1 2 /* 3 Provides an interface to the SuperLU_DIST_2.2 sparse solver 4 */ 5 6 #include <../src/mat/impls/aij/seq/aij.h> 7 #include <../src/mat/impls/aij/mpi/mpiaij.h> 8 #if defined(PETSC_HAVE_STDLIB_H) /* This is to get around weird problem with SuperLU on cray */ 9 #include <stdlib.h> 10 #endif 11 12 #if defined(PETSC_USE_64BIT_INDICES) 13 /* ugly SuperLU_Dist variable telling it to use long long int */ 14 #define _LONGINT 15 #endif 16 17 EXTERN_C_BEGIN 18 #if defined(PETSC_USE_COMPLEX) 19 #include <superlu_zdefs.h> 20 #else 21 #include <superlu_ddefs.h> 22 #endif 23 EXTERN_C_END 24 25 /* 26 GLOBAL - The sparse matrix and right hand side are all stored initially on process 0. Should be called centralized 27 DISTRIBUTED - The sparse matrix and right hand size are initially stored across the entire MPI communicator. 28 */ 29 typedef enum {GLOBAL,DISTRIBUTED} SuperLU_MatInputMode; 30 const char *SuperLU_MatInputModes[] = {"GLOBAL","DISTRIBUTED","SuperLU_MatInputMode","PETSC_",0}; 31 32 typedef struct { 33 int_t nprow,npcol,*row,*col; 34 gridinfo_t grid; 35 superlu_options_t options; 36 SuperMatrix A_sup; 37 ScalePermstruct_t ScalePermstruct; 38 LUstruct_t LUstruct; 39 int StatPrint; 40 SuperLU_MatInputMode MatInputMode; 41 SOLVEstruct_t SOLVEstruct; 42 fact_t FactPattern; 43 MPI_Comm comm_superlu; 44 #if defined(PETSC_USE_COMPLEX) 45 doublecomplex *val; 46 #else 47 double *val; 48 #endif 49 PetscBool matsolve_iscalled,matmatsolve_iscalled; 50 PetscBool CleanUpSuperLU_Dist; /* Flag to clean up (non-global) SuperLU objects during Destroy */ 51 } Mat_SuperLU_DIST; 52 53 extern PetscErrorCode MatFactorInfo_SuperLU_DIST(Mat,PetscViewer); 54 extern PetscErrorCode MatLUFactorNumeric_SuperLU_DIST(Mat,Mat,const MatFactorInfo*); 55 extern PetscErrorCode MatDestroy_SuperLU_DIST(Mat); 56 extern PetscErrorCode MatView_SuperLU_DIST(Mat,PetscViewer); 57 extern PetscErrorCode MatSolve_SuperLU_DIST(Mat,Vec,Vec); 58 extern PetscErrorCode MatLUFactorSymbolic_SuperLU_DIST(Mat,Mat,IS,IS,const MatFactorInfo*); 59 extern PetscErrorCode MatDestroy_MPIAIJ(Mat); 60 61 #undef __FUNCT__ 62 #define __FUNCT__ "MatGetDiagonal_SuperLU_DIST" 63 PetscErrorCode MatGetDiagonal_SuperLU_DIST(Mat A,Vec v) 64 { 65 PetscFunctionBegin; 66 SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"Mat type: SuperLU_DIST factor"); 67 PetscFunctionReturn(0); 68 } 69 70 #undef __FUNCT__ 71 #define __FUNCT__ "MatDestroy_SuperLU_DIST" 72 PetscErrorCode MatDestroy_SuperLU_DIST(Mat A) 73 { 74 PetscErrorCode ierr; 75 Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)A->spptr; 76 PetscBool flg; 77 78 PetscFunctionBegin; 79 if (lu && lu->CleanUpSuperLU_Dist) { 80 /* Deallocate SuperLU_DIST storage */ 81 if (lu->MatInputMode == GLOBAL) { 82 PetscStackCall("SuperLU_DIST:Destroy_CompCol_Matrix_dist",Destroy_CompCol_Matrix_dist(&lu->A_sup)); 83 } else { 84 PetscStackCall("SuperLU_DIST:Destroy_CompRowLoc_Matrix_dist",Destroy_CompRowLoc_Matrix_dist(&lu->A_sup)); 85 if (lu->options.SolveInitialized) { 86 #if defined(PETSC_USE_COMPLEX) 87 PetscStackCall("SuperLU_DIST:zSolveFinalize",zSolveFinalize(&lu->options, &lu->SOLVEstruct)); 88 #else 89 PetscStackCall("SuperLU_DIST:dSolveFinalize",dSolveFinalize(&lu->options, &lu->SOLVEstruct)); 90 #endif 91 } 92 } 93 PetscStackCall("SuperLU_DIST:Destroy_LU",Destroy_LU(A->cmap->N, &lu->grid, &lu->LUstruct)); 94 PetscStackCall("SuperLU_DIST:ScalePermstructFree",ScalePermstructFree(&lu->ScalePermstruct)); 95 PetscStackCall("SuperLU_DIST:LUstructFree",LUstructFree(&lu->LUstruct)); 96 97 /* Release the SuperLU_DIST process grid. */ 98 PetscStackCall("SuperLU_DIST:superlu_gridexit",superlu_gridexit(&lu->grid)); 99 ierr = MPI_Comm_free(&(lu->comm_superlu));CHKERRQ(ierr); 100 } 101 ierr = PetscFree(A->spptr);CHKERRQ(ierr); 102 103 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQAIJ,&flg);CHKERRQ(ierr); 104 if (flg) { 105 ierr = MatDestroy_SeqAIJ(A);CHKERRQ(ierr); 106 } else { 107 ierr = MatDestroy_MPIAIJ(A);CHKERRQ(ierr); 108 } 109 PetscFunctionReturn(0); 110 } 111 112 #undef __FUNCT__ 113 #define __FUNCT__ "MatSolve_SuperLU_DIST" 114 PetscErrorCode MatSolve_SuperLU_DIST(Mat