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
|
#! /usr/bin/python
# Copyright (c) 2021 Samsung Electronics Co., Ltd. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import numpy as np
import sys
import Queue
import utils
import signal
from pareto import ParetoData
class Hlps:
"""
Initialize Runner and Pareto data structure
"""
def __init__(self, runner, num_backends, num_samples):
self._runner = runner
self._num_backends = num_backends
self._num_samples = num_samples
self._marked = {}
self._extended_search = False
self._iteration = 0
self._pareto_obj = ParetoData()
"""
Method to generate new samples from a given sample v_vec.
The new samples bear a hamming distance hd from the provided sample.
"""
def gen_hamming(self, v_vec, hd=1, nsamples=None):
if nsamples is None:
nsamples = self._num_backends - 1
ret = np.zeros((nsamples, len(v_vec)), dtype=int)
v = v_vec
marked = np.full(len(v), False, dtype=bool)
cnt = 0
for r in range(nsamples):
ret[r] = v
rnd_pos = np.random.permutation(range(len(v)))
for i in range(hd):
pos = rnd_pos[i]
marked[pos] = True
for r in range(nsamples):
ret[r][pos] = (v[pos] - r - 1) % self._num_backends
return ret
"""
Method to generate all samples from a given sample v_vec, that
have a hamming distance of one with respect to it.
"""
def gen_hamming_one(self, v_vec, invert=False):
ret = np.zeros(((self._num_backends - 1) * len(v_vec), len(v_vec)), dtype=int)
if invert == False:
v = v_vec
else:
v = [1 - x for x in v_vec]
for nb in range(1, self._num_backends):
c = 0
for r in range((nb - 1) * len(v), nb * len(v)):
ret[r] = v
ret[r][c] = (v[c] - nb) % self._num_backends
c += 1
return ret
"""
Enable profiling over extended search space
"""
def enable_extended_search(self):
self._extended_search = True
for key in self._pareto_obj.get_pareto_keys():
config = self._pareto_obj.get_config(key)
extended_val = self._runner.get_extended_solution(config)
self._pareto_obj.set_config(key, extended_val)
self._iteration = 0
"""
HLPS algorithm implementation provided here.
Description: Starting with a random sample, fill up a sampling
queue with hamming neighbors. Fetch samples from queue,
each time checking for pareto optimality. Pareto-optimal samples
are then explored/exploited to generate new samples that are added to the queue.
Algorithm phase terminates when the queue is empty.
Repeat this phase in a multi-shot invokation for better results.
"""
def hlps_routine(self, config_ids):
# Initialize
solution_q = Queue.Queue()
visited = {}
nbits = self._runner.get_nbits(self._extended_search)
is_extended = self._runner.get_mode_extended()
nsolutions = self._num_backends**nbits
stop_insert = False
cnt = 0
q_add_cnt = 0
round_cnt = 0
def extended_solution(s):
return self._runner.get_extended_solution(s)
def mark_solution(s):
if is_extended == True and self._extended_search == False:
self._marked[extended_solution(s)] = True
else:
self._marked[s] = True
def is_marked(s):
if is_extended == True and self._extended_search == False:
return (extended_solution(s) in self._marked)
else:
return (s in self._marked)
def visit_solution(s):
if is_extended == True and self._extended_search == False:
visited[extended_solution(s)] = True
else:
visited[s] = True
def is_visited(s):
if is_extended == True and self._extended_search == False:
return (extended_solution(s) in visited)
else:
return (s in visited)
def sigint_handler(signum, frame):
print("Round cnt = ", round_cnt)
signal.signal(signal.SIGINT, sigint_handler)
if len(config_ids) > 0:
for solution in config_ids:
if is_extended == True and self._extended_search == True and self._iteration == 0:
s = extended_solution(solution)
else:
s = solution
s_vec = utils.int_to_vec(s, self._num_backends, nbits)
candidate = self.gen_hamming_one(s_vec)
for hd in range((self._num_backends - 1) * nbits):
candidate_int = int(''.join(str(x) for x in reversed(candidate[hd])),
self._num_backends)
if is_marked(candidate_int) == False:
solution_q.put(candidate_int)
mark_solution(candidate_int)
q_add_cnt += 1
else:
start_seed = int(np.random.rand() * (nsolutions))
solution_q.put(start_seed)
q_add_cnt += 1
self._iteration += 1
# Main routine
while not solution_q.empty():
s = solution_q.get()
mark_solution(s)
stop_insert = False
if (round_cnt % 100 == 0):
print("sample count = ", round_cnt)
if self._extended_search == True:
print("Queue size is ", solution_q.qsize())
if is_extended == True and self._extended_search == False:
time_val, memory_val = self._runner.profile_by_opname(s)
elif is_extended == True:
time_val, memory_val = self._runner.profile_by_opindex(s)
else:
time_val, memory_val = self._runner.profile_by_opname(s)
round_cnt += 1
utils.progressbar(round_cnt, nsolutions, prefix="% samples computed. : ")
self._pareto_obj.update_pareto_solutions(
s, time_val, memory_val, explore_flag=True)
for key in self._pareto_obj.get_pareto_keys():
pareto_sample = self._pareto_obj.get_config(key)
explore_sample = self._pareto_obj.get_exploration(key)
if is_visited(pareto_sample):
continue
visit_solution(pareto_sample)
s_vec = utils.int_to_vec(pareto_sample, self._num_backends, nbits)
if explore_sample == True:
# Explore solutions over a larger range
for hd in range(1, nbits + 1):
if stop_insert is True:
break
candidate = self.gen_hamming(s_vec, hd=hd)
for i in range(self._num_backends - 1):
if stop_insert is True:
break
candidate_int = int(
''.join(str(x) for x in reversed(candidate[i])),
self._num_backends)
try:
if is_marked(candidate_int) == False:
solution_q.put(candidate_int)
q_add_cnt += 1
except IndexError:
print("candidate[i] = ", candidate[i],
', candidate_int = ', candidate_int)
sys.exit(-1)
if (q_add_cnt >= self._num_samples):
print("Queue full in explore")
stop_insert = True
else:
# Exploit solutions within immediate neighborhood
candidate = self.gen_hamming_one(s_vec)
for j in range((self._num_backends - 1) * nbits):
if stop_insert is True:
break
candidate_int = int(
''.join(str(x) for x in reversed(candidate[j])),
self._num_backends)
if is_marked(candidate_int) == False:
solution_q.put(candidate_int)
q_add_cnt += 1
if (q_add_cnt >= self._num_samples):
print("Queue full in exploit")
stop_insert = True
self._pareto_obj.set_exploration(key)
pfront = set([
self._pareto_obj.get_config(key)
for key in self._pareto_obj.get_pareto_keys()
])
return pfront, q_add_cnt
"""
Method to dump results from HLPS
"""
def dump_results(self, dumpdata):
dumpdata = self._pareto_obj.dump_pareto_solutions(dumpdata)
dumpdata = self._runner.dump_config(dumpdata)
return dumpdata
|
