Learn C++ Multi-Threading in 5 Minutes

Summary: A crash course on the C++14 multi-threading constructs in a very non-verbose manner

The new C++ multi-threading constructs are very easy to learn. If you are familiar with C or C++ and want to start writing multithreaded programs, this article is for you!

I use C++14 as a reference, but what I describe is also supported in C++17 . I only cover common constructs. You should be able to write your own multithreaded programs after reading this.

Update (March 2020):

I created a video on this subject. You can watch it here:

Creating Threads

A thread can be created in several ways:

1. Using a function pointer
2. Using a functor
3. Using a lambda function

These methods are very similar with minor differences. I explain each method and their differences next.

Using a function pointer

Consider the following function which takes a vector reference

v

acm

v

beginIndex

endIndex

void accumulator_function2(const std::vector<int> &v, unsigned long long &acm, 
                            unsigned int beginIndex, unsigned int endIndex)
{
    acm = 0;
    for (unsigned int i = beginIndex; i < endIndex; ++i)
    {
        acm += v[i];
    }
}

A function calculating the sum of all elements between beginIndex and endIndex in a vector v

Now lets say you want to partition the vector in two sections and calculate the total sum of each section in a separate thread

t1

t2

//Pointer to function
    {
        unsigned long long acm1 = 0;
        unsigned long long acm2 = 0;
        std::thread t1(accumulator_function2, std::ref(v), 
                        std::ref(acm1), 0, v.size() / 2);
        std::thread t2(accumulator_function2, std::ref(v), 
                        std::ref(acm2), v.size() / 2, v.size());
        t1.join();
        t2.join();

        std::cout << "acm1: " << acm1 << endl;
        std::cout << "acm2: " << acm2 << endl;
        std::cout << "acm1 + acm2: " << acm1 + acm2 << endl;
    }

Creating threads using function pointers

What do you need to take away?

1. std::thread

   creates a new thread. The first parameter is the name of the function pointer

   accumulator_function2

   . Therefore, each thread will execute this function.
2. The rest of the parameters passed to

   std::thread

   constructor are the parameters that we need to pass to

   accumulator_function2

   .
3. Important: All parameters passed to

   accumulator_function2

   are passed by value unless you wrap them in

   std::ref.

   That’s why we wrapped

   v

    ,

   acm1

    , and

   acm2

   in

   std::ref

   .
4. Threads created by

   std::thread 

   do not have return values. If you want to return something, you should store it in one of the parameters passed by reference, i.e.

   acm

   .
5. Each thread starts as soon as it gets created.
6. We use

   join() 

   function to wait for a thread to finish

Using Functors

You can do exactly the same thing using functors. The following is the code that uses a functor:

class CAccumulatorFunctor3
{
  public:
    void operator()(const std::vector<int> &v, 
                    unsigned int beginIndex, unsigned int endIndex)
    {
        _acm = 0;
        for (unsigned int i = beginIndex; i < endIndex; ++i)
        {
            _acm += v[i];
        }
    }
    unsigned long long _acm;
};

Functor Definition

And the code that creates the threads is:

//Creating Thread using Functor
    {

        CAccumulatorFunctor3 accumulator1 = CAccumulatorFunctor3();
        CAccumulatorFunctor3 accumulator2 = CAccumulatorFunctor3();
        std::thread t1(std::ref(accumulator1), 
            std::ref(v), 0, v.size() / 2);
        std::thread t2(std::ref(accumulator2), 
            std::ref(v), v.size() / 2, v.size());
        t1.join();
        t2.join();

        std::cout << "acm1: " << accumulator1._acm << endl;
        std::cout << "acm2: " << accumulator2._acm << endl;
        std::cout << "accumulator1._acm + accumulator2._acm : " << 
            accumulator1._acm + accumulator2._acm << endl;
    }

Creating threads using functors

What do you need to take away?

Everything is very similar to function pointer, except that:

1. The first parameter is the functor object.
2. Instead of passing a reference to the functor to store the result, we can store its return value in a member variable inside the functor, i.e. in

   _acm

   .

Using Lambda Functions

As the third alternative we can define each thread in a lambda function as shown below:

{
        unsigned long long acm1 = 0;
        unsigned long long acm2 = 0;
        std::thread t1([&acm1, &v] {
            for (unsigned int i = 0; i < v.size() / 2; ++i)
            {
                acm1 += v[i];
            }
        });
        std::thread t2([&acm2, &v] {
            for (unsigned int i = v.size() / 2; i < v.size(); ++i)
            {
                acm2 += v[i];
            }
        });
        t1.join();
        t2.join();

        std::cout << "acm1: " << acm1 << endl;
        std::cout << "acm2: " << acm2 << endl;
        std::cout << "acm1 + acm2: " << acm1 + acm2 << endl;
    }

Creating threads using lambda functions

Again, everything is very similar to function pointer, except that:

1. As an alternative to pass a parameter, we can pass references to lambda functions using lambda capture.

Tasks, Futures, and Promises

As an alternative to

std::thread

Tasks work very similar to threads, but the main difference is that they can return a value. So, you can remember them as a more abstract way of defining your threads and use them when the threads return a value.

