/* param 1 = taille des matrices au depart
   param 2 = incrementation
   param 3 = taille des matrices ŕ l'arrive
   param 4 = nombre d'itérations 

mpicc -O3 -o cmult cmult.c
(using MPICH)

mpirun -arch LINUX -np 1 ./cmult 400 2 402 1

*/

#include <mpi.h>
#include <stdlib.h> 
#include <stdio.h> 
#include <math.h>
#include <string.h>

typedef struct Complexe {double re; double img;} complexe;

complexe zero;

/* __inline__ complexe complexe_add(complexe a, complexe b)
{
 return (complexe){a.re+b.re, a.img+b.img};
}


__inline__ complexe complexe_mult(complexe x, complexe y)
{
 return (complexe) {x.re*y.re-x.img*y.img, x.re*y.img+x.img*y.re};
}

*/


/* version generic */

__inline__ void* add(void *x, void *y)
{
complexe a=((complexe*) x)[0];
complexe b=((complexe*) y)[0];
return (void *) &((complexe){a.re+b.re, a.img+b.img});
}

__inline__ void* mult(void *x, void *y)
{
complexe a=((complexe*) x)[0];
complexe b=((complexe*) y)[0];
return (void *) &((complexe){a.re*b.re-a.img*b.img, a.re*b.img+a.img*b.re});
}

complexe random_complexe(void)
{
 complexe c = {c.re=(double)(rand()%1000), c.img=(double)(rand()%1000)};
 return c;
}


void multiply_generic(int n, void *zero, void* (*add)(void *a, void *b), void* (*mult)(void *a, void *b), void *a, void *b, void *c)
{
 register int i,j,k;
 void *tmp;
 void *cc;
 
 void *tt1, *tt2, *tt3;

  for (i=0; i<n; i++) 
   {
    for (j=0;j<n;j++)
     {
      tmp=zero;
      for (k=0;k<n;k++)
       {
        tmp=add(tmp,(mult(b+(k*n+j),a+(i*n+k))));
        /* tt1=malloc(sizeof(complexe));
        memcpy(tt1,mult(b+(k*n+j),a+(i*n+k)),sizeof(complexe));
	tt2=malloc(sizeof(complexe));
	memcpy(tt2,add(tmp,tt1),sizeof(complexe));
	free(tt1);
	memcpy(tmp,tt2,sizeof(complexe));
	free(tt2);	 */
       }
      cc=c+(i*n+j);
      cc=tmp;
      /* memcpy(cc,tmp,sizeof(complexe)); */
     }
  }
}

/*

void multiply_complex(int n, complexe *a, complexe *b, complexe *c)
{
 register int i,j,k,t1,t2;
 complexe tmp;

  for (i=0; i<n; i++) 
   {
    for (j=0;j<n;j++)
     {
      tmp=zero;
      for (k=0;k<n;k++) tmp=complexe_add(tmp,(complexe_mult(b[k*n+j],a[i*n+k])));
      c[i*n+j]=tmp;
     }
  } 
}

*/

int main(int argc, char* argv[])
{
 complexe *a, *b, *c;
 int i,j,k,t1,pas,maxi,iter,pid;
 double chrono=0;

 zero.re=(double)0;
 zero.img=(double)0;

 MPI_Init(&argc,&argv);
 MPI_Comm_rank(MPI_COMM_WORLD,&pid);

 i=atoi(argv[1]);
 pas=atoi(argv[2]);
 maxi=atoi(argv[3]);
 iter=atoi(argv[4]);

 srand(1000);
 while (i<maxi)
  {  
   /* initialisation des matrices */
   a=(complexe *)malloc(i*i*sizeof(complexe));
   b=(complexe *)malloc(i*i*sizeof(complexe));
   c=(complexe *)malloc(i*i*sizeof(complexe));
      
   for (j=0;j<i;j++)
    for (k=0,t1=j*i;k<i;k++)
     {
      a[t1+k]=random_complexe();
      b[t1+k]=random_complexe();
     }

  chrono=MPI_Wtime();  

  for (k=1;k<=iter;k++) 
   /* multiply_complex(i,a,b,c); */
   multiply_generic(i,&zero,add,mult,a,b,c);

  chrono=MPI_Wtime()-chrono;
  MPI_Barrier(MPI_COMM_WORLD);
  
  if (pid==0) printf("%f\n",chrono/iter);
  fflush(stdout);

  /* suppression des vieilles matrices */
  free((complexe *)a);
  free((complexe *)b);
  free((complexe *)c);

  i=i+pas;
  }

 MPI_Finalize();
 return 0;
}