/[escript]/trunk/paso/src/Solver_AMG.c
ViewVC logotype

Diff of /trunk/paso/src/Solver_AMG.c

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

revision 1889 by ksteube, Wed Oct 15 03:26:25 2008 UTC revision 1890 by artak, Fri Oct 17 00:14:22 2008 UTC
# Line 68  to Line 68  to
68     then AMG is applied to S again until S becomes empty     then AMG is applied to S again until S becomes empty
69    
70  */  */
71  Paso_Solver_AMG* Paso_Solver_getAMG(Paso_SparseMatrix *A_p,bool_t verbose) {  Paso_Solver_AMG* Paso_Solver_getAMG(Paso_SparseMatrix *A_p,bool_t verbose,dim_t level) {
72    Paso_Solver_AMG* out=NULL;    Paso_Solver_AMG* out=NULL;
73    dim_t n=A_p->numRows;    dim_t n=A_p->numRows;
74    dim_t n_block=A_p->row_block_size;    dim_t n_block=A_p->row_block_size;
# Line 79  Paso_Solver_AMG* Paso_Solver_getAMG(Paso Line 79  Paso_Solver_AMG* Paso_Solver_getAMG(Paso
79    dim_t i,k,j,j0;    dim_t i,k,j,j0;
80    Paso_SparseMatrix * schur=NULL;    Paso_SparseMatrix * schur=NULL;
81    Paso_SparseMatrix * schur_withFillIn=NULL;    Paso_SparseMatrix * schur_withFillIn=NULL;
   double A11,A12,A13,A21,A22,A23,A31,A32,A33,D,time0,time1,time2;  
82    schur_withFillIn=MEMALLOC(1,Paso_SparseMatrix);    schur_withFillIn=MEMALLOC(1,Paso_SparseMatrix);
83      
84      
85      double A11,A12,A13,A21,A22,A23,A31,A32,A33,D,time0,time1,time2;
86        
87    
88    /* identify independend set of rows/columns */    /* identify independend set of rows/columns */
# Line 101  Paso_Solver_AMG* Paso_Solver_getAMG(Paso Line 103  Paso_Solver_AMG* Paso_Solver_getAMG(Paso
103    out->b_F=NULL;    out->b_F=NULL;
104    out->x_C=NULL;    out->x_C=NULL;
105    out->b_C=NULL;    out->b_C=NULL;
106      out->A=Paso_SparseMatrix_getReference(A_p);
107      out->level=level;
108        
109   /* fprintf(stderr,"START OF MATRIX \n\n");   /* fprintf(stderr,"START OF MATRIX \n\n");
110    for (i = 0; i < A_p->numRows; ++i) {    for (i = 0; i < A_p->numRows; ++i) {
# Line 117  Paso_Solver_AMG* Paso_Solver_getAMG(Paso Line 121  Paso_Solver_AMG* Paso_Solver_getAMG(Paso
121       time0=Paso_timer();       time0=Paso_timer();
122       #pragma omp parallel for private(i) schedule(static)       #pragma omp parallel for private(i) schedule(static)
123       for (i=0;i<n;++i) mis_marker[i]=-1;       for (i=0;i<n;++i) mis_marker[i]=-1;
124       Paso_Pattern_coup(A_p,mis_marker);       Paso_Pattern_RS(A_p,mis_marker,0.25);
125         /*
126         for (i=0;i<n;++i) fprintf(stderr," i=%d mis[i]=%d \n",i,mis_marker[i]);
127         */
128       time2=Paso_timer()-time0;       time2=Paso_timer()-time0;
129       if (Paso_noError()) {       if (Paso_noError()) {
130          #pragma omp parallel for private(i) schedule(static)          #pragma omp parallel for private(i) schedule(static)
# Line 171  Paso_Solver_AMG* Paso_Solver_getAMG(Paso Line 178  Paso_Solver_AMG* Paso_Solver_getAMG(Paso
178                        if (ABS(rs[i])>0.) {                        if (ABS(rs[i])>0.) {
179                              out->inv_A_FF[i]=1./rs[i];                              out->inv_A_FF[i]=1./rs[i];
180                        } else {                        } else {
181                              Paso_setError(ZERO_DIVISION_ERROR, "Paso_Solver_getAMG: Break-down in AMG decomposition: non-regular main diagonal block.");                          out->inv_A_FF[i]=0;
182                        }                        }
183                        
184                         /* } else {
185                                Paso_setError(ZERO_DIVISION_ERROR, "Paso_Solver_getAMG: Break-down in AMG decomposition: non-regular main diagonal block.");
186                          }*/
187                  }                  }
188                }                }