A,Vec b_mpi,Vec x) 115 { 116 Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)A->spptr; 117 PetscErrorCode ierr; 118 PetscMPIInt size; 119 PetscInt m=A->rmap->n,M=A->rmap->N,N=A->cmap->N; 120 SuperLUStat_t stat; 121 double berr[1]; 122 PetscScalar *bptr; 123 PetscInt nrhs=1; 124 Vec x_seq; 125 IS iden; 126 VecScatter scat; 127 int info; /* SuperLU_Dist info code is ALWAYS an int, even with long long indices */ 128 static PetscBool cite = PETSC_FALSE; 129 130 PetscFunctionBegin; 131 ierr = PetscCitationsRegister("@article{lidemmel03,\n author = {Xiaoye S. Li and James W. Demmel},\n title = {{SuperLU_DIST}: A Scalable Distributed-Memory Sparse Direct\n Solver for Unsymmetric Linear Systems},\n journal = {ACM Trans. Mathematical Software},\n volume = {29},\n number = {2},\n pages = {110-140},\n year = 2003\n}\n",&cite);CHKERRQ(ierr); 132 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);CHKERRQ(ierr); 133 if (size > 1 && lu->MatInputMode == GLOBAL) { 134 /* global mat input, convert b to x_seq */ 135 ierr = VecCreateSeq(PETSC_COMM_SELF,N,&x_seq);CHKERRQ(ierr); 136 ierr = ISCreateStride(PETSC_COMM_SELF,N,0,1,&iden);CHKERRQ(ierr); 137 ierr = VecScatterCreate(b_mpi,iden,x_seq,iden,&scat);CHKERRQ(ierr); 138 ierr = ISDestroy(&iden);CHKERRQ(ierr); 139 140 ierr = VecScatterBegin(scat,b_mpi,x_seq,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 141 ierr = VecScatterEnd(scat,b_mpi,x_seq,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 142 ierr = VecGetArray(x_seq,&bptr);CHKERRQ(ierr); 143 } else { /* size==1 || distributed mat input */ 144 if (lu->options.SolveInitialized && !lu->matsolve_iscalled) { 145 /* see comments in MatMatSolve() */ 146 #if defined(PETSC_USE_COMPLEX) 147 PetscStackCall("SuperLU_DIST:zSolveFinalize",zSolveFinalize(&lu->options, &lu->SOLVEstruct)); 148 #else 149 PetscStackCall("SuperLU_DIST:dSolveFinalize",dSolveFinalize(&lu->options, &lu->SOLVEstruct)); 150 #endif 151 lu->options.SolveInitialized = NO; 152 } 153 ierr = VecCopy(b_mpi,x);CHKERRQ(ierr); 154 ierr = VecGetArray(x,&bptr);CHKERRQ(ierr); 155 } 156 157 if (lu->options.Fact != FACTORED) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"SuperLU_DIST options.Fact mush equal FACTORED"); 158 159 PetscStackCall("SuperLU_DIST:PStatInit",PStatInit(&stat)); /* Initialize the statistics variables. */ 160 if (lu->MatInputMode == GLOBAL) { 161 #if defined(PETSC_USE_COMPLEX) 162 PetscStackCall("SuperLU_DIST:pzgssvx_ABglobal",pzgssvx_ABglobal(&lu->options,&lu->A_sup,&lu->ScalePermstruct,(doublecomplex*)bptr,M,nrhs,&lu->grid,&lu->LUstruct,berr,&stat,&info)); 163 #else 164 PetscStackCall("SuperLU_DIST:pdgssvx_ABglobal",pdgssvx_ABglobal(&lu->options, &lu->A_sup, &lu->ScalePermstruct,bptr,M,nrhs,&lu->grid,&lu->LUstruct,berr,&stat,&info)); 165 #endif 166 } else { /* distributed mat input */ 167 #if defined(PETSC_USE_COMPLEX) 168 PetscStackCall("SuperLU_DIST:pzgssvx",pzgssvx(&lu->options,&lu->A_sup,&lu->ScalePermstruct,(doublecomplex*)bptr,m,nrhs,&lu->grid,&lu->LUstruct,&lu->SOLVEstruct,berr,&stat,&info)); 169 #else 170 PetscStackCall("SuperLU_DIST:pdgssvx",pdgssvx(&lu->options,&lu->A_sup,&lu->ScalePermstruct,bptr,m,nrhs,&lu->grid,&lu->LUstruct,&lu->SOLVEstruct,berr,&stat,&info)); 171 #endif 172 } 173 if (info) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"pdgssvx fails, info: %d\n",info); 174 175 if (lu->options.PrintStat) PStatPrint(&lu->options, &stat, &lu->grid); /* Print the statistics. */ 176 PetscStackCall("SuperLU_DIST:PStatFree",PStatFree(&stat)); 177 178 if (size > 1 && lu->MatInputMode == GLOBAL) { 179 /* convert seq x to mpi x */ 180 ierr = VecRestoreArray(x_seq,&bptr);CHKERRQ(ierr); 181 ierr = VecScatterBegin(scat,x_seq,x,INSERT_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 182 ierr = VecScatterEnd(scat,x_seq,x,INSERT_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 183 ierr = VecScatterDestroy(&scat);CHKERRQ(ierr); 184 ierr = VecDestroy(&x_seq);CHKERRQ(ierr); 185 } else { 186 ierr = VecRestoreArray(x,&bptr);CHKERRQ(ierr); 187 188 lu->matsolve_iscalled = PETSC_TRUE; 189 lu->matmatsolve_iscalled = PETSC_FALSE; 190 } 191 PetscFunctionReturn(0); 