C++ High Performance

Performance evaluation

Now that we have two different implementations, let's see how they measure up. The performance boost from using this parallelized version of copy_if is heavily dependent on how expensive the predicate is. Therefore, we measure with two predicates; is_odd, which is very inexpensive and is_prime, which is expensive.

The light is_odd predicate:	The heavy is_prime predicate:
auto is_odd = [](unsigned v) { return (v % 2) == 1; };	auto is_prime = [](unsigned v) { if (v < 2) return false; if (v == 2) return true; if (v % 2 == 0) return false; for (unsigned i=3; (i*i)<=v; i+=2) { if ((v % i) == 0) {return false; } } return true;};

The following table shows the performance as measured using an Intel i7 7700k ...