1 2 /* 3 Provides an interface to the MUMPS sparse solver 4 */ 5 #include <petscpkg_version.h> 6 #include <../src/mat/impls/aij/mpi/mpiaij.h> /*I "petscmat.h" I*/ 7 #include <../src/mat/impls/sbaij/mpi/mpisbaij.h> 8 #include <../src/mat/impls/sell/mpi/mpisell.h> 9 10 EXTERN_C_BEGIN 11 #if defined(PETSC_USE_COMPLEX) 12 #if defined(PETSC_USE_REAL_SINGLE) 13 #include <cmumps_c.h> 14 #else 15 #include <zmumps_c.h> 16 #endif 17 #else 18 #if defined(PETSC_USE_REAL_SINGLE) 19 #include <smumps_c.h> 20 #else 21 #include <dmumps_c.h> 22 #endif 23 #endif 24 EXTERN_C_END 25 #define JOB_INIT -1 26 #define JOB_FACTSYMBOLIC 1 27 #define JOB_FACTNUMERIC 2 28 #define JOB_SOLVE 3 29 #define JOB_END -2 30 31 /* calls to MUMPS */ 32 #if defined(PETSC_USE_COMPLEX) 33 #if defined(PETSC_USE_REAL_SINGLE) 34 #define MUMPS_c cmumps_c 35 #else 36 #define MUMPS_c zmumps_c 37 #endif 38 #else 39 #if defined(PETSC_USE_REAL_SINGLE) 40 #define MUMPS_c smumps_c 41 #else 42 #define MUMPS_c dmumps_c 43 #endif 44 #endif 45 46 /* MUMPS uses MUMPS_INT for nonzero indices such as irn/jcn, irn_loc/jcn_loc and uses int64_t for 47 number of nonzeros such as nnz, nnz_loc. We typedef MUMPS_INT to PetscMUMPSInt to follow the 48 naming convention in PetscMPIInt, PetscBLASInt etc. 49 */ 50 typedef MUMPS_INT PetscMUMPSInt; 51 52 #if PETSC_PKG_MUMPS_VERSION_GE(5,3,0) 53 #if defined(MUMPS_INTSIZE64) /* MUMPS_INTSIZE64 is in MUMPS headers if it is built in full 64-bit mode, therefore the macro is more reliable */ 54 #error "Petsc has not been tested with full 64-bit MUMPS and we choose to error out" 55 #endif 56 #else 57 #if defined(INTSIZE64) /* INTSIZE64 is a command line macro one used to build MUMPS in full 64-bit mode */ 58 #error "Petsc has not been tested with full 64-bit MUMPS and we choose to error out" 59 #endif 60 #endif 61 62 #define MPIU_MUMPSINT MPI_INT 63 #define PETSC_MUMPS_INT_MAX 2147483647 64 #define PETSC_MUMPS_INT_MIN -2147483648 65 66 /* Cast PetscInt to PetscMUMPSInt. Usually there is no overflow since <a> is row/col indices or some small integers*/ 67 static inline PetscErrorCode PetscMUMPSIntCast(PetscInt a,PetscMUMPSInt *b) 68 { 69 PetscFunctionBegin; 70 #if PetscDefined(USE_64BIT_INDICES) 71 PetscAssert(a <= PETSC_MUMPS_INT_MAX && a >= PETSC_MUMPS_INT_MIN,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"PetscInt too long for PetscMUMPSInt"); 72 #endif 73 *b = (PetscMUMPSInt)(a); 74 PetscFunctionReturn(0); 75 } 76 77 /* Put these utility routines here since they are only used in this file */ 78 static inline PetscErrorCode PetscOptionsMUMPSInt_Private(PetscOptionItems *PetscOptionsObject,const char opt[],const char text[],const char man[],PetscMUMPSInt currentvalue,PetscMUMPSInt *value,PetscBool *set,PetscMUMPSInt lb,PetscMUMPSInt ub) 79 { 80 PetscInt myval; 81 PetscBool myset; 82 PetscFunctionBegin; 83 /* PetscInt's size should be always >= PetscMUMPSInt's. It is safe to call PetscOptionsInt_Private to read a PetscMUMPSInt */ 84 PetscCall(PetscOptionsInt_Private(PetscOptionsObject,opt,text,man,(PetscInt)currentvalue,&myval,&myset,lb,ub)); 85 if (myset) PetscCall(PetscMUMPSIntCast(myval,value)); 86 if (set) *set = myset; 87 PetscFunctionReturn(0); 88 } 89 #define PetscOptionsMUMPSInt(a,b,c,d,e,f) PetscOptionsMUMPSInt_Private(PetscOptionsObject,a,b,c,d,e,f,PETSC_MUMPS_INT_MIN,PETSC_MUMPS_INT_MAX) 90 91 /* if using PETSc OpenMP support, we only call MUMPS on master ranks. Before/after the call, we change/restore CPUs the master ranks can run on */ 92 #if defined(PETSC_HAVE_OPENMP_SUPPORT) 93 #define PetscMUMPS_c(mumps) \ 94 do { \ 95 if (mumps->use_petsc_omp_support) { \ 96 if (mumps->is_omp_master) { \ 97 PetscCall(PetscOmpCtrlOmpRegionOnMasterBegin(mumps->omp_ctrl)); \ 98 MUMPS_c(&mumps->id); \ 99 PetscCall(PetscOmpCtrlOmpRegionOnMasterEnd(mumps->omp_ctrl)); \ 100 } \ 101 PetscCall(PetscOmpCtrlBarrier(mumps->omp_ctrl)); \ 102 /* Global info is same on all processes so we Bcast it within omp_comm. Local info is specific \ 103 to processes, so we only Bcast info[1], an error code and leave others (since they do not have \ 104 an easy translation between omp_comm and petsc_comm). See MUMPS-5.1.2 manual p82. \ 105 omp_comm is a small shared memory communicator, hence doing multiple Bcast as shown below is OK. \ 106 */ \ 107 PetscCallMPI(MPI_Bcast(mumps->id.infog, 40,MPIU_MUMPSINT, 0,mumps->omp_comm));\ 108 PetscCallMPI(MPI_Bcast(mumps->id.rinfog,20,MPIU_REAL, 0,mumps->omp_comm));\ 109 PetscCallMPI(MPI_Bcast(mumps->id.info, 1, MPIU_MUMPSINT, 0,mumps->omp_comm));\ 110 } else { \ 111 MUMPS_c(&mumps->id); \ 112 } \ 113 } while (0) 114 #else 115 #define PetscMUMPS_c(mumps) \ 116 do { MUMPS_c(&mumps->id); } while (0) 117 #endif 118 119 /* declare MumpsScalar */ 120 #if defined(PETSC_USE_COMPLEX) 121 #if defined(PETSC_USE_REAL_SINGLE) 122 #define MumpsScalar mumps_complex 123 #else 124 #define MumpsScalar mumps_double_complex 125 #endif 126 #else 127 #define MumpsScalar PetscScalar 128 #endif 129 130 /* macros s.t. indices match MUMPS documentation */ 131 #define ICNTL(I) icntl[(I)-1] 132 #define CNTL(I) cntl[(I)-1] 133 #define INFOG(I) infog[(I)-1] 134 #define INFO(I) info[(I)-1] 135 #define RINFOG(I) rinfog[(I)-1] 136 #define RINFO(I) rinfo[(I)-1] 137 138 typedef struct Mat_MUMPS Mat_MUMPS; 139 struct Mat_MUMPS { 140 #if defined(PETSC_USE_COMPLEX) 141 #if defined(PETSC_USE_REAL_SINGLE) 142 CMUMPS_STRUC_C id; 143 #else 144 ZMUMPS_STRUC_C id; 145 #endif 146 #else 147 #if defined(PETSC_USE_REAL_SINGLE) 148 SMUMPS_STRUC_C id; 149 #else 150 DMUMPS_STRUC_C id; 151 #endif 152 #endif 153 154 MatStructure matstruc; 155 PetscMPIInt myid,petsc_size; 156 PetscMUMPSInt *irn,*jcn; /* the (i,j,v) triplets passed to mumps. */ 157 PetscScalar *val,*val_alloc; /* For some matrices, we can directly access their data array without a buffer. For others, we need a buffer. So comes val_alloc. */ 158 PetscInt64 nnz; /* number of nonzeros. The type is called selective 64-bit in mumps */ 159 PetscMUMPSInt sym; 160 MPI_Comm mumps_comm; 161 PetscMUMPSInt ICNTL9_pre; /* check if ICNTL(9) is changed from previous MatSolve */ 162 VecScatter scat_rhs, scat_sol; /* used by MatSolve() */ 163 PetscMUMPSInt ICNTL20; /* use centralized (0) or distributed (10) dense RHS */ 164 PetscMUMPSInt lrhs_loc,nloc_rhs,*irhs_loc; 165 #if defined(PETSC_HAVE_OPENMP_SUPPORT) 166 PetscInt *rhs_nrow,max_nrhs; 167 PetscMPIInt *rhs_recvcounts,*rhs_disps; 168 PetscScalar *rhs_loc,*rhs_recvbuf; 169 #endif 170 Vec b_seq,x_seq; 171 PetscInt ninfo,*info; /* which INFO to display */ 172 PetscInt sizeredrhs; 173 PetscScalar *schur_sol; 174 PetscInt schur_sizesol; 175 PetscMUMPSInt *ia_alloc,*ja_alloc; /* work arrays used for the CSR struct for sparse rhs */ 176 PetscInt64 cur_ilen,cur_jlen; /* current len of ia_alloc[], ja_alloc[] */ 177 PetscErrorCode (*ConvertToTriples)(Mat,PetscInt,MatReuse,Mat_MUMPS*); 178 179 /* stuff used by petsc/mumps OpenMP support*/ 180 PetscBool use_petsc_omp_support; 181 PetscOmpCtrl omp_ctrl; /* an OpenMP controler that blocked processes will release their CPU (MPI_Barrier does not have this guarantee) */ 182 MPI_Comm petsc_comm,omp_comm; /* petsc_comm is petsc matrix's comm */ 183 PetscInt64 *recvcount; /* a collection of nnz on omp_master */ 184 PetscMPIInt tag,omp_comm_size; 185 PetscBool is_omp_master; /* is this rank the master of omp_comm */ 186 MPI_Request *reqs; 187 }; 188 189 /* Cast a 1-based CSR represented by (nrow, ia, ja) of type PetscInt to a CSR of type PetscMUMPSInt. 190 Here, nrow is number of rows, ia[] is row pointer and ja[] is column indices. 191 */ 192 static PetscErrorCode PetscMUMPSIntCSRCast(Mat_MUMPS *mumps,PetscInt nrow,PetscInt *ia,PetscInt *ja,PetscMUMPSInt **ia_mumps,PetscMUMPSInt **ja_mumps,PetscMUMPSInt *nnz_mumps) 193 { 194 PetscInt nnz=ia[nrow]-1; /* mumps uses 1-based indices. Uses PetscInt instead of PetscInt64 since mumps only uses PetscMUMPSInt for rhs */ 195 196 PetscFunctionBegin; 197 #if defined(PETSC_USE_64BIT_INDICES) 198 { 199 PetscInt i; 200 if (nrow+1 > mumps->cur_ilen) { /* realloc ia_alloc/ja_alloc to fit ia/ja */ 201 PetscCall(PetscFree(mumps->ia_alloc)); 202 PetscCall(PetscMalloc1(nrow+1,&mumps->ia_alloc)); 203 mumps->cur_ilen = nrow+1; 204 } 205 if (nnz > mumps->cur_jlen) { 206 PetscCall(PetscFree(mumps->ja_alloc)); 207 PetscCall(PetscMalloc1(nnz,&mumps->ja_alloc)); 208 mumps->cur_jlen = nnz; 209 } 210 for (i=0; i<nrow+1; i++) PetscCall(PetscMUMPSIntCast(ia[i],&(mumps->ia_alloc[i]))); 211 for (i=0; i<nnz; i++) PetscCall(PetscMUMPSIntCast(ja[i],&(mumps->ja_alloc[i]))); 212 *ia_mumps = mumps->ia_alloc; 213 *ja_mumps = mumps->ja_alloc; 214 } 215 #else 216 *ia_mumps = ia; 217 *ja_mumps = ja; 218 #endif 219 PetscCall(PetscMUMPSIntCast(nnz,nnz_mumps)); 220 PetscFunctionReturn(0); 221 } 222 223 static PetscErrorCode MatMumpsResetSchur_Private(Mat_MUMPS* mumps) 224 { 225 PetscFunctionBegin; 226 PetscCall(PetscFree(mumps->id.listvar_schur)); 227 PetscCall(PetscFree(mumps->id.redrhs)); 228 PetscCall(PetscFree(mumps->schur_sol)); 229 mumps->id.size_schur = 0; 230 mumps->id.schur_lld = 0; 231 mumps->id.ICNTL(19) = 0; 232 PetscFunctionReturn(0); 233 } 234 235 /* solve with rhs in mumps->id.redrhs and return in the same location */ 236 static PetscErrorCode MatMumpsSolveSchur_Private(Mat F) 237 { 238 Mat_MUMPS *mumps=(Mat_MUMPS*)F->data; 239 Mat S,B,X; 240 MatFactorSchurStatus schurstatus; 241 PetscInt sizesol; 242 243 PetscFunctionBegin; 244 PetscCall(MatFactorFactorizeSchurComplement(F)); 245 PetscCall(MatFactorGetSchurComplement(F,&S,&schurstatus)); 246 PetscCall(MatCreateSeqDense(PETSC_COMM_SELF,mumps->id.size_schur,mumps->id.nrhs,(PetscScalar*)mumps->id.redrhs,&B)); 247 PetscCall(MatSetType(B,((PetscObject)S)->type_name)); 248 #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 249 PetscCall(MatBindToCPU(B,S->boundtocpu)); 250 #endif 251 switch (schurstatus) { 252 case MAT_FACTOR_SCHUR_FACTORED: 253 PetscCall(MatCreateSeqDense(PETSC_COMM_SELF,mumps->id.size_schur,mumps->id.nrhs,(PetscScalar*)mumps->id.redrhs,&X)); 254 PetscCall(MatSetType(X,((PetscObject)S)->type_name)); 255 #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 256 PetscCall(MatBindToCPU(X,S->boundtocpu)); 257 #endif 258 if (!mumps->id.ICNTL(9)) { /* transpose solve */ 259 PetscCall(MatMatSolveTranspose(S,B,X)); 260 } else { 261 PetscCall(MatMatSolve(S,B,X)); 262 } 263 break; 264 case MAT_FACTOR_SCHUR_INVERTED: 265 sizesol = mumps->id.nrhs*mumps->id.size_schur; 266 if (!mumps->schur_sol || sizesol > mumps->schur_sizesol) { 267 PetscCall(PetscFree(mumps->schur_sol)); 268 PetscCall(PetscMalloc1(sizesol,&mumps->schur_sol)); 269 mumps->schur_sizesol = sizesol; 270 } 271 PetscCall(MatCreateSeqDense(PETSC_COMM_SELF,mumps->id.size_schur,mumps->id.nrhs,mumps->schur_sol,&X)); 272 PetscCall(MatSetType(X,((PetscObject)S)->type_name)); 273 #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 274 PetscCall(MatBindToCPU(X,S->boundtocpu)); 275 #endif 276 PetscCall(MatProductCreateWithMat(S,B,NULL,X)); 277 if (!mumps->id.ICNTL(9)) { /* transpose solve */ 278 PetscCall(MatProductSetType(X,MATPRODUCT_AtB)); 279 } else { 280 PetscCall(MatProductSetType(X,MATPRODUCT_AB)); 281 } 282 PetscCall(MatProductSetFromOptions(X)); 283 PetscCall(MatProductSymbolic(X)); 284 PetscCall(MatProductNumeric(X)); 285 286 PetscCall(MatCopy(X,B,SAME_NONZERO_PATTERN)); 287 break; 288 default: 289 SETERRQ(PetscObjectComm((PetscObject)F),PETSC_ERR_SUP,"Unhandled MatFactorSchurStatus %d",F->schur_status); 290 } 291 PetscCall(MatFactorRestoreSchurComplement(F,&S,schurstatus)); 292 PetscCall(MatDestroy(&B)); 293 PetscCall(MatDestroy(&X)); 294 PetscFunctionReturn(0); 295 } 296 297 static PetscErrorCode MatMumpsHandleSchur_Private(Mat F, PetscBool expansion) 298 { 299 Mat_MUMPS *mumps=(Mat_MUMPS*)F->data; 300 301 PetscFunctionBegin; 302 if (!mumps->id.ICNTL(19)) { /* do nothing when Schur complement has not been computed */ 303 PetscFunctionReturn(0); 304 } 305 if (!expansion) { /* prepare for the condensation step */ 306 PetscInt sizeredrhs = mumps->id.nrhs*mumps->id.size_schur; 307 /* allocate MUMPS internal array to store reduced right-hand sides */ 308 if (!mumps->id.redrhs || sizeredrhs > mumps->sizeredrhs) { 309 PetscCall(PetscFree(mumps->id.redrhs)); 310 mumps->id.lredrhs = mumps->id.size_schur; 311 PetscCall(PetscMalloc1(mumps->id.nrhs*mumps->id.lredrhs,&mumps->id.redrhs)); 312 mumps->sizeredrhs = mumps->id.nrhs*mumps->id.lredrhs; 313 } 314 mumps->id.ICNTL(26) = 1; /* condensation phase */ 315 } else { /* prepare for the expansion step */ 316 /* solve Schur complement (this has to be done by the MUMPS user, so basically us) */ 317 PetscCall(MatMumpsSolveSchur_Private(F)); 318 mumps->id.ICNTL(26) = 2; /* expansion phase */ 319 PetscMUMPS_c(mumps); 320 PetscCheck(mumps->id.INFOG(1) >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MUMPS in solve phase: INFOG(1)=%d",mumps->id.INFOG(1)); 321 /* restore defaults */ 322 mumps->id.ICNTL(26) = -1; 323 /* free MUMPS internal array for redrhs if we have solved for multiple rhs in order to save memory space */ 324 if (mumps->id.nrhs > 1) { 325 PetscCall(PetscFree(mumps->id.redrhs)); 326 mumps->id.lredrhs = 0; 327 mumps->sizeredrhs = 0; 328 } 329 } 330 PetscFunctionReturn(0); 331 } 332 333 /* 334 MatConvertToTriples_A_B - convert Petsc matrix to triples: row[nz], col[nz], val[nz] 335 336 input: 337 A - matrix in aij,baij or sbaij format 338 shift - 0: C style output triple; 1: Fortran style output triple. 339 reuse - MAT_INITIAL_MATRIX: spaces are allocated and values are set for the triple 340 MAT_REUSE_MATRIX: only the values in v array are updated 341 output: 342 nnz - dim of r, c, and v (number of local nonzero entries of A) 343 r, c, v - row and col index, matrix values (matrix triples) 344 345 The returned values r, c, and sometimes v are obtained in a single PetscMalloc(). Then in MatDestroy_MUMPS() it is 346 freed with PetscFree(mumps->irn); This is not ideal code, the fact that v is ONLY sometimes part of mumps->irn means 347 that the PetscMalloc() cannot easily be replaced with a PetscMalloc3(). 348 349 */ 350 351 PetscErrorCode MatConvertToTriples_seqaij_seqaij(Mat A,PetscInt shift,MatReuse reuse,Mat_MUMPS *mumps) 352 { 353 const PetscScalar *av; 354 const PetscInt *ai,*aj,*ajj,M=A->rmap->n; 355 PetscInt64 nz,rnz,i,j,k; 356 PetscMUMPSInt *row,*col; 357 Mat_SeqAIJ *aa=(Mat_SeqAIJ*)A->data; 358 359 PetscFunctionBegin; 360 PetscCall(MatSeqAIJGetArrayRead(A,&av)); 361 mumps->val = (PetscScalar*)av; 362 if (reuse == MAT_INITIAL_MATRIX) { 363 nz = aa->nz; 364 ai = aa->i; 365 aj = aa->j; 366 PetscCall(PetscMalloc2(nz,&row,nz,&col)); 367 for (i=k=0; i<M; i++) { 368 rnz = ai[i+1] - ai[i]; 369 ajj = aj + ai[i]; 370 for (j=0; j<rnz; j++) { 371 PetscCall(PetscMUMPSIntCast(i+shift,&row[k])); 372 PetscCall(PetscMUMPSIntCast(ajj[j] + shift,&col[k])); 373 k++; 374 } 375 } 376 mumps->irn = row; 377 mumps->jcn = col; 378 mumps->nnz = nz; 379 } 380 PetscCall(MatSeqAIJRestoreArrayRead(A,&av)); 381 PetscFunctionReturn(0); 382 } 383 384 PetscErrorCode MatConvertToTriples_seqsell_seqaij(Mat A,PetscInt shift,MatReuse reuse,Mat_MUMPS *mumps) 385 { 386 PetscInt64 nz,i,j,k,r; 387 Mat_SeqSELL *a=(Mat_SeqSELL*)A->data; 388 PetscMUMPSInt *row,*col; 389 390 PetscFunctionBegin; 391 mumps->val = a->val; 392 if (reuse == MAT_INITIAL_MATRIX) { 393 nz = a->sliidx[a->totalslices]; 394 PetscCall(PetscMalloc2(nz,&row,nz,&col)); 395 for (i=k=0; i<a->totalslices; i++) { 396 for (j=a->sliidx[i],r=0; j<a->sliidx[i+1]; j++,r=((r+1)&0x07)) { 397 PetscCall(PetscMUMPSIntCast(8*i+r+shift,&row[k++])); 398 } 399 } 400 for (i=0;i<nz;i++) PetscCall(PetscMUMPSIntCast(a->colidx[i]+shift,&col[i])); 401 mumps->irn = row; 402 mumps->jcn = col; 403 mumps->nnz = nz; 404 } 405 PetscFunctionReturn(0); 406 } 407 408 PetscErrorCode MatConvertToTriples_seqbaij_seqaij(Mat A,PetscInt shift,MatReuse reuse,Mat_MUMPS *mumps) 409 { 410 Mat_SeqBAIJ *aa=(Mat_SeqBAIJ*)A->data; 411 const PetscInt *ai,*aj,*ajj,bs2 = aa->bs2; 412 PetscInt64 M,nz,idx=0,rnz,i,j,k,m; 413 PetscInt bs; 414 PetscMUMPSInt *row,*col; 415 416 PetscFunctionBegin; 417 PetscCall(MatGetBlockSize(A,&bs)); 418 M = A->rmap->N/bs; 419 mumps->val = aa->a; 420 if (reuse == MAT_INITIAL_MATRIX) { 421 ai = aa->i; aj = aa->j; 422 nz = bs2*aa->nz; 423 PetscCall(PetscMalloc2(nz,&row,nz,&col)); 424 for (i=0; i<M; i++) { 425 ajj = aj + ai[i]; 426 rnz = ai[i+1] - ai[i]; 427 for (k=0; k<rnz; k++) { 428 for (j=0; j<bs; j++) { 429 for (m=0; m<bs; m++) { 430 PetscCall(PetscMUMPSIntCast(i*bs + m + shift,&row[idx])); 431 PetscCall(PetscMUMPSIntCast(bs*ajj[k] + j + shift,&col[idx])); 432 idx++; 433 } 434 } 435 } 436 } 437 mumps->irn = row; 438 mumps->jcn = col; 439 mumps->nnz = nz; 440 } 441 PetscFunctionReturn(0); 442 } 443 444 PetscErrorCode MatConvertToTriples_seqsbaij_seqsbaij(Mat A,PetscInt shift,MatReuse reuse,Mat_MUMPS *mumps) 445 { 446 const PetscInt *ai, *aj,*ajj; 447 PetscInt bs; 448 PetscInt64 nz,rnz,i,j,k,m; 449 PetscMUMPSInt *row,*col; 450 PetscScalar *val; 451 Mat_SeqSBAIJ *aa=(Mat_SeqSBAIJ*)A->data; 452 const PetscInt bs2=aa->bs2,mbs=aa->mbs; 453 #if defined(PETSC_USE_COMPLEX) 454 PetscBool hermitian; 455 #endif 456 457 PetscFunctionBegin; 458 #if defined(PETSC_USE_COMPLEX) 459 PetscCall(MatGetOption(A,MAT_HERMITIAN,&hermitian)); 460 PetscCheck(!hermitian,PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"MUMPS does not support Hermitian symmetric matrices for Choleksy"); 461 #endif 462 ai = aa->i; 463 aj = aa->j; 464 PetscCall(MatGetBlockSize(A,&bs)); 465 if (reuse == MAT_INITIAL_MATRIX) { 466 nz = aa->nz; 467 PetscCall(PetscMalloc2(bs2*nz,&row,bs2*nz,&col)); 468 if (bs>1) { 469 PetscCall(PetscMalloc1(bs2*nz,&mumps->val_alloc)); 470 mumps->val = mumps->val_alloc; 471 } else { 472 mumps->val = aa->a; 473 } 474 mumps->irn = row; 475 mumps->jcn = col; 476 } else { 477 if (bs == 1) mumps->val = aa->a; 478 row = mumps->irn; 479 col = mumps->jcn; 480 } 481 val = mumps->val; 482 483 nz = 0; 484 if (bs>1) { 485 for (i=0; i<mbs; i++) { 486 rnz = ai[i+1] - ai[i]; 487 ajj = aj + ai[i]; 488 for (j=0; j<rnz; j++) { 489 for (k=0; k<bs; k++) { 490 for (m=0; m<bs; m++) { 491 if (ajj[j]>i || k>=m) { 492 if (reuse == MAT_INITIAL_MATRIX) { 493 PetscCall(PetscMUMPSIntCast(i*bs + m + shift,&row[nz])); 494 PetscCall(PetscMUMPSIntCast(ajj[j]*bs + k + shift,&col[nz])); 495 } 496 val[nz++] = aa->a[(ai[i]+j)*bs2 + m + k*bs]; 497 } 498 } 499 } 500 } 501 } 502 } else if (reuse == MAT_INITIAL_MATRIX) { 503 for (i=0; i<mbs; i++) { 504 rnz = ai[i+1] - ai[i]; 505 ajj = aj + ai[i]; 506 for (j=0; j<rnz; j++) { 507 PetscCall(PetscMUMPSIntCast(i+shift,&row[nz])); 508 PetscCall(PetscMUMPSIntCast(ajj[j] + shift,&col[nz])); 