/[escript]/branches/arrexp_2137_win/paso/src/PCG.c
ViewVC logotype

Diff of /branches/arrexp_2137_win/paso/src/PCG.c

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

trunk/paso/src/PCG.c revision 757 by woo409, Mon Jun 26 13:12:56 2006 UTC temp/paso/src/PCG.c revision 1387 by trankine, Fri Jan 11 07:45:26 2008 UTC
# Line 1  Line 1 
 /* $Id$ */  
1    
2    /* $Id$ */
3    
4  /*  /*******************************************************
5  ********************************************************************************   *
6  *               Copyright   2006 by ACcESS MNRF                                *   *           Copyright 2003-2007 by ACceSS MNRF
7  *                                                                              *   *       Copyright 2007 by University of Queensland
8  *                 http://www.access.edu.au                                     *   *
9  *           Primary Business: Queensland, Australia                            *   *                http://esscc.uq.edu.au
10  *     Licensed under the Open Software License version 3.0             *   *        Primary Business: Queensland, Australia
11  *        http://www.opensource.org/licenses/osl-3.0.php                        *   *  Licensed under the Open Software License version 3.0
12  ********************************************************************************   *     http://www.opensource.org/licenses/osl-3.0.php
13  */   *
14     *******************************************************/
15    
16  /* PCG iterations */  /* PCG iterations */
17    
# Line 22  Line 23 
23  #include <omp.h>  #include <omp.h>
24  #endif  #endif
25    
26    #ifdef PASO_MPI
27    #include <mpi.h>
28    #endif
29    
30  /*  /*
31  *  *
32  *  Purpose  *  Purpose
# Line 72  err_t Paso_Solver_PCG( Line 77  err_t Paso_Solver_PCG(
77    dim_t i0;    dim_t i0;
78    bool_t breakFlag=FALSE, maxIterFlag=FALSE, convergeFlag=FALSE;    bool_t breakFlag=FALSE, maxIterFlag=FALSE, convergeFlag=FALSE;
79    err_t status = SOLVER_NO_ERROR;    err_t status = SOLVER_NO_ERROR;
80    dim_t n = A->num_cols * A-> col_block_size;    dim_t n = Paso_SystemMatrix_getTotalNumRows(A);
81    double *resid = tolerance, *rs=NULL, *p=NULL, *v=NULL, *x2=NULL ;    double *resid = tolerance, *rs=NULL, *p=NULL, *v=NULL, *x2=NULL ;
82    double tau_old,tau,beta,delta,gamma_1,gamma_2,alpha,sum_1,sum_2,sum_3,sum_4,sum_5,tol;    double tau_old,tau,beta,delta,gamma_1,gamma_2,alpha,sum_1,sum_2,sum_3,sum_4,sum_5,tol;
83    double norm_of_residual,norm_of_residual_global;    double norm_of_residual,norm_of_residual_global, loc_sum[2], sum[2];
84    register double r_tmp,d,rs_tmp,x2_tmp,x_tmp;    register double r_tmp,d,rs_tmp,x2_tmp,x_tmp;
85    
86  /*                                                                 */  /*                                                                 */
# Line 135  err_t Paso_Solver_PCG( Line 140  err_t Paso_Solver_PCG(
140             Performance_startMonitor(pp,PERFORMANCE_SOLVER);             Performance_startMonitor(pp,PERFORMANCE_SOLVER);
141             /* tau=v*r    */             /* tau=v*r    */
142             #pragma omp for private(i0) reduction(+:sum_1) schedule(static)             #pragma omp for private(i0) reduction(+:sum_1) schedule(static)
143             for (i0=0;i0<n;i0++) sum_1+=v[i0]*r[i0];             for (i0=0;i0<n;i0++) sum_1+=v[i0]*r[i0]; /* Limit to local values of v[] and r[] */
144               #ifdef PASO_MPI
145                /* In case we have many MPI processes, each of which may have several OMP threads:
146                   OMP master participates in an MPI reduction to get global sum_1 */
147                    #pragma omp master
148                {
149                  loc_sum[0] = sum_1;
150                  MPI_Allreduce(loc_sum, &sum_1, 1, MPI_DOUBLE, MPI_SUM, A->mpi_info->comm);
151                }
152               #endif
153             tau_old=tau;             tau_old=tau;
154             tau=sum_1;             tau=sum_1;
155             /* p=v+beta*p */             /* p=v+beta*p */
# Line 151  err_t Paso_Solver_PCG( Line 165  err_t Paso_Solver_PCG(
165             Performance_stopMonitor(pp,PERFORMANCE_SOLVER);             Performance_stopMonitor(pp,PERFORMANCE_SOLVER);
