Skip to content

Improve cunumeric.prod coverage and some test fixes #905

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
yimoj merged 3 commits into from
May 4, 2023
Merged

Improve cunumeric.prod coverage and some test fixes #905

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
8e41463
Improve prod coverage and some test fixes
yimoj Apr 24, 2023
78ff867
Compare the exceptions raised by NumPy and cuNumeric in negative tests
yimoj Apr 24, 2023
0c972dd
Add test for cunumeric.place mask implicit casting
yimoj May 3, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 21 additions & 2 deletions tests/integration/test_extract.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -205,8 +205,27 @@ def test_place_mask_reshape(shape, vals):
arr_np = mk_seq_array(np, shape)
arr_num = mk_seq_array(num, shape)

mask_np = (arr_np % 2).astype(bool)
mask_num = (arr_np % 2).astype(bool)
mask_np = np.arange(0, arr_np.size).astype(bool)
mask_num = num.arange(0, arr_num.size).astype(bool)

vals_np = np.array(vals).astype(arr_np.dtype)
vals_num = num.array(vals_np)

np.place(arr_np, mask_np, vals_np)
num.place(arr_num, mask_num, vals_num)

assert np.array_equal(arr_np, arr_num)


@pytest.mark.parametrize("dtype", (np.float32, np.complex64), ids=str)
def test_place_mask_dtype(dtype):
shape = (3, 2, 3)
vals = [42 + 3j]
arr_np = mk_seq_array(np, shape)
arr_num = mk_seq_array(num, shape)

mask_np = mk_seq_array(np, shape).astype(dtype)
mask_num = mk_seq_array(num, shape).astype(dtype)

vals_np = np.array(vals).astype(arr_np.dtype)
vals_num = num.array(vals_np)
Expand Down
143 changes: 77 additions & 66 deletions tests/integration/test_prod.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,10 +78,8 @@

ARR = ([], [[]], [[], []], np.inf, np.Inf, -10.3, 0, 200, 5 + 8j)

DTYPE = ["l", "L", "f", "e", "d"]
COMPLEX_TYPE = ["F"]
NEGATIVE_COMPLEX_TYPE = ["D"]
NEGATIVE_DTYPE = ["h", "i", "H", "I", "?", "b", "B"]
DTYPE = ("l", "L", "f", "e", "d")
INTEGER_DTYPE = ("h", "i", "H", "I", "?", "b", "B")


def to_dtype(s):
Expand All @@ -97,57 +95,26 @@ class TestProdNegative(object):
def test_array(self, arr):
assert allclose(np.prod(arr), num.prod(arr))

@pytest.mark.xfail
@pytest.mark.parametrize("dtype", NEGATIVE_DTYPE, ids=to_dtype)
def test_dtype_negative(self, dtype):
size = (5, 5, 5)
arr = np.random.random(size) * 10 + 2
arr_np = np.array(arr, dtype=dtype)
arr_num = num.array(arr_np)
out_np = np.prod(arr_np) # Numpy return product of all datas
out_num = num.prod(arr_num)
# cuNumeric return an array with a different data
assert allclose(out_np, out_num)

@pytest.mark.skip
@pytest.mark.parametrize("dtype", NEGATIVE_COMPLEX_TYPE, ids=to_dtype)
def test_dtype_complex_negative(self, dtype):
arr = (num.random.rand(5, 5) * 10 + 2) + (
num.random.rand(5, 5) * 10 * 1.0j + 0.2j
)
arr_np = np.array(arr, dtype=dtype)
arr_num = num.array(arr_np)
out_np = np.prod(arr_np)
out_num = num.prod(arr_num)
assert allclose(out_np, out_num)

def test_axis_out_bound(self):
expected_exc = np.AxisError
arr = [-1, 0, 1, 2, 10]
msg = r"bounds"
with pytest.raises(np.AxisError, match=msg):
with pytest.raises(expected_exc):
np.prod(arr, axis=2)
with pytest.raises(expected_exc):
num.prod(arr, axis=2)

@pytest.mark.xfail
@pytest.mark.parametrize("axis", ((-1, 1), (0, 1), (1, 2), (0, 2)))
def test_axis_tuple(self, axis):
size = (5, 5, 5)
arr_np = np.random.random(size) * 10
arr_num = num.array(arr_np)
out_np = np.prod(arr_np, axis=axis)
# cuNumeric raises NotImplementedError:
# Need support for reducing multiple dimensions.
# Numpy get results.
out_num = num.prod(arr_num, axis=axis)
assert allclose(out_np, out_num)

def test_out_negative(self):
expected_exc = ValueError
in_shape = (2, 3, 4)
out_shape = (2, 3, 3)
arr_num = num.random.random(in_shape) * 10
arr_out = num.random.random(out_shape) * 10
msg = r"shapes do not match"
with pytest.raises(ValueError, match=msg):
num.prod(arr_num, out=arr_out, axis=2)
arr_np = np.ndarray(in_shape)
out_np = np.ndarray(out_shape)
arr_num = num.ndarray(in_shape)
out_num = num.ndarray(out_shape)
with pytest.raises(expected_exc):
np.prod(arr_np, out=out_np, axis=2)
with pytest.raises(expected_exc):
num.prod(arr_num, out=out_num, axis=2)

def test_keepdims(self):
in_shape = (2, 3, 4)
Expand All @@ -157,23 +124,21 @@ def test_keepdims(self):
out_num = num.prod(arr_num, axis=2, keepdims=True)
assert allclose(out_np, out_num)

