def test_c_execution_with_global_temporaries():
# ensure that the "host" code of a bare ExecutableCTarget with
# global constant temporaries is None
from loopy.target.c import ExecutableCTarget
AS = lp.AddressSpace # noqa
n = 10
knl = lp.make_kernel(
"{[i]: 0 <= i < n}",
"""
a[i] = b[i]
""", [
lp.GlobalArg("a", shape=(n, ), dtype=np.int32),
lp.TemporaryVariable("b",
shape=(n, ),
initializer=np.arange(n, dtype=np.int32),
dtype=np.int32,
read_only=True,
address_space=AS.GLOBAL)
],
target=ExecutableCTarget())
knl = lp.fix_parameters(knl, n=n)
assert ("int b[%d]" % n) not in lp.generate_code_v2(knl).host_code()
assert np.allclose(knl(a=np.zeros(10, dtype=np.int32))[1], np.arange(10))
def test_c_execution_with_global_temporaries():
# ensure that the "host" code of a bare ExecutableCTarget with
# global constant temporaries is None
from loopy.target.c import ExecutableCTarget
from loopy.kernel.data import temp_var_scope as scopes
n = 10
knl = lp.make_kernel(
'{[i]: 0 <= i < n}',
"""
a[i] = b[i]
""", [
lp.GlobalArg('a', shape=(n, ), dtype=np.int32),
lp.TemporaryVariable('b',
shape=(n, ),
initializer=np.arange(n, dtype=np.int32),
dtype=np.int32,
read_only=True,
scope=scopes.GLOBAL)
],
target=ExecutableCTarget())
knl = lp.fix_parameters(knl, n=n)
assert ('int b[%d]' % n) not in lp.generate_code_v2(knl).host_code()
assert np.allclose(knl(a=np.zeros(10, dtype=np.int32))[1], np.arange(10))
def __get_kernel(order="C"):
return lp.make_kernel(
"{ [i,j]: 0<=i,j<n }",
"out[i, j] = 2*a[i, j]", [
lp.GlobalArg("out", np.float32, shape=("n", "n"), order=order),
lp.GlobalArg("a", np.float32, shape=("n", "n"), order=order),
"..."
],
target=ExecutableCTarget())
def __get_knl():
return lp.make_kernel('{[i]: 0 <= i < 10}',
"""
a[i] = b[i]
""",
[lp.GlobalArg('a', shape=(10,), dtype=np.int32),
lp.ConstantArg('b', shape=(10))],
target=ExecutableCTarget(),
name='cache_test')
def __get_knl():
return lp.make_kernel("{[i]: 0 <= i < 10}",
"""
a[i] = b[i]
""",
[lp.GlobalArg("a", shape=(10,), dtype=np.int32),
lp.ConstantArg("b", shape=(10))],
target=ExecutableCTarget(),
name="cache_test")
def test_c_target():
from loopy.target.c import ExecutableCTarget
knl = lp.make_kernel(
"{ [i]: 0<=i<n }",
"out[i] = 2*a[i]", [
lp.GlobalArg("out", np.float32, shape=lp.auto),
lp.GlobalArg("a", np.float32, shape=lp.auto), "..."
],
target=ExecutableCTarget())
assert np.allclose(
knl(a=np.arange(16, dtype=np.float32))[1],
2 * np.arange(16, dtype=np.float32))
def test_function_decl_extractor():
# ensure that we can tell the difference between pointers, constants, etc.
# in execution
from loopy.target.c import ExecutableCTarget
knl = lp.make_kernel('{[i]: 0 <= i < 10}',
"""
a[i] = b[i] + v
""",
[lp.GlobalArg('a', shape=(10,), dtype=np.int32),
lp.ConstantArg('b', shape=(10)),
lp.ValueArg('v', dtype=np.int32)],
target=ExecutableCTarget())
assert np.allclose(knl(b=np.arange(10), v=-1)[1], np.arange(10) - 1)
def __get_kernel(order="C"):
indicies = ["i", "j", "k"]
sizes = tuple(np.random.randint(1, 11, size=len(indicies)))
# create domain strings
domain_template = "{{ [{iname}]: 0 <= {iname} < {size} }}"
domains = []
for idx, size in zip(indicies, sizes):
domains.append(domain_template.format(iname=idx, size=size))
statement = "out[{indexed}] = 2 * a[{indexed}]".format(
indexed=", ".join(indicies))
return lp.make_kernel(
domains,
statement, [
lp.GlobalArg("out", np.float32, shape=sizes, order=order),
lp.GlobalArg("a", np.float32, shape=sizes, order=order), "..."
],
target=ExecutableCTarget())
