T operator()(I1 begin1, S1 end1, I2 begin2, S2 end2, T init, BOp1 bop1_ = BOp1{},
BOp2 bop2_ = BOp2{}, P1 proj1_ = P1{}, P2 proj2_ = P2{}) const
{
auto &&bop1 = as_function(bop1_);
auto &&bop2 = as_function(bop2_);
auto &&proj1 = as_function(proj1_);
auto &&proj2 = as_function(proj2_);
for (; begin1 != end1 && begin2 != end2; ++begin1, ++begin2)
init = bop1(init, bop2(proj1(*begin1), proj2(*begin2)));
return init;
}
T operator()(I1 begin1, S1 end1, I2 begin2, T init, BOp1 bop1_ = BOp1{},
BOp2 bop2_ = BOp2{}, P1 proj1_ = P1{}, P2 proj2_ = P2{}) const
{
auto &&bop1 = invokable(bop1_);
auto &&bop2 = invokable(bop2_);
auto &&proj1 = invokable(proj1_);
auto &&proj2 = invokable(proj2_);
for (; begin1 != end1; ++begin1, ++begin2)
init = bop1(init, bop2(proj1(*begin1), proj2(*begin2)));
return init;
}
duration_type benchmark_simd(F f, V& op0, V& op1, V& op2, V& res, std::size_t number)
{
std::size_t s = op0.size();
duration_type t_res = duration_type::max();
for (std::size_t count = 0; count < number; ++count)
{
auto start = std::chrono::steady_clock::now();
for (std::size_t i = 0; i <= (s - B::size); i += B::size)
{
B bop0(&op0[i], aligned_mode()),
bop1(&op1[i], aligned_mode()),
bop2(&op2[i], aligned_mode());
B bres = f(bop0, bop1, bop2);
bres.store_aligned(&res[i]);
}
auto end = std::chrono::steady_clock::now();
auto tmp = end - start;
t_res = tmp < t_res ? tmp : t_res;
}
return t_res;
}