1 #define PETSCKSP_DLL 2 3 /* 4 Defines the multigrid preconditioner interface. 5 */ 6 #include "src/ksp/pc/impls/mg/mgimpl.h" /*I "petscmg.h" I*/ 7 8 9 /* 10 MGMCycle_Private - Given an MG structure created with MGCreate() runs 11 one multiplicative cycle down through the levels and 12 back up. 13 14 Input Parameter: 15 . mg - structure created with MGCreate(). 16 */ 17 #undef __FUNCT__ 18 #define __FUNCT__ "MGMCycle_Private" 19 PetscErrorCode MGMCycle_Private(MG *mglevels,PetscTruth *converged) 20 { 21 MG mg = *mglevels,mgc; 22 PetscErrorCode ierr; 23 PetscInt cycles = mg->cycles; 24 PetscScalar zero = 0.0; 25 26 PetscFunctionBegin; 27 if (converged) *converged = PETSC_FALSE; 28 29 if (mg->eventsolve) {ierr = PetscLogEventBegin(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);} 30 ierr = KSPSolve(mg->smoothd,mg->b,mg->x);CHKERRQ(ierr); 31 if (mg->eventsolve) {ierr = PetscLogEventEnd(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);} 32 if (mg->level) { /* not the coarsest grid */ 33 ierr = (*mg->residual)(mg->A,mg->b,mg->x,mg->r);CHKERRQ(ierr); 34 35 /* if on finest level and have convergence criteria set */ 36 if (mg->level == mg->levels-1 && mg->ttol) { 37 PetscReal rnorm; 38 ierr = VecNorm(mg->r,NORM_2,&rnorm);CHKERRQ(ierr); 39 if (rnorm <= mg->ttol) { 40 *converged = PETSC_TRUE; 41 if (rnorm < mg->abstol) { 42 ierr = PetscLogInfo((0,"MGMCycle_Private:Linear solver has converged. Residual norm %g is less than absolute tolerance %g\n",rnorm,mg->abstol));CHKERRQ(ierr); 43 } else { 44 ierr = PetscLogInfo((0,"MGMCycle_Private:Linear solver has converged. Residual norm %g is less than relative tolerance times initial residual norm %g\n",rnorm,mg->ttol));CHKERRQ(ierr); 45 } 46 PetscFunctionReturn(0); 47 } 48 } 49 50 mgc = *(mglevels - 1); 51 ierr = MatRestrict(mg->restrct,mg->r,mgc->b);CHKERRQ(ierr); 52 ierr = VecSet(&zero,mgc->x);CHKERRQ(ierr); 53 while (cycles--) { 54 ierr = MGMCycle_Private(mglevels-1,converged);CHKERRQ(ierr); 55 } 56 ierr = MatInterpolateAdd(mg->interpolate,mgc->x,mg->x,mg->x);CHKERRQ(ierr); 57 if (mg->eventsolve) {ierr = PetscLogEventBegin(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);} 58 ierr = KSPSolve(mg->smoothu,mg->b,mg->x);CHKERRQ(ierr); 59 if (mg->eventsolve) {ierr = PetscLogEventEnd(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);} 60 } 61 PetscFunctionReturn(0); 62 } 63 64 /* 65 MGCreate_Private - Creates a MG structure for use with the 66 multigrid code. Level 0 is the coarsest. (But the 67 finest level is stored first in the array). 68 69 */ 70 #undef __FUNCT__ 71 #define __FUNCT__ "MGCreate_Private" 72 static PetscErrorCode MGCreate_Private(MPI_Comm comm,PetscInt levels,PC pc,MPI_Comm *comms,MG **result) 73 { 74 MG *mg; 75 PetscErrorCode ierr; 76 PetscInt i; 77 PetscMPIInt size; 78 char *prefix; 79 PC ipc; 80 81 PetscFunctionBegin; 82 ierr = PetscMalloc(levels*sizeof(MG),&mg);CHKERRQ(ierr); 83 ierr = PetscLogObjectMemory(pc,levels*(sizeof(MG)+sizeof(struct _MG)));CHKERRQ(ierr); 84 85 ierr = PCGetOptionsPrefix(pc,&prefix);CHKERRQ(ierr); 86 87 for (i=0; i<levels; i++) { 88 ierr = PetscNew(struct _MG,&mg[i]);CHKERRQ(ierr); 89 mg[i]->level = i; 90 mg[i]->levels = levels; 91 mg[i]->cycles = 1; 92 mg[i]->galerkin = PETSC_FALSE; 93 mg[i]->galerkinused = PETSC_FALSE; 94 mg[i]->default_smoothu = 1; 95 mg[i]->default_smoothd = 1; 96 97 if (comms) comm = comms[i]; 98 ierr = KSPCreate(comm,&mg[i]->smoothd);CHKERRQ(ierr); 99 ierr = KSPSetTolerances(mg[i]->smoothd,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT, mg[i]->default_smoothd);CHKERRQ(ierr); 100 ierr = KSPSetOptionsPrefix(mg[i]->smoothd,prefix);CHKERRQ(ierr); 101 102 /* do special stuff for coarse grid */ 103 if (!i && levels > 1) { 104 ierr = KSPAppendOptionsPrefix(mg[0]->smoothd,"mg_coarse_");CHKERRQ(ierr); 105 106 /* coarse solve is (redundant) LU by default */ 107 ierr = KSPSetType(mg[0]->smoothd,KSPPREONLY);CHKERRQ(ierr); 