/******************************************************************************
* FILE: loop11.c
* DESCRIPTION:
*   Demonstrates loop unrolling
******************************************************************************/

#include <stdio.h>
#include <sys/time.h>
#include <time.h>
#define	SEED		1995
#define	DIM1		400
#define	DIM2		400

float a[DIM1][DIM2], aa[DIM1][DIM2];
float b[DIM1][DIM2], bb[DIM1][DIM2];
float c[DIM1][DIM2], cc[DIM1][DIM2];
struct timeval start_time, end_time;

main()
{
   int i, j, k;
   initialize_data();

/*****************************************************************************/
/*				Untuned Loop			             */
/*****************************************************************************/
   gettimeofday(&start_time, (struct timeval*)0);

   for(i=0;i<DIM1;i++)
      for(j=0;j<DIM2;j++)
         for(k=0;k<4;k++)
            a[i][j] = a[i][j] + b[k][i] * c[k][j];

   gettimeofday(&end_time, (struct timeval*)0);  /* after time */
   printf("Execution time for the uncollapsed loop:\n");
   print_execution_times();

/*****************************************************************************/
/*				Tuned Loop				     */
/*****************************************************************************/
   gettimeofday(&start_time, (struct timeval*)0);

   for(i=0;i<DIM1;i++)
      for(j=0;j<DIM2;j++)
      {
         aa[i][j] = aa[i][j] + bb[0][i] * cc[0][j];
         aa[i][j] = aa[i][j] + bb[1][i] * cc[1][j];
         aa[i][j] = aa[i][j] + bb[2][i] * cc[2][j];
         aa[i][j] = aa[i][j] + bb[3][i] * cc[3][j];
      }

   gettimeofday(&end_time, (struct timeval*)0);  /* after time */
   printf("Execution time for the collapsed loop:\n");
   print_execution_times();
   finalize_data();
}

initialize_data()
{
   int i, j, k;
   srandom(SEED);
   for(i=0;i<DIM1;i++)
      for(j=0;j<DIM2;j++)
      {
            a[i][j] = aa[i][j] = (float) (random() % 100000);
            b[i][j] = bb[i][j] = (float) (random() % 100000);
            c[i][j] = cc[i][j] = (float) (random() % 100000);
      }
}

finalize_data()
{
   int i, j;
   FILE *fp, *fopen();

   fp = fopen("/dev/null", "w");
   i = random() % DIM1;
   j = random() % DIM2;
   fprintf(fp, "%f\n", a[i][j]);
   fprintf(fp, "%f\n", aa[i][j]);
   fclose(fp);
}

print_execution_times()
{
   int total_usecs;
   float total_time, total_flops;
   total_usecs = (end_time.tv_sec-start_time.tv_sec) * 1000000 +
         (end_time.tv_usec-start_time.tv_usec);
   printf("Total time was %.2f mSec and the ", ((float) total_usecs) / 1000.0);
   total_time = ((float) total_usecs) / 1000000.0;
   total_flops = (float) (DIM1 * DIM2 * 4 * 2);
   printf("total MFLOPS is %.2f\n\n", (total_flops / total_time) / 1000000.0);
}