509 nz++; 510 } 511 } 512 PetscCheck(nz == aa->nz,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Different numbers of nonzeros %" PetscInt64_FMT " != %" PetscInt_FMT,nz,aa->nz); 513 } 514 if (reuse == MAT_INITIAL_MATRIX) mumps->nnz = nz; 515 PetscFunctionReturn(0); 516 } 517 518 PetscErrorCode MatConvertToTriples_seqaij_seqsbaij(Mat A,PetscInt shift,MatReuse reuse,Mat_MUMPS *mumps) 519 { 520 const PetscInt *ai,*aj,*ajj,*adiag,M=A->rmap->n; 521 PetscInt64 nz,rnz,i,j; 522 const PetscScalar *av,*v1; 523 PetscScalar *val; 524 PetscMUMPSInt *row,*col; 525 Mat_SeqAIJ *aa=(Mat_SeqAIJ*)A->data; 526 PetscBool missing; 527 #if defined(PETSC_USE_COMPLEX) 528 PetscBool hermitian; 529 #endif 530 531 PetscFunctionBegin; 532 #if defined(PETSC_USE_COMPLEX) 533 PetscCall(MatGetOption(A,MAT_HERMITIAN,&hermitian)); 534 PetscCheck(!hermitian,PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"MUMPS does not support Hermitian symmetric matrices for Choleksy"); 535 #endif 536 PetscCall(MatSeqAIJGetArrayRead(A,&av)); 537 ai = aa->i; aj = aa->j; 538 adiag = aa->diag; 539 PetscCall(MatMissingDiagonal_SeqAIJ(A,&missing,NULL)); 540 if (reuse == MAT_INITIAL_MATRIX) { 541 /* count nz in the upper triangular part of A */ 542 nz = 0; 543 if (missing) { 544 for (i=0; i<M; i++) { 545 if (PetscUnlikely(adiag[i] >= ai[i+1])) { 546 for (j=ai[i];j<ai[i+1];j++) { 547 if (aj[j] < i) continue; 548 nz++; 549 } 550 } else { 551 nz += ai[i+1] - adiag[i]; 552 } 553 } 554 } else { 555 for (i=0; i<M; i++) nz += ai[i+1] - adiag[i]; 556 } 557 PetscCall(PetscMalloc2(nz,&row,nz,&col)); 558 PetscCall(PetscMalloc1(nz,&val)); 559 mumps->nnz = nz; 560 mumps->irn = row; 561 mumps->jcn = col; 562 mumps->val = mumps->val_alloc = val; 563 564 nz = 0; 565 if (missing) { 566 for (i=0; i<M; i++) { 567 if (PetscUnlikely(adiag[i] >= ai[i+1])) { 568 for (j=ai[i];j<ai[i+1];j++) { 569 if (aj[j] < i) continue; 570 PetscCall(PetscMUMPSIntCast(i+shift,&row[nz])); 571 PetscCall(PetscMUMPSIntCast(aj[j]+shift,&col[nz])); 572 val[nz] = av[j]; 573 nz++; 574 } 575 } else { 576 rnz = ai[i+1] - adiag[i]; 577 ajj = aj + adiag[i]; 578 v1 = av + adiag[i]; 579 for (j=0; j<rnz; j++) { 580 PetscCall(PetscMUMPSIntCast(i+shift,&row[nz])); 581 PetscCall(PetscMUMPSIntCast(ajj[j] + shift,&col[nz])); 582 val[nz++] = v1[j]; 583 } 584 } 585 } 586 } else { 587 for (i=0; i<M; i++) { 588 rnz = ai[i+1] - adiag[i]; 589 ajj = aj + adiag[i]; 590 v1 = av + adiag[i]; 591 for (j=0; j<rnz; j++) { 592 PetscCall(PetscMUMPSIntCast(i+shift,&row[nz])); 593 PetscCall(PetscMUMPSIntCast(ajj[j] + shift,&col[nz])); 594 val[nz++] = v1[j]; 595 } 596 } 597 } 598 } else { 599 nz = 0; 600 val = mumps->val; 601 if (missing) { 602 for (i=0; i <M; i++) { 603 if (PetscUnlikely(adiag[i] >= ai[i+1])) { 604 for (j=ai[i];j<ai[i+1];j++) { 605 if (aj[j] < i) continue; 606 val[nz++] = av[j]; 607 } 608 } else { 609 rnz = ai[i+1] - adiag[i]; 610 v1 = av + adiag[i]; 611 for (j=0; j<rnz; j++) { 612 val[nz++] = v1[j]; 613 } 614 } 615 } 616 } else { 617 for (i=0; i <M; i++) { 618 rnz = ai[i+1] - adiag[i]; 619 v1 = av + adiag[i]; 620 for (j=0; j<rnz; j++) { 621 val[nz++] = v1[j]; 622 } 623 } 624 } 625 } 626 PetscCall(MatSeqAIJRestoreArrayRead(A,&av)); 627 PetscFunctionReturn(0); 628 } 629 630 PetscErrorCode MatConvertToTriples_mpisbaij_mpisbaij(Mat A,PetscInt shift,MatReuse reuse,Mat_MUMPS *mumps) 631 { 632 const PetscInt *ai,*aj,*bi,*bj,*garray,*ajj,*bjj; 633 PetscInt bs; 634 PetscInt64 rstart,nz,i,j,k,m,jj,irow,countA,countB; 635 PetscMUMPSInt *row,*col; 636 const PetscScalar *av,*bv,*v1,*v2; 637 PetscScalar *val; 638 Mat_MPISBAIJ *mat = (Mat_MPISBAIJ*)A->data; 639 Mat_SeqSBAIJ *aa = (Mat_SeqSBAIJ*)(mat->A)->data; 640 Mat_SeqBAIJ *bb = (Mat_SeqBAIJ*)(mat->B)->data; 641 const PetscInt bs2=aa->bs2,mbs=aa->mbs; 642 #if defined(PETSC_USE_COMPLEX) 643 PetscBool hermitian; 644 #endif 645 646 PetscFunctionBegin; 647 #if defined(PETSC_USE_COMPLEX) 648 PetscCall(MatGetOption(A,MAT_HERMITIAN,&hermitian)); 649 PetscCheck(!hermitian,PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"MUMPS does not support Hermitian symmetric matrices for Choleksy"); 650 #endif 651 PetscCall(MatGetBlockSize(A,&bs)); 652 rstart = A->rmap->rstart; 653 ai = aa->i; 654 aj = aa->j; 655 bi = bb->i; 656 bj = bb->j; 657 av = aa->a; 658 bv = bb->a; 659 660 garray = mat->garray; 661 662 if (reuse == MAT_INITIAL_MATRIX) { 663 nz = (aa->nz+bb->nz)*bs2; /* just a conservative estimate */ 664 PetscCall(PetscMalloc2(nz,&row,nz,&col)); 665 PetscCall(PetscMalloc1(nz,&val)); 666 /* can not decide the exact mumps->nnz now because of the SBAIJ */ 667 mumps->irn = row; 668 mumps->jcn = col; 669 mumps->val = mumps->val_alloc = val; 670 } else { 671 val = mumps->val; 672 } 673 674 jj = 0; irow = rstart; 675 for (i=0; i<mbs; i++) { 676 ajj = aj + ai[i]; /* ptr to the beginning of this row */ 677 countA = ai[i+1] - ai[i]; 678 countB = bi[i+1] - bi[i]; 679 bjj = bj + bi[i]; 680 v1 = av + ai[i]*bs2; 681 v2 = bv + bi[i]*bs2; 682 683 if (bs>1) { 684 /* A-part */ 685 for (j=0; j<countA; j++) { 686 for (k=0; k<bs; k++) { 687 for (m=0; m<bs; m++) { 688 if (rstart + ajj[j]*bs>irow || k>=m) { 689 if (reuse == MAT_INITIAL_MATRIX) { 690 PetscCall(PetscMUMPSIntCast(irow + m + shift,&row[jj])); 691 PetscCall(PetscMUMPSIntCast(rstart + ajj[j]*bs + k + shift,&col[jj])); 692 } 693 val[jj++] = v1[j*bs2 + m + k*bs]; 694 } 695 } 696 } 697 } 698 699 /* B-part */ 700 for (j=0; j < countB; j++) { 701 for (k=0; k<bs; k++) { 702 for (m=0; m<bs; m++) { 703 if (reuse == MAT_INITIAL_MATRIX) { 704 PetscCall(PetscMUMPSIntCast(irow + m + shift,&row[jj])); 705 PetscCall(PetscMUMPSIntCast(garray[bjj[j]]*bs + k + shift,&col[jj])); 706 } 707 val[jj++] = v2[j*bs2 + m + k*bs]; 708 } 709 } 710 } 711 } else { 712 /* A-part */ 713 for (j=0; j<countA; j++) { 714 if (reuse == MAT_INITIAL_MATRIX) { 715 PetscCall(PetscMUMPSIntCast(irow + shift,&row[jj])); 716 PetscCall(PetscMUMPSIntCast(rstart + ajj[j] + shift,&col[jj])); 717 } 718 val[jj++] = v1[j]; 719 } 720 721 /* B-part */ 722 for (j=0; j < countB; j++) { 723 if (reuse == MAT_INITIAL_MATRIX) { 724 PetscCall(PetscMUMPSIntCast(irow + shift,&row[jj])); 725 PetscCall(PetscMUMPSIntCast(garray[bjj[j]] + shift,&col[jj])); 726 } 727 val[jj++] = v2[j]; 728 } 729 } 730 irow+=bs; 731 } 732 mumps->nnz = jj; 733 PetscFunctionReturn(0); 734 } 735 736 PetscErrorCode MatConvertToTriples_mpiaij_mpiaij(Mat A,PetscInt shift,MatReuse reuse,Mat_MUMPS *mumps) 737 { 738 const PetscInt *ai, *aj, *bi, *bj,*garray,m=A->rmap->n,*ajj,*bjj; 739 PetscInt64 rstart,nz,i,j,jj,irow,countA,countB; 740 PetscMUMPSInt *row,*col; 741 const PetscScalar *av, *bv,*v1,*v2; 742 PetscScalar *val; 743 Mat Ad,Ao; 744 Mat_SeqAIJ *aa; 745 Mat_SeqAIJ *bb; 746 747 PetscFunctionBegin; 748 PetscCall(MatMPIAIJGetSeqAIJ(A,&Ad,&Ao,&garray)); 749 PetscCall(MatSeqAIJGetArrayRead(Ad,&av)); 750 PetscCall(MatSeqAIJGetArrayRead(Ao,&bv)); 751 752 aa = (Mat_SeqAIJ*)(Ad)->data; 753 bb = (Mat_SeqAIJ*)(Ao)->data; 754 ai = aa->i; 755 aj = aa->j; 756 bi = bb->i; 757 bj = bb->j; 758 759 rstart = A->rmap->rstart; 760 761 if (reuse == MAT_INITIAL_MATRIX) { 762 nz = (PetscInt64)aa->nz + bb->nz; /* make sure the sum won't overflow PetscInt */ 763 PetscCall(PetscMalloc2(nz,&row,nz,&col)); 764 PetscCall(PetscMalloc1(nz,&val)); 765 mumps->nnz = nz; 766 mumps->irn = row; 767 mumps->jcn = col; 768 mumps->val = mumps->val_alloc = val; 769 } else { 770 val = mumps->val; 771 } 772 773 jj = 0; irow = rstart; 774 for (i=0; i<m; i++) { 775 ajj = aj + ai[i]; /* ptr to the beginning of this row */ 776 countA = ai[i+1] - ai[i]; 777 countB = bi[i+1] - bi[i]; 778 bjj = bj + bi[i]; 779 v1 = av + ai[i]; 780 v2 = bv + bi[i]; 781 782 /* A-part */ 783 for (j=0; j<countA; j++) { 784 if (reuse == MAT_INITIAL_MATRIX) { 785 PetscCall(PetscMUMPSIntCast(irow + shift,&row[jj])); 786 PetscCall(PetscMUMPSIntCast(rstart + ajj[j] + shift,&col[jj])); 787 } 788 val[jj++] = v1[j]; 789 } 790 791 /* B-part */ 792 for (j=0; j < countB; j++) { 793 if (reuse == MAT_INITIAL_MATRIX) { 794 PetscCall(PetscMUMPSIntCast(irow + shift,&row[jj])); 795 PetscCall(PetscMUMPSIntCast(garray[bjj[j]] + shift,&col[jj])); 796 } 797 val[jj++] = v2[j]; 798 } 799 irow++; 800 } 801 PetscCall(MatSeqAIJRestoreArrayRead(Ad,&av)); 802 PetscCall(MatSeqAIJRestoreArrayRead(Ao,&bv)); 803 PetscFunctionReturn(0); 804 } 805 806 PetscErrorCode MatConvertToTriples_mpibaij_mpiaij(Mat A,PetscInt shift,MatReuse reuse,Mat_MUMPS *mumps) 807 { 808 Mat_MPIBAIJ *mat = (Mat_MPIBAIJ*)A->data; 809 Mat_SeqBAIJ *aa = (Mat_SeqBAIJ*)(mat->A)->data; 810 Mat_SeqBAIJ *bb = (Mat_SeqBAIJ*)(mat->B)->data; 811 const PetscInt *ai = aa->i, *bi = bb->i, *aj = aa->j, *bj = bb->j,*ajj, *bjj; 812 const PetscInt *garray = mat->garray,mbs=mat->mbs,rstart=A->rmap->rstart; 813 const PetscInt bs2=mat->bs2; 814 PetscInt bs; 815 PetscInt64 nz,i,j,k,n,jj,irow,countA,countB,idx; 816 PetscMUMPSInt *row,*col; 817 const PetscScalar *av=aa->a, *bv=bb->a,*v1,*v2; 818 PetscScalar *val; 819 820 PetscFunctionBegin; 821 PetscCall(MatGetBlockSize(A,&bs)); 822 if (reuse == MAT_INITIAL_MATRIX) { 823 nz = bs2*(aa->nz + bb->nz); 824 PetscCall(PetscMalloc2(nz,&row,nz,&col)); 825 PetscCall(PetscMalloc1(nz,&val)); 826 mumps->nnz = nz; 827 mumps->irn = row; 828 mumps->jcn = col; 829 mumps->val = mumps->val_alloc = val; 830 } else { 831 val = mumps->val; 832 } 833 834 jj = 0; irow = rstart; 835 for (i=0; i<mbs; i++) { 836 countA = ai[i+1] - ai[i]; 837 countB = bi[i+1] - bi[i]; 838 ajj = aj + ai[i]; 839 bjj = bj + bi[i]; 840 v1 = av + bs2*ai[i]; 841 v2 = bv + bs2*bi[i]; 842 843 idx = 0; 844 /* A-part */ 845 for (k=0; k<countA; k++) { 846 for (j=0; j<bs; j++) { 847 for (n=0; n<bs; n++) { 848 if (reuse == MAT_INITIAL_MATRIX) { 849 PetscCall(PetscMUMPSIntCast(irow + n + shift,&row[jj])); 850 PetscCall(PetscMUMPSIntCast(rstart + bs*ajj[k] + j + shift,&col[jj])); 851 } 852 val[jj++] = v1[idx++]; 853 } 854 } 855 } 856 857 idx = 0; 858 /* B-part */ 859 for (k=0; k<countB; k++) { 860 for (j=0; j<bs; j++) { 861 for (n=0; n<bs; n++) { 862 if (reuse == MAT_INITIAL_MATRIX) { 863 PetscCall(PetscMUMPSIntCast(irow + n + shift,&row[jj])); 864 PetscCall(PetscMUMPSIntCast(bs*garray[bjj[k]] + j + shift,&col[jj])); 865 } 866 val[jj++] = v2[idx++]; 867 } 868 } 869 } 870 irow += bs; 871 } 872 PetscFunctionReturn(0); 873 } 874 875 PetscErrorCode MatConvertToTriples_mpiaij_mpisbaij(Mat A,PetscInt shift,MatReuse reuse,Mat_MUMPS *mumps) 876 { 877 const PetscInt *ai, *aj,*adiag, *bi, *bj,*garray,m=A->rmap->n,*ajj,*bjj; 878 PetscInt64 rstart,nz,nza,nzb,i,j,jj,irow,countA,countB; 879 PetscMUMPSInt *row,*col; 880 const PetscScalar *av, *bv,*v1,*v2; 881 PetscScalar *val; 882 Mat Ad,Ao; 883 Mat_SeqAIJ *aa; 884 Mat_SeqAIJ *bb; 885 #if defined(PETSC_USE_COMPLEX) 886 PetscBool hermitian; 887 #endif 888 889 PetscFunctionBegin; 890 #if defined(PETSC_USE_COMPLEX) 891 PetscCall(MatGetOption(A,MAT_HERMITIAN,&hermitian)); 892 PetscCheck(!hermitian,PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"MUMPS does not support Hermitian symmetric matrices for Choleksy"); 893 #endif 894 PetscCall(MatMPIAIJGetSeqAIJ(A,&Ad,&Ao,&garray)); 895 PetscCall(MatSeqAIJGetArrayRead(Ad,&av)); 896 PetscCall(MatSeqAIJGetArrayRead(Ao,&bv)); 897 898 aa = (Mat_SeqAIJ*)(Ad)->data; 899 bb = (Mat_SeqAIJ*)(Ao)->data; 900 ai = aa->i; 901 aj = aa->j; 902 adiag = aa->diag; 903 bi = bb->i; 904 bj = bb->j; 905 906 rstart = A->rmap->rstart; 907 908 if (reuse == MAT_INITIAL_MATRIX) { 909 nza = 0; /* num of upper triangular entries in mat->A, including diagonals */ 910 nzb = 0; /* num of upper triangular entries in mat->B */ 911 for (i=0; i<m; i++) { 912 nza += (ai[i+1] - adiag[i]); 913 countB = bi[i+1] - bi[i]; 914 bjj = bj + bi[i]; 915 for (j=0; j<countB; j++) { 916 if (garray[bjj[j]] > rstart) nzb++; 917 } 918 } 919 920 nz = nza + nzb; /* total nz of upper triangular part of mat */ 921 PetscCall(PetscMalloc2(nz,&row,nz,&col)); 922 PetscCall(PetscMalloc1(nz,&val)); 923 mumps->nnz = nz; 924 mumps->irn = row; 925 mumps->jcn = col; 926 mumps->val = mumps->val_alloc = val; 927 } else { 928 val = mumps->val; 929 } 930 931 jj = 0; irow = rstart; 932 for (i=0; i<m; i++) { 933 ajj = aj + adiag[i]; /* ptr to the beginning of the diagonal of this row */ 934 v1 = av + adiag[i]; 935 countA = ai[i+1] - adiag[i]; 936 countB = bi[i+1] - bi[i]; 937 bjj = bj + bi[i]; 938 v2 = bv + bi[i]; 939 940 /* A-part */ 941 for (j=0; j<countA; j++) { 942 if (reuse == MAT_INITIAL_MATRIX) { 943 PetscCall(PetscMUMPSIntCast(irow + shift,&row[jj])); 944 PetscCall(PetscMUMPSIntCast(rstart + ajj[j] + shift,&col[jj])); 945 } 946 val[jj++] = v1[j]; 947 } 948 949 /* B-part */ 950 for (j=0; j < countB; j++) { 951 if (garray[bjj[j]] > rstart) { 952 if (reuse == MAT_INITIAL_MATRIX) { 953 PetscCall(PetscMUMPSIntCast(irow + shift,&row[jj])); 954 PetscCall(PetscMUMPSIntCast(garray[bjj[j]] + shift,&col[jj])); 955 } 956 val[jj++] = v2[j]; 957 } 958 } 959 irow++; 960 } 961 PetscCall(MatSeqAIJRestoreArrayRead(Ad,&av)); 962 PetscCall(MatSeqAIJRestoreArrayRead(Ao,&bv)); 963 PetscFunctionReturn(0); 964 } 965 966 PetscErrorCode MatDestroy_MUMPS(Mat A) 967 { 968 Mat_MUMPS *mumps=(Mat_MUMPS*)A->data; 969 970 PetscFunctionBegin; 971 PetscCall(PetscFree2(mumps->id.sol_loc,mumps->id.isol_loc)); 972 PetscCall(VecScatterDestroy(&mumps->scat_rhs)); 973 PetscCall(VecScatterDestroy(&mumps->scat_sol)); 974 PetscCall(VecDestroy(&mumps->b_seq)); 975 PetscCall(VecDestroy(&mumps->x_seq)); 976 PetscCall(PetscFree(mumps->id.perm_in)); 977 PetscCall(PetscFree2(mumps->irn,mumps->jcn)); 978 PetscCall(PetscFree(mumps->val_alloc)); 979 PetscCall(PetscFree(mumps->info)); 980 PetscCall(MatMumpsResetSchur_Private(mumps)); 981 mumps->id.job = JOB_END; 982 PetscMUMPS_c(mumps); 983 PetscCheck(mumps->id.INFOG(1) >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MUMPS in MatDestroy_MUMPS: INFOG(1)=%d",mumps->id.INFOG(1)); 984 #if defined(PETSC_HAVE_OPENMP_SUPPORT) 985 if (mumps->use_petsc_omp_support) { 986 PetscCall(PetscOmpCtrlDestroy(&mumps->omp_ctrl)); 987 PetscCall(PetscFree2(mumps->rhs_loc,mumps->rhs_recvbuf)); 988 PetscCall(PetscFree3(mumps->rhs_nrow,mumps->rhs_recvcounts,mumps->rhs_disps)); 989 } 990 #endif 991 PetscCall(PetscFree(mumps->ia_alloc)); 992 PetscCall(PetscFree(mumps->ja_alloc)); 993 PetscCall(PetscFree(mumps->recvcount)); 994 PetscCall(PetscFree(mumps->reqs)); 995 PetscCall(PetscFree(mumps->irhs_loc)); 996 if (mumps->mumps_comm != MPI_COMM_NULL) PetscCall(PetscCommRestoreComm(PetscObjectComm((PetscObject)A),&mumps->mumps_comm)); 997 PetscCall(PetscFree(A->data)); 998 999 /* clear composed functions */ 1000 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatFactorGetSolverType_C",NULL)); 1001 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatFactorSetSchurIS_C",NULL)); 1002 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatFactorCreateSchurComplement_C",NULL)); 1003 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatMumpsSetIcntl_C",NULL)); 1004 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatMumpsGetIcntl_C",NULL)); 1005 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatMumpsSetCntl_C",NULL)); 1006 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatMumpsGetCntl_C",NULL)); 1007 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatMumpsGetInfo_C",NULL)); 1008 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatMumpsGetInfog_C",NULL)); 1009 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatMumpsGetRinfo_C",NULL)); 1010 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatMumpsGetRinfog_C",NULL)); 1011 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatMumpsGetInverse_C",NULL)); 1012 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatMumpsGetInverseTranspose_C",NULL)); 1013 PetscFunctionReturn(0); 1014 } 1015 1016 /* Set up the distributed RHS info for MUMPS. <nrhs> is the number of RHS. <array> points to start of RHS on the local processor. */ 1017 static PetscErrorCode MatMumpsSetUpDistRHSInfo(Mat A,PetscInt nrhs,const PetscScalar *array) 1018 { 1019 Mat_MUMPS *mumps=(Mat_MUMPS*)A->data; 1020 const PetscMPIInt ompsize=mumps->omp_comm_size; 1021 PetscInt i,m,M,rstart; 1022 1023 PetscFunctionBegin; 1024 PetscCall(MatGetSize(A,&M,NULL)); 1025 PetscCall(MatGetLocalSize(A,&m,NULL)); 1026 PetscCheck(M <= PETSC_MUMPS_INT_MAX,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"PetscInt too long for PetscMUMPSInt"); 1027 if (ompsize == 1) { 1028 if (!mumps->irhs_loc) { 1029 mumps->nloc_rhs = m; 1030 PetscCall(PetscMalloc1(m,&mumps->irhs_loc)); 1031 PetscCall(MatGetOwnershipRange(A,&rstart,NULL)); 1032 for (i=0; i<m; i++) mumps->irhs_loc[i] = rstart+i+1; /* use 1-based indices */ 1033 } 1034 mumps->id.rhs_loc = (MumpsScalar*)array; 1035 } else { 1036 #if defined(PETSC_HAVE_OPENMP_SUPPORT) 1037 const PetscInt *ranges; 1038 PetscMPIInt j,k,sendcount,*petsc_ranks,*omp_ranks; 1039 MPI_Group petsc_group,omp_group; 1040 PetscScalar *recvbuf=NULL; 1041 1042 if (mumps->is_omp_master) { 1043 /* Lazily initialize the omp stuff for distributed rhs */ 1044 if (!mumps->irhs_loc) { 1045 PetscCall(PetscMalloc2(ompsize,&omp_ranks,ompsize,&petsc_ranks)); 1046 PetscCall(PetscMalloc3(ompsize,&mumps->rhs_nrow,ompsize,&mumps->rhs_recvcounts,ompsize,&mumps->rhs_disps)); 1047 PetscCallMPI(MPI_Comm_group(mumps->petsc_comm,&petsc_group)); 1048 PetscCallMPI(MPI_Comm_group(mumps->omp_comm,&omp_group)); 1049 for (j=0; j<ompsize; j++) omp_ranks[j] = j; 1050 PetscCallMPI(MPI_Group_translate_ranks(omp_group,ompsize,omp_ranks,petsc_group,petsc_ranks)); 1051 1052 /* Populate mumps->irhs_loc[], rhs_nrow[] */ 1053 mumps->nloc_rhs = 0; 1054 PetscCall(MatGetOwnershipRanges(A,&ranges)); 1055 for (j=0; j<ompsize; j++) { 1056 mumps->rhs_nrow[j] = ranges[petsc_ranks[j]+1] - ranges[petsc_ranks[j]]; 1057 mumps->nloc_rhs += mumps->rhs_nrow[j]; 1058 } 1059 PetscCall(PetscMalloc1(mumps->nloc_rhs,&mumps->irhs_loc)); 1060 for (j=k=0; j<ompsize; j++) { 1061 for (i=ranges[petsc_ranks[j]]; i<ranges[petsc_ranks[j]+1]; i++,k++) mumps->irhs_loc[k] = i+1; /* uses 1-based indices */ 1062 } 1063 1064 PetscCall(PetscFree2(omp_ranks,petsc_ranks)); 1065 PetscCallMPI(MPI_Group_free(&petsc_group)); 1066 PetscCallMPI(MPI_Group_free(&omp_group)); 1067 } 1068 1069 /* Realloc buffers when current nrhs is bigger than what we have met */ 1070 if (nrhs > mumps->max_nrhs) { 1071 PetscCall(PetscFree2(mumps->rhs_loc,mumps->rhs_recvbuf)); 1072 PetscCall(PetscMalloc2(mumps->nloc_rhs*nrhs,&mumps->rhs_loc,mumps->nloc_rhs*nrhs,&mumps->rhs_recvbuf)); 1073 mumps->max_nrhs = nrhs; 1074 } 1075 1076 /* Setup recvcounts[], disps[], recvbuf on omp rank 0 for the upcoming MPI_Gatherv */ 1077 for (j=0; j<ompsize; j++) PetscCall(PetscMPIIntCast(mumps->rhs_nrow[j]*nrhs,&mumps->rhs_recvcounts[j])); 1078 mumps->rhs_disps[0] = 0; 1079 for (j=1; j<ompsize; j++) { 1080 mumps->rhs_disps[j] = mumps->rhs_disps[j-1] + mumps->rhs_recvcounts[j-1]; 1081 PetscCheck(mumps->rhs_disps[j] >= 0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"PetscMPIInt overflow!"); 1082 } 1083 recvbuf = (nrhs == 1) ? mumps->rhs_loc : mumps->rhs_recvbuf; /* Directly use rhs_loc[] as recvbuf. Single rhs is common in Ax=b */ 1084 } 1085 1086 PetscCall(PetscMPIIntCast(m*nrhs,&sendcount)); 1087 PetscCallMPI(MPI_Gatherv(array,sendcount,MPIU_SCALAR,recvbuf,mumps->rhs_recvcounts,mumps->rhs_disps,MPIU_SCALAR,0,mumps->omp_comm)); 1088 1089 if (mumps->is_omp_master) { 1090 if (nrhs > 1) { /* Copy & re-arrange data from rhs_recvbuf[] to mumps->rhs_loc[] only when there are multiple rhs */ 1091 PetscScalar *dst,*dstbase = mumps->rhs_loc; 1092 for (j=0; j<ompsize; j++) { 1093 const PetscScalar *src = mumps->rhs_recvbuf + mumps->rhs_disps[j]; 1094 dst = dstbase; 1095 for (i=0; i<nrhs; i++) { 1096 PetscCall(PetscArraycpy(dst,src,mumps->rhs_nrow[j])); 1097 src += mumps->rhs_nrow[j]; 1098 dst += mumps->nloc_rhs; 1099 } 1100 dstbase += mumps->rhs_nrow[j]; 1101 } 1102 } 1103 mumps->id.rhs_loc = (MumpsScalar*)mumps->rhs_loc; 1104 } 1105 #endif /* PETSC_HAVE_OPENMP_SUPPORT */ 1106 } 1107 mumps->id.nrhs = nrhs; 1108 mumps->id.nloc_rhs = mumps->nloc_rhs; 1109 mumps->id.lrhs_loc = mumps->nloc_rhs; 1110 mumps->id.irhs_loc = mumps->irhs_loc; 1111 PetscFunctionReturn(0); 1112 } 1113 1114 PetscErrorCode MatSolve_MUMPS(Mat A,Vec b,Vec x) 1115 { 1116 Mat_MUMPS *mumps=(Mat_MUMPS*)A->data; 1117 const PetscScalar *rarray = NULL; 1118 PetscScalar *array; 1119 IS is_iden,is_petsc; 1120 PetscInt i; 1121 PetscBool second_solve = PETSC_FALSE; 1122 static PetscBool cite1 = PETSC_FALSE,cite2 = PETSC_FALSE; 1123 1124 PetscFunctionBegin; 1125 PetscCall(PetscCitationsRegister("@article{MUMPS01,\n author = {P.