108 ierr = KSPGetPC(mg[0]->smoothd,&ipc);CHKERRQ(ierr); 109 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 110 if (size > 1) { 111 ierr = PCSetType(ipc,PCREDUNDANT);CHKERRQ(ierr); 112 ierr = PCRedundantGetPC(ipc,&ipc);CHKERRQ(ierr); 113 } 114 ierr = PCSetType(ipc,PCLU);CHKERRQ(ierr); 115 116 } else { 117 char tprefix[128]; 118 sprintf(tprefix,"mg_levels_%d_",(int)i); 119 ierr = KSPAppendOptionsPrefix(mg[i]->smoothd,tprefix);CHKERRQ(ierr); 120 } 121 ierr = PetscLogObjectParent(pc,mg[i]->smoothd);CHKERRQ(ierr); 122 mg[i]->smoothu = mg[i]->smoothd; 123 mg[i]->rtol = 0.0; 124 mg[i]->abstol = 0.0; 125 mg[i]->dtol = 0.0; 126 mg[i]->ttol = 0.0; 127 mg[i]->eventsetup = 0; 128 mg[i]->eventsolve = 0; 129 } 130 *result = mg; 131 PetscFunctionReturn(0); 132 } 133 134 #undef __FUNCT__ 135 #define __FUNCT__ "PCDestroy_MG" 136 static PetscErrorCode PCDestroy_MG(PC pc) 137 { 138 MG *mg = (MG*)pc->data; 139 PetscErrorCode ierr; 140 PetscInt i,n = mg[0]->levels; 141 142 PetscFunctionBegin; 143 if (mg[0]->galerkinused) { 144 Mat B; 145 for (i=0; i<n-1; i++) { 146 ierr = KSPGetOperators(mg[i]->smoothd,0,&B,0);CHKERRQ(ierr); 147 ierr = MatDestroy(B);CHKERRQ(ierr); 148 } 149 } 150 151 for (i=0; i<n-1; i++) { 152 if (mg[i+1]->r) {ierr = VecDestroy(mg[i+1]->r);CHKERRQ(ierr);} 153 if (mg[i]->b) {ierr = VecDestroy(mg[i]->b);CHKERRQ(ierr);} 154 if (mg[i]->x) {ierr = VecDestroy(mg[i]->x);CHKERRQ(ierr);} 155 if (mg[i+1]->restrct) {ierr = MatDestroy(mg[i+1]->restrct);CHKERRQ(ierr);} 156 if (mg[i+1]->interpolate) {ierr = MatDestroy(mg[i+1]->interpolate);CHKERRQ(ierr);} 157 } 158 159 for (i=0; i<n; i++) { 160 if (mg[i]->smoothd != mg[i]->smoothu) { 161 ierr = KSPDestroy(mg[i]->smoothd);CHKERRQ(ierr); 162 } 163 ierr = KSPDestroy(mg[i]->smoothu);CHKERRQ(ierr); 164 ierr = PetscFree(mg[i]);CHKERRQ(ierr); 165 } 166 ierr = PetscFree(mg);CHKERRQ(ierr); 167 PetscFunctionReturn(0); 168 } 169 170 171 172 EXTERN PetscErrorCode MGACycle_Private(MG*); 173 EXTERN PetscErrorCode MGFCycle_Private(MG*); 174 EXTERN PetscErrorCode MGKCycle_Private(MG*); 175 176 /* 177 PCApply_MG - Runs either an additive, multiplicative, Kaskadic 178 or full cycle of multigrid. 179 180 Note: 181 A simple wrapper which calls MGMCycle(),MGACycle(), or MGFCycle(). 182 */ 183 #undef __FUNCT__ 184 #define __FUNCT__ "PCApply_MG" 185 static PetscErrorCode PCApply_MG(PC pc,Vec b,Vec x) 186 { 187 MG *mg = (MG*)pc->data; 188 PetscScalar zero = 0.0; 189 PetscErrorCode ierr; 190 PetscInt levels = mg[0]->levels; 191 192 PetscFunctionBegin; 193 mg[levels-1]->b = b; 194 mg[levels-1]->x = x; 195 if (!mg[levels-1]->r && mg[0]->am == MGADDITIVE) { 196 Vec tvec; 197 ierr = VecDuplicate(mg[levels-1]->b,&tvec);CHKERRQ(ierr); 198 ierr = MGSetR(pc,levels-1,tvec);CHKERRQ(ierr); 199 ierr = VecDestroy(tvec);CHKERRQ(ierr); 200 } 201 if (mg[0]->am == MGMULTIPLICATIVE) { 202 ierr = VecSet(&zero,x);CHKERRQ(ierr); 203 ierr = MGMCycle_Private(mg+levels-1,PETSC_NULL);CHKERRQ(ierr); 204 } 205 else if (mg[0]->am == MGADDITIVE) { 206 ierr = MGACycle_Private(mg);CHKERRQ(ierr); 207 } 208 else if (mg[0]->am == MGKASKADE) { 209 ierr = MGKCycle_Private(mg);CHKERRQ(ierr); 210 } 211 else { 212 ierr = MGFCycle_Private(mg);CHKERRQ(ierr); 213 } 214 PetscFunctionReturn(0); 215 } 216 217 #undef __FUNCT__ 218 #define __FUNCT__ "PCApplyRichardson_MG" 219 static PetscErrorCode PCApplyRichardson_MG(PC pc,Vec b,Vec x,Vec w,PetscReal rtol,PetscReal abstol, PetscReal dtol,PetscInt its) 220 { 221 MG *mg = (MG*)pc->data; 222 PetscErrorCode ierr; 223 PetscInt levels = mg[0]->levels; 224 PetscTruth converged = PETSC_FALSE; 225 226 PetscFunctionBegin; 227 mg[levels-1]->b = b; 228 mg[levels-1]->x = x; 229 230 mg[levels-1]->rtol = rtol; 231 mg[levels-1]->abstol = abstol; 232 mg[levels-1]->dtol = dtol; 233 if (rtol) { 234 /* compute initial residual norm for relative convergence test */ 235 PetscReal rnorm; 236 ierr = (*mg[levels-1]->residual)(mg[levels-1]->A,b,x,w);CHKERRQ(ierr); 237 ierr = VecNorm(w,NORM_2,&rnorm);CHKERRQ(ierr); 238 mg[levels-1]->ttol = PetscMax(rtol*rnorm,abstol); 239 } else if (abstol) { 240 mg[levels-1]->ttol = abstol; 241 } else { 242 mg[levels-1]->ttol = 0.0; 243 } 244 245 while (its-- && !converged) { 246 ierr = MGMCycle_Private(mg+levels-1,&converged);CHKERRQ(ierr); 247 } 248 PetscFunctionReturn(0); 249 } 250 251 #undef __FUNCT__ 252 #define __FUNCT__ "PCSetFromOptions_MG" 253 PetscErrorCode PCSetFromOptions_MG(PC pc) 254 { 255 PetscErrorCode ierr; 256 PetscInt indx,m,levels = 1; 257 PetscTruth flg; 258 const char *type[] = {"additive","multiplicative","full","cascade","kascade"}; 259 260 PetscFunctionBegin; 261 262 ierr = PetscOptionsHead("Multigrid options");CHKERRQ(ierr); 263 if (!pc->data) { 264 ierr = PetscOptionsInt("-pc_mg_levels","Number of Levels","MGSetLevels",levels,&levels,&flg);CHKERRQ(ierr); 265 ierr = MGSetLevels(pc,levels,PETSC_NULL);CHKERRQ(ierr); 266 } 267 ierr = PetscOptionsInt("-pc_mg_cycles","1 for V cycle, 2 for W-cycle","MGSetCycles",1,&m,&flg);CHKERRQ(ierr); 268 if (flg) { 269 ierr = MGSetCycles(pc,m);CHKERRQ(ierr); 270 } 271 ierr = PetscOptionsName("-pc_mg_galerkin","Use Galerkin process to compute coarser operators","MGSetGalerkin",&flg);CHKERRQ(ierr); 272 if (flg) { 273 ierr = MGSetGalerkin(pc);CHKERRQ(ierr); 274 } 275 ierr = PetscOptionsInt("-pc_mg_smoothup","Number of post-smoothing steps","MGSetNumberSmoothUp",1,&m,&flg);CHKERRQ(ierr); 276 if (flg) { 277 ierr = MGSetNumberSmoothUp(pc,m);CHKERRQ(ierr); 278 } 279 ierr = PetscOptionsInt("-pc_mg_smoothdown","Number of pre-smoothing steps","MGSetNumberSmoothDown",1,&m,&flg);CHKERRQ(ierr); 280 if (flg) { 281 ierr = MGSetNumberSmoothDown(pc,m);CHKERRQ(ierr); 282 } 283 ierr = PetscOptionsEList("-pc_mg_type","Multigrid type","MGSetType",type,5,type[1],&indx,&flg);CHKERRQ(ierr); 284 if (flg) { 285 MGType mg = (MGType) 0; 286 switch (indx) { 287 case 0: 288 mg = MGADDITIVE; 289 break; 290 case 1: 291 mg = MGMULTIPLICATIVE; 292 break; 293 case 2: 294 mg = MGFULL; 295 break; 296 case 3: 297 mg = MGKASKADE; 298 break; 299 case 4: 300 mg = MGKASKADE; 301 break; 302 } 303 ierr = MGSetType(pc,mg);CHKERRQ(ierr); 304 } 305 ierr = PetscOptionsName("-pc_mg_log","Log times for each multigrid level","None",&flg);CHKERRQ(ierr); 306 if (flg) { 307 MG *mg = (MG*)pc->data; 308 PetscInt i; 309 char eventname[128]; 310 if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling"); 311 levels = mg[0]->levels; 312 for (i=0; i<levels; i++) { 313 sprintf(eventname,"MSetup Level %d",(int)i); 314 ierr = PetscLogEventRegister(&mg[i]->eventsetup,eventname,pc->cookie);CHKERRQ(ierr); 315 sprintf(eventname,"MGSolve Level %d to 0",(int)i); 316 ierr = PetscLogEventRegister(&mg[i]->eventsolve,eventname,pc->cookie);CHKERRQ(ierr); 317 } 318 } 319 ierr = PetscOptionsTail();CHKERRQ(ierr); 320 PetscFunctionReturn(0); 321 } 322 323 #undef __FUNCT__ 324 #define __FUNCT__ "PCView_MG" 325 static PetscErrorCode PCView_MG(PC pc,PetscViewer viewer) 326 { 327 MG *mg = (MG*)pc->data; 328 PetscErrorCode ierr; 329 PetscInt levels = mg[0]->levels,i; 330 const char *cstring; 331 PetscTruth iascii; 332 333 PetscFunctionBegin; 334 ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&iascii);CHKERRQ(ierr); 335 if (iascii) { 336 if (mg[0]->am == MGMULTIPLICATIVE) cstring = "multiplicative"; 337 else if (mg[0]->am == MGADDITIVE) cstring = "additive"; 338 else if (mg[0]->am == MGFULL) cstring = "full"; 339 else if (mg[0]->am == MGKASKADE) cstring = "Kaskade"; 340 else cstring = "unknown"; 341 ierr = PetscViewerASCIIPrintf(viewer," MG: type is %s, levels=%D cycles=%D, pre-smooths=%D, post-smooths=%D\n", 342 cstring,levels,mg[0]->cycles,mg[0]->default_smoothd,mg[0]->default_smoothu);CHKERRQ(ierr); 343 if (mg[0]->galerkin) { 344 ierr = PetscViewerASCIIPrintf(viewer," Using Galerkin computed coarse grid