Below is the same example written using tasks:

#include <future>
//Tasks, Future, and Promises
    {
        auto f1 = [](std::vector<int> &v, 
            unsigned int left, unsigned int right) {
            unsigned long long acm = 0;
            for (unsigned int i = left; i < right; ++i)
            {
                acm += v[i];
            }

            return acm;
        };

        auto t1 = std::async(f1, std::ref(v), 
            0, v.size() / 2);
        auto t2 = std::async(f1, std::ref(v), 
            v.size() / 2, v.size());

        //You can do other things here!
        unsigned long long acm1 = t1.get();
        unsigned long long acm2 = t2.get();

        std::cout << "acm1: " << acm1 << endl;
        std::cout << "acm2: " << acm2 << endl;
        std::cout << "acm1 + acm2: " << acm1 + acm2 << endl;
    }

What do you need to take away?

1. Tasks are defined and created using

   std::async

    , (instead of threads that are created using

   std::thread

   )
2. The returned value from

   std::async

   is called a

   std::future

   . Don’t get scared by its name. It just means

   t1

   and

   t2

   are variables whose value will be assigned to in the future. We get their values by calling

   t1.get() 

   and

   t2.get()

3. If the future values are not ready, upon calling

   get() 

   the main thread blocks until the future value becomes ready (similar to

   join()

   ).
4. Notice that the function that we passed to

   std::async

   returns a value. This value is passed through a type called std::promise. Again, don’t get scared by its name. For the most part, you don’t need to know details of

   std::promise

   or define any variable of type

   std::promise

   . The C++ library does that behind the scenes.
5. Each task by default starts as soon as it is created (there is a way to change this which I don’t cover).

Summary of Creating Threads

There you have it. Creating threads is as simple as what I explained above. You can either use

std::thread

1. Use function pointers
2. Use functors
3. Use lambda functions

Or you can use

std::async

task

std::future

Shared Memory and Shared Resources

In short, threads should be careful when they read/write into shared memory and resources (such as files) to avoid race conditions.

C++14 provides several constructs to synchronize threads to avoid such race conditions.

Using Mutex, lock,() and unlock() (Not recommended)

The following code shows how we create a critical section such that each thread accesses

std::cout

std::mutex g_display_mutex;
thread_function()
{

    g_display_mutex.lock();
    std::thread::id this_id = std::this_thread::get_id();
    std::cout << "My thread id is: " << this_id  << endl;
    g_display_mutex.unlock();
}

What do you need to take away?

1. A mutex is created

   std::mutex

2. A critical section (i.e. guaranteed to be run only by a single thread at each time) is created using

   lock()

3. The critical section ends upon calling

   unlock()

4. Each thread waits at

   lock() 

   and only enters the critical section if no other thread is inside that section.

While the above method works, it is not recommended because:

1. It is not exception safe: if the code before lock generates an exception,

   unlock()

   will not be executed, and we never release the mutex which might cause deadlock
2. We always have to be careful not to forget to call

   unlock()

Using std::lock_guard (recommended)

Don’t get scared by its name

lock_guard

Below is the same critical section using lock_guard:

std::mutex g_display_mutex;
thread_function()
{
    std::lock_guard<std::mutex> guard(g_display_mutex);
    std::thread::id this_id = std::this_thread::get_id();
    std::cout << "From thread " << this_id  << endl;
}

critical section using lock_guard

What do you need to take away?

1. The code coming after std::lock_guard creation is automatically locked. No need for explicit

   lock()

   and

   unlock()

   function calls.
2. The critical section automatically ends when

   std::lock_guard

   goes out of scope. This makes it exception safe, and also we don’t need to remember to call

   unlock()

3. lock_guard

   still requires using a variable of type

   std::mutex

   in its constructor.

How Many Threads Should We Create?

You can create as many threads as you want, but it would probably be pointless if the number of active threads is more than the number of available CPU cores.

In order to get the maximum number of cores you can call:

std::thread::hardware_cuncurrency()

{
    unsigned int c = std::thread::hardware_concurrency();
    std::cout << " number of cores: " << c << endl;;
}

What I Didn’t Cover

I covered most of what you need to create threads. There are several other details that are less common which I don’t include here, but you can study them on your own:

1. std::move
2. details of std::promise
3. std::packaged_task
4. Conditional variables

Hope this helps you learning C++ multi threading quickly.

If you liked this article please click on the clap and give me feedback.