Newer
Older
data
)
knl = lp.split_iname(knl, 'i', 4, inner_tag='vec')
knl = lp.split_array_axis(knl, names, 0, 4)
knl = lp.tag_array_axes(knl, names, 'N0,vec')
queue = cl.CommandQueue(ctx)
if exception is not None:
with pytest.raises(exception):
knl(queue, **kwargs)
else:
result = knl(queue, **kwargs)[1][0]
assert np.array_equal(result.flatten('C'), answer)
ans = np.zeros(12, dtype=np.int32)
ans[7:] = 1
from loopy.diagnostic import LoopyError
# 1) test a conditional on a vector iname -- currently unimplemented as it
# would require creating a 'shadow' vector iname temporary
create_and_test('a[i] = 1', 'i > 6', ans, exception=LoopyError)
# 2) condition on a vector variable -- unimplemented as the i_inner in the
# condition currently isn't resolved
create_and_test('a[i] = 1', 'b[i] > 6', ans, b=np.arange(
12, dtype=np.int32).reshape((3, 4)))
def test_check_for_variable_access_ordering():
knl = lp.make_kernel(
"{[i]: 0<=i<n}",
"""
a[i] = 12
a[i+1] = 13
""")
knl = lp.preprocess_kernel(knl)
from loopy.diagnostic import VariableAccessNotOrdered
with pytest.raises(VariableAccessNotOrdered):
lp.get_one_scheduled_kernel(knl)
def test_check_for_variable_access_ordering_with_aliasing():
knl = lp.make_kernel(
"{[i]: 0<=i<n}",
"""
a[i] = 12
b[i+1] = 13
""",
[
lp.TemporaryVariable("a", shape="n+1", base_storage="tmp"),
lp.TemporaryVariable("b", shape="n+1", base_storage="tmp"),
])
knl = lp.preprocess_kernel(knl)
from loopy.diagnostic import VariableAccessNotOrdered
with pytest.raises(VariableAccessNotOrdered):
lp.get_one_scheduled_kernel(knl)
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
@pytest.mark.parametrize(("second_index", "expect_barrier"),
[
("2*i", True),
("2*i+1", False),
])
def test_no_barriers_for_nonoverlapping_access(second_index, expect_barrier):
knl = lp.make_kernel(
"{[i]: 0<=i<128}",
"""
a[2*i] = 12 {id=first}
a[%s] = 13 {id=second,dep=first}
""" % second_index,
[
lp.TemporaryVariable("a", lp.auto, shape=(256,),
scope=lp.temp_var_scope.LOCAL),
])
knl = lp.tag_inames(knl, "i:l.0")
knl = lp.preprocess_kernel(knl)
knl = lp.get_one_scheduled_kernel(knl)
assert barrier_between(knl, "first", "second") == expect_barrier
def test_half_complex_conditional(ctx_factory):
ctx = ctx_factory()
queue = cl.CommandQueue(ctx)
knl = lp.make_kernel(
"{[i]: 0 <= i < 10}",
"""
tmp[i] = if(i < 5, 0, 0j)
""")
knl(queue)
if __name__ == "__main__":
if len(sys.argv) > 1:
exec(sys.argv[1])
else:
main([__file__])