matrices\n");CHKERRQ(ierr); 345 } 346 for (i=0; i<levels; i++) { 347 if (!i) { 348 ierr = PetscViewerASCIIPrintf(viewer,"Coarse gride solver -- level %D -------------------------------\n",i);CHKERRQ(ierr); 349 } else { 350 ierr = PetscViewerASCIIPrintf(viewer,"Down solver (pre-smoother) on level %D -------------------------------\n",i);CHKERRQ(ierr); 351 } 352 ierr = PetscViewerASCIIPushTab(viewer);CHKERRQ(ierr); 353 ierr = KSPView(mg[i]->smoothd,viewer);CHKERRQ(ierr); 354 ierr = PetscViewerASCIIPopTab(viewer);CHKERRQ(ierr); 355 if (i && mg[i]->smoothd == mg[i]->smoothu) { 356 ierr = PetscViewerASCIIPrintf(viewer,"Up solver (post-smoother) same as down solver (pre-smoother)\n");CHKERRQ(ierr); 357 } else if (i){ 358 ierr = PetscViewerASCIIPrintf(viewer,"Up solver (post-smoother) on level %D -------------------------------\n",i);CHKERRQ(ierr); 359 ierr = PetscViewerASCIIPushTab(viewer);CHKERRQ(ierr); 360 ierr = KSPView(mg[i]->smoothu,viewer);CHKERRQ(ierr); 361 ierr = PetscViewerASCIIPopTab(viewer);CHKERRQ(ierr); 362 } 363 } 364 } else { 365 SETERRQ1(PETSC_ERR_SUP,"Viewer type %s not supported for PCMG",((PetscObject)viewer)->type_name); 366 } 367 PetscFunctionReturn(0); 368 } 369 370 /* 371 Calls setup for the KSP on each level 372 */ 373 #undef __FUNCT__ 374 #define __FUNCT__ "PCSetUp_MG" 375 static PetscErrorCode PCSetUp_MG(PC pc) 376 { 377 MG *mg = (MG*)pc->data; 378 PetscErrorCode ierr; 379 PetscInt i,n = mg[0]->levels; 380 PC cpc; 381 PetscTruth preonly,lu,redundant,monitor = PETSC_FALSE,dump; 382 PetscViewer ascii; 383 MPI_Comm comm; 384 Mat dA,dB; 385 MatStructure uflag; 386 Vec tvec; 387 388 PetscFunctionBegin; 389 if (!pc->setupcalled) { 390 ierr = PetscOptionsHasName(0,"-pc_mg_monitor",&monitor);CHKERRQ(ierr); 391 392 for (i=0; i<n; i++) { 393 if (monitor) { 394 ierr = PetscObjectGetComm((PetscObject)mg[i]->smoothd,&comm);CHKERRQ(ierr); 395 ierr = PetscViewerASCIIOpen(comm,"stdout",&ascii);CHKERRQ(ierr); 396 ierr = PetscViewerASCIISetTab(ascii,n-i);CHKERRQ(ierr); 397 ierr = KSPSetMonitor(mg[i]->smoothd,KSPDefaultMonitor,ascii,(PetscErrorCode(*)(void*))PetscViewerDestroy);CHKERRQ(ierr); 398 } 399 ierr = KSPSetFromOptions(mg[i]->smoothd);CHKERRQ(ierr); 400 } 401 for (i=1; i<n; i++) { 402 if (mg[i]->smoothu && mg[i]->smoothu != mg[i]->smoothd) { 403 if (monitor) { 404 ierr = PetscObjectGetComm((PetscObject)mg[i]->smoothu,&comm);CHKERRQ(ierr); 405 ierr = PetscViewerASCIIOpen(comm,"stdout",&ascii);CHKERRQ(ierr); 406 ierr = PetscViewerASCIISetTab(ascii,n-i);CHKERRQ(ierr); 407 ierr = KSPSetMonitor(mg[i]->smoothu,KSPDefaultMonitor,ascii,(PetscErrorCode(*)(void*))PetscViewerDestroy);CHKERRQ(ierr); 408 } 409 ierr = KSPSetFromOptions(mg[i]->smoothu);CHKERRQ(ierr); 410 } 411 } 412 for (i=1; i<n; i++) { 413 if (mg[i]->restrct && !mg[i]->interpolate) { 414 ierr = MGSetInterpolate(pc,i,mg[i]->restrct);CHKERRQ(ierr); 415 } 416 if (!mg[i]->restrct && mg[i]->interpolate) { 417 ierr = MGSetRestriction(pc,i,mg[i]->interpolate);CHKERRQ(ierr); 418 } 419 #if defined(PETSC_USE_DEBUG) 420 if (!mg[i]->restrct || !mg[i]->interpolate) { 421 SETERRQ1(PETSC_ERR_ARG_WRONGSTATE,"Need to set restriction or interpolation on level %d",(int)i); 422 } 423 #endif 424 } 425 for (i=0; i<n-1; i++) { 426 if (!mg[i]->r && i) { 427 ierr = VecDuplicate(mg[i]->b,&tvec);CHKERRQ(ierr); 428 ierr = MGSetR(pc,i,tvec);CHKERRQ(ierr); 429 ierr = VecDestroy(tvec);CHKERRQ(ierr); 430 } 431 if (!mg[i]->x) { 432 ierr = VecDuplicate(mg[i]->b,&tvec);CHKERRQ(ierr); 433 ierr = MGSetX(pc,i,tvec);CHKERRQ(ierr); 434 ierr = VecDestroy(tvec);CHKERRQ(ierr); 435 } 436 } 437 } 438 439 /* If user did not provide fine grid operators, use those from PC */ 440 /* BUG BUG BUG This will work ONLY the first time called: hence if the user changes 441 the PC matrices between solves PCMG will continue to use first set provided */ 442 ierr = KSPGetOperators(mg[n-1]->smoothd,&dA,&dB,&uflag);CHKERRQ(ierr); 443 if (!dA && !dB) { 444 ierr = PetscLogInfo((pc,"PCSetUp_MG: Using outer operators to define finest grid operator \n because MGGetSmoother(pc,nlevels-1,&ksp);KSPSetOperators(ksp,...); was not called.