166             Performance_startMonitor(pp,PERFORMANCE_MVM);             Performance_startMonitor(pp,PERFORMANCE_MVM);
167         Paso_SystemMatrix_MatrixVector_CSR_OFFSET0(ONE, A, p,ZERO,v);         Paso_SystemMatrix_MatrixVector_CSR_OFFSET0(ONE, A, p,ZERO,v);
168           Paso_SystemMatrix_MatrixVector_CSR_OFFSET0(ONE, A, p,ZERO,v);
169             Performance_stopMonitor(pp,PERFORMANCE_MVM);             Performance_stopMonitor(pp,PERFORMANCE_MVM);
170             Performance_startMonitor(pp,PERFORMANCE_SOLVER);             Performance_startMonitor(pp,PERFORMANCE_SOLVER);
171             /* delta=p*v */             /* delta=p*v */
172             #pragma omp for private(i0) reduction(+:sum_2) schedule(static)             #pragma omp for private(i0) reduction(+:sum_2) schedule(static)
173             for (i0=0;i0<n;i0++) sum_2+=v[i0]*p[i0];             for (i0=0;i0<n;i0++) sum_2+=v[i0]*p[i0];
174               #ifdef PASO_MPI
175                   #pragma omp master
176               {
177                 loc_sum[0] = sum_2;
178                 MPI_Allreduce(loc_sum, &sum_2, 1, MPI_DOUBLE, MPI_SUM, A->mpi_info->comm);
179               }
180               #endif
181             delta=sum_2;             delta=sum_2;
182    
183        
# Line 169  err_t Paso_Solver_PCG( Line 191  err_t Paso_Solver_PCG(
191                       d=r[i0]-rs[i0];                       d=r[i0]-rs[i0];
192                       sum_3+=d*d;                       sum_3+=d*d;
193                       sum_4+=d*rs[i0];                       sum_4+=d*rs[i0];
194                  }                 }
195                   #ifdef PASO_MPI
196                       #pragma omp master
197                   {
198                     loc_sum[0] = sum_3;
199                     loc_sum[1] = sum_4;
200                     MPI_Allreduce(loc_sum, sum, 2, MPI_DOUBLE, MPI_SUM, A->mpi_info->comm);
201                     sum_3=sum[0];
202                     sum_4=sum[1];
203                   }
204                    #endif
205                  gamma_1= ( (ABS(sum_3)<= ZERO) ? 0 : -sum_4/sum_3) ;                  gamma_1= ( (ABS(sum_3)<= ZERO) ? 0 : -sum_4/sum_3) ;
206                  gamma_2= ONE-gamma_1;                  gamma_2= ONE-gamma_1;
207                  #pragma omp for private(i0,x2_tmp,x_tmp,rs_tmp) schedule(static)                  #pragma omp for private(i0,x2_tmp,x_tmp,rs_tmp) schedule(static)
# Line 180  err_t Paso_Solver_PCG( Line 212  err_t Paso_Solver_PCG(
212                  }                  }
213                  #pragma omp for private(i0) reduction(+:sum_5) schedule(static)                  #pragma omp for private(i0) reduction(+:sum_5) schedule(static)
214                  for (i0=0;i0<n;++i0) sum_5+=rs[i0]*rs[i0];                  for (i0=0;i0<n;++i0) sum_5+=rs[i0]*rs[i0];
215                    #ifdef PASO_MPI
216                       #pragma omp master
217                   {
218                      loc_sum[0] = sum_5;
219                      MPI_Allreduce(loc_sum, &sum_5, 1, MPI_DOUBLE, MPI_SUM, A->mpi_info->comm);
220                   }
221                    #endif
222                  norm_of_residual=sqrt(sum_5);                  norm_of_residual=sqrt(sum_5);
223                  convergeFlag = norm_of_residual <= tol;                  convergeFlag = norm_of_residual <= tol;
224                  maxIterFlag = num_iter == maxit;                  maxIterFlag = num_iter == maxit;
# Line 209  err_t Paso_Solver_PCG( Line 248  err_t Paso_Solver_PCG(
248    /*     End of PCG */    /*     End of PCG */
249    return status;    return status;
250  }  }
   
 /*  
  * $Log$  
  * Revision 1.2  2005/09/15 03:44:40  jgs  
  * Merge of development branch dev-02 back to main trunk on 2005-09-15  
  *  
  * Revision 1.1.2.1  2005/09/05 06:29:49  gross  
  * These files have been extracted from finley to define a stand alone libray for iterative  
  * linear solvers on the ALTIX. main entry through Paso_solve. this version compiles but  
  * has not been tested yet.  
  *  
  *  
  */  

Legend:
Removed from v.757  
changed lines
  Added in v.1387

  ViewVC Help
Powered by ViewVC 1.1.26