#error Please comment out the next two lines under linux, then comment this error
#include "stdafx.h" //Visual studio expects this line to be the first one, comment out if different compiler
#include <windows.h> // Include if under windows

#ifndef WIN32
#include <sys/time.h>
#endif
#include <stdlib.h>
#include <stdio.h>
#include "rdtsc.h"
#include <time.h>

#define NUM_RUNS 1000

int m, n, k;
double *A, *B, *C;

void compute() {
  int i,j,h;
  for(i = 0; i < m; ++i) {
    for(j = 0; j < n; ++j) {
      double out = 0;
      for(h = 0; h < k; ++h) {
        out = out + A[i*k+h] * B[h*n+j];
      }
      C[i*n+j] += out;
    }
  }
}

double rdtsc() {
  int i;
  tsc_counter a, b;

/*   CPUID instruction serializes the pipeline */
/*   Therefore, splits the code to be timed from the code running before */

  CPUID(); RDTSC(a); CPUID(); RDTSC(b);
  CPUID(); RDTSC(a); CPUID(); RDTSC(b);

  RDTSC(a);
  for(i=0; i<NUM_RUNS; ++i)
    { compute(); }
  RDTSC(b);

  double cycles = ((double)COUNTER_DIFF(b, a)) / ((double) NUM_RUNS);
  return cycles;
}

double c_clock() {
  int i;
  clock_t a,b;

  a=clock();
  for(i=0; i<NUM_RUNS; ++i)
    { compute(); }
  b=clock();

  return (double)(b-a)/((double) NUM_RUNS);
}

#ifndef WIN32
double timeofday() {
  int i;
  struct timeval a, b;

  gettimeofday(&a, NULL);
  for(i=0; i<NUM_RUNS; ++i)
    { compute(); }
  gettimeofday(&b, NULL);
  return(double)((b.tv_sec - a.tv_sec) + (b.tv_usec - a.tv_usec)/1000000.0)/(double) NUM_RUNS;
}
#else
double gettickcount() {
  int i;
  double a, b;

  a = (double)GetTickCount();
  for(i=0; i<NUM_RUNS; ++i)
    { compute(); }
  b = (double)GetTickCount();
  return(double)((b-a)/(double) NUM_RUNS);
	return 0;
}

double queryperfcounter() {
  int i;
  LARGE_INTEGER a,b;

  QueryPerformanceCounter(&a);
  for(i=0; i<NUM_RUNS; ++i)
    { compute(); }
  QueryPerformanceCounter(&b);

  return(double)((double)(b.QuadPart-a.QuadPart)/(double) NUM_RUNS);
}
#endif

int main(int argc, char **argv){
  const double FREQUENCY = 2666000000.0;
  int i;
  
  if (argc!=4) {printf("usage: mmm <m> <k> <n>\n"); return -1;}
  m = atoi(argv[1]);
  k = atoi(argv[2]);
  n = atoi(argv[3]);
  printf("m=%d k=%d n=%d\n",m,k,n);

  A = (double*)malloc(m*k*sizeof(double));
  B = (double*)malloc(k*n*sizeof(double));
  C = (double*)malloc(m*n*sizeof(double));

  for(i =0; i<m*k;i++) A[i]=i;
  for(i =0; i<k*n;i++) B[i]=i;

  double r = rdtsc();
  printf("RDTSC instruction :\n %lf clock ticks measured => %lf seconds, assuming frequency is %lf MHz. (change in file if different)\n\n", r, r/(FREQUENCY), (FREQUENCY)/1e6);

  double c = c_clock();
  printf("C clock() function :\n %lf clock ticks measured. On some systems, this number seems to be actually computed from a timer in seconds then transformed into clock ticks using the variable CLOCKS_PER_SEC. Unfortunately, it appears that CLOCKS_PER_SEC is sometimes improperly. (According to this variable, your computer should be running at %lf MHz). In any case, dividing by this value should give a correct timing: %lf seconds. \n\n",c, (double) CLOCKS_PER_SEC/1e6, c/ CLOCKS_PER_SEC);

#ifndef WIN32
  double t = timeofday();
  printf("C gettimeofday() function:\n %lf seconds measured\n\n",t);
#else
  double t = gettickcount();
  printf("Windows getTickCount() function:\n %lf milliseconds measured\n\n",t);

  __int64 f;
  QueryPerformanceFrequency((LARGE_INTEGER *)&f);

  double p = queryperfcounter();
  printf("Windows QueryPerformanceCounter() function:\n %lf clock ticks measured => %lf seconds, with reported CPU frequency %lf MHz\n\n",p,p/f,(double)f);
#endif

  printf("Type anything to leave\n");
  scanf("%d",&i);
  return 0;
}