Intel Threading Building Blocks

Name

auto_partitioner Class — A class that models an adaptive partitioner that monitors the work-stealing actions of the task_scheduler to manage the number of splits performed.

Synopsis

#include "tbb/partitioner.h"

class auto_partitioner;

Description

The class auto_partitioner models an adaptive partitioner that limits the number of splits needed for load balancing by reacting to work-stealing events.

The range is first divided into S_I subranges, where S_I is proportional to the number of threads created by the task scheduler. These subranges are executed to completion by tasks unless they are stolen. If a subrange is stolen by an idle thread, the auto_partitioner further subdivides the range to create additional subranges.

The auto_partitioner creates additional subranges only if threads are actively stealing work. If the load is well balanced, the use of only a few large initial subranges reduces the overheads incurred when splitting and joining ranges. However, if there is a load imbalance that results in work stealing, the auto_partitioner creates additional subranges that can be stolen to more finely balance the load.

The auto_partitioner therefore attempts to minimize the number of range splits, while providing ample opportunities for work stealing.

auto_partitioner()

An empty default constructor.

auto_partitioner (auto_partitioner & partitioner, split)

A splitting constructor that divides the auto_partitioner partitioner into two partitioners.

template<typenameRange > bool should_execute_range ...