189             } /* end parallel region */             } /* end parallel region */
# Line 180  Paso_Solver_AMG* Paso_Solver_getAMG(Paso Line 191  Paso_Solver_AMG* Paso_Solver_getAMG(Paso
191             if( Paso_noError()) {             if( Paso_noError()) {
192                /* if there are no nodes in the coarse level there is no more work to do */                /* if there are no nodes in the coarse level there is no more work to do */
193                out->n_C=n-out->n_F;                out->n_C=n-out->n_F;
194                if (out->n_C>0) {                 /*if (out->n_C>11) {*/
195                   if (level>0) {
196                     out->rows_in_C=MEMALLOC(out->n_C,index_t);                     out->rows_in_C=MEMALLOC(out->n_C,index_t);
197                     out->mask_C=MEMALLOC(n,index_t);                     out->mask_C=MEMALLOC(n,index_t);
198                     if (! (Paso_checkPtr(out->mask_C) || Paso_checkPtr(out->rows_in_C) ) ) {                     if (! (Paso_checkPtr(out->mask_C) || Paso_checkPtr(out->rows_in_C) ) ) {
# Line 243  Paso_Solver_AMG* Paso_Solver_getAMG(Paso Line 255  Paso_Solver_AMG* Paso_Solver_getAMG(Paso
255                                  /* update A_CC block to get Schur complement and then apply AMG to it */                                  /* update A_CC block to get Schur complement and then apply AMG to it */
256                                  Paso_Solver_updateIncompleteSchurComplement(schur_withFillIn,out->A_CF,out->inv_A_FF,out->A_FF_pivot,out->A_FC);                                  Paso_Solver_updateIncompleteSchurComplement(schur_withFillIn,out->A_CF,out->inv_A_FF,out->A_FF_pivot,out->A_FC);
257                                  time1=Paso_timer()-time1;                                  time1=Paso_timer()-time1;
258                                  out->AMG_of_Schur=Paso_Solver_getAMG(schur_withFillIn,verbose);                                  out->AMG_of_Schur=Paso_Solver_getAMG(schur_withFillIn,verbose,level-1);
259                                                                    
260                                  Paso_SparseMatrix_free(schur);                                  Paso_SparseMatrix_free(schur);
261                                  Paso_SparseMatrix_free(schur_withFillIn);                                  /* Paso_SparseMatrix_free(schur_withFillIn);*/
262                              }                              }
263                              /* allocate work arrays for AMG application */                              /* allocate work arrays for AMG application */
264                              if (Paso_noError()) {                              if (Paso_noError()) {
# Line 254  Paso_Solver_AMG* Paso_Solver_getAMG(Paso Line 266  Paso_Solver_AMG* Paso_Solver_getAMG(Paso
266                                out->b_F=MEMALLOC(n_block*out->n_F,double);                                out->b_F=MEMALLOC(n_block*out->n_F,double);
267                                out->x_C=MEMALLOC(n_block*out->n_C,double);                                out->x_C=MEMALLOC(n_block*out->n_C,double);
268                                out->b_C=MEMALLOC(n_block*out->n_C,double);                                out->b_C=MEMALLOC(n_block*out->n_C,double);
269    
270                                if (! (Paso_checkPtr(out->x_F) || Paso_checkPtr(out->b_F) || Paso_checkPtr(out->x_C) || Paso_checkPtr(out->b_C) ) ) {                                if (! (Paso_checkPtr(out->x_F) || Paso_checkPtr(out->b_F) || Paso_checkPtr(out->x_C) || Paso_checkPtr(out->b_C) ) ) {
271                                    #pragma omp parallel                                    #pragma omp parallel
272                                    {                                    {
