/* program app */
#include <stdio.h>
#include "Tmk.h"

struct shared {
	int sum;
	int turn;
	int* array;
}* shared;

main(int argc, char **argv)
{
	int start, end, i, p;
	int arrayDim = 100;

	Tmk_startup(argc, argv);

	/* Allocate shared memory */
	if (Tmk_proc_id == 0) {
		shared = (struct shared *) Tmk_alloc(sizeof(shared));
		if (shared == NULL)
			Tmk_exit(-1);
		/* share common pointer with all procs */
		Tmk_distribute(&shared, sizeof(shared));

		shared->array = (int *) Tmk_malloc(arrayDim * sizeof(int));
		if (shared->array == NULL)
			Tmk_exit(-1);
		shared->turn = 0;
		shared->sum = 0;
	}

	Tmk_barrier(0);

	/* Determine array range for each processor */
	{
		int id0 = Tmk_proc_id, id1 = Tmk_proc_id + 1;
		int perProc = arrayDim / Tmk_nprocs;
		int leftOver = arrayDim % Tmk_nprocs;
		start = id0 * perProc + id0 * leftOver / Tmk_nprocs;
		end = id1 * perProc + id1 * leftOver / Tmk_nprocs;
	}

	for (i = start; i < end; i++)
		shared->array[i] = i;

	Tmk_barrier(0);

	/* Print array elements, in the natural output order */
	for (p = 0; p < Tmk_nprocs; p++) {
		if (shared->turn == tmk-proc_id) {
			for (i = start; i < end; i++)
				printf("%d: %d\n", i, shared->array[i]);
			shared->turn++;
		}
		Tmk_barrier(0);
	}

	/* Compute local sum, then add to global sum */
	{
		int mySum = 0;
		for (i = start; i < end; i++) 
			mySum += shared->array[i];

		Tmk_lock_acquire(0);
		shared->sum += mySum;
		Tmk_lock_release(0);
	}

	if (Tmk_proc_id == 0) {
		Tmk_free(shared->array);
		Tmk_free(shared);
		printf("Sum is %d\n", shared->sum);
	}

	Tmk_exit(0);
}