\n")); 445 ierr = KSPSetOperators(mg[n-1]->smoothd,pc->mat,pc->pmat,uflag);CHKERRQ(ierr); 446 } 447 448 if (mg[0]->galerkin) { 449 Mat B; 450 mg[0]->galerkinused = PETSC_TRUE; 451 /* currently only handle case where mat and pmat are the same on coarser levels */ 452 ierr = KSPGetOperators(mg[n-1]->smoothd,&dA,&dB,&uflag);CHKERRQ(ierr); 453 if (!pc->setupcalled) { 454 for (i=n-2; i>-1; i--) { 455 ierr = MatPtAP(dB,mg[i+1]->interpolate,MAT_INITIAL_MATRIX,1.0,&B);CHKERRQ(ierr); 456 ierr = KSPSetOperators(mg[i]->smoothd,B,B,uflag);CHKERRQ(ierr); 457 dB = B; 458 } 459 } else { 460 for (i=n-2; i>-1; i--) { 461 ierr = KSPGetOperators(mg[i]->smoothd,0,&B,0);CHKERRQ(ierr); 462 ierr = MatPtAP(dB,mg[i]->interpolate,MAT_REUSE_MATRIX,1.0,&B);CHKERRQ(ierr); 463 ierr = KSPSetOperators(mg[i]->smoothd,B,B,uflag);CHKERRQ(ierr); 464 dB = B; 465 } 466 } 467 } 468 469 for (i=1; i<n; i++) { 470 if (mg[i]->smoothu == mg[i]->smoothd) { 471 /* if doing only down then initial guess is zero */ 472 ierr = KSPSetInitialGuessNonzero(mg[i]->smoothd,PETSC_TRUE);CHKERRQ(ierr); 473 } 474 if (mg[i]->eventsetup) {ierr = PetscLogEventBegin(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);} 475 ierr = KSPSetUp(mg[i]->smoothd);CHKERRQ(ierr); 476 if (mg[i]->eventsetup) {ierr = PetscLogEventEnd(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);} 477 } 478 for (i=1; i<n; i++) { 479 if (mg[i]->smoothu && mg[i]->smoothu != mg[i]->smoothd) { 480 PC uppc,downpc; 481 Mat downmat,downpmat,upmat,uppmat; 482 MatStructure matflag; 483 484 /* check if operators have been set for up, if not use down operators to set them */ 485 ierr = KSPGetPC(mg[i]->smoothu,&uppc);CHKERRQ(ierr); 486 ierr = PCGetOperators(uppc,&upmat,&uppmat,PETSC_NULL);CHKERRQ(ierr); 487 if (!upmat) { 488 ierr = KSPGetPC(mg[i]->smoothd,&downpc);CHKERRQ(ierr); 489 ierr = PCGetOperators(downpc,&downmat,&downpmat,&matflag);CHKERRQ(ierr); 490 ierr = KSPSetOperators(mg[i]->smoothu,downmat,downpmat,matflag);CHKERRQ(ierr); 491 } 492 493 ierr = KSPSetInitialGuessNonzero(mg[i]->smoothu,PETSC_TRUE);CHKERRQ(ierr); 494 if (mg[i]->eventsetup) {ierr = PetscLogEventBegin(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);} 495 ierr = KSPSetUp(mg[i]->smoothu);CHKERRQ(ierr); 496 if (mg[i]->eventsetup) {ierr = PetscLogEventEnd(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);} 497 } 498 } 499 500 /* 501 If coarse solver is not direct method then DO NOT USE preonly 502 */ 503 ierr = PetscTypeCompare((PetscObject)mg[0]->smoothd,KSPPREONLY,&preonly);CHKERRQ(ierr); 504 if (preonly) { 505 ierr = KSPGetPC(mg[0]->smoothd,&cpc);CHKERRQ(ierr); 506 ierr = PetscTypeCompare((PetscObject)cpc,PCLU,&lu);CHKERRQ(ierr); 507 ierr = PetscTypeCompare((PetscObject)cpc,PCREDUNDANT,&redundant);CHKERRQ(ierr); 508 if (!lu && !redundant) { 509 ierr = KSPSetType(mg[0]->smoothd,KSPGMRES);CHKERRQ(ierr); 510 } 511 } 512 513 if (!pc->setupcalled) { 514 if (monitor) { 515 ierr = PetscObjectGetComm((PetscObject)mg[0]->smoothd,&comm);CHKERRQ(ierr); 516 ierr = PetscViewerASCIIOpen(comm,"stdout",&ascii);CHKERRQ(ierr); 517 ierr = PetscViewerASCIISetTab(ascii,n);CHKERRQ(ierr); 518 ierr = KSPSetMonitor(mg[0]->smoothd,KSPDefaultMonitor,ascii,(PetscErrorCode(*)(void*))PetscViewerDestroy);CHKERRQ(ierr); 519 } 520 ierr = KSPSetFromOptions(mg[0]->smoothd);CHKERRQ(ierr); 521 } 522 523 if (mg[0]->eventsetup) {ierr = PetscLogEventBegin(mg[0]->eventsetup,0,0,0,0);CHKERRQ(ierr);} 524 ierr = KSPSetUp(mg[0]->smoothd);CHKERRQ(ierr); 525 if (mg[0]->eventsetup) {ierr = PetscLogEventEnd(mg[0]->eventsetup,0,0,0,0);CHKERRQ(ierr);} 526 527 /* 528 Dump the interpolation/restriction matrices to matlab plus the 529 Jacobian/stiffness on each level. This allows Matlab users