-
Notifications
You must be signed in to change notification settings - Fork 159
/
parallel.cpp
273 lines (232 loc) · 8.46 KB
/
parallel.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
#include "parallel.h"
#include <list>
#include <thread>
#include <condition_variable>
#include <vector>
#include <cassert>
// From https://github.com/mmp/pbrt-v3/blob/master/src/core/parallel.cpp
static std::vector<std::thread> threads;
static bool shutdownThreads = false;
struct ParallelForLoop;
static ParallelForLoop *workList = nullptr;
static std::mutex workListMutex;
struct ParallelForLoop {
ParallelForLoop(std::function<void(int64_t)> func1D, int64_t maxIndex, int chunkSize)
: func1D(std::move(func1D)), maxIndex(maxIndex), chunkSize(chunkSize) {
}
ParallelForLoop(const std::function<void(Vector2i)> &f, const Vector2i count)
: func2D(f), maxIndex(count[0] * count[1]), chunkSize(1) {
nX = count[0];
}
std::function<void(int64_t)> func1D;
std::function<void(Vector2i)> func2D;
const int64_t maxIndex;
const int chunkSize;
int64_t nextIndex = 0;
int activeWorkers = 0;
ParallelForLoop *next = nullptr;
int nX = -1;
bool Finished() const {
return nextIndex >= maxIndex && activeWorkers == 0;
}
};
void Barrier::Wait() {
std::unique_lock<std::mutex> lock(mutex);
assert(count > 0);
if (--count == 0) {
// This is the last thread to reach the barrier; wake up all of the
// other ones before exiting.
cv.notify_all();
} else {
// Otherwise there are still threads that haven't reached it. Give
// up the lock and wait to be notified.
cv.wait(lock, [this] { return count == 0; });
}
}
static std::condition_variable workListCondition;
static void worker_thread_func(const int tIndex, std::shared_ptr<Barrier> barrier) {
ThreadIndex = tIndex;
// The main thread sets up a barrier so that it can be sure that all
// workers have called ProfilerWorkerThreadInit() before it continues
// (and actually starts the profiling system).
barrier->Wait();
// Release our reference to the Barrier so that it's freed once all of
// the threads have cleared it.
barrier.reset();
std::unique_lock<std::mutex> lock(workListMutex);
while (!shutdownThreads) {
if (!workList) {
// Sleep until there are more tasks to run
workListCondition.wait(lock);
} else {
// Get work from _workList_ and run loop iterations
ParallelForLoop &loop = *workList;
// Run a chunk of loop iterations for _loop_
// Find the set of loop iterations to run next
int64_t indexStart = loop.nextIndex;
int64_t indexEnd = std::min(indexStart + loop.chunkSize, loop.maxIndex);
// Update _loop_ to reflect iterations this thread will run
loop.nextIndex = indexEnd;
if (loop.nextIndex == loop.maxIndex)
workList = loop.next;
loop.activeWorkers++;
// Run loop indices in _[indexStart, indexEnd)_
lock.unlock();
for (int64_t index = indexStart; index < indexEnd; ++index) {
if (loop.func1D) {
loop.func1D(index);
}
// Handle other types of loops
else {
assert(loop.func2D != nullptr);
loop.func2D(Vector2i{int(index % loop.nX),
int(index / loop.nX)});
}
}
lock.lock();
// Update _loop_ to reflect completion of iterations
loop.activeWorkers--;
if (loop.Finished()) {
workListCondition.notify_all();
}
}
}
}
void parallel_for_host(const std::function<void(int64_t)> &func,
int64_t count,
int chunkSize) {
// Run iterations immediately if not using threads or if _count_ is small
if (threads.empty() || count < chunkSize) {
for (int64_t i = 0; i < count; ++i) {
func(i);
}
return;
}
// Create and enqueue _ParallelForLoop_ for this loop
ParallelForLoop loop(func, count, chunkSize);
workListMutex.lock();
