1 /* 2 Provides an interface to the PaStiX sparse solver 3 */ 4 #include <../src/mat/impls/aij/seq/aij.h> 5 #include <../src/mat/impls/aij/mpi/mpiaij.h> 6 #include <../src/mat/impls/sbaij/seq/sbaij.h> 7 #include <../src/mat/impls/sbaij/mpi/mpisbaij.h> 8 9 #if defined(PETSC_USE_COMPLEX) 10 #define _H_COMPLEX 11 #endif 12 13 EXTERN_C_BEGIN 14 #include <pastix.h> 15 EXTERN_C_END 16 17 #if defined(PETSC_USE_COMPLEX) 18 #if defined(PETSC_USE_REAL_SINGLE) 19 #define PASTIX_CALL c_pastix 20 #define PASTIX_CHECKMATRIX c_pastix_checkMatrix 21 #else 22 #define PASTIX_CALL z_pastix 23 #define PASTIX_CHECKMATRIX z_pastix_checkMatrix 24 #endif 25 26 #else /* PETSC_USE_COMPLEX */ 27 28 #if defined(PETSC_USE_REAL_SINGLE) 29 #define PASTIX_CALL s_pastix 30 #define PASTIX_CHECKMATRIX s_pastix_checkMatrix 31 #else 32 #define PASTIX_CALL d_pastix 33 #define PASTIX_CHECKMATRIX d_pastix_checkMatrix 34 #endif 35 36 #endif /* PETSC_USE_COMPLEX */ 37 38 typedef PetscScalar PastixScalar; 39 40 typedef struct Mat_Pastix_ { 41 pastix_data_t *pastix_data; /* Pastix data storage structure */ 42 MatStructure matstruc; 43 PetscInt n; /* Number of columns in the matrix */ 44 PetscInt *colptr; /* Index of first element of each column in row and val */ 45 PetscInt *row; /* Row of each element of the matrix */ 46 PetscScalar *val; /* Value of each element of the matrix */ 47 PetscInt *perm; /* Permutation tabular */ 48 PetscInt *invp; /* Reverse permutation tabular */ 49 PetscScalar *rhs; /* Rhight-hand-side member */ 50 PetscInt rhsnbr; /* Rhight-hand-side number (must be 1) */ 51 PetscInt iparm[64]; /* Integer parameters */ 52 double dparm[64]; /* Floating point parameters */ 53 MPI_Comm pastix_comm; /* PaStiX MPI communicator */ 54 PetscMPIInt commRank; /* MPI rank */ 55 PetscMPIInt commSize; /* MPI communicator size */ 56 PetscBool CleanUpPastix; /* Boolean indicating if we call PaStiX clean step */ 57 VecScatter scat_rhs; 58 VecScatter scat_sol; 59 Vec b_seq; 60 PetscBool isAIJ; 61 PetscErrorCode (*Destroy)(Mat); 62 } Mat_Pastix; 63 64 extern PetscErrorCode MatDuplicate_Pastix(Mat,MatDuplicateOption,Mat*); 65 66 #undef __FUNCT__ 67 #define __FUNCT__ "MatConvertToCSC" 68 /* 69 convert Petsc seqaij matrix to CSC: colptr[n], row[nz], val[nz] 70 71 input: 72 A - matrix in seqaij or mpisbaij (bs=1) format 73 valOnly - FALSE: spaces are allocated and values are set for the CSC 74 TRUE: Only fill values 75 output: 76 n - Size of the matrix 77 colptr - Index of first element of each column in row and val 78 row - Row of each element of the matrix 79 values - Value of each element of the matrix 80 */ 81 PetscErrorCode MatConvertToCSC(Mat A,PetscBool valOnly,PetscInt *n,PetscInt **colptr,PetscInt **row,PetscScalar **values) 82 { 83 Mat_SeqAIJ *aa = (Mat_SeqAIJ*)A->data; 84 PetscInt *rowptr = aa->i; 85 PetscInt *col = aa->j; 86 PetscScalar *rvalues = aa->a; 87 PetscInt m = A->rmap->N; 88 PetscInt nnz; 89 PetscInt i,j, k; 90 PetscInt base = 1; 91 PetscInt idx; 92 PetscErrorCode ierr; 93 PetscInt colidx; 94 PetscInt *colcount; 95 PetscBool isSBAIJ; 96 PetscBool isSeqSBAIJ; 97 PetscBool isMpiSBAIJ; 98 PetscBool isSym; 99 PetscBool flg; 100 PetscInt icntl; 101 PetscInt verb; 102 PetscInt check; 103 104 PetscFunctionBegin; 105 ierr = MatIsSymmetric(A,0.0,&isSym);CHKERRQ(ierr); 106 ierr = PetscObjectTypeCompare((PetscObject)A,MATSBAIJ,&isSBAIJ);CHKERRQ(ierr); 107 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQSBAIJ,&isSeqSBAIJ);CHKERRQ(ierr); 108 ierr = PetscObjectTypeCompare((PetscObject)A,MATMPISBAIJ,&isMpiSBAIJ);CHKERRQ(ierr); 109 110 *n = A->cmap->N; 111 112 /* PaStiX only needs triangular matrix if matrix is symmetric 113 */ 114 if (isSym && !