{"id":6061,"date":"2020-12-19T11:11:55","date_gmt":"2020-12-19T11:11:55","guid":{"rendered":"https:\/\/www.modernescpp.com\/index.php\/performancecomparison-of-condition-variables-and-atomics-in-c-20\/"},"modified":"2023-06-26T09:35:04","modified_gmt":"2023-06-26T09:35:04","slug":"performancecomparison-of-condition-variables-and-atomics-in-c-20","status":"publish","type":"post","link":"https:\/\/www.modernescpp.com\/index.php\/performancecomparison-of-condition-variables-and-atomics-in-c-20\/","title":{"rendered":"Performance Comparison of Condition Variables and Atomics in C++20"},"content":{"rendered":"

After the introduction to std::atomic_flag<\/code> in my last post, Synchronization with Atomics in C++20<\/a>, I want to dive deeper. Today, I created a ping-pong game using condition variables, std::atomic_flag<\/code> and std::atomic<bool><\/code>. Let’s play.<\/p>\n

<\/p>\n

 \"TimelineCpp20CoreLanguage\"<\/p>\n

The key question I want to answer in this post is: What is the fastest way to synchronize threads in C++20? I use in this post three different data types: std::condition_variable<\/code>, std::atomic_flag<\/code>, and std::atomic<bool><\/code>.<\/p>\n

To get comparable numbers, I implement a ping-pong game. One thread executes a ping<\/code> function, and the other thread a pong<\/code> function. For simplicity reasons, I call the thread executing the ping<\/code> function the ping thread and the other thread the pong thread. The ping thread waits for the notification of the pong threads and returns the notification to the pong thread. The game stops after 1,000,000 ball changes. I perform each game five times to get comparable performance numbers.<\/p>\n

I made my performance test with the brand new Visual Studio compiler because it already supports synchronization with atomics. Additionally, I compiled the examples with maximum optimization (\/Ox<\/code>).<\/p>\n

