...one of the most highly
regarded and expertly designed C++ library projects in the
world. — Herb Sutter and Andrei
// In header: <boost/compute/algorithm/reduce.hpp> template<typename InputIterator, typename OutputIterator, typename BinaryFunction> void reduce(InputIterator first, InputIterator last, OutputIterator result, BinaryFunction function, command_queue & queue = system::default_queue()); template<typename InputIterator, typename OutputIterator> void reduce(InputIterator first, InputIterator last, OutputIterator result, command_queue & queue = system::default_queue());
Returns the result of applying
function to the elements in the range [
If no function is specified,
plus will be used.
reduce() algorithm assumes that the binary reduction function is associative. When used with non-associative functions the result may be non-deterministic and vary in precision. Notably this affects the
plus<float>() function as floating-point addition is not associative and may produce slightly different results than a serial algorithm.
This algorithm supports both host and device iterators for the result argument. This allows for values to be reduced and copied to the host all with a single function call.
For example, to calculate the sum of the values in a device vector and copy the result to a value on the host:
Note that while the the
reduce() algorithm is conceptually identical to the
accumulate() algorithm, its implementation is substantially more efficient on parallel hardware. For more information, see the documentation on the
Space complexity on GPUs: \Omega(n)
Space complexity on CPUs: \Omega(1)