CODEWITHSUNDEEP
stlc++17
mlund
mlund

Reputation: 78

C++17 parallel algorithms on containers with object data

I am attempting to use C++17 parallel algorithms with containers holding non-fundamental types as illustrated in the minimal example below, compiled using GCC9.2.1/Intel TBB on Ubuntu 19.10. The sequential policy is OK, but compilation fails with par since the lambda is expected to accept double as the second argument. The issue persists on icc 19.0.1. My question is if the code is valid or if this issue is simply because of the early development stage of the parallel implementation?

#include <numeric>
#include <algorithm>
#include <execution>
#include <vector>

struct Data {
    double radius;
};

int main() {
    double sum;
    std::vector<double> v1;
    std::vector<Data> v2;

    // ok
    sum = std::reduce(std::execution::par, v1.begin(), v1.end(), 0.0, [](double sum, auto i) { return sum + i; });

    // ok
    sum = std::reduce(std::execution::seq, v2.begin(), v2.end(), 0.0, [](double sum, const Data &i) { return sum + i.radius; });

    // compile error
    sum = std::reduce(std::execution::par, v2.begin(), v2.end(), 0.0, [](double sum, const Data &i) { return sum + i.radius; });
}

Upvotes: 1

Views: 139

Answers (1)

rafix07
rafix07

Reputation: 20969

BinaryOp for std::reduce should be commutative, below both operations should be supported:

double + Data // your lambda supports only this 
Data + double // this can be performed only by adding some conversions

If you want to make conversion double->Data you should add proper constructor. For conversion Data->double you should add proper conversion operator:

struct Data {
    double radius;

    // double -> Data
    Data (double d) : radius(d) {}

    // Data -> double
    operator double() const {
        return radius;
    }
};

Live demo

Upvotes: 2

Related Questions