~R. Amestoy and I.~S. Duff and J.-Y. L'Excellent and J. Koster},\n title = {A fully asynchronous multifrontal solver using distributed dynamic scheduling},\n journal = {SIAM Journal on Matrix Analysis and Applications},\n volume = {23},\n number = {1},\n pages = {15--41},\n year = {2001}\n}\n",&cite1)); 1126 PetscCall(PetscCitationsRegister("@article{MUMPS02,\n author = {P.~R. Amestoy and A. Guermouche and J.-Y. L'Excellent and S. Pralet},\n title = {Hybrid scheduling for the parallel solution of linear systems},\n journal = {Parallel Computing},\n volume = {32},\n number = {2},\n pages = {136--156},\n year = {2006}\n}\n",&cite2)); 1127 1128 if (A->factorerrortype) { 1129 PetscCall(PetscInfo(A,"MatSolve is called with singular matrix factor, INFOG(1)=%d, INFO(2)=%d\n",mumps->id.INFOG(1),mumps->id.INFO(2))); 1130 PetscCall(VecSetInf(x)); 1131 PetscFunctionReturn(0); 1132 } 1133 1134 mumps->id.nrhs = 1; 1135 if (mumps->petsc_size > 1) { 1136 if (mumps->ICNTL20 == 10) { 1137 mumps->id.ICNTL(20) = 10; /* dense distributed RHS */ 1138 PetscCall(VecGetArrayRead(b,&rarray)); 1139 PetscCall(MatMumpsSetUpDistRHSInfo(A,1,rarray)); 1140 } else { 1141 mumps->id.ICNTL(20) = 0; /* dense centralized RHS; Scatter b into a sequential rhs vector*/ 1142 PetscCall(VecScatterBegin(mumps->scat_rhs,b,mumps->b_seq,INSERT_VALUES,SCATTER_FORWARD)); 1143 PetscCall(VecScatterEnd(mumps->scat_rhs,b,mumps->b_seq,INSERT_VALUES,SCATTER_FORWARD)); 1144 if (!mumps->myid) { 1145 PetscCall(VecGetArray(mumps->b_seq,&array)); 1146 mumps->id.rhs = (MumpsScalar*)array; 1147 } 1148 } 1149 } else { /* petsc_size == 1 */ 1150 mumps->id.ICNTL(20) = 0; /* dense centralized RHS */ 1151 PetscCall(VecCopy(b,x)); 1152 PetscCall(VecGetArray(x,&array)); 1153 mumps->id.rhs = (MumpsScalar*)array; 1154 } 1155 1156 /* 1157 handle condensation step of Schur complement (if any) 1158 We set by default ICNTL(26) == -1 when Schur indices have been provided by the user. 1159 According to MUMPS (5.0.0) manual, any value should be harmful during the factorization phase 1160 Unless the user provides a valid value for ICNTL(26), MatSolve and MatMatSolve routines solve the full system. 1161 This requires an extra call to PetscMUMPS_c and the computation of the factors for S 1162 */ 1163 if (mumps->id.size_schur > 0 && (mumps->id.ICNTL(26) < 0 || mumps->id.ICNTL(26) > 2)) { 1164 PetscCheck(mumps->petsc_size <= 1,PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"Parallel Schur complements not yet supported from PETSc"); 1165 second_solve = PETSC_TRUE; 1166 PetscCall(MatMumpsHandleSchur_Private(A,PETSC_FALSE)); 1167 } 1168 /* solve phase */ 1169 /*-------------*/ 1170 mumps->id.job = JOB_SOLVE; 1171 PetscMUMPS_c(mumps); 1172 PetscCheck(mumps->id.INFOG(1) >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MUMPS in solve phase: INFOG(1)=%d",mumps->id.INFOG(1)); 1173 1174 /* handle expansion step of Schur complement (if any) */ 1175 if (second_solve) { 1176 PetscCall(MatMumpsHandleSchur_Private(A,PETSC_TRUE)); 1177 } 1178 1179 if (mumps->petsc_size > 1) { /* convert mumps distributed solution to petsc mpi x */ 1180 if (mumps->scat_sol && mumps->ICNTL9_pre != mumps->id.ICNTL(9)) { 1181 /* when id.ICNTL(9) changes, the contents of lsol_loc may change (not its size, lsol_loc), recreates scat_sol */ 1182 PetscCall(VecScatterDestroy(&mumps->scat_sol)); 1183 } 1184 if (!mumps->scat_sol) { /* create scatter scat_sol */ 1185 PetscInt *isol2_loc=NULL; 1186 PetscCall(ISCreateStride(PETSC_COMM_SELF,mumps->id.lsol_loc,0,1,&is_iden)); /* from */ 1187 PetscCall(PetscMalloc1(mumps->id.lsol_loc,&isol2_loc)); 1188 for (i=0; i<mumps->id.lsol_loc; i++) isol2_loc[i] = mumps->id.isol_loc[i]-1; /* change Fortran style to C style */ 1189 PetscCall(ISCreateGeneral(PETSC_COMM_SELF,mumps->id.lsol_loc,isol2_loc,PETSC_OWN_POINTER,&is_petsc)); /* to */ 1190 PetscCall(VecScatterCreate(mumps->x_seq,is_iden,x,is_petsc,&mumps->scat_sol)); 1191 PetscCall(ISDestroy(&is_iden)); 1192 PetscCall(ISDestroy(&is_petsc)); 1193 mumps->ICNTL9_pre = mumps->id.ICNTL(9); /* save current value of id.ICNTL(9) */ 1194 } 1195 1196 PetscCall(VecScatterBegin(mumps->scat_sol,mumps->x_seq,x,INSERT_VALUES,SCATTER_FORWARD)); 1197 PetscCall(VecScatterEnd(mumps->scat_sol,mumps->x_seq,x,INSERT_VALUES,SCATTER_FORWARD)); 1198 } 1199 1200 if (mumps->petsc_size > 1) { 1201 if (mumps->ICNTL20 == 10) { 1202 PetscCall(VecRestoreArrayRead(b,&rarray)); 1203 } else if (!mumps->myid) { 1204 PetscCall(VecRestoreArray(mumps->b_seq,&array)); 1205 } 1206 } else PetscCall(VecRestoreArray(x,&array)); 1207 1208 PetscCall(PetscLogFlops(2.0*mumps->id.RINFO(3))); 1209 PetscFunctionReturn(0); 1210 } 1211 1212 PetscErrorCode MatSolveTranspose_MUMPS(Mat A,Vec b,Vec x) 1213 { 1214 Mat_MUMPS *mumps=(Mat_MUMPS*)A->data; 1215 1216 PetscFunctionBegin; 1217 mumps->id.ICNTL(9) = 0; 1218 PetscCall(MatSolve_MUMPS(A,b,x)); 1219 mumps->id.ICNTL(9) = 1; 1220 PetscFunctionReturn(0); 1221 } 1222 1223 PetscErrorCode MatMatSolve_MUMPS(Mat A,Mat B,Mat X) 1224 { 1225 Mat Bt = NULL; 1226 PetscBool denseX,denseB,flg,flgT; 1227 Mat_MUMPS *mumps=(Mat_MUMPS*)A->data; 1228 PetscInt i,nrhs,M; 1229 PetscScalar *array; 1230 const PetscScalar *rbray; 1231 PetscInt lsol_loc,nlsol_loc,*idxx,iidx = 0; 1232 PetscMUMPSInt *isol_loc,*isol_loc_save; 1233 PetscScalar *bray,*sol_loc,*sol_loc_save; 1234 IS is_to,is_from; 1235 PetscInt k,proc,j,m,myrstart; 1236 const PetscInt *rstart; 1237 Vec v_mpi,msol_loc; 1238 VecScatter scat_sol; 1239 Vec b_seq; 1240 VecScatter scat_rhs; 1241 PetscScalar *aa; 1242 PetscInt spnr,*ia,*ja; 1243 Mat_MPIAIJ *b = NULL; 1244 1245 PetscFunctionBegin; 1246 PetscCall(PetscObjectTypeCompareAny((PetscObject)X,&denseX,MATSEQDENSE,MATMPIDENSE,NULL)); 1247 PetscCheck(denseX,PetscObjectComm((PetscObject)X),PETSC_ERR_ARG_WRONG,"Matrix X must be MATDENSE matrix"); 1248 1249 PetscCall(PetscObjectTypeCompareAny((PetscObject)B,&denseB,MATSEQDENSE,MATMPIDENSE,NULL)); 1250 if (denseB) { 1251 PetscCheck(B->rmap->n == X->rmap->n,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Matrix B and X must have same row distribution"); 1252 mumps->id.ICNTL(20)= 0; /* dense RHS */ 1253 } else { /* sparse B */ 1254 PetscCheck(X != B,PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_IDN,"X and B must be different matrices"); 1255 PetscCall(PetscObjectTypeCompare((PetscObject)B,MATTRANSPOSEMAT,&flgT)); 1256 if (flgT) { /* input B is transpose of actural RHS matrix, 1257 because mumps requires sparse compressed COLUMN storage! See MatMatTransposeSolve_MUMPS() */ 1258 PetscCall(MatTransposeGetMat(B,&Bt)); 1259 } else SETERRQ(PetscObjectComm((PetscObject)B),PETSC_ERR_ARG_WRONG,"Matrix B must be MATTRANSPOSEMAT matrix"); 1260 mumps->id.ICNTL(20)= 1; /* sparse RHS */ 1261 } 1262 1263 PetscCall(MatGetSize(B,&M,&nrhs)); 1264 mumps->id.nrhs = nrhs; 1265 mumps->id.lrhs = M; 1266 mumps->id.rhs = NULL; 1267 1268 if (mumps->petsc_size == 1) { 1269 PetscScalar *aa; 1270 PetscInt spnr,*ia,*ja; 1271 PetscBool second_solve = PETSC_FALSE; 1272 1273 PetscCall(MatDenseGetArray(X,&array)); 1274 mumps->id.rhs = (MumpsScalar*)array; 1275 1276 if (denseB) { 1277 /* copy B to X */ 1278 PetscCall(MatDenseGetArrayRead(B,&rbray)); 1279 PetscCall(PetscArraycpy(array,rbray,M*nrhs)); 1280 PetscCall(MatDenseRestoreArrayRead(B,&rbray)); 1281 } else { /* sparse B */ 1282 PetscCall(MatSeqAIJGetArray(Bt,&aa)); 1283 PetscCall(MatGetRowIJ(Bt,1,PETSC_FALSE,PETSC_FALSE,&spnr,(const PetscInt**)&ia,(const PetscInt**)&ja,&flg)); 1284 PetscCheck(flg,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Cannot get IJ structure"); 1285 PetscCall(PetscMUMPSIntCSRCast(mumps,spnr,ia,ja,&mumps->id.irhs_ptr,&mumps->id.irhs_sparse,&mumps->id.nz_rhs)); 1286 mumps->id.rhs_sparse = (MumpsScalar*)aa; 1287 } 1288 /* handle condensation step of Schur complement (if any) */ 1289 if (mumps->id.size_schur > 0 && (mumps->id.ICNTL(26) < 0 || mumps->id.ICNTL(26) > 2)) { 1290 second_solve = PETSC_TRUE; 1291 PetscCall(MatMumpsHandleSchur_Private(A,PETSC_FALSE)); 1292 } 1293 /* solve phase */ 1294 /*-------------*/ 1295 mumps->id.job = JOB_SOLVE; 1296 PetscMUMPS_c(mumps); 1297 PetscCheck(mumps->id.INFOG(1) >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MUMPS in solve phase: INFOG(1)=%d",mumps->id.INFOG(1)); 1298 1299 /* handle expansion step of Schur complement (if any) */ 1300 if (second_solve) { 1301 PetscCall(MatMumpsHandleSchur_Private(A,PETSC_TRUE)); 1302 } 1303 if (!denseB) { /* sparse B */ 1304 PetscCall(MatSeqAIJRestoreArray(Bt,&aa)); 1305 PetscCall(MatRestoreRowIJ(Bt,1,PETSC_FALSE,PETSC_FALSE,&spnr,(const PetscInt**)&ia,(const PetscInt**)&ja,&flg)); 1306 PetscCheck(flg,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Cannot restore IJ structure"); 1307 } 1308 PetscCall(MatDenseRestoreArray(X,&array)); 1309 PetscFunctionReturn(0); 1310 } 1311 1312 /*--------- parallel case: MUMPS requires rhs B to be centralized on the host! --------*/ 1313 PetscCheckFalse(mumps->petsc_size > 1 && mumps->id.ICNTL(19),PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"Parallel Schur complements not yet supported from PETSc"); 1314 1315 /* create msol_loc to hold mumps local solution */ 1316 isol_loc_save = mumps->id.isol_loc; /* save it for MatSolve() */ 1317 sol_loc_save = (PetscScalar*)mumps->id.sol_loc; 1318 1319 lsol_loc = mumps->id.lsol_loc; 1320 nlsol_loc = nrhs*lsol_loc; /* length of sol_loc */ 1321 PetscCall(PetscMalloc2(nlsol_loc,&sol_loc,lsol_loc,&isol_loc)); 1322 mumps->id.sol_loc = (MumpsScalar*)sol_loc; 1323 mumps->id.isol_loc = isol_loc; 1324 1325 PetscCall(VecCreateSeqWithArray(PETSC_COMM_SELF,1,nlsol_loc,(PetscScalar*)sol_loc,&msol_loc)); 1326 1327 if (denseB) { 1328 if (mumps->ICNTL20 == 10) { 1329 mumps->id.ICNTL(20) = 10; /* dense distributed RHS */ 1330 PetscCall(MatDenseGetArrayRead(B,&rbray)); 1331 PetscCall(MatMumpsSetUpDistRHSInfo(A,nrhs,rbray)); 1332 PetscCall(MatDenseRestoreArrayRead(B,&rbray)); 1333 PetscCall(MatGetLocalSize(B,&m,NULL)); 1334 PetscCall(VecCreateMPIWithArray(PetscObjectComm((PetscObject)B),1,nrhs*m,nrhs*M,NULL,&v_mpi)); 1335 } else { 1336 mumps->id.ICNTL(20) = 0; /* dense centralized RHS */ 1337 /* TODO: Because of non-contiguous indices, the created vecscatter scat_rhs is not done in MPI_Gather, resulting in 1338 very inefficient communication. An optimization is to use VecScatterCreateToZero to gather B to rank 0. Then on rank 1339 0, re-arrange B into desired order, which is a local operation. 1340 */ 1341 1342 /* scatter v_mpi to b_seq because MUMPS before 5.3.0 only supports centralized rhs */ 1343 /* wrap dense rhs matrix B into a vector v_mpi */ 1344 PetscCall(MatGetLocalSize(B,&m,NULL)); 1345 PetscCall(MatDenseGetArray(B,&bray)); 1346 PetscCall(VecCreateMPIWithArray(PetscObjectComm((PetscObject)B),1,nrhs*m,nrhs*M,(const PetscScalar*)bray,&v_mpi)); 1347 PetscCall(MatDenseRestoreArray(B,&bray)); 1348 1349 /* scatter v_mpi to b_seq in proc[0]. MUMPS requires rhs to be centralized on the host! */ 1350 if (!mumps->myid) { 1351 PetscInt *idx; 1352 /* idx: maps from k-th index of v_mpi to (i,j)-th global entry of B */ 1353 PetscCall(PetscMalloc1(nrhs*M,&idx)); 1354 PetscCall(MatGetOwnershipRanges(B,&rstart)); 1355 k = 0; 1356 for (proc=0; proc<mumps->petsc_size; proc++) { 1357 for (j=0; j<nrhs; j++) { 1358 for (i=rstart[proc]; i<rstart[proc+1]; i++) idx[k++] = j*M + i; 1359 } 1360 } 1361 1362 PetscCall(VecCreateSeq(PETSC_COMM_SELF,nrhs*M,&b_seq)); 1363 PetscCall(ISCreateGeneral(PETSC_COMM_SELF,nrhs*M,idx,PETSC_OWN_POINTER,&is_to)); 1364 PetscCall(ISCreateStride(PETSC_COMM_SELF,nrhs*M,0,1,&is_from)); 1365 } else { 1366 PetscCall(VecCreateSeq(PETSC_COMM_SELF,0,&b_seq)); 1367 PetscCall(ISCreateStride(PETSC_COMM_SELF,0,0,1,&is_to)); 1368 PetscCall(ISCreateStride(PETSC_COMM_SELF,0,0,1,&is_from)); 1369 } 1370 PetscCall(VecScatterCreate(v_mpi,is_from,b_seq,is_to,&scat_rhs)); 1371 PetscCall(VecScatterBegin(scat_rhs,v_mpi,b_seq,INSERT_VALUES,SCATTER_FORWARD)); 1372 PetscCall(ISDestroy(&is_to)); 1373 PetscCall(ISDestroy(&is_from)); 1374 PetscCall(VecScatterEnd(scat_rhs,v_mpi,b_seq,INSERT_VALUES,SCATTER_FORWARD)); 1375 1376 if (!mumps->myid) { /* define rhs on the host */ 1377 PetscCall(VecGetArray(b_seq,&bray)); 1378 mumps->id.rhs = (MumpsScalar*)bray; 1379 PetscCall(VecRestoreArray(b_seq,&bray)); 1380 } 1381 } 1382 } else { /* sparse B */ 1383 b = (Mat_MPIAIJ*)Bt->data; 1384 1385 /* wrap dense X into a vector v_mpi */ 1386 PetscCall(MatGetLocalSize(X,&m,NULL)); 1387 PetscCall(MatDenseGetArray(X,&bray)); 1388 PetscCall(VecCreateMPIWithArray(PetscObjectComm((PetscObject)X),1,nrhs*m,nrhs*M,(const PetscScalar*)bray,&v_mpi)); 1389 PetscCall(MatDenseRestoreArray(X,&bray)); 1390 1391 if (!mumps->myid) { 1392 PetscCall(MatSeqAIJGetArray(b->A,&aa)); 1393 PetscCall(MatGetRowIJ(b->A,1,PETSC_FALSE,PETSC_FALSE,&spnr,(const PetscInt**)&ia,(const PetscInt**)&ja,&flg)); 1394 PetscCheck(flg,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Cannot get IJ structure"); 1395 PetscCall(PetscMUMPSIntCSRCast(mumps,spnr,ia,ja,&mumps->id.irhs_ptr,&mumps->id.irhs_sparse,&mumps->id.nz_rhs)); 1396 mumps->id.rhs_sparse = (MumpsScalar*)aa; 1397 } else { 1398 mumps->id.irhs_ptr = NULL; 1399 mumps->id.irhs_sparse = NULL; 1400 mumps->id.nz_rhs = 0; 1401 mumps->id.rhs_sparse = NULL; 1402 } 1403 } 1404 1405 /* solve phase */ 1406 /*-------------*/ 1407 mumps->id.job = JOB_SOLVE; 1408 PetscMUMPS_c(mumps); 1409 PetscCheck(mumps->id.INFOG(1) >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MUMPS in solve phase: INFOG(1)=%d",mumps->id.INFOG(1)); 1410 1411 /* scatter mumps distributed solution to petsc vector v_mpi, which shares local arrays with solution matrix X */ 1412 PetscCall(MatDenseGetArray(X,&array)); 1413 PetscCall(VecPlaceArray(v_mpi,array)); 1414 1415 /* create scatter scat_sol */ 1416 PetscCall(MatGetOwnershipRanges(X,&rstart)); 1417 /* iidx: index for scatter mumps solution to petsc X */ 1418 1419 PetscCall(ISCreateStride(PETSC_COMM_SELF,nlsol_loc,0,1,&is_from)); 1420 PetscCall(PetscMalloc1(nlsol_loc,&idxx)); 1421 for (i=0; i<lsol_loc; i++) { 1422 isol_loc[i] -= 1; /* change Fortran style to C style. isol_loc[i+j*lsol_loc] contains x[isol_loc[i]] in j-th vector */ 1423 1424 for (proc=0; proc<mumps->petsc_size; proc++) { 1425 if (isol_loc[i] >= rstart[proc] && isol_loc[i] < rstart[proc+1]) { 1426 myrstart = rstart[proc]; 1427 k = isol_loc[i] - myrstart; /* local index on 1st column of petsc vector X */ 1428 iidx = k + myrstart*nrhs; /* maps mumps isol_loc[i] to petsc index in X */ 1429 m = rstart[proc+1] - rstart[proc]; /* rows of X for this proc */ 1430 break; 1431 } 1432 } 1433 1434 for (j=0; j<nrhs; j++) idxx[i+j*lsol_loc] = iidx + j*m; 1435 } 1436 PetscCall(ISCreateGeneral(PETSC_COMM_SELF,nlsol_loc,idxx,PETSC_COPY_VALUES,&is_to)); 1437 PetscCall(VecScatterCreate(msol_loc,is_from,v_mpi,is_to,&scat_sol)); 1438 PetscCall(VecScatterBegin(scat_sol,msol_loc,v_mpi,INSERT_VALUES,SCATTER_FORWARD)); 1439 PetscCall(ISDestroy(&is_from)); 1440 PetscCall(ISDestroy(&is_to)); 1441 PetscCall(VecScatterEnd(scat_sol,msol_loc,v_mpi,INSERT_VALUES,SCATTER_FORWARD)); 1442 PetscCall(MatDenseRestoreArray(X,&array)); 1443 1444 /* free spaces */ 1445 mumps->id.sol_loc = (MumpsScalar*)sol_loc_save; 1446 mumps->id.isol_loc = isol_loc_save; 1447 1448 PetscCall(PetscFree2(sol_loc,isol_loc)); 1449 PetscCall(PetscFree(idxx)); 1450 PetscCall(VecDestroy(&msol_loc)); 1451 PetscCall(VecDestroy(&v_mpi)); 1452 if (!denseB) { 1453 if (!mumps->myid) { 1454 b = (Mat_MPIAIJ*)Bt->data; 1455 PetscCall(MatSeqAIJRestoreArray(b->A,&aa)); 1456 PetscCall(MatRestoreRowIJ(b->A,1,PETSC_FALSE,PETSC_FALSE,&spnr,(const PetscInt**)&ia,(const PetscInt**)&ja,&flg)); 1457 PetscCheck(flg,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Cannot restore IJ structure"); 1458 } 1459 } else { 1460 if (mumps->ICNTL20 == 0) { 1461 PetscCall(VecDestroy(&b_seq)); 1462 PetscCall(VecScatterDestroy(&scat_rhs)); 1463 } 1464 } 1465 PetscCall(VecScatterDestroy(&scat_sol)); 1466 PetscCall(PetscLogFlops(2.0*nrhs*mumps->id.RINFO(3))); 1467 PetscFunctionReturn(0); 1468 } 1469 1470 PetscErrorCode MatMatTransposeSolve_MUMPS(Mat A,Mat Bt,Mat X) 1471 { 1472 PetscBool flg; 1473 Mat B; 1474 1475 PetscFunctionBegin; 1476 PetscCall(PetscObjectTypeCompareAny((PetscObject)Bt,&flg,MATSEQAIJ,MATMPIAIJ,NULL)); 1477 PetscCheck(flg,PetscObjectComm((PetscObject)Bt),PETSC_ERR_ARG_WRONG,"Matrix Bt must be MATAIJ matrix"); 1478 1479 /* Create B=Bt^T that uses Bt's data structure */ 1480 PetscCall(MatCreateTranspose(Bt,&B)); 1481 1482 PetscCall(MatMatSolve_MUMPS(A,B,X)); 1483 PetscCall(MatDestroy(&B)); 1484 PetscFunctionReturn(0); 1485 } 1486 1487 #if !defined(PETSC_USE_COMPLEX) 1488 /* 1489 input: 1490 F: numeric factor 1491 output: 1492 nneg: total number of negative pivots 1493 nzero: total number of zero pivots 1494 npos: (global dimension of F) - nneg - nzero 1495 */ 1496 PetscErrorCode MatGetInertia_SBAIJMUMPS(Mat F,PetscInt *nneg,PetscInt *nzero,PetscInt *npos) 1497 { 1498 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 1499 PetscMPIInt size; 1500 1501 PetscFunctionBegin; 1502 PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)F),&size)); 1503 /* MUMPS 4.3.1 calls ScaLAPACK when ICNTL(13)=0 (default), which does not offer the possibility to compute the inertia of a dense matrix. Set ICNTL(13)=1 to skip ScaLAPACK */ 1504 PetscCheckFalse(size > 1 && mumps->id.ICNTL(13) != 1,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"ICNTL(13)=%d. -mat_mumps_icntl_13 must be set as 1 for correct global matrix inertia",mumps->id.INFOG(13)); 1505 1506 if (nneg) *nneg = mumps->id.INFOG(12); 1507 if (nzero || npos) { 1508 PetscCheck(mumps->id.ICNTL(24) == 1,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"-mat_mumps_icntl_24 must be set as 1 for null pivot row detection"); 1509 if (nzero) *nzero = mumps->id.INFOG(28); 1510 if (npos) *npos = F->rmap->N - (mumps->id.INFOG(12) + mumps->id.INFOG(28)); 1511 } 1512 PetscFunctionReturn(0); 1513 } 1514 #endif 1515 1516 PetscErrorCode MatMumpsGatherNonzerosOnMaster(MatReuse reuse,Mat_MUMPS *mumps) 1517 { 1518 PetscInt i,nreqs; 1519 PetscMUMPSInt *irn,*jcn; 1520 PetscMPIInt count; 1521 PetscInt64 totnnz,remain; 1522 const PetscInt osize=mumps->omp_comm_size; 1523 PetscScalar *val; 1524 1525 PetscFunctionBegin; 1526 if (osize > 1) { 1527 if (reuse == MAT_INITIAL_MATRIX) { 1528 /* master first gathers counts of nonzeros to receive */ 1529 if (mumps->is_omp_master) PetscCall(PetscMalloc1(osize,&mumps->recvcount)); 1530 PetscCallMPI(MPI_Gather(&mumps->nnz,1,MPIU_INT64,mumps->recvcount,1,MPIU_INT64,0/*master*/,mumps->omp_comm)); 1531 1532 /* Then each computes number of send/recvs */ 1533 if (mumps->is_omp_master) { 1534 /* Start from 1 since self communication is not done in MPI */ 1535 nreqs = 0; 1536 for (i=1; i<osize; i++) nreqs += (mumps->recvcount[i]+PETSC_MPI_INT_MAX-1)/PETSC_MPI_INT_MAX; 1537 } else { 1538 nreqs = (mumps->nnz+PETSC_MPI_INT_MAX-1)/PETSC_MPI_INT_MAX; 1539 } 1540 PetscCall(PetscMalloc1(nreqs*3,&mumps->reqs)); /* Triple the requests since we send irn, jcn and val seperately */ 1541 1542 /* The following code is doing a very simple thing: omp_master rank gathers irn/jcn/val from others. 