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 #include <petsctime.h> 9 #if defined(PETSC_HAVE_STDLIB_H) /* This is to get around weird problem with SuperLU on cray */ 10 #include <stdlib.h> 11 #endif 12 13 #if defined(PETSC_USE_64BIT_INDICES) 14 /* ugly SuperLU_Dist variable telling it to use long long int */ 15 #define _LONGINT 16 #endif 17 18 EXTERN_C_BEGIN 19 #if defined(PETSC_USE_COMPLEX) 20 #include <superlu_zdefs.h> 21 #else 22 #include <superlu_ddefs.h> 23 #endif 24 EXTERN_C_END 25 26 /* 27 GLOBAL - The sparse matrix and right hand side are all stored initially on process 0. Should be called centralized 28 DISTRIBUTED - The sparse matrix and right hand size are initially stored across the entire MPI communicator. 29 */ 30 typedef enum {GLOBAL,DISTRIBUTED} SuperLU_MatInputMode; 31 const char *SuperLU_MatInputModes[] = {"GLOBAL","DISTRIBUTED","SuperLU_MatInputMode","PETSC_",0}; 32 33 typedef struct { 34 int_t nprow,npcol,*row,*col; 35 gridinfo_t grid; 36 superlu_options_t options; 37 SuperMatrix A_sup; 38 ScalePermstruct_t ScalePermstruct; 39 LUstruct_t LUstruct; 40 int StatPrint; 41 SuperLU_MatInputMode MatInputMode; 42 SOLVEstruct_t SOLVEstruct; 43 fact_t FactPattern; 44 MPI_Comm comm_superlu; 45 #if defined(PETSC_USE_COMPLEX) 46 doublecomplex *val; 47 #else 48 double *val; 49 #endif 50 PetscBool matsolve_iscalled,matmatsolve_iscalled; 51 PetscBool CleanUpSuperLU_Dist; /* Flag to clean up (non-global) SuperLU objects during Destroy */ 52 } Mat_SuperLU_DIST; 53 54 extern PetscErrorCode MatFactorInfo_SuperLU_DIST(Mat,PetscViewer); 55 extern PetscErrorCode MatLUFactorNumeric_SuperLU_DIST(Mat,Mat,const MatFactorInfo*); 56 extern PetscErrorCode MatDestroy_SuperLU_DIST(Mat); 57 extern PetscErrorCode MatView_SuperLU_DIST(Mat,PetscViewer); 58 extern PetscErrorCode MatSolve_SuperLU_DIST(Mat,Vec,Vec); 59 extern PetscErrorCode MatLUFactorSymbolic_SuperLU_DIST(Mat,Mat,IS,IS,const MatFactorInfo*); 60 extern PetscErrorCode MatDestroy_MPIAIJ(Mat); 61 62 #undef __FUNCT__ 63 #define __FUNCT__ "MatDestroy_SuperLU_DIST" 64 PetscErrorCode MatDestroy_SuperLU_DIST(Mat A) 65 { 66 PetscErrorCode ierr; 67 Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)A->spptr; 68 PetscBool flg; 69 70 PetscFunctionBegin; 71 if (lu && lu->CleanUpSuperLU_Dist) { 72 /* Deallocate SuperLU_DIST storage */ 73 if (lu->MatInputMode == GLOBAL) { 74 PetscStackCall("SuperLU_DIST:Destroy_CompCol_Matrix_dist",Destroy_CompCol_Matrix_dist(&lu->A_sup)); 75 } else { 76 PetscStackCall("SuperLU_DIST:Destroy_CompRowLoc_Matrix_dist",Destroy_CompRowLoc_Matrix_dist(&lu->A_sup)); 77 if (lu->options.SolveInitialized) { 78 #if defined(PETSC_USE_COMPLEX) 79 PetscStackCall("SuperLU_DIST:zSolveFinalize",zSolveFinalize(&lu->options, &lu->SOLVEstruct)); 80 #else 81 PetscStackCall("SuperLU_DIST:dSolveFinalize",dSolveFinalize(&lu->options, &lu->SOLVEstruct)); 82 #endif 83 } 84 } 85 PetscStackCall("SuperLU_DIST:Destroy_LU",Destroy_LU(A->cmap->N, &lu->grid, &lu->LUstruct)); 86 PetscStackCall("SuperLU_DIST:ScalePermstructFree",ScalePermstructFree(&lu->ScalePermstruct)); 87 PetscStackCall("SuperLU_DIST:LUstructFree",LUstructFree(&lu->LUstruct)); 88 89 /* Release the SuperLU_DIST process grid. */ 90 PetscStackCall("SuperLU_DIST:superlu_gridexit",superlu_gridexit(&lu->grid)); 91 ierr = MPI_Comm_free(&(lu->comm_superlu));CHKERRQ(ierr); 92 } 93 ierr = PetscFree(A->spptr);CHKERRQ(ierr); 94 95 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQAIJ,&flg);CHKERRQ(ierr); 96 if (flg) { 97 ierr = MatDestroy_SeqAIJ(A);CHKERRQ(ierr); 98 } else { 99 ierr = MatDestroy_MPIAIJ(A);CHKERRQ(ierr); 100 } 101 PetscFunctionReturn(0); 102 } 103 104 #undef __FUNCT__ 105 #define __FUNCT__ "MatSolve_SuperLU_DIST" 106 PetscErrorCode MatSolve_SuperLU_DIST(Mat A,Vec b_mpi,Vec x) 107 { 108 Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)A->spptr; 