コード例 #1
0
LIS_INT lis_bicr_quad(LIS_SOLVER solver)
{
  LIS_MATRIX A,At;
  LIS_PRECON M;
  LIS_VECTOR b,x;
  LIS_VECTOR r,rtld, z,ztld,p, ptld, ap, map, az, aptld;
  LIS_QUAD_PTR alpha, beta, rho, rho_old, tmpdot1;
  LIS_REAL   bnrm2, nrm2, tol;
  LIS_INT iter,maxiter,n,output,conv;
  double times,ptimes;

  LIS_DEBUG_FUNC_IN;

  A       = solver->A;
  At      = solver->A;
  M       = solver->precon;
  b       = solver->b;
  x       = solver->x;
  n       = A->n;
  maxiter = solver->options[LIS_OPTIONS_MAXITER];
  output  = solver->options[LIS_OPTIONS_OUTPUT];
  conv    = solver->options[LIS_OPTIONS_CONV_COND];
  ptimes  = 0.0;

  r       = solver->work[0];
  rtld    = solver->work[1];
  z       = solver->work[2];
  ztld    = solver->work[3];
  p       = solver->work[4];
  ptld    = solver->work[5];
  ap      = solver->work[6];
  az      = solver->work[7];
  map     = solver->work[8];
  aptld   = solver->work[9];

  LIS_QUAD_SCALAR_MALLOC(alpha,0,1);
  LIS_QUAD_SCALAR_MALLOC(beta,1,1);
  LIS_QUAD_SCALAR_MALLOC(rho,2,1);
  LIS_QUAD_SCALAR_MALLOC(rho_old,3,1);
  LIS_QUAD_SCALAR_MALLOC(tmpdot1,4,1);

  /* Initial Residual */
  if( lis_solver_get_initial_residual(solver,NULL,NULL,r,&bnrm2) )
  {
    LIS_DEBUG_FUNC_OUT;
    return LIS_SUCCESS;
  }
  tol     = solver->tol;

  lis_solver_set_shadowresidual(solver,r,rtld);

  lis_psolve(solver, r, z);
  lis_psolvet(solver, rtld, ztld);
  lis_vector_copyex_mm(z,p);
  lis_vector_copyex_mm(ztld,ptld);
  LIS_MATVEC(A,z,ap);
  lis_vector_dotex_mmm(ap,ztld,&rho_old);

  for( iter=1; iter<=maxiter; iter++ )
  {
    /* aptld = A^T * ptld */
    /* map   = M^-1 * ap  */
    LIS_MATVECT(A,ptld,aptld);
    times = lis_wtime();
    lis_psolve(solver, ap, map);
    ptimes += lis_wtime()-times;

    /* tmpdot1 = <map,aptld> */
    lis_vector_dotex_mmm(map,aptld,&tmpdot1);
    /* test breakdown */
    if( tmpdot1.hi[0]==0.0 && tmpdot1.lo[0]==0.0 )
    {
      solver->retcode   = LIS_BREAKDOWN;
      solver->iter      = iter;
      solver->resid     = nrm2;
      LIS_DEBUG_FUNC_OUT;
      return LIS_BREAKDOWN;
    }

    /* alpha = rho_old / tmpdot1 */
    /* x     = x + alpha*p   */
    /* r     = r - alpha*ap  */
    lis_quad_div((LIS_QUAD *)alpha.hi,(LIS_QUAD *)rho_old.hi,(LIS_QUAD *)tmpdot1.hi);
    lis_vector_axpyex_mmm(alpha,p,x);
    lis_quad_minus((LIS_QUAD *)alpha.hi);
    lis_vector_axpyex_mmm(alpha,ap,r);
    /* convergence check */
    lis_solver_get_residual[conv](r,solver,&nrm2);

    if( output )
    {
      if( output & LIS_PRINT_MEM ) solver->residual[iter] = nrm2;
      if( output & LIS_PRINT_OUT && A->my_rank==0 ) lis_print_rhistory(iter,nrm2);
    }

    if( tol >= nrm2 )
    {
      solver->retcode    = LIS_SUCCESS;
      solver->iter       = iter;
      solver->resid      = nrm2;
      solver->ptimes     = ptimes;
      LIS_DEBUG_FUNC_OUT;
      return LIS_SUCCESS;
    }
    
    /* rtld = rtld - alpha*aptld */
    /* z    = z - alpha*map      */
    /* ztld = M^-T * rtld        */
    /* az   = A * z              */
    /* rho = <az,ztld>           */
    lis_vector_axpyex_mmm(alpha,aptld,rtld);
    lis_vector_axpyex_mmm(alpha,map,z);
    times = lis_wtime();
    lis_psolvet(solver, rtld, ztld);
    ptimes += lis_wtime()-times;
    LIS_MATVEC(A,z,az);
    lis_vector_dotex_mmm(az,ztld,&rho);

    /* test breakdown */
    if( rho.hi[0]==0.0 && rho.lo[0]==0.0 )
    {
      solver->retcode   = LIS_BREAKDOWN;
      solver->iter      = iter;
      solver->resid     = nrm2;
      LIS_DEBUG_FUNC_OUT;
      return LIS_BREAKDOWN;
    }

    /* beta = rho / rho_old    */
    /* p    = z    + beta*p    */
    /* ptld = ztld + beta*ptld */
    /* ap   = az   + beta*ap   */
    lis_quad_div((LIS_QUAD *)beta.hi,(LIS_QUAD *)rho.hi,(LIS_QUAD *)rho_old.hi);
    lis_vector_xpayex_mmm(z,beta,p);
    lis_vector_xpayex_mmm(ztld,beta,ptld);
    lis_vector_xpayex_mmm(az,beta,ap);

    rho_old.hi[0] = rho.hi[0];
    rho_old.lo[0] = rho.lo[0];
  }

  solver->retcode   = LIS_MAXITER;
  solver->iter      = iter;
  solver->resid     = nrm2;
  LIS_DEBUG_FUNC_OUT;
  return LIS_MAXITER;
}
コード例 #2
0
LIS_INT lis_crs_quad(LIS_SOLVER solver)
{
	LIS_MATRIX A;
	LIS_PRECON M;
	LIS_VECTOR b,x;
	LIS_VECTOR r,rtld, p, q, u, z, ap, map, uq, auq;
	LIS_QUAD_PTR alpha, beta, rho, rho_old, tmpdot1, one;
	LIS_REAL   bnrm2, nrm2, tol;
	LIS_INT iter,maxiter,n,output,conv;
	double times,ptimes;