to 530 easily check if the Galerkin condition A_c = R A_f R^T is satisfied */ 531 ierr = PetscOptionsHasName(pc->prefix,"-pc_mg_dump_matlab",&dump);CHKERRQ(ierr); 532 if (dump) { 533 for (i=1; i<n; i++) { 534 ierr = MatView(mg[i]->restrct,PETSC_VIEWER_SOCKET_(pc->comm));CHKERRQ(ierr); 535 } 536 for (i=0; i<n; i++) { 537 ierr = KSPGetPC(mg[i]->smoothd,&pc);CHKERRQ(ierr); 538 ierr = MatView(pc->mat,PETSC_VIEWER_SOCKET_(pc->comm));CHKERRQ(ierr); 539 } 540 } 541 ierr = PetscOptionsHasName(pc->prefix,"-pc_mg_dump_binary",&dump);CHKERRQ(ierr); 542 if (dump) { 543 for (i=1; i<n; i++) { 544 ierr = MatView(mg[i]->restrct,PETSC_VIEWER_BINARY_(pc->comm));CHKERRQ(ierr); 545 } 546 for (i=0; i<n; i++) { 547 ierr = KSPGetPC(mg[i]->smoothd,&pc);CHKERRQ(ierr); 548 ierr = MatView(pc->mat,PETSC_VIEWER_BINARY_(pc->comm));CHKERRQ(ierr); 549 } 550 } 551 PetscFunctionReturn(0); 552 } 553 554 /* -------------------------------------------------------------------------------------*/ 555 556 #undef __FUNCT__ 557 #define __FUNCT__ "MGSetLevels" 558 /*@C 559 MGSetLevels - Sets the number of levels to use with MG. 560 Must be called before any other MG routine. 561 562 Collective on PC 563 564 Input Parameters: 565 + pc - the preconditioner context 566 . levels - the number of levels 567 - comms - optional communicators for each level; this is to allow solving the coarser problems 568 on smaller sets of processors. Use PETSC_NULL_OBJECT for default in Fortran 569 570 Level: intermediate 571 572 Notes: 573 If the number of levels is one then the multigrid uses the -mg_levels prefix 574 for setting the level options rather than the -mg_coarse prefix. 575 576 .keywords: MG, set, levels, multigrid 577 578 .seealso: MGSetType(), MGGetLevels() 579 @*/ 580 PetscErrorCode PETSCKSP_DLLEXPORT MGSetLevels(PC pc,PetscInt levels,MPI_Comm *comms) 581 { 582 PetscErrorCode ierr; 583 MG *mg; 584 585 PetscFunctionBegin; 586 PetscValidHeaderSpecific(pc,PC_COOKIE,1); 587 588 if (pc->data) { 589 SETERRQ(PETSC_ERR_ORDER,"Number levels already set for MG\n\ 590 make sure that you call MGSetLevels() before KSPSetFromOptions()"); 591 } 592 ierr = MGCreate_Private(pc->comm,levels,pc,comms,&mg);CHKERRQ(ierr); 593 mg[0]->am = MGMULTIPLICATIVE; 594 pc->data = (void*)mg; 595 pc->ops->applyrichardson = PCApplyRichardson_MG; 596 PetscFunctionReturn(0); 597 } 598 599 #undef __FUNCT__ 600 #define __FUNCT__ "MGGetLevels" 601 /*@ 602 MGGetLevels - Gets the number of levels to use with MG. 603 604 Not Collective 605 606 Input Parameter: 607 . pc - the preconditioner context 608 609 Output parameter: 610 . levels - the number of levels 611 612 Level: advanced 613 614 .keywords: MG, get, levels, multigrid 615 616 .seealso: MGSetLevels() 617 @*/ 618 PetscErrorCode PETSCKSP_DLLEXPORT MGGetLevels(PC pc,PetscInt *levels) 619 { 620 MG *mg; 621 622 PetscFunctionBegin; 623 PetscValidHeaderSpecific(pc,PC_COOKIE,1); 624 PetscValidIntPointer(levels,2); 625 626 mg = (MG*)pc->data; 627 *levels = mg[0]->levels; 628 PetscFunctionReturn(0); 629 } 630 631 #undef __FUNCT__ 632 #define __FUNCT__ "MGSetType" 633 /*@ 634 MGSetType - Determines the form of multigrid to use: 635 multiplicative, additive, full, or the Kaskade algorithm. 636 637 Collective on PC 638 639 Input Parameters: 640 + pc - the preconditioner context 641 - form - multigrid form, one of MGMULTIPLICATIVE, MGADDITIVE, 642 MGFULL, MGKASKADE 643 644 Options Database Key: 645 . -pc_mg_type <form> - Sets <form>, one of multiplicative, 646 additive, full, kaskade 647 648 Level: advanced 649 650 .keywords: MG, set, method, multiplicative, additive, full, Kaskade, multigrid 651 652 .seealso: MGSetLevels() 653 @*/ 654 PetscErrorCode PETSCKSP_DLLEXPORT MGSetType(PC pc,MGType form) 655 { 656 MG *mg; 657 658 PetscFunctionBegin; 659 PetscValidHeaderSpecific(pc,PC_COOKIE,1); 660 mg = (MG*)pc->data; 661 662 if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling"); 663 mg[0]->am = form; 664 if (form == MGMULTIPLICATIVE) pc->ops->applyrichardson = PCApplyRichardson_MG; 665 else pc->ops->applyrichardson = 0; 666 PetscFunctionReturn(0); 667 } 668 669 #undef __FUNCT__ 670 #define __FUNCT__ "MGSetCycles" 671 /*@ 672 MGSetCycles - Sets the type cycles to use. Use MGSetCyclesOnLevel() for more 673 complicated cycling. 