def init_partition(circuit, block_ids=None) -> Data:
"""
randomly partition the nodes equally, if the block_ids
is not specified.
:return: data container for the current partition
"""
pmax = get_pmax(circuit)
nets_size = circuit.get_nets_size()
n = circuit.get_cells_size()
if block_ids is None:
random_cids = random.sample(range(n), n)
block_ids = [cid % 2 for i, cid in enumerate(random_cids)]
# initialize a data container for the current partition
data = Data(pmax, nets_size, block_ids)
# update the nets distribution in the current partition
update_distribution(circuit, data)
# update the gain for each node in the current partition
calculate_gains(circuit, data)
# update the cutsize of the current partition
data.cutsize = calculate_cutsize(circuit, data)
logging.info("initial cutsize = {}".format(data.cutsize))
# intialize best partition, and prev mincut
data.store_best_cut()
data.prev_mincut = data.mincut
return data
def kl_inner_stop(circuit: Circuit, data: Data, app=None, genetic=False):
"""
it is the end of the current pass, restore the best cut of this pass;
terminates the outer loop if mincut doesn't improve or it reaches 6 iterations
"""
data.restore_best_cut() # restore block data structures
# update the distribution for the restored best cut
update_distribution(circuit, data)
# update the gains for the restored best cut
calculate_gains(circuit, data)
if genetic:
return
logging.info("iteration {}: best mincut = {}".format(
data.iteration, data.cutsize))
if app is not None:
app.update_canvas(data)
data.iteration += 1
# continue for up to 6 iterations or until mincut stops improving
if data.iteration <= 6 and data.mincut != data.prev_mincut:
data.prev_mincut = data.mincut
if app is not None:
app.root.after(1000, kl_inner_loop, circuit, data, app)
elif app is not None:
app.update_partition_button(True)