109 PetscErrorCode ierr; 110 PetscMPIInt size; 111 PetscInt m=A->rmap->n,M=A->rmap->N,N=A->cmap->N; 112 SuperLUStat_t stat; 113 double berr[1]; 114 PetscScalar *bptr; 115 PetscInt nrhs=1; 116 Vec x_seq; 117 IS iden; 118 VecScatter scat; 119 int info; /* SuperLU_Dist info code is ALWAYS an int, even with long long indices */ 120 121 PetscFunctionBegin; 122 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);CHKERRQ(ierr); 123 if (size > 1 && lu->MatInputMode == GLOBAL) { 124 /* global mat input, convert b to x_seq */ 125 ierr = VecCreateSeq(PETSC_COMM_SELF,N,&x_seq);CHKERRQ(ierr); 126 ierr = ISCreateStride(PETSC_COMM_SELF,N,0,1,&iden);CHKERRQ(ierr); 127 ierr = VecScatterCreate(b_mpi,iden,x_seq,iden,&scat);CHKERRQ(ierr); 128 ierr = ISDestroy(&iden);CHKERRQ(ierr); 129 130 ierr = VecScatterBegin(scat,b_mpi,x_seq,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 131 ierr = VecScatterEnd(scat,b_mpi,x_seq,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 132 ierr = VecGetArray(x_seq,&bptr);CHKERRQ(ierr); 133 } else { /* size==1 || distributed mat input */ 134 if (lu->options.SolveInitialized && !lu->matsolve_iscalled) { 135 /* see comments in MatMatSolve() */ 136 #if defined(PETSC_USE_COMPLEX) 137 PetscStackCall("SuperLU_DIST:zSolveFinalize",zSolveFinalize(&lu->options, &lu->SOLVEstruct)); 138 #else 139 PetscStackCall("SuperLU_DIST:dSolveFinalize",dSolveFinalize(&lu->options, &lu->SOLVEstruct)); 140 #endif 141 lu->options.SolveInitialized = NO; 142 } 143 ierr = VecCopy(b_mpi,x);CHKERRQ(ierr); 144 ierr = VecGetArray(x,&bptr);CHKERRQ(ierr); 145 } 146 147 if (lu->options.Fact != FACTORED) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"SuperLU_DIST options.Fact mush equal FACTORED"); 148 149 PetscStackCall("SuperLU_DIST:PStatInit",PStatInit(&stat)); /* Initialize the statistics variables. */ 150 if (lu->MatInputMode == GLOBAL) { 151 #if defined(PETSC_USE_COMPLEX) 152 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)); 153 #else 154 PetscStackCall("SuperLU_DIST:pdgssvx_ABglobal",pdgssvx_ABglobal(&lu->options, &lu->A_sup, &lu->ScalePermstruct,bptr,M,nrhs,&lu->grid,&lu->LUstruct,berr,&stat,&info)); 155 #endif 156 } else { /* distributed mat input */ 157 #if defined(PETSC_USE_COMPLEX) 158 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)); 159 #else 160 PetscStackCall("SuperLU_DIST:pdgssvx",pdgssvx(&lu->options,&lu->A_sup,&lu->ScalePermstruct,bptr,m,nrhs,&lu->grid,&lu->LUstruct,&lu->SOLVEstruct,berr,&stat,&info)); 161 #endif 162 } 163 if (info) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"pdgssvx fails, info: %d\n",info); 164 165 if (lu->options.PrintStat) PStatPrint(&lu->options, &stat, &lu->grid); /* Print the statistics. */ 166 PetscStackCall("SuperLU_DIST:PStatFree",PStatFree(&stat)); 167 168 if (size > 1 && lu->MatInputMode == GLOBAL) { 169 /* convert seq x to mpi x */ 170 ierr = VecRestoreArray(x_seq,&bptr);CHKERRQ(ierr); 171 ierr = VecScatterBegin(scat,x_seq,x,INSERT_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 172 ierr = VecScatterEnd(scat,x_seq,x,INSERT_VALUES,SCATTER_REVERSE);CHKERRQ(ierr); 173 ierr = VecScatterDestroy(&scat);CHKERRQ(ierr); 174 ierr = VecDestroy(&x_seq);CHKERRQ(ierr); 175 } else { 176 ierr = VecRestoreArray(x,&bptr);CHKERRQ(ierr); 177 178 lu->matsolve_iscalled = PETSC_TRUE; 179 lu->matmatsolve_iscalled = PETSC_FALSE; 180 } 181 PetscFunctionReturn(0); 182 } 183 184 #undef __FUNCT__ 185 #define __FUNCT__ "MatMatSolve_SuperLU_DIST" 186 PetscErrorCode MatMatSolve_SuperLU_DIST(Mat A,Mat B_mpi,Mat X) 187 { 188 Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)A->spptr; 189 PetscErrorCode ierr; 190 PetscMPIInt size; 191 PetscInt M=A->rmap->N,m=A->rmap->n,nrhs; 192 SuperLUStat_t stat; 193 double berr[1]; 194 PetscScalar *bptr; 195 int info; /* SuperLU_Dist info code is ALWAYS