1 #include <../src/ksp/pc/impls/gamg/gamg.h> /*I "petscpc.h" I*/ 2 #include <petsc-private/kspimpl.h> 3 4 typedef struct { 5 PetscReal dummy; /* empty struct; save for later */ 6 } PC_GAMG_Classical; 7 8 9 #undef __FUNCT__ 10 #define __FUNCT__ "PCGAMGClassicalCreateGhostVector_Private" 11 PetscErrorCode PCGAMGClassicalCreateGhostVector_Private(Mat G,Vec *gvec,PetscInt **global) 12 { 13 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)G->data; 14 PetscErrorCode ierr; 15 PetscBool isMPIAIJ; 16 17 PetscFunctionBegin; 18 ierr = PetscObjectTypeCompare((PetscObject)G, MATMPIAIJ, &isMPIAIJ); CHKERRQ(ierr); 19 if (isMPIAIJ) { 20 if (gvec)ierr = VecDuplicate(aij->lvec,gvec);CHKERRQ(ierr); 21 if (global)*global = aij->garray; 22 } else { 23 /* no off-processor nodes */ 24 if (gvec)*gvec = NULL; 25 if (global)*global = NULL; 26 } 27 PetscFunctionReturn(0); 28 } 29 30 #undef __FUNCT__ 31 #define __FUNCT__ "PCGAMGClassicalGraphSplitting_Private" 32 /* 33 Split the relevant graph into diagonal and off-diagonal parts in local numbering; for now this 34 a roundabout private interface to the mats' internal diag and offdiag mats. 35 */ 36 PetscErrorCode PCGAMGClassicalGraphSplitting_Private(Mat G,Mat *Gd, Mat *Go) 37 { 38 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)G->data; 39 PetscErrorCode ierr; 40 PetscBool isMPIAIJ; 41 PetscFunctionBegin; 42 ierr = PetscObjectTypeCompare((PetscObject)G, MATMPIAIJ, &isMPIAIJ ); CHKERRQ(ierr); 43 if (isMPIAIJ) { 44 *Gd = aij->A; 45 *Go = aij->B; 46 } else { 47 *Gd = G; 48 *Go = NULL; 49 } 50 PetscFunctionReturn(0); 51 } 52 53 #undef __FUNCT__ 54 #define __FUNCT__ "PCGAMGGraph_Classical" 55 PetscErrorCode PCGAMGGraph_Classical(PC pc,const Mat A,Mat *G) 56 { 57 PetscInt s,f,n,idx,lidx,gidx; 58 PetscInt r,c,ncols; 59 const PetscInt *rcol; 60 const PetscScalar *rval; 61 PetscInt *gcol; 62 PetscScalar *gval; 63 PetscReal rmax; 64 PetscInt cmax = 0; 65 PC_MG *mg; 66 PC_GAMG *gamg; 67 PetscErrorCode ierr; 68 PetscInt *gsparse,*lsparse; 69 PetscScalar *Amax; 70 MatType mtype; 71 72 PetscFunctionBegin; 73 mg = (PC_MG *)pc->data; 74 gamg = (PC_GAMG *)mg->innerctx; 75 76 ierr = MatGetOwnershipRange(A,&s,&f);CHKERRQ(ierr); 77 n=f-s; 78 ierr = PetscMalloc(sizeof(PetscInt)*n,&lsparse);CHKERRQ(ierr); 79 ierr = PetscMalloc(sizeof(PetscInt)*n,&gsparse);CHKERRQ(ierr); 80 ierr = PetscMalloc(sizeof(PetscScalar)*n,&Amax);CHKERRQ(ierr); 81 82 for (r = 0;r < n;r++) { 83 lsparse[r] = 0; 84 gsparse[r] = 0; 85 } 86 87 for (r = s;r < f;r++) { 88 /* determine the maximum off-diagonal in each row */ 89 rmax = 0.; 90 ierr = MatGetRow(A,r,&ncols,&rcol,&rval);CHKERRQ(ierr); 91 for (c = 0; c < ncols; c++) { 92 if (PetscRealPart(-rval[c]) > rmax && rcol[c] != r) { 93 rmax = PetscRealPart(-rval[c]); 94 } 95 } 96 Amax[r-s] = rmax; 97 if (ncols > cmax) cmax = ncols; 98 lidx = 0; 99 gidx = 0; 100 /* create the local and global sparsity patterns */ 101 for (c = 0; c < ncols; c++) { 102 if (PetscRealPart(-rval[c]) > gamg->threshold*PetscRealPart(Amax[r-s])) { 103 if (rcol[c] < f && rcol[c] >= s) { 104 lidx++; 105 } else { 106 gidx++; 107 } 108 } 109 } 110 ierr = MatRestoreRow(A,r,&ncols,&rcol,&rval);CHKERRQ(ierr); 111 lsparse[r-s] = lidx; 112 gsparse[r-s] = gidx; 113 } 114 ierr = PetscMalloc(sizeof(PetscScalar)*cmax,&gval);CHKERRQ(ierr); 115 ierr = PetscMalloc(sizeof(PetscInt)*cmax,&gcol);CHKERRQ(ierr); 116 117 ierr = MatCreate(PetscObjectComm((PetscObject)A),G); CHKERRQ(ierr); 118 ierr = MatGetType(A,&mtype);CHKERRQ(ierr); 119 ierr = MatSetType(*G,mtype);CHKERRQ(ierr); 120 ierr = MatSetSizes(*G,n,n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 121 ierr = MatMPIAIJSetPreallocation(*G,0,lsparse,0,gsparse);CHKERRQ(ierr); 122 ierr = MatSeqAIJSetPreallocation(*G,0,lsparse);CHKERRQ(ierr); 123 for (r = s;r < f;r++) { 124 ierr = MatGetRow(A,r,&ncols,&rcol,&rval);CHKERRQ(ierr); 125 idx = 0; 126 for (c = 0; c < ncols; c++) { 127 /* classical strength of connection */ 128 if (PetscRealPart(-rval[c]) > gamg->threshold*PetscRealPart(Amax[r-s])) { 129 gcol[idx] = rcol[c]; 130 gval[idx] = rval[c]; 131 idx++; 132 } 133 } 134 ierr = MatSetValues(*G,1,&r,idx,gcol,gval,INSERT_VALUES);CHKERRQ(ierr); 135 ierr = MatRestoreRow(A,r,&ncols,&rcol,&rval);CHKERRQ(ierr); 136 } 137 ierr = MatAssemblyBegin(*G, MAT_FINAL_ASSEMBLY); CHKERRQ(ierr); 138 ierr = MatAssemblyEnd(*G, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 139 140 ierr = PetscFree(gval);CHKERRQ(ierr); 141 ierr = PetscFree(gcol);CHKERRQ(ierr); 142 ierr = PetscFree(lsparse);CHKERRQ(ierr); 143 ierr = PetscFree(gsparse);CHKERRQ(ierr); 144 ierr = PetscFree(Amax);CHKERRQ(ierr); 145 PetscFunctionReturn(0); 146 } 147 148 149 #undef __FUNCT__ 150 #define __FUNCT__ "PCGAMGCoarsen_Classical" 151 PetscErrorCode PCGAMGCoarsen_Classical(PC pc,Mat *G,PetscCoarsenData **agg_lists) 152 { 153 PetscErrorCode ierr; 154 MatCoarsen crs; 155 MPI_Comm fcomm = ((PetscObject)pc)->comm; 156 157 PetscFunctionBegin; 158 159 160 /* construct the graph if necessary */ 161 if (!G) { 162 SETERRQ(fcomm,PETSC_ERR_ARG_WRONGSTATE,"Must set Graph in PC in PCGAMG before coarsening"); 163 } 164 165 ierr = MatCoarsenCreate(fcomm,&crs);CHKERRQ(ierr); 166 ierr = MatCoarsenSetFromOptions(crs);CHKERRQ(ierr); 167 ierr = MatCoarsenSetAdjacency(crs,*G);CHKERRQ(ierr); 168 ierr = MatCoarsenSetStrictAggs(crs,PETSC_TRUE);CHKERRQ(ierr); 169 ierr = MatCoarsenApply(crs);CHKERRQ(ierr); 170 ierr = MatCoarsenGetData(crs,agg_lists);CHKERRQ(ierr); 171 ierr = MatCoarsenDestroy(&crs);CHKERRQ(ierr); 172 173 PetscFunctionReturn(0); 174 } 175 176 #undef __FUNCT__ 177 #define __FUNCT__ "PCGAMGClassicalGhost_Private" 178 /* 179 Find all ghost nodes that are coarse and output the fine/coarse splitting for those as well 180 181 Input: 182 G - graph; 183 gvec - Global Vector 184 avec - Local part of the scattered vec 185 bvec - Global part of the scattered vec 186 187 Output: 188 findx - indirection t 189 190 */ 191 PetscErrorCode PCGAMGClassicalGhost_Private(Mat G,Vec v,Vec gv) 192 { 193 PetscErrorCode ierr; 194 Mat_MPIAIJ *aij = (Mat_MPIAIJ*)G->data; 