(isSBAIJ || isSeqSBAIJ || isMpiSBAIJ)) nnz = (aa->nz - *n)/2 + *n; 115 else nnz = aa->nz; 116 117 if (!valOnly) { 118 ierr = PetscMalloc(((*n)+1) *sizeof(PetscInt) ,colptr);CHKERRQ(ierr); 119 ierr = PetscMalloc(nnz *sizeof(PetscInt) ,row);CHKERRQ(ierr); 120 ierr = PetscMalloc1(nnz ,values);CHKERRQ(ierr); 121 122 if (isSBAIJ || isSeqSBAIJ || isMpiSBAIJ) { 123 ierr = PetscMemcpy (*colptr, rowptr, ((*n)+1)*sizeof(PetscInt));CHKERRQ(ierr); 124 for (i = 0; i < *n+1; i++) (*colptr)[i] += base; 125 ierr = PetscMemcpy (*row, col, (nnz)*sizeof(PetscInt));CHKERRQ(ierr); 126 for (i = 0; i < nnz; i++) (*row)[i] += base; 127 ierr = PetscMemcpy (*values, rvalues, (nnz)*sizeof(PetscScalar));CHKERRQ(ierr); 128 } else { 129 ierr = PetscMalloc1((*n),&colcount);CHKERRQ(ierr); 130 131 for (i = 0; i < m; i++) colcount[i] = 0; 132 /* Fill-in colptr */ 133 for (i = 0; i < m; i++) { 134 for (j = rowptr[i]; j < rowptr[i+1]; j++) { 135 if (!isSym || col[j] <= i) colcount[col[j]]++; 136 } 137 } 138 139 (*colptr)[0] = base; 140 for (j = 0; j < *n; j++) { 141 (*colptr)[j+1] = (*colptr)[j] + colcount[j]; 142 /* in next loop we fill starting from (*colptr)[colidx] - base */ 143 colcount[j] = -base; 144 } 145 146 /* Fill-in rows and values */ 147 for (i = 0; i < m; i++) { 148 for (j = rowptr[i]; j < rowptr[i+1]; j++) { 149 if (!isSym || col[j] <= i) { 150 colidx = col[j]; 151 idx = (*colptr)[colidx] + colcount[colidx]; 152 (*row)[idx] = i + base; 153 (*values)[idx] = rvalues[j]; 154 colcount[colidx]++; 155 } 156 } 157 } 158 ierr = PetscFree(colcount);CHKERRQ(ierr); 159 } 160 } else { 161 /* Fill-in only values */ 162 for (i = 0; i < m; i++) { 163 for (j = rowptr[i]; j < rowptr[i+1]; j++) { 164 colidx = col[j]; 165 if ((isSBAIJ || isSeqSBAIJ || isMpiSBAIJ) ||!isSym || col[j] <= i) { 166 /* look for the value to fill */ 167 for (k = (*colptr)[colidx] - base; k < (*colptr)[colidx + 1] - base; k++) { 168 if (((*row)[k]-base) == i) { 169 (*values)[k] = rvalues[j]; 170 break; 171 } 172 } 173 /* data structure of sparse matrix has changed */ 174 if (k == (*colptr)[colidx + 1] - base) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_PLIB,"overflow on k %D",k); 175 } 176 } 177 } 178 } 179 180 icntl =-1; 181 check = 0; 182 ierr = PetscOptionsGetInt(((PetscObject) A)->prefix, "-mat_pastix_check", &icntl, &flg);CHKERRQ(ierr); 183 if ((flg && icntl >= 0) || PetscLogPrintInfo) check = icntl; 184 185 if (check == 1) { 186 PetscScalar *tmpvalues; 187 PetscInt *tmprows,*tmpcolptr; 188 tmpvalues = (PetscScalar*)malloc(nnz*sizeof(PetscScalar)); if (!tmpvalues) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_MEM,"Unable to allocate memory"); 189 tmprows = (PetscInt*) malloc(nnz*sizeof(PetscInt)); if (!tmprows) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_MEM,"Unable to allocate memory"); 190 tmpcolptr = (PetscInt*) malloc((*n+1)*sizeof(PetscInt)); if (!tmpcolptr) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_MEM,"Unable to allocate memory"); 