192 } 193 194 #undef __FUNCT__ 195 #define __FUNCT__ "MatMatSolve_SuperLU_DIST" 196 PetscErrorCode MatMatSolve_SuperLU_DIST(Mat A,Mat B_mpi,Mat X) 197 { 198 Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)A->spptr; 199 PetscErrorCode ierr; 200 PetscMPIInt size; 201 PetscInt M=A->rmap->N,m=A->rmap->n,nrhs; 202 SuperLUStat_t stat; 203 double berr[1]; 204 PetscScalar *bptr; 205 int info; /* SuperLU_Dist info code is ALWAYS an int, even with long long indices */ 206 PetscBool flg; 207 208 PetscFunctionBegin; 209 if (lu->options.Fact != FACTORED) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"SuperLU_DIST options.Fact mush equal FACTORED"); 210 ierr = PetscObjectTypeCompareAny((PetscObject)B_mpi,&flg,MATSEQDENSE,MATMPIDENSE,NULL);CHKERRQ(ierr); 211 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix B must be MATDENSE matrix"); 212 ierr = PetscObjectTypeCompareAny((PetscObject)X,&flg,MATSEQDENSE,MATMPIDENSE,NULL);CHKERRQ(ierr); 213 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix X must be MATDENSE matrix"); 214 215 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);CHKERRQ(ierr); 216 if (size > 1 && lu->MatInputMode == GLOBAL) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatInputMode=GLOBAL for nproc %d>1 is not supported",size); 217 /* size==1 or distributed mat input */ 218 if (lu->options.SolveInitialized && !lu->matmatsolve_iscalled) { 219 /* communication pattern of SOLVEstruct is unlikely created for matmatsolve, 220 thus destroy it and create a new SOLVEstruct. 221 Otherwise it may result in memory corruption or incorrect solution 222 See src/mat/examples/tests/ex125.c */ 223 #if defined(PETSC_USE_COMPLEX) 224 PetscStackCall("SuperLU_DIST:zSolveFinalize",zSolveFinalize(&lu->options, &lu->SOLVEstruct)); 225 #else 226 PetscStackCall("SuperLU_DIST:dSolveFinalize",dSolveFinalize(&lu->options, &lu->SOLVEstruct)); 227 #endif 228 lu->options.SolveInitialized = NO; 229 } 230 ierr = MatCopy(B_mpi,X,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 231 232 ierr = MatGetSize(B_mpi,NULL,&nrhs);CHKERRQ(ierr); 233 234 PetscStackCall("SuperLU_DIST:PStatInit",PStatInit(&stat)); /* Initialize the statistics variables. */ 235 ierr = MatDenseGetArray(X,&bptr);CHKERRQ(ierr); 236 if (lu->MatInputMode == GLOBAL) { /* size == 1 */ 237 #if defined(PETSC_USE_COMPLEX) 238 PetscStackCall("SuperLU_DIST:pzgssvx_ABglobal",pzgssvx_ABglobal(&lu->options, &lu->A_sup, &lu->ScalePermstruct,(doublecomplex*)bptr, M, nrhs,&lu->grid, &lu->LUstruct, berr, &stat, &info)); 239 #else 240 PetscStackCall("SuperLU_DIST:pdgssvx_ABglobal",pdgssvx_ABglobal(&lu->options, &lu->A_sup, &lu->ScalePermstruct,bptr, M, nrhs, &lu->grid, &lu->LUstruct, berr, &stat, &info)); 241 #endif 242 } else { /* distributed mat input */ 243 #if defined(PETSC_USE_COMPLEX) 244 PetscStackCall("SuperLU_DIST:pzgssvx",pzgssvx(&lu->options,&lu->A_sup,&lu->ScalePermstruct,(doublecomplex*)bptr,m,nrhs,&lu->grid, &lu->LUstruct,&lu->SOLVEstruct,berr,&stat,&info)); 245 #else 246 PetscStackCall("SuperLU_DIST:pdgssvx",pdgssvx(&lu->options,&lu->A_sup,&lu->ScalePermstruct,bptr,m,nrhs,&lu->grid,&lu->LUstruct,&lu->SOLVEstruct,berr,&stat,&info)); 247 #endif 248 } 249 if (info) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"pdgssvx fails, info: %d\n",info); 250 ierr = MatDenseRestoreArray(X,&bptr);CHKERRQ(ierr); 251 252 if (lu->options.PrintStat) PStatPrint(&lu->options, &stat, &lu->grid); /* Print the statistics. */ 253 PetscStackCall("SuperLU_DIST:PStatFree",PStatFree(&stat)); 254 lu->matsolve_iscalled = PETSC_FALSE; 255 lu->matmatsolve_iscalled = PETSC_TRUE; 256 PetscFunctionReturn(0); 257 } 258 259 260 #undef __FUNCT__ 261 #define __FUNCT__ "MatLUFactorNumeric_SuperLU_DIST" 262 PetscErrorCode MatLUFactorNumeric_SuperLU_DIST(Mat F,Mat A,const MatFactorInfo *info) 263 { 264 Mat *tseq,A_seq = NULL; 265 Mat_SeqAIJ *aa,*bb; 266 Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)(F)->spptr; 267 PetscErrorCode ierr; 268 PetscInt M=A->rmap->N,N=A->cmap->N,i,*ai,*aj,*bi,*bj,nz,rstart,*garray, 269 m=A->rmap->n, colA_start,j,jcol,jB,countA,countB,*bjj,*ajj; 270 int sinfo; /* SuperLU_Dist