	LIS_DEBUG_FUNC_IN;

	A       = solver->A;
	M       = solver->precon;
	b       = solver->b;
	x       = solver->x;
	n       = A->n;
	maxiter = solver->options[LIS_OPTIONS_MAXITER];
	output  = solver->options[LIS_OPTIONS_OUTPUT];
	conv    = solver->options[LIS_OPTIONS_CONV_COND];
	ptimes  = 0.0;

	r       = solver->work[0];
	rtld    = solver->work[1];
	p       = solver->work[2];
	z       = solver->work[3];
	u       = solver->work[3];
	uq      = solver->work[3];
	q       = solver->work[4];
	ap      = solver->work[4];
	map     = solver->work[5];
	auq     = solver->work[5];
	LIS_QUAD_SCALAR_MALLOC(alpha,0,1);
	LIS_QUAD_SCALAR_MALLOC(beta,1,1);
	LIS_QUAD_SCALAR_MALLOC(rho,2,1);
	LIS_QUAD_SCALAR_MALLOC(rho_old,3,1);
	LIS_QUAD_SCALAR_MALLOC(tmpdot1,4,1);
	LIS_QUAD_SCALAR_MALLOC(one,6,1);

	/* Initial Residual */
	if( lis_solver_get_initial_residual(solver,NULL,NULL,r,&bnrm2) )
	{
		LIS_DEBUG_FUNC_OUT;
		return LIS_SUCCESS;
	}
	tol     = solver->tol;

	lis_solver_set_shadowresidual(solver,r,p);

	LIS_MATVECT(A,p,rtld);
	lis_vector_set_allex_nm(0.0,q);
	lis_vector_set_allex_nm(0.0,p);
	rho_old.hi[0] = 1.0;
	rho_old.lo[0] = 0.0;
	one.hi[0]   = 1.0;
	one.lo[0]   = 0.0;

	
	for( iter=1; iter<=maxiter; iter++ )
	{
		/* z   = M^-1 * r  */
		/* rho = <rtld,z>  */
		times = lis_wtime();
		lis_psolve(solver, r, z);
		ptimes += lis_wtime()-times;
		lis_vector_dotex_mmm(rtld,z,&rho);

		/* test breakdown */
		if( rho.hi[0]==0.0 && rho.lo[0]==0.0 )
		{
			solver->retcode   = LIS_BREAKDOWN;
			solver->iter      = iter;
			solver->resid     = nrm2;
			LIS_DEBUG_FUNC_OUT;
			return LIS_BREAKDOWN;
		}

		/* beta    = rho / rho_old         */
		/* u       = z + beta*q            */
		/* p       = u + beta*(q + beta*p) */
		/* ap      = A * p                 */
		/* map     = M^-1 * ap             */
		/* tmpdot1 = <rtld,map>            */
		lis_quad_div((LIS_QUAD *)beta.hi,(LIS_QUAD *)rho.hi,(LIS_QUAD *)rho_old.hi);
		lis_vector_axpyzex_mmmm(beta,q,z,u);
		lis_vector_xpayex_mmm(q,beta,p);
		lis_vector_xpayex_mmm(u,beta,p);
		LIS_MATVEC(A,p,ap);
		times = lis_wtime();
		lis_psolve(solver, ap, map);
		ptimes += lis_wtime()-times;
		lis_vector_dotex_mmm(rtld,map,&tmpdot1);
		/* test breakdown */
		if( tmpdot1.hi[0]==0.0 && tmpdot1.lo[0]==0.0 )
		{
			solver->retcode   = LIS_BREAKDOWN;
			solver->iter      = iter;
			solver->resid     = nrm2;
			LIS_DEBUG_FUNC_OUT;
			return LIS_BREAKDOWN;
		}
		
		/* alpha = rho / tmpdot1 */
		/* q     = u - alpha*map */
		/* uq    = u + q         */
		/* auq   = A * uq        */
		/* x     = x + alpha*uq  */
		/* r     = r - alpha*auq */
		lis_quad_div((LIS_QUAD *)alpha.hi,(LIS_QUAD *)rho.hi,(LIS_QUAD *)tmpdot1.hi);
		lis_quad_minus((LIS_QUAD *)alpha.hi);
		lis_vector_axpyzex_mmmm(alpha,map,u,q);
		lis_vector_axpyzex_mmmm(one,u,q,uq);
		LIS_MATVEC(A,uq,auq);
		lis_quad_minus((LIS_QUAD *)alpha.hi);
		lis_vector_axpyex_mmm(alpha,uq,x);
		lis_quad_minus((LIS_QUAD *)alpha.hi);
		lis_vector_axpyex_mmm(alpha,auq,r);

		/* convergence check */
		lis_solver_get_residual[conv](r,solver,&nrm2);
		if( output )
		{
			if( output & LIS_PRINT_MEM ) solver->residual[iter] = nrm2;
			if( output & LIS_PRINT_OUT ) printf("iter: %5d  residual = %e\n", iter, nrm2);
		}
		
		if( tol >= nrm2 )
		{
			solver->retcode    = LIS_SUCCESS;
			solver->iter       = iter;
			solver->resid      = nrm2;
			solver->ptimes     = ptimes;
			LIS_DEBUG_FUNC_OUT;
			return LIS_SUCCESS;
		}
		
		rho_old.hi[0] = rho.hi[0];
		rho_old.lo[0] = rho.lo[0];
	}

	solver->retcode   = LIS_MAXITER;
	solver->iter      = iter;
	solver->resid     = nrm2;
	LIS_DEBUG_FUNC_OUT;
	return LIS_MAXITER;
}
コード例 #3
0
LIS_INT lis_bicg_switch(LIS_SOLVER solver)
{
  LIS_MATRIX A,At;
  LIS_PRECON M;
  LIS_VECTOR b,x;
  LIS_VECTOR r,rtld, z,ztld,p, ptld, q, qtld;
  LIS_QUAD_PTR alpha, beta, rho, rho_old, tmpdot1;
  LIS_REAL   bnrm2, nrm2, tol, tol2;
  LIS_INT iter,maxiter,n,output,conv;
  LIS_INT iter2,maxiter2;
  double times,ptimes;

