Exemple #1
0
LIS_INT main(LIS_INT argc, char* argv[])
{
        LIS_SCALAR *a,*q,*r;
	LIS_INT m,n,nn;
	LIS_INT	i,j,ii,jj,nnz,qriter;
	double time,time0;
	LIS_REAL qrerr;	

 	LIS_DEBUG_FUNC_IN;

	lis_initialize(&argc, &argv);

	if( argc < 3 )
	{
	  printf("Usage: %s m n\n", argv[0]);
	  CHKERR(1);
	}

	m  = atoi(argv[1]);
	n  = atoi(argv[2]);
	if( m<=0 || n<=0 )
	{
	  printf("m=%d <=0 or n=%d <=0\n", m,n);
	  CHKERR(1);
	}
	
	printf("\n");

	/* create arrays */

	nn = m*n;

	a = (LIS_SCALAR *)malloc(nn*nn*sizeof(LIS_SCALAR));
	q = (LIS_SCALAR *)malloc(nn*nn*sizeof(LIS_SCALAR));
	r = (LIS_SCALAR *)malloc(nn*nn*sizeof(LIS_SCALAR));

	/* define two-dimensional Laplacian */

	lis_array_set_all(nn*nn,(LIS_SCALAR)0.0,a);

	nnz = 0;
	for(ii=0;ii<nn;ii++)
	  {
	    i = ii/m;
	    j = ii - i*m;
	    if( i>0 )   { jj = ii - m; a[ii + jj * nn] = -1.0; nnz++;} 
	    if( i<n-1 ) { jj = ii + m; a[ii + jj * nn] = -1.0; nnz++;}
	    if( j>0 )   { jj = ii - 1; a[ii + jj * nn] = -1.0; nnz++;}
	    if( j<m-1 ) { jj = ii + 1; a[ii + jj * nn] = -1.0; nnz++;}
	    jj = ii; a[ii + jj * nn] = 4.0; nnz++;
	  }

	printf("matrix size = %d x %d (%d nonzero entries)\n\n", nn,nn,nnz);

	/* solve eigenproblem */

	time0 = lis_wtime();
	lis_array_qr(nn,a,q,r,&qriter,&qrerr);
	time = lis_wtime() - time0;

	printf("QR    : number of iterations = %d\n", qriter);
	printf("QR    : elapsed time         = %e sec.\n", time);
	printf("QR    :   eigensolver        = %e sec.\n", time);
#ifdef _LONG__DOUBLE
	printf("QR    : 2-norm of A(2,1)     = %Le\n\n", qrerr);
#else
	printf("QR    : 2-norm of A(2,1)     = %e\n\n", qrerr);
#endif

	lis_finalize();

	LIS_DEBUG_FUNC_OUT;
	return 0;
}
LIS_INT lis_ecg(LIS_ESOLVER esolver)
{
  LIS_MATRIX        A;
  LIS_VECTOR        x;
  LIS_SCALAR        evalue;
  LIS_INT               emaxiter;
  LIS_REAL          tol;
  LIS_INT               iter,iter3,nsolver,i,j,output;
  LIS_INT               nprocs,my_rank;
  LIS_REAL          nrm2,resid,resid3;
  LIS_SCALAR        lshift;
  LIS_VECTOR        b,D,r,w,p,Aw,Ax,Ap,ones,Ds;
  LIS_SCALAR        *SA, *SB, *SW, *v3, *SAv3, *SBv3, *z3, *q3, *SBz3, evalue3, ievalue3;
  LIS_SOLVER        solver;
  LIS_PRECON        precon;
  LIS_MATRIX        A0;
  LIS_VECTOR        x0,z,q;
  double	    times,itimes,ptimes,p_c_times,p_i_times;
  LIS_INT           nsol, precon_type;
  char              solvername[128], preconname[128];