info flag is always an int even with long long indices */ 271 PetscMPIInt size; 272 SuperLUStat_t stat; 273 double *berr=0; 274 IS isrow; 275 Mat F_diag=NULL; 276 #if defined(PETSC_USE_COMPLEX) 277 doublecomplex *av, *bv; 278 #else 279 double *av, *bv; 280 #endif 281 282 PetscFunctionBegin; 283 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);CHKERRQ(ierr); 284 285 if (lu->MatInputMode == GLOBAL) { /* global mat input */ 286 if (size > 1) { /* convert mpi A to seq mat A */ 287 ierr = ISCreateStride(PETSC_COMM_SELF,M,0,1,&isrow);CHKERRQ(ierr); 288 ierr = MatGetSubMatrices(A,1,&isrow,&isrow,MAT_INITIAL_MATRIX,&tseq);CHKERRQ(ierr); 289 ierr = ISDestroy(&isrow);CHKERRQ(ierr); 290 291 A_seq = *tseq; 292 ierr = PetscFree(tseq);CHKERRQ(ierr); 293 aa = (Mat_SeqAIJ*)A_seq->data; 294 } else { 295 PetscBool flg; 296 ierr = PetscObjectTypeCompare((PetscObject)A,MATMPIAIJ,&flg);CHKERRQ(ierr); 297 if (flg) { 298 Mat_MPIAIJ *At = (Mat_MPIAIJ*)A->data; 299 A = At->A; 300 } 301 aa = (Mat_SeqAIJ*)A->data; 302 } 303 304 /* Convert Petsc NR matrix to SuperLU_DIST NC. 305 Note: memories of lu->val, col and row are allocated by CompRow_to_CompCol_dist()! */ 306 if (lu->options.Fact != DOFACT) {/* successive numeric factorization, sparsity pattern is reused. */ 307 PetscStackCall("SuperLU_DIST:Destroy_CompCol_Matrix_dist",Destroy_CompCol_Matrix_dist(&lu->A_sup)); 308 if (lu->FactPattern == SamePattern_SameRowPerm) { 309 lu->options.Fact = SamePattern_SameRowPerm; /* matrix has similar numerical values */ 310 } else { /* lu->FactPattern == SamePattern */ 311 PetscStackCall("SuperLU_DIST:Destroy_LU",Destroy_LU(N, &lu->grid, &lu->LUstruct)); 312 lu->options.Fact = SamePattern; 313 } 314 } 315 #if defined(PETSC_USE_COMPLEX) 316 PetscStackCall("SuperLU_DIST:zCompRow_to_CompCol_dist",zCompRow_to_CompCol_dist(M,N,aa->nz,(doublecomplex*)aa->a,(int_t*)aa->j,(int_t*)aa->i,&lu->val,&lu->col, &lu->row)); 317 #else 318 PetscStackCall("SuperLU_DIST:dCompRow_to_CompCol_dist",dCompRow_to_CompCol_dist(M,N,aa->nz,aa->a,(int_t*)aa->j,(int_t*)aa->i,&lu->val, &lu->col, &lu->row)); 319 #endif 320 321 /* Create compressed column matrix A_sup. */ 322 #if defined(PETSC_USE_COMPLEX) 323 PetscStackCall("SuperLU_DIST:zCreate_CompCol_Matrix_dist",zCreate_CompCol_Matrix_dist(&lu->A_sup, M, N, aa->nz, lu->val, lu->col, lu->row, SLU_NC, SLU_Z, SLU_GE)); 324 #else 325 PetscStackCall("SuperLU_DIST:dCreate_CompCol_Matrix_dist",dCreate_CompCol_Matrix_dist(&lu->A_sup, M, N, aa->nz, lu->val, lu->col, lu->row, SLU_NC, SLU_D, SLU_GE)); 326 #endif 327 } else { /* distributed mat input */ 328 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)A->data; 329 aa=(Mat_SeqAIJ*)(mat->A)->data; 330 bb=(Mat_SeqAIJ*)(mat->B)->data; 331 ai=aa->i; aj=aa->j; 332 bi=bb->i; bj=bb->j; 333 #if defined(PETSC_USE_COMPLEX) 334 av=(doublecomplex*)aa->a; 335 bv=(doublecomplex*)bb->a; 336 #else 337 av=aa->a; 338 bv=bb->a; 339 #endif 340 rstart = A->rmap->rstart; 341 nz = aa->nz + bb->nz; 342 garray = mat->garray; 343 344 if (lu->options.Fact == DOFACT) { /* first numeric factorization */ 345 #if defined(PETSC_USE_COMPLEX) 346 PetscStackCall("SuperLU_DIST:zallocateA_dist",zallocateA_dist(m, nz, &lu->val, &lu->col, &lu->row)); 347 #else 348 PetscStackCall("SuperLU_DIST:dallocateA_dist",dallocateA_dist(m, nz, &lu->val, &lu->col, &lu->row)); 349 #endif 350 } else { /* successive numeric factorization, sparsity pattern and perm_c are reused. */ 351 /* Destroy_CompRowLoc_Matrix_dist(&lu->A_sup); */ /* this leads to crash! However, see SuperLU_DIST_2.5/EXAMPLE/pzdrive2.c */ 352 if (lu->FactPattern == SamePattern_SameRowPerm) { 353 lu->options.Fact = SamePattern_SameRowPerm; /* matrix has similar numerical values */ 354 } else { 355 PetscStackCall("SuperLU_DIST:Destroy_LU",Destroy_LU(N, &lu->grid, &lu->LUstruct)); /* Deallocate storage associated with the L and U matrices. */ 356 lu->options.Fact = SamePattern; 357 } 358 } 359 nz = 0; 360 for (i=0; i<m; i++) { 361 lu->row[i] = nz; 362 countA = ai[i+1] - ai[i]; 363 countB = bi[i+1] - bi[i]; 364 if (aj) { 365 ajj = aj + ai[i]; /* ptr to the beginning of this row */ 366 } else { 367 