  LIS_DEBUG_FUNC_IN;


  A       = solver->A;
  At      = solver->A;
  M        = solver->precon;
  b        = solver->b;
  x        = solver->x;
  n        = A->n;
  maxiter  = solver->options[LIS_OPTIONS_MAXITER];
  maxiter2 = solver->options[LIS_OPTIONS_SWITCH_MAXITER];
  output   = solver->options[LIS_OPTIONS_OUTPUT];
  conv    = solver->options[LIS_OPTIONS_CONV_COND];
  tol      = solver->params[LIS_PARAMS_RESID-LIS_OPTIONS_LEN];
  tol2     = solver->params[LIS_PARAMS_SWITCH_RESID-LIS_OPTIONS_LEN];
  ptimes   = 0.0;

  r        = solver->work[0];
  rtld     = solver->work[1];
  z        = solver->work[2];
  ztld     = solver->work[3];
  p        = solver->work[4];
  ptld     = solver->work[5];
  q        = solver->work[2];
  qtld     = solver->work[3];

  LIS_QUAD_SCALAR_MALLOC(alpha,0,1);
  LIS_QUAD_SCALAR_MALLOC(beta,1,1);
  LIS_QUAD_SCALAR_MALLOC(rho,2,1);
  LIS_QUAD_SCALAR_MALLOC(rho_old,3,1);
  LIS_QUAD_SCALAR_MALLOC(tmpdot1,4,1);
  rho_old.hi[0] = 1.0;
  rho_old.lo[0] = 0.0;


  /* Initial Residual */
  if( lis_solver_get_initial_residual(solver,NULL,NULL,r,&bnrm2) )
  {
    LIS_DEBUG_FUNC_OUT;
    return LIS_SUCCESS;
  }
  tol2     = solver->tol_switch;

  lis_solver_set_shadowresidual(solver,r,rtld);

  lis_vector_set_allex_nm(0.0, p);
  lis_vector_set_allex_nm(0.0, ptld);

  r->precision = LIS_PRECISION_DEFAULT;
  rtld->precision = LIS_PRECISION_DEFAULT;
  p->precision = LIS_PRECISION_DEFAULT;
  ptld->precision = LIS_PRECISION_DEFAULT;

  for( iter=1; iter<=maxiter2; iter++ )
  {
    /* z    = M^-1 * r */
    /* ztld = M^-T * rtld */
    times = lis_wtime();
    lis_psolve(solver, r, z);
    lis_psolvet(solver, rtld, ztld);
    ptimes += lis_wtime()-times;

    /* rho = <z,rtld> */
    lis_vector_dot(z,rtld,&rho.hi[0]);

    /* test breakdown */
    if( rho.hi[0]==0.0 )
    {
      solver->retcode   = LIS_BREAKDOWN;
      solver->iter      = iter;
      solver->iter2     = iter;
      solver->resid     = nrm2;
      LIS_DEBUG_FUNC_OUT;
      return LIS_BREAKDOWN;
    }

    /* beta = (rho / rho_old) */
    beta.hi[0] = rho.hi[0] / rho_old.hi[0];

    /* p    = z    + beta*p    */
    /* ptld = ztld + beta*ptld */
    
    /* q    = A   * p    */
    /* qtld = A^T * ptld */
    lis_vector_xpay(z,beta.hi[0],p);
    LIS_MATVEC(A,p,q);

    lis_vector_xpay(ztld,beta.hi[0],ptld);
    LIS_MATVECT(At,ptld,qtld);

    
    /* tmpdot1 = <ptld,q> */
    lis_vector_dot(ptld,q,&tmpdot1.hi[0]);

    /* test breakdown */
    if( tmpdot1.hi[0]==0.0 )
    {
      solver->retcode   = LIS_BREAKDOWN;
      solver->iter      = iter;
      solver->iter2     = iter;
      solver->resid     = nrm2;
      LIS_DEBUG_FUNC_OUT;
      return LIS_BREAKDOWN;
    }
    
    /* alpha = rho / tmpdot1 */
    alpha.hi[0] = rho.hi[0] / tmpdot1.hi[0];
    
    /* x = x + alpha*p */
    lis_vector_axpy(alpha.hi[0],p,x);
    
    /* r    = r    - alpha*q    */
    lis_vector_axpy(-alpha.hi[0],q,r);
    
    /* convergence check */
    lis_solver_get_residual[conv](r,solver,&nrm2);

    if( output )
    {
      if( output & LIS_PRINT_MEM ) solver->residual[iter] = nrm2;
      if( output & LIS_PRINT_OUT && A->my_rank==0 ) lis_print_rhistory(iter,nrm2);
    }

    if( nrm2 <= tol2 )
    {
      solver->iter       = iter;
      solver->iter2     = iter;
      solver->ptimes     = ptimes;
      break;
    }
    
    /* rtld = rtld - alpha*qtld */
    lis_vector_axpy(-alpha.hi[0],qtld,rtld);

    rho_old.hi[0] = rho.hi[0];
  }

  r->precision = LIS_PRECISION_QUAD;
  rtld->precision = LIS_PRECISION_QUAD;
  p->precision = LIS_PRECISION_QUAD;
  ptld->precision = LIS_PRECISION_QUAD;

/*  solver->precon->precon_type = 0;*/
  solver->options[LIS_OPTIONS_INITGUESS_ZEROS] = LIS_FALSE;
  lis_vector_copyex_mn(x,solver->xx);
  rho_old.hi[0] = 1.0;
  lis_solver_get_initial_residual(solver,NULL,NULL,r,&bnrm2);
  tol     = solver->tol;

  lis_solver_set_shadowresidual(solver,r,rtld);

  lis_vector_set_allex_nm(0.0, p);
  lis_vector_set_allex_nm(0.0, ptld);

  for( iter2=iter+1; iter2<=maxiter; iter2++ )
  {
    /* z    = M^-1 * r */
    /* ztld = M^-T * rtld */
    times = lis_wtime();
    lis_psolve(solver, r, z);
    lis_psolvet(solver, rtld, ztld);
/*    memset(z->value_lo,0,n*sizeof(LIS_SCALAR));
    memset(ztld->value_lo,0,n*sizeof(LIS_SCALAR));*/
    ptimes += lis_wtime()-times;

