-
Notifications
You must be signed in to change notification settings - Fork 8
/
bench.cpp
334 lines (291 loc) · 7.95 KB
/
bench.cpp
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
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
#include <iostream>
#include <thread>
#include <string>
#include <vector>
#include <set>
#include <memory>
#include <unistd.h> // for sleep()
#include <getopt.h>
#include "policy.hpp"
#include "asm.hpp"
#include "rcu.hpp"
#include "timer.hpp"
using namespace std;
// some global variables - set defaults here
static int g_verbose = false;
static size_t g_nthreads = 1;
static uint64_t g_duration_sec = 10;
static void
_die(const char *filename,
const char *func,
int line,
const string &msg) __attribute__((noreturn));
static void
_die(const char *filename,
const char *func,
int line,
const string &msg)
{
cerr << filename << ":" << line << ": " << func << " - panic: " << msg << endl;
abort();
}
#define die(x) _die(__FILE__, __func__, __LINE__, x)
class worker {
friend class benchmark;
public:
worker(const string &name) : name(name), nops(0) {}
virtual ~worker() {}
protected:
static void
thread_fn(unique_ptr<worker> &w,
const atomic<bool> &start_flag,
const atomic<bool> &stop_flag)
{
while (!start_flag.load())
nop_pause();
w->run(stop_flag);
}
// returns the number of ops
virtual void run(const atomic<bool> &stop_flag) = 0;
string name;
atomic<size_t> nops;
};
class benchmark {
public:
virtual ~benchmark() {}
void
do_bench()
{
init();
auto workers = make_workers();
atomic<bool> start_flag(false);
atomic<bool> stop_flag(false);
vector<thread> thds;
for (auto &w : workers)
thds.emplace_back(worker::thread_fn, ref(w), ref(start_flag), ref(stop_flag));
start_flag.store(true);
timer t;
sleep(g_duration_sec);
stop_flag.store(true);
for (auto &t : thds)
t.join();
const uint64_t elasped_usec = t.lap();
const double elasped_sec = double(elasped_usec) / 1000000.0;
size_t agg_ops = 0;
for (auto &w : workers) {
if (g_verbose)
cout << w->name << " : " << double(w->nops)/elasped_sec << " ops/sec" << endl;
agg_ops += w->nops;
}
if (g_verbose)
cout << "total : " << double(agg_ops)/elasped_sec << " ops/sec" << endl;
else
// output for runner.py
cout << double(agg_ops)/elasped_sec << endl;
cleanup();
}
protected:
virtual void init() = 0;
virtual void cleanup() = 0;
virtual vector<unique_ptr<worker>> make_workers() = 0;
};
template <typename Impl>
class read_only_benchmark : public benchmark {
typedef linked_list<int, Impl> llist;
static const size_t NElems = 100;
class ro_worker : public worker {
public:
ro_worker(llist *list) : worker("reader"), list(list), nelems_seen(0) {}
inline size_t get_nelems_seen() const { return nelems_seen; }
protected:
void
run(const atomic<bool> &stop_flag) OVERRIDE
{
while (!stop_flag.load()) {
//vector<int> l(list->begin(), list->end());
//nelems_seen += l.size(); // so GCC doesn't optimize the vector away
nelems_seen += list->size();
nops++;
}
}
private:
llist *list;
size_t nelems_seen;
};
protected:
void
init() OVERRIDE
{
for (size_t i = 0; i < NElems; i++)
list.push_back(i);
}
void
cleanup() OVERRIDE
{
list.clear();
}
vector<unique_ptr<worker>>
make_workers() OVERRIDE
{
vector<unique_ptr<worker>> ret;
for (size_t i = 0; i < g_nthreads; i++)
ret.emplace_back(new ro_worker(&list));
return move(ret);
}
private:
llist list;
};
template <typename Impl>
class queue_benchmark : public benchmark {
typedef linked_list<int, Impl> llist;
static const size_t NElemsInitial = 100000;
class producer : public worker {
public:
producer(llist *list) : worker("producer"), list(list) {}
protected:
void
run(const atomic<bool> &stop_flag) OVERRIDE