1543 MPI_Gatherv would be enough if it supports big counts > 2^31-1. Since it does not, and mumps->nnz 1544 might be a prime number > 2^31-1, we have to slice the message. Note omp_comm_size 1545 is very small, the current approach should have no extra overhead compared to MPI_Gatherv. 1546 */ 1547 nreqs = 0; /* counter for actual send/recvs */ 1548 if (mumps->is_omp_master) { 1549 for (i=0,totnnz=0; i<osize; i++) totnnz += mumps->recvcount[i]; /* totnnz = sum of nnz over omp_comm */ 1550 PetscCall(PetscMalloc2(totnnz,&irn,totnnz,&jcn)); 1551 PetscCall(PetscMalloc1(totnnz,&val)); 1552 1553 /* Self communication */ 1554 PetscCall(PetscArraycpy(irn,mumps->irn,mumps->nnz)); 1555 PetscCall(PetscArraycpy(jcn,mumps->jcn,mumps->nnz)); 1556 PetscCall(PetscArraycpy(val,mumps->val,mumps->nnz)); 1557 1558 /* Replace mumps->irn/jcn etc on master with the newly allocated bigger arrays */ 1559 PetscCall(PetscFree2(mumps->irn,mumps->jcn)); 1560 PetscCall(PetscFree(mumps->val_alloc)); 1561 mumps->nnz = totnnz; 1562 mumps->irn = irn; 1563 mumps->jcn = jcn; 1564 mumps->val = mumps->val_alloc = val; 1565 1566 irn += mumps->recvcount[0]; /* recvcount[0] is old mumps->nnz on omp rank 0 */ 1567 jcn += mumps->recvcount[0]; 1568 val += mumps->recvcount[0]; 1569 1570 /* Remote communication */ 1571 for (i=1; i<osize; i++) { 1572 count = PetscMin(mumps->recvcount[i],PETSC_MPI_INT_MAX); 1573 remain = mumps->recvcount[i] - count; 1574 while (count>0) { 1575 PetscCallMPI(MPI_Irecv(irn,count,MPIU_MUMPSINT,i,mumps->tag,mumps->omp_comm,&mumps->reqs[nreqs++])); 1576 PetscCallMPI(MPI_Irecv(jcn,count,MPIU_MUMPSINT,i,mumps->tag,mumps->omp_comm,&mumps->reqs[nreqs++])); 1577 PetscCallMPI(MPI_Irecv(val,count,MPIU_SCALAR, i,mumps->tag,mumps->omp_comm,&mumps->reqs[nreqs++])); 1578 irn += count; 1579 jcn += count; 1580 val += count; 1581 count = PetscMin(remain,PETSC_MPI_INT_MAX); 1582 remain -= count; 1583 } 1584 } 1585 } else { 1586 irn = mumps->irn; 1587 jcn = mumps->jcn; 1588 val = mumps->val; 1589 count = PetscMin(mumps->nnz,PETSC_MPI_INT_MAX); 1590 remain = mumps->nnz - count; 1591 while (count>0) { 1592 PetscCallMPI(MPI_Isend(irn,count,MPIU_MUMPSINT,0,mumps->tag,mumps->omp_comm,&mumps->reqs[nreqs++])); 1593 PetscCallMPI(MPI_Isend(jcn,count,MPIU_MUMPSINT,0,mumps->tag,mumps->omp_comm,&mumps->reqs[nreqs++])); 1594 PetscCallMPI(MPI_Isend(val,count,MPIU_SCALAR, 0,mumps->tag,mumps->omp_comm,&mumps->reqs[nreqs++])); 1595 irn += count; 1596 jcn += count; 1597 val += count; 1598 count = PetscMin(remain,PETSC_MPI_INT_MAX); 1599 remain -= count; 1600 } 1601 } 1602 } else { 1603 nreqs = 0; 1604 if (mumps->is_omp_master) { 1605 val = mumps->val + mumps->recvcount[0]; 1606 for (i=1; i<osize; i++) { /* Remote communication only since self data is already in place */ 1607 count = PetscMin(mumps->recvcount[i],PETSC_MPI_INT_MAX); 1608 remain = mumps->recvcount[i] - count; 1609 while (count>0) { 1610 PetscCallMPI(MPI_Irecv(val,count,MPIU_SCALAR,i,mumps->tag,mumps->omp_comm,&mumps->reqs[nreqs++])); 1611 val += count; 1612 count = PetscMin(remain,PETSC_MPI_INT_MAX); 1613 remain -= count; 1614 } 1615 } 1616 } else { 1617 val = mumps->val; 1618 count = PetscMin(mumps->nnz,PETSC_MPI_INT_MAX); 1619 remain = mumps->nnz - count; 1620 while (count>0) { 1621 PetscCallMPI(MPI_Isend(val,count,MPIU_SCALAR,0,mumps->tag,mumps->omp_comm,&mumps->reqs[nreqs++])); 1622 val += count; 1623 count = PetscMin(remain,PETSC_MPI_INT_MAX); 1624 remain -= count; 1625 } 1626 } 1627 } 1628 PetscCallMPI(MPI_Waitall(nreqs,mumps->reqs,MPI_STATUSES_IGNORE)); 1629 mumps->tag++; /* It is totally fine for above send/recvs to share one mpi tag */ 1630 } 1631 PetscFunctionReturn(0); 1632 } 1633 1634 PetscErrorCode MatFactorNumeric_MUMPS(Mat F,Mat A,const MatFactorInfo *info) 1635 { 1636 Mat_MUMPS *mumps =(Mat_MUMPS*)(F)->data; 1637 PetscBool isMPIAIJ; 1638 1639 PetscFunctionBegin; 1640 if (mumps->id.INFOG(1) < 0 && !(mumps->id.INFOG(1) == -16 && mumps->id.INFOG(1) == 0)) { 1641 if (mumps->id.INFOG(1) == -6) { 1642 PetscCall(PetscInfo(A,"MatFactorNumeric is called with singular matrix structure, INFOG(1)=%d, INFO(2)=%d\n",mumps->id.INFOG(1),mumps->id.INFO(2))); 1643 } 1644 PetscCall(PetscInfo(A,"MatFactorNumeric is called after analysis phase fails, INFOG(1)=%d, INFO(2)=%d\n",mumps->id.INFOG(1),mumps->id.INFO(2))); 1645 PetscFunctionReturn(0); 1646 } 1647 1648 PetscCall((*mumps->ConvertToTriples)(A, 1, MAT_REUSE_MATRIX, mumps)); 1649 PetscCall(MatMumpsGatherNonzerosOnMaster(MAT_REUSE_MATRIX,mumps)); 1650 1651 /* numerical factorization phase */ 1652 /*-------------------------------*/ 1653 mumps->id.job = JOB_FACTNUMERIC; 1654 if (!mumps->id.ICNTL(18)) { /* A is centralized */ 1655 if (!mumps->myid) { 1656 mumps->id.a = (MumpsScalar*)mumps->val; 1657 } 1658 } else { 1659 mumps->id.a_loc = (MumpsScalar*)mumps->val; 1660 } 1661 PetscMUMPS_c(mumps); 1662 if (mumps->id.INFOG(1) < 0) { 1663 if (A->erroriffailure) { 1664 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MUMPS in numerical factorization phase: INFOG(1)=%d, INFO(2)=%d",mumps->id.INFOG(1),mumps->id.INFO(2)); 1665 } else { 1666 if (mumps->id.INFOG(1) == -10) { /* numerically singular matrix */ 1667 PetscCall(PetscInfo(F,"matrix is numerically singular, INFOG(1)=%d, INFO(2)=%d\n",mumps->id.INFOG(1),mumps->id.INFO(2))); 1668 F->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 1669 } else if (mumps->id.INFOG(1) == -13) { 1670 PetscCall(PetscInfo(F,"MUMPS in numerical factorization phase: INFOG(1)=%d, cannot allocate required memory %d megabytes\n",mumps->id.INFOG(1),mumps->id.INFO(2))); 1671 F->factorerrortype = MAT_FACTOR_OUTMEMORY; 1672 } else if (mumps->id.INFOG(1) == -8 || mumps->id.INFOG(1) == -9 || (-16 < mumps->id.INFOG(1) && mumps->id.INFOG(1) < -10)) { 1673 PetscCall(PetscInfo(F,"MUMPS in numerical factorization phase: INFOG(1)=%d, INFO(2)=%d, problem with workarray \n",mumps->id.INFOG(1),mumps->id.INFO(2))); 1674 F->factorerrortype = MAT_FACTOR_OUTMEMORY; 1675 } else { 1676 PetscCall(PetscInfo(F,"MUMPS in numerical factorization phase: INFOG(1)=%d, INFO(2)=%d\n",mumps->id.INFOG(1),mumps->id.INFO(2))); 1677 F->factorerrortype = MAT_FACTOR_OTHER; 1678 } 1679 } 1680 } 1681 PetscCheckFalse(!mumps->myid && mumps->id.ICNTL(16) > 0,PETSC_COMM_SELF,PETSC_ERR_LIB," mumps->id.ICNTL(16):=%d",mumps->id.INFOG(16)); 1682 1683 F->assembled = PETSC_TRUE; 1684 1685 if (F->schur) { /* reset Schur status to unfactored */ 1686 #if defined(PETSC_HAVE_CUDA) 1687 F->schur->offloadmask = PETSC_OFFLOAD_CPU; 1688 #endif 1689 if (mumps->id.ICNTL(19) == 1) { /* stored by rows */ 1690 mumps->id.ICNTL(19) = 2; 1691 PetscCall(MatTranspose(F->schur,MAT_INPLACE_MATRIX,&F->schur)); 1692 } 1693 PetscCall(MatFactorRestoreSchurComplement(F,NULL,MAT_FACTOR_SCHUR_UNFACTORED)); 1694 } 1695 1696 /* just to be sure that ICNTL(19) value returned by a call from MatMumpsGetIcntl is always consistent */ 1697 if (!mumps->sym && mumps->id.ICNTL(19) && mumps->id.ICNTL(19) != 1) mumps->id.ICNTL(19) = 3; 1698 1699 if (!mumps->is_omp_master) mumps->id.INFO(23) = 0; 1700 if (mumps->petsc_size > 1) { 1701 PetscInt lsol_loc; 1702 PetscScalar *sol_loc; 1703 1704 PetscCall(PetscObjectTypeCompare((PetscObject)A,MATMPIAIJ,&isMPIAIJ)); 1705 1706 /* distributed solution; Create x_seq=sol_loc for repeated use */ 1707 if (mumps->x_seq) { 1708 PetscCall(VecScatterDestroy(&mumps->scat_sol)); 1709 PetscCall(PetscFree2(mumps->id.sol_loc,mumps->id.isol_loc)); 1710 PetscCall(VecDestroy(&mumps->x_seq)); 1711 } 1712 lsol_loc = mumps->id.INFO(23); /* length of sol_loc */ 1713 PetscCall(PetscMalloc2(lsol_loc,&sol_loc,lsol_loc,&mumps->id.isol_loc)); 1714 mumps->id.lsol_loc = lsol_loc; 1715 mumps->id.sol_loc = (MumpsScalar*)sol_loc; 1716 PetscCall(VecCreateSeqWithArray(PETSC_COMM_SELF,1,lsol_loc,sol_loc,&mumps->x_seq)); 1717 } 1718 PetscCall(PetscLogFlops(mumps->id.RINFO(2))); 1719 PetscFunctionReturn(0); 1720 } 1721 1722 /* Sets MUMPS options from the options database */ 1723 PetscErrorCode PetscSetMUMPSFromOptions(Mat F, Mat A) 1724 { 1725 Mat_MUMPS *mumps = (Mat_MUMPS*)F->data; 1726 PetscMUMPSInt icntl=0; 1727 PetscInt info[80],i,ninfo=80,rbs,cbs; 1728 PetscBool flg=PETSC_FALSE; 1729 1730 PetscFunctionBegin; 1731 PetscOptionsBegin(PetscObjectComm((PetscObject)A),((PetscObject)A)->prefix,"MUMPS Options","Mat"); 1732 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_1","ICNTL(1): output stream for error messages","None",mumps->id.ICNTL(1),&icntl,&flg)); 1733 if (flg) mumps->id.ICNTL(1) = icntl; 1734 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_2","ICNTL(2): output stream for diagnostic printing, statistics, and warning","None",mumps->id.ICNTL(2),&icntl,&flg)); 1735 if (flg) mumps->id.ICNTL(2) = icntl; 1736 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_3","ICNTL(3): output stream for global information, collected on the host","None",mumps->id.ICNTL(3),&icntl,&flg)); 1737 if (flg) mumps->id.ICNTL(3) = icntl; 1738 1739 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_4","ICNTL(4): level of printing (0 to 4)","None",mumps->id.ICNTL(4),&icntl,&flg)); 1740 if (flg) mumps->id.ICNTL(4) = icntl; 1741 if (mumps->id.ICNTL(4) || PetscLogPrintInfo) mumps->id.ICNTL(3) = 6; /* resume MUMPS default id.ICNTL(3) = 6 */ 1742 1743 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_6","ICNTL(6): permutes to a zero-free diagonal and/or scale the matrix (0 to 7)","None",mumps->id.ICNTL(6),&icntl,&flg)); 1744 if (flg) mumps->id.ICNTL(6) = icntl; 1745 1746 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_7","ICNTL(7): computes a symmetric permutation in sequential analysis. 0=AMD, 2=AMF, 3=Scotch, 4=PORD, 5=Metis, 6=QAMD, and 7=auto(default)","None",mumps->id.ICNTL(7),&icntl,&flg)); 1747 if (flg) { 1748 PetscCheckFalse(icntl == 1 || icntl < 0 || icntl > 7,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"Valid values are 0=AMD, 2=AMF, 3=Scotch, 4=PORD, 5=Metis, 6=QAMD, and 7=auto"); 1749 mumps->id.ICNTL(7) = icntl; 1750 } 1751 1752 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_8","ICNTL(8): scaling strategy (-2 to 8 or 77)","None",mumps->id.ICNTL(8),&mumps->id.ICNTL(8),NULL)); 1753 /* PetscCall(PetscOptionsInt("-mat_mumps_icntl_9","ICNTL(9): computes the solution using A or A^T","None",mumps->id.ICNTL(9),&mumps->id.ICNTL(9),NULL)); handled by MatSolveTranspose_MUMPS() */ 1754 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_10","ICNTL(10): max num of refinements","None",mumps->id.ICNTL(10),&mumps->id.ICNTL(10),NULL)); 1755 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_11","ICNTL(11): statistics related to an error analysis (via -ksp_view)","None",mumps->id.ICNTL(11),&mumps->id.ICNTL(11),NULL)); 1756 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_12","ICNTL(12): an ordering strategy for symmetric matrices (0 to 3)","None",mumps->id.ICNTL(12),&mumps->id.ICNTL(12),NULL)); 1757 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_13","ICNTL(13): parallelism of the root node (enable ScaLAPACK) and its splitting","None",mumps->id.ICNTL(13),&mumps->id.ICNTL(13),NULL)); 1758 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_14","ICNTL(14): percentage increase in the estimated working space","None",mumps->id.ICNTL(14),&mumps->id.ICNTL(14),NULL)); 1759 PetscCall(MatGetBlockSizes(A,&rbs,&cbs)); 1760 if (rbs == cbs && rbs > 1) mumps->id.ICNTL(15) = -rbs; 1761 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_15","ICNTL(15): compression of the input matrix resulting from a block format","None",mumps->id.ICNTL(15),&mumps->id.ICNTL(15),&flg)); 1762 if (flg) { 1763 PetscCheck(mumps->id.ICNTL(15) <= 0,PETSC_COMM_SELF,PETSC_ERR_SUP,"Positive -mat_mumps_icntl_15 not handled"); 1764 PetscCheck((-mumps->id.ICNTL(15) % cbs == 0) && (-mumps->id.ICNTL(15) % rbs == 0),PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"The opposite of -mat_mumps_icntl_15 must be a multiple of the column and row blocksizes"); 1765 } 1766 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_19","ICNTL(19): computes the Schur complement","None",mumps->id.ICNTL(19),&mumps->id.ICNTL(19),NULL)); 1767 if (mumps->id.ICNTL(19) <= 0 || mumps->id.ICNTL(19) > 3) { /* reset any schur data (if any) */ 1768 PetscCall(MatDestroy(&F->schur)); 1769 PetscCall(MatMumpsResetSchur_Private(mumps)); 1770 } 1771 1772 /* Two MPICH Fortran MPI_IN_PLACE binding bugs prevented the use of 'mpich + mumps'. One happened with "mpi4py + mpich + mumps", 1773 and was reported by Firedrake. See https://bitbucket.org/mpi4py/mpi4py/issues/162/mpi4py-initialization-breaks-fortran 1774 and a petsc-maint mailing list thread with subject 'MUMPS segfaults in parallel because of ...' 1775 This bug was fixed by https://github.com/pmodels/mpich/pull/4149. But the fix brought a new bug, 1776 see https://github.com/pmodels/mpich/issues/5589. This bug was fixed by https://github.com/pmodels/mpich/pull/5590. 1777 In short, we could not use distributed RHS with MPICH until v4.0b1. 1778 */ 1779 #if PETSC_PKG_MUMPS_VERSION_LT(5,3,0) || (defined(PETSC_HAVE_MPICH_NUMVERSION) && (PETSC_HAVE_MPICH_NUMVERSION < 40000101)) 1780 mumps->ICNTL20 = 0; /* Centralized dense RHS*/ 1781 #else 1782 mumps->ICNTL20 = 10; /* Distributed dense RHS*/ 1783 #endif 1784 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_20","ICNTL(20): give mumps centralized (0) or distributed (10) dense right-hand sides","None",mumps->ICNTL20,&mumps->ICNTL20,&flg)); 1785 PetscCheckFalse(flg && mumps->ICNTL20 != 10 && mumps->ICNTL20 != 0,PETSC_COMM_SELF,PETSC_ERR_SUP,"ICNTL(20)=%d is not supported by the PETSc/MUMPS interface. Allowed values are 0, 10",(int)mumps->ICNTL20); 1786 #if PETSC_PKG_MUMPS_VERSION_LT(5,3,0) 1787 PetscCheckFalse(flg && mumps->ICNTL20 == 10,PETSC_COMM_SELF,PETSC_ERR_SUP,"ICNTL(20)=10 is not supported before MUMPS-5.3.0"); 1788 #endif 1789 /* PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_21","ICNTL(21): the distribution (centralized or distributed) of the solution vectors","None",mumps->id.ICNTL(21),&mumps->id.ICNTL(21),NULL)); we only use distributed solution vector */ 1790 1791 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_22","ICNTL(22): in-core/out-of-core factorization and solve (0 or 1)","None",mumps->id.ICNTL(22),&mumps->id.ICNTL(22),NULL)); 1792 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_23","ICNTL(23): max size of the working memory (MB) that can allocate per processor","None",mumps->id.ICNTL(23),&mumps->id.ICNTL(23),NULL)); 1793 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_24","ICNTL(24): detection of null pivot rows (0 or 1)","None",mumps->id.ICNTL(24),&mumps->id.ICNTL(24),NULL)); 1794 if (mumps->id.ICNTL(24)) { 1795 mumps->id.ICNTL(13) = 1; /* turn-off ScaLAPACK to help with the correct detection of null pivots */ 1796 } 1797 1798 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_25","ICNTL(25): computes a solution of a deficient matrix and a null space basis","None",mumps->id.ICNTL(25),&mumps->id.ICNTL(25),NULL)); 1799 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_26","ICNTL(26): drives the solution phase if a Schur complement matrix","None",mumps->id.ICNTL(26),&mumps->id.ICNTL(26),NULL)); 1800 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_27","ICNTL(27): controls the blocking size for multiple right-hand sides","None",mumps->id.ICNTL(27),&mumps->id.ICNTL(27),NULL)); 1801 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_28","ICNTL(28): use 1 for sequential analysis and ictnl(7) ordering, or 2 for parallel analysis and ictnl(29) ordering","None",mumps->id.ICNTL(28),&mumps->id.ICNTL(28),NULL)); 1802 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_29","ICNTL(29): parallel ordering 1 = ptscotch, 2 = parmetis","None",mumps->id.ICNTL(29),&mumps->id.ICNTL(29),NULL)); 1803 /* PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_30","ICNTL(30): compute user-specified set of entries in inv(A)","None",mumps->id.ICNTL(30),&mumps->id.ICNTL(30),NULL)); */ /* call MatMumpsGetInverse() directly */ 1804 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_31","ICNTL(31): indicates which factors may be discarded during factorization","None",mumps->id.ICNTL(31),&mumps->id.ICNTL(31),NULL)); 