#include <iostream>
#include <unistd.h>
#include <sys/mman.h>
#include <string>
#include <omp.h>

#define GIG (1024ul*1024*1024)
#define VOLUME (1024ul*1024*1024)

// Real values is in the ~65536, only use half of it to be safe
// sysctl vm.max_map_count
#define MAXMMAPS 32768

int main(int argc, char **argv) {
  auto page_size = sysconf(_SC_PAGESIZE);
  std::cout << "The system have a page size of " << page_size << std::endl;
  // Src is using anonymous mapping
  long nmaps = 1024;
  if (argc > 1) {
    nmaps = std::stoi(argv[1]);
    if (VOLUME / page_size % nmaps != 0) {
      std::cout << "nmaps is not perfect multiple, quit" << std::endl;
      return 0;
    }
  }

  long mmap_sz = VOLUME / nmaps;
  std::cout << "Each mapped region is of size(pages) " << mmap_sz/page_size << std::endl;

  auto dst = (long*)malloc(VOLUME);

  uint8_t *hint = (uint8_t*)0x600000000000UL;
  hint -= VOLUME;
  auto src = (long*)hint;
  for (long i = 0; i < nmaps; ++i) {
    auto r = mmap(hint, mmap_sz, PROT_READ | PROT_WRITE, MAP_ANON | MAP_PRIVATE, -1, 0);
    if (r == MAP_FAILED || r != hint)
      printf("MMAP failed somehow\n");
    hint += mmap_sz;
  }

#pragma omp parallel for
  for (long i = 0; i < VOLUME/sizeof(long); ++i) {
    src[i] = i;
    dst[i] = 0;
  }

  int ITER = 100;
  double st = omp_get_wtime();

  for (int t = 0; t < ITER; ++t) {
#pragma omp parallel for
    for (long i = 0; i < VOLUME/sizeof(long); ++i)
      dst[i] = src[i];
  }

  st = (omp_get_wtime() - st) / ITER;

  printf("Average time(s) %f\n", st);
//  std::cout << "Average time(s) " << st << std::endl;
  printf("Average throughput(GB/s) %f\n", VOLUME * 2 / st * 1e-9);
//  std::cout << "Average throughput(GB/s) " << VOLUME * 2 / st * 1e-9 << std::endl;
  return 0;
}