#include <omp.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>

#define N 10000000                  /* number of vector elements */

int main (int argc, char *argv[]) 
{
  int	tid, nthreads, i;
  double scalar = 4.0;
  double *a, *b, *c;
  
  double t0, t1, dt;

//Serial Initialization
  t0 = omp_get_wtime();
  //allocation of C arrays
  a=(double*) malloc(N*sizeof(double));
  b=(double*) malloc(N*sizeof(double));
  c=(double*) malloc(N*sizeof(double));
  //serial initialization:
  //OS will allocate all data close to initial thread
  for (i=0;i<N;i++) a[i]=b[i]=c[i]=0.0;
  //saxpying with poor memory placement
  #pragma omp parallel for
  for(i=0;i<N;i++) a[i]=b[i]+scalar*c[i];
  //timing  
  t1 = omp_get_wtime();
  dt = t1 - t0;
  printf("SAXPY with   Serial Initialization: Elapsed time = %lf\n",dt);

//Parallel Initialization
  t0 = omp_get_wtime();
  //allocation of C arrays
  //double *a, *b, *c;
  a=(double*) malloc(N*sizeof(double));
  b=(double*) malloc(N*sizeof(double));
  c=(double*) malloc(N*sizeof(double));
  //parallel initialization: data allocated where used
  # pragma omp parallel for
  for (i=0;i<N;i++) a[i]=b[i]=c[i]=0.0;
  //saxpying with optimal memory placement
  #pragma omp parallel for
  for(i=0;i<N;i++) a[i]=b[i]+scalar*c[i];
  //timing  
  t1 = omp_get_wtime();
  dt = t1 - t0;
  printf("SAXPY with Parallel Initialization: Elapsed time = %lf\n",dt);
}