191 192 ierr = PetscMemcpy(tmpcolptr,*colptr,(*n+1)*sizeof(PetscInt));CHKERRQ(ierr); 193 ierr = PetscMemcpy(tmprows,*row,nnz*sizeof(PetscInt));CHKERRQ(ierr); 194 ierr = PetscMemcpy(tmpvalues,*values,nnz*sizeof(PetscScalar));CHKERRQ(ierr); 195 ierr = PetscFree(*row);CHKERRQ(ierr); 196 ierr = PetscFree(*values);CHKERRQ(ierr); 197 198 icntl=-1; 199 verb = API_VERBOSE_NOT; 200 /* "iparm[IPARM_VERBOSE] : level of printing (0 to 2)" */ 201 ierr = PetscOptionsGetInt(((PetscObject) A)->prefix, "-mat_pastix_verbose", &icntl, &flg);CHKERRQ(ierr); 202 if ((flg && icntl >= 0) || PetscLogPrintInfo) verb = icntl; 203 PASTIX_CHECKMATRIX(MPI_COMM_WORLD,verb,((isSym != 0) ? API_SYM_YES : API_SYM_NO),API_YES,*n,&tmpcolptr,&tmprows,(PastixScalar**)&tmpvalues,NULL,1); 204 205 ierr = PetscMemcpy(*colptr,tmpcolptr,(*n+1)*sizeof(PetscInt));CHKERRQ(ierr); 206 ierr = PetscMalloc1(((*colptr)[*n]-1),row);CHKERRQ(ierr); 207 ierr = PetscMemcpy(*row,tmprows,((*colptr)[*n]-1)*sizeof(PetscInt));CHKERRQ(ierr); 208 ierr = PetscMalloc1(((*colptr)[*n]-1),values);CHKERRQ(ierr); 209 ierr = PetscMemcpy(*values,tmpvalues,((*colptr)[*n]-1)*sizeof(PetscScalar));CHKERRQ(ierr); 210 free(tmpvalues); 211 free(tmprows); 212 free(tmpcolptr); 213 214 } 215 PetscFunctionReturn(0); 216 } 217 218 219 220 #undef __FUNCT__ 221 #define __FUNCT__ "MatDestroy_Pastix" 222 /* 223 Call clean step of PaStiX if lu->CleanUpPastix == true. 224 Free the CSC matrix. 225 */ 226 PetscErrorCode MatDestroy_Pastix(Mat A) 227 { 228 Mat_Pastix *lu=(Mat_Pastix*)A->spptr; 229 PetscErrorCode ierr; 230 PetscMPIInt size=lu->commSize; 231 232 PetscFunctionBegin; 233 if (lu && lu->CleanUpPastix) { 234 /* Terminate instance, deallocate memories */ 235 if (size > 1) { 236 ierr = VecScatterDestroy(&lu->scat_rhs);CHKERRQ(ierr); 237 ierr = VecDestroy(&lu->b_seq);CHKERRQ(ierr); 238 ierr = VecScatterDestroy(&lu->scat_sol);CHKERRQ(ierr); 239 } 240 241 lu->iparm[IPARM_START_TASK]=API_TASK_CLEAN; 242 lu->iparm[IPARM_END_TASK] =API_TASK_CLEAN; 243 244 PASTIX_CALL(&(lu->pastix_data), 245 lu->pastix_comm, 246 lu->n, 247 lu->colptr, 248 lu->row, 249 (PastixScalar*)lu->val, 250 lu->perm, 251 lu->invp, 252 (PastixScalar*)lu->rhs, 253 lu->rhsnbr, 254 lu->iparm, 255 lu->dparm); 256 257 ierr = PetscFree(lu->colptr);CHKERRQ(ierr); 258 ierr = PetscFree(lu->row);CHKERRQ(ierr); 259 ierr = PetscFree(lu->val);CHKERRQ(ierr); 260 ierr = PetscFree(lu->perm);CHKERRQ(ierr); 261 ierr = PetscFree(lu->invp);CHKERRQ(ierr); 262 ierr = MPI_Comm_free(&(lu->pastix_comm));CHKERRQ(ierr); 263 } 264 if (lu && lu->Destroy) { 265 ierr = (lu->Destroy)(A);CHKERRQ(ierr); 266 } 267 ierr = PetscFree(A->spptr);CHKERRQ(ierr); 268 PetscFunctionReturn(0); 269 } 270 271 #undef __FUNCT__ 272 #define __FUNCT__ "MatSolve_PaStiX" 273 /* 274 Gather right-hand-side. 275 Call for Solve step. 276 Scatter solution. 