\"windowsCompiler\"<\/p>\n

Let me start with C++11.<\/p>\n<\/p>\n

Condition Variables<\/h2>\n

 <\/p>\n

<\/p>\n

\n
\/\/ pingPongConditionVariable.cpp<\/span>\r\n\r\n#include <condition_variable><\/span>\r\n#include <iostream><\/span>\r\n#include <atomic><\/span>\r\n#include <thread><\/span>\r\n\r\nbool<\/span> dataReady{false<\/span>};\r\n\r\nstd::<\/span>mutex mutex_;\r\nstd::<\/span>condition_variable condVar1;          \/\/ (1)<\/span>\r\nstd::<\/span>condition_variable condVar2;          \/\/ (2)<\/span>\r\n\r\nstd::<\/span>atomic<<\/span>int<\/span>><\/span> counter{};\r\nconstexpr int<\/span> countlimit =<\/span> 1<\/span>'<\/span>000<\/span>'<\/span>000<\/span>;\r\n\r\nvoid<\/span> ping<\/span>() {\r\n\r\n    while<\/span>(counter <=<\/span> countlimit) {\r\n        {\r\n            std::<\/span>unique_lock<<\/span>std::<\/span>mutex><\/span> lck(mutex_);\r\n            condVar1.wait(lck, []{return<\/span> dataReady ==<\/span> false<\/span>;});\r\n            dataReady =<\/span> true<\/span>;\r\n        }\r\n        ++<\/span>counter;                          <\/span>\r\n        condVar2.notify_one();              \/\/ (3)<\/span>\r\n  }\r\n}\r\n\r\nvoid<\/span> pong<\/span>() {\r\n\r\n    while<\/span>(counter <<\/span> countlimit) {  \r\n        {\r\n            std::<\/span>unique_lock<<\/span>std::<\/span>mutex><\/span> lck(mutex_);\r\n            condVar2.wait(lck, []{return<\/span> dataReady ==<\/span> true<\/span>;});\r\n            dataReady =<\/span> false<\/span>;\r\n        }\r\n        condVar1.notify_one();            \/\/ (3)<\/span>\r\n  }\r\n\r\n}\r\n\r\nint<\/span> main<\/span>(){\r\n\r\n    auto<\/span> start =<\/span> std::<\/span>chrono::<\/span>system_clock::<\/span>now();  \r\n\r\n    std::<\/span>thread<\/span> t1(ping);\r\n    std::<\/span>thread<\/span> t2(pong);\r\n\r\n    t1.join();\r\n    t2.join();\r\n  \r\n    std::<\/span>chrono::<\/span>duration<<\/span>double<\/span>><\/span> dur =<\/span> std::<\/span>chrono::<\/span>system_clock::<\/span>now() -<\/span> start;\r\n    std::<\/span>cout <<<\/span> \"Duration: \"<\/span> <<<\/span> dur.count() <<<\/span> \" seconds\"<\/span> <<<\/span> std::<\/span>endl;\r\n\r\n}\r\n<\/pre>\n<\/div>\n
 <\/p>\n
I use two condition variables in the program: condVar1<\/code> and condVar2 <\/code>(lines 1 and 2). The ping thread waits for the notification of condVar1<\/code> and sends its notification with condVar2<\/code>. dataReady<\/code> protects against spurious and lost wakeups (see “C++ Core Guidelines: Be Aware of the Traps of Condition Variables<\/a>“). The ping-pong game ends when counter<\/code> reaches the countlimit<\/code>. The nofication_one<\/code> calls (lines 3) and the counter are thread-safe and are, therefore, outside the critical region.<\/p>\n
Here are the numbers:<\/p>\n
\"pingPongConditionVariable\"<\/p>\n
The average execution time is 0.52 seconds.<\/p>\n
Porting this play to std::atomic_flags<\/code>‘s in C++20 is straightforward.<\/p>\n
std::atomic_flag<\/code><\/h2>\n
Here is the play using two atomic flags.<\/p>\n
Two Atomic Flags<\/h3>\n
In the following program, I replace the waiting on the condition variable with the waiting on the atomic flag and the notification of the condition variable with the setting of the atomic flag followed by the notification.<\/p>\n
<\/p>\n
\n
\/\/ pingPongAtomicFlags.cpp<\/span>\r\n\r\n#include <iostream><\/span>\r\n#include <atomic><\/span>\r\n#include <thread><\/span>\r\n\r\nstd::<\/span>atomic_flag condAtomicFlag1{};\r\nstd::<\/span>atomic_flag condAtomicFlag2{};\r\n\r\nstd::<\/span>atomic<<\/span>int<\/span>><\/span> counter{};\r\nconstexpr int<\/span> countlimit =<\/span> 1<\/span>'<\/span>000<\/span>'<\/span>000<\/span>;\r\n\r\nvoid<\/span> ping<\/span>() {\r\n    while<\/span>(counter <=<\/span> countlimit) {\r\n        condAtomicFlag1.wait(false<\/span>);               \/\/ (1)<\/span>\r\n        condAtomicFlag1.clear();                   \/\/ (2)<\/span>\r\n\r\n        ++<\/span>counter;\r\n        \r\n        