-
Notifications
You must be signed in to change notification settings - Fork 1
/
environment.py
222 lines (191 loc) · 6.58 KB
/
environment.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
import numpy as np
import subprocess
import random
import os
import struct
import parse
import socket
HOST = '127.0.0.1'
PORT = 1992
os.environ["CUDA_VISIBLE_DEVICES"] = "3"
sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
sock.bind((HOST, PORT))
class Player:
def __init__(self, log_dir="./data/log/"):
self._iter = 1
self._log_dir = log_dir
def reset(self):
return;
def step(self):
return;
class RandomPlayer(Player):
def __init__(self, log_dir):
Player.__init__(self, log_dir)
self._total_reward = 0.0
self._sock = sock
self._sock.listen(5)
f = open('rlconfig', 'r')
self._llc = f.readline().split(' ')[2].replace("\n", "")
self._src = f.readline().split(' ')[2].replace("\n", "")
self._target = f.readline().split(' ')[2].replace("\n", "")
f.close()
def reset(self):
self._iter = 1
self._p = subprocess.Popen([self._llc, '-debug-only=regallocdl', '--regalloc=drl', self._src, '-o', self._target],shell=False, stdout=subprocess.PIPE)
def step(self):
terminal = False
actions = set()
maxlength = 0
while (terminal == False):
print "start accept " + str(self._iter)
conn, addr = self._sock.accept()
data = conn.recv(1024)
if (data[0] == 'e'):
terminal = True
break
reward_map, sortedr, slotend = self.getState(data)
maxlength = max(slotend, maxlength)
action, score = self.doAction(sortedr)
conn.send(str(action))
self._iter = self._iter + 1
self._total_reward = self._total_reward + int(score)
for g in reward_map.keys():
actions.add(g)
print "the totalreward is " + str(self._total_reward)
self._iter = 1
self._total_reward = 0.0
self.terminate()
return actions, maxlength
def getState(self, data):
data = struct.unpack("!i", data)[0]
if int(data) != self._iter:
print "c++ iter: " + data + " python iter: " + str(self._iter)
sys.exit(0)
state, reward_map, sortedr, maxlength, _ = parse.fileToImage("state.txt", self._iter)
return reward_map, sortedr, maxlength
def doAction(self, reward_map):
action = reward_map[-1][0]
score = reward_map[-1][1]
return action, score
def terminate(self):
self._p.terminate()
self._p.wait()
print "process finish"
class Gplayer(Player):
def __init__(self, idx2regs, regs2idx, maxlength, tofile, log_dir):
Player.__init__(self, log_dir)
self._sock = sock
self._sock.listen(5)
self._idx2Regs = idx2regs
self._regs2idx = regs2idx
self._maxlength = maxlength + 1
self._actionsize = len(idx2regs)
self._tofile = tofile
f = open('rlconfig', 'r')
self._llc = f.readline().split(' ')[2].replace("\n", "")
self._src = f.readline().split(' ')[2].replace("\n", "")
self._target = f.readline().split(' ')[2].replace("\n", "")
f.close()
def terprocess(self):
print "terminal the process in python"
#self._p.terminate()
self._p.wait()
def reset(self):
self._p = subprocess.Popen([self._llc, '-debug-only=regallocdl', '--regalloc=drl', self._src, '-o', self._target],shell=False, stdout=subprocess.PIPE)
print "start accept " + str(self._iter)
self._iter = 1
self._conn, addr = self._sock.accept()
data = self._conn.recv(1024)
state, reward_map = self.getState(data)
return state, reward_map
def step(self, action):
#for log action data
self.log(action)
#prepare action and send action
action = self._idx2Regs[action]
self._conn.send(str(action))
self._iter = self._iter + 1
self._conn, addr = self._sock.accept()
data = self._conn.recv(1024)
if data[0] == 'e':
self.terprocess()
return [], True, []
#get the next state after do action
state, reward_map = self.getState(data)
return state, False, reward_map
def getState(self, data):
data = struct.unpack("!i", data)[0]
# unpack the socket data to test if it is terminated
if int(data) != self._iter:
print "c++ iter: " + data + " python iter: " + str(self._iter)
sys.exit(0)
# parse the state data which is outputed from compiler
state, reward_map, _ = parse.getstate("state.txt", self._iter, self._maxlength, self._actionsize, self._regs2idx, self._tofile)
return state, reward_map
# test the action is valid
def among(self, distri, reward_map, ac, valid):
ac = self._idx2Regs[ac]
if valid and reward_map.get(str(ac)) != None:
reward = reward_map[str(ac)]
ac = self._regs2idx[str(ac)]
return int(reward), ac, True
elif valid and reward_map.get(str(ac)) == None:
ac = self._regs2idx[str(ac)]
return 0.000001, ac, False
elif not valid and reward_map.get(str(ac)) != None:
reward = reward_map[str(ac)]
ac = self._regs2idx[str(ac)]
return int(reward), ac, True
# find the register which is in the candidate list
finalidx2reg = {}
index = 0
actions = []
for i in self._regs2idx.keys():
if reward_map.get(str(i)) != None:
actions.insert(index, distri[self._regs2idx[str(i)]])
finalidx2reg[str(index)] = i
index = index + 1
#the comment part is the epislon greedy
#if random.random() < 0.05:
actions = softmax(actions)
action = np.random.choice(index, 1, p=actions)
action = action[0]
#else:
# actions = softmax(actions)
# action = np.random.choice(index, 1, p=actions)
# action = action[0]
#action = np.argmax(np.array(actions))
action = finalidx2reg[str(action)]
reward = reward_map[str(action)]
action = self._regs2idx[str(action)]
return int(reward), action, True
def log(self, action):
name = self._log_dir + "action" + str(self._iter) + ".txt"
f = open(name, "w")
f.write(str(action))
f.close()
# helper function for simple reinforcement learning example
def among(distri, ac, valid):
reward_map = {"1": 3, "0": 5}
if valid and reward_map.get(str(ac)) != None:
return ac, True
elif valid and reward_map.get(str(ac)) == None:
return ac, False
elif not valid and reward_map.get(str(ac)) != None:
return ac, True
index = 2
actions = []
for i in range(index):
actions.append(distri[0][i])
#if random.random() < 0.0005:
actions = softmax(actions)
action = np.random.choice(index, 1, p=actions)
action = action[0]
#else:
#action = np.argmax(np.array(actions))
return action, True
def softmax(x):
x = x - np.max(x)
exp_x = np.exp(x)
softmax_x = exp_x / np.sum(exp_x)
return softmax_x