Probing performance with micro-benchmarks

The outcome of the previous section may leave you somewhat daunted: the processor is very complex and, apparently, needs a lot of hand-holding on the part of the programmer to operate at peak efficiency. Let us start small and see how fast a processor can do some basic operations. To that end, we will use the same Google Benchmark tool we have used in the last chapter. Here is a benchmark for the simple addition of two arrays:

01_superscalar.C

#include "benchmark/benchmark.h"
void BM_add(benchmark::State& state) {
     srand(1);
     const unsigned int N = state.range(0);
     std::vector<unsigned long> v1(N), v2(N);
     for (size_t i = 0; i < N; ++i) {
           v1[i] = rand();
           v2[i] = rand(...