277 */ 278 PetscErrorCode MatSolve_PaStiX(Mat A,Vec b,Vec x) 279 { 280 Mat_Pastix *lu=(Mat_Pastix*)A->spptr; 281 PetscScalar *array; 282 Vec x_seq; 283 PetscErrorCode ierr; 284 285 PetscFunctionBegin; 286 lu->rhsnbr = 1; 287 x_seq = lu->b_seq; 288 if (lu->commSize > 1) { 289 /* PaStiX only supports centralized rhs. Scatter b into a seqential rhs vector */ 290 ierr = VecScatterBegin(lu->scat_rhs,b,x_seq,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 291 ierr = VecScatterEnd(lu->scat_rhs,b,x_seq,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 292 ierr = VecGetArray(x_seq,&array);CHKERRQ(ierr); 293 } else { /* size == 1 */ 294 ierr = VecCopy(b,x);CHKERRQ(ierr); 295 ierr = VecGetArray(x,&array);CHKERRQ(ierr); 296 } 297 lu->rhs = array; 298 if (lu->commSize == 1) { 299 ierr = VecRestoreArray(x,&array);CHKERRQ(ierr); 300 } else { 301 ierr = VecRestoreArray(x_seq,&array);CHKERRQ(ierr); 302 } 303 304 /* solve phase */ 305 /*-------------*/ 306 lu->iparm[IPARM_START_TASK] = API_TASK_SOLVE; 307 lu->iparm[IPARM_END_TASK] = API_TASK_REFINE; 308 lu->iparm[IPARM_RHS_MAKING] = API_RHS_B; 309 310 PASTIX_CALL(&(lu->pastix_data), 311 lu->pastix_comm, 312 lu->n, 313 lu->colptr, 314 lu->row, 315 (PastixScalar*)lu->val, 316 lu->perm, 317 lu->invp, 318 (PastixScalar*)lu->rhs, 319 lu->rhsnbr, 320 lu->iparm, 321 lu->dparm); 322 323 if (lu->iparm[IPARM_ERROR_NUMBER] < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by PaStiX in solve phase: lu->iparm[IPARM_ERROR_NUMBER] = %d\n",lu->iparm[IPARM_ERROR_NUMBER]); 324 325 if (lu->commSize == 1) { 326 ierr = VecRestoreArray(x,&(lu->rhs));CHKERRQ(ierr); 327 } else { 328 ierr = VecRestoreArray(x_seq,&(lu->rhs));CHKERRQ(ierr); 329 } 330 331 if (lu->commSize > 1) { /* convert PaStiX centralized solution to petsc mpi x */ 332 ierr = VecScatterBegin(lu->scat_sol,x_seq,x,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 333 ierr = VecScatterEnd(lu->scat_sol,x_seq,x,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 334 } 335 PetscFunctionReturn(0); 336 } 337 338 /* 339 Numeric factorisation using PaStiX solver. 340 341 */ 342 #undef __FUNCT__ 343 #define __FUNCT__ "MatFactorNumeric_PaStiX" 344 PetscErrorCode MatFactorNumeric_PaStiX(Mat F,Mat A,const MatFactorInfo *info) 345 { 346 Mat_Pastix *lu =(Mat_Pastix*)(F)->spptr; 347 Mat *tseq; 348 PetscErrorCode ierr = 0; 349 PetscInt icntl; 350 PetscInt M=A->rmap->N; 351 PetscBool valOnly,flg, isSym; 352 Mat F_diag; 353 IS is_iden; 354 Vec b; 355 IS isrow; 356 PetscBool isSeqAIJ,isSeqSBAIJ,isMPIAIJ; 357 358 PetscFunctionBegin; 359 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQAIJ,&isSeqAIJ);CHKERRQ(ierr); 360 ierr = PetscObjectTypeCompare((PetscObject)A,MATMPIAIJ,&isMPIAIJ);CHKERRQ(ierr); 361 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQSBAIJ,&isSeqSBAIJ);CHKERRQ(ierr); 362 if (lu->matstruc == DIFFERENT_NONZERO_PATTERN) { 363 (F)->ops->solve = MatSolve_PaStiX; 364 365 /* Initialize a PASTIX instance */ 366 ierr = MPI_Comm_dup(PetscObjectComm((PetscObject)A),&(lu->pastix_comm));CHKERRQ(ierr); 367 ierr = MPI_Comm_rank(lu->pastix_comm, &lu->commRank);CHKERRQ(ierr); 368 ierr = MPI_Comm_size(lu->pastix_comm, &lu->commSize);CHKERRQ(ierr); 369 370 /* Set pastix options */ 371 lu->iparm[IPARM_MODIFY_PARAMETER] = API_NO; 372 lu->iparm[IPARM_START_TASK] = API_TASK_INIT; 373 lu->iparm[IPARM_END_TASK] = API_TASK_INIT; 374 375 lu->rhsnbr = 1; 376 377 /* Call to set default pastix options */ 378 PASTIX_CALL(&(lu->pastix_data), 379 lu->pastix_comm, 380 lu->n, 381 lu->colptr, 382 lu->row, 383 (PastixScalar*)lu->val, 384 lu->perm, 385 lu->invp, 386 (PastixScalar*)lu->rhs, 387 lu->rhsnbr, 388 lu->iparm, 389 lu->dparm); 390 391 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)A),((PetscObject)A)->prefix,"PaStiX Options","Mat");CHKERRQ(ierr); 392 393 icntl = -1; 394 395 