  A = esolver->A;
  x = esolver->x;
  
  emaxiter = esolver->options[LIS_EOPTIONS_MAXITER];
  tol = esolver->params[LIS_EPARAMS_RESID - LIS_EOPTIONS_LEN]; 
  output  = esolver->options[LIS_EOPTIONS_OUTPUT];
  lshift = esolver->lshift;

  if( A->my_rank==0 ) printf("local shift = %e\n", lshift);
  if (lshift != 0) lis_matrix_shift_diagonal(A, lshift);

  SA = (LIS_SCALAR *)lis_malloc(3*3*sizeof(LIS_SCALAR), "lis_ecg::SA");
  SB = (LIS_SCALAR *)lis_malloc(3*3*sizeof(LIS_SCALAR), "lis_ecg::SB");
  SW = (LIS_SCALAR *)lis_malloc(3*3*sizeof(LIS_SCALAR), "lis_ecg::SW");
  v3 = (LIS_SCALAR *)lis_malloc(3*sizeof(LIS_SCALAR), "lis_ecg::v3");
  SAv3 = (LIS_SCALAR *)lis_malloc(3*sizeof(LIS_SCALAR), "lis_ecg::SAv3");
  SBv3 = (LIS_SCALAR *)lis_malloc(3*sizeof(LIS_SCALAR), "lis_ecg::SBv3");
  SBz3 = (LIS_SCALAR *)lis_malloc(3*sizeof(LIS_SCALAR), "lis_ecg::SBz3");
  z3 = (LIS_SCALAR *)lis_malloc(3*sizeof(LIS_SCALAR), "lis_ecg::z3");
  q3 = (LIS_SCALAR *)lis_malloc(3*sizeof(LIS_SCALAR), "lis_ecg::q3");

  b = esolver->work[0];
  D = esolver->work[1];
  Ds = esolver->work[2];
  r = esolver->work[3];
  w = esolver->work[4];
  p = esolver->work[5];
  Aw = esolver->work[6];
  Ax = esolver->work[7];
  Ap = esolver->work[8];

  lis_vector_set_all(1.0,b);
  lis_vector_nrm2(b, &nrm2);
  lis_vector_scale(1/nrm2, b);
  lis_solver_create(&solver);
  lis_solver_set_option("-i bicg -p ilu",solver);
  lis_solver_set_optionC(solver);
  lis_solver_get_solver(solver, &nsol);
  lis_solver_get_precon(solver, &precon_type);
  lis_get_solvername(nsol, solvername);
  lis_get_preconname(precon_type, preconname);
  printf("solver     : %s %d\n", solvername, nsol);
  printf("precon     : %s %d\n", preconname, precon_type);
  lis_solve(A, b, x, solver);
  lis_vector_copy(b,Ax);

  lis_vector_nrm2(x, &nrm2);
  lis_vector_set_all(0.0,p);
  lis_vector_set_all(0.0,Ap);

  lis_precon_create(solver, &precon);
  solver->precon = precon;

  iter=0;

  while (iter<emaxiter)
    {
      iter = iter + 1;

      lis_vector_dot(x,Ax,&evalue);
      lis_vector_axpyz(-(evalue),x,Ax,r); 
      lis_vector_nrm2(r, &nrm2);
      resid = fabs(nrm2/(evalue));

      if( output )
	{
	  if( output & LIS_EPRINT_MEM ) esolver->residual[iter] = resid;
	  if( output & LIS_EPRINT_OUT && A->my_rank==0 ) printf("iter: %5d  residual = %e\n", iter, resid);
	}

      if (resid<tol) break;  

      lis_psolve(solver, x, w);
      lis_vector_copy(x,Aw);
      lis_vector_nrm2(w, &nrm2);

      lis_vector_dot(w,Aw,&SA[0]);
      lis_vector_dot(x,Aw,&SA[3]);
      lis_vector_dot(p,Aw,&SA[6]);
      SA[1] = SA[3];
      lis_vector_dot(x,Ax,&SA[4]);
      lis_vector_dot(p,Ax,&SA[7]);
      SA[2] = SA[6];
      SA[5] = SA[7];
      lis_vector_dot(p,Ap,&SA[8]);

