C++ High Performance

Performance evaluation

Now we create another, rather stupid, transform_func that takes more time depending on the input value, and a range of increasing values using std::iota() like this:

const auto transform_func = [](float v) {  auto sum = v;  auto i_max = v / 100'000;  // The larger "v" is, the more to compute  for (size_t i = 0; i < i_max; ++i) { sum += (i*i*i*sum); }  return sum;};auto n = size_t{ 10'000'000 };auto src = std::vector<float>(n);std::iota(src.begin(), src.end(), 0.0f); // "src" goes from 0 to n

If we evaluate them with different chunk sizes, as well as std::transform() and the old par_transform_naive(), we get the following computation times:

Function	Chunk size	Number of tasks	Microseconds	Speed up
std::transform() ...