an int, even with long long indices */ 196 PetscBool flg; 197 198 PetscFunctionBegin; 199 if (lu->options.Fact != FACTORED) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"SuperLU_DIST options.Fact mush equal FACTORED"); 200 ierr = PetscObjectTypeCompareAny((PetscObject)B_mpi,&flg,MATSEQDENSE,MATMPIDENSE,NULL);CHKERRQ(ierr); 201 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix B must be MATDENSE matrix"); 202 ierr = PetscObjectTypeCompareAny((PetscObject)X,&flg,MATSEQDENSE,MATMPIDENSE,NULL);CHKERRQ(ierr); 203 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix X must be MATDENSE matrix"); 204 205 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);CHKERRQ(ierr); 206 if (size > 1 && lu->MatInputMode == GLOBAL) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatInputMode=GLOBAL for nproc %d>1 is not supported",size); 207 /* size==1 or distributed mat input */ 208 if (lu->options.SolveInitialized && !lu->matmatsolve_iscalled) { 209 /* communication pattern of SOLVEstruct is unlikely created for matmatsolve, 210 thus destroy it and create a new SOLVEstruct. 211 Otherwise it may result in memory corruption or incorrect solution 212 See src/mat/examples/tests/ex125.c */ 213 #if defined(PETSC_USE_COMPLEX) 214 PetscStackCall("SuperLU_DIST:zSolveFinalize",zSolveFinalize(&lu->options, &lu->SOLVEstruct)); 215 #else 216 PetscStackCall("SuperLU_DIST:dSolveFinalize",dSolveFinalize(&lu->options, &lu->SOLVEstruct)); 217 #endif 218 lu->options.SolveInitialized = NO; 219 } 220 ierr = MatCopy(B_mpi,X,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 221 222 ierr = MatGetSize(B_mpi,NULL,&nrhs);CHKERRQ(ierr); 223 224 PetscStackCall("SuperLU_DIST:PStatInit",PStatInit(&stat)); /* Initialize the statistics variables. */ 225 ierr = MatDenseGetArray(X,&bptr);CHKERRQ(ierr); 226 if (lu->MatInputMode == GLOBAL) { /* size == 1 */ 227 #if defined(PETSC_USE_COMPLEX) 228 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)); 229 #else 230 PetscStackCall("SuperLU_DIST:pdgssvx_ABglobal",pdgssvx_ABglobal(&lu->options, &lu->A_sup, &lu->ScalePermstruct,bptr, M, nrhs, &lu->grid, &lu->LUstruct, berr, &stat, &info)); 231 #endif 232 } else { /* distributed mat input */ 233 #if defined(PETSC_USE_COMPLEX) 234 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)); 235 #else 236 PetscStackCall("SuperLU_DIST:pdgssvx",pdgssvx(&lu->options,&lu->A_sup,&lu->ScalePermstruct,bptr,m,nrhs,&lu->grid,&lu->LUstruct,&lu->SOLVEstruct,berr,&stat,&info)); 237 #endif 238 } 239 if (info) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"pdgssvx fails, info: %d\n",info); 240 ierr = MatDenseRestoreArray(X,&bptr);CHKERRQ(ierr); 241 242 if (lu->options.PrintStat) PStatPrint(&lu->options, &stat, &lu->grid); /* Print the statistics. */ 243 PetscStackCall("SuperLU_DIST:PStatFree",PStatFree(&stat)); 244 lu->matsolve_iscalled = PETSC_FALSE; 245 lu->matmatsolve_iscalled = PETSC_TRUE; 246 PetscFunctionReturn(0); 247 } 248 249 250 #undef __FUNCT__ 251 #define __FUNCT__ "MatLUFactorNumeric_SuperLU_DIST" 252 PetscErrorCode MatLUFactorNumeric_SuperLU_DIST(Mat F,Mat A,const MatFactorInfo *info) 253 { 254 Mat *tseq,A_seq = NULL; 255 Mat_SeqAIJ *aa,*bb; 256 Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)(F)->spptr; 257 PetscErrorCode ierr; 258 PetscInt M=A->rmap->N,N=A->cmap->N,i,*ai,*aj,*bi,*bj,nz,rstart,*garray, 259 m=A->rmap->n, colA_start,j,jcol,jB,countA,countB,*bjj,*ajj; 260 int sinfo; /* SuperLU_Dist info flag is always an int even with long long indices */ 261 PetscMPIInt size; 262 SuperLUStat_t stat; 263 double *berr=0; 264 IS isrow; 265 PetscLogDouble time0,time,time_min,time_max; 266 Mat F_diag=NULL; 267 #if defined(PETSC_USE_COMPLEX) 268 doublecomplex *av, *bv; 269 #else 270 double *av, *bv; 271 #endif 272 273 PetscFunctionBegin; 274 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);CHKERRQ(ierr); 