lu->iparm[IPARM_VERBOSE] = API_VERBOSE_NOT; 396 397 ierr = PetscOptionsInt("-mat_pastix_verbose","iparm[IPARM_VERBOSE] : level of printing (0 to 2)","None",lu->iparm[IPARM_VERBOSE],&icntl,&flg);CHKERRQ(ierr); 398 if ((flg && icntl >= 0) || PetscLogPrintInfo) { 399 lu->iparm[IPARM_VERBOSE] = icntl; 400 } 401 icntl=-1; 402 ierr = PetscOptionsInt("-mat_pastix_threadnbr","iparm[IPARM_THREAD_NBR] : Number of thread by MPI node","None",lu->iparm[IPARM_THREAD_NBR],&icntl,&flg);CHKERRQ(ierr); 403 if ((flg && icntl > 0)) { 404 lu->iparm[IPARM_THREAD_NBR] = icntl; 405 } 406 PetscOptionsEnd(); 407 valOnly = PETSC_FALSE; 408 } else { 409 if (isSeqAIJ || isMPIAIJ) { 410 ierr = PetscFree(lu->colptr);CHKERRQ(ierr); 411 ierr = PetscFree(lu->row);CHKERRQ(ierr); 412 ierr = PetscFree(lu->val);CHKERRQ(ierr); 413 valOnly = PETSC_FALSE; 414 } else valOnly = PETSC_TRUE; 415 } 416 417 lu->iparm[IPARM_MATRIX_VERIFICATION] = API_YES; 418 419 /* convert mpi A to seq mat A */ 420 ierr = ISCreateStride(PETSC_COMM_SELF,M,0,1,&isrow);CHKERRQ(ierr); 421 ierr = MatGetSubMatrices(A,1,&isrow,&isrow,MAT_INITIAL_MATRIX,&tseq);CHKERRQ(ierr); 422 ierr = ISDestroy(&isrow);CHKERRQ(ierr); 423 424 ierr = MatConvertToCSC(*tseq,valOnly, &lu->n, &lu->colptr, &lu->row, &lu->val);CHKERRQ(ierr); 425 ierr = MatIsSymmetric(*tseq,0.0,&isSym);CHKERRQ(ierr); 426 ierr = MatDestroyMatrices(1,&tseq);CHKERRQ(ierr); 427 428 if (!lu->perm) { 429 ierr = PetscMalloc1((lu->n),&(lu->perm));CHKERRQ(ierr); 430 ierr = PetscMalloc1((lu->n),&(lu->invp));CHKERRQ(ierr); 431 } 432 433 if (isSym) { 434 /* On symmetric matrix, LLT */ 435 lu->iparm[IPARM_SYM] = API_SYM_YES; 436 lu->iparm[IPARM_FACTORIZATION] = API_FACT_LDLT; 437 } else { 438 /* On unsymmetric matrix, LU */ 439 lu->iparm[IPARM_SYM] = API_SYM_NO; 440 lu->iparm[IPARM_FACTORIZATION] = API_FACT_LU; 441 } 442 443 /*----------------*/ 444 if (lu->matstruc == DIFFERENT_NONZERO_PATTERN) { 445 if (!(isSeqAIJ || isSeqSBAIJ)) { 446 /* PaStiX only supports centralized rhs. Create scatter scat_rhs for repeated use in MatSolve() */ 447 ierr = VecCreateSeq(PETSC_COMM_SELF,A->cmap->N,&lu->b_seq);CHKERRQ(ierr); 448 ierr = ISCreateStride(PETSC_COMM_SELF,A->cmap->N,0,1,&is_iden);CHKERRQ(ierr); 449 ierr = MatCreateVecs(A,NULL,&b);CHKERRQ(ierr); 450 ierr = VecScatterCreate(b,is_iden,lu->b_seq,is_iden,&lu->scat_rhs);CHKERRQ(ierr); 451 ierr = VecScatterCreate(lu->b_seq,is_iden,b,is_iden,&lu->scat_sol);CHKERRQ(ierr); 452 ierr = ISDestroy(&is_iden);CHKERRQ(ierr); 453 ierr = VecDestroy(&b);CHKERRQ(ierr); 454 } 455 lu->iparm[IPARM_START_TASK] = API_TASK_ORDERING; 456 lu->iparm[IPARM_END_TASK] = API_TASK_NUMFACT; 457 458 PASTIX_CALL(&(lu->pastix_data), 459 lu->pastix_comm, 460 lu->n, 461 lu->colptr, 462 lu->row, 463 (PastixScalar*)lu->val, 464 lu->perm, 465 lu->invp, 466 (PastixScalar*)lu->rhs, 467 lu->rhsnbr, 468 lu->iparm, 469 lu->dparm); 470 if (lu->iparm[IPARM_ERROR_NUMBER] < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by PaStiX in analysis phase: iparm(IPARM_ERROR_NUMBER)=%d\n",lu->iparm[IPARM_ERROR_NUMBER]); 471 } else { 472 lu->iparm[IPARM_START_TASK] = API_TASK_NUMFACT; 473 lu->iparm[IPARM_END_TASK] = API_TASK_NUMFACT; 474 PASTIX_CALL(&(lu->pastix_data), 475 lu->pastix_comm, 476 lu->n, 477 lu->colptr, 478 lu->row, 479 (PastixScalar*)lu->val, 480 lu->perm, 481 lu->invp, 482 (PastixScalar*)lu->rhs, 483 lu->rhsnbr, 484 lu->iparm, 485 lu->dparm); 486 487 if (lu->iparm[IPARM_ERROR_NUMBER] < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by PaStiX in analysis phase: iparm(IPARM_ERROR_NUMBER)=%d\n",lu->iparm[IPARM_ERROR_NUMBER]); 488 } 489 490 if (lu->commSize > 1) { 491 if ((F)->factortype == MAT_FACTOR_LU) { 492 F_diag = ((Mat_MPIAIJ*)(F)->data)->A; 493 } else { 494 F_diag = ((Mat_MPISBAIJ*)(F)->data)->A; 495 } 496 F_diag->assembled = PETSC_TRUE; 497 } 498 (F)->assembled = PETSC_TRUE; 499 lu->matstruc = SAME_NONZERO_PATTERN; 500 lu->CleanUpPastix = PETSC_TRUE; 501 PetscFunctionReturn(0); 502 } 503 504 /* Note the Petsc r and c permutations are ignored */ 505 #undef __FUNCT__ 506 #define __FUNCT__ "MatLUFactorSymbolic_AIJPASTIX" 507 PetscErrorCode MatLUFactorSymbolic_AIJPASTIX(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info) 508 { 509 Mat_Pastix *lu = (Mat_Pastix*)F->spptr; 510 511 PetscFunctionBegin; 512 lu->iparm[IPARM_FACTORIZATION] = API_FACT_LU; 513 lu->iparm[IPARM_SYM] = API_SYM_YES; 514 lu->matstruc = DIFFERENT_NONZERO_PATTERN; 515 F->ops->lufactornumeric = MatFactorNumeric_PaStiX; 516 PetscFunctionReturn(0); 517 } 518 519 520 /* Note the Petsc r permutation is ignored */ 521 #undef __FUNCT__ 522 #define __FUNCT__ "MatCholeskyFactorSymbolic_SBAIJPASTIX" 523 PetscErrorCode MatCholeskyFactorSymbolic_SBAIJPASTIX(Mat F,Mat A,IS r,const MatFactorInfo *info) 524 { 525 Mat_Pastix *lu = (Mat_Pastix*)(F)->spptr; 526 527 PetscFunctionBegin; 528 lu->iparm[IPARM_FACTORIZATION] = API_FACT_LLT; 529 lu->iparm[IPARM_SYM] = API_SYM_NO; 530 lu->matstruc = DIFFERENT_NONZERO_PATTERN; 531 (F)->ops->choleskyfactornumeric = MatFactorNumeric_PaStiX; 532 PetscFunctionReturn(0); 533 } 534 535 #undef __FUNCT__ 536 #define __FUNCT__ "MatView_PaStiX" 537 PetscErrorCode MatView_PaStiX(Mat A,PetscViewer viewer) 538 { 539 PetscErrorCode ierr; 540 PetscBool iascii; 541 PetscViewerFormat format; 542 543 PetscFunctionBegin; 544 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 545 if (iascii) { 546 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 547 if (format == PETSC_VIEWER_ASCII_INFO) { 548 Mat_Pastix *lu=(Mat_Pastix*)A->spptr; 549 550 ierr = PetscViewerASCIIPrintf(viewer,"PaStiX run parameters:\n");CHKERRQ(ierr); 551 ierr = PetscViewerASCIIPrintf(viewer," Matrix type : %s \n",((lu->iparm[IPARM_SYM] == API_SYM_YES) ? "Symmetric" : "Unsymmetric"));CHKERRQ(ierr); 552 ierr = PetscViewerASCIIPrintf(viewer," Level of printing (0,1,2): %d \n",lu->iparm[IPARM_VERBOSE]);CHKERRQ(ierr); 553 ierr = PetscViewerASCIIPrintf(viewer," Number of refinements iterations : %d \n",lu->iparm[IPARM_NBITER]);CHKERRQ(ierr); 554 ierr = PetscPrintf(PETSC_COMM_SELF," Error : %g \n",lu->dparm[DPARM_RELATIVE_ERROR]);CHKERRQ(ierr); 555 } 556 } 557 PetscFunctionReturn(0); 558 } 559 560 561 /*MC 562 MATSOLVERPASTIX - A solver package providing direct solvers (LU) for distributed 563 and sequential matrices via the external package PaStiX. 