    /* rho = <z,rtld> */
    lis_vector_dotex_mmm(z,rtld,&rho);
/*    printf("rho = %e %e\n",rho.hi[0],rho.lo[0]);*/

    /* test breakdown */
    if( rho.hi[0]==0.0 && rho.lo[0]==0.0 )
    {
      solver->retcode   = LIS_BREAKDOWN;
      solver->iter      = iter2;
      solver->resid     = nrm2;
      LIS_DEBUG_FUNC_OUT;
      return LIS_BREAKDOWN;
    }

    /* beta = (rho / rho_old) */
    lis_quad_div((LIS_QUAD *)beta.hi,(LIS_QUAD *)rho.hi,(LIS_QUAD *)rho_old.hi);

    /* p    = z    + beta*p    */
    /* ptld = ztld + beta*ptld */
    
    /* q    = A   * p    */
    /* qtld = A^T * ptld */
    lis_vector_xpayex_mmm(z,beta,p);
    LIS_MATVEC(A,p,q);

    lis_vector_xpayex_mmm(ztld,beta,ptld);
    LIS_MATVECT(At,ptld,qtld);

    
    /* tmpdot1 = <ptld,q> */
    lis_vector_dotex_mmm(ptld,q,&tmpdot1);

    /* test breakdown */
    if( tmpdot1.hi[0]==0.0 && tmpdot1.lo[0]==0.0 )
    {
      solver->retcode   = LIS_BREAKDOWN;
      solver->iter      = iter2;
      solver->resid     = nrm2;
      LIS_DEBUG_FUNC_OUT;
      return LIS_BREAKDOWN;
    }
    
    /* alpha = rho / tmpdot1 */
    lis_quad_div((LIS_QUAD *)alpha.hi,(LIS_QUAD *)rho.hi,(LIS_QUAD *)tmpdot1.hi);
    
    /* x = x + alpha*p */
    lis_vector_axpyex_mmm(alpha,p,x);
    
    /* r    = r    - alpha*q    */
    lis_quad_minus((LIS_QUAD *)alpha.hi);
    lis_vector_axpyex_mmm(alpha,q,r);
    
    /* convergence check */
    lis_solver_get_residual[conv](r,solver,&nrm2);
    if( output )
    {
      if( output & LIS_PRINT_MEM ) solver->residual[iter2] = nrm2;
      if( output & LIS_PRINT_OUT && A->my_rank==0 ) lis_print_rhistory(iter,nrm2);
    }

    if( tol > nrm2 )
    {
      solver->retcode    = LIS_SUCCESS;
      solver->iter       = iter2;
      solver->iter2      = iter;
      solver->resid      = nrm2;
      solver->ptimes     = ptimes;
      LIS_DEBUG_FUNC_OUT;
      return LIS_SUCCESS;
    }
    
    /* rtld = rtld - alpha*qtld */
    lis_vector_axpyex_mmm(alpha,qtld,rtld);

    rho_old.hi[0] = rho.hi[0];
    rho_old.lo[0] = rho.lo[0];
  }

  solver->retcode   = LIS_MAXITER;
  solver->iter      = iter2;
  solver->iter2     = iter;
  solver->resid     = nrm2;
  LIS_DEBUG_FUNC_OUT;
  return LIS_MAXITER;
}
コード例 #4
0
LIS_INT lis_bicg(LIS_SOLVER solver)
{
  LIS_MATRIX A,At;
  LIS_PRECON M;
  LIS_VECTOR b,x;
  LIS_VECTOR r,rtld, z,ztld,p, ptld, q, qtld;
  LIS_SCALAR alpha, beta, rho, rho_old, tmpdot1;
  LIS_REAL   bnrm2, nrm2, tol;
  LIS_INT iter,maxiter,n,output,conv;
  double times,ptimes;

  LIS_DEBUG_FUNC_IN;

  A       = solver->A;
  At      = solver->A;
  M       = solver->precon;
  b       = solver->b;
  x       = solver->x;
  n       = A->n;
  maxiter = solver->options[LIS_OPTIONS_MAXITER];
  output  = solver->options[LIS_OPTIONS_OUTPUT];
  conv    = solver->options[LIS_OPTIONS_CONV_COND];
  ptimes  = 0.0;

  r       = solver->work[0];
  rtld    = solver->work[1];
  z       = solver->work[2];
  ztld    = solver->work[3];
  p       = solver->work[4];
  ptld    = solver->work[5];
  q       = solver->work[2];
  qtld    = solver->work[3];
  rho_old = (LIS_SCALAR)1.0;



  /* Initial Residual */
  if( lis_solver_get_initial_residual(solver,NULL,NULL,r,&bnrm2) )
  {
    LIS_DEBUG_FUNC_OUT;
    return LIS_SUCCESS;
  }
  tol     = solver->tol;

  lis_solver_set_shadowresidual(solver,r,rtld);

  lis_vector_set_all(0, p);
  lis_vector_set_all(0, ptld);

  for( iter=1; iter<=maxiter; iter++ )
  {
    /* z    = M^-1 * r */
    /* ztld = M^-T * rtld */
    times = lis_wtime();
    lis_psolve(solver, r, z);
    lis_psolvet(solver, rtld, ztld);
    ptimes += lis_wtime()-times;

    /* rho = <z,rtld> */
    lis_vector_dot(z,rtld,&rho);
/*    printf("rho = %e\n",rho);*/

    /* test breakdown */
    if( rho==0.0 )
    {
      solver->retcode   = LIS_BREAKDOWN;
      solver->iter      = iter;
      solver->resid     = nrm2;
      LIS_DEBUG_FUNC_OUT;
      return LIS_BREAKDOWN;
    }

    /* beta = (rho / rho_old) */
    beta = rho / rho_old;

    /* p    = z    + beta*p    */
    /* ptld = ztld + beta*ptld */
    
    /* q    = A   * p    */
    /* qtld = A^T * ptld */
    lis_vector_xpay(z,beta,p);
    LIS_MATVEC(A,p,q);

    lis_vector_xpay(ztld,beta,ptld);
    LIS_MATVECT(At,ptld,qtld);