1805 /* PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_32","ICNTL(32): performs the forward elemination of the right-hand sides during factorization","None",mumps->id.ICNTL(32),&mumps->id.ICNTL(32),NULL)); -- not supported by PETSc API */ 1806 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_33","ICNTL(33): compute determinant","None",mumps->id.ICNTL(33),&mumps->id.ICNTL(33),NULL)); 1807 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_35","ICNTL(35): activates Block Low Rank (BLR) based factorization","None",mumps->id.ICNTL(35),&mumps->id.ICNTL(35),NULL)); 1808 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_36","ICNTL(36): choice of BLR factorization variant","None",mumps->id.ICNTL(36),&mumps->id.ICNTL(36),NULL)); 1809 PetscCall(PetscOptionsMUMPSInt("-mat_mumps_icntl_38","ICNTL(38): estimated compression rate of LU factors with BLR","None",mumps->id.ICNTL(38),&mumps->id.ICNTL(38),NULL)); 1810 1811 PetscCall(PetscOptionsReal("-mat_mumps_cntl_1","CNTL(1): relative pivoting threshold","None",mumps->id.CNTL(1),&mumps->id.CNTL(1),NULL)); 1812 PetscCall(PetscOptionsReal("-mat_mumps_cntl_2","CNTL(2): stopping criterion of refinement","None",mumps->id.CNTL(2),&mumps->id.CNTL(2),NULL)); 1813 PetscCall(PetscOptionsReal("-mat_mumps_cntl_3","CNTL(3): absolute pivoting threshold","None",mumps->id.CNTL(3),&mumps->id.CNTL(3),NULL)); 1814 PetscCall(PetscOptionsReal("-mat_mumps_cntl_4","CNTL(4): value for static pivoting","None",mumps->id.CNTL(4),&mumps->id.CNTL(4),NULL)); 1815 PetscCall(PetscOptionsReal("-mat_mumps_cntl_5","CNTL(5): fixation for null pivots","None",mumps->id.CNTL(5),&mumps->id.CNTL(5),NULL)); 1816 PetscCall(PetscOptionsReal("-mat_mumps_cntl_7","CNTL(7): dropping parameter used during BLR","None",mumps->id.CNTL(7),&mumps->id.CNTL(7),NULL)); 1817 1818 PetscCall(PetscOptionsString("-mat_mumps_ooc_tmpdir", "out of core directory", "None", mumps->id.ooc_tmpdir, mumps->id.ooc_tmpdir, sizeof(mumps->id.ooc_tmpdir), NULL)); 1819 1820 PetscCall(PetscOptionsIntArray("-mat_mumps_view_info","request INFO local to each processor","",info,&ninfo,NULL)); 1821 if (ninfo) { 1822 PetscCheck(ninfo <= 80,PETSC_COMM_SELF,PETSC_ERR_USER,"number of INFO %" PetscInt_FMT " must <= 80",ninfo); 1823 PetscCall(PetscMalloc1(ninfo,&mumps->info)); 1824 mumps->ninfo = ninfo; 1825 for (i=0; i<ninfo; i++) { 1826 PetscCheckFalse(info[i] < 0 || info[i]>80,PETSC_COMM_SELF,PETSC_ERR_USER,"index of INFO %" PetscInt_FMT " must between 1 and 80",ninfo); 1827 else mumps->info[i] = info[i]; 1828 } 1829 } 1830 1831 PetscOptionsEnd(); 1832 PetscFunctionReturn(0); 1833 } 1834 1835 PetscErrorCode PetscInitializeMUMPS(Mat A,Mat_MUMPS *mumps) 1836 { 1837 PetscInt nthreads=0; 1838 1839 PetscFunctionBegin; 1840 mumps->petsc_comm = PetscObjectComm((PetscObject)A); 1841 PetscCallMPI(MPI_Comm_size(mumps->petsc_comm,&mumps->petsc_size)); 1842 PetscCallMPI(MPI_Comm_rank(mumps->petsc_comm,&mumps->myid));/* "if (!myid)" still works even if mumps_comm is different */ 1843 1844 PetscCall(PetscOptionsHasName(NULL,((PetscObject)A)->prefix,"-mat_mumps_use_omp_threads",&mumps->use_petsc_omp_support)); 1845 if (mumps->use_petsc_omp_support) nthreads = -1; /* -1 will let PetscOmpCtrlCreate() guess a proper value when user did not supply one */ 1846 PetscCall(PetscOptionsGetInt(NULL,((PetscObject)A)->prefix,"-mat_mumps_use_omp_threads",&nthreads,NULL)); 1847 if (mumps->use_petsc_omp_support) { 1848 #if defined(PETSC_HAVE_OPENMP_SUPPORT) 1849 PetscCall(PetscOmpCtrlCreate(mumps->petsc_comm,nthreads,&mumps->omp_ctrl)); 1850 PetscCall(PetscOmpCtrlGetOmpComms(mumps->omp_ctrl,&mumps->omp_comm,&mumps->mumps_comm,&mumps->is_omp_master)); 1851 #else 1852 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP_SYS,"the system does not have PETSc OpenMP support but you added the -%smat_mumps_use_omp_threads option. Configure PETSc with --with-openmp --download-hwloc (or --with-hwloc) to enable it, see more in MATSOLVERMUMPS manual",((PetscObject)A)->prefix?((PetscObject)A)->prefix:""); 1853 #endif 1854 } else { 1855 mumps->omp_comm = PETSC_COMM_SELF; 1856 mumps->mumps_comm = mumps->petsc_comm; 1857 mumps->is_omp_master = PETSC_TRUE; 1858 } 1859 PetscCallMPI(MPI_Comm_size(mumps->omp_comm,&mumps->omp_comm_size)); 1860 mumps->reqs = NULL; 1861 mumps->tag = 0; 1862 1863 /* It looks like MUMPS does not dup the input comm. Dup a new comm for MUMPS to avoid any tag mismatches. */ 1864 if (mumps->mumps_comm != MPI_COMM_NULL) { 1865 PetscCall(PetscCommGetComm(PetscObjectComm((PetscObject)A),&mumps->mumps_comm)); 1866 } 1867 1868 mumps->id.comm_fortran = MPI_Comm_c2f(mumps->mumps_comm); 1869 mumps->id.job = JOB_INIT; 1870 mumps->id.par = 1; /* host participates factorizaton and solve */ 1871 mumps->id.sym = mumps->sym; 1872 1873 PetscMUMPS_c(mumps); 1874 PetscCheck(mumps->id.INFOG(1) >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MUMPS in PetscInitializeMUMPS: INFOG(1)=%d",mumps->id.INFOG(1)); 1875 1876 /* copy MUMPS default control values from master to slaves. Although slaves do not call MUMPS, they may access these values in code. 1877 For example, ICNTL(9) is initialized to 1 by MUMPS and slaves check ICNTL(9) in MatSolve_MUMPS. 1878 */ 1879 PetscCallMPI(MPI_Bcast(mumps->id.icntl,40,MPI_INT, 0,mumps->omp_comm)); 1880 PetscCallMPI(MPI_Bcast(mumps->id.cntl, 15,MPIU_REAL,0,mumps->omp_comm)); 1881 1882 mumps->scat_rhs = NULL; 1883 mumps->scat_sol = NULL; 1884 1885 /* set PETSc-MUMPS default options - override MUMPS default */ 1886 mumps->id.ICNTL(3) = 0; 1887 mumps->id.ICNTL(4) = 0; 1888 if (mumps->petsc_size == 1) { 1889 mumps->id.ICNTL(18) = 0; /* centralized assembled matrix input */ 1890 mumps->id.ICNTL(7) = 7; /* automatic choice of ordering done by the package */ 1891 } else { 1892 mumps->id.ICNTL(18) = 3; /* distributed assembled matrix input */ 1893 mumps->id.ICNTL(21) = 1; /* distributed solution */ 1894 } 1895 1896 /* schur */ 1897 mumps->id.size_schur = 0; 1898 mumps->id.listvar_schur = NULL; 1899 mumps->id.schur = NULL; 1900 mumps->sizeredrhs = 0; 1901 mumps->schur_sol = NULL; 1902 mumps->schur_sizesol = 0; 1903 PetscFunctionReturn(0); 1904 } 1905 1906 PetscErrorCode MatFactorSymbolic_MUMPS_ReportIfError(Mat F,Mat A,const MatFactorInfo *info,Mat_MUMPS *mumps) 1907 { 1908 PetscFunctionBegin; 1909 if (mumps->id.INFOG(1) < 0) { 1910 if (A->erroriffailure) { 1911 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MUMPS in analysis phase: INFOG(1)=%d",mumps->id.INFOG(1)); 1912 } else { 1913 if (mumps->id.INFOG(1) == -6) { 1914 PetscCall(PetscInfo(F,"matrix is singular in structure, INFOG(1)=%d, INFO(2)=%d\n",mumps->id.INFOG(1),mumps->id.INFO(2))); 1915 F->factorerrortype = MAT_FACTOR_STRUCT_ZEROPIVOT; 1916 } else if (mumps->id.INFOG(1) == -5 || mumps->id.INFOG(1) == -7) { 1917 PetscCall(PetscInfo(F,"problem of workspace, INFOG(1)=%d, INFO(2)=%d\n",mumps->id.INFOG(1),mumps->id.INFO(2))); 1918 F->factorerrortype = MAT_FACTOR_OUTMEMORY; 1919 } else if (mumps->id.INFOG(1) == -16 && mumps->id.INFOG(1) == 0) { 1920 PetscCall(PetscInfo(F,"Empty matrix\n")); 1921 } else { 1922 PetscCall(PetscInfo(F,"Error reported by MUMPS in analysis phase: INFOG(1)=%d, INFO(2)=%d\n",mumps->id.INFOG(1),mumps->id.INFO(2))); 1923 F->factorerrortype = MAT_FACTOR_OTHER; 1924 } 1925 } 1926 } 1927 PetscFunctionReturn(0); 1928 } 1929 1930 PetscErrorCode MatLUFactorSymbolic_AIJMUMPS(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info) 1931 { 1932 Mat_MUMPS *mumps = (Mat_MUMPS*)F->data; 1933 Vec b; 1934 const PetscInt M = A->rmap->N; 1935 1936 PetscFunctionBegin; 1937 if (mumps->matstruc == SAME_NONZERO_PATTERN) { 1938 /* F is assembled by a previous call of MatLUFactorSymbolic_AIJMUMPS() */ 1939 PetscFunctionReturn(0); 1940 } 1941 1942 /* Set MUMPS options from the options database */ 1943 PetscCall(PetscSetMUMPSFromOptions(F,A)); 1944 1945 PetscCall((*mumps->ConvertToTriples)(A, 1, MAT_INITIAL_MATRIX, mumps)); 1946 PetscCall(MatMumpsGatherNonzerosOnMaster(MAT_INITIAL_MATRIX,mumps)); 1947 1948 /* analysis phase */ 1949 /*----------------*/ 1950 mumps->id.job = JOB_FACTSYMBOLIC; 1951 mumps->id.n = M; 1952 switch (mumps->id.ICNTL(18)) { 1953 case 0: /* centralized assembled matrix input */ 1954 if (!mumps->myid) { 1955 mumps->id.nnz = mumps->nnz; 1956 mumps->id.irn = mumps->irn; 1957 mumps->id.jcn = mumps->jcn; 1958 if (mumps->id.ICNTL(6)>1) mumps->id.a = (MumpsScalar*)mumps->val; 1959 if (r) { 1960 mumps->id.ICNTL(7) = 1; 1961 if (!mumps->myid) { 1962 const PetscInt *idx; 1963 PetscInt i; 1964 1965 PetscCall(PetscMalloc1(M,&mumps->id.perm_in)); 1966 PetscCall(ISGetIndices(r,&idx)); 1967 for (i=0; i<M; i++) PetscCall(PetscMUMPSIntCast(idx[i]+1,&(mumps->id.perm_in[i]))); /* perm_in[]: start from 1, not 0! */ 1968 PetscCall(ISRestoreIndices(r,&idx)); 1969 } 1970 } 1971 } 1972 break; 1973 case 3: /* distributed assembled matrix input (size>1) */ 1974 mumps->id.nnz_loc = mumps->nnz; 1975 mumps->id.irn_loc = mumps->irn; 1976 mumps->id.jcn_loc = mumps->jcn; 1977 if (mumps->id.ICNTL(6)>1) mumps->id.a_loc = (MumpsScalar*)mumps->val; 1978 if (mumps->ICNTL20 == 0) { /* Centralized rhs. Create scatter scat_rhs for repeated use in MatSolve() */ 1979 PetscCall(MatCreateVecs(A,NULL,&b)); 1980 PetscCall(VecScatterCreateToZero(b,&mumps->scat_rhs,&mumps->b_seq)); 1981 PetscCall(VecDestroy(&b)); 1982 } 1983 break; 1984 } 1985 PetscMUMPS_c(mumps); 1986 PetscCall(MatFactorSymbolic_MUMPS_ReportIfError(F,A,info,mumps)); 1987 1988 F->ops->lufactornumeric = MatFactorNumeric_MUMPS; 1989 F->ops->solve = MatSolve_MUMPS; 1990 F->ops->solvetranspose = MatSolveTranspose_MUMPS; 1991 F->ops->matsolve = MatMatSolve_MUMPS; 1992 F->ops->mattransposesolve = MatMatTransposeSolve_MUMPS; 1993 1994 mumps->matstruc = SAME_NONZERO_PATTERN; 1995 PetscFunctionReturn(0); 1996 } 1997 1998 /* Note the Petsc r and c permutations are ignored */ 1999 PetscErrorCode MatLUFactorSymbolic_BAIJMUMPS(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info) 2000 { 2001 Mat_MUMPS *mumps = (Mat_MUMPS*)F->data; 2002 Vec b; 2003 const PetscInt M = A->rmap->N; 2004 2005 PetscFunctionBegin; 2006 if (mumps->matstruc == SAME_NONZERO_PATTERN) { 2007 /* F is assembled by a previous call of MatLUFactorSymbolic_AIJMUMPS() */ 2008 PetscFunctionReturn(0); 2009 } 2010 2011 /* Set MUMPS options from the options database */ 2012 PetscCall(PetscSetMUMPSFromOptions(F,A)); 2013 2014 PetscCall((*mumps->ConvertToTriples)(A, 1, MAT_INITIAL_MATRIX, mumps)); 2015 PetscCall(MatMumpsGatherNonzerosOnMaster(MAT_INITIAL_MATRIX,mumps)); 2016 2017 /* analysis phase */ 2018 /*----------------*/ 2019 mumps->id.job = JOB_FACTSYMBOLIC; 2020 mumps->id.n = M; 2021 switch (mumps->id.ICNTL(18)) { 2022 case 0: /* centralized assembled matrix input */ 2023 if (!mumps->myid) { 2024 mumps->id.nnz = mumps->nnz; 2025 mumps->id.irn = mumps->irn; 2026 mumps->id.jcn = mumps->jcn; 2027 if (mumps->id.ICNTL(6)>1) { 2028 mumps->id.a = (MumpsScalar*)mumps->val; 2029 } 2030 } 2031 break; 2032 case 3: /* distributed assembled matrix input (size>1) */ 2033 mumps->id.nnz_loc = mumps->nnz; 2034 mumps->id.irn_loc = mumps->irn; 2035 mumps->id.jcn_loc = mumps->jcn; 2036 if (mumps->id.ICNTL(6)>1) { 2037 mumps->id.a_loc = (MumpsScalar*)mumps->val; 2038 } 2039 if (mumps->ICNTL20 == 0) { /* Centralized rhs. Create scatter scat_rhs for repeated use in MatSolve() */ 2040 PetscCall(MatCreateVecs(A,NULL,&b)); 2041 PetscCall(VecScatterCreateToZero(b,&mumps->scat_rhs,&mumps->b_seq)); 2042 PetscCall(VecDestroy(&b)); 2043 } 2044 break; 2045 } 2046 PetscMUMPS_c(mumps); 2047 PetscCall(MatFactorSymbolic_MUMPS_ReportIfError(F,A,info,mumps)); 2048 2049 F->ops->lufactornumeric = MatFactorNumeric_MUMPS; 2050 F->ops->solve = MatSolve_MUMPS; 2051 F->ops->solvetranspose = MatSolveTranspose_MUMPS; 2052 2053 mumps->matstruc = SAME_NONZERO_PATTERN; 2054 PetscFunctionReturn(0); 2055 } 2056 2057 /* Note the Petsc r permutation and factor info are ignored */ 2058 PetscErrorCode MatCholeskyFactorSymbolic_MUMPS(Mat F,Mat A,IS r,const MatFactorInfo *info) 2059 { 2060 Mat_MUMPS *mumps = (Mat_MUMPS*)F->data; 2061 Vec b; 2062 const PetscInt M = A->rmap->N; 2063 2064 PetscFunctionBegin; 2065 if (mumps->matstruc == SAME_NONZERO_PATTERN) { 2066 /* F is assembled by a previous call of MatLUFactorSymbolic_AIJMUMPS() */ 2067 PetscFunctionReturn(0); 2068 } 2069 2070 /* Set MUMPS options from the options database */ 2071 PetscCall(PetscSetMUMPSFromOptions(F,A)); 2072 2073 PetscCall((*mumps->ConvertToTriples)(A, 1, MAT_INITIAL_MATRIX, mumps)); 2074 PetscCall(MatMumpsGatherNonzerosOnMaster(MAT_INITIAL_MATRIX,mumps)); 2075 2076 /* analysis phase */ 2077 /*----------------*/ 2078 mumps->id.job = JOB_FACTSYMBOLIC; 2079 mumps->id.n = M; 2080 switch (mumps->id.ICNTL(18)) { 2081 case 0: /* centralized assembled matrix input */ 2082 if (!mumps->myid) { 2083 mumps->id.nnz = mumps->nnz; 2084 mumps->id.irn = mumps->irn; 2085 mumps->id.jcn = mumps->jcn; 2086 if (mumps->id.ICNTL(6)>1) { 2087 mumps->id.a = (MumpsScalar*)mumps->val; 2088 } 2089 } 2090 break; 2091 case 3: /* distributed assembled matrix input (size>1) */ 2092 mumps->id.nnz_loc = mumps->nnz; 2093 mumps->id.irn_loc = mumps->irn; 2094 mumps->id.jcn_loc = mumps->jcn; 2095 if (mumps->id.ICNTL(6)>1) { 2096 mumps->id.a_loc = (MumpsScalar*)mumps->val; 2097 } 2098 if (mumps->ICNTL20 == 0) { /* Centralized rhs. Create scatter scat_rhs for repeated use in MatSolve() */ 2099 PetscCall(MatCreateVecs(A,NULL,&b)); 2100 PetscCall(VecScatterCreateToZero(b,&mumps->scat_rhs,&mumps->b_seq)); 2101 PetscCall(VecDestroy(&b)); 2102 } 2103 break; 2104 } 2105 PetscMUMPS_c(mumps); 2106 PetscCall(MatFactorSymbolic_MUMPS_ReportIfError(F,A,info,mumps)); 2107 2108 F->ops->choleskyfactornumeric = MatFactorNumeric_MUMPS; 2109 F->ops->solve = MatSolve_MUMPS; 2110 F->ops->solvetranspose = MatSolve_MUMPS; 2111 F->ops->matsolve = MatMatSolve_MUMPS; 2112 F->ops->mattransposesolve = MatMatTransposeSolve_MUMPS; 2113 #if defined(PETSC_USE_COMPLEX) 2114 F->ops->getinertia = NULL; 2115 #else 2116 F->ops->getinertia = MatGetInertia_SBAIJMUMPS; 2117 #endif 2118 2119 mumps->matstruc = SAME_NONZERO_PATTERN; 2120 PetscFunctionReturn(0); 2121 } 2122 2123 PetscErrorCode MatView_MUMPS(Mat A,PetscViewer viewer) 2124 { 2125 PetscBool iascii; 2126 PetscViewerFormat format; 2127 Mat_MUMPS *mumps=(Mat_MUMPS*)A->data; 2128 2129 PetscFunctionBegin; 2130 /* check if matrix is mumps type */ 2131 if (A->ops->solve != MatSolve_MUMPS) PetscFunctionReturn(0); 2132 2133 PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii)); 2134 if (iascii) { 2135 PetscCall(PetscViewerGetFormat(viewer,&format)); 2136 if (format == PETSC_VIEWER_ASCII_INFO) { 2137 PetscCall(PetscViewerASCIIPrintf(viewer,"MUMPS run parameters:\n")); 2138 PetscCall(PetscViewerASCIIPrintf(viewer," SYM (matrix type): %d\n",mumps->id.sym)); 2139 PetscCall(PetscViewerASCIIPrintf(viewer," PAR (host participation): %d\n",mumps->id.par)); 2140 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(1) (output for error): %d\n",mumps->id.ICNTL(1))); 2141 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(2) (output of diagnostic msg): %d\n",mumps->id.ICNTL(2))); 2142 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(3) (output for global info): %d\n",mumps->id.ICNTL(3))); 2143 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(4) (level of printing): %d\n",mumps->id.ICNTL(4))); 2144 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(5) (input mat struct): %d\n",mumps->id.ICNTL(5))); 2145 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(6) (matrix prescaling): %d\n",mumps->id.ICNTL(6))); 2146 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(7) (sequential matrix ordering):%d\n",mumps->id.ICNTL(7))); 2147 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(8) (scaling strategy): %d\n",mumps->id.ICNTL(8))); 2148 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(10) (max num of refinements): %d\n",mumps->id.ICNTL(10))); 2149 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(11) (error analysis): %d\n",mumps->id.ICNTL(11))); 2150 if (mumps->id.ICNTL(11)>0) { 2151 PetscCall(PetscViewerASCIIPrintf(viewer," RINFOG(4) (inf norm of input mat): %g\n",mumps->id.RINFOG(4))); 2152 PetscCall(PetscViewerASCIIPrintf(viewer," RINFOG(5) (inf norm of solution): %g\n",mumps->id.RINFOG(5))); 2153 PetscCall(PetscViewerASCIIPrintf(viewer," RINFOG(6) (inf norm of residual): %g\n",mumps->id.RINFOG(6))); 2154 PetscCall(PetscViewerASCIIPrintf(viewer," RINFOG(7),RINFOG(8) (backward error est): %g, %g\n",mumps->id.RINFOG(7),mumps->id.RINFOG(8))); 2155 PetscCall(PetscViewerASCIIPrintf(viewer," RINFOG(9) (error estimate): %g \n",mumps->id.RINFOG(9))); 2156 PetscCall(PetscViewerASCIIPrintf(viewer," RINFOG(10),RINFOG(11)(condition numbers): %g, %g\n",mumps->id.RINFOG(10),mumps->id.RINFOG(11))); 2157 } 2158 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(12) (efficiency control): %d\n",mumps->id.ICNTL(12))); 2159 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(13) (sequential factorization of the root node): %d\n",mumps->id.ICNTL(13))); 2160 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(14) (percentage of estimated workspace increase): %d\n",mumps->id.ICNTL(14))); 2161 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(15) (compression of the input matrix): %d\n",mumps->id.ICNTL(15))); 2162 /* ICNTL(15-17) not used */ 2163 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(18) (input mat struct): %d\n",mumps->id.ICNTL(18))); 2164 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(19) (Schur complement info): %d\n",mumps->id.ICNTL(19))); 2165 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(20) (RHS sparse pattern): %d\n",mumps->id.ICNTL(20))); 2166 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(21) (solution struct): %d\n",mumps->id.ICNTL(21))); 2167 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(22) (in-core/out-of-core facility): %d\n",mumps->id.ICNTL(22))); 2168 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(23) (max size of memory can be allocated locally):%d\n",mumps->id.ICNTL(23))); 2169 2170 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(24) (detection of null pivot rows): %d\n",mumps->id.ICNTL(24))); 2171 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(25) (computation of a null space basis): %d\n",mumps->id.ICNTL(25))); 2172 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(26) (Schur options for RHS or solution): %d\n",mumps->id.ICNTL(26))); 2173 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(27) (blocking size for multiple RHS): %d\n",mumps->id.ICNTL(27))); 2174 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(28) (use