condAtomicFlag2.test_and_set();           \/\/ (4)<\/span>\r\n        condAtomicFlag2.notify_one();             \/\/ (3)<\/span>\r\n    }\r\n}\r\n\r\nvoid<\/span> pong<\/span>() {\r\n    while<\/span>(counter <<\/span> countlimit) {\r\n        condAtomicFlag2.wait(false<\/span>);\r\n        condAtomicFlag2.clear();\r\n        \r\n        condAtomicFlag1.test_and_set();\r\n        condAtomicFlag1.notify_one();\r\n    }\r\n}\r\n\r\nint<\/span> main<\/span>() {\r\n\r\n     auto<\/span> start =<\/span> std::<\/span>chrono::<\/span>system_clock::<\/span>now();  \r\n\r\n    condAtomicFlag1.test_and_set();                    \/\/ (5)<\/span>\r\n    std::<\/span>thread<\/span> t1(ping);\r\n    std::<\/span>thread<\/span> t2(pong);\r\n\r\n    t1.join();\r\n    t2.join();\r\n\r\n    std::<\/span>chrono::<\/span>duration<<\/span>double<\/span>><\/span> dur =<\/span> std::<\/span>chrono::<\/span>system_clock::<\/span>now() -<\/span> start;\r\n    std::<\/span>cout <<<\/span> \"Duration: \"<\/span> <<<\/span> dur.count() <<<\/span> \" seconds\"<\/span> <<<\/span> std::<\/span>endl;\r\n\r\n}\r\n<\/pre>\n<\/div>\n
 <\/p>\n
A call condAtomicFlag1.wait(false)<\/code> (1) blocks, if the value of the atomic flag is false<\/code>. On the contrary, it returns if condAtomicFlag1<\/code> has the value true<\/code>. The boolean value serves as a kind of predicate and must, therefore, be set back to false<\/code> (2). Before the notification (3) is sent to the pong thread, condAtomicFlag1<\/code> is set to true <\/code>(4). The initial setting of condAtomicFlag1<\/code> to true<\/code> (5) starts the game. <\/span><\/p>\n
Thanks to std::atomic_flag<\/code> the game end earlier.<\/p>\n
\"pingPongAtomicFlags\"<\/p>\n
On average, a game takes 0.32 seconds.<\/p>\n
When you analyze the program, you may recognize that one atomics flag is sufficient for the play.<\/p>\n
One Atomic Flag<\/h3>\n
Using one atomic flag makes the play easier to understand.<\/p>\n
 <\/p>\n
<\/p>\n
\n
\/\/ pingPongAtomicFlag.cpp<\/span>\r\n\r\n#include <iostream><\/span>\r\n#include <atomic><\/span>\r\n#include <thread><\/span>\r\n\r\nstd::<\/span>atomic_flag condAtomicFlag{};\r\n\r\nstd::<\/span>atomic<<\/span>int<\/span>><\/span> counter{};\r\nconstexpr int<\/span> countlimit =<\/span> 1<\/span>'<\/span>000<\/span>'<\/span>000<\/span>;\r\n\r\nvoid<\/span> ping<\/span>() {\r\n    while<\/span>(counter <=<\/span> countlimit) {\r\n        condAtomicFlag.wait(true<\/span>);\r\n        condAtomicFlag.test_and_set();\r\n        \r\n        ++<\/span>counter;\r\n        \r\n        condAtomicFlag.notify_one();\r\n    }\r\n}\r\n\r\nvoid<\/span> pong<\/span>() {\r\n    while<\/span>(counter <<\/span> countlimit) {\r\n        condAtomicFlag.wait(false<\/span>);\r\n        condAtomicFlag.clear();\r\n        condAtomicFlag.notify_one();\r\n    }\r\n}\r\n\r\nint<\/span> main<\/span>() {\r\n\r\n     auto<\/span> start =<\/span> std::<\/span>chrono::<\/span>system_clock::<\/span>now();  \r\n\r\n    \r\n    condAtomicFlag.test_and_set();\r\n    std::<\/span>thread<\/span> t1(ping);\r\n    std::<\/span>thread<\/span> t2(pong);\r\n\r\n    t1.join();\r\n    t2.join();\r\n\r\n    std::<\/span>chrono::<\/span>duration<<\/span>double<\/span>><\/span> dur =<\/span> std::<\/span>chrono::<\/span>system_clock::<\/span>now() -<\/span> start;\r\n    std::<\/span>cout <<<\/span> \"Duration: \"<\/span> <<<\/span> dur.count() <<<\/span> \" seconds\"<\/span> <<<\/span> std::<\/span>endl;\r\n\r\n}\r\n<\/pre>\n<\/div>\n
 <\/p>\n
In this case, the ping thread blocks on true<\/code> but the pong thread blocks on false<\/code>. Using one or two atomic flags make no difference from the performance perspective.<\/p>\n
 \"pingPongAtomicFlag\"<\/p>\n
The average execution time is 0.31 seconds.<\/p>\n
 I used in this example std::atomic_flag<\/code> such as an atomic boolean. Let’s give it another try with std::atomic<bool><\/code>.<\/p>\n
std::atomic<bool><\/code><\/h2>\n
From the readability perspective, I prefer the following C++20 implementation based on std::atomic<bool>.<\/code><\/p>\n
