Boost C++ Libraries

...one of the most highly regarded and expertly designed C++ library projects in the world. Herb Sutter and Andrei Alexandrescu, C++ Coding Standards

PrevUpHomeNext

Parallel - Fork-Join -- EXPERIMENTAL

Fork-Join
Reference -- EXPERIMENTAL
[Warning] Warning

These features are experimental and subject to change in future versions. There are not too much tests yet, so it is possible that you can find out some trivial bugs :(

[Note] Note

These features are based on the n4088 - Task Region R3 C++1y proposal from P. Halpern, A. Robison, A. Laksberg, H. Sutter, et al. The text that follows has been adapted from this paper to show the differences.

The major difference respect to the standard proposal is that we are able to use a common executor for several task regions.

This module introduces a C++11/c++14 library function template task_region and a library class task_region_handle with member functions run and wait that together enable developers to write expressive and portable fork-join parallel code.

The working draft for the Parallelism TS N4105 augments the STL algorithms with the inclusion of parallel execution policies. Programmers use these as a basis to write additional high-level algorithms that can be implemented in terms of the provided parallel algorithms. However, the scope of n4105 does not include lower-level mechanisms to express arbitrary fork-join parallelism

The task_region, run and the wait functions provided by this library are based on the task_group concept that is a part of the common subset of the PPL and the TBB libraries.

Consider an example of a parallel traversal of a tree, where a user-provided function compute is applied to each node of the tree, returning the sum of the results:

template<typename Func>
int traverse(node *n, Func&& compute)
{
  int left = 0, right = 0;
  task_region([&](task_region_handle& tr) {
    if (n->left)
      tr.run([&] { left = traverse(n->left, compute); });
    if (n->right)
      tr.run([&] { right = traverse(n->right, compute); });
  });
  return compute(n) + left + right;
}

The example above demonstrates the use of two of the functions proposed in this paper, task_region and task_region_handle::run. The task_region function delineates a region in a program code potentially containing invocations of tasks spawned by the run member function of the task_region_handle class.

The run function spawns a task, a unit of work that is allowed to execute in parallel with respect to the caller. Any parallel tasks spawned by run within the task_region are joined back to a single thread of execution at the end of the task_region.

run takes a user-provided function object f and starts it asynchronously - i.e. it may return before the execution of f completes. The implementation's scheduler may choose to run f immediately or delay running f until compute resources become available.

A task_region_handle can be constructed only by task_region because it has no public constructors. Thus, run can be invoked (directly or indirectly) only from a user-provided function passed to task_region:

void g();
void f(task_region_handle& tr)
{
  tr.run(g); // OK, invoked from within task_region in h
}
void h()
{
  task_region(f);
}

int main()
{
  task_region_handle tr; // Error: no public constructor
  tr.run(g); // No way to call run outside of a task_region
  return 0;
}

This is surely the worst implementation of the Fibonacci function. Anyway, here it is, as it is simple and shows the fork-join structure clearly. Fibonacci(n) = Fibonacci(n-1) + Fibonacci(n-2), so the task decomposition is trivial.

int fib_task_region(int n)
{
  using boost::experimental::parallel::task_region;
  using boost::experimental::parallel::task_region_handle;

  if (n == 0) return 0;
  if (n == 1) return 1;

  int n1;
  int n2;

  task_region([&](task_region_handle& trh)
      {
        trh.run([&]
            {
              n1 = fib_task_region(n - 1);
            });

        n2 = fib_task_region(n - 2);
      });

  return n1 + n2;
}

int main()
{
  for (int i = 0; i<10; ++i) {
    std::cout << fib_task_region(i) << " ";
  }
  std::cout << std::endl;
}

The previous example make use of an implementation defined way to spawn the tasks. Often the user wants to master how the task must be spawned. There is an overload of task_region that accept an additional Executor parameter and a function that takes as parameter a task_region_handle_gen<Executor>. task_region_handle_gen<Executor> run uses this executor to spawn the tasks.