674 675 Collective on PC 676 677 Input Parameters: 678 + pc - the multigrid context 679 - n - the number of cycles 680 681 Options Database Key: 682 $ -pc_mg_cycles n - 1 denotes a V-cycle; 2 denotes a W-cycle. 683 684 Level: advanced 685 686 .keywords: MG, set, cycles, V-cycle, W-cycle, multigrid 687 688 .seealso: MGSetCyclesOnLevel() 689 @*/ 690 PetscErrorCode PETSCKSP_DLLEXPORT MGSetCycles(PC pc,PetscInt n) 691 { 692 MG *mg; 693 PetscInt i,levels; 694 695 PetscFunctionBegin; 696 PetscValidHeaderSpecific(pc,PC_COOKIE,1); 697 mg = (MG*)pc->data; 698 if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling"); 699 levels = mg[0]->levels; 700 701 for (i=0; i<levels; i++) { 702 mg[i]->cycles = n; 703 } 704 PetscFunctionReturn(0); 705 } 706 707 #undef __FUNCT__ 708 #define __FUNCT__ "MGSetGalerkin" 709 /*@ 710 MGSetGalerkin - Causes the coarser grid matrices to be computed from the 711 finest grid via the Galerkin process: A_i-1 = r_i * A_i * r_i^t 712 713 Collective on PC 714 715 Input Parameters: 716 + pc - the multigrid context 717 - n - the number of cycles 718 719 Options Database Key: 720 $ -pc_mg_galerkin 721 722 Level: intermediate 723 724 .keywords: MG, set, Galerkin 725 726 @*/ 727 PetscErrorCode PETSCKSP_DLLEXPORT MGSetGalerkin(PC pc) 728 { 729 MG *mg; 730 PetscInt i,levels; 731 732 PetscFunctionBegin; 733 PetscValidHeaderSpecific(pc,PC_COOKIE,1); 734 mg = (MG*)pc->data; 735 if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling"); 736 levels = mg[0]->levels; 737 738 for (i=0; i<levels; i++) { 739 mg[i]->galerkin = PETSC_TRUE; 740 } 741 PetscFunctionReturn(0); 742 } 743 744 #undef __FUNCT__ 745 #define __FUNCT__ "MGCheck" 746 /*@ 747 MGCheck - Checks that all components of the MG structure have 748 been set. 749 750 Collective on PC 751 752 Input Parameters: 753 . mg - the MG structure 754 755 Level: advanced 756 757 .keywords: MG, check, set, multigrid 758 @*/ 759 PetscErrorCode PETSCKSP_DLLEXPORT MGCheck(PC pc) 760 { 761 MG *mg; 762 PetscInt i,n,count = 0; 763 764 PetscFunctionBegin; 765 PetscValidHeaderSpecific(pc,PC_COOKIE,1); 766 mg = (MG*)pc->data; 767 768 if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling"); 769 770 n = mg[0]->levels; 771 772 for (i=1; i<n; i++) { 773 if (!mg[i]->restrct) { 774 (*PetscErrorPrintf)("No restrict set level %D \n",n-i); count++; 775 } 776 if (!mg[i]->interpolate) { 777 (*PetscErrorPrintf)("No interpolate set level %D \n",n-i); count++; 778 } 779 if (!mg[i]->r) { 780 (*PetscErrorPrintf)("No r set level %D \n",n-i); count++; 781 } 782 if (!mg[i-1]->x) { 783 (*PetscErrorPrintf)("No x set level %D \n",n-i); count++; 784 } 785 if (!mg[i-1]->b) { 786 (*PetscErrorPrintf)("No b set level %D \n",n-i); count++; 787 } 788 } 789 PetscFunctionReturn(count); 790 } 791 792 793 #undef __FUNCT__ 794 #define __FUNCT__ "MGSetNumberSmoothDown" 795 /*@ 796 MGSetNumberSmoothDown - Sets the number of pre-smoothing steps to 797 use on all levels. Use MGGetSmootherDown() to set different 798 pre-smoothing steps on different levels. 