parallel or sequential ordering): %d\n",mumps->id.ICNTL(28))); 2175 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(29) (parallel ordering): %d\n",mumps->id.ICNTL(29))); 2176 2177 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(30) (user-specified set of entries in inv(A)): %d\n",mumps->id.ICNTL(30))); 2178 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(31) (factors is discarded in the solve phase): %d\n",mumps->id.ICNTL(31))); 2179 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(33) (compute determinant): %d\n",mumps->id.ICNTL(33))); 2180 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(35) (activate BLR based factorization): %d\n",mumps->id.ICNTL(35))); 2181 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(36) (choice of BLR factorization variant): %d\n",mumps->id.ICNTL(36))); 2182 PetscCall(PetscViewerASCIIPrintf(viewer," ICNTL(38) (estimated compression rate of LU factors): %d\n",mumps->id.ICNTL(38))); 2183 2184 PetscCall(PetscViewerASCIIPrintf(viewer," CNTL(1) (relative pivoting threshold): %g \n",mumps->id.CNTL(1))); 2185 PetscCall(PetscViewerASCIIPrintf(viewer," CNTL(2) (stopping criterion of refinement): %g \n",mumps->id.CNTL(2))); 2186 PetscCall(PetscViewerASCIIPrintf(viewer," CNTL(3) (absolute pivoting threshold): %g \n",mumps->id.CNTL(3))); 2187 PetscCall(PetscViewerASCIIPrintf(viewer," CNTL(4) (value of static pivoting): %g \n",mumps->id.CNTL(4))); 2188 PetscCall(PetscViewerASCIIPrintf(viewer," CNTL(5) (fixation for null pivots): %g \n",mumps->id.CNTL(5))); 2189 PetscCall(PetscViewerASCIIPrintf(viewer," CNTL(7) (dropping parameter for BLR): %g \n",mumps->id.CNTL(7))); 2190 2191 /* information local to each processor */ 2192 PetscCall(PetscViewerASCIIPrintf(viewer, " RINFO(1) (local estimated flops for the elimination after analysis): \n")); 2193 PetscCall(PetscViewerASCIIPushSynchronized(viewer)); 2194 PetscCall(PetscViewerASCIISynchronizedPrintf(viewer," [%d] %g \n",mumps->myid,mumps->id.RINFO(1))); 2195 PetscCall(PetscViewerFlush(viewer)); 2196 PetscCall(PetscViewerASCIIPrintf(viewer, " RINFO(2) (local estimated flops for the assembly after factorization): \n")); 2197 PetscCall(PetscViewerASCIISynchronizedPrintf(viewer," [%d] %g \n",mumps->myid,mumps->id.RINFO(2))); 2198 PetscCall(PetscViewerFlush(viewer)); 2199 PetscCall(PetscViewerASCIIPrintf(viewer, " RINFO(3) (local estimated flops for the elimination after factorization): \n")); 2200 PetscCall(PetscViewerASCIISynchronizedPrintf(viewer," [%d] %g \n",mumps->myid,mumps->id.RINFO(3))); 2201 PetscCall(PetscViewerFlush(viewer)); 2202 2203 PetscCall(PetscViewerASCIIPrintf(viewer, " INFO(15) (estimated size of (in MB) MUMPS internal data for running numerical factorization): \n")); 2204 PetscCall(PetscViewerASCIISynchronizedPrintf(viewer," [%d] %d\n",mumps->myid,mumps->id.INFO(15))); 2205 PetscCall(PetscViewerFlush(viewer)); 2206 2207 PetscCall(PetscViewerASCIIPrintf(viewer, " INFO(16) (size of (in MB) MUMPS internal data used during numerical factorization): \n")); 2208 PetscCall(PetscViewerASCIISynchronizedPrintf(viewer," [%d] %d\n",mumps->myid,mumps->id.INFO(16))); 2209 PetscCall(PetscViewerFlush(viewer)); 2210 2211 PetscCall(PetscViewerASCIIPrintf(viewer, " INFO(23) (num of pivots eliminated on this processor after factorization): \n")); 2212 PetscCall(PetscViewerASCIISynchronizedPrintf(viewer," [%d] %d\n",mumps->myid,mumps->id.INFO(23))); 2213 PetscCall(PetscViewerFlush(viewer)); 2214 2215 if (mumps->ninfo && mumps->ninfo <= 80) { 2216 PetscInt i; 2217 for (i=0; i<mumps->ninfo; i++) { 2218 PetscCall(PetscViewerASCIIPrintf(viewer, " INFO(%" PetscInt_FMT "): \n",mumps->info[i])); 2219 PetscCall(PetscViewerASCIISynchronizedPrintf(viewer," [%d] %d\n",mumps->myid,mumps->id.INFO(mumps->info[i]))); 2220 PetscCall(PetscViewerFlush(viewer)); 2221 } 2222 } 2223 PetscCall(PetscViewerASCIIPopSynchronized(viewer)); 2224 2225 if (!mumps->myid) { /* information from the host */ 2226 PetscCall(PetscViewerASCIIPrintf(viewer," RINFOG(1) (global estimated flops for the elimination after analysis): %g \n",mumps->id.RINFOG(1))); 2227 PetscCall(PetscViewerASCIIPrintf(viewer," RINFOG(2) (global estimated flops for the assembly after factorization): %g \n",mumps->id.RINFOG(2))); 2228 PetscCall(PetscViewerASCIIPrintf(viewer," RINFOG(3) (global estimated flops for the elimination after factorization): %g \n",mumps->id.RINFOG(3))); 2229 PetscCall(PetscViewerASCIIPrintf(viewer," (RINFOG(12) RINFOG(13))*2^INFOG(34) (determinant): (%g,%g)*(2^%d)\n",mumps->id.RINFOG(12),mumps->id.RINFOG(13),mumps->id.INFOG(34))); 2230 2231 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(3) (estimated real workspace for factors on all processors after analysis): %d\n",mumps->id.INFOG(3))); 2232 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(4) (estimated integer workspace for factors on all processors after analysis): %d\n",mumps->id.INFOG(4))); 2233 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(5) (estimated maximum front size in the complete tree): %d\n",mumps->id.INFOG(5))); 2234 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(6) (number of nodes in the complete tree): %d\n",mumps->id.INFOG(6))); 2235 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(7) (ordering option effectively used after analysis): %d\n",mumps->id.INFOG(7))); 2236 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(8) (structural symmetry in percent of the permuted matrix after analysis): %d\n",mumps->id.INFOG(8))); 2237 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(9) (total real/complex workspace to store the matrix factors after factorization): %d\n",mumps->id.INFOG(9))); 2238 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(10) (total integer space store the matrix factors after factorization): %d\n",mumps->id.INFOG(10))); 2239 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(11) (order of largest frontal matrix after factorization): %d\n",mumps->id.INFOG(11))); 2240 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(12) (number of off-diagonal pivots): %d\n",mumps->id.INFOG(12))); 2241 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(13) (number of delayed pivots after factorization): %d\n",mumps->id.INFOG(13))); 2242 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(14) (number of memory compress after factorization): %d\n",mumps->id.INFOG(14))); 2243 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(15) (number of steps of iterative refinement after solution): %d\n",mumps->id.INFOG(15))); 2244 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(16) (estimated size (in MB) of all MUMPS internal data for factorization after analysis: value on the most memory consuming processor): %d\n",mumps->id.INFOG(16))); 2245 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(17) (estimated size of all MUMPS internal data for factorization after analysis: sum over all processors): %d\n",mumps->id.INFOG(17))); 2246 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(18) (size of all MUMPS internal data allocated during factorization: value on the most memory consuming processor): %d\n",mumps->id.INFOG(18))); 2247 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(19) (size of all MUMPS internal data allocated during factorization: sum over all processors): %d\n",mumps->id.INFOG(19))); 2248 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(20) (estimated number of entries in the factors): %d\n",mumps->id.INFOG(20))); 2249 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(21) (size in MB of memory effectively used during factorization - value on the most memory consuming processor): %d\n",mumps->id.INFOG(21))); 2250 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(22) (size in MB of memory effectively used during factorization - sum over all processors): %d\n",mumps->id.INFOG(22))); 2251 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(23) (after analysis: value of ICNTL(6) effectively used): %d\n",mumps->id.INFOG(23))); 2252 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(24) (after analysis: value of ICNTL(12) effectively used): %d\n",mumps->id.INFOG(24))); 2253 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(25) (after factorization: number of pivots modified by static pivoting): %d\n",mumps->id.INFOG(25))); 2254 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(28) (after factorization: number of null pivots encountered): %d\n",mumps->id.INFOG(28))); 2255 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(29) (after factorization: effective number of entries in the factors (sum over all processors)): %d\n",mumps->id.INFOG(29))); 2256 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(30, 31) (after solution: size in Mbytes of memory used during solution phase): %d, %d\n",mumps->id.INFOG(30),mumps->id.INFOG(31))); 2257 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(32) (after analysis: type of analysis done): %d\n",mumps->id.INFOG(32))); 2258 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(33) (value used for ICNTL(8)): %d\n",mumps->id.INFOG(33))); 2259 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(34) (exponent of the determinant if determinant is requested): %d\n",mumps->id.INFOG(34))); 2260 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(35) (after factorization: number of entries taking into account BLR factor compression - sum over all processors): %d\n",mumps->id.INFOG(35))); 2261 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(36) (after analysis: estimated size of all MUMPS internal data for running BLR in-core - value on the most memory consuming processor): %d\n",mumps->id.INFOG(36))); 2262 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(37) (after analysis: estimated size of all MUMPS internal data for running BLR in-core - sum over all processors): %d\n",mumps->id.INFOG(37))); 2263 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(38) (after analysis: estimated size of all MUMPS internal data for running BLR out-of-core - value on the most memory consuming processor): %d\n",mumps->id.INFOG(38))); 2264 PetscCall(PetscViewerASCIIPrintf(viewer," INFOG(39) (after analysis: estimated size of all MUMPS internal data for running BLR out-of-core - sum over all processors): %d\n",mumps->id.INFOG(39))); 2265 } 2266 } 2267 } 2268 PetscFunctionReturn(0); 2269 } 2270 2271 PetscErrorCode MatGetInfo_MUMPS(Mat A,MatInfoType flag,MatInfo *info) 2272 { 2273 Mat_MUMPS *mumps =(Mat_MUMPS*)A->data; 2274 2275 PetscFunctionBegin; 2276 info->block_size = 1.0; 2277 info->nz_allocated = mumps->id.INFOG(20); 2278 info->nz_used = mumps->id.INFOG(20); 2279 info->nz_unneeded = 0.0; 2280 info->assemblies = 0.0; 2281 info->mallocs = 0.0; 2282 info->memory = 0.0; 2283 info->fill_ratio_given = 0; 2284 info->fill_ratio_needed = 0; 2285 info->factor_mallocs = 0; 2286 PetscFunctionReturn(0); 2287 } 2288 2289 /* -------------------------------------------------------------------------------------------*/ 2290 PetscErrorCode MatFactorSetSchurIS_MUMPS(Mat F, IS is) 2291 { 2292 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 2293 const PetscScalar *arr; 2294 const PetscInt *idxs; 2295 PetscInt size,i; 2296 2297 PetscFunctionBegin; 2298 PetscCall(ISGetLocalSize(is,&size)); 2299 if (mumps->petsc_size > 1) { 2300 PetscBool ls,gs; /* gs is false if any rank other than root has non-empty IS */ 2301 2302 ls = mumps->myid ? (size ? PETSC_FALSE : PETSC_TRUE) : PETSC_TRUE; /* always true on root; false on others if their size != 0 */ 2303 PetscCallMPI(MPI_Allreduce(&ls,&gs,1,MPIU_BOOL,MPI_LAND,mumps->petsc_comm)); 2304 PetscCheck(gs,PETSC_COMM_SELF,PETSC_ERR_SUP,"MUMPS distributed parallel Schur complements not yet supported from PETSc"); 2305 } 2306 2307 /* Schur complement matrix */ 2308 PetscCall(MatDestroy(&F->schur)); 2309 PetscCall(MatCreateSeqDense(PETSC_COMM_SELF,size,size,NULL,&F->schur)); 2310 PetscCall(MatDenseGetArrayRead(F->schur,&arr)); 2311 mumps->id.schur = (MumpsScalar*)arr; 2312 mumps->id.size_schur = size; 2313 mumps->id.schur_lld = size; 2314 PetscCall(MatDenseRestoreArrayRead(F->schur,&arr)); 2315 if (mumps->sym == 1) { 2316 PetscCall(MatSetOption(F->schur,MAT_SPD,PETSC_TRUE)); 2317 } 2318 2319 /* MUMPS expects Fortran style indices */ 2320 PetscCall(PetscFree(mumps->id.listvar_schur)); 2321 PetscCall(PetscMalloc1(size,&mumps->id.listvar_schur)); 2322 PetscCall(ISGetIndices(is,&idxs)); 2323 for (i=0; i<size; i++) PetscCall(PetscMUMPSIntCast(idxs[i]+1,&(mumps->id.listvar_schur[i]))); 2324 PetscCall(ISRestoreIndices(is,&idxs)); 2325 if (mumps->petsc_size > 1) { 2326 mumps->id.ICNTL(19) = 1; /* MUMPS returns Schur centralized on the host */ 2327 } else { 2328 if (F->factortype == MAT_FACTOR_LU) { 2329 mumps->id.ICNTL(19) = 3; /* MUMPS returns full matrix */ 2330 } else { 2331 mumps->id.ICNTL(19) = 2; /* MUMPS returns lower triangular part */ 2332 } 2333 } 2334 /* set a special value of ICNTL (not handled my MUMPS) to be used in the solve phase by PETSc */ 2335 mumps->id.ICNTL(26) = -1; 2336 PetscFunctionReturn(0); 2337 } 2338 2339 /* -------------------------------------------------------------------------------------------*/ 2340 PetscErrorCode MatFactorCreateSchurComplement_MUMPS(Mat F,Mat* S) 2341 { 2342 Mat St; 2343 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 2344 PetscScalar *array; 2345 #if defined(PETSC_USE_COMPLEX) 2346 PetscScalar im = PetscSqrtScalar((PetscScalar)-1.0); 2347 #endif 2348 2349 PetscFunctionBegin; 2350 PetscCheck(mumps->id.ICNTL(19),PetscObjectComm((PetscObject)F),PETSC_ERR_ORDER,"Schur complement mode not selected! You should call MatFactorSetSchurIS to enable it"); 2351 PetscCall(MatCreate(PETSC_COMM_SELF,&St)); 2352 PetscCall(MatSetSizes(St,PETSC_DECIDE,PETSC_DECIDE,mumps->id.size_schur,mumps->id.size_schur)); 2353 PetscCall(MatSetType(St,MATDENSE)); 2354 PetscCall(MatSetUp(St)); 2355 PetscCall(MatDenseGetArray(St,&array)); 2356 if (!mumps->sym) { /* MUMPS always return a full matrix */ 2357 if (mumps->id.ICNTL(19) == 1) { /* stored by rows */ 2358 PetscInt i,j,N=mumps->id.size_schur; 2359 for (i=0;i<N;i++) { 2360 for (j=0;j<N;j++) { 2361 #if !defined(PETSC_USE_COMPLEX) 2362 PetscScalar val = mumps->id.schur[i*N+j]; 2363 #else 2364 PetscScalar val = mumps->id.schur[i*N+j].r + im*mumps->id.schur[i*N+j].i; 2365 #endif 2366 array[j*N+i] = val; 2367 } 2368 } 2369 } else { /* stored by columns */ 2370 PetscCall(PetscArraycpy(array,mumps->id.schur,mumps->id.size_schur*mumps->id.size_schur)); 2371 } 2372 } else { /* either full or lower-triangular (not packed) */ 2373 if (mumps->id.ICNTL(19) == 2) { /* lower triangular stored by columns */ 2374 PetscInt i,j,N=mumps->id.size_schur; 2375 for (i=0;i<N;i++) { 2376 for (j=i;j<N;j++) { 2377 #if !defined(PETSC_USE_COMPLEX) 2378 PetscScalar val = mumps->id.schur[i*N+j]; 2379 #else 2380 PetscScalar val = mumps->id.schur[i*N+j].r + im*mumps->id.schur[i*N+j].i; 2381 #endif 2382 array[i*N+j] = val; 2383 array[j*N+i] = val; 2384 } 2385 } 2386 } else if (mumps->id.ICNTL(19) == 3) { /* full matrix */ 2387 PetscCall(PetscArraycpy(array,mumps->id.schur,mumps->id.size_schur*mumps->id.size_schur)); 2388 } else { /* ICNTL(19) == 1 lower triangular stored by rows */ 2389 PetscInt i,j,N=mumps->id.size_schur; 2390 for (i=0;i<N;i++) { 2391 for (j=0;j<i+1;j++) { 2392 #if !defined(PETSC_USE_COMPLEX) 2393 PetscScalar val = mumps->id.schur[i*N+j]; 2394 #else 2395 PetscScalar val = mumps->id.schur[i*N+j].r + im*mumps->id.schur[i*N+j].i; 2396 #endif 2397 array[i*N+j] = val; 2398 array[j*N+i] = val; 2399 } 2400 } 2401 } 2402 } 2403 PetscCall(MatDenseRestoreArray(St,&array)); 2404 *S = St; 2405 PetscFunctionReturn(0); 2406 } 2407 2408 /* -------------------------------------------------------------------------------------------*/ 2409 PetscErrorCode MatMumpsSetIcntl_MUMPS(Mat F,PetscInt icntl,PetscInt ival) 2410 { 2411 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 2412 2413 PetscFunctionBegin; 2414 PetscCall(PetscMUMPSIntCast(ival,&mumps->id.ICNTL(icntl))); 2415 PetscFunctionReturn(0); 2416 } 2417 2418 PetscErrorCode MatMumpsGetIcntl_MUMPS(Mat F,PetscInt icntl,PetscInt *ival) 2419 { 2420 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 2421 2422 PetscFunctionBegin; 2423 *ival = mumps->id.ICNTL(icntl); 2424 PetscFunctionReturn(0); 2425 } 2426 2427 /*@ 2428 MatMumpsSetIcntl - Set MUMPS parameter ICNTL() 2429 2430 Logically Collective on Mat 2431 2432 Input Parameters: 2433 + F - the factored matrix obtained by calling MatGetFactor() from PETSc-MUMPS interface 2434 . icntl - index of MUMPS parameter array ICNTL() 2435 - ival - value of MUMPS ICNTL(icntl) 2436 2437 Options Database: 2438 . -mat_mumps_icntl_<icntl> <ival> - change the option numbered icntl to ival 2439 2440 Level: beginner 2441 2442 References: 2443 . * - MUMPS Users' Guide 2444 2445 .seealso: MatGetFactor(), MatMumpsGetIcntl(), MatMumpsSetCntl(), MatMumpsGetCntl(), MatMumpsGetInfo(), MatMumpsGetInfog(), MatMumpsGetRinfo(), MatMumpsGetRinfog() 2446 @*/ 2447 PetscErrorCode MatMumpsSetIcntl(Mat F,PetscInt icntl,PetscInt ival) 2448 { 2449 PetscFunctionBegin; 2450 PetscValidType(F,1); 2451 PetscCheck(F->factortype,PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_WRONGSTATE,"Only for factored matrix"); 2452 PetscValidLogicalCollectiveInt(F,icntl,2); 2453 PetscValidLogicalCollectiveInt(F,ival,3); 2454 PetscTryMethod(F,"MatMumpsSetIcntl_C",(Mat,PetscInt,PetscInt),(F,icntl,ival)); 2455 PetscFunctionReturn(0); 2456 } 2457 2458 /*@ 2459 MatMumpsGetIcntl - Get MUMPS parameter ICNTL() 2460 2461 Logically Collective on Mat 2462 2463 Input Parameters: 2464 + F - the factored matrix obtained by calling MatGetFactor() from PETSc-MUMPS interface 2465 - icntl - index of MUMPS parameter array ICNTL() 2466 2467 Output Parameter: 2468 . ival - value of MUMPS ICNTL(icntl) 2469 2470 Level: beginner 2471 2472 References: 2473 . * - MUMPS Users' Guide 2474 2475 .seealso: MatGetFactor(), MatMumpsSetIcntl(), MatMumpsSetCntl(), MatMumpsGetCntl(), MatMumpsGetInfo(), MatMumpsGetInfog(), MatMumpsGetRinfo(), MatMumpsGetRinfog() 2476 @*/ 2477 PetscErrorCode MatMumpsGetIcntl(Mat F,PetscInt icntl,PetscInt *ival) 2478 { 2479 PetscFunctionBegin; 2480 PetscValidType(F,1); 2481 PetscCheck(F->factortype,PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_WRONGSTATE,"Only for factored matrix"); 2482 PetscValidLogicalCollectiveInt(F,icntl,2); 2483 PetscValidIntPointer(ival,3); 2484 PetscUseMethod(F,"MatMumpsGetIcntl_C",(Mat,PetscInt,PetscInt*),(F,icntl,ival)); 2485 PetscFunctionReturn(0); 2486 } 2487 2488 /* -------------------------------------------------------------------------------------------*/ 