How to properly measure code?

Question

C++ / C#

I wrote my bike to measure the speed of arithmetic operations https://github.com/DevAlone/CppCodeMeasurer
I measure time using std::chrono (highresolutiontimer.hpp file). I run the test function like this:

template <typename F, typename... Args>
Timer::precision CodeMeasurer::measure(F func, Args... args)
{
// делаем N повторений и записываем результат
    for (size_t i = 0; i < NUM_OF_REPETITION; i++) {
        t1.reset();
        func(args...);
        result = t1.elapsed();
        results[i] = result;
    }

// считаем среднее исключая крайние замеры
    result = 0;
    for (size_t i = NUM_OF_SKIPPED; i < NUM_OF_REPETITION - NUM_OF_SKIPPED; i++) {
        result += results[i];
    }
    result /= NUM_OF_REPETITION - NUM_OF_SKIPPED * 2;

    return result;
}

Passing a function through a pointer shouldn't affect the results.
During the first tests, I noticed that there is a difference between the first and subsequent runs of the function, probably because of the cache, so I only take into account the results "in the middle".
The function under test itself performs operations many times in a row:

void exec()
    {
        for (size_t i = 0; i < NUM_OF_REPETITION; i++) {
            val1 += val2;
            val2 -= val2;
            val1 *= val2;
        }
    }

I compile it like this:
g++ -O0 -o runIt ../main.cpp
So that a too smart compiler doesn't cut out my function with its optimizations I
run it like this:
sudo nice -n -20 ./runIt
Have I taken everything into account for the purity of the experiment?