{
while (!stop_flag.load()) {
list->push_back(1);
nops++;
}
}
private:
llist *list;
};
class consumer : public worker {
public:
consumer(llist *list) : worker("consumer"), list(list), nelems_popped(0) {}
inline size_t get_nelems_popped() const { return nelems_popped; }
protected:
void
run(const atomic<bool> &stop_flag) OVERRIDE
{
while (!stop_flag.load()) {
auto ret = list->try_pop_front();
if (ret.first)
nelems_popped++;
nops++; // count regardless of removal or not
}
}
private:
llist *list;
size_t nelems_popped;
};
protected:
void
init() OVERRIDE
{
for (size_t i = 0; i < NElemsInitial; i++)
list.push_back(i);
}
void
cleanup() OVERRIDE
{
list.clear();
}
vector<unique_ptr<worker>>
make_workers() OVERRIDE
{
vector<unique_ptr<worker>> ret;
for (size_t i = 0; i < g_nthreads / 2; i++)
ret.emplace_back(new producer(&list));
for (size_t i = g_nthreads / 2; i < g_nthreads; i++)
ret.emplace_back(new consumer(&list));
return move(ret);
}
private:
llist list;
};
int
main(int argc, char **argv)
{
#ifndef NDEBUG
cerr << "Warning: benchmarks being run w/ assertions" << endl;
#endif
string bench_type = "readonly";
string policy_type = "global_lock";
for (;;) {
static struct option long_options[] =
{
{"verbose", no_argument, &g_verbose, 1 },
{"bench", required_argument, 0, 'b'},
{"policy", required_argument, 0, 'p'},
{"num-threads", required_argument, 0, 't'},
{"runtime", required_argument, 0, 'r'},
{0, 0, 0, 0}
};
int option_index = 0;
int c = getopt_long(argc, argv, "vb:t:r:", long_options, &option_index);
if (c == -1)
break;
switch (c) {
case 0:
if (long_options[option_index].flag != 0)
break;
abort();
break;
case 'b':
bench_type = optarg;
break;
case 'p':
policy_type = optarg;
break;
case 't':
g_nthreads = strtoul(optarg, NULL, 10);
if (g_nthreads <= 0)
die("need --num-threads > 0");
break;
case 'r':
g_duration_sec = strtoul(optarg, NULL, 10);
if (g_duration_sec <= 0)
die("need --runtime > 0");
break;
case '?':
/* getopt_long already printed an error message. */
break;
default:
abort();
}
}
const set<string> valid_bench_types =
{"readonly", "queue"};
const set<string> valid_policy_types =
{"global_lock", "per_node_lock", "lock_free", "lock_free_rcu"};
if (!valid_bench_types.count(bench_type))
die("invalid --bench");
if (!valid_policy_types.count(policy_type))
die("invalid --policy");
unique_ptr<benchmark> p;
// XXX(stephentu): there must be a better way to do this, but leave it for
// now
if (bench_type == "readonly") {
if (policy_type == "global_lock")
p.reset(new read_only_benchmark<typename ll_policy<int>::global_lock>);
else if (policy_type == "per_node_lock")
p.reset(new read_only_benchmark<typename ll_policy<int>::per_node_lock>);
else if (policy_type == "lock_free")
p.reset(new read_only_benchmark<typename ll_policy<int>::lock_free>);
else if (policy_type == "lock_free_rcu")
p.reset(new read_only_benchmark<typename ll_policy<int>::lock_free_rcu>);
} else if (bench_type == "queue") {
if (policy_type == "global_lock")
p.reset(new queue_benchmark<typename ll_policy<int>::global_lock>);
else if (policy_type == "per_node_lock")
p.reset(new queue_benchmark<typename ll_policy<int>::per_node_lock>);
else if (policy_type == "lock_free")
p.reset(new queue_benchmark<typename ll_policy<int>::lock_free>);
else if (policy_type == "lock_free_rcu")
p.reset(new queue_benchmark<typename ll_policy<int>::lock_free_rcu>);
}
if (g_verbose) {
cout << "bench configuration:" << endl
<< " bench : " << bench_type << endl
<< " policy : " << policy_type << endl
<< " num-threads: " << g_nthreads << endl
<< " runtime : " << g_duration_sec << " sec" << endl;
}
p->do_bench();
return 0;
}