1 #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/ 2 #include <../src/mat/impls/sbaij/seq/sbaij.h> 3 #include <../src/mat/impls/dense/seq/dense.h> 4 5 #if defined(PETSC_HAVE_MKL_INTEL_ILP64) 6 #define MKL_ILP64 7 #endif 8 #include <mkl_pardiso.h> 9 10 PETSC_EXTERN void PetscSetMKL_PARDISOThreads(int); 11 12 /* 13 * Possible mkl_pardiso phases that controls the execution of the solver. 14 * For more information check mkl_pardiso manual. 15 */ 16 #define JOB_ANALYSIS 11 17 #define JOB_ANALYSIS_NUMERICAL_FACTORIZATION 12 18 #define JOB_ANALYSIS_NUMERICAL_FACTORIZATION_SOLVE_ITERATIVE_REFINEMENT 13 19 #define JOB_NUMERICAL_FACTORIZATION 22 20 #define JOB_NUMERICAL_FACTORIZATION_SOLVE_ITERATIVE_REFINEMENT 23 21 #define JOB_SOLVE_ITERATIVE_REFINEMENT 33 22 #define JOB_SOLVE_FORWARD_SUBSTITUTION 331 23 #define JOB_SOLVE_DIAGONAL_SUBSTITUTION 332 24 #define JOB_SOLVE_BACKWARD_SUBSTITUTION 333 25 #define JOB_RELEASE_OF_LU_MEMORY 0 26 #define JOB_RELEASE_OF_ALL_MEMORY -1 27 28 #define IPARM_SIZE 64 29 30 #if defined(PETSC_USE_64BIT_INDICES) 31 #if defined(PETSC_HAVE_MKL_INTEL_ILP64) 32 #define INT_TYPE long long int 33 #define MKL_PARDISO pardiso 34 #define MKL_PARDISO_INIT pardisoinit 35 #else 36 /* this is the case where the MKL BLAS/LAPACK are 32 bit integers but the 64 bit integer version of 37 of Pardiso code is used; hence the need for the 64 below*/ 38 #define INT_TYPE long long int 39 #define MKL_PARDISO pardiso_64 40 #define MKL_PARDISO_INIT pardiso_64init 41 void pardiso_64init(void *pt, INT_TYPE *mtype, INT_TYPE iparm []) 42 { 43 int iparm_copy[IPARM_SIZE], mtype_copy, i; 44 45 mtype_copy = *mtype; 46 pardisoinit(pt, &mtype_copy, iparm_copy); 47 for (i=0; i<IPARM_SIZE; i++) iparm[i] = iparm_copy[i]; 48 } 49 #endif 50 #else 51 #define INT_TYPE int 52 #define MKL_PARDISO pardiso 53 #define MKL_PARDISO_INIT pardisoinit 54 #endif 55 56 57 /* 58 * Internal data structure. 59 * For more information check mkl_pardiso manual. 60 */ 61 typedef struct { 62 63 /* Configuration vector*/ 64 INT_TYPE iparm[IPARM_SIZE]; 65 66 /* 67 * Internal mkl_pardiso memory location. 68 * After the first call to mkl_pardiso do not modify pt, as that could cause a serious memory leak. 69 */ 70 void *pt[IPARM_SIZE]; 71 72 /* Basic mkl_pardiso info*/ 73 INT_TYPE phase, maxfct, mnum, mtype, n, nrhs, msglvl, err; 74 75 /* Matrix structure*/ 76 void *a; 77 INT_TYPE *ia, *ja; 78 79 /* Number of non-zero elements*/ 80 INT_TYPE nz; 81 82 /* Row permutaton vector*/ 83 INT_TYPE *perm; 84 85 /* Define if matrix preserves sparse structure.*/ 86 MatStructure matstruc; 87 88 PetscBool needsym; 89 PetscBool freeaij; 90 91 /* Schur complement */ 92 PetscScalar *schur; 93 PetscInt schur_size; 94 PetscInt *schur_idxs; 95 PetscScalar *schur_work; 96 PetscBLASInt schur_work_size; 97 PetscBool solve_interior; 98 99 /* True if mkl_pardiso function have been used.*/ 100 PetscBool CleanUp; 101 102 /* Conversion to a format suitable for MKL */ 103 PetscErrorCode (*Convert)(Mat, PetscBool, MatReuse, PetscBool*, INT_TYPE*, INT_TYPE**, INT_TYPE**, PetscScalar**); 104 } Mat_MKL_PARDISO; 105 106 PetscErrorCode MatMKLPardiso_Convert_seqsbaij(Mat A,PetscBool sym,MatReuse reuse,PetscBool *free,INT_TYPE *nnz,INT_TYPE **r,INT_TYPE **c,PetscScalar **v) 107 { 108 Mat_SeqSBAIJ *aa = (Mat_SeqSBAIJ*)A->data; 109 PetscInt bs = A->rmap->bs,i; 110 PetscErrorCode ierr; 111 112 PetscFunctionBegin; 113 if (!sym) { 114 SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_PLIB,"This should not happen"); 115 } 116 *v = aa->a; 117 if (bs == 1) { /* already in the correct format */ 118 /* though PetscInt and INT_TYPE are of the same size since they are defined differently the Intel compiler requires a cast */ 119 *r = (INT_TYPE*)aa->i; 120 *c = (INT_TYPE*)aa->j; 121 *nnz = (INT_TYPE)aa->nz; 122 *free = PETSC_FALSE; 123 } else if (reuse == MAT_INITIAL_MATRIX) { 124 PetscInt m = A->rmap->n,nz = aa->nz; 125 PetscInt *row,*col; 126 ierr = PetscMalloc2(m+1,&row,nz,&col);CHKERRQ(ierr); 127 for (i=0; i<m+1; i++) { 128 row[i] = aa->i[i]+1; 129 } 130 for (i=0; i<nz; i++) { 131 col[i] = aa->j[i]+1; 132 } 133 *r = (INT_TYPE*)row; 134 *c = (INT_TYPE*)col; 135 *nnz = (INT_TYPE)nz; 136 *free = PETSC_TRUE; 137 } 138 PetscFunctionReturn(0); 139 } 140 141 PetscErrorCode MatMKLPardiso_Convert_seqbaij(Mat A,PetscBool sym,MatReuse reuse,PetscBool *free,INT_TYPE *nnz,INT_TYPE **r,INT_TYPE **c,PetscScalar **v) 142 { 143 Mat_SeqBAIJ *aa = (Mat_SeqBAIJ*)A->data; 144 PetscInt bs = A->rmap->bs,i; 145 PetscErrorCode ierr; 146 147 PetscFunctionBegin; 148 if (!sym) { 149 *v = aa->a; 150 if (bs == 1) { /* already in the correct format */ 151 /* though PetscInt and INT_TYPE are of the same size since they are defined differently the Intel compiler requires a cast */ 152 *r = (INT_TYPE*)aa->i; 153 *c = (INT_TYPE*)aa->j; 154 *nnz = (INT_TYPE)aa->nz; 155 *free = PETSC_FALSE; 156 PetscFunctionReturn(0); 157 } else if (reuse == MAT_INITIAL_MATRIX) { 158 PetscInt m = A->rmap->n,nz = aa->nz; 159 PetscInt *row,*col; 160 ierr = PetscMalloc2(m+1,&row,nz,&col);CHKERRQ(ierr); 161 for (i=0; i<m+1; i++) { 162 row[i] = aa->i[i]+1; 163 } 164 for (i=0; i<nz; i++) { 165 col[i] = aa->j[i]+1; 166 } 167 *r = (INT_TYPE*)row; 168 *c = (INT_TYPE*)col; 169 *nnz = (INT_TYPE)nz; 170 } 171 *free = PETSC_TRUE; 172 } else { 173 SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_PLIB,"This should not happen"); 174 } 175 PetscFunctionReturn(0); 176 } 177 178 PetscErrorCode MatMKLPardiso_Convert_seqaij(Mat A,PetscBool sym,MatReuse reuse,PetscBool *free,INT_TYPE *nnz,INT_TYPE **r,INT_TYPE **c,PetscScalar **v) 179 { 180 Mat_SeqAIJ *aa = (Mat_SeqAIJ*)A->data; 181 PetscScalar *aav; 182 PetscErrorCode ierr; 183 184 PetscFunctionBegin; 185 ierr = MatSeqAIJGetArrayRead(A,(const PetscScalar**)&aav);CHKERRQ(ierr); 186 if (!sym) { /* already in the correct format */ 187 *v = aav; 188 *r = (INT_TYPE*)aa->i; 189 *c = (INT_TYPE*)aa->j; 190 *nnz = (INT_TYPE)aa->nz; 191 *free = PETSC_FALSE; 192 } else if (reuse == MAT_INITIAL_MATRIX) { /* need to get the triangular part */ 193 PetscScalar *vals,*vv; 194 PetscInt *row,*col,*jj; 195 PetscInt m = A->rmap->n,nz,i; 196 197 nz = 0; 198 for (i=0; i<m; i++) nz += aa->i[i+1] - aa->diag[i]; 199 ierr = PetscMalloc2(m+1,&row,nz,&col);CHKERRQ(ierr); 200 ierr = PetscMalloc1(nz,&vals);CHKERRQ(ierr); 201 jj = col; 202 vv = vals; 203 204 row[0] = 0; 205 for (i=0; i<m; i++) { 206 PetscInt *aj = aa->j + aa->diag[i]; 207 PetscScalar *av = aav + aa->diag[i]; 208 PetscInt rl = aa->i[i+1] - aa->diag[i],j; 209 210 for (j=0; j<rl; j++) { 211 *jj = *aj; jj++; aj++; 212 *vv = *av; vv++; av++; 213 } 214 row[i+1] = row[i] + rl; 215 } 216 *v = vals; 217 *r = (INT_TYPE*)row; 218 *c = (INT_TYPE*)col; 219 *nnz = (INT_TYPE)nz; 220 *free = PETSC_TRUE; 221 } else { 222 PetscScalar *vv; 223 PetscInt m = A->rmap->n,i; 224 225 vv = *v; 226 for (i=0; i<m; i++) { 227 PetscScalar *av = aav + aa->diag[i]; 228 PetscInt rl = aa->i[i+1] - aa->diag[i],j; 229 for (j=0; j<rl; j++) { 230 *vv = *av; vv++; av++; 231 } 232 } 233 *free = PETSC_TRUE; 234 } 235 ierr = MatSeqAIJRestoreArrayRead(A,(const PetscScalar**)&aav);CHKERRQ(ierr); 236 PetscFunctionReturn(0); 237 } 238 239 240 static PetscErrorCode MatMKLPardisoSolveSchur_Private(Mat F, PetscScalar *B, PetscScalar *X) 241 { 242 Mat_MKL_PARDISO *mpardiso = (Mat_MKL_PARDISO*)F->data; 243 Mat S,Xmat,Bmat; 244 MatFactorSchurStatus schurstatus; 245 PetscErrorCode ierr; 246 247 PetscFunctionBegin; 248 ierr = MatFactorFactorizeSchurComplement(F);CHKERRQ(ierr); 249 ierr = MatFactorGetSchurComplement(F,&S,&schurstatus);CHKERRQ(ierr); 250 if (X == B && schurstatus == MAT_FACTOR_SCHUR_INVERTED) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"X and B cannot point to the same address"); 251 ierr = MatCreateSeqDense(PETSC_COMM_SELF,mpardiso->schur_size,mpardiso->nrhs,B,&Bmat);CHKERRQ(ierr); 252 ierr = MatCreateSeqDense(PETSC_COMM_SELF,mpardiso->schur_size,mpardiso->nrhs,X,&Xmat);CHKERRQ(ierr); 253 ierr = MatSetType(Bmat,((PetscObject)S)->type_name);CHKERRQ(ierr); 254 ierr = MatSetType(Xmat,((PetscObject)S)->type_name);CHKERRQ(ierr); 255 #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 256 ierr = MatPinToCPU(Xmat,S->pinnedtocpu);CHKERRQ(ierr); 257 ierr = MatPinToCPU(Bmat,S->pinnedtocpu);CHKERRQ(ierr); 258 #endif 259 260 #if defined(PETSC_USE_COMPLEX) 261 if (mpardiso->iparm[12-1] == 1) SETERRQ(PetscObjectComm((PetscObject)F),PETSC_ERR_SUP,"Hermitian solve not implemented yet"); 262 #endif 263 264 switch (schurstatus) { 265 case MAT_FACTOR_SCHUR_FACTORED: 266 if (!mpardiso->iparm[12-1]) { 267 ierr = MatMatSolve(S,Bmat,Xmat);CHKERRQ(ierr); 268 } else { /* transpose solve */ 269 ierr = MatMatSolveTranspose(S,Bmat,Xmat);CHKERRQ(ierr); 270 } 271 break; 272 case MAT_FACTOR_SCHUR_INVERTED: 273 if (!mpardiso->iparm[12-1]) { 274 ierr = MatMatMult(S,Bmat,MAT_REUSE_MATRIX,PETSC_DEFAULT,&Xmat);CHKERRQ(ierr); 275 } else { /* transpose solve */ 276 ierr = MatTransposeMatMult(S,Bmat,MAT_REUSE_MATRIX,PETSC_DEFAULT,&Xmat);CHKERRQ(ierr); 277 } 278 break; 279 default: 280 SETERRQ1(PetscObjectComm((PetscObject)F),PETSC_ERR_SUP,"Unhandled MatFactorSchurStatus %D",F->schur_status); 281 break; 282 } 283 ierr = MatFactorRestoreSchurComplement(F,&S,schurstatus);CHKERRQ(ierr); 284 ierr = MatDestroy(&Bmat);CHKERRQ(ierr); 285 ierr = MatDestroy(&Xmat);CHKERRQ(ierr); 286 PetscFunctionReturn(0); 287 } 288 289 PetscErrorCode MatFactorSetSchurIS_MKL_PARDISO(Mat F, IS is) 290 { 291 Mat_MKL_PARDISO *mpardiso = (Mat_MKL_PARDISO*)F->data; 292 const PetscScalar *arr; 293 const PetscInt *idxs; 294 PetscInt size,i; 295 PetscMPIInt csize; 296 PetscBool sorted; 297 PetscErrorCode ierr; 298 299 PetscFunctionBegin; 300 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)F),&csize);CHKERRQ(ierr); 301 if (csize > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MKL_PARDISO parallel Schur complements not yet supported from PETSc"); 302 ierr = ISSorted(is,&sorted);CHKERRQ(ierr); 303 if (!sorted) { 304 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"IS for MKL_PARDISO Schur complements needs to be sorted"); 305 } 306 ierr = ISGetLocalSize(is,&size);CHKERRQ(ierr); 307 ierr = PetscFree(mpardiso->schur_work);CHKERRQ(ierr); 308 ierr = PetscBLASIntCast(PetscMax(mpardiso->n,2*size),&mpardiso->schur_work_size);CHKERRQ(ierr); 309 ierr = PetscMalloc1(mpardiso->schur_work_size,&mpardiso->schur_work);CHKERRQ(ierr); 310 ierr = MatDestroy(&F->schur);CHKERRQ(ierr); 311 ierr = MatCreateSeqDense(PETSC_COMM_SELF,size,size,NULL,&F->schur);CHKERRQ(ierr); 312 ierr = MatDenseGetArrayRead(F->schur,&arr);CHKERRQ(ierr); 313 mpardiso->schur = (PetscScalar*)arr; 314 mpardiso->schur_size = size; 315 ierr = MatDenseRestoreArrayRead(F->schur,&arr);CHKERRQ(ierr); 316 if (mpardiso->mtype == 2) { 317 ierr = MatSetOption(F->schur,MAT_SPD,PETSC_TRUE);CHKERRQ(ierr); 318 } 319 320 ierr = PetscFree(mpardiso->schur_idxs);CHKERRQ(ierr); 321 ierr = PetscMalloc1(size,&mpardiso->schur_idxs);CHKERRQ(ierr); 322 ierr = PetscArrayzero(mpardiso->perm,mpardiso->n);CHKERRQ(ierr); 323 ierr = ISGetIndices(is,&idxs);CHKERRQ(ierr); 324 ierr = PetscArraycpy(mpardiso->schur_idxs,idxs,size);CHKERRQ(ierr); 325 for (i=0;i<size;i++) mpardiso->perm[idxs[i]] = 1; 326 ierr = ISRestoreIndices(is,&idxs);CHKERRQ(ierr); 327 if (size) { /* turn on Schur switch if the set of indices is not empty */ 328 mpardiso->iparm[36-1] = 2; 329 } 330 PetscFunctionReturn(0); 331 } 332 333 PetscErrorCode MatDestroy_MKL_PARDISO(Mat A) 334 { 335 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 336 PetscErrorCode ierr; 337 338 PetscFunctionBegin; 339 if (mat_mkl_pardiso->CleanUp) { 340 mat_mkl_pardiso->phase = JOB_RELEASE_OF_ALL_MEMORY; 341 342 MKL_PARDISO (mat_mkl_pardiso->pt, 343 &mat_mkl_pardiso->maxfct, 344 &mat_mkl_pardiso->mnum, 345 &mat_mkl_pardiso->mtype, 346 &mat_mkl_pardiso->phase, 347 &mat_mkl_pardiso->n, 348 NULL, 349 NULL, 350 NULL, 351 NULL, 352 &mat_mkl_pardiso->nrhs, 353 mat_mkl_pardiso->iparm, 354 &mat_mkl_pardiso->msglvl, 355 NULL, 356 NULL, 357 &mat_mkl_pardiso->err); 358 } 359 ierr = PetscFree(mat_mkl_pardiso->perm);CHKERRQ(ierr); 360 ierr = PetscFree(mat_mkl_pardiso->schur_work);CHKERRQ(ierr); 361 ierr = PetscFree(mat_mkl_pardiso->schur_idxs);CHKERRQ(ierr); 362 if (mat_mkl_pardiso->freeaij) { 363 ierr = PetscFree2(mat_mkl_pardiso->ia,mat_mkl_pardiso->ja);CHKERRQ(ierr); 364 if (mat_mkl_pardiso->iparm[34] == 1) { 365 ierr = PetscFree(mat_mkl_pardiso->a);CHKERRQ(ierr); 366 } 367 } 368 ierr = PetscFree(A->data);CHKERRQ(ierr); 369 370 /* clear composed functions */ 371 ierr = PetscObjectComposeFunction((PetscObject)A,"MatFactorGetSolverType_C",NULL);CHKERRQ(ierr); 372 ierr = PetscObjectComposeFunction((PetscObject)A,"MatFactorSetSchurIS_C",NULL);CHKERRQ(ierr); 373 ierr = PetscObjectComposeFunction((PetscObject)A,"MatMkl_PardisoSetCntl_C",NULL);CHKERRQ(ierr); 374 PetscFunctionReturn(0); 375 } 376 377 static PetscErrorCode MatMKLPardisoScatterSchur_Private(Mat_MKL_PARDISO *mpardiso, PetscScalar *whole, PetscScalar *schur, PetscBool reduce) 378 { 379 PetscFunctionBegin; 380 if (reduce) { /* data given for the whole matrix */ 381 PetscInt i,m=0,p=0; 382 for (i=0;i<mpardiso->nrhs;i++) { 383 PetscInt j; 384 for (j=0;j<mpardiso->schur_size;j++) { 385 schur[p+j] = whole[m+mpardiso->schur_idxs[j]]; 386 } 387 m += mpardiso->n; 388 p += mpardiso->schur_size; 389 } 390 } else { /* from Schur to whole */ 391 PetscInt i,m=0,p=0; 392 for (i=0;i<mpardiso->nrhs;i++) { 393 PetscInt j; 394 for (j=0;j<mpardiso->schur_size;j++) { 395 whole[m+mpardiso->schur_idxs[j]] = schur[p+j]; 396 } 397 m += mpardiso->n; 398 p += mpardiso->schur_size; 399 } 400 } 401 PetscFunctionReturn(0); 402 } 403 404 PetscErrorCode MatSolve_MKL_PARDISO(Mat A,Vec b,Vec x) 405 { 406 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 407 PetscErrorCode ierr; 408 PetscScalar *xarray; 409 const PetscScalar *barray; 410 411 PetscFunctionBegin; 412 mat_mkl_pardiso->nrhs = 1; 413 ierr = VecGetArray(x,&xarray);CHKERRQ(ierr); 414 ierr = VecGetArrayRead(b,&barray);CHKERRQ(ierr); 415 416 if (!mat_mkl_pardiso->schur) mat_mkl_pardiso->phase = JOB_SOLVE_ITERATIVE_REFINEMENT; 417 else mat_mkl_pardiso->phase = JOB_SOLVE_FORWARD_SUBSTITUTION; 418 419 if (barray == xarray) { /* if the two vectors share the same memory */ 420 PetscScalar *work; 421 if (!mat_mkl_pardiso->schur_work) { 422 ierr = PetscMalloc1(mat_mkl_pardiso->n,&work);CHKERRQ(ierr); 423 } else { 424 work = mat_mkl_pardiso->schur_work; 425 } 426 mat_mkl_pardiso->iparm[6-1] = 1; 427 MKL_PARDISO (mat_mkl_pardiso->pt, 428 &mat_mkl_pardiso->maxfct, 429 &mat_mkl_pardiso->mnum, 430 &mat_mkl_pardiso->mtype, 431 &mat_mkl_pardiso->phase, 432 &mat_mkl_pardiso->n, 433 mat_mkl_pardiso->a, 434 mat_mkl_pardiso->ia, 435 mat_mkl_pardiso->ja, 436 NULL, 437 &mat_mkl_pardiso->nrhs, 438 mat_mkl_pardiso->iparm, 439 &mat_mkl_pardiso->msglvl, 440 (void*)xarray, 441 (void*)work, 442 &mat_mkl_pardiso->err); 443 if (!mat_mkl_pardiso->schur_work) { 444 ierr = PetscFree(work);CHKERRQ(ierr); 445 } 446 } else { 447 mat_mkl_pardiso->iparm[6-1] = 0; 448 MKL_PARDISO (mat_mkl_pardiso->pt, 449 &mat_mkl_pardiso->maxfct, 450 &mat_mkl_pardiso->mnum, 451 &mat_mkl_pardiso->mtype, 452 &mat_mkl_pardiso->phase, 453 &mat_mkl_pardiso->n, 454 mat_mkl_pardiso->a, 455 mat_mkl_pardiso->ia, 456 mat_mkl_pardiso->ja, 457 mat_mkl_pardiso->perm, 458 &mat_mkl_pardiso->nrhs, 459 mat_mkl_pardiso->iparm, 460 &mat_mkl_pardiso->msglvl, 461 (void*)barray, 462 (void*)xarray, 463 &mat_mkl_pardiso->err); 464 } 465 ierr = VecRestoreArrayRead(b,&barray);CHKERRQ(ierr); 466 467 if (mat_mkl_pardiso->err < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 468 469 if (mat_mkl_pardiso->schur) { /* solve Schur complement and expand solution */ 470 PetscInt shift = mat_mkl_pardiso->schur_size; 471 472 /* if inverted, uses BLAS *MM subroutines, otherwise LAPACK *TRS */ 473 if (A->schur_status != MAT_FACTOR_SCHUR_INVERTED) shift = 0; 474 475 if (!mat_mkl_pardiso->solve_interior) { 476 /* solve Schur complement */ 477 ierr = MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work,PETSC_TRUE);CHKERRQ(ierr); 478 ierr = MatMKLPardisoSolveSchur_Private(A,mat_mkl_pardiso->schur_work,mat_mkl_pardiso->schur_work+shift);CHKERRQ(ierr); 479 ierr = MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work+shift,PETSC_FALSE);CHKERRQ(ierr); 480 } else { /* if we are solving for the interior problem, any value in barray[schur] forward-substituted to xarray[schur] will be neglected */ 481 PetscInt