      lis_vector_dot(w,w,&SB[0]);
      lis_vector_dot(x,w,&SB[3]);
      lis_vector_dot(p,w,&SB[6]);
      SB[1] = SB[3];
      lis_vector_dot(x,x,&SB[4]);
      lis_vector_dot(p,x,&SB[7]);
      SB[2] = SB[6];
      SB[5] = SB[7];
      lis_vector_dot(p,p,&SB[8]);
      
      lis_array_set_all(3, 1.0, v3);

      iter3=0;
      while (iter3<emaxiter)
	{
	  iter3 = iter3 + 1;
	  lis_array_nrm2(3, v3, &nrm2); 
	  lis_array_scale(3, 1/nrm2, v3);
	  lis_array_matvec(3, SB, v3, SBv3, LIS_INS_VALUE);
	  lis_array_invvec(3, SA, SBv3, z3);
	  lis_array_dot2(3, SBv3, z3, &ievalue3);
	  if (ievalue3==0) 
	    {
	      printf("ievalue3 is zero\n");
	      lis_precon_destroy(precon);
	      lis_solver_destroy(solver);
	      esolver->iter       = iter;
	      esolver->resid      = resid;
	      esolver->evalue[0] = evalue;

	      if (lshift != 0) lis_matrix_shift_diagonal(A, -lshift);
	      lis_free(SA);
	      lis_free(SB);
	      lis_free(SW);
	      lis_free(v3);
	      lis_free(SAv3);
	      lis_free(SBv3);
	      lis_free(SBz3);
	      lis_free(z3);
	      lis_free(q3);
	      return LIS_BREAKDOWN;
	    }
	  lis_array_axpyz(3, -ievalue3, SBv3, z3, q3);
	  lis_array_nrm2(3, q3, &resid3); 
	  resid3 = fabs(resid3 / ievalue3);
	  if (resid3<1e-12) break;   
	  lis_array_copy(3,z3,v3);
	}

      evalue3 = 1 / ievalue3;
      
      lis_vector_scale(v3[0],w);  
      lis_vector_axpy(v3[2],p,w);
      lis_vector_xpay(w,v3[1],x);
      lis_vector_copy(w,p);
      
      lis_vector_scale(v3[0],Aw);  
      lis_vector_axpy(v3[2],Ap,Aw);
      lis_vector_xpay(Aw,v3[1],Ax);
      lis_vector_copy(Aw,Ap);
      
      lis_vector_nrm2(x,&nrm2);
      lis_vector_scale(1/nrm2,x);
      lis_vector_scale(1/nrm2,Ax);
      
      lis_vector_nrm2(p,&nrm2);
      lis_vector_scale(1/nrm2,p);
      lis_vector_scale(1/nrm2,Ap);
      
      lis_solver_get_timeex(solver,&times,&itimes,&ptimes,&p_c_times,&p_i_times);
      esolver->ptimes += solver->ptimes;
      esolver->itimes += solver->itimes;
      esolver->p_c_times += solver->p_c_times;
      esolver->p_i_times += solver->p_i_times;

    }

  lis_precon_destroy(precon);
  lis_solver_destroy(solver);

  esolver->iter       = iter;
  esolver->resid      = resid;
  esolver->evalue[0] = evalue;

  if (lshift != 0) lis_matrix_shift_diagonal(A, -lshift);

  lis_free(SA);
  lis_free(SB);
  lis_free(SW);
  lis_free(v3);
  lis_free(SAv3);
  lis_free(SBv3);
  lis_free(SBz3);
  lis_free(z3);
  lis_free(q3);

  if (resid<tol) 
    {
      esolver->retcode = LIS_SUCCESS;
      return LIS_SUCCESS;
    }
  else
    {
      esolver->retcode = LIS_MAXITER;
      return LIS_MAXITER;
    }
}