ajj = NULL; 368 } 369 bjj = bj + bi[i]; 370 371 /* B part, smaller col index */ 372 if (aj) { 373 colA_start = rstart + ajj[0]; /* the smallest global col index of A */ 374 } else { /* superlu_dist does not require matrix has diagonal entries, thus aj=NULL would work */ 375 colA_start = rstart; 376 } 377 jB = 0; 378 for (j=0; j<countB; j++) { 379 jcol = garray[bjj[j]]; 380 if (jcol > colA_start) { 381 jB = j; 382 break; 383 } 384 lu->col[nz] = jcol; 385 lu->val[nz++] = *bv++; 386 if (j==countB-1) jB = countB; 387 } 388 389 /* A part */ 390 for (j=0; j<countA; j++) { 391 lu->col[nz] = rstart + ajj[j]; 392 lu->val[nz++] = *av++; 393 } 394 395 /* B part, larger col index */ 396 for (j=jB; j<countB; j++) { 397 lu->col[nz] = garray[bjj[j]]; 398 lu->val[nz++] = *bv++; 399 } 400 } 401 lu->row[m] = nz; 402 #if defined(PETSC_USE_COMPLEX) 403 PetscStackCall("SuperLU_DIST:zCreate_CompRowLoc_Matrix_dist",zCreate_CompRowLoc_Matrix_dist(&lu->A_sup, M, N, nz, m, rstart,lu->val, lu->col, lu->row, SLU_NR_loc, SLU_Z, SLU_GE)); 404 #else 405 PetscStackCall("SuperLU_DIST:dCreate_CompRowLoc_Matrix_dist",dCreate_CompRowLoc_Matrix_dist(&lu->A_sup, M, N, nz, m, rstart,lu->val, lu->col, lu->row, SLU_NR_loc, SLU_D, SLU_GE)); 406 #endif 407 } 408 409 /* Factor the matrix. */ 410 PetscStackCall("SuperLU_DIST:PStatInit",PStatInit(&stat)); /* Initialize the statistics variables. */ 411 if (lu->MatInputMode == GLOBAL) { /* global mat input */ 412 #if defined(PETSC_USE_COMPLEX) 413 PetscStackCall("SuperLU_DIST:pzgssvx_ABglobal",pzgssvx_ABglobal(&lu->options, &lu->A_sup, &lu->ScalePermstruct, 0, M, 0,&lu->grid, &lu->LUstruct, berr, &stat, &sinfo)); 414 #else 415 PetscStackCall("SuperLU_DIST:pdgssvx_ABglobal",pdgssvx_ABglobal(&lu->options, &lu->A_sup, &lu->ScalePermstruct, 0, M, 0,&lu->grid, &lu->LUstruct, berr, &stat, &sinfo)); 416 #endif 417 } else { /* distributed mat input */ 418 #if defined(PETSC_USE_COMPLEX) 419 PetscStackCall("SuperLU_DIST:pzgssvx",pzgssvx(&lu->options, &lu->A_sup, &lu->ScalePermstruct, 0, m, 0, &lu->grid,&lu->LUstruct, &lu->SOLVEstruct, berr, &stat, &sinfo)); 420 #else 421 PetscStackCall("SuperLU_DIST:pdgssvx",pdgssvx(&lu->options, &lu->A_sup, &lu->ScalePermstruct, 0, m, 0, &lu->grid,&lu->LUstruct, &lu->SOLVEstruct, berr, &stat, &sinfo)); 422 #endif 423 } 424 425 if (sinfo > 0) { 426 if (A->erroriffailure) { 427 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot in row %D",sinfo); 428 } else { 429 if (sinfo <= lu->A_sup.ncol) { 430 F->errortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 431 ierr = PetscInfo1(F,"U(i,i) is exactly zero, i= %D\n",sinfo);CHKERRQ(ierr); 432 } else if (sinfo > lu->A_sup.ncol) { 433 /* 434 number of bytes allocated when memory allocation 435 failure occurred, plus A->ncol. 436 */ 437 F->errortype = MAT_FACTOR_OUTMEMORY; 438 ierr = PetscInfo1(F,"Number of bytes allocated when memory allocation fails %D\n",sinfo);CHKERRQ(ierr); 439 } 440 } 441 } else if (sinfo < 0) { 442 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB, "info = %D, argument in p*gssvx() had an illegal value", sinfo); 443 } 444 445 if (lu->MatInputMode == GLOBAL && size > 1) { 446 ierr = MatDestroy(&A_seq);CHKERRQ(ierr); 447 } 448 449 if (lu->options.PrintStat) { 450 PStatPrint(&lu->options, &stat, &lu->grid); /* Print the statistics. */ 451 } 452 PStatFree(&stat); 453 if (size > 1) { 454 F_diag = ((Mat_MPIAIJ*)(F)->data)->A; 455 F_diag->assembled = PETSC_TRUE; 456 } 457 (F)->assembled = PETSC_TRUE; 458 (F)->preallocated = PETSC_TRUE; 459 lu->options.Fact = FACTORED; /* The factored form of A is supplied. Local option used by this func. only */ 460 PetscFunctionReturn(0); 461 } 462 463 /* Note the Petsc r and c permutations are ignored */ 464 #undef __FUNCT__ 465 #define __FUNCT__ "MatLUFactorSymbolic_SuperLU_DIST" 466 PetscErrorCode MatLUFactorSymbolic_SuperLU_DIST(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info) 467 { 468 Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)F->spptr; 469 PetscInt M = A->rmap->N,N=A->cmap->N; 470 471 PetscFunctionBegin; 472 /* Initialize the SuperLU process grid. */ 473 