i; 482 for (i=0;i<mat_mkl_pardiso->schur_size;i++) { 483 xarray[mat_mkl_pardiso->schur_idxs[i]] = 0.; 484 } 485 } 486 487 /* expansion phase */ 488 mat_mkl_pardiso->iparm[6-1] = 1; 489 mat_mkl_pardiso->phase = JOB_SOLVE_BACKWARD_SUBSTITUTION; 490 MKL_PARDISO (mat_mkl_pardiso->pt, 491 &mat_mkl_pardiso->maxfct, 492 &mat_mkl_pardiso->mnum, 493 &mat_mkl_pardiso->mtype, 494 &mat_mkl_pardiso->phase, 495 &mat_mkl_pardiso->n, 496 mat_mkl_pardiso->a, 497 mat_mkl_pardiso->ia, 498 mat_mkl_pardiso->ja, 499 mat_mkl_pardiso->perm, 500 &mat_mkl_pardiso->nrhs, 501 mat_mkl_pardiso->iparm, 502 &mat_mkl_pardiso->msglvl, 503 (void*)xarray, 504 (void*)mat_mkl_pardiso->schur_work, /* according to the specs, the solution vector is always used */ 505 &mat_mkl_pardiso->err); 506 507 if (mat_mkl_pardiso->err < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 508 mat_mkl_pardiso->iparm[6-1] = 0; 509 } 510 ierr = VecRestoreArray(x,&xarray);CHKERRQ(ierr); 511 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 512 PetscFunctionReturn(0); 513 } 514 515 PetscErrorCode MatSolveTranspose_MKL_PARDISO(Mat A,Vec b,Vec x) 516 { 517 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 518 PetscInt oiparm12; 519 PetscErrorCode ierr; 520 521 PetscFunctionBegin; 522 oiparm12 = mat_mkl_pardiso->iparm[12 - 1]; 523 mat_mkl_pardiso->iparm[12 - 1] = 2; 524 ierr = MatSolve_MKL_PARDISO(A,b,x);CHKERRQ(ierr); 525 mat_mkl_pardiso->iparm[12 - 1] = oiparm12; 526 PetscFunctionReturn(0); 527 } 528 529 PetscErrorCode MatMatSolve_MKL_PARDISO(Mat A,Mat B,Mat X) 530 { 531 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)(A)->data; 532 PetscErrorCode ierr; 533 const PetscScalar *barray; 534 PetscScalar *xarray; 535 PetscBool flg; 536 537 PetscFunctionBegin; 538 ierr = PetscObjectBaseTypeCompare((PetscObject)B,MATSEQDENSE,&flg);CHKERRQ(ierr); 539 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix B must be MATSEQDENSE matrix"); 540 ierr = PetscObjectBaseTypeCompare((PetscObject)X,MATSEQDENSE,&flg);CHKERRQ(ierr); 541 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix X must be MATSEQDENSE matrix"); 542 543 ierr = MatGetSize(B,NULL,(PetscInt*)&mat_mkl_pardiso->nrhs);CHKERRQ(ierr); 544 545 if (mat_mkl_pardiso->nrhs > 0) { 546 ierr = MatDenseGetArrayRead(B,&barray);CHKERRQ(ierr); 547 ierr = MatDenseGetArray(X,&xarray);CHKERRQ(ierr); 548 549 if (barray == xarray) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"B and X cannot share the same memory location"); 550 if (!mat_mkl_pardiso->schur) mat_mkl_pardiso->phase = JOB_SOLVE_ITERATIVE_REFINEMENT; 551 else mat_mkl_pardiso->phase = JOB_SOLVE_FORWARD_SUBSTITUTION; 552 mat_mkl_pardiso->iparm[6-1] = 0; 553 554 MKL_PARDISO (mat_mkl_pardiso->pt, 555 &mat_mkl_pardiso->maxfct, 556 &mat_mkl_pardiso->mnum, 557 &mat_mkl_pardiso->mtype, 558 &mat_mkl_pardiso->phase, 559 &mat_mkl_pardiso->n, 560 mat_mkl_pardiso->a, 561 mat_mkl_pardiso->ia, 562 mat_mkl_pardiso->ja, 563 mat_mkl_pardiso->perm, 564 &mat_mkl_pardiso->nrhs, 565 mat_mkl_pardiso->iparm, 566 &mat_mkl_pardiso->msglvl, 567 (void*)barray, 568 (void*)xarray, 569 &mat_mkl_pardiso->err); 570 if (mat_mkl_pardiso->err < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 571 572 ierr = MatDenseRestoreArrayRead(B,&barray);CHKERRQ(ierr); 573 if (mat_mkl_pardiso->schur) { /* solve Schur complement and expand solution */ 574 PetscScalar *o_schur_work = NULL; 575 PetscInt shift = mat_mkl_pardiso->schur_size*mat_mkl_pardiso->nrhs,scale; 576 PetscInt mem = mat_mkl_pardiso->n*mat_mkl_pardiso->nrhs; 577 578 /* allocate extra memory if it is needed */ 579 scale = 1; 580 if (A->schur_status == MAT_FACTOR_SCHUR_INVERTED) scale = 2; 581 582 mem *= scale; 583 if (mem > mat_mkl_pardiso->schur_work_size) { 584 o_schur_work = mat_mkl_pardiso->schur_work; 585 ierr = PetscMalloc1(mem,&mat_mkl_pardiso->schur_work);CHKERRQ(ierr); 586 } 587 588 /* if inverted, uses BLAS *MM subroutines, otherwise LAPACK *TRS */ 589 if (A->schur_status != MAT_FACTOR_SCHUR_INVERTED) shift = 0; 590 591 /* solve Schur complement */ 592 if (!mat_mkl_pardiso->solve_interior) { 593 ierr = MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work,PETSC_TRUE);CHKERRQ(ierr); 594 ierr = MatMKLPardisoSolveSchur_Private(A,mat_mkl_pardiso->schur_work,mat_mkl_pardiso->schur_work+shift);CHKERRQ(ierr); 595 ierr = MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work+shift,PETSC_FALSE);CHKERRQ(ierr); 596 } else { /* if we are solving for the interior problem, any value in barray[schur,n] forward-substituted to xarray[schur,n] will be neglected */ 597 PetscInt i,n,m=0; 598 for (n=0;n<mat_mkl_pardiso->nrhs;n++) { 599 for (i=0;i<mat_mkl_pardiso->schur_size;i++) { 600 xarray[mat_mkl_pardiso->schur_idxs[i]+m] = 0.; 601 } 602 m += mat_mkl_pardiso->n; 603 } 604 } 605 606 /* expansion phase */ 607 mat_mkl_pardiso->iparm[6-1] = 1; 608 mat_mkl_pardiso->phase = JOB_SOLVE_BACKWARD_SUBSTITUTION; 609 MKL_PARDISO (mat_mkl_pardiso->pt, 610 &mat_mkl_pardiso->maxfct, 611 &mat_mkl_pardiso->mnum, 612 &mat_mkl_pardiso->mtype, 613 &mat_mkl_pardiso->phase, 614 &mat_mkl_pardiso->n, 615 mat_mkl_pardiso->a, 616 mat_mkl_pardiso->ia, 617 mat_mkl_pardiso->ja, 618 mat_mkl_pardiso->perm, 619 &mat_mkl_pardiso->nrhs, 620 mat_mkl_pardiso->iparm, 621 &mat_mkl_pardiso->msglvl, 622 (void*)xarray, 623 (void*)mat_mkl_pardiso->schur_work, /* according to the specs, the solution vector is always used */ 624 &mat_mkl_pardiso->err); 625 if (o_schur_work) { /* restore original schur_work (minimal size) */ 626 ierr = PetscFree(mat_mkl_pardiso->schur_work);CHKERRQ(ierr); 627 mat_mkl_pardiso->schur_work = o_schur_work; 628 } 629 if (mat_mkl_pardiso->err < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 630 mat_mkl_pardiso->iparm[6-1] = 0; 631 } 632 } 633 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 634 PetscFunctionReturn(0); 635 } 636 637 PetscErrorCode MatFactorNumeric_MKL_PARDISO(Mat F,Mat A,const MatFactorInfo *info) 638 { 639 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)(F)->data; 640 PetscErrorCode ierr; 641 642 PetscFunctionBegin; 643 mat_mkl_pardiso->matstruc = SAME_NONZERO_PATTERN; 644 ierr = (*mat_mkl_pardiso->Convert)(A,mat_mkl_pardiso->needsym,MAT_REUSE_MATRIX,&mat_mkl_pardiso->freeaij,&mat_mkl_pardiso->nz,&mat_mkl_pardiso->ia,&mat_mkl_pardiso->ja,(PetscScalar**)&mat_mkl_pardiso->a);CHKERRQ(ierr); 645 646 mat_mkl_pardiso->phase = JOB_NUMERICAL_FACTORIZATION; 647 MKL_PARDISO (mat_mkl_pardiso->pt, 648 &mat_mkl_pardiso->maxfct, 649 &mat_mkl_pardiso->mnum, 650 &mat_mkl_pardiso->mtype, 651 &mat_mkl_pardiso->phase, 652 &mat_mkl_pardiso->n, 653 mat_mkl_pardiso->a, 654 mat_mkl_pardiso->ia, 655 mat_mkl_pardiso->ja, 656 mat_mkl_pardiso->perm, 657 &mat_mkl_pardiso->nrhs, 658 mat_mkl_pardiso->iparm, 659 &mat_mkl_pardiso->msglvl, 660 NULL, 661 (void*)mat_mkl_pardiso->schur, 662 &mat_mkl_pardiso->err); 663 if (mat_mkl_pardiso->err < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 664 665 /* report flops */ 666 if (mat_mkl_pardiso->iparm[18] > 0) { 667 ierr = PetscLogFlops(PetscPowRealInt(10.,6)*mat_mkl_pardiso->iparm[18]);CHKERRQ(ierr); 668 } 669 670 if (F->schur) { /* schur output from pardiso is in row major format */ 671 #if defined(PETSC_HAVE_CUDA) 672 F->schur->valid_GPU_matrix = PETSC_OFFLOAD_CPU; 673 #endif 674 ierr = MatFactorRestoreSchurComplement(F,NULL,MAT_FACTOR_SCHUR_UNFACTORED);CHKERRQ(ierr); 675 ierr = MatTranspose(F->schur,MAT_INPLACE_MATRIX,&F->schur);CHKERRQ(ierr); 676 } 677 mat_mkl_pardiso->matstruc = SAME_NONZERO_PATTERN; 678 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 679 PetscFunctionReturn(0); 680 } 681 682 PetscErrorCode PetscSetMKL_PARDISOFromOptions(Mat F, Mat A) 683 { 684 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)F->data; 685 PetscErrorCode ierr; 686 PetscInt icntl,threads=1; 687 PetscBool flg; 688 689 PetscFunctionBegin; 690 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)A),((PetscObject)A)->prefix,"MKL_PARDISO Options","Mat");CHKERRQ(ierr); 691 692 ierr = PetscOptionsInt("-mat_mkl_pardiso_65","Number of threads to use within PARDISO","None",threads,&threads,&flg);CHKERRQ(ierr); 693 if (flg) PetscSetMKL_PARDISOThreads((int)threads); 694 695 ierr = PetscOptionsInt("-mat_mkl_pardiso_66","Maximum number of factors with identical