    
    /* tmpdot1 = <ptld,q> */
    lis_vector_dot(ptld,q,&tmpdot1);
/*    printf("tmpdot1 = %e\n",tmpdot1);*/

    /* test breakdown */
    if( tmpdot1==0.0 )
    {
      solver->retcode   = LIS_BREAKDOWN;
      solver->iter      = iter;
      solver->resid     = nrm2;
      LIS_DEBUG_FUNC_OUT;
      return LIS_BREAKDOWN;
    }
    
    /* alpha = rho / tmpdot1 */
    alpha = rho / tmpdot1;
    
    /* x = x + alpha*p */
    lis_vector_axpy(alpha,p,x);
    
    /* r    = r    - alpha*q    */
    lis_vector_axpy(-alpha,q,r);
    
    /* convergence check */
    lis_solver_get_residual[conv](r,solver,&nrm2);

    if( output )
    {
      if( output & LIS_PRINT_MEM ) solver->residual[iter] = nrm2;
      if( output & LIS_PRINT_OUT && A->my_rank==0 ) lis_print_rhistory(iter,nrm2);
    }

    if( tol >= nrm2 )
    {
      solver->retcode    = LIS_SUCCESS;
      solver->iter       = iter;
      solver->resid      = nrm2;
      solver->ptimes     = ptimes;
      LIS_DEBUG_FUNC_OUT;
      return LIS_SUCCESS;
    }
    
    /* rtld = rtld - alpha*qtld */
    lis_vector_axpy(-alpha,qtld,rtld);

    rho_old = rho;
  }

  solver->retcode   = LIS_MAXITER;
  solver->iter      = iter;
  solver->resid     = nrm2;
  LIS_DEBUG_FUNC_OUT;
  return LIS_MAXITER;
}
コード例 #5
0
LIS_INT lis_bicrstab(LIS_SOLVER solver)
{
	LIS_MATRIX A;
	LIS_PRECON M;
	LIS_VECTOR b,x;
	LIS_VECTOR r,rtld, p, s, ap, ms, map, ams, z;
	LIS_SCALAR alpha, beta, omega, rho, rho_old, tmpdot1, tmpdot2;
	LIS_REAL   bnrm2, nrm2, tol;
	LIS_INT iter,maxiter,n,output,conv;
	double times,ptimes;

	LIS_DEBUG_FUNC_IN;

	A       = solver->A;
	M       = solver->precon;
	b       = solver->b;
	x       = solver->x;
	n       = A->n;
	maxiter = solver->options[LIS_OPTIONS_MAXITER];
	output  = solver->options[LIS_OPTIONS_OUTPUT];
	conv    = solver->options[LIS_OPTIONS_CONV_COND];
	ptimes  = 0.0;

	rtld    = solver->work[0];
	r       = solver->work[1];
	s       = solver->work[2];
	ms      = solver->work[3];
	ams     = solver->work[4];
	p       = solver->work[5];
	ap      = solver->work[6];
	map     = solver->work[7];
	z       = solver->work[8];

	/* Initial Residual */
	if( lis_solver_get_initial_residual(solver,NULL,NULL,r,&bnrm2) )
	{
		LIS_DEBUG_FUNC_OUT;
		return LIS_SUCCESS;
	}
	tol     = solver->tol;

	lis_solver_set_shadowresidual(solver,r,p);

	LIS_MATVECT(A,p,rtld);
	times = lis_wtime();
	lis_psolve(solver, r, z);
	ptimes += lis_wtime()-times;
	lis_vector_copy(z,p);
	lis_vector_dot(rtld,z,&rho_old);
	
	for( iter=1; iter<=maxiter; iter++ )
	{
		/* ap      = A * p             */
		/* map     = M^-1 * ap         */
		/* tmpdot1 = <rtld,map>        */
		/* alpha   = rho_old / tmpdot1 */
		/* s       = r - alpha*ap      */
		LIS_MATVEC(A,p,ap);
		times = lis_wtime();
		lis_psolve(solver, ap, map);
		ptimes += lis_wtime()-times;
		lis_vector_dot(rtld,map,&tmpdot1);
		alpha = rho_old / tmpdot1;
		lis_vector_axpyz(-alpha,ap,r,s);

		/* Early check for tolerance */
		lis_solver_get_residual[conv](s,solver,&nrm2);
		if( nrm2 <= tol )
		{
			if( output )
			{
				if( output & LIS_PRINT_MEM ) solver->residual[iter] = nrm2;
				if( output & LIS_PRINT_OUT && A->my_rank==0 ) printf("iter: %5d  residual = %e\n", iter, nrm2);
			}

			lis_vector_axpy(alpha,p,x);
			solver->retcode    = LIS_SUCCESS;
			solver->iter       = iter;
			solver->resid      = nrm2;
			solver->ptimes     = ptimes;
			LIS_DEBUG_FUNC_OUT;
			return LIS_SUCCESS;
		}

		/* ms      = z - alpha*map     */
		/* ams     = A * ms            */
		/* tmpdot1 = <ams,s>           */
		/* tmpdot2 = <ams,ams>         */
		/* omega   = tmpdot1 / tmpdot2 */
		lis_vector_axpyz(-alpha,map,z,ms);
		LIS_MATVEC(A,ms,ams);
		lis_vector_dot(ams,s,&tmpdot1);
		lis_vector_dot(ams,ams,&tmpdot2);
		omega   = tmpdot1 / tmpdot2;

		/* x = x + alpha*p  + omega*ms  */
		/* r = s - omega*ams            */
		lis_vector_axpy(alpha,p,x);
		lis_vector_axpy(omega,ms,x);
		lis_vector_axpyz(-omega,ams,s,r);
		
		/* convergence check */
		lis_solver_get_residual[conv](r,solver,&nrm2);
		if( output )
		{
			if( output & LIS_PRINT_MEM ) solver->residual[iter] = nrm2;
			if( output & LIS_PRINT_OUT && A->my_rank==0 ) printf("iter: %5d  residual = %e\n", iter, nrm2);
		}

		if( tol >= nrm2 )
		{
			solver->retcode    = LIS_SUCCESS;
			solver->iter       = iter;
			solver->resid      = nrm2;
			solver->ptimes     = ptimes;
			LIS_DEBUG_FUNC_OUT;
			return LIS_SUCCESS;
		}
		