195 PetscBool isMPIAIJ; 196 197 PetscFunctionBegin; 198 ierr = PetscObjectTypeCompare((PetscObject)G, MATMPIAIJ, &isMPIAIJ ); CHKERRQ(ierr); 199 if (isMPIAIJ) { 200 ierr = VecScatterBegin(aij->Mvctx,v,gv,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 201 ierr = VecScatterEnd(aij->Mvctx,v,gv,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 202 } 203 PetscFunctionReturn(0); 204 } 205 206 #undef __FUNCT__ 207 #define __FUNCT__ "PCGAMGProlongator_Classical" 208 PetscErrorCode PCGAMGProlongator_Classical(PC pc, const Mat A, const Mat G, PetscCoarsenData *agg_lists,Mat *P) 209 { 210 PetscErrorCode ierr; 211 MPI_Comm comm; 212 PetscReal *Amax_pos,*Amax_neg; 213 Mat lA,gA; /* on and off diagonal matrices */ 214 PetscInt fn; /* fine local blocked sizes */ 215 PetscInt cn; /* coarse local blocked sizes */ 216 PetscInt gn; /* size of the off-diagonal fine vector */ 217 PetscInt fs,fe; /* fine (row) ownership range*/ 218 PetscInt cs,ce; /* coarse (column) ownership range */ 219 PetscInt i,j; /* indices! */ 220 PetscBool iscoarse; /* flag for determining if a node is coarse */ 221 PetscInt *lcid,*gcid; /* on and off-processor coarse unknown IDs */ 222 PetscInt *lsparse,*gsparse; /* on and off-processor sparsity patterns for prolongator */ 223 PetscScalar pij; 224 const PetscScalar *rval; 225 const PetscInt *rcol; 226 PetscScalar g_pos,g_neg,a_pos,a_neg,diag,invdiag,alpha,beta; 227 Vec F; /* vec of coarse size */ 228 Vec C; /* vec of fine size */ 229 Vec gF; /* vec of off-diagonal fine size */ 230 MatType mtype; 231 PetscInt c_indx; 232 PetscScalar c_scalar; 233 PetscInt ncols,col; 234 PetscInt row_f,row_c; 235 PetscInt cmax=0,idx; 236 PetscScalar *pvals; 237 PetscInt *pcols; 238 PC_MG *mg = (PC_MG*)pc->data; 239 PC_GAMG *gamg = (PC_GAMG*)mg->innerctx; 240 241 PetscFunctionBegin; 242 comm = ((PetscObject)pc)->comm; 243 ierr = MatGetOwnershipRange(A,&fs,&fe); CHKERRQ(ierr); 244 fn = (fe - fs); 245 246 ierr = MatGetVecs(A,&F,NULL);CHKERRQ(ierr); 247 248 /* get the number of local unknowns and the indices of the local unknowns */ 249 250 ierr = PetscMalloc(sizeof(PetscInt)*fn,&lsparse);CHKERRQ(ierr); 251 ierr = PetscMalloc(sizeof(PetscInt)*fn,&gsparse);CHKERRQ(ierr); 252 ierr = PetscMalloc(sizeof(PetscInt)*fn,&lcid);CHKERRQ(ierr); 253 ierr = PetscMalloc(sizeof(PetscReal)*fn,&Amax_pos);CHKERRQ(ierr); 254 ierr = PetscMalloc(sizeof(PetscReal)*fn,&Amax_neg);CHKERRQ(ierr); 255 256 /* count the number of coarse unknowns */ 257 cn = 0; 258 for (i=0;i<fn;i++) { 259 /* filter out singletons */ 260 ierr = PetscCDEmptyAt(agg_lists,i,&iscoarse); CHKERRQ(ierr); 261 lcid[i] = -1; 262 if (!iscoarse) { 263 cn++; 264 } 265 } 266 267 /* create the coarse vector */ 268 ierr = VecCreateMPI(comm,cn,PETSC_DECIDE,&C);CHKERRQ(ierr); 269 ierr = VecGetOwnershipRange(C,&cs,&ce);CHKERRQ(ierr); 270 271 /* construct a global vector indicating the global indices of the coarse unknowns */ 272 cn = 0; 273 for (i=0;i<fn;i++) { 274 ierr = PetscCDEmptyAt(agg_lists,i,&iscoarse); CHKERRQ(ierr); 