sparsity structure that must be kept in memory at the same time","None",mat_mkl_pardiso->maxfct,&icntl,&flg);CHKERRQ(ierr); 696 if (flg) mat_mkl_pardiso->maxfct = icntl; 697 698 ierr = PetscOptionsInt("-mat_mkl_pardiso_67","Indicates the actual matrix for the solution phase","None",mat_mkl_pardiso->mnum,&icntl,&flg);CHKERRQ(ierr); 699 if (flg) mat_mkl_pardiso->mnum = icntl; 700 701 ierr = PetscOptionsInt("-mat_mkl_pardiso_68","Message level information","None",mat_mkl_pardiso->msglvl,&icntl,&flg);CHKERRQ(ierr); 702 if (flg) mat_mkl_pardiso->msglvl = icntl; 703 704 ierr = PetscOptionsInt("-mat_mkl_pardiso_69","Defines the matrix type","None",mat_mkl_pardiso->mtype,&icntl,&flg);CHKERRQ(ierr); 705 if (flg) { 706 void *pt[IPARM_SIZE]; 707 mat_mkl_pardiso->mtype = icntl; 708 MKL_PARDISO_INIT(pt, &mat_mkl_pardiso->mtype, mat_mkl_pardiso->iparm); 709 #if defined(PETSC_USE_REAL_SINGLE) 710 mat_mkl_pardiso->iparm[27] = 1; 711 #else 712 mat_mkl_pardiso->iparm[27] = 0; 713 #endif 714 mat_mkl_pardiso->iparm[34] = 1; /* use 0-based indexing */ 715 } 716 ierr = PetscOptionsInt("-mat_mkl_pardiso_1","Use default values (if 0)","None",mat_mkl_pardiso->iparm[0],&icntl,&flg);CHKERRQ(ierr); 717 718 if (flg && icntl != 0) { 719 ierr = PetscOptionsInt("-mat_mkl_pardiso_2","Fill-in reducing ordering for the input matrix","None",mat_mkl_pardiso->iparm[1],&icntl,&flg);CHKERRQ(ierr); 720 if (flg) mat_mkl_pardiso->iparm[1] = icntl; 721 722 ierr = PetscOptionsInt("-mat_mkl_pardiso_4","Preconditioned CGS/CG","None",mat_mkl_pardiso->iparm[3],&icntl,&flg);CHKERRQ(ierr); 723 if (flg) mat_mkl_pardiso->iparm[3] = icntl; 724 725 ierr = PetscOptionsInt("-mat_mkl_pardiso_5","User permutation","None",mat_mkl_pardiso->iparm[4],&icntl,&flg);CHKERRQ(ierr); 726 if (flg) mat_mkl_pardiso->iparm[4] = icntl; 727 728 ierr = PetscOptionsInt("-mat_mkl_pardiso_6","Write solution on x","None",mat_mkl_pardiso->iparm[5],&icntl,&flg);CHKERRQ(ierr); 729 if (flg) mat_mkl_pardiso->iparm[5] = icntl; 730 731 ierr = PetscOptionsInt("-mat_mkl_pardiso_8","Iterative refinement step","None",mat_mkl_pardiso->iparm[7],&icntl,&flg);CHKERRQ(ierr); 732 if (flg) mat_mkl_pardiso->iparm[7] = icntl; 733 734 ierr = PetscOptionsInt("-mat_mkl_pardiso_10","Pivoting perturbation","None",mat_mkl_pardiso->iparm[9],&icntl,&flg);CHKERRQ(ierr); 735 if (flg) mat_mkl_pardiso->iparm[9] = icntl; 736 737 ierr = PetscOptionsInt("-mat_mkl_pardiso_11","Scaling vectors","None",mat_mkl_pardiso->iparm[10],&icntl,&flg);CHKERRQ(ierr); 738 if (flg) mat_mkl_pardiso->iparm[10] = icntl; 739 740 ierr = PetscOptionsInt("-mat_mkl_pardiso_12","Solve with transposed or conjugate transposed matrix A","None",mat_mkl_pardiso->iparm[11],&icntl,&flg);CHKERRQ(ierr); 741 if (flg) mat_mkl_pardiso->iparm[11] = icntl; 742 743 ierr = PetscOptionsInt("-mat_mkl_pardiso_13","Improved accuracy using (non-) symmetric weighted matching","None",mat_mkl_pardiso->iparm[12],&icntl,&flg);CHKERRQ(ierr); 744 if (flg) mat_mkl_pardiso->iparm[12] = icntl; 745 746 ierr = PetscOptionsInt("-mat_mkl_pardiso_18","Numbers of non-zero elements","None",mat_mkl_pardiso->iparm[17],&icntl,&flg);CHKERRQ(ierr); 747 if (flg) mat_mkl_pardiso->iparm[17] = icntl; 748 749 ierr = PetscOptionsInt("-mat_mkl_pardiso_19","Report number of floating point operations (0 to disable)","None",mat_mkl_pardiso->iparm[18],&icntl,&flg);CHKERRQ(ierr); 750 if (flg) mat_mkl_pardiso->iparm[18] = icntl; 751 752 ierr = PetscOptionsInt("-mat_mkl_pardiso_21","Pivoting for symmetric indefinite matrices","None",mat_mkl_pardiso->iparm[20],&icntl,&flg);CHKERRQ(ierr); 753 if (flg) mat_mkl_pardiso->iparm[20] = icntl; 754 755 ierr = PetscOptionsInt("-mat_mkl_pardiso_24","Parallel factorization control","None",mat_mkl_pardiso->iparm[23],&icntl,&flg);CHKERRQ(ierr); 756 if (flg) mat_mkl_pardiso->iparm[23] = icntl; 757 758 ierr = PetscOptionsInt("-mat_mkl_pardiso_25","Parallel forward/backward solve control","None",mat_mkl_pardiso->iparm[24],&icntl,&flg);CHKERRQ(ierr); 759 if (flg) mat_mkl_pardiso->iparm[24] = icntl; 760 761 ierr = PetscOptionsInt("-mat_mkl_pardiso_27","Matrix