def test_regref_transform(self, setup_eng):
"""Test circuit with regref transforms doesn't raise any errors"""
eng, prog = setup_eng(2)
with prog.context as q:
ops.MeasureX | q[0]
ops.Sgate(q[0]) | q[1]
# symbolic hermitian conjugate together with register reference
ops.Dgate(q[0]).H | q[1]
ops.Sgate(ops.RR(q[0], lambda x: x ** 2)) | q[1]
ops.Dgate(ops.RR(q[0], lambda x: -x)).H | q[1]
eng.run(prog)
def test_regref_no_func_str(self):
"""Test a regreftransform with no function string raises exception"""
prog = Program(2)
with prog.context as q:
ops.Sgate(0.43) | q[0]
ops.MeasureX | q[0]
ops.Zgate(ops.RR(q[0], lambda x: 2 * x)) | q[1]
with pytest.raises(ValueError, match="not supported by Blackbird"):
io.to_blackbird(prog)
def test_parameters_with_operations(batch_size, setup_eng, backend, hbar):
"""Test using different types of Parameters with different classes
of ParOperations."""
# TODO: this test is very hard to understand. Should be split up,
# and separated into many individual tests that do one thing only.
# TODO: In the original test suite, this test appears to 'work' on the
# Gaussian backend, even though it never explicitly ignores non-Gaussian
# operations. It even worked on the Fock backends, even though the Fock
# backend doesn't support the ThermalLoss channel?!
# TODO: This test doesn't assert anything!!!!!
eng, prog = setup_eng(2)
r = prog.register
@sf.convert
def func1(x):
return abs(2 * x**2 - 3 * x + 1)
@sf.convert
def func2(x, y):
return abs(2 * x * y - y**2 + 0.5)
kwargs = {}
# RegRefTransforms for deferred measurements (note that some operations
# expect nonnegative parameter values)
rr_inputs = [ops.RR(r[0], lambda x: x**2), func1(r[0]), func2(*r)]
rr_pars = tuple(Parameter(k) for k in rr_inputs)
# other types of parameters
other_inputs = [0.14]
if isinstance(eng.backend, TFBackend):
# HACK: add placeholders for tf.Variables
other_inputs.append(np.inf)
if batch_size is not None:
# test batched input
other_inputs.append(np.random.uniform(size=(batch_size, )))
other_pars = tuple(Parameter(k) for k in other_inputs)
def check(G, par, measure=False):
"""Check a ParOperation/Parameters combination"""
# construct the op using the given tuple of Parameters as args
G = G(*par)
with prog:
if measure:
r[0].val = 0.1
r[1].val = 0.2
if G.ns == 1:
G | r[0]
else:
G | (r[0], r[1])
prog.optimize()
try:
eng.run(prog, **kwargs)
except NotApplicableError as err:
# catch unapplicable op/backend combinations here
logging.debug(err)
# unsuccessful run means the queue was not emptied.
eng.reset_queue()
except NotImplementedError as err:
# catch not yet implemented op/backend combinations here
logging.debug(err)
# unsuccessful run means the queue was not emptied.
eng.reset_queue()
prog.locked = False
scalar_arg_preparations = (
ops.Coherent,
ops.Squeezed,
ops.DisplacedSqueezed,
ops.Thermal,
ops.Catstate,
)
# Fock requires an integer parameter
testset = set(ops.one_args_gates + ops.two_args_gates + ops.channels +
scalar_arg_preparations) - {ops.ThermalLossChannel}
print(testset)
for G in testset:
sig = inspect.signature(G.__init__)
n_args = (len(sig.parameters) - 1
) # number of parameters __init__ takes, minus self
if n_args < 1:
logging.debug(
"Unexpected number of args ({}) for {}, check the testset.".
format(n_args, G))
# shortcut, only test cartesian products up to two parameter types
# n_args = min(n_args, 2)
# check all combinations of Parameter types
for p in itertools.product(rr_pars + other_pars, repeat=n_args):
# TODO: For now decompose() works on unevaluated Parameters.
# This causes an error if a :class:`.RegRefTransform`-based Parameter is used, and
# decompose() tries to do arithmetic on it.
if isinstance(p[0].x, ops.RR):
continue
if isinstance(eng.backend, TFBackend):
# declare tensorflow Variables here (otherwise they will be on a different graph when we do backend.reset below)
# replace placeholders with tf.Variable instances
def f(q):
if np.any(q.x == np.inf): # detect the HACK
if f.v is None: # only construct one instance, only when needed
f.v = Parameter(
tf.Variable(0.8, name="sf_parameter"))
return f.v
return q
f.v = None # kind of like a static variable in a function in C
p = tuple(map(f, p))
# create new tf.Session (which will be associated with the default graph) and initialize variables
session = tf.Session()
kwargs["session"] = session
session.run(tf.global_variables_initializer())
check(G, p, measure=True)
session.close()
else:
check(G, p, measure=True)
eng.backend.reset()
