#include <iostream>
#include <chrono>

void foo(int* a, int* b) {
  *a+=1;
  *b+=1;
  *a+=1;
}

void goo(int* __restrict a, int* b) {
  *a+=1;
  *b+=1;
  *a+=1;
}

void measure() {
  int x = 1;
  int y = 2;

  auto start_foo = std::chrono::high_resolution_clock::now();  
 
  for(int i = 0; i < 10000000; ++i) {
    foo(&x, &y);
  }
  
  auto end_foo = std::chrono::high_resolution_clock::now();
  
  
  std::chrono::duration<double> duration_foo = end_foo - start_foo;
  std::cout << "foo Runtime(secs): " << duration_foo.count() << std::endl;

  auto start_goo = std::chrono::high_resolution_clock::now();

  for(int i = 0; i < 10000000; ++i) {
    goo(&x, &y);
  }

  auto end_goo = std::chrono::high_resolution_clock::now(); 
  std::chrono::duration<double> duration_goo = end_goo - start_goo;

  std::cout << "goo Runtime(secs): " << duration_goo.count() << std::endl;
  std::cout << "Without/With: " << duration_foo.count() / duration_goo.count() << std::endl << std::endl;
}

int main() {
  for (int i = 0; i < 10; ++i) {
    measure();
  }
}

I see that goo is optimized as expected via __restirct, but the output confused me a lot since there are cases when goo performs much more worse than foo. In 3/10 cases they perform similiarly, in 3/10 cases foo performs better, and only in 4/10 cases we have expected result.

Please, explain why?