/*
#pragma omp parallel private(local_sum)
    {

  }

In front of the loop (inside the parallel region) there is a 'work-sharing" 
for directive (Chandra et al., p.113) that cares about the splitting of the 
work.  This unburdens the programer from setting the index range of the 
for-loop for each thread.

I attach the code.

*/
#include <stdio.h>
#include <sys/times.h>
#include <unistd.h>
#include <omp.h>

#define N 100000
#define CT 100

int main()
{
  int i, p, ict, ticks;
  struct tms t;
  clock_t elapsed_ticks;
  double elapsed_time;
  double local_sum, sum, x[N], y[N];

  ticks = sysconf(CLK_TCK);

#pragma omp parallel for
  for (i=0; i< N; i++){
    x[i] = i;
    y[i] = 1.0;
  };

#pragma omp parallel
  {
    p = omp_get_num_threads();
  }

  /* --------------- CASE 1 --------------- */

  elapsed_ticks = times(&t);

  for (ict=0; ict<CT; ict++){

    sum = 0.0;
#pragma omp parallel for
    for (i=0; i< N; i++){
      sum = sum + x[i]*y[i];
    }

  }

  elapsed_ticks = times(&t) - elapsed_ticks;
  elapsed_time = 1e6*(double)elapsed_ticks/(double)ticks/(double)CT;

  printf("Case 1: parallel for shared(sum)\n");
  printf("number of procs: %d\n",p);
  printf("elapsed ticks:   %d\n",elapsed_ticks);
  printf("elapsed time:  %10.4f musec\n",elapsed_time);
  printf("sum =  %16.2f\n\n",sum);

  /* --------------- CASE 2 --------------- */

#if 0
  elapsed_ticks = times(&t);

  for (ict=0; ict<CT; ict++){

    sum = 0.0;
#pragma omp parallel for
    for (i=0; i< N; i++){
#pragma omp critical
      { sum = sum + x[i]*y[i]; }
    }

  }

  elapsed_ticks = times(&t) - elapsed_ticks;
  elapsed_time = 1e6*(double)elapsed_ticks/(double)ticks/(double)CT;

  printf("Case 2: critical section for summing sum\n");
  printf("number of procs: %d\n",p);
  printf("elapsed ticks:   %d\n",elapsed_ticks);
  printf("elapsed time:  %10.4f musec\n",elapsed_time);
  printf("sum =  %16.2f\n\n",sum);
#endif

  /* --------------- CASE 3 --------------- */

  elapsed_ticks = times(&t);
  for (ict=0; ict<CT; ict++){

    sum = 0.0;
#pragma omp parallel private(local_sum)
    {
      local_sum = 0.0;
#pragma omp for
      for (i=0; i< N; i++){
	local_sum = local_sum + x[i]*y[i];
      }
#pragma omp critical
      { sum = sum + local_sum;}
    }
 
  }

  elapsed_ticks = times(&t) - elapsed_ticks;
  elapsed_time = 1e6*(double)elapsed_ticks/(double)ticks/(double)CT;

  printf("Case 3: parallel private(local_sum)\n");
  printf("number of procs: %d\n",p);
  printf("elapsed ticks:   %d\n",elapsed_ticks);
  printf("elapsed time:  %10.4f musec\n",elapsed_time);
  printf("sum =  %16.2f\n\n",sum);

  /* --------------- CASE 4 --------------- */

  elapsed_ticks = times(&t);
  for (ict=0; ict<CT; ict++){

    sum = 0.0;
#pragma omp parallel for reduction(+ : sum)
    for (i=0; i< N; i++){
      sum = sum + x[i]*y[i];
    }

  }

  elapsed_ticks = times(&t) - elapsed_ticks;
  elapsed_time = 1e6*(double)elapsed_ticks/(double)ticks/(double)CT;

  printf("Case 4: parallel for reduction(+ : sum)\n");
  printf("number of procs: %d\n",p);
  printf("elapsed ticks:   %d\n",elapsed_ticks);
  printf("elapsed time:  %10.4f musec\n",elapsed_time);
  printf("sum =  %16.2f\n\n",sum);

}
/*

#!/usr/bin/tcsh
date
foreach i ($argv[*])
    setenv OMP_NUM_THREADS $i
    ./mydots
echo --------------------------------------------
end

*/