# Line 328  Paso_Solver_AMG* Paso_Solver_getAMG(Paso Line 341  Paso_Solver_AMG* Paso_Solver_getAMG(Paso
341  void Paso_Solver_solveAMG(Paso_Solver_AMG * amg, double * x, double * b) {  void Paso_Solver_solveAMG(Paso_Solver_AMG * amg, double * x, double * b) {
342       dim_t i,k;       dim_t i,k;
343       dim_t n_block=amg->n_block;       dim_t n_block=amg->n_block;
344         double *r=MEMALLOC(amg->n,double);
345       if (amg->n_C==0) {       double *x0=MEMALLOC(amg->n,double);
346        
347         if (amg->level==0) {
348          /* x=invA_FF*b  */          /* x=invA_FF*b  */
349          Paso_Solver_applyBlockDiagonalMatrix(n_block,amg->n_F,amg->inv_A_FF,amg->A_FF_pivot,x,b);          Paso_Solver_applyBlockDiagonalMatrix(n_block,amg->n_F,amg->inv_A_FF,amg->A_FF_pivot,x,b);
350       } else {       } else {
351             fprintf(stdout,"LEVEL %d \n",amg->level);
352          /* presmoothing on (Shure, x, b, r) */          /* presmoothing on (Shure, x, b, r) */
353            /****************/
354             Paso_Solver_GS* GS=MEMALLOC(1,Paso_Solver_GS);
355             GS=Paso_Solver_getGS(amg->A,-1);
356             Paso_Solver_solveGS(GS,x,b);
357            
358             #pragma omp parallel for private(i) schedule(static)
359             for (i=0;i<amg->n;++i) r[i]=b[i];
360            
361             /*r=b-Ax*/
362             Paso_SparseMatrix_MatrixVector_CSR_OFFSET0(-1.,amg->A,x,1.,r);
363           /****************/        
364          /* b->[b_F,b_C]     */          /* b->[b_F,b_C]     */
365          if (n_block==1) {          if (n_block==1) {
366             #pragma omp parallel for private(i) schedule(static)             #pragma omp parallel for private(i) schedule(static)
367             for (i=0;i<amg->n_F;++i) amg->b_F[i]=b[amg->rows_in_F[i]];             for (i=0;i<amg->n_F;++i) amg->b_F[i]=r[amg->rows_in_F[i]];    /* was b istead of r */
368             #pragma omp parallel for private(i) schedule(static)            #pragma omp parallel for private(i) schedule(static)
369             for (i=0;i<amg->n_C;++i) amg->b_C[i]=b[amg->rows_in_C[i]];             for (i=0;i<amg->n_C;++i) amg->b_C[i]=r[amg->rows_in_C[i]];
370          } else {          } else {
371             #pragma omp parallel for private(i,k) schedule(static)             #pragma omp parallel for private(i,k) schedule(static)
372             for (i=0;i<amg->n_F;++i)             for (i=0;i<amg->n_F;++i)
373                   for (k=0;k<n_block;k++) amg->b_F[amg->n_block*i+k]=b[n_block*amg->rows_in_F[i]+k];                   for (k=0;k<n_block;k++) amg->b_F[amg->n_block*i+k]=r[n_block*amg->rows_in_F[i]+k];
374             #pragma omp parallel for private(i,k) schedule(static)             #pragma omp parallel for private(i,k) schedule(static)
375             for (i=0;i<amg->n_C;++i)             for (i=0;i<amg->n_C;++i)
376                   for (k=0;k<n_block;k++) amg->b_C[amg->n_block*i+k]=b[n_block*amg->rows_in_C[i]+k];                   for (k=0;k<n_block;k++) amg->b_C[amg->n_block*i+k]=r[n_block*amg->rows_in_C[i]+k];
377          }          }
378    
379            /****************/
380            /* Coursening part */
381            
382            /* r_F=A_FF^-1*r_F  */
383            /*Paso_Solver_applyBlockDiagonalMatrix(n_block,amg->n_F,amg->inv_A_FF,amg->A_FF_pivot,amg->b_F,amg->b_F);
384            fprintf(stderr,"2\n");
385            */
386            /* r_C=r_C-A_CF*r_F */
387            /*Paso_SparseMatrix_MatrixVector_CSR_OFFSET0(-1.,amg->A_CF,amg->b_F,1.,amg->b_C);
388            fprintf(stderr,"3\n");
389            */
390           /****************/
391    
392            
393          /* x_F=invA_FF*b_F  */          /* x_F=invA_FF*b_F  */
