/* -*- Mode: c -*-
 *
 *  Copyright 1993 Massachusetts Institute of Technology
 * 
 *  Permission to use, copy, modify, distribute, and sell this software and its
 *  documentation for any purpose is hereby granted without fee, provided that
 *  the above copyright notice appear in all copies and that both that
 *  copyright notice and this permission notice appear in supporting
 *  documentation, and that the name of M.I.T. not be used in advertising or
 *  publicity pertaining to distribution of the software without specific,
 *  written prior permission.  M.I.T. makes no representations about the
 *  suitability of this software for any purpose.  It is provided "as is"
 *  without express or implied warranty.
 * 
 */

/* Written by Jeremy Lin, jhlin@lcs.mit.edu */

#include <stdlib.h>
#include <stdio.h>
#include "measure.h"

#define mmxtest(data1, data2, result)  __asm__("movq %1,%%mm0\n\tpaddb %2, %%mm0\n\tmovq %%mm0,%0": "=m" (result) : "m" (data1), "m" (data2))

#define NUMQWORDS 100000
#define NUMBYTES NUMQWORDS * 4
#define TESTSIZE (NUMQWORDS * 8) + 8

void main(void)
{
  int i, j;
  unsigned char *scratch1, *scratch2, *scratch3, *data1, *data2, *result;
  MSR start, end;

/* Make sure you have quad-word aligned memory */
  scratch1 = (unsigned char *)malloc(16);
  scratch2 = (unsigned char *)malloc(16);
  scratch3 = (unsigned char *)malloc(16);

  data1=(unsigned char *)((unsigned long)(scratch1+8) & (unsigned long)(~7));
  data2=(unsigned char *)((unsigned long)(scratch2+8) & (unsigned long)(~7));
  result=(unsigned char *)((unsigned long)(scratch3+8) & (unsigned long)(~7));


/* Create Test Data */

  for(i = 0; i < NUMBYTES; i++) {
    data1[i] = i;
    data2[i] = i * 5;
  }

/* Add the 8-bit quantities using MMX */

  getTSC(start);
  for(j = 0; j < 1000; j++)
    for(i = 0; i < NUMBYTES; i += 8) {
      mmxtest(data1[i], data2[i], result[i]);
    }
  getTSC(end);

/* Print data and Number of cycles used */

  printf("MMX: %d\n", MSRdiff(end, start, 8));
  for(i = 390000; i < 390012; i++)
//    printf("%x %x %x\n", data1[i], data2[i], result[i]);
  __asm__("emms" : : );

/* Add the 8-bit quantities 'normally" */

  getTSC(start);
  for(j = 0; j < 1000; j++) 
    for(i = 0; i < NUMBYTES; i++) 
      result[i] = data1[i] + data2[i];
  getTSC(end);

/* Print data and Number of cycles used */

  printf("Normal: %d\n", MSRdiff(end, start, 8));
  for(i = 390000; i < 390012; i++)
//    printf("%x %x %x\n", data1[i], data2[i], result[i]);
  free(scratch1);
  free(scratch2);
  free(scratch3);
}


.