564 565 Use ./configure --download-pastix to have PETSc installed with PaStiX 566 567 Options Database Keys: 568 + -mat_pastix_verbose <0,1,2> - print level 569 - -mat_pastix_threadnbr <integer> - Set the thread number by MPI task. 570 571 Level: beginner 572 573 .seealso: PCFactorSetMatSolverPackage(), MatSolverPackage 574 575 M*/ 576 577 578 #undef __FUNCT__ 579 #define __FUNCT__ "MatGetInfo_PaStiX" 580 PetscErrorCode MatGetInfo_PaStiX(Mat A,MatInfoType flag,MatInfo *info) 581 { 582 Mat_Pastix *lu =(Mat_Pastix*)A->spptr; 583 584 PetscFunctionBegin; 585 info->block_size = 1.0; 586 info->nz_allocated = lu->iparm[IPARM_NNZEROS]; 587 info->nz_used = lu->iparm[IPARM_NNZEROS]; 588 info->nz_unneeded = 0.0; 589 info->assemblies = 0.0; 590 info->mallocs = 0.0; 591 info->memory = 0.0; 592 info->fill_ratio_given = 0; 593 info->fill_ratio_needed = 0; 594 info->factor_mallocs = 0; 595 PetscFunctionReturn(0); 596 } 597 598 #undef __FUNCT__ 599 #define __FUNCT__ "MatFactorGetSolverPackage_pastix" 600 PetscErrorCode MatFactorGetSolverPackage_pastix(Mat A,const MatSolverPackage *type) 601 { 602 PetscFunctionBegin; 603 *type = MATSOLVERPASTIX; 604 PetscFunctionReturn(0); 605 } 606 607 /* 608 The seq and mpi versions of this function are the same 609 */ 610 #undef __FUNCT__ 611 #define __FUNCT__ "MatGetFactor_seqaij_pastix" 612 PETSC_EXTERN PetscErrorCode MatGetFactor_seqaij_pastix(Mat A,MatFactorType ftype,Mat *F) 613 { 614 Mat B; 615 PetscErrorCode ierr; 616 Mat_Pastix *pastix; 617 618 PetscFunctionBegin; 619 if (ftype != MAT_FACTOR_LU) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Cannot use PETSc AIJ matrices with PaStiX Cholesky, use SBAIJ matrix"); 620 /* Create the factorization matrix */ 621 ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr); 622 ierr = MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N);CHKERRQ(ierr); 623 ierr = MatSetType(B,((PetscObject)A)->type_name);CHKERRQ(ierr); 624 ierr = MatSeqAIJSetPreallocation(B,0,NULL);CHKERRQ(ierr); 625 626 B->ops->lufactorsymbolic = MatLUFactorSymbolic_AIJPASTIX; 627 B->ops->view = MatView_PaStiX; 628 B->ops->getinfo = MatGetInfo_PaStiX; 629 630 ierr = PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverPackage_C",MatFactorGetSolverPackage_pastix);CHKERRQ(ierr); 631 632 B->factortype = MAT_FACTOR_LU; 633 634 ierr = PetscNewLog(B,&pastix);CHKERRQ(ierr); 635 636 pastix->CleanUpPastix = PETSC_FALSE; 637 pastix->isAIJ = PETSC_TRUE; 638 pastix->scat_rhs = NULL; 639 pastix->scat_sol = NULL; 640 pastix->Destroy = B->ops->destroy; 641 B->ops->destroy = MatDestroy_Pastix; 642 B->spptr = (void*)pastix; 643 644 *F = B; 645 PetscFunctionReturn(0); 646 } 647 648 #undef __FUNCT__ 649 #define __FUNCT__ "MatGetFactor_mpiaij_pastix" 650 PETSC_EXTERN PetscErrorCode MatGetFactor_mpiaij_pastix(Mat A,MatFactorType ftype,Mat *F) 651 { 652 Mat B; 653 PetscErrorCode ierr; 654 Mat_Pastix *pastix; 655 656 PetscFunctionBegin; 657 if (ftype != MAT_FACTOR_LU) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Cannot use PETSc AIJ matrices with PaStiX Cholesky, use SBAIJ matrix"); 658 /* Create the factorization matrix */ 659 ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr); 660 ierr = MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N);CHKERRQ(ierr); 661 ierr = MatSetType(B,((PetscObject)A)->type_name);CHKERRQ(ierr); 662 ierr = MatSeqAIJSetPreallocation(B,0,NULL);CHKERRQ(ierr); 663 ierr = MatMPIAIJSetPreallocation(B,0,NULL,0,NULL);CHKERRQ(ierr); 664 665 B->ops->lufactorsymbolic = MatLUFactorSymbolic_AIJPASTIX; 666 B->ops->view = MatView_PaStiX; 667 668 ierr = PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverPackage_C",MatFactorGetSolverPackage_pastix);CHKERRQ(ierr); 