template <class Ex>
int fib_task_region_gen( Ex& ex, int n)
{
  using boost::experimental::parallel::task_region;
  using boost::experimental::parallel::task_region_handle_gen;

  if (n == 0) return 0;
  if (n == 1) return 1;

  int n1;
  int n2;

  task_region(ex, [&](task_region_handle_gen<Ex>& trh) // (2)
      {
        trh.run([&]
            {
              n1 = fib_task_region(n - 1);
            });

        n2 = fib_task_region(n - 2);
      });

  return n1 + n2;
}

int main()
{
  boost::basic_thread_pool tp; // (1)
  for (int i = 0; i<10; ++i) {
    std::cout << fib_task_region_gen(tp,i) << " ";
  }
  std::cout << std::endl;
  return 0;
}

The specific executor is declared in line (1) and it is used in line (2).

namespace boost
{
namespace experimental
{
namespace parallel
{
inline namespace v1
{

  class exception_list;

} // v1
} // parallel
} // experimental
} // boost
namespace boost
{
namespace experimental
{
namespace parallel
{
inline namespace v1
{

  class exception_list: public std::exception
  {
  public:
    typedef 'implementation defined' const_iterator;

    ~exception_list() noexcept {}

    void add(exception_ptr const& e);
    size_t size() const noexcept;
    const_iterator begin() const noexcept;
    const_iterator end() const noexcept;
    const char* what() const noexcept;

  };

} // v1
} // parallel
} // experimental
} // boost
namespace boost
{
namespace experimental
{
namespace parallel
{
inline namespace v2
{

  class task_canceled_exception;

  template <class Executor>
  class task_region_handle_gen;

  using default_executor = 'implementation defined';

  class task_region_handle;

  template <typename Executor, typename F>
    void task_region_final(Executor& ex, F&& f);
  template <typename F>
    void task_region_final(F&& f);

  template <typename Executor, typename F>
    void task_region(Executor& ex, F&& f);
  template <typename F>
    void task_region(F&& f);

} // v2
} // parallel
} // experimental
} // boost
namespace boost
{
namespace experimental
{
namespace parallel
{
inline namespace v2
{

  class task_canceled_exception: public std::exception
  {
  public:
    task_canceled_exception() noexcept;
    task_canceled_exception(const task_canceled_exception&) noexcept;
    task_canceled_exception& operator=(const task_canceled_exception&) noexcept;
    virtual const char* what() const noexcept;
  };

} // v2
} // parallel
} // experimental
} // boost
namespace boost
{
namespace experimental
{
namespace parallel
{
inline namespace v2
{

  template <class Executor>
  class task_region_handle_gen
  {
  protected:
    task_region_handle_gen(Executor& ex);

    ~task_region_handle_gen();

  public:
    task_region_handle_gen(const task_region_handle_gen&) = delete;
    task_region_handle_gen& operator=(const task_region_handle_gen&) = delete;
    task_region_handle_gen* operator&() const = delete;

    template<typename F>
    void run(F&& f);

    void wait();
  };

} // v2
} // parallel
} // experimental
} // boost
namespace boost
{
namespace experimental
{
namespace parallel
{
inline namespace v2
{

  using default_executor = 'implementation defined';

} // v2
} // parallel
} // experimental
} // boost
namespace boost
{
namespace experimental
{
namespace parallel
{
inline namespace v2
{

  class task_region_handle :
    public task_region_handle_gen<default_executor>
  {
  protected:
    task_region_handle();
    task_region_handle(const task_region_handle&) = delete;
    task_region_handle& operator=(const task_region_handle&) = delete;
    task_region_handle* operator&() const = delete;

  };

} // v2
} // parallel
} // experimental
} // boost
namespace boost
{
namespace experimental
{
namespace parallel
{
inline namespace v2
{

  template <typename Executor, typename F>
    void task_region_final(Executor& ex, F&& f);
  template <typename F>
    void task_region_final(F&& f);

} // v2
} // parallel
} // experimental
} // boost
namespace boost
{
namespace experimental
{
namespace parallel
{
inline namespace v2
{

  template <typename Executor, typename F>
    void task_region(Executor& ex, F&& f);
  template <typename F>
    void task_region(F&& f);

} // v2
} // parallel
} // experimental
} // boost

PrevUpHomeNext