PetscStackCall("SuperLU_DIST:superlu_gridinit",superlu_gridinit(lu->comm_superlu, lu->nprow, lu->npcol, &lu->grid)); 474 475 /* Initialize ScalePermstruct and LUstruct. */ 476 PetscStackCall("SuperLU_DIST:ScalePermstructInit",ScalePermstructInit(M, N, &lu->ScalePermstruct)); 477 PetscStackCall("SuperLU_DIST:LUstructInit",LUstructInit(N, &lu->LUstruct)); 478 F->ops->lufactornumeric = MatLUFactorNumeric_SuperLU_DIST; 479 F->ops->solve = MatSolve_SuperLU_DIST; 480 F->ops->matsolve = MatMatSolve_SuperLU_DIST; 481 lu->CleanUpSuperLU_Dist = PETSC_TRUE; 482 PetscFunctionReturn(0); 483 } 484 485 #undef __FUNCT__ 486 #define __FUNCT__ "MatFactorGetSolverPackage_aij_superlu_dist" 487 PetscErrorCode MatFactorGetSolverPackage_aij_superlu_dist(Mat A,const MatSolverPackage *type) 488 { 489 PetscFunctionBegin; 490 *type = MATSOLVERSUPERLU_DIST; 491 PetscFunctionReturn(0); 492 } 493 494 #undef __FUNCT__ 495 #define __FUNCT__ "MatGetFactor_aij_superlu_dist" 496 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_superlu_dist(Mat A,MatFactorType ftype,Mat *F) 497 { 498 Mat B; 499 Mat_SuperLU_DIST *lu; 500 PetscErrorCode ierr; 501 PetscInt M=A->rmap->N,N=A->cmap->N,indx; 502 PetscMPIInt size; 503 superlu_options_t options; 504 PetscBool flg; 505 const char *colperm[] = {"NATURAL","MMD_AT_PLUS_A","MMD_ATA","METIS_AT_PLUS_A","PARMETIS"}; 506 const char *rowperm[] = {"LargeDiag","NATURAL"}; 507 const char *factPattern[] = {"SamePattern","SamePattern_SameRowPerm"}; 508 PetscBool set; 509 510 PetscFunctionBegin; 511 /* Create the factorization matrix */ 512 ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr); 513 ierr = MatSetSizes(B,A->rmap->n,A->cmap->n,M,N);CHKERRQ(ierr); 514 ierr = MatSetType(B,((PetscObject)A)->type_name);CHKERRQ(ierr); 515 ierr = MatSeqAIJSetPreallocation(B,0,NULL); 516 ierr = MatMPIAIJSetPreallocation(B,0,NULL,0,NULL);CHKERRQ(ierr); 517 518 B->ops->lufactorsymbolic = MatLUFactorSymbolic_SuperLU_DIST; 519 B->ops->view = MatView_SuperLU_DIST; 520 B->ops->destroy = MatDestroy_SuperLU_DIST; 521 B->ops->getdiagonal = MatGetDiagonal_SuperLU_DIST; 522 523 ierr = PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverPackage_C",MatFactorGetSolverPackage_aij_superlu_dist);CHKERRQ(ierr); 524 525 B->factortype = MAT_FACTOR_LU; 526 527 ierr = PetscNewLog(B,&lu);CHKERRQ(ierr); 528 B->spptr = lu; 529 530 /* Set the default input options: 531 options.Fact = DOFACT; 532 options.Equil = YES; 533 options.ParSymbFact = NO; 534 options.ColPerm = METIS_AT_PLUS_A; 535 options.RowPerm = LargeDiag; 536 options.ReplaceTinyPivot = YES; 537 options.IterRefine = DOUBLE; 538 options.Trans = NOTRANS; 539 options.SolveInitialized = NO; -hold the communication pattern used MatSolve() and MatMatSolve() 540 options.RefineInitialized = NO; 541 options.PrintStat = YES; 542 */ 543 set_default_options_dist(&options); 544 545 ierr = MPI_Comm_dup(PetscObjectComm((PetscObject)A),&(lu->comm_superlu));CHKERRQ(ierr); 546 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);CHKERRQ(ierr); 547 /* Default num of process columns and rows */ 548 lu->npcol = (int_t) (0.5 + PetscSqrtReal((PetscReal)size)); 549 if (!lu->npcol) lu->npcol = 1; 550 while (lu->npcol > 0) { 551 lu->nprow = (int_t) (size/lu->npcol); 552 if (size == lu->nprow * lu->npcol) break; 553 lu->npcol--; 554 } 555 556 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)A),((PetscObject)A)->prefix,"SuperLU_Dist Options","Mat");CHKERRQ(ierr); 557 ierr = PetscOptionsInt("-mat_superlu_dist_r","Number rows in processor partition","None",lu->nprow,(PetscInt*)&lu->nprow,NULL);CHKERRQ(ierr); 558 ierr = PetscOptionsInt("-mat_superlu_dist_c","Number columns in processor partition","None",lu->npcol,(PetscInt*)&lu->npcol,NULL);CHKERRQ(ierr); 559 if (size != lu->nprow * lu->npcol) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Number of processes %d must equal to nprow %d * npcol %d",size,lu->nprow,lu->npcol); 560 561 lu->MatInputMode = DISTRIBUTED; 562 563 ierr = PetscOptionsEnum("-mat_superlu_dist_matinput","Matrix input mode (global or distributed)","None",SuperLU_MatInputModes,(PetscEnum)lu->MatInputMode,(PetscEnum*)&lu->MatInputMode,NULL);CHKERRQ(ierr); 