-ffast-math

[lgarithm]

#include <array>
#include <cstdio>

template <typename T> struct vec3 {
    std::array<T, 3> _val;
};

template <typename T> inline double dot3d(const vec3<T> &p, const vec3<T> &q)
{
    T t = 0;
    for (int i = 0; i < 3; ++i) { t += p._val[i] * q._val[i]; }
    return t;
}

void test_1()
{
    using T = double;
    vec3<T> p1 = {1, 2, 3};
    vec3<T> p2 = {4, 5, 6};
    double s = 0;
    const int n = 1000000000;
    for (int i = 0; i < n; ++i) {
        double l = dot3d(p1, p2);
        s += l;
    }
    printf("%f\n", s);
}

int main()
{
    test_1();
    return 0;
}

clang++ can speed up by X25 with -fffast-math

[lgarithm]

objdump -d shows the optimized code used addpd instead of addsd