-
Notifications
You must be signed in to change notification settings - Fork 0
/
test.cpp
259 lines (213 loc) · 8.03 KB
/
test.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
/*
Copyright 2020 Florian Lemaitre
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
#include <mutex>
#include <thread>
#include <random>
#include <cstdint>
#include <cstdio>
#include <x86intrin.h>
#include "dsurcu.h"
struct Counter {
std::atomic<uint64_t> time{};
std::vector<uint64_t>* writer_data = nullptr;
};
template <class rcu, class Derived>
struct Worker {
using RCU = rcu;
std::atomic<Counter*>* time;
std::vector<uint64_t>* writer_data;
long niter;
double read_ratio;
Worker(std::atomic<Counter*>& time, std::vector<uint64_t>& writer_data, long niter, double read_ratio) : time(&time), writer_data(&writer_data), niter(niter), read_ratio(read_ratio) {}
void consume(std::atomic<uint64_t>*) const = delete;
static void reclaim(Counter*) {}
__attribute__((noinline))
void operator()() const {
double reads = 0., accesses = 1.;
for (long i = 0; i < niter; i += 1) {
Counter* p;
if (reads <= accesses * read_ratio) {
RCU::read_lock();
p = time->load(std::memory_order_relaxed);
static_cast<Derived const*>(this)->consume(&p->time);
RCU::read_unlock();
reads += 1.;
} else {
p = dsurcu::rcu_allocate<Counter>();
p->time.store(_rdtsc(), std::memory_order_relaxed);
p->writer_data = writer_data;
p = time->exchange(p, std::memory_order_acq_rel);
dsurcu::Node::get(p)->reclaim = [](dsurcu::Node* node) {
Derived::reclaim(node->data<Counter>());
dsurcu::Node::del<Counter>(node);
};
RCU::queue(p);
RCU::quiescent();
}
accesses += 1.;
}
}
};
template <class RCU>
struct WorkerRead : public Worker<RCU, WorkerRead<RCU>> {
using Worker<RCU, WorkerRead<RCU>>::Worker;
using Worker<RCU, WorkerRead<RCU>>::operator=;
void consume(std::atomic<uint64_t>* p) const {
uint64_t dummy = p->load(std::memory_order_relaxed);
asm (""::"r"(dummy));
}
};
template <class RCU>
struct WorkerWrite : public Worker<RCU, WorkerWrite<RCU>> {
using Worker<RCU, WorkerWrite<RCU>>::Worker;
using Worker<RCU, WorkerWrite<RCU>>::operator=;
void consume(std::atomic<uint64_t>* p) const {
uint64_t zero = 0;
asm("":"+rm"(zero));
p->store(zero, std::memory_order_relaxed);
}
};
template <class RCU>
struct WorkerRDTSC : public Worker<RCU, WorkerRDTSC<RCU>> {
using Worker<RCU, WorkerRDTSC<RCU>>::Worker;
using Worker<RCU, WorkerRDTSC<RCU>>::operator=;
void consume(std::atomic<uint64_t>* p) const {
p->store(_rdtsc(), std::memory_order_relaxed);
}
static void reclaim(Counter* p) {
uint64_t t1 = _rdtsc();
uint64_t t0 = p->time.load(std::memory_order_acquire);
p->writer_data->emplace_back(t1 - t0);
}
};
template <class Worker>
void bench(int nthreads, long niter, double read_ratio) {
using RCU = typename Worker::RCU;
RCU::init();
uint64_t rcu_wait = 0, rcu_busy = 0, rcu_end = 0, rcu_iter = 0;
// RCU thread: when there is some tasks to complete, wait for readers to finish their critical sections
std::atomic<bool> stop{false};
std::thread rcu([&]() {
do {
uint64_t t0 = _rdtsc();
RCU::process_queue_wait();
uint64_t t1 = _rdtsc();
RCU::process_queue_busy();
uint64_t t2 = _rdtsc();
rcu_wait += t1 - t0;
rcu_busy += t2 - t1;
rcu_iter += 1;
// stop is required
} while (!stop.load());
rcu_end = _rdtsc();
});
std::vector<uint64_t> writer_data;