275 276 if (lu->options.PrintStat) { /* collect time for mat conversion */ 277 ierr = MPI_Barrier(PetscObjectComm((PetscObject)A));CHKERRQ(ierr); 278 ierr = PetscTime(&time0);CHKERRQ(ierr); 279 } 280 281 if (lu->MatInputMode == GLOBAL) { /* global mat input */ 282 if (size > 1) { /* convert mpi A to seq mat A */ 283 ierr = ISCreateStride(PETSC_COMM_SELF,M,0,1,&isrow);CHKERRQ(ierr); 284 ierr = MatGetSubMatrices(A,1,&isrow,&isrow,MAT_INITIAL_MATRIX,&tseq);CHKERRQ(ierr); 285 ierr = ISDestroy(&isrow);CHKERRQ(ierr); 286 287 A_seq = *tseq; 288 ierr = PetscFree(tseq);CHKERRQ(ierr); 289 aa = (Mat_SeqAIJ*)A_seq->data; 290 } else { 291 PetscBool flg; 292 ierr = PetscObjectTypeCompare((PetscObject)A,MATMPIAIJ,&flg);CHKERRQ(ierr); 293 if (flg) { 294 Mat_MPIAIJ *At = (Mat_MPIAIJ*)A->data; 295 A = At->A; 296 } 297 aa = (Mat_SeqAIJ*)A->data; 298 } 299 300 /* Convert Petsc NR matrix to SuperLU_DIST NC. 301 Note: memories of lu->val, col and row are allocated by CompRow_to_CompCol_dist()! */ 302 if (lu->options.Fact != DOFACT) {/* successive numeric factorization, sparsity pattern is reused. */ 303 PetscStackCall("SuperLU_DIST:Destroy_CompCol_Matrix_dist",Destroy_CompCol_Matrix_dist(&lu->A_sup)); 304 if (lu->FactPattern == SamePattern_SameRowPerm) { 305 lu->options.Fact = SamePattern_SameRowPerm; /* matrix has similar numerical values */ 306 } else { /* lu->FactPattern == SamePattern */ 307 PetscStackCall("SuperLU_DIST:Destroy_LU",Destroy_LU(N, &lu->grid, &lu->LUstruct)); 308 lu->options.Fact = SamePattern; 309 } 310 } 311 #if defined(PETSC_USE_COMPLEX) 312 PetscStackCall("SuperLU_DIST:zCompRow_to_CompCol_dist",zCompRow_to_CompCol_dist(M,N,aa->nz,(doublecomplex*)aa->a,aa->j,aa->i,&lu->val,&lu->col, &lu->row)); 313 #else 314 PetscStackCall("SuperLU_DIST:dCompRow_to_CompCol_dist",dCompRow_to_CompCol_dist(M,N,aa->nz,aa->a,aa->j,aa->i,&lu->val, &lu->col, &lu->row)); 315 #endif 316 317 /* Create compressed column matrix A_sup. */ 318 #if defined(PETSC_USE_COMPLEX) 319 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)); 320 #else 321 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)); 322 #endif 323 } else { /* distributed mat input */ 324 Mat_MPIAIJ *mat = (Mat_MPIAIJ*)A->data; 325 aa=(Mat_SeqAIJ*)(mat->A)->data; 326 bb=(Mat_SeqAIJ*)(mat->B)->data; 327 ai=aa->i; aj=aa->j; 328 bi=bb->i; bj=bb->j; 329 #if defined(PETSC_USE_COMPLEX) 330 av=(doublecomplex*)aa->a; 331 bv=(doublecomplex*)bb->a; 332 #else 333 av=aa->a; 334 bv=bb->a; 335 #endif 336 rstart = A->rmap->rstart; 337 nz = aa->nz + bb->nz; 338 garray = mat->garray; 339 340 if (lu->options.Fact == DOFACT) { /* first numeric factorization */ 341 #if defined(PETSC_USE_COMPLEX) 342 PetscStackCall("SuperLU_DIST:zallocateA_dist",zallocateA_dist(m, nz, &lu->val, &lu->col, &lu->row)); 343 #else 344 PetscStackCall("SuperLU_DIST:dallocateA_dist",dallocateA_dist(m, nz, &lu->val, &lu->col, &lu->row)); 345 #endif 346 } else { /* successive numeric factorization, sparsity pattern and perm_c are reused. */ 347 /* Destroy_CompRowLoc_Matrix_dist(&lu->A_sup); */ /* this leads to crash! However, see SuperLU_DIST_2.5/EXAMPLE/pzdrive2.c */ 348 if (lu->FactPattern == SamePattern_SameRowPerm) { 349 lu->options.Fact = SamePattern_SameRowPerm; /* matrix has similar numerical values */ 350 } else { 351 PetscStackCall("SuperLU_DIST:Destroy_LU",Destroy_LU(N, &lu->grid, &lu->LUstruct)); /* Deallocate storage associated with the L and U matrices. */ 352 lu->options.Fact = SamePattern; 353 } 354 } 355 nz = 0; 356 for (i=0; i<m; i++) { 357 lu->row[i] = nz; 358 countA = ai[i+1] - ai[i]; 359 countB = bi[i+1] - bi[i]; 360 ajj = aj + ai[i]; /* ptr to the beginning of this row */ 361 bjj = bj + bi[i]; 362 363 /* B part, smaller col index */ 364 colA_start = rstart + ajj[0]; /* the smallest global col index of A */ 