394          Paso_Solver_applyBlockDiagonalMatrix(n_block,amg->n_F,amg->inv_A_FF,amg->A_FF_pivot,amg->x_F,amg->b_F);          Paso_Solver_applyBlockDiagonalMatrix(n_block,amg->n_F,amg->inv_A_FF,amg->A_FF_pivot,amg->x_F,amg->b_F);
395            
396          /* b_C=b_C-A_CF*x_F */          /* b_C=b_C-A_CF*x_F */
397          Paso_SparseMatrix_MatrixVector_CSR_OFFSET0(-1.,amg->A_CF,amg->x_F,1.,amg->b_C);          Paso_SparseMatrix_MatrixVector_CSR_OFFSET0(-1.,amg->A_CF,amg->x_F,1.,amg->b_C);
398            
399          /* x_C=AMG(b_C)     */          /* x_C=AMG(b_C)     */
400          Paso_Solver_solveAMG(amg->AMG_of_Schur,amg->x_C,amg->b_C);          Paso_Solver_solveAMG(amg->AMG_of_Schur,amg->x_C,amg->b_C);
401            
402          /* b_F=b_F-A_FC*x_C */          /* b_F=b_F-A_FC*x_C */
403          Paso_SparseMatrix_MatrixVector_CSR_OFFSET0(-1.,amg->A_FC,amg->x_C,1.,amg->b_F);          Paso_SparseMatrix_MatrixVector_CSR_OFFSET0(-1.,amg->A_FC,amg->x_C,1.,amg->b_F);
404          /* x_F=invA_FF*b_F  */          /* x_F=invA_FF*b_F  */
405          Paso_Solver_applyBlockDiagonalMatrix(n_block,amg->n_F,amg->inv_A_FF,amg->A_FF_pivot,amg->x_F,amg->b_F);          Paso_Solver_applyBlockDiagonalMatrix(n_block,amg->n_F,amg->inv_A_FF,amg->A_FF_pivot,amg->x_F,amg->b_F);
406          /* x<-[x_F,x_C]     */          /* x<-[x_F,x_C]     */
407            
408            /* */
409            
410            
411          if (n_block==1) {          if (n_block==1) {
412             #pragma omp parallel for private(i) schedule(static)             #pragma omp parallel for private(i) schedule(static)
413             for (i=0;i<amg->n;++i) {             for (i=0;i<amg->n;++i) {
414                if (amg->mask_C[i]>-1) {                if (amg->mask_C[i]>-1) {
415                    x[i]=amg->x_C[amg->mask_C[i]];                    x[i]+=amg->x_C[amg->mask_C[i]];
416                } else {                } else {
417                    x[i]=amg->x_F[amg->mask_F[i]];                    x[i]+=amg->x_F[amg->mask_F[i]];
418                }                }
419             }             }
420          } else {          } else {
421             #pragma omp parallel for private(i,k) schedule(static)             #pragma omp parallel for private(i,k) schedule(static)
422             for (i=0;i<amg->n;++i) {             for (i=0;i<amg->n;++i) {
423                   if (amg->mask_C[i]>-1) {                   if (amg->mask_C[i]>-1) {
424                       for (k=0;k<n_block;k++) x[n_block*i+k]=amg->x_C[n_block*amg->mask_C[i]+k];                       for (k=0;k<n_block;k++) x[n_block*i+k]+=amg->x_C[n_block*amg->mask_C[i]+k];
425                   } else {                   } else {
426                       for (k=0;k<n_block;k++) x[n_block*i+k]=amg->x_F[n_block*amg->mask_F[i]+k];                       for (k=0;k<n_block;k++) x[n_block*i+k]+=amg->x_F[n_block*amg->mask_F[i]+k];
427                   }                   }
428             }             }
429          }          }
430          /* all done */          /* all done */
431         /*post smoothing*/
432    /*     Paso_Solver_solveGS(GS,x0,b);
433         if (n_block==1) {
434               #pragma omp parallel for private(i) schedule(static)
435               for (i=0;i<amg->n;++i) x[i]+=x0[i];
436            } else {
437               #pragma omp parallel for private(i,k) schedule(static)
438               for (i=0;i<amg->n;++i) {
439                         for (k=0;k<n_block;k++) x[n_block*i+k]+=x0[n_block*i+k];
440               }
441            }
442    */    
443         Paso_Solver_GS_free(GS);
444       }       }
445         MEMFREE(x0);
446         MEMFREE(r);
447       return;       return;
448  }  }
449    

Legend:
Removed from v.1889  
changed lines
  Added in v.1890

  ViewVC Help
Powered by ViewVC 1.1.26