275 if (!iscoarse) { 276 lcid[i] = cs+cn; 277 cn++; 278 } else { 279 lcid[i] = -1; 280 } 281 *((PetscInt *)&c_scalar) = lcid[i]; 282 c_indx = fs+i; 283 ierr = VecSetValues(F,1,&c_indx,&c_scalar,INSERT_VALUES);CHKERRQ(ierr); 284 } 285 286 ierr = VecAssemblyBegin(F);CHKERRQ(ierr); 287 ierr = VecAssemblyEnd(F);CHKERRQ(ierr); 288 289 /* determine the biggest off-diagonal entries in each row */ 290 for (i=fs;i<fe;i++) { 291 Amax_pos[i-fs] = 0.; 292 Amax_neg[i-fs] = 0.; 293 ierr = MatGetRow(A,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 294 for(j=0;j<ncols;j++){ 295 if ((PetscRealPart(-rval[j]) > Amax_neg[i-fs]) && i != rcol[j]) Amax_neg[i-fs] = PetscAbsScalar(rval[j]); 296 if ((PetscRealPart(rval[j]) > Amax_pos[i-fs]) && i != rcol[j]) Amax_pos[i-fs] = PetscAbsScalar(rval[j]); 297 } 298 if (ncols > cmax) cmax = ncols; 299 ierr = MatRestoreRow(A,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 300 } 301 ierr = PetscMalloc(sizeof(PetscInt)*cmax,&pcols);CHKERRQ(ierr); 302 ierr = PetscMalloc(sizeof(PetscScalar)*cmax,&pvals);CHKERRQ(ierr); 303 304 /* split the operator into two */ 305 ierr = PCGAMGClassicalGraphSplitting_Private(A,&lA,&gA);CHKERRQ(ierr); 306 307 /* scatter to the ghost vector */ 308 ierr = PCGAMGClassicalCreateGhostVector_Private(A,&gF,NULL);CHKERRQ(ierr); 309 ierr = PCGAMGClassicalGhost_Private(A,F,gF);CHKERRQ(ierr); 310 311 if (gA) { 312 ierr = VecGetSize(gF,&gn);CHKERRQ(ierr); 313 ierr = PetscMalloc(sizeof(PetscInt)*gn,&gcid);CHKERRQ(ierr); 314 for (i=0;i<gn;i++) { 315 ierr = VecGetValues(gF,1,&i,&c_scalar);CHKERRQ(ierr); 316 gcid[i] = *((PetscInt *)&c_scalar); 317 } 318 } 319 320 ierr = VecDestroy(&F);CHKERRQ(ierr); 321 ierr = VecDestroy(&gF);CHKERRQ(ierr); 322 ierr = VecDestroy(&C);CHKERRQ(ierr); 323 324 /* count the on and off processor sparsity patterns for the prolongator */ 325 for (i=0;i<fn;i++) { 326 /* on */ 327 lsparse[i] = 0; 328 gsparse[i] = 0; 329 if (lcid[i] >= 0) { 330 lsparse[i] = 1; 331 gsparse[i] = 0; 332 } else { 333 ierr = MatGetRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 334 for (j = 0;j < ncols;j++) { 335 col = rcol[j]; 336 if (lcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) { 337 lsparse[i] += 1; 338 } 339 } 340 ierr = MatRestoreRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 341 /* off */ 342 if (gA) { 343 ierr = MatGetRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 344 for (j = 0; j < ncols; j++) { 345 col = rcol[j]; 346 if (gcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) { 347 gsparse[i] += 1; 348 } 349 } 350 ierr = MatRestoreRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 351 } 352 } 353 } 354 355 /* preallocate and create the prolongator */ 356 ierr = MatCreate(comm,P); CHKERRQ(ierr); 357 ierr = MatGetType(G,&mtype);CHKERRQ(ierr); 358 ierr = MatSetType(*P,mtype);CHKERRQ(ierr); 359 360 ierr = MatSetSizes(*P,fn,cn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 361 ierr = MatMPIAIJSetPreallocation(*P,0,lsparse,0,gsparse);CHKERRQ(ierr); 362 ierr = MatSeqAIJSetPreallocation(*P,0,lsparse);CHKERRQ(ierr); 363 364 /* loop over local fine nodes -- get the diagonal, the sum of positive and negative strong and weak weights, and set up the row */ 365 for (i = 0;i < fn;i++) { 366 /* determine on or off */ 367 row_f = i + fs; 368 row_c = lcid[i]; 369 if (row_c >= 0) { 370 pij = 1.; 371 ierr = MatSetValues(*P,1,&row_f,1,&row_c,&pij,INSERT_VALUES);CHKERRQ(ierr); 372 } else { 373 g_pos = 0.; 374 g_neg = 0.; 375 a_pos = 0.; 376 a_neg = 0.; 377 diag = 0.; 378 379 /* local connections */ 380 ierr = MatGetRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 381 for (j = 0; j < ncols; j++) { 382 col = rcol[j]; 383 if (lcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) { 384 if (PetscRealPart(rval[j]) > 0.) { 385 g_pos += rval[j]; 386 } else { 387 g_neg += rval[j]; 388 } 389 } 390 if (col != i) { 391 if (PetscRealPart(rval[j]) > 0.) { 392 a_pos += rval[j]; 393 } else { 394 a_neg += rval[j]; 395 } 396 } else { 397 diag = rval[j]; 398 } 399 } 400 ierr = MatRestoreRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 401 402 /* ghosted connections */ 403 if (gA) { 404 ierr = MatGetRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 405 for (j = 0; j < ncols; j++) { 406 col = rcol[j]; 407 if (gcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) { 408 if (PetscRealPart(rval[j]) > 0.) { 409 g_pos += rval[j]; 410 } else { 411 g_neg += rval[j]; 412 } 413 } 414 if (PetscRealPart(rval[j]) > 0.) { 415 a_pos += rval[j]; 416 } else { 417 a_neg += rval[j]; 418 } 419 } 420 ierr = MatRestoreRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 421 } 422 423 if (g_neg == 0.) { 424 alpha = 0.; 425 } else { 426 alpha = -a_neg/g_neg; 427 } 428 429 if (g_pos == 0.) { 430 diag += a_pos; 431 beta = 0.; 432 } else { 433 beta = -a_pos/g_pos; 434 } 435 if (diag == 0.) { 436 invdiag = 0.; 437 } else invdiag = 1. / diag; 438 /* on */ 439 ierr = MatGetRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 440 idx = 0; 441 for (j = 0;j < ncols;j++) { 442 col = rcol[j]; 443 if (lcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) { 444 row_f = i + fs; 445 row_c = lcid[col]; 446 /* set the values for on-processor ones */ 447 if (PetscRealPart(rval[j]) < 0.) { 448 pij = rval[j]*alpha*invdiag; 449 } else { 450 pij = rval[j]*beta*invdiag; 451 } 452 if (PetscAbsScalar(pij) != 0.) { 453 pvals[idx] = pij; 454 pcols[idx] = row_c; 455 idx++; 456 } 457 } 458 } 459 ierr = MatRestoreRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 460 /* off */ 461 if (gA) { 462 ierr = MatGetRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 463 for (j = 0; j < ncols; j++) { 464 col = rcol[j]; 465 if (gcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) { 466 row_f = i + fs; 467 row_c = gcid[col]; 468 /* set the values for on-processor ones */ 469 if (PetscRealPart(rval[j]) < 0.) { 470 pij = rval[j]*alpha*invdiag; 471 } else { 472 pij = rval[j]*beta*invdiag; 473 } 474 if (PetscAbsScalar(pij) != 0.) { 475 pvals[idx] = pij; 476 pcols[idx] = row_c; 477 idx++; 