 <\/p>\n
<\/p>\n
\n
\/\/ pingPongAtomicBool.cpp<\/span>\r\n\r\n#include <iostream><\/span>\r\n#include <atomic><\/span>\r\n#include <thread><\/span>\r\n\r\nstd::<\/span>atomic<<\/span>bool<\/span>><\/span> atomicBool{};\r\n\r\nstd::<\/span>atomic<<\/span>int<\/span>><\/span> counter{};\r\nconstexpr int<\/span> countlimit =<\/span> 1<\/span>'<\/span>000<\/span>'<\/span>000<\/span>;\r\n\r\nvoid<\/span> ping<\/span>() {\r\n    while<\/span>(counter <=<\/span> countlimit) {\r\n        atomicBool.wait(true<\/span>);\r\n        atomicBool.store(true<\/span>);\r\n\r\n        ++<\/span>counter;\r\n        \r\n        atomicBool.notify_one();\r\n    }\r\n}\r\n\r\nvoid<\/span> pong<\/span>() {\r\n    while<\/span>(counter <<\/span> countlimit) {\r\n        atomicBool.wait(false<\/span>);\r\n        atomicBool.store(false<\/span>);\r\n        atomicBool.notify_one();\r\n    }\r\n}\r\n\r\nint<\/span> main<\/span>() {\r\n\r\n    std::<\/span>cout <<<\/span> std::<\/span>boolalpha <<<\/span> std::<\/span>endl;\r\n\r\n    std::<\/span>cout <<<\/span> \"atomicBool.is_lock_free(): \"<\/span>              \/\/ (1)<\/span>\r\n              <<<\/span> atomicBool.is_lock_free()  <<<\/span> std::<\/span>endl; \r\n\r\n    std::<\/span>cout <<<\/span> std::<\/span>endl;\r\n\r\n    auto<\/span> start =<\/span> std::<\/span>chrono::<\/span>system_clock::<\/span>now();\r\n\r\n    atomicBool.store(true<\/span>);\r\n    std::<\/span>thread<\/span> t1(ping);\r\n    std::<\/span>thread<\/span> t2(pong);\r\n\r\n    t1.join();\r\n    t2.join();\r\n\r\n    std::<\/span>chrono::<\/span>duration<<\/span>double<\/span>><\/span> dur =<\/span> std::<\/span>chrono::<\/span>system_clock::<\/span>now() -<\/span> start;\r\n    std::<\/span>cout <<<\/span> \"Duration: \"<\/span> <<<\/span> dur.count() <<<\/span> \" seconds\"<\/span> <<<\/span> std::<\/span>endl;\r\n\r\n}\r\n<\/pre>\n<\/div>\n
 <\/p>\n
std::atomic<bool><\/code> can internally use a locking mechanism such as a mutex. As I assumed, my Windows runtime is lock-free (1).<\/p>\n
\"pingPongAtomicBool\"<\/p>\n
On average, the execution time is 0.38 seconds.<\/p>\n
All Numbers<\/h2>\n
As expected, condition variables are the slowest way, and atomic flag is the fastest way to synchronize threads. The performance of a std::atomic<bool><\/code> is in-between.<\/p>\n
\"PerformanceComparison\"<\/p>\n
What’s next?<\/h2>\n
With C++20, we have a few new mechanisms for thread coordination. In my next post<\/a>, I will look deeper into latches, barriers, and semaphores. They also allow it to play Ping-Pong.<\/p>\n
 <\/p>\n<\/p>\n
 <\/p>\n
 <\/p>\n
 <\/p>\n
 <\/div>\n","protected":false},"excerpt":{"rendered":"
After the introduction to std::atomic_flag in my last post, Synchronization with Atomics in C++20, I want to dive deeper. Today, I created a ping-pong game using condition variables, std::atomic_flag and std::atomic<bool>. Let’s play.<\/p>\n","protected":false},"author":21,"featured_media":5945,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[375],"tags":[434,451],"class_list":["post-6061","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-c-20","tag-atomics","tag-condition-variables"],"_links":{"self":[{"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/posts\/6061","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/users\/21"}],"replies":[{"embeddable":true,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/comments?post=6061"}],"version-history":[{"count":1,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/posts\/6061\/revisions"}],"predecessor-version":[{"id":6718,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/posts\/6061\/revisions\/6718"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/media\/5945"}],"wp:attachment":[{"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/media?parent=6061"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/categories?post=6061"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/tags?post=6061"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}