#include <pthread.h>
#include <assert.h>
#include <string.h>
#include <stdlib.h>
#include <stdio.h>

#ifdef FEWER_KERNELS
#define THREADS 8
#define BLOCKSIZE (8 * 1024 * 1024)
#define REPS 32
#else
#define THREADS 8
#define BLOCKSIZE (8 * 256)
#define REPS (32 * 1024 * 4)
#endif

/* A worker thread.  Launch a pile of offload kernels and do some work.  */

void *
target_incr (void *arg)
{
  int *myarr = (int *) arg;

  for (int r = 0; r < REPS; r++)
    {
      #pragma omp target map(tofrom: myarr[0:BLOCKSIZE])
      {
	#pragma omp for
	for (int i = 0; i < BLOCKSIZE; i++)
	  {
	    myarr[i]++; 
	  }
      }
    }

  return NULL;
}

int
main (int argc, char* argv[])
{
  int *arr[THREADS];
  pthread_t workerthread[THREADS];
  int i;

  for (i = 0; i < THREADS; i++)
    {
      arr[i] = malloc (BLOCKSIZE * sizeof (int));
      memset (arr[i], 0, BLOCKSIZE * sizeof (int));
    }

  for (i = 0; i < THREADS; i++)
    {
      int *tmp = arr[i];
      #pragma omp target enter data map(to: tmp[0:BLOCKSIZE])
    }

  for (i = 0; i < THREADS; i++)
    pthread_create (&workerthread[i], NULL, target_incr, (void *) arr[i]);

  for (i = 0; i < THREADS; i++)
    {
      void *rv;
      pthread_join (workerthread[i], &rv);
      assert (rv == NULL);
    }

  for (i = 0; i < THREADS; i++)
    {
      int *tmp = arr[i];
      #pragma omp target exit data map(from: tmp[0:BLOCKSIZE])
    }

  for (i = 0; i < THREADS; i++)
    for (int j = 0; j < BLOCKSIZE; j++)
      assert (arr[i][j] == REPS);

  for (i = 0; i < THREADS; i++)
    free (arr[i]);

  return 0;
}