-
Notifications
You must be signed in to change notification settings - Fork 1
/
lazy.py
262 lines (204 loc) · 6.52 KB
/
lazy.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
import threading
import time
import networkx as nx
_graph = nx.DiGraph()
debug = False
verbose = False
parallelize = False
def log(s):
if verbose:
print(s)
class Task(object):
def __init__(self, iter_base=0.1, iter_func=lambda x: x):
self.spins = []
self.iter_base = iter_base
self.iter_func = iter_func
def spin(self):
self.iter_base = 0.1 / (len(self.spins) + 1)
self.spins.append(0)
i = self.iter_base
while True:
time.sleep(i)
i = self.iter_func(i)
self.spins[-1] += 1
yield None
class Thread(threading.Thread):
def __init__(self, func, *args, **kwargs):
threading.Thread.__init__(self)
self.func = func
self.args = args
self.kwargs = kwargs
def run(self):
self.out = self.func(*self.args, **self.kwargs)
def _run_ops_in_parallel(ops):
log("running in parallel {}".format(ops))
threads = [Thread(op.run) for op in ops]
[t.start() for t in threads]
[t.join() for t in threads]
def _get_required_controlflow_graph(data):
nodes = nx.ancestors(_graph, data)
nodes.add(data)
needed = [n for n in nodes if n.type == "Data" and n.data == None]
log("need to resolve {}".format(needed))
controlflow = nx.DiGraph()
for n in needed:
for i in _graph.predecessors(n):
log("running {} to resolve {}".format(i, n))
controlflow.add_node(i)
# Check if we draw an edge to a consumer
# of the output of this node
for s in _graph.successors(n):
# Data output is data we need
for d in _graph.successors(s):
if d in needed:
controlflow.add_edge(i, s)
# fake a node to be root node (its a No-op)
root_op = Operation(lambda: None)
controlflow.add_node(root_op)
for node in controlflow.nodes:
if node == root_op:
continue
if len(controlflow.in_edges(node)) == 0:
controlflow.add_edge(root_op, node)
return controlflow
# Data acquisition
def _execute_graph(data):
log("executing graph")
controlflow = _get_required_controlflow_graph(data)
splits = []
for n, idom in nx.immediate_dominators(controlflow, root_op).items():
if n == idom:
continue
if n in controlflow.successors(idom):
continue
splits.append((idom, n))
# Graph mode
order = list(nx.topological_sort(controlflow))
iteration = 0
max_threads = 4
while iteration < len(order):
ops = []
for i in range(max_threads):
if iteration >= len(order):
break
next_op = order[iteration]
if sum([op in nx.ancestors(controlflow, next_op) for op in ops]) > 0:
break
ops.append(order[iteration])
iteration += 1
_run_ops_in_parallel(ops)
class Data(object):
@classmethod
def getId(cls, _impl=[-1]):
_impl[0] += 1
return _impl[0]
def __init__(self, d=None):
self.data = d
self.type = "Data"
self.id = self.getId()
self.executor = None
log("creating {}".format(self))
def __eq__(self, other):
return self.id == other.id and self.type == other.type
def __hash__(self):
return hash(self.id)
def get(self):
if self.data is None:
log("getting data")
if parallelize and self.data is None:
_execute_graph(self)
if self.executor and self.data is None:
self.executor(self)
else: # Default execution
if self.data is None:
for p in _graph.predecessors(self):
p.run()
return self.data
# Returns the calculated controlflow graph
def dump_cf(self):
return _get_required_controlflow_graph(self)
def set(self, data):
self.data = data
for d in nx.descendants(_graph, self):
if d.type == "Data":
d.data = None
def __repr__(self):
return "Data_{}".format(self.id)
class Operation(object):
@classmethod
def getId(cls, _impl=[-1]):
_impl[0] += 1
return _impl[0]
def __init__(self, f):
self.func = f
self.name = f.__name__
self.type = "Operation"
self.id = self.getId()
args = f.__code__.co_argcount
self.inputs = [Data() for d in range(args)]
self.output = Data()
log("creating {}".format(self))
def __eq__(self, other):
return self.id == other.id and self.type == other.type
def __hash__(self):
return hash(self.id)
def run(self):
log("running {}".format(self))
inputs = [x.get() for x in self.inputs]
self.output.data = self.func(*inputs)
def __repr__(self):
return "{}_{}".format(
self.name, self.id, ", ".join([str(i) for i in self.inputs]), self.output
)
def _insert_op_to_graph(op):
_graph.add_node(op)
for inp in op.inputs:
_graph.add_node(inp)
_graph.add_edge(inp, op)
_graph.add_edge(op, op.output)
# todo handle kwargs
def synchronous(fn):
def wrapper(*args):
op = Operation(fn)
for i in range(len(op.inputs)):
if not isinstance(args[i], Data):
op.inputs[i].data = args[i]
else:
op.inputs[i] = args[i]
_insert_op_to_graph(op)
return op.output
return wrapper
def asynchronous(fn):
def wrapper(*args):
op = Operation(fn)
op.inputs[0] = Data(Task())
for i in range(len(args)):
if not isinstance(args[i], Data):
op.inputs[i + 1].data = args[i]
else:
op.inputs[i + 1] = args[i]
_insert_op_to_graph(op)
return op.output
return wrapper
pygraphviz = None
plt = None
def _draw(G):
global pygraphviz
global plt
if not pygraphviz:
import pygraphviz
if not plt:
import matplotlib.pyplot as plt
log("num nodes {}".format(len(G)))
node_sizes = [400 for i in range(len(G))]
pos = nx.nx_agraph.graphviz_layout(G, prog="dot")
color = lambda x: "red" if x.type == "Operation" else "green" if x.data else "grey"
colors = [color(node) for node in G.nodes()]
nx.draw(G, pos, with_labels=True, node_size=node_sizes, node_color=colors)
ax = plt.gca()
ax.set_axis_off()
plt.show()
def draw():
_draw(_graph)
def dump(self):
return _graph