		/* z   = M^-1 * r */
		/* rho = <rtld,z> */
		times = lis_wtime();
		lis_psolve(solver, r, z);
		ptimes += lis_wtime()-times;
		lis_vector_dot(rtld,z,&rho);
		if( rho==0.0 )
		{
			solver->retcode   = LIS_BREAKDOWN;
			solver->iter      = iter;
			solver->resid     = nrm2;
			LIS_DEBUG_FUNC_OUT;
			return LIS_BREAKDOWN;
		}

		/* beta = (rho / rho_old) * (alpha / omega) */
		/* p    = z + beta*(p - omega*map)          */
		beta = (rho / rho_old) * (alpha / omega);
		lis_vector_axpy(-omega,map,p);
		lis_vector_xpay(z,beta,p);

		rho_old = rho;
	}

	solver->retcode   = LIS_MAXITER;
	solver->iter      = iter;
	solver->resid     = nrm2;
	LIS_DEBUG_FUNC_OUT;
	return LIS_MAXITER;
}
コード例 #6
0
LIS_INT lis_bicrsafe_quad(LIS_SOLVER solver)
{
	LIS_MATRIX A;
	LIS_PRECON M;
	LIS_VECTOR b,x;
	LIS_VECTOR r, rtld, artld, mr, amr, p, ap, map;
	LIS_VECTOR y, my, u, au, z;
	LIS_QUAD_PTR alpha, beta, rho, rho_old;
	LIS_QUAD_PTR qsi, eta, one;
	LIS_QUAD_PTR tmp, tmpdot[5];
	LIS_REAL   bnrm2, nrm2, tol;
	LIS_INT iter,maxiter,n,output,conv;
	double times,ptimes;

	LIS_DEBUG_FUNC_IN;

	A       = solver->A;
	M       = solver->precon;
	b       = solver->b;
	x       = solver->x;
	n       = A->n;
	maxiter = solver->options[LIS_OPTIONS_MAXITER];
	output  = solver->options[LIS_OPTIONS_OUTPUT];
	conv    = solver->options[LIS_OPTIONS_CONV_COND];
	ptimes  = 0.0;

	rtld    = solver->work[0];
	r       = solver->work[1];
	mr      = solver->work[2];
	amr     = solver->work[3];
	p       = solver->work[4];
	ap      = solver->work[5];
	map     = solver->work[6];
	my      = solver->work[7];
	y       = solver->work[8];
	u       = solver->work[9];
	z       = solver->work[10];
	au      = solver->work[11];
	artld   = solver->work[12];

	LIS_QUAD_SCALAR_MALLOC(alpha,0,1);
	LIS_QUAD_SCALAR_MALLOC(beta,1,1);
	LIS_QUAD_SCALAR_MALLOC(rho,2,1);
	LIS_QUAD_SCALAR_MALLOC(rho_old,3,1);
	LIS_QUAD_SCALAR_MALLOC(qsi,4,1);
	LIS_QUAD_SCALAR_MALLOC(eta,5,1);
	LIS_QUAD_SCALAR_MALLOC(tmp,6,1);
	LIS_QUAD_SCALAR_MALLOC(tmpdot[0],7,1);
	LIS_QUAD_SCALAR_MALLOC(tmpdot[1],8,1);
	LIS_QUAD_SCALAR_MALLOC(tmpdot[2],9,1);
	LIS_QUAD_SCALAR_MALLOC(tmpdot[3],10,1);
	LIS_QUAD_SCALAR_MALLOC(tmpdot[4],11,1);
	LIS_QUAD_SCALAR_MALLOC(one,13,1);


	/* Initial Residual */
	if( lis_solver_get_initial_residual(solver,NULL,NULL,r,&bnrm2) )
	{
		LIS_DEBUG_FUNC_OUT;
		return LIS_SUCCESS;
	}
	tol     = solver->tol;

	lis_solver_set_shadowresidual(solver,r,rtld);

	LIS_MATVECT(A,rtld,artld);
	times = lis_wtime();
	lis_psolve(solver, r, mr);
	ptimes += lis_wtime()-times;
	LIS_MATVEC(A,mr,amr);
	lis_vector_dotex_mmm(rtld,amr,&rho_old);
	lis_vector_copyex_mm(amr,ap);
	lis_vector_copyex_mm(mr,p);
	one.hi[0] = -1.0;
	one.lo[0] = 0.0;

	
	for( iter=1; iter<=maxiter; iter++ )
	{
		/* map  = M^-1 * ap */
		times = lis_wtime();
		lis_psolve(solver, ap, map);
		ptimes += lis_wtime()-times;

		/* tmpdot[0] = <artld,map> */
		/* alpha = rho_old / tmpdot[0] */
		lis_vector_dotex_mmm(artld,map,&tmpdot[0]);
		lis_quad_div((LIS_QUAD *)alpha.hi,(LIS_QUAD *)rho_old.hi,(LIS_QUAD *)tmpdot[0].hi);


		/* tmpdot[0] = <y,y>           */
		/* tmpdot[1] = <amr,r>         */
		/* tmpdot[2] = <y,r>           */
		/* tmpdot[3] = <amr,y>         */
		/* tmpdot[4] = <amr,amr>       */
		lis_vector_dotex_mmm(y,y,&tmpdot[0]);
		lis_vector_dotex_mmm(amr,r,&tmpdot[1]);
		lis_vector_dotex_mmm(y,r,&tmpdot[2]);
		lis_vector_dotex_mmm(amr,y,&tmpdot[3]);
		lis_vector_dotex_mmm(amr,amr,&tmpdot[4]);
		if(iter==1)
		{
			lis_quad_div((LIS_QUAD *)qsi.hi,(LIS_QUAD *)tmpdot[1].hi,(LIS_QUAD *)tmpdot[4].hi);
			eta.hi[0] = 0.0;
			eta.lo[0] = 0.0;
		}
		else
		{
			lis_quad_mul((LIS_QUAD *)tmp.hi,(LIS_QUAD *)tmpdot[4].hi,(LIS_QUAD *)tmpdot[0].hi);
			lis_quad_sqr((LIS_QUAD *)qsi.hi,(LIS_QUAD *)tmpdot[3].hi);
			lis_quad_sub((LIS_QUAD *)tmp.hi,(LIS_QUAD *)tmp.hi,(LIS_QUAD *)qsi.hi);

