def get_test_array(shape, dtype, no_zeros=False, high=None):
if not isinstance(shape, tuple):
shape = (shape,)
dtype = dtypes.normalize_type(dtype)
if dtypes.is_integer(dtype):
low = 1 if no_zeros else 0
if high is None:
high = 100 # will work even with signed chars
get_arr = lambda: numpy.random.randint(low, high, shape).astype(dtype)
else:
low = 0.01 if no_zeros else 0
if high is None:
high = 1.0
get_arr = lambda: numpy.random.uniform(low, high, shape).astype(dtype)
if dtypes.is_complex(dtype):
return get_arr() + 1j * get_arr()
else:
return get_arr()
def get_test_array(shape, dtype, no_zeros=False, high=None):
if not isinstance(shape, tuple):
shape = (shape, )
dtype = dtypes.normalize_type(dtype)
if dtypes.is_integer(dtype):
low = 1 if no_zeros else 0
if high is None:
high = 100 # will work even with signed chars
get_arr = lambda: numpy.random.randint(low, high, shape).astype(dtype)
else:
low = 0.01 if no_zeros else 0
if high is None:
high = 1.0
get_arr = lambda: numpy.random.uniform(low, high, shape).astype(dtype)
if dtypes.is_complex(dtype):
return get_arr() + 1j * get_arr()
else:
return get_arr()
def deduce(name):
node = self.nodes[name]
if node.children is None:
# Values received from user may point to the same object.
# Therefore we're playing it safe and not assigning them.
node.value.fill_with(values_dict[name])
return
for child in node.children:
deduce(child)
# derive type
child_dtypes = [self.nodes[child].value.dtype for child in node.children]
tr = node.tr_to_children
derive_types = tr.derive_i_from_os if node.type == NODE_OUTPUT else tr.derive_o_from_is
node.value.dtype = dtypes.normalize_type(derive_types(*child_dtypes))
# derive shape
child_shapes = [self.nodes[child].value.shape for child in node.children
if hasattr(self.nodes[child].value, 'shape')]
assert len(set(child_shapes)) == 1
node.value.shape = child_shapes[0]
to_device = (to_device1, to_device2)
from_device = (from_device1, from_device2, from_device3, from_device4,
from_device5)
for to_d, from_d in itertools.product(to_device, from_device):
a_device = to_d(a)
a_copy = ctx.copy_array(a_device)
a_back = from_d(a_copy)
assert diff_is_negligible(a, a_back)
@pytest.mark.parametrize(
"dtype",
TEST_DTYPES,
ids=[dtypes.normalize_type(dtype).name for dtype in TEST_DTYPES])
def test_dtype_support(ctx, dtype):
# Test passes if either context correctly reports that it does not support given dtype,
# or it successfully compiles kernel that operates with this dtype.
N = 256
if not ctx.supports_dtype(dtype):
pytest.skip()
module = ctx.compile("""
KERNEL void test(
GLOBAL_MEM ${ctype} *dest, GLOBAL_MEM ${ctype} *a, GLOBAL_MEM ${ctype} *b)
{
const int i = get_global_id(0);
${ctype} temp = ${func.mul(dtype, dtype)}(a[i], b[i]);
ctx.synchronize()
return y
to_device = (to_device1, to_device2)
from_device = (from_device1, from_device2, from_device3, from_device4, from_device5)
for to_d, from_d in itertools.product(to_device, from_device):
a_device = to_d(a)
a_copy = ctx.copy_array(a_device)
a_back = from_d(a_copy)
assert diff_is_negligible(a, a_back)
@pytest.mark.parametrize(
"dtype", TEST_DTYPES,
ids=[dtypes.normalize_type(dtype).name for dtype in TEST_DTYPES])
def test_dtype_support(ctx, dtype):
# Test passes if either context correctly reports that it does not support given dtype,
# or it successfully compiles kernel that operates with this dtype.
N = 256
if not ctx.supports_dtype(dtype):
pytest.skip()
module = ctx.compile(
"""
KERNEL void test(
GLOBAL_MEM ${ctype} *dest, GLOBAL_MEM ${ctype} *a, GLOBAL_MEM ${ctype} *b)
{
const int i = get_global_id(0);
def __init__(self, dtype):
self.dtype = dtypes.normalize_type(dtype) if dtype is not None else None
self.is_array = False
def __init__(self, shape, dtype):
self.shape = wrap_in_tuple(shape) if shape is not None else None
self.dtype = dtypes.normalize_type(dtype) if dtype is not None else None
self.is_array = True