2489 PetscErrorCode MatMumpsSetCntl_MUMPS(Mat F,PetscInt icntl,PetscReal val) 2490 { 2491 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 2492 2493 PetscFunctionBegin; 2494 mumps->id.CNTL(icntl) = val; 2495 PetscFunctionReturn(0); 2496 } 2497 2498 PetscErrorCode MatMumpsGetCntl_MUMPS(Mat F,PetscInt icntl,PetscReal *val) 2499 { 2500 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 2501 2502 PetscFunctionBegin; 2503 *val = mumps->id.CNTL(icntl); 2504 PetscFunctionReturn(0); 2505 } 2506 2507 /*@ 2508 MatMumpsSetCntl - Set MUMPS parameter CNTL() 2509 2510 Logically Collective on Mat 2511 2512 Input Parameters: 2513 + F - the factored matrix obtained by calling MatGetFactor() from PETSc-MUMPS interface 2514 . icntl - index of MUMPS parameter array CNTL() 2515 - val - value of MUMPS CNTL(icntl) 2516 2517 Options Database: 2518 . -mat_mumps_cntl_<icntl> <val> - change the option numbered icntl to ival 2519 2520 Level: beginner 2521 2522 References: 2523 . * - MUMPS Users' Guide 2524 2525 .seealso: MatGetFactor(), MatMumpsSetIcntl(), MatMumpsGetIcntl(), MatMumpsGetCntl(), MatMumpsGetInfo(), MatMumpsGetInfog(), MatMumpsGetRinfo(), MatMumpsGetRinfog() 2526 @*/ 2527 PetscErrorCode MatMumpsSetCntl(Mat F,PetscInt icntl,PetscReal val) 2528 { 2529 PetscFunctionBegin; 2530 PetscValidType(F,1); 2531 PetscCheck(F->factortype,PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_WRONGSTATE,"Only for factored matrix"); 2532 PetscValidLogicalCollectiveInt(F,icntl,2); 2533 PetscValidLogicalCollectiveReal(F,val,3); 2534 PetscTryMethod(F,"MatMumpsSetCntl_C",(Mat,PetscInt,PetscReal),(F,icntl,val)); 2535 PetscFunctionReturn(0); 2536 } 2537 2538 /*@ 2539 MatMumpsGetCntl - Get MUMPS parameter CNTL() 2540 2541 Logically Collective on Mat 2542 2543 Input Parameters: 2544 + F - the factored matrix obtained by calling MatGetFactor() from PETSc-MUMPS interface 2545 - icntl - index of MUMPS parameter array CNTL() 2546 2547 Output Parameter: 2548 . val - value of MUMPS CNTL(icntl) 2549 2550 Level: beginner 2551 2552 References: 2553 . * - MUMPS Users' Guide 2554 2555 .seealso: MatGetFactor(), MatMumpsSetIcntl(), MatMumpsGetIcntl(), MatMumpsSetCntl(), MatMumpsGetInfo(), MatMumpsGetInfog(), MatMumpsGetRinfo(), MatMumpsGetRinfog() 2556 @*/ 2557 PetscErrorCode MatMumpsGetCntl(Mat F,PetscInt icntl,PetscReal *val) 2558 { 2559 PetscFunctionBegin; 2560 PetscValidType(F,1); 2561 PetscCheck(F->factortype,PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_WRONGSTATE,"Only for factored matrix"); 2562 PetscValidLogicalCollectiveInt(F,icntl,2); 2563 PetscValidRealPointer(val,3); 2564 PetscUseMethod(F,"MatMumpsGetCntl_C",(Mat,PetscInt,PetscReal*),(F,icntl,val)); 2565 PetscFunctionReturn(0); 2566 } 2567 2568 PetscErrorCode MatMumpsGetInfo_MUMPS(Mat F,PetscInt icntl,PetscInt *info) 2569 { 2570 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 2571 2572 PetscFunctionBegin; 2573 *info = mumps->id.INFO(icntl); 2574 PetscFunctionReturn(0); 2575 } 2576 2577 PetscErrorCode MatMumpsGetInfog_MUMPS(Mat F,PetscInt icntl,PetscInt *infog) 2578 { 2579 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 2580 2581 PetscFunctionBegin; 2582 *infog = mumps->id.INFOG(icntl); 2583 PetscFunctionReturn(0); 2584 } 2585 2586 PetscErrorCode MatMumpsGetRinfo_MUMPS(Mat F,PetscInt icntl,PetscReal *rinfo) 2587 { 2588 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 2589 2590 PetscFunctionBegin; 2591 *rinfo = mumps->id.RINFO(icntl); 2592 PetscFunctionReturn(0); 2593 } 2594 2595 PetscErrorCode MatMumpsGetRinfog_MUMPS(Mat F,PetscInt icntl,PetscReal *rinfog) 2596 { 2597 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 2598 2599 PetscFunctionBegin; 2600 *rinfog = mumps->id.RINFOG(icntl); 2601 PetscFunctionReturn(0); 2602 } 2603 2604 PetscErrorCode MatMumpsGetInverse_MUMPS(Mat F,Mat spRHS) 2605 { 2606 Mat Bt = NULL,Btseq = NULL; 2607 PetscBool flg; 2608 Mat_MUMPS *mumps =(Mat_MUMPS*)F->data; 2609 PetscScalar *aa; 2610 PetscInt spnr,*ia,*ja,M,nrhs; 2611 2612 PetscFunctionBegin; 2613 PetscValidPointer(spRHS,2); 2614 PetscCall(PetscObjectTypeCompare((PetscObject)spRHS,MATTRANSPOSEMAT,&flg)); 2615 if (flg) { 2616 PetscCall(MatTransposeGetMat(spRHS,&Bt)); 2617 } else SETERRQ(PetscObjectComm((PetscObject)spRHS),PETSC_ERR_ARG_WRONG,"Matrix spRHS must be type MATTRANSPOSEMAT matrix"); 2618 2619 PetscCall(MatMumpsSetIcntl(F,30,1)); 2620 2621 if (mumps->petsc_size > 1) { 2622 Mat_MPIAIJ *b = (Mat_MPIAIJ*)Bt->data; 2623 Btseq = b->A; 2624 } else { 2625 Btseq = Bt; 2626 } 2627 2628 PetscCall(MatGetSize(spRHS,&M,&nrhs)); 2629 mumps->id.nrhs = nrhs; 2630 mumps->id.lrhs = M; 2631 mumps->id.rhs = NULL; 2632 2633 if (!mumps->myid) { 2634 PetscCall(MatSeqAIJGetArray(Btseq,&aa)); 2635 PetscCall(MatGetRowIJ(Btseq,1,PETSC_FALSE,PETSC_FALSE,&spnr,(const PetscInt**)&ia,(const PetscInt**)&ja,&flg)); 2636 PetscCheck(flg,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Cannot get IJ structure"); 2637 PetscCall(PetscMUMPSIntCSRCast(mumps,spnr,ia,ja,&mumps->id.irhs_ptr,&mumps->id.irhs_sparse,&mumps->id.nz_rhs)); 2638 mumps->id.rhs_sparse = (MumpsScalar*)aa; 2639 } else { 2640 mumps->id.irhs_ptr = NULL; 2641 mumps->id.irhs_sparse = NULL; 2642 mumps->id.nz_rhs = 0; 2643 mumps->id.rhs_sparse = NULL; 2644 } 2645 mumps->id.ICNTL(20) = 1; /* rhs is sparse */ 2646 mumps->id.ICNTL(21) = 0; /* solution is in assembled centralized format */ 2647 2648 /* solve phase */ 2649 /*-------------*/ 2650 mumps->id.job = JOB_SOLVE; 2651 PetscMUMPS_c(mumps); 2652 if (mumps->id.INFOG(1) < 0) 2653 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MUMPS in solve phase: INFOG(1)=%d INFO(2)=%d",mumps->id.INFOG(1),mumps->id.INFO(2)); 2654 2655 if (!mumps->myid) { 2656 PetscCall(MatSeqAIJRestoreArray(Btseq,&aa)); 2657 PetscCall(MatRestoreRowIJ(Btseq,1,PETSC_FALSE,PETSC_FALSE,&spnr,(const PetscInt**)&ia,(const PetscInt**)&ja,&flg)); 2658 PetscCheck(flg,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Cannot get IJ structure"); 2659 } 2660 PetscFunctionReturn(0); 2661 } 2662 2663 /*@ 2664 MatMumpsGetInverse - Get user-specified set of entries in inverse of A 2665 2666 Logically Collective on Mat 2667 2668 Input Parameters: 2669 + F - the factored matrix obtained by calling MatGetFactor() from PETSc-MUMPS interface 2670 - spRHS - sequential sparse matrix in MATTRANSPOSEMAT format holding specified indices in processor[0] 2671 2672 Output Parameter: 2673 . spRHS - requested entries of inverse of A 2674 2675 Level: beginner 2676 2677 References: 2678 . * - MUMPS Users' Guide 2679 2680 .seealso: MatGetFactor(), MatCreateTranspose() 2681 @*/ 2682 PetscErrorCode MatMumpsGetInverse(Mat F,Mat spRHS) 2683 { 2684 PetscFunctionBegin; 2685 PetscValidType(F,1); 2686 PetscCheck(F->factortype,PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_WRONGSTATE,"Only for factored matrix"); 2687 PetscUseMethod(F,"MatMumpsGetInverse_C",(Mat,Mat),(F,spRHS)); 2688 PetscFunctionReturn(0); 2689 } 2690 2691 PetscErrorCode MatMumpsGetInverseTranspose_MUMPS(Mat F,Mat spRHST) 2692 { 2693 Mat spRHS; 2694 2695 PetscFunctionBegin; 2696 PetscCall(MatCreateTranspose(spRHST,&spRHS)); 2697 PetscCall(MatMumpsGetInverse_MUMPS(F,spRHS)); 2698 PetscCall(MatDestroy(&spRHS)); 2699 PetscFunctionReturn(0); 2700 } 2701 2702 /*@ 2703 MatMumpsGetInverseTranspose - Get user-specified set of entries in inverse of matrix A^T 2704 2705 Logically Collective on Mat 2706 2707 Input Parameters: 2708 + F - the factored matrix of A obtained by calling MatGetFactor() from PETSc-MUMPS interface 2709 - spRHST - sequential sparse matrix in MATAIJ format holding specified indices of A^T in processor[0] 2710 2711 Output Parameter: 2712 . spRHST - requested entries of inverse of A^T 2713 2714 Level: beginner 2715 2716 References: 2717 . * - MUMPS Users' Guide 2718 2719 .seealso: MatGetFactor(), MatCreateTranspose(), MatMumpsGetInverse() 2720 @*/ 2721 PetscErrorCode MatMumpsGetInverseTranspose(Mat F,Mat spRHST) 2722 { 2723 PetscBool flg; 2724 2725 PetscFunctionBegin; 2726 PetscValidType(F,1); 2727 PetscCheck(F->factortype,PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_WRONGSTATE,"Only for factored matrix"); 2728 PetscCall(PetscObjectTypeCompareAny((PetscObject)spRHST,&flg,MATSEQAIJ,MATMPIAIJ,NULL)); 2729 PetscCheck(flg,PetscObjectComm((PetscObject)spRHST),PETSC_ERR_ARG_WRONG,"Matrix spRHST must be MATAIJ matrix"); 2730 2731 PetscUseMethod(F,"MatMumpsGetInverseTranspose_C",(Mat,Mat),(F,spRHST)); 2732 PetscFunctionReturn(0); 2733 } 2734 2735 /*@ 2736 MatMumpsGetInfo - Get MUMPS parameter INFO() 2737 2738 Logically Collective on Mat 2739 2740 Input Parameters: 2741 + F - the factored matrix obtained by calling MatGetFactor() from PETSc-MUMPS interface 2742 - icntl - index of MUMPS parameter array INFO() 2743 2744 Output Parameter: 2745 . ival - value of MUMPS INFO(icntl) 2746 2747 Level: beginner 2748 2749 References: 2750 . * - MUMPS Users' Guide 2751 2752 .seealso: MatGetFactor(), MatMumpsSetIcntl(), MatMumpsGetIcntl(), MatMumpsSetCntl(), MatMumpsGetCntl(), MatMumpsGetInfog(), MatMumpsGetRinfo(), MatMumpsGetRinfog() 2753 @*/ 2754 PetscErrorCode MatMumpsGetInfo(Mat F,PetscInt icntl,PetscInt *ival) 2755 { 2756 PetscFunctionBegin; 2757 PetscValidType(F,1); 2758 PetscCheck(F->factortype,PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_WRONGSTATE,"Only for factored matrix"); 2759 PetscValidIntPointer(ival,3); 2760 PetscUseMethod(F,"MatMumpsGetInfo_C",(Mat,PetscInt,PetscInt*),(F,icntl,ival)); 2761 PetscFunctionReturn(0); 2762 } 2763 2764 /*@ 2765 MatMumpsGetInfog - Get MUMPS parameter INFOG() 2766 2767 Logically Collective on Mat 2768 2769 Input Parameters: 2770 + F - the factored matrix obtained by calling MatGetFactor() from PETSc-MUMPS interface 2771 - icntl - index of MUMPS parameter array INFOG() 2772 2773 Output Parameter: 2774 . ival - value of MUMPS INFOG(icntl) 2775 2776 Level: beginner 2777 2778 References: 2779 . * - MUMPS Users' Guide 2780 2781 .seealso: MatGetFactor(), MatMumpsSetIcntl(), MatMumpsGetIcntl(), MatMumpsSetCntl(), MatMumpsGetCntl(), MatMumpsGetInfo(), MatMumpsGetRinfo(), MatMumpsGetRinfog() 2782 @*/ 2783 PetscErrorCode MatMumpsGetInfog(Mat F,PetscInt icntl,PetscInt *ival) 2784 { 2785 PetscFunctionBegin; 2786 PetscValidType(F,1); 2787 PetscCheck(F->factortype,PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_WRONGSTATE,"Only for factored matrix"); 2788 PetscValidIntPointer(ival,3); 2789 PetscUseMethod(F,"MatMumpsGetInfog_C",(Mat,PetscInt,PetscInt*),(F,icntl,ival)); 2790 PetscFunctionReturn(0); 2791 } 2792 2793 /*@ 2794 MatMumpsGetRinfo - Get MUMPS parameter RINFO() 2795 2796 Logically Collective on Mat 2797 2798 Input Parameters: 2799 + F - the factored matrix obtained by calling MatGetFactor() from PETSc-MUMPS interface 2800 - icntl - index of MUMPS parameter array RINFO() 2801 2802 Output Parameter: 2803 . val - value of MUMPS RINFO(icntl) 2804 2805 Level: beginner 2806 2807 References: 2808 . * - MUMPS Users' Guide 2809 2810 .seealso: MatGetFactor(), MatMumpsSetIcntl(), MatMumpsGetIcntl(), MatMumpsSetCntl(), MatMumpsGetCntl(), MatMumpsGetInfo(), MatMumpsGetInfog(), MatMumpsGetRinfog() 2811 @*/ 2812 PetscErrorCode MatMumpsGetRinfo(Mat F,PetscInt icntl,PetscReal *val) 2813 { 2814 PetscFunctionBegin; 2815 PetscValidType(F,1); 2816 PetscCheck(F->factortype,PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_WRONGSTATE,"Only for factored matrix"); 2817 PetscValidRealPointer(val,3); 2818 PetscUseMethod(F,"MatMumpsGetRinfo_C",(Mat,PetscInt,PetscReal*),(F,icntl,val)); 2819 PetscFunctionReturn(0); 2820 } 2821 2822 /*@ 2823 MatMumpsGetRinfog - Get MUMPS parameter RINFOG() 2824 2825 Logically Collective on Mat 2826 2827 Input Parameters: 2828 + F - the factored matrix obtained by calling MatGetFactor() from PETSc-MUMPS interface 2829 - icntl - index of MUMPS parameter array RINFOG() 2830 2831 Output Parameter: 2832 . val - value of MUMPS RINFOG(icntl) 2833 2834 Level: beginner 2835 2836 References: 2837 . * - MUMPS Users' Guide 2838 2839 .seealso: MatGetFactor(), MatMumpsSetIcntl(), MatMumpsGetIcntl(), MatMumpsSetCntl(), MatMumpsGetCntl(), MatMumpsGetInfo(), MatMumpsGetInfog(), MatMumpsGetRinfo() 2840 @*/ 2841 PetscErrorCode MatMumpsGetRinfog(Mat F,PetscInt icntl,PetscReal *val) 2842 { 2843 PetscFunctionBegin; 2844 PetscValidType(F,1); 2845 PetscCheck(F->factortype,PetscObjectComm((PetscObject)F),PETSC_ERR_ARG_WRONGSTATE,"Only for factored matrix"); 2846 PetscValidRealPointer(val,3); 2847 PetscUseMethod(F,"MatMumpsGetRinfog_C",(Mat,PetscInt,PetscReal*),(F,icntl,val)); 2848 PetscFunctionReturn(0); 2849 } 2850 2851 /*MC 2852 MATSOLVERMUMPS - A matrix type providing direct solvers (LU and Cholesky) for 2853 distributed and sequential matrices via the external package MUMPS. 2854 2855 Works with MATAIJ and MATSBAIJ matrices 2856 2857 Use ./configure --download-mumps --download-scalapack --download-parmetis --download-metis --download-ptscotch to have PETSc installed with MUMPS 2858 2859 Use ./configure --with-openmp --download-hwloc (or --with-hwloc) to enable running MUMPS in MPI+OpenMP hybrid mode and non-MUMPS in flat-MPI mode. See details below. 2860 2861 Use -pc_type cholesky or lu -pc_factor_mat_solver_type mumps to use this direct solver 2862 2863 Options Database Keys: 2864 + -mat_mumps_icntl_1 - ICNTL(1): output stream for error messages 2865 . -mat_mumps_icntl_2 - ICNTL(2): output stream for diagnostic printing, statistics, and warning 2866 . -mat_mumps_icntl_3 - ICNTL(3): output stream for global information, collected on the host 2867 . -mat_mumps_icntl_4 - ICNTL(4): level of printing (0 to 4) 2868 . -mat_mumps_icntl_6 - ICNTL(6): permutes to a zero-free diagonal and/or scale the matrix (0 to 7) 2869 . -mat_mumps_icntl_7 - ICNTL(7): computes a symmetric permutation in sequential analysis, 0=AMD, 2=AMF, 3=Scotch, 4=PORD, 5=Metis, 6=QAMD, and 7=auto 2870 Use -pc_factor_mat_ordering_type <type> to have PETSc perform the ordering (sequential only) 2871 . -mat_mumps_icntl_8 - ICNTL(8): scaling strategy (-2 to 8 or 77) 2872 . -mat_mumps_icntl_10 - ICNTL(10): max num of refinements 2873 . -mat_mumps_icntl_11 - ICNTL(11): statistics related to an error analysis (via -ksp_view) 2874 . -mat_mumps_icntl_12 - ICNTL(12): an ordering strategy for symmetric matrices (0 to 3) 2875 . -mat_mumps_icntl_13 - ICNTL(13): parallelism of the root node (enable ScaLAPACK) and its splitting 2876 . -mat_mumps_icntl_14 - ICNTL(14): percentage increase in the estimated working space 2877 . -mat_mumps_icntl_15 - ICNTL(15): compression of the input matrix resulting from a block format 2878 . -mat_mumps_icntl_19 - ICNTL(19): computes the Schur complement 2879 . -mat_mumps_icntl_20 - ICNTL(20): give MUMPS centralized (0) or distributed (10) dense RHS 2880 . -mat_mumps_icntl_22 - ICNTL(22): in-core/out-of-core factorization and solve (0 or 1) 2881 . -mat_mumps_icntl_23 - ICNTL(23): max size of the working memory (MB) that can allocate per processor 2882 . -mat_mumps_icntl_24 - ICNTL(24): detection of null pivot rows (0 or 1) 2883 . -mat_mumps_icntl_25 - ICNTL(25): compute a solution of a deficient matrix and a null space basis 2884 . -mat_mumps_icntl_26 - ICNTL(26): drives the solution phase if a Schur complement matrix 2885 . -mat_mumps_icntl_28 - ICNTL(28): use 1 for sequential analysis and ictnl(7) ordering, or 2 for parallel analysis and ictnl(29) ordering 2886 . -mat_mumps_icntl_29 - ICNTL(29): parallel ordering 1 = ptscotch, 2 = parmetis 2887 . -mat_mumps_icntl_30 - ICNTL(30): compute user-specified set of entries in inv(A) 2888 . -mat_mumps_icntl_31 - ICNTL(31): indicates which factors may be discarded during factorization 2889 . -mat_mumps_icntl_33 - ICNTL(33): compute determinant 2890 . -mat_mumps_icntl_35 - ICNTL(35): level of activation of BLR (Block Low-Rank) feature 2891 . -mat_mumps_icntl_36 - ICNTL(36): controls the choice of BLR factorization variant 2892 . -mat_mumps_icntl_38 - ICNTL(38): sets the estimated compression rate of LU factors with BLR 2893 . -mat_mumps_cntl_1 - CNTL(1): relative pivoting threshold 2894 . -mat_mumps_cntl_2 - CNTL(2): stopping criterion of refinement 2895 . -mat_mumps_cntl_3 - CNTL(3): absolute pivoting threshold 2896 . -mat_mumps_cntl_4 - CNTL(4): value for static pivoting 2897 . -mat_mumps_cntl_5 - CNTL(5): fixation for null pivots 2898 . -mat_mumps_cntl_7 - CNTL(7): precision of the dropping parameter used during BLR factorization 2899 - -mat_mumps_use_omp_threads [m] - run MUMPS in MPI+OpenMP hybrid mode as if omp_set_num_threads(m) is called before calling MUMPS. 2900 Default might be the number of cores per CPU package (socket) as reported by hwloc and suggested by the MUMPS manual. 2901 2902 Level: beginner 2903 2904 Notes: 2905 MUMPS Cholesky does not handle (complex) Hermitian matrices http://mumps.enseeiht.fr/doc/userguide_5.2.1.pdf so using it will error if the matrix is Hermitian. 2906 2907 When a MUMPS factorization fails inside a KSP solve, for example with a KSP_DIVERGED_PC_FAILED, one can find the MUMPS information about the failure by calling 2908 $ KSPGetPC(ksp,&pc); 2909 $ PCFactorGetMatrix(pc,&mat); 2910 $ MatMumpsGetInfo(mat,....); 2911 $ MatMumpsGetInfog(mat,....); etc. 2912 Or you can run with -ksp_error_if_not_converged and the program will be stopped and the information printed in the error message. 2913 2914 Using MUMPS with 64-bit integers 2915 MUMPS provides 64-bit integer support in two build modes: 2916 full 64-bit: here MUMPS is built with C preprocessing flag -DINTSIZE64 and Fortran compiler option -i8, -fdefault-integer-8 or equivalent, and 2917 requires all dependent libraries MPI, ScaLAPACK, LAPACK and BLAS built the same way with 64-bit integers (for example ILP64 Intel MKL and MPI). 2918 2919 selective 64-bit: with the default MUMPS build, 64-bit integers have been introduced where needed. In compressed sparse row (CSR) storage of matrices, 2920 MUMPS stores column indices in 32-bit, but row offsets in 64-bit, so you can have a huge number of non-zeros, but must have less than 2^31 rows and 2921 columns. This can lead to significant memory and performance gains with respect to a full 64-bit integer MUMPS version. This requires a regular (32-bit 2922 integer) build of all dependent libraries MPI, ScaLAPACK, LAPACK and BLAS. 2923 2924 With --download-mumps=1, PETSc always build MUMPS in selective 64-bit mode, which can be used by both --with-64-bit-indices=0/1 variants of PETSc. 2925 2926 Two modes to run MUMPS/PETSc with OpenMP 2927 $ Set OMP_NUM_THREADS and run with fewer MPI ranks than cores. For example, if you want to have 16 OpenMP 2928 $ threads per rank, then you may use "export OMP_NUM_THREADS=16 && mpirun -n 4 ./test". 