affinity_partitioner

[algorithms.affinity_partitioner]

Hints that loop iterations should be assigned to threads in a way that optimizes for cache affinity.

An affinity_partitioner hints that execution of a loop template should use the same task affinity pattern for splitting the work as used by previous execution of the loop (or another loop) with the same affinity_partitioner object.

affinity_partitioner uses proportional splitting when it is enabled for a Range type.

Unlike the other partitioners, it is important that the same affinity_partitioner object be passed to the loop templates to be optimized for affinity.

The affinity_partitioner class satisfies the CopyConstructibe requirement from the ISO C++ [utility.arg.requirements] section.

// Defined in header <tbb/partitioner.h>

namespace tbb {

   class affinity_partitioner {
   public:
       affinity_partitioner() = default;
       ~affinity_partitioner() = default;
   };

}

See also: