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
// Copyright (c) 2014 baidu-rpc authors.
// Author: Ge,Jun (gejun@baidu.com)
#include <stdlib.h>
#include <unistd.h>
#include <stdio.h>
#include <signal.h>
#include <gtest/gtest.h>
#include "base/time.h"
#include "base/macros.h"
#include "base/errno.h"
#include <syscall.h>
#include <limits.h> // INT_MAX
#include "base/atomicops.h"
#include "bthread/bthread.h"
#include <bthread/sys_futex.h>
#include <bthread/processor.h>
namespace {
volatile bool stop = false;
base::atomic<int> nthread(0);
void* read_thread(void* arg) {
base::atomic<int>* m = (base::atomic<int>*)arg;
int njob = 0;
while (!stop) {
int x;
while (!stop && (x = *m) != 0) {
if (x > 0) {
while ((x = m->fetch_sub(1)) > 0) {
++njob;
const long start = base::cpuwide_time_ns();
while (base::cpuwide_time_ns() < start + 10000) {
}
if (stop) {
return new int(njob);
}
}
m->fetch_add(1);
} else {
cpu_relax();
}
}
++nthread;
bthread::futex_wait_private(m/*lock1*/, 0/*consumed_njob*/, NULL);
--nthread;
}
return new int(njob);
}
TEST(FutexTest, rdlock_performance) {
const size_t N = 100000;
base::atomic<int> lock1(0);
pthread_t rth[8];
for (size_t i = 0; i < ARRAY_SIZE(rth); ++i) {
ASSERT_EQ(0, pthread_create(&rth[i], NULL, read_thread, &lock1));
}
const size_t t1 = base::cpuwide_time_ns();
for (size_t i = 0; i < N; ++i) {
if (nthread) {
lock1.fetch_add(1);
bthread::futex_wake_private(&lock1, 1);
} else {
lock1.fetch_add(1);
if (nthread) {
bthread::futex_wake_private(&lock1, 1);
}
}
}
const size_t t2 = base::cpuwide_time_ns();
bthread_usleep(3000000);
stop = true;
for (int i = 0; i < 10; ++i) {
bthread::futex_wake_private(&lock1, INT_MAX);
sched_yield();
}
int njob = 0;
int* res;
for (size_t i = 0; i < ARRAY_SIZE(rth); ++i) {
pthread_join(rth[i], (void**)&res);
njob += *res;
delete res;
}
printf("wake %lu times, %ldns each, lock1=%d njob=%d\n",
N, (t2-t1)/N, lock1.load(), njob);
ASSERT_EQ(N, (size_t)(lock1.load() + njob));
}
TEST(FutexTest, futex_wake_before_wait) {
int lock1 = 0;
timespec timeout = { 1, 0 };
ASSERT_EQ(0, bthread::futex_wake_private(&lock1, INT_MAX));
ASSERT_EQ(-1, bthread::futex_wait_private(&lock1, 0, &timeout));
ASSERT_EQ(ETIMEDOUT, errno);
}
void* dummy_waiter(void* lock) {
bthread::futex_wait_private(lock, 0, NULL);
return NULL;
}
TEST(FutexTest, futex_wake_many_waiters_perf) {
int lock1 = 0;
size_t N = 0;
pthread_t th;
for (; N < 1000 && !pthread_create(&th, NULL, dummy_waiter, &lock1); ++N) {}
sleep(1);
int nwakeup = 0;
base::Timer tm;
tm.start();
for (size_t i = 0; i < N; ++i) {
nwakeup += bthread::futex_wake_private(&lock1, 1);
}
tm.stop();
printf("N=%lu, futex_wake a thread = %ldns\n", N, tm.n_elapsed() / N);
ASSERT_EQ(N, (size_t)nwakeup);
const size_t REP = 10000;
nwakeup = 0;
tm.start();
for (size_t i = 0; i < REP; ++i) {
nwakeup += bthread::futex_wake_private(&lock1, 1);
}
tm.stop();
ASSERT_EQ(0, nwakeup);
printf("futex_wake nop = %ldns\n", tm.n_elapsed() / REP);
}
base::atomic<int> nevent(0);
void* waker(void* lock) {
bthread_usleep(10000);
const size_t REP = 100000;
int nwakeup = 0;
base::Timer tm;
tm.start();
for (size_t i = 0; i < REP; ++i) {
nwakeup += bthread::futex_wake_private(lock, 1);
}
tm.stop();
EXPECT_EQ(0, nwakeup);
printf("futex_wake nop = %ldns\n", tm.n_elapsed() / REP);
return NULL;
}
void* batch_waker(void* lock) {
bthread_usleep(10000);
const size_t REP = 100000;
int nwakeup = 0;
base::Timer tm;
tm.start();
for (size_t i = 0; i < REP; ++i) {
if (nevent.fetch_add(1, base::memory_order_relaxed) == 0) {
nwakeup += bthread::futex_wake_private(lock, 1);
int expected = 1;
while (1) {
int last_expected = expected;
if (nevent.compare_exchange_strong(expected, 0, base::memory_order_relaxed)) {
break;
}
nwakeup += bthread::futex_wake_private(lock, expected - last_expected);
}
}
}
tm.stop();
EXPECT_EQ(0, nwakeup);
printf("futex_wake nop = %ldns\n", tm.n_elapsed() / REP);
return NULL;
}
TEST(FutexTest, many_futex_wake_nop_perf) {
pthread_t th[8];
int lock1;
std::cout << "[Direct wake]" << std::endl;
for (size_t i = 0; i < ARRAY_SIZE(th); ++i) {
ASSERT_EQ(0, pthread_create(&th[i], NULL, waker, &lock1));
}
for (size_t i = 0; i < ARRAY_SIZE(th); ++i) {
ASSERT_EQ(0, pthread_join(th[i], NULL));
}
std::cout << "[Batch wake]" << std::endl;
for (size_t i = 0; i < ARRAY_SIZE(th); ++i) {
ASSERT_EQ(0, pthread_create(&th[i], NULL, batch_waker, &lock1));
}
for (size_t i = 0; i < ARRAY_SIZE(th); ++i) {
ASSERT_EQ(0, pthread_join(th[i], NULL));
}
}
} // namespace