669 670 B->factortype = MAT_FACTOR_LU; 671 672 ierr = PetscNewLog(B,&pastix);CHKERRQ(ierr); 673 674 pastix->CleanUpPastix = PETSC_FALSE; 675 pastix->isAIJ = PETSC_TRUE; 676 pastix->scat_rhs = NULL; 677 pastix->scat_sol = NULL; 678 pastix->Destroy = B->ops->destroy; 679 B->ops->destroy = MatDestroy_Pastix; 680 B->spptr = (void*)pastix; 681 682 *F = B; 683 PetscFunctionReturn(0); 684 } 685 686 #undef __FUNCT__ 687 #define __FUNCT__ "MatGetFactor_seqsbaij_pastix" 688 PETSC_EXTERN PetscErrorCode MatGetFactor_seqsbaij_pastix(Mat A,MatFactorType ftype,Mat *F) 689 { 690 Mat B; 691 PetscErrorCode ierr; 692 Mat_Pastix *pastix; 693 694 PetscFunctionBegin; 695 if (ftype != MAT_FACTOR_CHOLESKY) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Cannot use PETSc SBAIJ matrices with PaStiX LU, use AIJ matrix"); 696 /* Create the factorization matrix */ 697 ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr); 698 ierr = MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N);CHKERRQ(ierr); 699 ierr = MatSetType(B,((PetscObject)A)->type_name);CHKERRQ(ierr); 700 ierr = MatSeqSBAIJSetPreallocation(B,1,0,NULL);CHKERRQ(ierr); 701 ierr = MatMPISBAIJSetPreallocation(B,1,0,NULL,0,NULL);CHKERRQ(ierr); 702 703 B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_SBAIJPASTIX; 704 B->ops->view = MatView_PaStiX; 705 706 ierr = PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverPackage_C",MatFactorGetSolverPackage_pastix);CHKERRQ(ierr); 707 708 B->factortype = MAT_FACTOR_CHOLESKY; 709 710 ierr = PetscNewLog(B,&pastix);CHKERRQ(ierr); 711 712 pastix->CleanUpPastix = PETSC_FALSE; 713 pastix->isAIJ = PETSC_TRUE; 714 pastix->scat_rhs = NULL; 715 pastix->scat_sol = NULL; 716 pastix->Destroy = B->ops->destroy; 717 B->ops->destroy = MatDestroy_Pastix; 718 B->spptr = (void*)pastix; 719 720 *F = B; 721 PetscFunctionReturn(0); 722 } 723 724 #undef __FUNCT__ 725 #define __FUNCT__ "MatGetFactor_mpisbaij_pastix" 726 PETSC_EXTERN PetscErrorCode MatGetFactor_mpisbaij_pastix(Mat A,MatFactorType ftype,Mat *F) 727 { 728 Mat B; 729 PetscErrorCode ierr; 730 Mat_Pastix *pastix; 731 732 PetscFunctionBegin; 733 if (ftype != MAT_FACTOR_CHOLESKY) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Cannot use PETSc SBAIJ matrices with PaStiX LU, use AIJ matrix"); 734 735 /* Create the factorization matrix */ 736 ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr); 737 ierr = MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N);CHKERRQ(ierr); 738 ierr = MatSetType(B,((PetscObject)A)->type_name);CHKERRQ(ierr); 739 ierr = MatSeqSBAIJSetPreallocation(B,1,0,NULL);CHKERRQ(ierr); 740 ierr = MatMPISBAIJSetPreallocation(B,1,0,NULL,0,NULL);CHKERRQ(ierr); 741 742 B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_SBAIJPASTIX; 743 B->ops->view = MatView_PaStiX; 744 745 ierr = PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverPackage_C",MatFactorGetSolverPackage_pastix);CHKERRQ(ierr); 746 747 B->factortype = MAT_FACTOR_CHOLESKY; 748 749 ierr = PetscNewLog(B,&pastix);CHKERRQ(ierr); 750 751 pastix->CleanUpPastix = PETSC_FALSE; 752 pastix->isAIJ = PETSC_TRUE; 753 pastix->scat_rhs = NULL; 754 pastix->scat_sol = NULL; 755 pastix->Destroy = B->ops->destroy; 756 B->ops->destroy = MatDestroy_Pastix; 757 B->spptr = (void*)pastix; 758 759 *F = B; 760 PetscFunctionReturn(0); 761 } 762 763