loop.next = workList;
workList = &loop;
workListMutex.unlock();
// Notify worker threads of work to be done
std::unique_lock<std::mutex> lock(workListMutex);
workListCondition.notify_all();
// Help out with parallel loop iterations in the current thread
while (!loop.Finished()) {
// Run a chunk of loop iterations for _loop_
// Find the set of loop iterations to run next
int64_t indexStart = loop.nextIndex;
int64_t indexEnd = std::min(indexStart + loop.chunkSize, loop.maxIndex);
// Update _loop_ to reflect iterations this thread will run
loop.nextIndex = indexEnd;
if (loop.nextIndex == loop.maxIndex) {
workList = loop.next;
}
loop.activeWorkers++;
// Run loop indices in _[indexStart, indexEnd)_
lock.unlock();
for (int64_t index = indexStart; index < indexEnd; ++index) {
if (loop.func1D) {
loop.func1D(index);
}
// Handle other types of loops
else {
assert(loop.func2D != nullptr);
loop.func2D(Vector2i{int(index % loop.nX),
int(index / loop.nX)});
}
}
lock.lock();
// Update _loop_ to reflect completion of iterations
loop.activeWorkers--;
}
}
thread_local int ThreadIndex;
void parallel_for_host(
std::function<void(Vector2i)> func, const Vector2i count) {
// Launch worker threads if needed
if (threads.empty() || count.x * count.y <= 1) {
for (int y = 0; y < count.y; ++y) {
for (int x = 0; x < count.x; ++x) {
func(Vector2i{x, y});
}
}
return;
}
ParallelForLoop loop(std::move(func), count);
{
std::lock_guard<std::mutex> lock(workListMutex);
loop.next = workList;
workList = &loop;
}
std::unique_lock<std::mutex> lock(workListMutex);
workListCondition.notify_all();
// Help out with parallel loop iterations in the current thread
while (!loop.Finished()) {
// Run a chunk of loop iterations for _loop_
// Find the set of loop iterations to run next
int64_t indexStart = loop.nextIndex;
int64_t indexEnd = std::min(indexStart + loop.chunkSize, loop.maxIndex);
// Update _loop_ to reflect iterations this thread will run
loop.nextIndex = indexEnd;
if (loop.nextIndex == loop.maxIndex) {
workList = loop.next;
}
loop.activeWorkers++;
// Run loop indices in _[indexStart, indexEnd)_
lock.unlock();
for (int64_t index = indexStart; index < indexEnd; ++index) {
if (loop.func1D) {
loop.func1D(index);
}
// Handle other types of loops
else {
assert(loop.func2D != nullptr);
loop.func2D(Vector2i{int(index % loop.nX),
int(index / loop.nX)});
}
}
lock.lock();
// Update _loop_ to reflect completion of iterations
loop.activeWorkers--;
}
}
int num_system_cores() {
// return 1;
int ret = std::thread::hardware_concurrency();
if (ret == 0) {
return 16;
}
return ret;
}
void parallel_init() {
assert(threads.size() == 0);
int nThreads = num_system_cores();
ThreadIndex = 0;
// Create a barrier so that we can be sure all worker threads get past
// their call to ProfilerWorkerThreadInit() before we return from this
// function. In turn, we can be sure that the profiling system isn't
// started until after all worker threads have done that.
std::shared_ptr<Barrier> barrier = std::make_shared<Barrier>(nThreads);
// Launch one fewer worker thread than the total number we want doing
// work, since the main thread helps out, too.
for (int i = 0; i < nThreads - 1; ++i) {
threads.push_back(std::thread(worker_thread_func, i + 1, barrier));
}
barrier->Wait();
}
void parallel_cleanup() {
if (threads.empty()) {
return;
}
{
std::lock_guard<std::mutex> lock(workListMutex);
shutdownThreads = true;
workListCondition.notify_all();
}
for (std::thread &thread : threads) {
thread.join();
}
threads.erase(threads.begin(), threads.end());
shutdownThreads = false;
}