478 } 479 } 480 } 481 ierr = MatRestoreRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr); 482 } 483 ierr = MatSetValues(*P,1,&row_f,idx,pcols,pvals,INSERT_VALUES);CHKERRQ(ierr); 484 } 485 } 486 487 ierr = MatAssemblyBegin(*P, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 488 ierr = MatAssemblyEnd(*P, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 489 490 ierr = PetscFree(lsparse);CHKERRQ(ierr); 491 ierr = PetscFree(gsparse);CHKERRQ(ierr); 492 ierr = PetscFree(pcols);CHKERRQ(ierr); 493 ierr = PetscFree(pvals);CHKERRQ(ierr); 494 ierr = PetscFree(Amax_pos);CHKERRQ(ierr); 495 ierr = PetscFree(Amax_neg);CHKERRQ(ierr); 496 ierr = PetscFree(lcid);CHKERRQ(ierr); 497 if (gA) {ierr = PetscFree(gcid);CHKERRQ(ierr);} 498 499 PetscFunctionReturn(0); 500 } 501 502 #undef __FUNCT__ 503 #define __FUNCT__ "PCGAMGDestroy_Classical" 504 PetscErrorCode PCGAMGDestroy_Classical(PC pc) 505 { 506 PetscErrorCode ierr; 507 PC_MG *mg = (PC_MG*)pc->data; 508 PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx; 509 510 PetscFunctionBegin; 511 ierr = PetscFree(pc_gamg->subctx);CHKERRQ(ierr); 512 PetscFunctionReturn(0); 513 } 514 515 #undef __FUNCT__ 516 #define __FUNCT__ "PCGAMGSetFromOptions_Classical" 517 PetscErrorCode PCGAMGSetFromOptions_Classical(PC pc) 518 { 519 PetscFunctionBegin; 520 PetscFunctionReturn(0); 521 } 522 523 #undef __FUNCT__ 524 #define __FUNCT__ "PCGAMGSetData_Classical" 525 PetscErrorCode PCGAMGSetData_Classical(PC pc, Mat A) 526 { 527 PC_MG *mg = (PC_MG*)pc->data; 528 PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx; 529 530 PetscFunctionBegin; 531 /* no data for classical AMG */ 532 pc_gamg->data = NULL; 533 pc_gamg->data_cell_cols = 0; 534 pc_gamg->data_cell_rows = 0; 535 pc_gamg->data_sz = 0; 536 PetscFunctionReturn(0); 537 } 538 539 /* -------------------------------------------------------------------------- */ 540 /* 541 PCCreateGAMG_Classical 542 543 */ 544 #undef __FUNCT__ 545 #define __FUNCT__ "PCCreateGAMG_Classical" 546 PetscErrorCode PCCreateGAMG_Classical(PC pc) 547 { 548 PetscErrorCode ierr; 549 PC_MG *mg = (PC_MG*)pc->data; 550 PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx; 551 PC_GAMG_Classical *pc_gamg_classical; 552 553 PetscFunctionBegin; 554 if (pc_gamg->subctx) { 555 /* call base class */ 556 ierr = PCDestroy_GAMG(pc);CHKERRQ(ierr); 557 } 558 559 /* create sub context for SA */ 560 ierr = PetscNewLog(pc, PC_GAMG_Classical, &pc_gamg_classical);CHKERRQ(ierr); 561 pc_gamg->subctx = pc_gamg_classical; 562 pc->ops->setfromoptions = PCGAMGSetFromOptions_Classical; 563 /* reset does not do anything; setup not virtual */ 564 565 /* set internal function pointers */ 566 pc_gamg->ops->destroy = PCGAMGDestroy_Classical; 567 pc_gamg->ops->graph = PCGAMGGraph_Classical; 568 pc_gamg->ops->coarsen = PCGAMGCoarsen_Classical; 569 pc_gamg->ops->prolongator = PCGAMGProlongator_Classical; 570 pc_gamg->ops->optprol = NULL; 571 572 pc_gamg->ops->createdefaultdata = PCGAMGSetData_Classical; 573 PetscFunctionReturn(0); 574 } 575