			lis_quad_mul((LIS_QUAD *)qsi.hi,(LIS_QUAD *)tmpdot[0].hi,(LIS_QUAD *)tmpdot[1].hi);
			lis_quad_mul((LIS_QUAD *)eta.hi,(LIS_QUAD *)tmpdot[2].hi,(LIS_QUAD *)tmpdot[3].hi);
			lis_quad_sub((LIS_QUAD *)qsi.hi,(LIS_QUAD *)qsi.hi,(LIS_QUAD *)eta.hi);
			lis_quad_div((LIS_QUAD *)qsi.hi,(LIS_QUAD *)qsi.hi,(LIS_QUAD *)tmp.hi);

			lis_quad_mul((LIS_QUAD *)eta.hi,(LIS_QUAD *)tmpdot[4].hi,(LIS_QUAD *)tmpdot[2].hi);
			lis_quad_mul((LIS_QUAD *)tmpdot[0].hi,(LIS_QUAD *)tmpdot[3].hi,(LIS_QUAD *)tmpdot[1].hi);
			lis_quad_sub((LIS_QUAD *)eta.hi,(LIS_QUAD *)eta.hi,(LIS_QUAD *)tmpdot[0].hi);
			lis_quad_div((LIS_QUAD *)eta.hi,(LIS_QUAD *)eta.hi,(LIS_QUAD *)tmp.hi);
		}

		/* u    = qsi*map + eta*my + eta*beta*u */
		/* au   = A * u                         */
		lis_quad_mul((LIS_QUAD *)tmp.hi,(LIS_QUAD *)eta.hi,(LIS_QUAD *)beta.hi);
		lis_vector_scaleex_mm(tmp,u);
		lis_vector_axpyex_mmm(qsi,map,u);
		lis_vector_axpyex_mmm(eta,my,u);
		LIS_MATVEC(A,u,au);

		/* z = qsi*mr + eta*z - alpha*u */
		lis_vector_scaleex_mm(eta,z);
		lis_vector_axpyex_mmm(qsi,mr,z);
		lis_quad_minus((LIS_QUAD *)alpha.hi);
		lis_vector_axpyex_mmm(alpha,u,z);

		/* y  = qsi*amr + eta*y - alpha*au */
		/* my = M^-1 * y */
		lis_vector_scaleex_mm(eta,y);
		lis_vector_axpyex_mmm(qsi,amr,y);
		lis_vector_axpyex_mmm(alpha,au,y);
		times = lis_wtime();
		lis_psolve(solver, y, my);
		ptimes += lis_wtime()-times;

		/* x = x + alpha*p + z */
		lis_quad_minus((LIS_QUAD *)alpha.hi);
		lis_vector_axpyex_mmm(alpha,p,x);
		lis_quad_minus((LIS_QUAD *)one.hi);
		lis_vector_axpyex_mmm(one,z,x);
		
		/* r = r - alpha*ap - y */
		lis_quad_minus((LIS_QUAD *)alpha.hi);
		lis_quad_minus((LIS_QUAD *)one.hi);
		lis_vector_axpyex_mmm(alpha,ap,r);
		lis_vector_axpyex_mmm(one,y,r);
		
		/* convergence check */
		lis_solver_get_residual[conv](r,solver,&nrm2);
		if( output )
		{
			if( output & LIS_PRINT_MEM ) solver->residual[iter] = nrm2;
			if( output & LIS_PRINT_OUT && A->my_rank==0 ) lis_print_rhistory(iter,nrm2);
		}

		if( tol >= nrm2 )
		{
			solver->retcode    = LIS_SUCCESS;
			solver->iter       = iter;
			solver->resid      = nrm2;
			solver->ptimes     = ptimes;
			LIS_DEBUG_FUNC_OUT;
			return LIS_SUCCESS;
		}

		/* mr  = mr - alpha*map - my */
		/* amr = A * mr              */
		/* rho = <rtld,amr> */
		lis_vector_axpyex_mmm(alpha,map,mr);
		lis_vector_axpyex_mmm(one,my,mr);
		LIS_MATVEC(A,mr,amr);
		lis_vector_dotex_mmm(rtld,amr,&rho);
		if( rho.hi[0]==0.0 && rho.lo[0]==0.0 )
		{
			solver->retcode   = LIS_BREAKDOWN;
			solver->iter      = iter;
			solver->resid     = nrm2;
			LIS_DEBUG_FUNC_OUT;
			return LIS_BREAKDOWN;
		}

		/* beta = (rho / rho_old) * (alpha / qsi) */
		lis_quad_minus((LIS_QUAD *)alpha.hi);
		lis_quad_div((LIS_QUAD *)beta.hi,(LIS_QUAD *)rho.hi,(LIS_QUAD *)rho_old.hi);
		lis_quad_div((LIS_QUAD *)tmp.hi,(LIS_QUAD *)alpha.hi,(LIS_QUAD *)qsi.hi);
		lis_quad_mul((LIS_QUAD *)beta.hi,(LIS_QUAD *)beta.hi,(LIS_QUAD *)tmp.hi);


		/* p  = mr + beta*(p - u)    */
		/* ap = amr + beta*(ap - au) */
		lis_vector_axpyex_mmm(one,u,p);
		lis_vector_xpayex_mmm(mr,beta,p);
		lis_vector_axpyex_mmm(one,au,ap);
		lis_vector_xpayex_mmm(amr,beta,ap);

		rho_old.hi[0] = rho.hi[0];
		rho_old.lo[0] = rho.lo[0];
	}

	solver->retcode   = LIS_MAXITER;
	solver->iter      = iter;
	solver->resid     = nrm2;
	LIS_DEBUG_FUNC_OUT;
	return LIS_MAXITER;
}
コード例 #7
0
LIS_INT lis_bicrsafe(LIS_SOLVER solver)
{
	LIS_MATRIX A;
	LIS_PRECON M;
	LIS_VECTOR b,x;
	LIS_VECTOR r, rtld, artld, mr, amr, p, ap, map;
	LIS_VECTOR y, my, u, au, z;
	LIS_SCALAR alpha, beta, rho, rho_old;
	LIS_SCALAR qsi, eta;
	LIS_SCALAR tmp, tmpdot[5];
	LIS_REAL   bnrm2, nrm2, tol;
	LIS_INT iter,maxiter,n,output,conv;
	double times,ptimes;

