//////////////////////////////////////////////////////////////////////////////////
// File : main.c  (for gameoflife)
// Date : November 2013
// Author :  Alexandre Joannou <alexandre.joannou@lip6.fr>
//
// This application is an emulation of the game of life automaton.
// The world size is defined by the HEIGHT and WIDTH parameters.
// There is one task per processor, and each task compute HEIGHT/nbprocs lines.
// The number of processors must be a power of 2 not larger than HEIGHT.
//////////////////////////////////////////////////////////////////////////////////

#include "stdio.h"
#include "limits.h"
#include "user_barrier.h"
#include "mapping_info.h"

#define WIDTH           128
#define HEIGHT          128
#define NB_ITERATION    1000000000

#define PRINTF(...) ({ if ( proc_id==0) { giet_shr_printf(__VA_ARGS__); } })

giet_sqt_barrier_t barrier;

volatile unsigned int init_ok;

#define OLD 0
#define NEW 1
#define DSP 2

typedef unsigned char uint8_t;
typedef unsigned int size_t;

uint8_t world[3][HEIGHT][WIDTH];

/////////////////////////////////////////////////
void init_world(size_t base_line, size_t nb_line)
{
   size_t x,y;
   for (y = base_line ; y < base_line + nb_line; y++)
   {
      for(x = 0; x < WIDTH ; x++) 
      {
         world[OLD][y][x] = (giet_rand() >> (x % 8)) & 0x1;
      }
   }
}

/////////////////////////////////////////////////
uint8_t number_of_alive_neigh(size_t x, size_t y)
{
   uint8_t nb = 0;

   nb += world[OLD][(y - 1) % HEIGHT][(x - 1) % WIDTH];
   nb += world[OLD][ y              ][(x - 1) % WIDTH];
   nb += world[OLD][(y + 1) % HEIGHT][(x - 1) % WIDTH];
   nb += world[OLD][(y - 1) % HEIGHT][ x             ];
   nb += world[OLD][(y + 1) % HEIGHT][ x             ];
   nb += world[OLD][(y - 1) % HEIGHT][(x + 1) % WIDTH];
   nb += world[OLD][ y              ][(x + 1) % WIDTH];
   nb += world[OLD][(y + 1) % HEIGHT][(x + 1) % WIDTH];

   return nb;
}

/////////////////////////////////////////////////
uint8_t compute_cell(size_t x, size_t y)
{
   uint8_t nb_neighbours_alive = number_of_alive_neigh(x,y);
   if (world[OLD][y][x] == 1) 
   {
      if (nb_neighbours_alive == 2 || nb_neighbours_alive == 3)  return 1;
   }
   else 
   {
      if (nb_neighbours_alive == 3) return 1;
      else                          return world[OLD][y][x];
   }
   return 0;
}

//////////////////////////////////////////////////////
void compute_new_gen(size_t base_line, size_t nb_line)
{
   size_t x,y;
   for (y = base_line; y < base_line + nb_line; y++)
   {
      for(x = 0; x < WIDTH ; x++) 
      {
         world[NEW][y][x] = compute_cell(x,y);  
      }
   }
}

////////////////////////////////////////////////////
void display_world(size_t base_line, size_t nb_line)
{
   size_t x,y;
   for (y = base_line; y < base_line + nb_line; y++)
   {
      for(x = 0; x < WIDTH ; x++) 
      {
         world[DSP][y][x] = world[OLD][y][x]*255;  
      }
   }

   giet_fbf_sync_write( base_line * WIDTH , 
                        &world[DSP][base_line][0], 
                        nb_line * WIDTH );
}

/////////////////////////////////////////////////////
void grow_old_world(size_t base_line, size_t nb_line)
{
   size_t x,y;
   for (y = base_line; y < base_line + nb_line; y++)
   {
      for(x = 0; x < WIDTH ; x++) 
      {
         world[OLD][y][x] = world[NEW][y][x];  
      }
   }
}

////////////////////////////////////////
__attribute__((constructor)) void main()
{
   // get processor identifier
   unsigned int x;
   unsigned int y;
   unsigned int p;
   giet_proc_xyp( &x, &y, &p );

   // get processors number
   unsigned int x_size;
   unsigned int y_size;
   unsigned int n_local_procs;
   giet_procs_number( &x_size, &y_size, &n_local_procs );

   // compute continuous processor index
   unsigned int proc_id = (((x * y_size) + y) * n_local_procs) + p;  

   unsigned int n_clusters     = x_size * y_size;            // number of clusters
   unsigned int n_global_procs = n_clusters * n_local_procs; // number of processors
   size_t i;

   if ( n_global_procs > HEIGHT )
   {
       PRINTF("[GAMEOFLIFE ERROR] Number or processors too large :"
              " nb_procs = %d / image heigth = %d\n", n_global_procs, HEIGHT );
       giet_exit("error");
   }

   size_t       nb_line       = HEIGHT / n_global_procs;
   size_t       base_line     = nb_line * proc_id; 
   
   PRINTF("\n*** Starting barrier initialisation at cycle %d ***\n"
          " nprocs = %d / nlines = %d\n", 
          giet_proctime() , n_global_procs, HEIGHT );

   // barrier initialization
   if ( proc_id == 0 )
   {
      sqt_barrier_init( &barrier , x_size , y_size , n_local_procs );
      init_ok = 1;
   }
   else
   {
      while ( init_ok == 0 ) asm volatile("nop\n nop\n nop");
   }

   PRINTF("\n*** Starting world initialisation at cycle %d ***\n",
          giet_proctime() );

   //  parallel world  initialization
   init_world( base_line , nb_line );
   display_world( base_line , nb_line );

   sqt_barrier_wait( &barrier );

   PRINTF("\n*** Starting life at cycle %d ***\n", 
          giet_proctime() );
   
   for (i = 0; i < NB_ITERATION; i++)
   {
      compute_new_gen( base_line, nb_line );
      grow_old_world( base_line, nb_line );
      display_world( base_line, nb_line );

      sqt_barrier_wait( &barrier );

      PRINTF(" - iteration %d completed\n", i );
   }

   PRINTF("\n*** End of main at cycle %d ***\n", giet_proctime());

   giet_exit("Completed");
} // end main()

// Local Variables:
// tab-width: 3
// c-basic-offset: 3
// c-file-offsets:((innamespace . 0)(inline-open . 0))
// indent-tabs-mode: nil
// End:

// vim: filetype=cpp:expandtab:shiftwidth=3:tabstop=3:softtabstop=3