799 800 Collective on PC 801 802 Input Parameters: 803 + mg - the multigrid context 804 - n - the number of smoothing steps 805 806 Options Database Key: 807 . -pc_mg_smoothdown <n> - Sets number of pre-smoothing steps 808 809 Level: advanced 810 811 .keywords: MG, smooth, down, pre-smoothing, steps, multigrid 812 813 .seealso: MGSetNumberSmoothUp() 814 @*/ 815 PetscErrorCode PETSCKSP_DLLEXPORT MGSetNumberSmoothDown(PC pc,PetscInt n) 816 { 817 MG *mg; 818 PetscErrorCode ierr; 819 PetscInt i,levels; 820 821 PetscFunctionBegin; 822 PetscValidHeaderSpecific(pc,PC_COOKIE,1); 823 mg = (MG*)pc->data; 824 if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling"); 825 levels = mg[0]->levels; 826 827 for (i=0; i<levels; i++) { 828 /* make sure smoother up and down are different */ 829 ierr = MGGetSmootherUp(pc,i,PETSC_NULL);CHKERRQ(ierr); 830 ierr = KSPSetTolerances(mg[i]->smoothd,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT,n);CHKERRQ(ierr); 831 mg[i]->default_smoothd = n; 832 } 833 PetscFunctionReturn(0); 834 } 835 836 #undef __FUNCT__ 837 #define __FUNCT__ "MGSetNumberSmoothUp" 838 /*@ 839 MGSetNumberSmoothUp - Sets the number of post-smoothing steps to use 840 on all levels. Use MGGetSmootherUp() to set different numbers of 841 post-smoothing steps on different levels. 842 843 Collective on PC 844 845 Input Parameters: 846 + mg - the multigrid context 847 - n - the number of smoothing steps 848 849 Options Database Key: 850 . -pc_mg_smoothup <n> - Sets number of post-smoothing steps 851 852 Level: advanced 853 854 Note: this does not set a value on the coarsest grid, since we assume that 855 there is no seperate smooth up on the coarsest grid. 856 857 .keywords: MG, smooth, up, post-smoothing, steps, multigrid 858 859 .seealso: MGSetNumberSmoothDown() 860 @*/ 861 PetscErrorCode PETSCKSP_DLLEXPORT MGSetNumberSmoothUp(PC pc,PetscInt n) 862 { 863 MG *mg; 864 PetscErrorCode ierr; 865 PetscInt i,levels; 866 867 PetscFunctionBegin; 868 PetscValidHeaderSpecific(pc,PC_COOKIE,1); 869 mg = (MG*)pc->data; 870 if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling"); 871 levels = mg[0]->levels; 872 873 for (i=1; i<levels; i++) { 874 /* make sure smoother up and down are different */ 875 ierr = MGGetSmootherUp(pc,i,PETSC_NULL);CHKERRQ(ierr); 876 ierr = KSPSetTolerances(mg[i]->smoothu,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT,n);CHKERRQ(ierr); 877 mg[i]->default_smoothu = n; 878 } 879 PetscFunctionReturn(0); 880 } 881 882 /* ----------------------------------------------------------------------------------------*/ 883 884 /*MC 885 PCMG - Use geometric multigrid preconditioning. This preconditioner requires you provide additional 886 information about the coarser grid matrices and restriction/interpolation operators. 887 888 Options Database Keys: 889 + -pc_mg_levels <nlevels> - number of levels including finest 890 . -pc_mg_cycles 1 or 2 - for V or W-cycle 891 . -pc_mg_smoothup <n> - number of smoothing steps after interpolation 892 . -pc_mg_smoothdown <n> - number of smoothing steps before applying restriction operator 893 . -pc_mg_type <additive,multiplicative,full,cascade> - multiplicative is the default 894 . -pc_mg_log - log information about time spent on each level of the solver 895 . -pc_mg_monitor - print information on the multigrid convergence 896 - -pc_mg_dump_matlab - dumps the matrices for each level and the restriction/interpolation matrices 897 to the Socket viewer for reading from Matlab. 898 899 Notes: 900 901 Level: intermediate 902 903 Concepts: multigrid 904 905 .seealso: PCCreate(), PCSetType(), PCType (for list of available types), PC, PCMGType, 906 MGSetLevels(), MGGetLevels(), MGSetType(), MPSetCycles(), MGSetNumberSmoothDown(), 907 MGSetNumberSmoothUp(), MGGetCoarseSolve(), MGSetResidual(), MGSetInterpolation(), 908 MGSetRestriction(), MGGetSmoother(), MGGetSmootherUp(), MGGetSmootherDown(), 909 MGSetCyclesOnLevel(), MGSetRhs(), MGSetX(), MGSetR() 910 M*/ 911 912 EXTERN_C_BEGIN 913 #undef __FUNCT__ 914 #define __FUNCT__ "PCCreate_MG" 915 PetscErrorCode PETSCKSP_DLLEXPORT PCCreate_MG(PC pc) 916 { 917 PetscFunctionBegin; 918 pc->ops->apply = PCApply_MG; 919 pc->ops->setup = PCSetUp_MG; 920 pc->ops->destroy = PCDestroy_MG; 921 pc->ops->setfromoptions = PCSetFromOptions_MG; 922 pc->ops->view = PCView_MG; 923 924 pc->data = (void*)0; 925 PetscFunctionReturn(0); 926 } 927 EXTERN_C_END 928