	LIS_DEBUG_FUNC_IN;

	A       = solver->A;
	M       = solver->precon;
	b       = solver->b;
	x       = solver->x;
	n       = A->n;
	maxiter = solver->options[LIS_OPTIONS_MAXITER];
	output  = solver->options[LIS_OPTIONS_OUTPUT];
	conv    = solver->options[LIS_OPTIONS_CONV_COND];
	ptimes  = 0.0;

	rtld    = solver->work[0];
	r       = solver->work[1];
	mr      = solver->work[2];
	amr     = solver->work[3];
	p       = solver->work[4];
	ap      = solver->work[5];
	map     = solver->work[6];
	my      = solver->work[7];
	y       = solver->work[8];
	u       = solver->work[9];
	z       = solver->work[10];
	au      = solver->work[11];
	artld   = solver->work[12];


	/* Initial Residual */
	if( lis_solver_get_initial_residual(solver,NULL,NULL,r,&bnrm2) )
	{
		LIS_DEBUG_FUNC_OUT;
		return LIS_SUCCESS;
	}
	tol     = solver->tol;

	lis_solver_set_shadowresidual(solver,r,rtld);

	LIS_MATVECT(A,rtld,artld);
	times = lis_wtime();
	lis_psolve(solver, r, mr);
	ptimes += lis_wtime()-times;
	LIS_MATVEC(A,mr,amr);
	lis_vector_dot(rtld,amr,&rho_old);
	lis_vector_copy(amr,ap);
	lis_vector_copy(mr,p);
	beta = 0.0;

	
	for( iter=1; iter<=maxiter; iter++ )
	{
		/* map  = M^-1 * ap */
		times = lis_wtime();
		lis_psolve(solver, ap, map);
		ptimes += lis_wtime()-times;

		/* tmpdot[0] = <artld,map> */
		/* alpha = rho_old / tmpdot[0] */
		lis_vector_dot(artld,map,&tmpdot[0]);
		alpha = rho_old / tmpdot[0];


		/* tmpdot[0] = <y,y>           */
		/* tmpdot[1] = <amr,r>         */
		/* tmpdot[2] = <y,r>           */
		/* tmpdot[3] = <amr,y>         */
		/* tmpdot[4] = <amr,amr>       */
		lis_vector_dot(y,y,&tmpdot[0]);
		lis_vector_dot(amr,r,&tmpdot[1]);
		lis_vector_dot(y,r,&tmpdot[2]);
		lis_vector_dot(amr,y,&tmpdot[3]);
		lis_vector_dot(amr,amr,&tmpdot[4]);
		if(iter==1)
		{
			qsi = tmpdot[1] / tmpdot[4];
			eta = 0.0;
		}
		else
		{
			tmp = tmpdot[4]*tmpdot[0] - tmpdot[3]*tmpdot[3];
			qsi = (tmpdot[0]*tmpdot[1] - tmpdot[2]*tmpdot[3]) / tmp;
			eta = (tmpdot[4]*tmpdot[2] - tmpdot[3]*tmpdot[1]) / tmp;
		}

		/* u    = qsi*map + eta*my + eta*beta*u */
		/* au   = A * u                         */
		lis_vector_scale(eta*beta,u);
		lis_vector_axpy(qsi,map,u);
		lis_vector_axpy(eta,my,u);
		LIS_MATVEC(A,u,au);

		/* z = qsi*mr + eta*z - alpha*u */
		lis_vector_scale(eta,z);
		lis_vector_axpy(qsi,mr,z);
		lis_vector_axpy(-alpha,u,z);

		/* y  = qsi*amr + eta*y - alpha*au */
		/* my = M^-1 * y */
		lis_vector_scale(eta,y);
		lis_vector_axpy(qsi,amr,y);
		lis_vector_axpy(-alpha,au,y);
		times = lis_wtime();
		lis_psolve(solver, y, my);
		ptimes += lis_wtime()-times;

		/* x = x + alpha*p + z */
		lis_vector_axpy(alpha,p,x);
		lis_vector_axpy(1.0,z,x);
		
		/* r = r - alpha*ap - y */
		lis_vector_axpy(-alpha,ap,r);
		lis_vector_axpy(-1.0,y,r);
		
		/* convergence check */
		lis_solver_get_residual[conv](r,solver,&nrm2);
		if( output )
		{
			if( output & LIS_PRINT_MEM ) solver->residual[iter] = nrm2;
			if( output & LIS_PRINT_OUT && A->my_rank==0 ) lis_print_rhistory(iter,nrm2);
		}

		if( tol >= nrm2 )
		{
			solver->retcode    = LIS_SUCCESS;
			solver->iter       = iter;
			solver->resid      = nrm2;
			solver->ptimes     = ptimes;
			LIS_DEBUG_FUNC_OUT;
			return LIS_SUCCESS;
		}

		/* mr  = mr - alpha*map - my */
		/* amr = A * mr              */
		/* rho = <rtld,amr> */
		lis_vector_axpy(-alpha,map,mr);
		lis_vector_axpy(-1.0,my,mr);
		LIS_MATVEC(A,mr,amr);
		lis_vector_dot(rtld,amr,&rho);
		if( rho==0.0 )
		{
			solver->retcode   = LIS_BREAKDOWN;
			solver->iter      = iter;
			solver->resid     = nrm2;
			LIS_DEBUG_FUNC_OUT;
			return LIS_BREAKDOWN;
		}

		/* beta = (rho / rho_old) * (alpha / qsi) */
		beta = (rho / rho_old) * (alpha / qsi);


		/* p  = mr + beta*(p - u)    */
		/* ap = amr + beta*(ap - au) */
		lis_vector_axpy(-1.0,u,p);
		lis_vector_xpay(mr,beta,p);
		lis_vector_axpy(-1.0,au,ap);
		lis_vector_xpay(amr,beta,ap);

		rho_old = rho;
	}

	solver->retcode   = LIS_MAXITER;
	solver->iter      = iter;
	solver->resid     = nrm2;
	LIS_DEBUG_FUNC_OUT;
	return LIS_MAXITER;
}