Function mem_bench

src/test.cpp:218–249

}

void mem_bench() {
  constexpr size_t N_THREADS = 32;
  constexpr size_t MB_PER_THREAD = 1024;
  constexpr size_t ELS_PER_THREAD = (MB_PER_THREAD * 1024 * 1024) / sizeof(float);
  constexpr size_t N = N_THREADS * ELS_PER_THREAD;

  std::cout << "Using " << N_THREADS << " threads" << std::endl;
  std::cout << "Allocating " << N_THREADS * MB_PER_THREAD << " MB (" << N << " floats)" << std::endl;
  float* data = allocateAlignedArray(N);

  std::cout << "Filling data..." << std::endl;
#pragma omp parallel for num_threads(N_THREADS)
  for (size_t i = 0; i < N_THREADS; i++) {
    fill_random(data + i * ELS_PER_THREAD, ELS_PER_THREAD, (unsigned long)i);
  }
  std::cout << "Running memory bandwidth test..." << std::endl;

  float totalSum = 0.0;
  uint64_t start = get_timestamp_ms();
#pragma omp parallel for simd reduction(+:totalSum) schedule(guided) aligned(data: 64) num_threads(N_THREADS)
  for (size_t i = 0; i < N; i++) {
    totalSum += data[i];
  }
    
  uint64_t end = get_timestamp_ms();
  float elapsed_s = (end - start) / 1000.0;
  float mb_per_s = N_THREADS * MB_PER_THREAD / elapsed_s;

  std::cout << "Total sum: " << totalSum << std::endl;
  std::cout << "Elapsed time: " << elapsed_s << " s" << std::endl;
  std::cout << "Memory bandwidth: " << mb_per_s << " MB/s" << std::endl;
}

// 64 is the typical cache line size
struct alignas(64) ThreadData {