564 if (lu->MatInputMode == DISTRIBUTED && size == 1) lu->MatInputMode = GLOBAL; 565 566 ierr = PetscOptionsBool("-mat_superlu_dist_equil","Equilibrate matrix","None",options.Equil ? PETSC_TRUE : PETSC_FALSE,&flg,&set);CHKERRQ(ierr); 567 if (set && !flg) options.Equil = NO; 568 569 ierr = PetscOptionsEList("-mat_superlu_dist_rowperm","Row permutation","None",rowperm,2,rowperm[0],&indx,&flg);CHKERRQ(ierr); 570 if (flg) { 571 switch (indx) { 572 case 0: 573 options.RowPerm = LargeDiag; 574 break; 575 case 1: 576 options.RowPerm = NOROWPERM; 577 break; 578 } 579 } 580 581 ierr = PetscOptionsEList("-mat_superlu_dist_colperm","Column permutation","None",colperm,5,colperm[3],&indx,&flg);CHKERRQ(ierr); 582 if (flg) { 583 switch (indx) { 584 case 0: 585 options.ColPerm = NATURAL; 586 break; 587 case 1: 588 options.ColPerm = MMD_AT_PLUS_A; 589 break; 590 case 2: 591 options.ColPerm = MMD_ATA; 592 break; 593 case 3: 594 options.ColPerm = METIS_AT_PLUS_A; 595 break; 596 case 4: 597 options.ColPerm = PARMETIS; /* only works for np>1 */ 598 break; 599 default: 600 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown column permutation"); 601 } 602 } 603 604 options.ReplaceTinyPivot = NO; 605 ierr = PetscOptionsBool("-mat_superlu_dist_replacetinypivot","Replace tiny pivots","None",options.ReplaceTinyPivot ? PETSC_TRUE : PETSC_FALSE,&flg,&set);CHKERRQ(ierr); 606 if (set && flg) options.ReplaceTinyPivot = YES; 607 608 options.ParSymbFact = NO; 609 ierr = PetscOptionsBool("-mat_superlu_dist_parsymbfact","Parallel symbolic factorization","None",PETSC_FALSE,&flg,&set);CHKERRQ(ierr); 610 if (set && flg && size>1) { 611 if (lu->MatInputMode == GLOBAL) { 612 #if defined(PETSC_USE_INFO) 613 ierr = PetscInfo(A,"Warning: '-mat_superlu_dist_parsymbfact' is ignored because MatInputMode=GLOBAL\n");CHKERRQ(ierr); 614 #endif 615 } else { 616 #if defined(PETSC_HAVE_PARMETIS) 617 options.ParSymbFact = YES; 618 options.ColPerm = PARMETIS; /* in v2.2, PARMETIS is forced for ParSymbFact regardless of user ordering setting */ 619 #else 620 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"parsymbfact needs PARMETIS"); 621 #endif 622 } 623 } 624 625 lu->FactPattern = SamePattern; 626 ierr = PetscOptionsEList("-mat_superlu_dist_fact","Sparsity pattern for repeated matrix factorization","None",factPattern,2,factPattern[0],&indx,&flg);CHKERRQ(ierr); 627 if (flg) { 628 switch (indx) { 629 case 0: 630 lu->FactPattern = SamePattern; 631 break; 632 case 1: 633 lu->FactPattern = SamePattern_SameRowPerm; 634 break; 635 } 636 } 637 638 options.IterRefine = NOREFINE; 639 ierr = PetscOptionsBool("-mat_superlu_dist_iterrefine","Use iterative refinement","None",options.IterRefine == NOREFINE ? PETSC_FALSE : PETSC_TRUE ,&flg,&set);CHKERRQ(ierr); 640 if (set) { 641 if (flg) options.IterRefine = SLU_DOUBLE; 642 else options.IterRefine = NOREFINE; 643 } 644 645 if (PetscLogPrintInfo) options.PrintStat = YES; 646 else options.PrintStat = NO; 647 ierr = PetscOptionsBool("-mat_superlu_dist_statprint","Print factorization information","None",(PetscBool)options.PrintStat,(PetscBool*)&options.PrintStat,NULL);CHKERRQ(ierr); 648 PetscOptionsEnd(); 649 650 lu->options = options; 651 lu->options.Fact = DOFACT; 652 lu->matsolve_iscalled = PETSC_FALSE; 653 lu->matmatsolve_iscalled = PETSC_FALSE; 654 655 *F = B; 656 PetscFunctionReturn(0); 657 } 658 659 #undef __FUNCT__ 660 #define __FUNCT__ "MatSolverPackageRegister_SuperLU_DIST" 661 PETSC_EXTERN PetscErrorCode MatSolverPackageRegister_SuperLU_DIST(void) 662 { 663 PetscErrorCode ierr; 664 PetscFunctionBegin; 665 ierr = MatSolverPackageRegister(MATSOLVERSUPERLU_DIST,MATMPIAIJ, MAT_FACTOR_LU,MatGetFactor_aij_superlu_dist);CHKERRQ(ierr); 666 ierr = MatSolverPackageRegister(MATSOLVERSUPERLU_DIST,MATSEQAIJ, MAT_FACTOR_LU,MatGetFactor_aij_superlu_dist);CHKERRQ(ierr); 667 PetscFunctionReturn(0); 668 } 669 670 #undef __FUNCT__ 671 #define __FUNCT__ "MatFactorInfo_SuperLU_DIST" 672 PetscErrorCode MatFactorInfo_SuperLU_DIST(Mat A,PetscViewer viewer) 673 { 674 Mat_SuperLU_DIST *lu=(Mat_SuperLU_DIST*)A->spptr; 675 superlu_options_t options; 676 PetscErrorCode ierr; 