365 jB = 0; 366 for (j=0; j<countB; j++) { 367 jcol = garray[bjj[j]]; 368 if (jcol > colA_start) { 369 jB = j; 370 break; 371 } 372 lu->col[nz] = jcol; 373 lu->val[nz++] = *bv++; 374 if (j==countB-1) jB = countB; 375 } 376 377 /* A part */ 378 for (j=0; j<countA; j++) { 379 lu->col[nz] = rstart + ajj[j]; 380 lu->val[nz++] = *av++; 381 } 382 383 /* B part, larger col index */ 384 for (j=jB; j<countB; j++) { 385 lu->col[nz] = garray[bjj[j]]; 386 lu->val[nz++] = *bv++; 387 } 388 } 389 lu->row[m] = nz; 390 #if defined(PETSC_USE_COMPLEX) 391 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)); 392 #else 393 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)); 394 #endif 395 } 396 if (lu->options.PrintStat) { 397 ierr = PetscTime(&time);CHKERRQ(ierr); 398 time0 = time - time0; 399 } 400 401 /* Factor the matrix. */ 402 PetscStackCall("SuperLU_DIST:PStatInit",PStatInit(&stat)); /* Initialize the statistics variables. */ 403 if (lu->MatInputMode == GLOBAL) { /* global mat input */ 404 #if defined(PETSC_USE_COMPLEX) 405 PetscStackCall("SuperLU_DIST:pzgssvx_ABglobal",pzgssvx_ABglobal(&lu->options, &lu->A_sup, &lu->ScalePermstruct, 0, M, 0,&lu->grid, &lu->LUstruct, berr, &stat, &sinfo)); 406 #else 407 PetscStackCall("SuperLU_DIST:pdgssvx_ABglobal",pdgssvx_ABglobal(&lu->options, &lu->A_sup, &lu->ScalePermstruct, 0, M, 0,&lu->grid, &lu->LUstruct, berr, &stat, &sinfo)); 408 #endif 409 } else { /* distributed mat input */ 410 #if defined(PETSC_USE_COMPLEX) 411 PetscStackCall("SuperLU_DIST:pzgssvx",pzgssvx(&lu->options, &lu->A_sup, &lu->ScalePermstruct, 0, m, 0, &lu->grid,&lu->LUstruct, &lu->SOLVEstruct, berr, &stat, &sinfo)); 412 if (sinfo) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"pzgssvx fails, info: %d\n",sinfo); 413 #else 414 PetscStackCall("SuperLU_DIST:pdgssvx",pdgssvx(&lu->options, &lu->A_sup, &lu->ScalePermstruct, 0, m, 0, &lu->grid,&lu->LUstruct, &lu->SOLVEstruct, berr, &stat, &sinfo)); 415 if (sinfo) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"pdgssvx fails, info: %d\n",sinfo); 416 #endif 417 } 418 419 if (lu->MatInputMode == GLOBAL && size > 1) { 420 ierr = MatDestroy(&A_seq);CHKERRQ(ierr); 421 } 422 423 if (lu->options.PrintStat) { 424 ierr = MPI_Reduce(&time0,&time_max,1,MPI_DOUBLE,MPI_MAX,0,PetscObjectComm((PetscObject)A)); 425 ierr = MPI_Reduce(&time0,&time_min,1,MPI_DOUBLE,MPI_MIN,0,PetscObjectComm((PetscObject)A)); 426 ierr = MPI_Reduce(&time0,&time,1,MPI_DOUBLE,MPI_SUM,0,PetscObjectComm((PetscObject)A)); 427 time = time/size; /* average time */ 428 ierr = PetscPrintf(PetscObjectComm((PetscObject)A), " Mat conversion(PETSc->SuperLU_DIST) time (max/min/avg): \n %g / %g / %g\n",time_max,time_min,time);CHKERRQ(ierr); 429 PStatPrint(&lu->options, &stat, &lu->grid); /* Print the statistics. */ 430 } 431 PStatFree(&stat); 432 if (size > 1) { 433 F_diag = ((Mat_MPIAIJ*)(F)->data)->A; 434 F_diag->assembled = PETSC_TRUE; 435 } 436 (F)->assembled = PETSC_TRUE; 437 (F)->preallocated = PETSC_TRUE; 438 lu->options.Fact = FACTORED; /* The factored form of A is supplied. Local option used by this func. only */ 439 PetscFunctionReturn(0); 440 } 441 442 /* Note the Petsc r and c permutations are ignored */ 443 #undef __FUNCT__ 444 #define __FUNCT__ "MatLUFactorSymbolic_SuperLU_DIST" 445 PetscErrorCode MatLUFactorSymbolic_SuperLU_DIST(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info) 446 { 447 Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)F->spptr; 448 PetscInt M = A->rmap->N,N=A->cmap->N; 449 450 PetscFunctionBegin; 451 /* Initialize the SuperLU process grid. */ 452 PetscStackCall("SuperLU_DIST:superlu_gridinit",superlu_gridinit(lu->comm_superlu, lu->nprow, lu->npcol, &lu->grid)); 453 454 /* Initialize ScalePermstruct and LUstruct. */ 455 PetscStackCall("SuperLU_DIST:ScalePermstructInit",ScalePermstructInit(M, N, &lu->ScalePermstruct)); 