Minimal Android OpenMP Benchmark

android_openmp_bench.cpp

// Results one OnePlus two (snapdragon 810 - 8 cores ARMv8):
// with optimization -fopenmp -Ofast -mcpu=cortex-a57.cortex-a53
// Serial time: 19.480000 ms
// OpenMP time: 10.000000 ms

{
  const int S = 10e6;
  auto A = new int[S];
  auto B = new int[S];

  timer timer;

  const int K = 100;

  timer.start();
  for (int k = 0; k < K; k++)
  for (int i = 0; i < S; i++)
    A[i] = A[i] + B[i];
  timer.end();
  __android_log_print(ANDROID_LOG_INFO, "bench", "Serial time %f ms", float(timer.ms()) / K);
  
  timer.start();
   omp_set_num_threads(4);
  for (int k = 0; k < K; k++)
#pragma omp parallel for simd
    for (int i = 0; i < S; i++)
      A[i] = A[i] + B[i];
  timer.end();
  __android_log_print(ANDROID_LOG_INFO, "bench", "OpenMP time %f ms", float(timer.ms()) / K);
}

4 threads/cores for 2X speedup only?
Doesn't that imply twice overhead introduced by OpenMP?

I think that today this benchmark is a outdated. Compilers have evolved and processors too.