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
|
// Copyright 2023 The Chromium Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include <thread>
#include <vector>
#include "base/debug/allocation_trace.h"
#include "base/strings/stringprintf.h"
#include "base/timer/lap_timer.h"
#include "testing/gtest/include/gtest/gtest.h"
#include "testing/perf/perf_result_reporter.h"
namespace base {
namespace debug {
namespace {
// Change kTimeLimit to something higher if you need more time to capture a
// trace.
constexpr base::TimeDelta kTimeLimit = base::Seconds(3);
constexpr int kWarmupRuns = 100;
constexpr int kTimeCheckInterval = 1000;
constexpr char kMetricStackTraceDuration[] = ".duration_per_run";
constexpr char kMetricStackTraceThroughput[] = ".throughput";
enum class HandlerFunctionSelector { OnAllocation, OnFree };
// An executor to perform the actual notification of the recorder. The correct
// handler function is selected using template specialization based on the
// HandlerFunctionSelector.
template <HandlerFunctionSelector HandlerFunction>
struct HandlerFunctionExecutor {
void operator()(base::debug::tracer::AllocationTraceRecorder& recorder) const;
};
template <>
struct HandlerFunctionExecutor<HandlerFunctionSelector::OnAllocation> {
void operator()(
base::debug::tracer::AllocationTraceRecorder& recorder) const {
// Since the recorder just stores the value, we can use any value for
// address and size that we want.
recorder.OnAllocation(
&recorder, sizeof(recorder),
base::allocator::dispatcher::AllocationSubsystem::kPartitionAllocator,
nullptr);
}
};
template <>
struct HandlerFunctionExecutor<HandlerFunctionSelector::OnFree> {
void operator()(
base::debug::tracer::AllocationTraceRecorder& recorder) const {
recorder.OnFree(&recorder);
}
};
} // namespace
class AllocationTraceRecorderPerfTest
: public testing::TestWithParam<
std::tuple<HandlerFunctionSelector, size_t>> {
protected:
// The result data of a single thread. From the results of all the single
// threads the final results will be calculated.
struct ResultData {
TimeDelta time_per_lap;
float laps_per_second = 0.0;
int number_of_laps = 0;
};
// The data of a single test thread.
struct ThreadRunnerData {
std::thread thread;
ResultData result_data;
};
// Create and setup the result reporter.
const char* GetHandlerDescriptor(HandlerFunctionSelector handler_function);
perf_test::PerfResultReporter SetUpReporter(
HandlerFunctionSelector handler_function,
size_t number_of_allocating_threads);
// Select the correct test function which shall be used for the current test.
using TestFunction =
void (*)(base::debug::tracer::AllocationTraceRecorder& recorder,
ResultData& result_data);
static TestFunction GetTestFunction(HandlerFunctionSelector handler_function);
template <HandlerFunctionSelector HandlerFunction>
static void TestFunctionImplementation(
base::debug::tracer::AllocationTraceRecorder& recorder,
ResultData& result_data);
// The test management function. Using the the above auxiliary functions it is
// responsible to setup the result reporter, select the correct test function,
// spawn the specified number of worker threads and post process the results.
void PerformTest(HandlerFunctionSelector handler_function,
size_t number_of_allocating_threads);
};
const char* AllocationTraceRecorderPerfTest::GetHandlerDescriptor(
HandlerFunctionSelector handler_function) {
switch (handler_function) {
case HandlerFunctionSelector::OnAllocation:
return "OnAllocation";
case HandlerFunctionSelector::OnFree:
return "OnFree";
}
}
perf_test::PerfResultReporter AllocationTraceRecorderPerfTest::SetUpReporter(
HandlerFunctionSelector handler_function,
size_t number_of_allocating_threads) {
const std::string story_name = base::StringPrintf(
"(%s;%zu-threads)", GetHandlerDescriptor(handler_function),
number_of_allocating_threads);
perf_test::PerfResultReporter reporter("AllocationRecorderPerf", story_name);
reporter.RegisterImportantMetric(kMetricStackTraceDuration, "ns");
reporter.RegisterImportantMetric(kMetricStackTraceThroughput, "runs/s");
return reporter;
}
AllocationTraceRecorderPerfTest::TestFunction
AllocationTraceRecorderPerfTest::GetTestFunction(
HandlerFunctionSelector handler_function) {
switch (handler_function) {
case HandlerFunctionSelector::OnAllocation:
return TestFunctionImplementation<HandlerFunctionSelector::OnAllocation>;
case HandlerFunctionSelector::OnFree:
return TestFunctionImplementation<HandlerFunctionSelector::OnFree>;
}
}
void AllocationTraceRecorderPerfTest::PerformTest(
HandlerFunctionSelector handler_function,
size_t number_of_allocating_threads) {
perf_test::PerfResultReporter reporter =
SetUpReporter(handler_function, number_of_allocating_threads);
TestFunction test_function = GetTestFunction(handler_function);
base::debug::tracer::AllocationTraceRecorder the_recorder;
std::vector<ThreadRunnerData> notifying_threads;
notifying_threads.reserve(number_of_allocating_threads);
// Setup the threads. After creation, each thread immediately starts running.
// We expect the creation of the threads to be so quick that the delay from
// first to last thread is negligible.
for (size_t i = 0; i < number_of_allocating_threads; ++i) {
auto& last_item = notifying_threads.emplace_back();
last_item.thread = std::thread{test_function, std::ref(the_recorder),
std::ref(last_item.result_data)};
}
TimeDelta average_time_per_lap;
float average_laps_per_second = 0;
// Wait for each thread to finish and collect its result data.
for (auto& item : notifying_threads) {
item.thread.join();
// When finishing, each threads writes its results into result_data. So,
// from here we gather its performance statistics.
average_time_per_lap += item.result_data.time_per_lap;
average_laps_per_second += item.result_data.laps_per_second;
}
average_time_per_lap /= number_of_allocating_threads;
average_laps_per_second /= number_of_allocating_threads;
reporter.AddResult(kMetricStackTraceDuration, average_time_per_lap);
reporter.AddResult(kMetricStackTraceThroughput, average_laps_per_second);
}
template <HandlerFunctionSelector HandlerFunction>
void AllocationTraceRecorderPerfTest::TestFunctionImplementation(
base::debug::tracer::AllocationTraceRecorder& recorder,
ResultData& result_data) {
LapTimer timer(kWarmupRuns, kTimeLimit, kTimeCheckInterval,
LapTimer::TimerMethod::kUseTimeTicks);
HandlerFunctionExecutor<HandlerFunction> handler_executor;
timer.Start();
do {
handler_executor(recorder);
timer.NextLap();
} while (!timer.HasTimeLimitExpired());
result_data.time_per_lap = timer.TimePerLap();
result_data.laps_per_second = timer.LapsPerSecond();
result_data.number_of_laps = timer.NumLaps();
}
INSTANTIATE_TEST_SUITE_P(
,
AllocationTraceRecorderPerfTest,
::testing::Combine(::testing::Values(HandlerFunctionSelector::OnAllocation,
HandlerFunctionSelector::OnFree),
::testing::Values(1, 5, 10, 20, 40, 80)));
TEST_P(AllocationTraceRecorderPerfTest, TestNotification) {
const auto parameters = GetParam();
const HandlerFunctionSelector handler_function = std::get<0>(parameters);
const size_t number_of_threads = std::get<1>(parameters);
PerformTest(handler_function, number_of_threads);
}
} // namespace debug
} // namespace base
|
