// matmul_thread1.java  for compare with  matmul_thread4
import java.lang.management.ManagementFactory;
import java.lang.management.ThreadMXBean;

public class matmul_thread1
{
  public matmul_thread1()
  {
    int nthread = 1;
    double cpu_start[] = new double[nthread]; // for each thread           
    double cpu_end[]   = new double[nthread];
    int i, j, k;
    int DIM = 2048;
    double a[][] = new double[DIM][DIM];
    double b[][] = new double[DIM][DIM];
    double c[][] = new double[DIM][DIM];
    double sum;
    double twall = 0.0, tcpu = 0.0;
    ThreadMXBean tb = ManagementFactory.getThreadMXBean();

    System.out.println("matmul_thread1.java running");
    System.out.println("DIM="+DIM+", nthread="+nthread);

    for(i=0; i<DIM; i++)
    {
      for(j=0; j<DIM; j++)
      {
        a[i][j] = (double)(i+j);
        b[i][j] = (double)(i+j+1);
      }
    }

    twall = System.currentTimeMillis()/1000.0;
    cpu_start[0] = tb.getCurrentThreadCpuTime()/1.0E9;

    for(i=0; i<DIM; i++)
    {
      for(j=0; j<DIM; j++)
      {
        sum=0.0;
        for(k=0; k<DIM; k++)
        {
          sum += a[i][k]*b[k][j];
        } // end k
        c[i][j]=sum;
      } // end j
    } // end i

    twall = System.currentTimeMillis()/1000.0 - twall;
    cpu_end[0] = tb.getCurrentThreadCpuTime()/1.0E9;
    tcpu = cpu_end[0] - cpu_start[0];
    System.out.println("c[1][1]="+c[1][1]+
		       ", c[2][2]="+c[2][2]);
    System.out.println("c[513][513]="+c[513][513]+
		       ", c[514][514]="+c[514][514]);
    System.out.println("c[1025][1025]="+c[1025][1025]+
		       ", c[1026][1026]="+c[1026][1026]);
    System.out.println("c[1538][1538]="+c[1538][1538]+
		       ", c[1539][1539]="+c[1539][1539]);
    System.out.println("took "+twall+" wall sec, "+tcpu+" cpu sec");
 
  } // end matmul_thread1

  public static void main (String[] args)
  {
    new matmul_thread1();
  }

} // end matmul_thread1.java