std::experimental::parallel:: reduce

The behavior is non-deterministic if binary_op is not associative or not commutative.

The behavior is undefined if binary_op modifies any element or invalidates any iterator in [ first , last ) .

Parameters

Return value

Generalized sum of init and * first , * ( first + 1 ) , ... * ( last - 1 ) over binary_op ,

where generalized sum GSUM(op, a 1 , ..., a N ) is defined as follows:

• if N=1 , a 1
• if N > 1 , op(GSUM(op, b 1 , ..., b K ), GSUM(op, b M , ..., b N )) where

• b 1 , ..., b N may be any permutation of a1, ..., aN and
• 1 < K+1 = M ≤ N

in other words, the elements of the range may be grouped and rearranged in arbitrary order.

Complexity

O(last - first) applications of binary_op .

Exceptions

• If execution of a function invoked as part of the algorithm throws an exception,

• if policy is parallel_vector_execution_policy , std::terminate is called.
• if policy is sequential_execution_policy or parallel_execution_policy , the algorithm exits with an exception_list containing all uncaught exceptions. If there was only one uncaught exception, the algorithm may rethrow it without wrapping in exception_list . It is unspecified how much work the algorithm will perform before returning after the first exception was encountered.
• if policy is some other type, the behavior is implementation-defined.

• If the algorithm fails to allocate memory (either for itself or to construct an exception_list when handling a user exception), std::bad_alloc is thrown.

Notes

If the range is empty, init is returned, unmodified.

• If policy is an instance of sequential_execution_policy , all operations are performed in the calling thread.
• If policy is an instance of parallel_execution_policy , operations may be performed in unspecified number of threads, indeterminately sequenced with each other.
• If policy is an instance of parallel_vector_execution_policy , execution may be both parallelized and vectorized: function body boundaries are not respected and user code may be overlapped and combined in arbitrary manner (in particular, this implies that a user-provided Callable must not acquire a mutex to access a shared resource).

Example

#include <chrono>
#include <experimental/execution_policy>
#include <experimental/numeric>
#include <iostream>
#include <numeric>
#include <vector>
 
int main()
{
    std::vector<double> v(10'000'007, 0.5);
 
    {
        auto t1 = std::chrono::high_resolution_clock::now();
        double result = std::accumulate(v.begin(), v.end(), 0.0);
        auto t2 = std::chrono::high_resolution_clock::now();
        std::chrono::duration<double, std::milli> ms = t2 - t1;
        std::cout << std::fixed << "std::accumulate result " << result
                  << " took " << ms.count() << " ms\n";
    }
 
    {
        auto t1 = std::chrono::high_resolution_clock::now();
        double result = std::experimental::parallel::reduce(
                            std::experimental::parallel::par,
                            v.begin(), v.end());
        auto t2 = std::chrono::high_resolution_clock::now();
        std::chrono::duration<double, std::milli> ms = t2 - t1;
        std::cout << "parallel::reduce result "
                  << result << " took " << ms.count() << " ms\n";
    }
}

std::accumulate result 5000003.50000 took 12.7365 ms
parallel::reduce result 5000003.50000 took 5.06423 ms

See also