writer_data.reserve(niter * nthreads * std::max(0., 1. - read_ratio) + nthreads + 1);
std::vector<std::thread> workers;
std::vector<uint64_t> worker_data(nthreads);
workers.reserve(nthreads);
std::atomic<Counter*> time;
{ Counter* p = dsurcu::rcu_allocate<Counter>();
p->time.store(_rdtsc(), std::memory_order_relaxed);
p->writer_data = &writer_data;
time.store(p, std::memory_order_relaxed);
}
std::atomic<int> barrier{nthreads};
for (int t = 0; t < nthreads; ++t) {
workers.emplace_back([&, t]() {
RCU::thread_init();
Worker work(time, writer_data, niter, read_ratio);
barrier.fetch_sub(1, std::memory_order_release);
int b;
do {
_mm_pause();
b = barrier.load(std::memory_order_relaxed);
} while (b > 0);
std::atomic_thread_fence(std::memory_order_acquire);
uint64_t t0 = _rdtsc();
work();
uint64_t t1 = _rdtsc();
RCU::thread_fini();
worker_data[t] = t1 - t0;
});
}
// wait for all workers to finish
for (auto& w : workers) {
w.join();
}
// stop the RCU thread
stop.store(true);
RCU::wake();
rcu.join();
RCU::fini();
Counter* p = time.load(std::memory_order_acquire);
uint64_t t0 = p->time.load(std::memory_order_relaxed);
uint64_t t1 = rcu_end;
dsurcu::rcu_deallocate(p);
double lateness = t1 - t0;
printf(" RCU: %lu times (wait: % 6lg\tbusy: % 6lg\t% 6lg%%)\n", rcu_iter, double(rcu_wait) / double(rcu_iter), double(rcu_busy) / double(rcu_iter), 100. * double(rcu_busy) / double(rcu_busy + rcu_wait));
if (worker_data.size() > 0) {
int n = worker_data.size();
std::sort(worker_data.begin(), worker_data.end());
double min = double(worker_data.front()) / double(niter);
double max = double(worker_data.back()) / double(niter);
double med = double(worker_data[0.5*n]) / double(niter);
uint64_t s1 = 0;
for (uint64_t delay : worker_data) {
s1 += delay;
}
double avg = double(s1) / double(n * niter);
printf(" processing time: % 6lg\t(min: % 6lg\tmed: % 6lg\tmax: % 6lg)\n", avg, min, med, max);
}
if (writer_data.size() > 2) {
long n = writer_data.size();
std::sort(writer_data.begin(), writer_data.end());
double min = writer_data.front();
double max = writer_data.back();
double p50 = writer_data[0.50*n];
double p90 = writer_data[0.90*n];
double p99 = writer_data[0.99*n];
uint64_t s1 = 0, s2 = 0;
for (uint64_t delay : writer_data) {
s1 += delay;
s2 += delay*delay;
}
double avg = double(s1) / double(n);
double var = (double(s2) - double(s1*s1) / double(n)) / double(n);
double err = std::sqrt(var);
printf(" Delay stat (%ld writes):\n", n);
printf(" avg: % 6lg +- % 6lg\n", avg, err);
printf(" min: % 6lg\n", min);
printf(" 50%%: % 6lg\n", p50);
printf(" 90%%: % 6lg\n", p90);
printf(" 99%%: % 6lg\n", p99);
printf(" max: % 6lg\n", max);
printf(" lateness: % 6lg\n", lateness);
}
}
template <class RCU>
void bench_rcu(int nthreads, long niter, float ratio) {
printf(" Read:\n");
bench<WorkerRead<RCU>>(nthreads, niter, ratio);
printf(" Write:\n");
bench<WorkerWrite<RCU>>(nthreads, niter, ratio);
printf(" RDTSC:\n");
bench<WorkerRDTSC<RCU>>(nthreads, niter, ratio);
}
int main() {
int nthreads = 10;
long niter = 1e8;
double ratio = 0.99;
printf("Noop:\n");
bench_rcu<dsurcu::Noop>(nthreads, niter, ratio);
printf("\nLocalEpochWeak:\n");
bench_rcu<dsurcu::LocalEpochWeak>(nthreads, niter, ratio);
printf("\nGlobalEpoch:\n");
bench_rcu<dsurcu::GlobalEpoch>(nthreads, niter, ratio);
printf("\nLocalEpoch:\n");
bench_rcu<dsurcu::LocalEpoch>(nthreads, niter, ratio);
}