gt_storage.zeros(dtype=np.float64,
shape=(10, 10, 10),
default_origin=(0, 0, 0),
backend=backend)
inp = 7.0
outp.host_to_device()
@dace.program(
device=dace.DeviceType.GPU if "gpu" in backend else dace.DeviceType.CPU
)
def call_stencil_object(locoutp, locinp):
defn(locoutp, par=locinp)
call_stencil_object(locoutp=outp, locinp=inp)
outp.device_to_host(force=True)
assert np.allclose(outp, 7.0)
def zeros(self, *args, **kwargs):
"""Create a storage with all fields initialized to 0."""
keywords = self._kwargs.copy()
keywords.update(kwargs)
return storage.zeros(*args, **keywords)
def test_optional_arg_provide_aot(backend):
@gtscript.stencil(backend=backend)
def stencil(
inp: gtscript.Field[np.float64],
unused_field: gtscript.Field[np.float64],
outp: gtscript.Field[np.float64],
unused_par: float,
):
with computation(PARALLEL), interval(...):
outp = inp # noqa F841: local variable 'outp' is assigned to but never used
frozen_stencil = stencil.freeze(
domain=(3, 3, 10),
origin={
"inp": (2, 2, 0),
"outp": (2, 2, 0),
"unused_field": (0, 0, 0)
},
)
inp = gt_storage.from_array(data=7.0,
dtype=np.float64,
shape=(10, 10, 10),
default_origin=(0, 0, 0),
backend=backend)
outp = gt_storage.zeros(dtype=np.float64,
shape=(10, 10, 10),
default_origin=(0, 0, 0),
backend=backend)
unused_field = gt_storage.zeros(dtype=np.float64,
shape=(10, 10, 10),
default_origin=(0, 0, 0),
backend=backend)
inp.host_to_device()
outp.host_to_device()
storage = dace.StorageType.GPU_Global if "gpu" in backend else dace.StorageType.CPU_Heap
@dace.program(
device=dace.DeviceType.GPU if "gpu" in backend else dace.DeviceType.CPU
)
def call_frozen_stencil(
inp: dace.data.Array(
shape=inp.shape,
strides=tuple(s // inp.itemsize for s in inp.strides),
dtype=dace.float64,
storage=storage,
),
outp: dace.data.Array(
shape=outp.shape,
strides=tuple(s // outp.itemsize for s in outp.strides),
dtype=dace.float64,
storage=storage,
),
unused_field: dace.data.Array(
shape=unused_field.shape,
strides=tuple(s // unused_field.itemsize
for s in unused_field.strides),
dtype=dace.float64,
storage=storage,
), # type: ignore
unused_par: dace.float64, # type: ignore
):
frozen_stencil(inp=inp,
unused_field=unused_field,
outp=outp,
unused_par=unused_par)
csdfg = call_frozen_stencil.compile()
csdfg(inp=inp, outp=outp, unused_field=unused_field, unused_par=7.0)
outp.device_to_host(force=True)
assert np.allclose(inp, 7.0)
assert np.allclose(np.asarray(outp)[2:5, 2:5, :], 7.0)
assert np.sum(np.asarray(outp), axis=(0, 1, 2)) == 90 * 7.0