456 PetscStackCall("SuperLU_DIST:LUstructInit",LUstructInit(M, N, &lu->LUstruct)); 457 F->ops->lufactornumeric = MatLUFactorNumeric_SuperLU_DIST; 458 F->ops->solve = MatSolve_SuperLU_DIST; 459 F->ops->matsolve = MatMatSolve_SuperLU_DIST; 460 lu->CleanUpSuperLU_Dist = PETSC_TRUE; 461 PetscFunctionReturn(0); 462 } 463 464 #undef __FUNCT__ 465 #define __FUNCT__ "MatFactorGetSolverPackage_aij_superlu_dist" 466 PetscErrorCode MatFactorGetSolverPackage_aij_superlu_dist(Mat A,const MatSolverPackage *type) 467 { 468 PetscFunctionBegin; 469 *type = MATSOLVERSUPERLU_DIST; 470 PetscFunctionReturn(0); 471 } 472 473 #undef __FUNCT__ 474 #define __FUNCT__ "MatGetFactor_aij_superlu_dist" 475 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_superlu_dist(Mat A,MatFactorType ftype,Mat *F) 476 { 477 Mat B; 478 Mat_SuperLU_DIST *lu; 479 PetscErrorCode ierr; 480 PetscInt M=A->rmap->N,N=A->cmap->N,indx; 481 PetscMPIInt size; 482 superlu_options_t options; 483 PetscBool flg; 484 const char *colperm[] = {"NATURAL","MMD_AT_PLUS_A","MMD_ATA","METIS_AT_PLUS_A","PARMETIS"}; 485 const char *rowperm[] = {"LargeDiag","NATURAL"}; 486 const char *factPattern[] = {"SamePattern","SamePattern_SameRowPerm"}; 487 488 PetscFunctionBegin; 489 /* Create the factorization matrix */ 490 ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr); 491 ierr = MatSetSizes(B,A->rmap->n,A->cmap->n,M,N);CHKERRQ(ierr); 492 ierr = MatSetType(B,((PetscObject)A)->type_name);CHKERRQ(ierr); 493 ierr = MatSeqAIJSetPreallocation(B,0,NULL); 494 ierr = MatMPIAIJSetPreallocation(B,0,NULL,0,NULL);CHKERRQ(ierr); 495 496 B->ops->lufactorsymbolic = MatLUFactorSymbolic_SuperLU_DIST; 497 B->ops->view = MatView_SuperLU_DIST; 498 B->ops->destroy = MatDestroy_SuperLU_DIST; 499 500 ierr = PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverPackage_C",MatFactorGetSolverPackage_aij_superlu_dist);CHKERRQ(ierr); 501 502 B->factortype = MAT_FACTOR_LU; 503 504 ierr = PetscNewLog(B,Mat_SuperLU_DIST,&lu);CHKERRQ(ierr); 505 B->spptr = lu; 506 507 /* Set the default input options: 508 options.Fact = DOFACT; 509 options.Equil = YES; 510 options.ParSymbFact = NO; 511 options.ColPerm = METIS_AT_PLUS_A; 512 options.RowPerm = LargeDiag; 513 options.ReplaceTinyPivot = YES; 514 options.IterRefine = DOUBLE; 515 options.Trans = NOTRANS; 516 options.SolveInitialized = NO; -hold the communication pattern used MatSolve() and MatMatSolve() 517 options.RefineInitialized = NO; 518 options.PrintStat = YES; 519 */ 520 set_default_options_dist(&options); 521 522 ierr = MPI_Comm_dup(PetscObjectComm((PetscObject)A),&(lu->comm_superlu));CHKERRQ(ierr); 523 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);CHKERRQ(ierr); 524 /* Default num of process columns and rows */ 525 ierr = PetscMPIIntCast((PetscInt)(0.5 + PetscSqrtReal((PetscReal)size)),&lu->npcol);CHKERRQ(ierr); 526 if (!lu->npcol) lu->npcol = 1; 527 while (lu->npcol > 0) { 528 ierr = PetscMPIIntCast(size/lu->npcol,&lu->nprow);CHKERRQ(ierr); 529 if (size == lu->nprow * lu->npcol) break; 530 lu->npcol--; 531 } 532 533 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)A),((PetscObject)A)->prefix,"SuperLU_Dist Options","Mat");CHKERRQ(ierr); 534 ierr = PetscOptionsInt("-mat_superlu_dist_r","Number rows in processor partition","None",lu->nprow,&lu->nprow,NULL);CHKERRQ(ierr); 535 ierr = PetscOptionsInt("-mat_superlu_dist_c","Number columns in processor partition","None",lu->npcol,&lu->npcol,NULL);CHKERRQ(ierr); 536 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); 537 538 lu->MatInputMode = DISTRIBUTED; 539 540 ierr = PetscOptionsEnum("-mat_superlu_dist_matinput","Matrix input mode (global or distributed)","None",SuperLU_MatInputModes,(PetscEnum)lu->MatInputMode,(PetscEnum*)&lu->MatInputMode,NULL);CHKERRQ(ierr); 541 if (lu->MatInputMode == DISTRIBUTED && size == 1) lu->MatInputMode = GLOBAL; 542 543 ierr = PetscOptionsBool("-mat_superlu_dist_equil","Equilibrate matrix","None",PETSC_TRUE,&flg,0);CHKERRQ(ierr); 544 if (!flg) options.Equil = NO; 545 546 ierr = PetscOptionsEList("-mat_superlu_dist_rowperm","Row permutation","None",rowperm,2,rowperm[0],&indx,&flg);CHKERRQ(ierr); 547 if (flg) { 548 switch (indx) { 549 case 0: 550 options.RowPerm = LargeDiag; 551 break; 552 case 1: 553 options.RowPerm = NOROWPERM; 554 break; 555 } 556 } 557 558 ierr = PetscOptionsEList("-mat_superlu_dist_colperm","Column permutation","None",colperm,5,colperm[3],&indx,&flg);CHKERRQ(ierr); 559 if (flg) { 560 switch (indx) { 561 case 0: 562 options.ColPerm = NATURAL; 563 break; 564 case 1: 565 options.ColPerm = MMD_AT_PLUS_A; 566 break; 567 case 2: 568 options.ColPerm = MMD_ATA; 569 break; 570 case 3: 571 options.ColPerm = METIS_AT_PLUS_A; 572 break; 573 case 4: 574 options.ColPerm = PARMETIS; /* only works for np>1 */ 575 break; 576 default: 577 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown column permutation"); 578 } 579 } 580 581 ierr = PetscOptionsBool("-mat_superlu_dist_replacetinypivot","Replace tiny pivots","None",PETSC_TRUE,&flg,0);CHKERRQ(ierr); 582 if (!flg) options.ReplaceTinyPivot = NO; 583 584 options.ParSymbFact = NO; 585 586 ierr = PetscOptionsBool("-mat_superlu_dist_parsymbfact","Parallel symbolic factorization","None",PETSC_FALSE,&flg,0);CHKERRQ(ierr); 587 if (flg) { 588 #if defined(PETSC_HAVE_PARMETIS) 589 options.ParSymbFact = YES; 590 options.ColPerm = PARMETIS; /* in v2.2, PARMETIS is forced for ParSymbFact regardless of user ordering setting */ 591 #else 592 printf("parsymbfact needs PARMETIS"); 593 #endif 594 } 595 596 lu->FactPattern = SamePattern_SameRowPerm; 597 598 ierr = PetscOptionsEList("-mat_superlu_dist_fact","Sparsity pattern for repeated matrix factorization","None",factPattern,2,factPattern[1],&indx,&flg);CHKERRQ(ierr); 599 if (flg) { 600 switch (indx) { 601 case 0: 602 lu->FactPattern = SamePattern; 603 break; 604 case 1: 605 lu->FactPattern = SamePattern_SameRowPerm; 606 break; 607 } 608 } 609 610 options.IterRefine = NOREFINE; 611 ierr = PetscOptionsBool("-mat_superlu_dist_iterrefine","Use iterative refinement","None",PETSC_FALSE,&flg,0);CHKERRQ(ierr); 612 if (flg) options.IterRefine = SLU_DOUBLE; 613 614 if (PetscLogPrintInfo) options.PrintStat = YES; 615 else options.PrintStat = NO; 616 ierr = PetscOptionsBool("-mat_superlu_dist_statprint","Print factorization information","None", 617 (PetscBool)options.PrintStat,(PetscBool*)&options.PrintStat,0);CHKERRQ(ierr); 618 PetscOptionsEnd(); 619 620 lu->options = options; 621 lu->options.Fact = DOFACT; 622 lu->matsolve_iscalled = PETSC_FALSE; 623 lu->matmatsolve_iscalled = PETSC_FALSE; 624 625 *F = B; 626 PetscFunctionReturn(0); 627 } 628 629 #undef __FUNCT__ 630 #define __FUNCT__ "MatGetFactor_seqaij_superlu_dist" 631 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_superlu_dist(Mat A,MatFactorType ftype,Mat *F) 632 { 633 PetscErrorCode ierr; 634 635 PetscFunctionBegin; 636 ierr = MatGetFactor_aij_superlu_dist(A,ftype,F);CHKERRQ(ierr); 637 PetscFunctionReturn(0); 638 } 639 640 #undef __FUNCT__ 641 #define __FUNCT__ "MatGetFactor_mpiaij_superlu_dist" 642 PETSC_EXTERN PetscErrorCode MatGetFactor_mpiaij_superlu_dist(Mat A,MatFactorType ftype,Mat *F) 643 { 644 PetscErrorCode ierr; 645 646 PetscFunctionBegin; 647 ierr = MatGetFactor_aij_superlu_dist(A,ftype,F);CHKERRQ(ierr); 648 PetscFunctionReturn(0); 649 } 650 651 #undef __FUNCT__ 652 #define __FUNCT__ "MatFactorInfo_SuperLU_DIST" 653 PetscErrorCode MatFactorInfo_SuperLU_DIST(Mat A,PetscViewer viewer) 654 { 655 Mat_SuperLU_DIST *lu=(Mat_SuperLU_DIST*)A->spptr; 656 superlu_options_t options; 657 PetscErrorCode ierr; 658 659 PetscFunctionBegin; 660 /* check if matrix is superlu_dist