677 678 PetscFunctionBegin; 679 /* check if matrix is superlu_dist type */ 680 if (A->ops->solve != MatSolve_SuperLU_DIST) PetscFunctionReturn(0); 681 682 options = lu->options; 683 ierr = PetscViewerASCIIPrintf(viewer,"SuperLU_DIST run parameters:\n");CHKERRQ(ierr); 684 ierr = PetscViewerASCIIPrintf(viewer," Process grid nprow %D x npcol %D \n",lu->nprow,lu->npcol);CHKERRQ(ierr); 685 ierr = PetscViewerASCIIPrintf(viewer," Equilibrate matrix %s \n",PetscBools[options.Equil != NO]);CHKERRQ(ierr); 686 ierr = PetscViewerASCIIPrintf(viewer," Matrix input mode %d \n",lu->MatInputMode);CHKERRQ(ierr); 687 ierr = PetscViewerASCIIPrintf(viewer," Replace tiny pivots %s \n",PetscBools[options.ReplaceTinyPivot != NO]);CHKERRQ(ierr); 688 ierr = PetscViewerASCIIPrintf(viewer," Use iterative refinement %s \n",PetscBools[options.IterRefine == SLU_DOUBLE]);CHKERRQ(ierr); 689 ierr = PetscViewerASCIIPrintf(viewer," Processors in row %d col partition %d \n",lu->nprow,lu->npcol);CHKERRQ(ierr); 690 ierr = PetscViewerASCIIPrintf(viewer," Row permutation %s \n",(options.RowPerm == NOROWPERM) ? "NATURAL" : "LargeDiag");CHKERRQ(ierr); 691 692 switch (options.ColPerm) { 693 case NATURAL: 694 ierr = PetscViewerASCIIPrintf(viewer," Column permutation NATURAL\n");CHKERRQ(ierr); 695 break; 696 case MMD_AT_PLUS_A: 697 ierr = PetscViewerASCIIPrintf(viewer," Column permutation MMD_AT_PLUS_A\n");CHKERRQ(ierr); 698 break; 699 case MMD_ATA: 700 ierr = PetscViewerASCIIPrintf(viewer," Column permutation MMD_ATA\n");CHKERRQ(ierr); 701 break; 702 case METIS_AT_PLUS_A: 703 ierr = PetscViewerASCIIPrintf(viewer," Column permutation METIS_AT_PLUS_A\n");CHKERRQ(ierr); 704 break; 705 case PARMETIS: 706 ierr = PetscViewerASCIIPrintf(viewer," Column permutation PARMETIS\n");CHKERRQ(ierr); 707 break; 708 default: 709 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown column permutation"); 710 } 711 712 ierr = PetscViewerASCIIPrintf(viewer," Parallel symbolic factorization %s \n",PetscBools[options.ParSymbFact != NO]);CHKERRQ(ierr); 713 714 if (lu->FactPattern == SamePattern) { 715 ierr = PetscViewerASCIIPrintf(viewer," Repeated factorization SamePattern\n");CHKERRQ(ierr); 716 } else { 717 ierr = PetscViewerASCIIPrintf(viewer," Repeated factorization SamePattern_SameRowPerm\n");CHKERRQ(ierr); 718 } 719 PetscFunctionReturn(0); 720 } 721 722 #undef __FUNCT__ 723 #define __FUNCT__ "MatView_SuperLU_DIST" 724 PetscErrorCode MatView_SuperLU_DIST(Mat A,PetscViewer viewer) 725 { 726 PetscErrorCode ierr; 727 PetscBool iascii; 728 PetscViewerFormat format; 729 730 PetscFunctionBegin; 731 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 732 if (iascii) { 733 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 734 if (format == PETSC_VIEWER_ASCII_INFO) { 735 ierr = MatFactorInfo_SuperLU_DIST(A,viewer);CHKERRQ(ierr); 736 } 737 } 738 PetscFunctionReturn(0); 739 } 740 741 742 /*MC 743 MATSOLVERSUPERLU_DIST - Parallel direct solver package for LU factorization 744 745 Use ./configure --download-superlu_dist --download-parmetis --download-metis --download-ptscotch to have PETSc installed with SuperLU_DIST 746 747 Use -pc_type lu -pc_factor_mat_solver_package superlu_dist to us this direct solver 748 749 Works with AIJ matrices 750 751 Options Database Keys: 752 + -mat_superlu_dist_r <n> - number of rows in processor partition 753 . -mat_superlu_dist_c <n> - number of columns in processor partition 754 . -mat_superlu_dist_matinput <0,1> - matrix input mode; 0=global, 1=distributed 755 . -mat_superlu_dist_equil - equilibrate the matrix 756 . -mat_superlu_dist_rowperm <LargeDiag,NATURAL> - row permutation 757 . -mat_superlu_dist_colperm <MMD_AT_PLUS_A,MMD_ATA,NATURAL> - column permutation 758 . -mat_superlu_dist_replacetinypivot - replace tiny pivots 759 . -mat_superlu_dist_fact <SamePattern> - (choose one of) SamePattern SamePattern_SameRowPerm 760 . -mat_superlu_dist_iterrefine - use iterative refinement 761 - -mat_superlu_dist_statprint - print factorization information 762 763 Level: beginner 764 765 .seealso: PCLU 766 767 .seealso: PCFactorSetMatSolverPackage(), MatSolverPackage 768 769 M*/ 770 771