2929 2930 $ -mat_mumps_use_omp_threads [m] and run your code with as many MPI ranks as the number of cores. For example, 2931 $ if a compute node has 32 cores and you run on two nodes, you may use "mpirun -n 64 ./test -mat_mumps_use_omp_threads 16" 2932 2933 To run MUMPS in MPI+OpenMP hybrid mode (i.e., enable multithreading in MUMPS), but still run the non-MUMPS part 2934 (i.e., PETSc part) of your code in the so-called flat-MPI (aka pure-MPI) mode, you need to configure PETSc with --with-openmp --download-hwloc 2935 (or --with-hwloc), and have an MPI that supports MPI-3.0's process shared memory (which is usually available). Since MUMPS calls BLAS 2936 libraries, to really get performance, you should have multithreaded BLAS libraries such as Intel MKL, AMD ACML, Cray libSci or OpenBLAS 2937 (PETSc will automatically try to utilized a threaded BLAS if --with-openmp is provided). 2938 2939 If you run your code through a job submission system, there are caveats in MPI rank mapping. We use MPI_Comm_split_type() to obtain MPI 2940 processes on each compute node. Listing the processes in rank ascending order, we split processes on a node into consecutive groups of 2941 size m and create a communicator called omp_comm for each group. Rank 0 in an omp_comm is called the master rank, and others in the omp_comm 2942 are called slave ranks (or slaves). Only master ranks are seen to MUMPS and slaves are not. We will free CPUs assigned to slaves (might be set 2943 by CPU binding policies in job scripts) and make the CPUs available to the master so that OMP threads spawned by MUMPS can run on the CPUs. 2944 In a multi-socket compute node, MPI rank mapping is an issue. Still use the above example and suppose your compute node has two sockets, 2945 if you interleave MPI ranks on the two sockets, in other words, even ranks are placed on socket 0, and odd ranks are on socket 1, and bind 2946 MPI ranks to cores, then with -mat_mumps_use_omp_threads 16, a master rank (and threads it spawns) will use half cores in socket 0, and half 2947 cores in socket 1, that definitely hurts locality. On the other hand, if you map MPI ranks consecutively on the two sockets, then the 2948 problem will not happen. Therefore, when you use -mat_mumps_use_omp_threads, you need to keep an eye on your MPI rank mapping and CPU binding. 2949 For example, with the Slurm job scheduler, one can use srun --cpu-bind=verbose -m block:block to map consecutive MPI ranks to sockets and 2950 examine the mapping result. 2951 2952 PETSc does not control thread binding in MUMPS. So to get best performance, one still has to set OMP_PROC_BIND and OMP_PLACES in job scripts, 2953 for example, export OMP_PLACES=threads and export OMP_PROC_BIND=spread. One does not need to export OMP_NUM_THREADS=m in job scripts as PETSc 2954 calls omp_set_num_threads(m) internally before calling MUMPS. 2955 2956 References: 2957 + * - Heroux, Michael A., R. Brightwell, and Michael M. Wolf. "Bi-modal MPI and MPI+ threads computing on scalable multicore systems." IJHPCA (Submitted) (2011). 2958 - * - Gutierrez, Samuel K., et al. "Accommodating Thread-Level Heterogeneity in Coupled Parallel Applications." Parallel and Distributed Processing Symposium (IPDPS), 2017 IEEE International. IEEE, 2017. 2959 2960 .seealso: PCFactorSetMatSolverType(), MatSolverType, MatMumpsSetIcntl(), MatMumpsGetIcntl(), MatMumpsSetCntl(), MatMumpsGetCntl(), MatMumpsGetInfo(), MatMumpsGetInfog(), MatMumpsGetRinfo(), MatMumpsGetRinfog(), KSPGetPC(), PCFactorGetMatrix() 2961 2962 M*/ 2963 2964 static PetscErrorCode MatFactorGetSolverType_mumps(Mat A,MatSolverType *type) 2965 { 2966 PetscFunctionBegin; 2967 *type = MATSOLVERMUMPS; 2968 PetscFunctionReturn(0); 2969 } 2970 2971 /* MatGetFactor for Seq and MPI AIJ matrices */ 2972 static PetscErrorCode MatGetFactor_aij_mumps(Mat A,MatFactorType ftype,Mat *F) 2973 { 2974 Mat B; 2975 Mat_MUMPS *mumps; 2976 PetscBool isSeqAIJ; 2977 PetscMPIInt size; 2978 2979 PetscFunctionBegin; 2980 #if defined(PETSC_USE_COMPLEX) 2981 PetscCheckFalse(A->hermitian && !A->symmetric && ftype == MAT_FACTOR_CHOLESKY,PETSC_COMM_SELF,PETSC_ERR_SUP,"Hermitian CHOLESKY Factor is not supported"); 2982 #endif 2983 /* Create the factorization matrix */ 2984 PetscCall(PetscObjectBaseTypeCompare((PetscObject)A,MATSEQAIJ,&isSeqAIJ)); 2985 PetscCall(MatCreate(PetscObjectComm((PetscObject)A),&B)); 2986 PetscCall(MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N)); 2987 PetscCall(PetscStrallocpy("mumps",&((PetscObject)B)->type_name)); 2988 PetscCall(MatSetUp(B)); 2989 2990 PetscCall(PetscNewLog(B,&mumps)); 2991 2992 B->ops->view = MatView_MUMPS; 2993 B->ops->getinfo = MatGetInfo_MUMPS; 2994 2995 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverType_C",MatFactorGetSolverType_mumps)); 2996 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorSetSchurIS_C",MatFactorSetSchurIS_MUMPS)); 2997 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorCreateSchurComplement_C",MatFactorCreateSchurComplement_MUMPS)); 2998 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsSetIcntl_C",MatMumpsSetIcntl_MUMPS)); 2999 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetIcntl_C",MatMumpsGetIcntl_MUMPS)); 3000 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsSetCntl_C",MatMumpsSetCntl_MUMPS)); 3001 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetCntl_C",MatMumpsGetCntl_MUMPS)); 3002 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInfo_C",MatMumpsGetInfo_MUMPS)); 3003 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInfog_C",MatMumpsGetInfog_MUMPS)); 3004 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetRinfo_C",MatMumpsGetRinfo_MUMPS)); 3005 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetRinfog_C",MatMumpsGetRinfog_MUMPS)); 3006 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInverse_C",MatMumpsGetInverse_MUMPS)); 3007 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInverseTranspose_C",MatMumpsGetInverseTranspose_MUMPS)); 3008 3009 if (ftype == MAT_FACTOR_LU) { 3010 B->ops->lufactorsymbolic = MatLUFactorSymbolic_AIJMUMPS; 3011 B->factortype = MAT_FACTOR_LU; 3012 if (isSeqAIJ) mumps->ConvertToTriples = MatConvertToTriples_seqaij_seqaij; 3013 else mumps->ConvertToTriples = MatConvertToTriples_mpiaij_mpiaij; 3014 PetscCall(PetscStrallocpy(MATORDERINGEXTERNAL,(char**)&B->preferredordering[MAT_FACTOR_LU])); 3015 mumps->sym = 0; 3016 } else { 3017 B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_MUMPS; 3018 B->factortype = MAT_FACTOR_CHOLESKY; 3019 if (isSeqAIJ) mumps->ConvertToTriples = MatConvertToTriples_seqaij_seqsbaij; 3020 else mumps->ConvertToTriples = MatConvertToTriples_mpiaij_mpisbaij; 3021 PetscCall(PetscStrallocpy(MATORDERINGEXTERNAL,(char**)&B->preferredordering[MAT_FACTOR_CHOLESKY])); 3022 #if defined(PETSC_USE_COMPLEX) 3023 mumps->sym = 2; 3024 #else 3025 if (A->spd_set && A->spd) mumps->sym = 1; 3026 else mumps->sym = 2; 3027 #endif 3028 } 3029 3030 /* set solvertype */ 3031 PetscCall(PetscFree(B->solvertype)); 3032 PetscCall(PetscStrallocpy(MATSOLVERMUMPS,&B->solvertype)); 3033 PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)A),&size)); 3034 if (size == 1) { 3035 /* MUMPS option -mat_mumps_icntl_7 1 is automatically set if PETSc ordering is passed into symbolic factorization */ 3036 B->canuseordering = PETSC_TRUE; 3037 } 3038 B->ops->destroy = MatDestroy_MUMPS; 3039 B->data = (void*)mumps; 3040 3041 PetscCall(PetscInitializeMUMPS(A,mumps)); 3042 3043 *F = B; 3044 mumps->matstruc = DIFFERENT_NONZERO_PATTERN; 3045 PetscFunctionReturn(0); 3046 } 3047 3048 /* MatGetFactor for Seq and MPI SBAIJ matrices */ 3049 static PetscErrorCode MatGetFactor_sbaij_mumps(Mat A,MatFactorType ftype,Mat *F) 3050 { 3051 Mat B; 3052 Mat_MUMPS *mumps; 3053 PetscBool isSeqSBAIJ; 3054 PetscMPIInt size; 3055 3056 PetscFunctionBegin; 3057 #if defined(PETSC_USE_COMPLEX) 3058 PetscCheck(!A->hermitian || A->symmetric,PETSC_COMM_SELF,PETSC_ERR_SUP,"Hermitian CHOLESKY Factor is not supported"); 3059 #endif 3060 PetscCall(MatCreate(PetscObjectComm((PetscObject)A),&B)); 3061 PetscCall(MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N)); 3062 PetscCall(PetscStrallocpy("mumps",&((PetscObject)B)->type_name)); 3063 PetscCall(MatSetUp(B)); 3064 3065 PetscCall(PetscNewLog(B,&mumps)); 3066 PetscCall(PetscObjectTypeCompare((PetscObject)A,MATSEQSBAIJ,&isSeqSBAIJ)); 3067 if (isSeqSBAIJ) { 3068 mumps->ConvertToTriples = MatConvertToTriples_seqsbaij_seqsbaij; 3069 } else { 3070 mumps->ConvertToTriples = MatConvertToTriples_mpisbaij_mpisbaij; 3071 } 3072 3073 B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_MUMPS; 3074 B->ops->view = MatView_MUMPS; 3075 B->ops->getinfo = MatGetInfo_MUMPS; 3076 3077 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverType_C",MatFactorGetSolverType_mumps)); 3078 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorSetSchurIS_C",MatFactorSetSchurIS_MUMPS)); 3079 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorCreateSchurComplement_C",MatFactorCreateSchurComplement_MUMPS)); 3080 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsSetIcntl_C",MatMumpsSetIcntl_MUMPS)); 3081 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetIcntl_C",MatMumpsGetIcntl_MUMPS)); 3082 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsSetCntl_C",MatMumpsSetCntl_MUMPS)); 3083 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetCntl_C",MatMumpsGetCntl_MUMPS)); 3084 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInfo_C",MatMumpsGetInfo_MUMPS)); 3085 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInfog_C",MatMumpsGetInfog_MUMPS)); 3086 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetRinfo_C",MatMumpsGetRinfo_MUMPS)); 3087 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetRinfog_C",MatMumpsGetRinfog_MUMPS)); 3088 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInverse_C",MatMumpsGetInverse_MUMPS)); 3089 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInverseTranspose_C",MatMumpsGetInverseTranspose_MUMPS)); 3090 3091 B->factortype = MAT_FACTOR_CHOLESKY; 3092 #if defined(PETSC_USE_COMPLEX) 3093 mumps->sym = 2; 3094 #else 3095 if (A->spd_set && A->spd) mumps->sym = 1; 3096 else mumps->sym = 2; 3097 #endif 3098 3099 /* set solvertype */ 3100 PetscCall(PetscFree(B->solvertype)); 3101 PetscCall(PetscStrallocpy(MATSOLVERMUMPS,&B->solvertype)); 3102 PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)A),&size)); 3103 if (size == 1) { 3104 /* MUMPS option -mat_mumps_icntl_7 1 is automatically set if PETSc ordering is passed into symbolic factorization */ 3105 B->canuseordering = PETSC_TRUE; 3106 } 3107 PetscCall(PetscStrallocpy(MATORDERINGEXTERNAL,(char**)&B->preferredordering[MAT_FACTOR_CHOLESKY])); 3108 B->ops->destroy = MatDestroy_MUMPS; 3109 B->data = (void*)mumps; 3110 3111 PetscCall(PetscInitializeMUMPS(A,mumps)); 3112 3113 *F = B; 3114 mumps->matstruc = DIFFERENT_NONZERO_PATTERN; 3115 PetscFunctionReturn(0); 3116 } 3117 3118 static PetscErrorCode MatGetFactor_baij_mumps(Mat A,MatFactorType ftype,Mat *F) 3119 { 3120 Mat B; 3121 Mat_MUMPS *mumps; 3122 PetscBool isSeqBAIJ; 3123 PetscMPIInt size; 3124 3125 PetscFunctionBegin; 3126 /* Create the factorization matrix */ 3127 PetscCall(PetscObjectTypeCompare((PetscObject)A,MATSEQBAIJ,&isSeqBAIJ)); 3128 PetscCall(MatCreate(PetscObjectComm((PetscObject)A),&B)); 3129 PetscCall(MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N)); 3130 PetscCall(PetscStrallocpy("mumps",&((PetscObject)B)->type_name)); 3131 PetscCall(MatSetUp(B)); 3132 3133 PetscCall(PetscNewLog(B,&mumps)); 3134 if (ftype == MAT_FACTOR_LU) { 3135 B->ops->lufactorsymbolic = MatLUFactorSymbolic_BAIJMUMPS; 3136 B->factortype = MAT_FACTOR_LU; 3137 if (isSeqBAIJ) mumps->ConvertToTriples = MatConvertToTriples_seqbaij_seqaij; 3138 else mumps->ConvertToTriples = MatConvertToTriples_mpibaij_mpiaij; 3139 mumps->sym = 0; 3140 PetscCall(PetscStrallocpy(MATORDERINGEXTERNAL,(char**)&B->preferredordering[MAT_FACTOR_LU])); 3141 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Cannot use PETSc BAIJ matrices with MUMPS Cholesky, use SBAIJ or AIJ matrix instead"); 3142 3143 B->ops->view = MatView_MUMPS; 3144 B->ops->getinfo = MatGetInfo_MUMPS; 3145 3146 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverType_C",MatFactorGetSolverType_mumps)); 3147 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorSetSchurIS_C",MatFactorSetSchurIS_MUMPS)); 3148 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorCreateSchurComplement_C",MatFactorCreateSchurComplement_MUMPS)); 3149 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsSetIcntl_C",MatMumpsSetIcntl_MUMPS)); 3150 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetIcntl_C",MatMumpsGetIcntl_MUMPS)); 3151 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsSetCntl_C",MatMumpsSetCntl_MUMPS)); 3152 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetCntl_C",MatMumpsGetCntl_MUMPS)); 3153 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInfo_C",MatMumpsGetInfo_MUMPS)); 3154 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInfog_C",MatMumpsGetInfog_MUMPS)); 3155 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetRinfo_C",MatMumpsGetRinfo_MUMPS)); 3156 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetRinfog_C",MatMumpsGetRinfog_MUMPS)); 3157 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInverse_C",MatMumpsGetInverse_MUMPS)); 3158 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInverseTranspose_C",MatMumpsGetInverseTranspose_MUMPS)); 3159 3160 /* set solvertype */ 3161 PetscCall(PetscFree(B->solvertype)); 3162 PetscCall(PetscStrallocpy(MATSOLVERMUMPS,&B->solvertype)); 3163 PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)A),&size)); 3164 if (size == 1) { 3165 /* MUMPS option -mat_mumps_icntl_7 1 is automatically set if PETSc ordering is passed into symbolic factorization */ 3166 B->canuseordering = PETSC_TRUE; 3167 } 3168 B->ops->destroy = MatDestroy_MUMPS; 3169 B->data = (void*)mumps; 3170 3171 PetscCall(PetscInitializeMUMPS(A,mumps)); 3172 3173 *F = B; 3174 mumps->matstruc = DIFFERENT_NONZERO_PATTERN; 3175 PetscFunctionReturn(0); 3176 } 3177 3178 /* MatGetFactor for Seq and MPI SELL matrices */ 3179 static PetscErrorCode MatGetFactor_sell_mumps(Mat A,MatFactorType ftype,Mat *F) 3180 { 3181 Mat B; 3182 Mat_MUMPS *mumps; 3183 PetscBool isSeqSELL; 3184 PetscMPIInt size; 3185 3186 PetscFunctionBegin; 3187 /* Create the factorization matrix */ 3188 PetscCall(PetscObjectTypeCompare((PetscObject)A,MATSEQSELL,&isSeqSELL)); 3189 PetscCall(MatCreate(PetscObjectComm((PetscObject)A),&B)); 3190 PetscCall(MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N)); 3191 PetscCall(PetscStrallocpy("mumps",&((PetscObject)B)->type_name)); 3192 PetscCall(MatSetUp(B)); 3193 3194 PetscCall(PetscNewLog(B,&mumps)); 3195 3196 B->ops->view = MatView_MUMPS; 3197 B->ops->getinfo = MatGetInfo_MUMPS; 3198 3199 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverType_C",MatFactorGetSolverType_mumps)); 3200 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorSetSchurIS_C",MatFactorSetSchurIS_MUMPS)); 3201 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorCreateSchurComplement_C",MatFactorCreateSchurComplement_MUMPS)); 3202 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsSetIcntl_C",MatMumpsSetIcntl_MUMPS)); 3203 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetIcntl_C",MatMumpsGetIcntl_MUMPS)); 3204 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsSetCntl_C",MatMumpsSetCntl_MUMPS)); 3205 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetCntl_C",MatMumpsGetCntl_MUMPS)); 3206 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInfo_C",MatMumpsGetInfo_MUMPS)); 3207 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetInfog_C",MatMumpsGetInfog_MUMPS)); 3208 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetRinfo_C",MatMumpsGetRinfo_MUMPS)); 3209 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMumpsGetRinfog_C",MatMumpsGetRinfog_MUMPS)); 3210 3211 if (ftype == MAT_FACTOR_LU) { 3212 B->ops->lufactorsymbolic = MatLUFactorSymbolic_AIJMUMPS; 3213 B->factortype = MAT_FACTOR_LU; 3214 if (isSeqSELL) mumps->ConvertToTriples = MatConvertToTriples_seqsell_seqaij; 3215 else SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"To be implemented"); 3216 mumps->sym = 0; 3217 PetscCall(PetscStrallocpy(MATORDERINGEXTERNAL,(char**)&B->preferredordering[MAT_FACTOR_LU])); 3218 } else SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"To be implemented"); 3219 3220 /* set solvertype */ 3221 PetscCall(PetscFree(B->solvertype)); 3222 PetscCall(PetscStrallocpy(MATSOLVERMUMPS,&B->solvertype)); 3223 PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)A),&size)); 3224 if (size == 1) { 3225 /* MUMPS option -mat_mumps_icntl_7 1 is automatically set if PETSc ordering is passed into symbolic factorization */ 3226 B->canuseordering = PETSC_TRUE; 3227 } 3228 B->ops->destroy = MatDestroy_MUMPS; 3229 B->data = (void*)mumps; 3230 3231 PetscCall(PetscInitializeMUMPS(A,mumps)); 3232 3233 *F = B; 3234 mumps->matstruc = DIFFERENT_NONZERO_PATTERN; 3235 PetscFunctionReturn(0); 3236 } 3237 3238 PETSC_EXTERN PetscErrorCode MatSolverTypeRegister_MUMPS(void) 3239 { 3240 PetscFunctionBegin; 3241 PetscCall(MatSolverTypeRegister(MATSOLVERMUMPS,MATMPIAIJ,MAT_FACTOR_LU,MatGetFactor_aij_mumps)); 3242 PetscCall(MatSolverTypeRegister(MATSOLVERMUMPS,MATMPIAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_aij_mumps)); 3243 PetscCall(MatSolverTypeRegister(MATSOLVERMUMPS,MATMPIBAIJ,MAT_FACTOR_LU,MatGetFactor_baij_mumps)); 3244 PetscCall(MatSolverTypeRegister(MATSOLVERMUMPS,MATMPIBAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_baij_mumps)); 3245 PetscCall(MatSolverTypeRegister(MATSOLVERMUMPS,MATMPISBAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_sbaij_mumps)); 3246 PetscCall(MatSolverTypeRegister(MATSOLVERMUMPS,MATSEQAIJ,MAT_FACTOR_LU,MatGetFactor_aij_mumps)); 3247 PetscCall(MatSolverTypeRegister(MATSOLVERMUMPS,MATSEQAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_aij_mumps)); 3248 PetscCall(MatSolverTypeRegister(MATSOLVERMUMPS,MATSEQBAIJ,MAT_FACTOR_LU,MatGetFactor_baij_mumps)); 3249 PetscCall(MatSolverTypeRegister(MATSOLVERMUMPS,MATSEQBAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_baij_mumps)); 3250 PetscCall(MatSolverTypeRegister(MATSOLVERMUMPS,MATSEQSBAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_sbaij_mumps)); 3251 PetscCall(MatSolverTypeRegister(MATSOLVERMUMPS,MATSEQSELL,MAT_FACTOR_LU,MatGetFactor_sell_mumps)); 3252 PetscFunctionReturn(0); 3253 } 3254