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
|
// Copyright The OpenTelemetry Authors
// SPDX-License-Identifier: Apache-2.0
#include <benchmark/benchmark.h>
#include <atomic>
#include <cstdint>
#include <thread>
#include <vector>
#if defined(__i386__) || defined(__x86_64__)
# if defined(__clang__) || defined(__INTEL_COMPILER)
# include <emmintrin.h> // for _mm_pause
# endif
#endif
#include "opentelemetry/common/macros.h"
#include "opentelemetry/common/spin_lock_mutex.h"
namespace
{
using opentelemetry::common::SpinLockMutex;
constexpr int TightLoopLocks = 10000;
// Runs a thrash-test where we spin up N threads, each of which will
// attempt to lock-mutate-unlock a total of `TightLoopLocks` times.
//
// lock: A lambda denoting how to lock. Accepts a reference to `SpinLockType`.
// unlock: A lambda denoting how to unlock. Accepts a reference to `SpinLockType`.
template <typename SpinLockType, typename LockF, typename UnlockF>
inline void SpinThrash(benchmark::State &s, SpinLockType &spinlock, LockF lock, UnlockF unlock)
{
auto num_threads = s.range(0);
// Value we will increment, fighting over a spinlock.
// The contention is meant to be brief, as close to our expected
// use cases of "updating pointers" or "pushing an event onto a buffer".
std::int64_t value OPENTELEMETRY_MAYBE_UNUSED = 0;
std::vector<std::thread> threads;
threads.reserve(num_threads);
// Timing loop
for (auto _ : s)
{
for (auto i = 0; i < num_threads; i++)
{
threads.emplace_back([&] {
// Increment value once each time the lock is acquired. Spin a few times
// to ensure maximum thread contention.
for (int i = 0; i < TightLoopLocks; i++)
{
lock(spinlock);
value++;
unlock(spinlock);
}
});
}
// Join threads
for (auto &thread : threads)
thread.join();
threads.clear();
}
}
// Benchmark of full spin-lock implementation.
static void BM_SpinLockThrashing(benchmark::State &s)
{
SpinLockMutex spinlock;
SpinThrash(s, spinlock, [](SpinLockMutex &m) { m.lock(); }, [](SpinLockMutex &m) { m.unlock(); });
}
// Naive `while(try_lock()) {}` implementation of lock.
static void BM_NaiveSpinLockThrashing(benchmark::State &s)
{
SpinLockMutex spinlock;
SpinThrash(
s, spinlock,
[](SpinLockMutex &m) {
while (!m.try_lock())
{
// Left this comment to keep the same format on old and new versions of clang-format
}
},
[](SpinLockMutex &m) { m.unlock(); });
}
// Simple `while(try_lock()) { yield-processor }`
static void BM_ProcYieldSpinLockThrashing(benchmark::State &s)
{
SpinLockMutex spinlock;
SpinThrash<SpinLockMutex>(
s, spinlock,
[](SpinLockMutex &m) {
while (!m.try_lock())
{
#if defined(_MSC_VER)
YieldProcessor();
#elif defined(__i386__) || defined(__x86_64__)
# if defined(__clang__) || defined(__INTEL_COMPILER)
_mm_pause();
# else
__builtin_ia32_pause();
# endif
#elif defined(__armel__) || defined(__ARMEL__)
asm volatile("nop" ::: "memory");
#elif defined(__arm__) || defined(__aarch64__) // arm big endian / arm64
__asm__ __volatile__("yield" ::: "memory");
#endif
}
},
[](SpinLockMutex &m) { m.unlock(); });
}
// SpinLock thrashing with thread::yield().
static void BM_ThreadYieldSpinLockThrashing(benchmark::State &s)
{
#if defined(__cpp_lib_atomic_value_initialization) && \
__cpp_lib_atomic_value_initialization >= 201911L
std::atomic_flag mutex{};
#else
std::atomic_flag mutex = ATOMIC_FLAG_INIT;
#endif
SpinThrash<std::atomic_flag>(
s, mutex,
[](std::atomic_flag &l) {
uint32_t try_count = 0;
while (l.test_and_set(std::memory_order_acq_rel))
{
++try_count;
if (try_count % 32)
{
std::this_thread::yield();
}
}
std::this_thread::yield();
},
[](std::atomic_flag &l) { l.clear(std::memory_order_release); });
}
// Run the benchmarks at 2x thread/core and measure the amount of time to thrash around.
BENCHMARK(BM_SpinLockThrashing)
->RangeMultiplier(2)
->Range(1, std::thread::hardware_concurrency())
->MeasureProcessCPUTime()
->UseRealTime()
->Unit(benchmark::kMillisecond);
BENCHMARK(BM_ProcYieldSpinLockThrashing)
->RangeMultiplier(2)
->Range(1, std::thread::hardware_concurrency())
->MeasureProcessCPUTime()
->UseRealTime()
->Unit(benchmark::kMillisecond);
BENCHMARK(BM_NaiveSpinLockThrashing)
->RangeMultiplier(2)
->Range(1, std::thread::hardware_concurrency())
->MeasureProcessCPUTime()
->UseRealTime()
->Unit(benchmark::kMillisecond);
BENCHMARK(BM_ThreadYieldSpinLockThrashing)
->RangeMultiplier(2)
->Range(1, std::thread::hardware_concurrency())
->MeasureProcessCPUTime()
->UseRealTime()
->Unit(benchmark::kMillisecond);
} // namespace
BENCHMARK_MAIN();
|