type */ 661 if (A->ops->solve != MatSolve_SuperLU_DIST) PetscFunctionReturn(0); 662 663 options = lu->options; 664 ierr = PetscViewerASCIIPrintf(viewer,"SuperLU_DIST run parameters:\n");CHKERRQ(ierr); 665 ierr = PetscViewerASCIIPrintf(viewer," Process grid nprow %D x npcol %D \n",lu->nprow,lu->npcol);CHKERRQ(ierr); 666 ierr = PetscViewerASCIIPrintf(viewer," Equilibrate matrix %s \n",PetscBools[options.Equil != NO]);CHKERRQ(ierr); 667 ierr = PetscViewerASCIIPrintf(viewer," Matrix input mode %d \n",lu->MatInputMode);CHKERRQ(ierr); 668 ierr = PetscViewerASCIIPrintf(viewer," Replace tiny pivots %s \n",PetscBools[options.ReplaceTinyPivot != NO]);CHKERRQ(ierr); 669 ierr = PetscViewerASCIIPrintf(viewer," Use iterative refinement %s \n",PetscBools[options.IterRefine == SLU_DOUBLE]);CHKERRQ(ierr); 670 ierr = PetscViewerASCIIPrintf(viewer," Processors in row %d col partition %d \n",lu->nprow,lu->npcol);CHKERRQ(ierr); 671 ierr = PetscViewerASCIIPrintf(viewer," Row permutation %s \n",(options.RowPerm == NOROWPERM) ? "NATURAL" : "LargeDiag");CHKERRQ(ierr); 672 673 switch (options.ColPerm) { 674 case NATURAL: 675 ierr = PetscViewerASCIIPrintf(viewer," Column permutation NATURAL\n");CHKERRQ(ierr); 676 break; 677 case MMD_AT_PLUS_A: 678 ierr = PetscViewerASCIIPrintf(viewer," Column permutation MMD_AT_PLUS_A\n");CHKERRQ(ierr); 679 break; 680 case MMD_ATA: 681 ierr = PetscViewerASCIIPrintf(viewer," Column permutation MMD_ATA\n");CHKERRQ(ierr); 682 break; 683 case METIS_AT_PLUS_A: 684 ierr = PetscViewerASCIIPrintf(viewer," Column permutation METIS_AT_PLUS_A\n");CHKERRQ(ierr); 685 break; 686 case PARMETIS: 687 ierr = PetscViewerASCIIPrintf(viewer," Column permutation PARMETIS\n");CHKERRQ(ierr); 688 break; 689 default: 690 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown column permutation"); 691 } 692 693 ierr = PetscViewerASCIIPrintf(viewer," Parallel symbolic factorization %s \n",PetscBools[options.ParSymbFact != NO]);CHKERRQ(ierr); 694 695 if (lu->FactPattern == SamePattern) { 696 ierr = PetscViewerASCIIPrintf(viewer," Repeated factorization SamePattern\n");CHKERRQ(ierr); 697 } else { 698 ierr = PetscViewerASCIIPrintf(viewer," Repeated factorization SamePattern_SameRowPerm\n");CHKERRQ(ierr); 699 } 700 PetscFunctionReturn(0); 701 } 702 703 #undef __FUNCT__ 704 #define __FUNCT__ "MatView_SuperLU_DIST" 705 PetscErrorCode MatView_SuperLU_DIST(Mat A,PetscViewer viewer) 706 { 707 PetscErrorCode ierr; 708 PetscBool iascii; 709 PetscViewerFormat format; 710 711 PetscFunctionBegin; 712 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 713 if (iascii) { 714 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 715 if (format == PETSC_VIEWER_ASCII_INFO) { 716 ierr = MatFactorInfo_SuperLU_DIST(A,viewer);CHKERRQ(ierr); 717 } 718 } 719 PetscFunctionReturn(0); 720 } 721 722 723 /*MC 724 MATSOLVERSUPERLU_DIST - Parallel direct solver package for LU factorization 725 726 Works with AIJ matrices 727 728 Options Database Keys: 729 + -mat_superlu_dist_r <n> - number of rows in processor partition 730 . -mat_superlu_dist_c <n> - number of columns in processor partition 731 . -mat_superlu_dist_matinput <0,1> - matrix input mode; 0=global, 1=distributed 732 . -mat_superlu_dist_equil - equilibrate the matrix 733 . -mat_superlu_dist_rowperm <LargeDiag,NATURAL> - row permutation 734 . -mat_superlu_dist_colperm <MMD_AT_PLUS_A,MMD_ATA,NATURAL> - column permutation 735 . -mat_superlu_dist_replacetinypivot - replace tiny pivots 736 . -mat_superlu_dist_fact <SamePattern> - (choose one of) SamePattern SamePattern_SameRowPerm 737 . -mat_superlu_dist_iterrefine - use iterative refinement 738 - -mat_superlu_dist_statprint - print factorization information 739 740 Level: beginner 741 742 .seealso: PCLU 743 744 .seealso: PCFactorSetMatSolverPackage(), MatSolverPackage 745 746 M*/ 747 748