// Compilation:
// spu-gcc -I/opt/cell/sdk/usr/spu/include/ -L/opt/cell/sdk/usr/spu/lib -O3 -o lme large_matrix_example.c -llarge_matrix
#include <stdio.h>
#include <spu_timer.h>
#include <liblarge_matrix.h>
//#include <spu_intrinsics.h>

#define TIMEBASE 79800000 // from /proc/cpuinfo

#define N 128

float a[N][N] __attribute__((aligned(16)));
float b[N][N] __attribute__((aligned(16)));
float c[N][N] __attribute__((aligned(16)));

int main()
{ uint64_t clk0,clk1;
  int i,j;

  spu_clock_start();

  for (i=0;i<N;i++)
    for (j=0;j<N;j++)
    { a[i][j] = i+j;
      b[i][j] = i-j;
      c[i][j] = 0.0;
    }

  clk0 = spu_clock_read();

  madd_matrix_matrix(N,N,N,&a[0][0],N,&b[0][0],N,&c[0][0],N);

  clk1 = spu_clock_read();

  printf("Time (clocks ): %llu\n",clk1-clk0);
  printf("Time (seconds): %lf\n",(double)(clk1-clk0)/TIMEBASE);

  spu_clock_stop();
  return 0;
}