@pytest.mark.xfail
def test_initial_scalar_list(self):
@pytest.mark.parametrize(
"initial",
([2, 3], pytest.param([3], marks=pytest.mark.xfail)),
ids=str,
)
def test_initial_list(self, initial):
expected_exc = ValueError
arr = [[1, 2], [3, 4]]
initial_value = [3]

out_num = num.prod(arr, initial=initial_value) # array(72)
# Numpy raises ValueError:
# Input object to FillWithScalar is not a scalar
out_np = np.prod(arr, initial=initial_value)

assert allclose(out_np, out_num)

def test_initial_list(self):
arr = [[1, 2], [3, 4]]
initial_value = [2, 3]
with pytest.raises(ValueError):
num.prod(arr, initial=initial_value)
with pytest.raises(expected_exc):
np.prod(arr, initial=initial)
# when LEGATE_TEST=1, cuNumeric casts list to scalar and proceeds
with pytest.raises(expected_exc):
num.prod(arr, initial=initial)

def test_initial_empty_array(self):
size = (1, 0)
Expand Down Expand Up @@ -206,6 +171,7 @@ def test_basic(self, size):
out_np = np.prod(arr_np)
out_num = np.prod(arr_num)
assert allclose(out_np, out_num)
assert allclose(out_num, arr_num.prod())

@pytest.mark.parametrize("dtype", DTYPE, ids=to_dtype)
def test_dtype(self, dtype):
Expand All @@ -217,15 +183,43 @@ def test_dtype(self, dtype):
out_num = num.prod(arr_num)
assert allclose(out_np, out_num)

@pytest.mark.parametrize("dtype", COMPLEX_TYPE, ids=to_dtype)
@pytest.mark.xfail(reason="numpy and cunumeric return different dtypes")
@pytest.mark.parametrize("dtype", INTEGER_DTYPE, ids=to_dtype)
def test_dtype_integer_precision(self, dtype):
arr_np = np.arange(0, 5).astype(dtype)
arr_num = num.arange(0, 5).astype(dtype)
out_np = np.prod(arr_np)
out_num = num.prod(arr_num)
assert allclose(out_num, arr_num.prod())
# When input precision is less than default platform integer
# NumPy returns the product with dtype of platform integer
# cuNumeric returns the product with dtype of the input array
assert allclose(out_np, out_num)

@pytest.mark.parametrize(
"dtype",
(
"F",
pytest.param("D", marks=pytest.mark.xfail),
pytest.param("G", marks=pytest.mark.xfail),
),
ids=to_dtype,
)
def test_dtype_complex(self, dtype):
arr = (num.random.rand(5, 5) * 10 + 2) + (
num.random.rand(5, 5) * 10 * 1.0j + 0.2j
arr = (np.random.rand(5, 5) * 10 + 2) + (
np.random.rand(5, 5) * 10 * 1.0j + 0.2j
)
arr_np = np.array(arr, dtype=dtype)
arr_num = num.array(arr_np)
arr_num = num.array(arr, dtype=dtype)
out_np = np.prod(arr_np)
# cunumeric always returns [1+0.j] when LEGATE_TEST=1
out_num = num.prod(arr_num)
# When running tests with CUNUMERIC_TEST=1 and dtype is complex256,
# allclose hits assertion error:
# File "/legate/cunumeric/cunumeric/eager.py", line 293,
# in to_deferred_array
# assert self.runtime.is_supported_type(self.array.dtype)
# AssertionError
assert allclose(out_np, out_num)

@pytest.mark.parametrize("axis", (_ for _ in range(-2, 3, 1)))
Expand All @@ -237,6 +231,23 @@ def test_axis_basic(self, axis):
out_np = np.prod(arr_np, axis=axis)
assert allclose(out_np, out_num)

@pytest.mark.xfail(reason="cunumeric raises exceptions when LEGATE_TEST=1")
@pytest.mark.parametrize(
"axis", ((-1, 1), (0, 1), (1, 2), (0, 2)), ids=str
)
def test_axis_tuple(self, axis):
size = (5, 5, 5)
arr_np = np.random.random(size) * 10
arr_num = num.array(arr_np)
out_np = np.prod(arr_np, axis=axis)
# when LEGATE_TEST = 1 cuNumeric raises two types of exceptions
# (-1, 1): ValueError: Invalid promotion on dimension 2 for a 1-D store
# others:
# NotImplementedError: Need support for reducing multiple dimensions
# Numpy get results.
out_num = num.prod(arr_num, axis=axis)
assert allclose(out_np, out_num)

@pytest.mark.parametrize("size", SIZES)
def test_out_basic(self, size):
arr_np = np.random.random(size)
Expand Down