checker","None",mat_mkl_pardiso->iparm[26],&icntl,&flg);CHKERRQ(ierr); 762 if (flg) mat_mkl_pardiso->iparm[26] = icntl; 763 764 ierr = PetscOptionsInt("-mat_mkl_pardiso_31","Partial solve and computing selected components of the solution vectors","None",mat_mkl_pardiso->iparm[30],&icntl,&flg);CHKERRQ(ierr); 765 if (flg) mat_mkl_pardiso->iparm[30] = icntl; 766 767 ierr = PetscOptionsInt("-mat_mkl_pardiso_34","Optimal number of threads for conditional numerical reproducibility (CNR) mode","None",mat_mkl_pardiso->iparm[33],&icntl,&flg);CHKERRQ(ierr); 768 if (flg) mat_mkl_pardiso->iparm[33] = icntl; 769 770 ierr = PetscOptionsInt("-mat_mkl_pardiso_60","Intel MKL_PARDISO mode","None",mat_mkl_pardiso->iparm[59],&icntl,&flg);CHKERRQ(ierr); 771 if (flg) mat_mkl_pardiso->iparm[59] = icntl; 772 } 773 PetscOptionsEnd(); 774 PetscFunctionReturn(0); 775 } 776 777 PetscErrorCode MatFactorMKL_PARDISOInitialize_Private(Mat A, MatFactorType ftype, Mat_MKL_PARDISO *mat_mkl_pardiso) 778 { 779 PetscErrorCode ierr; 780 PetscInt i,bs; 781 PetscBool match; 782 783 PetscFunctionBegin; 784 for (i=0; i<IPARM_SIZE; i++) mat_mkl_pardiso->iparm[i] = 0; 785 for (i=0; i<IPARM_SIZE; i++) mat_mkl_pardiso->pt[i] = 0; 786 /* Default options for both sym and unsym */ 787 mat_mkl_pardiso->iparm[ 0] = 1; /* Solver default parameters overriden with provided by iparm */ 788 mat_mkl_pardiso->iparm[ 1] = 2; /* Metis reordering */ 789 mat_mkl_pardiso->iparm[ 5] = 0; /* Write solution into x */ 790 mat_mkl_pardiso->iparm[ 7] = 0; /* Max number of iterative refinement steps */ 791 mat_mkl_pardiso->iparm[17] = -1; /* Output: Number of nonzeros in the factor LU */ 792 mat_mkl_pardiso->iparm[18] = -1; /* Output: Mflops for LU factorization */ 793 #if 0 794 mat_mkl_pardiso->iparm[23] = 1; /* Parallel factorization control*/ 795 #endif 796 ierr = PetscObjectTypeCompareAny((PetscObject)A,&match,MATSEQBAIJ,MATSEQSBAIJ,"");CHKERRQ(ierr); 797 ierr = MatGetBlockSize(A,&bs);CHKERRQ(ierr); 798 if (!match || bs == 1) { 799 mat_mkl_pardiso->iparm[34] = 1; /* Cluster Sparse Solver use C-style indexing for ia and ja arrays */ 800 mat_mkl_pardiso->n = A->rmap->N; 801 } else { 802 mat_mkl_pardiso->iparm[34] = 0; /* Cluster Sparse Solver use Fortran-style indexing for ia and ja arrays */ 803 mat_mkl_pardiso->iparm[36] = bs; 804 mat_mkl_pardiso->n = A->rmap->N/bs; 805 } 806 mat_mkl_pardiso->iparm[39] = 0; /* Input: matrix/rhs/solution stored on master */ 807 808 mat_mkl_pardiso->CleanUp = PETSC_FALSE; 809 mat_mkl_pardiso->maxfct = 1; /* Maximum number of numerical factorizations. */ 810 mat_mkl_pardiso->mnum = 1; /* Which factorization to use. */ 811 mat_mkl_pardiso->msglvl = 0; /* 0: do not print 1: Print statistical information in file */ 812 mat_mkl_pardiso->phase = -1; 813 mat_mkl_pardiso->err = 0; 814 815 mat_mkl_pardiso->nrhs = 1; 816 mat_mkl_pardiso->err = 0; 817 mat_mkl_pardiso->phase = -1; 818 819 if (ftype == MAT_FACTOR_LU) { 820 mat_mkl_pardiso->iparm[ 9] = 13; /* Perturb the pivot elements with 1E-13 */ 821 mat_mkl_pardiso->iparm[10] = 1; /* Use nonsymmetric permutation and scaling MPS */ 822 mat_mkl_pardiso->iparm[12] = 1; /* Switch on Maximum Weighted Matching algorithm (default for non-symmetric) */ 823 } else { 824 mat_mkl_pardiso->iparm[ 9] = 8; /* Perturb the pivot elements with 1E-8 */ 825 mat_mkl_pardiso->iparm[10] = 0; /* Use nonsymmetric permutation and scaling MPS */ 826 mat_mkl_pardiso->iparm[12] = 1; /* Switch on Maximum Weighted Matching algorithm (default for non-symmetric) */ 827 #if defined(PETSC_USE_DEBUG) 828 mat_mkl_pardiso->iparm[26] = 1; /* Matrix checker */ 829 #endif 830 } 831 ierr = PetscMalloc1(A->rmap->N*sizeof(INT_TYPE), &mat_mkl_pardiso->perm);CHKERRQ(ierr); 832 for (i=0; i<A->rmap->N; i++) { 833 mat_mkl_pardiso->perm[i] = 0; 834 } 835 mat_mkl_pardiso->schur_size = 0; 836 PetscFunctionReturn(0); 837 } 838 839 PetscErrorCode MatFactorSymbolic_AIJMKL_PARDISO_Private(Mat F,Mat A,const MatFactorInfo *info) 840 { 841 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)F->data; 842 PetscErrorCode ierr; 843 844 PetscFunctionBegin; 845 mat_mkl_pardiso->matstruc = DIFFERENT_NONZERO_PATTERN; 846 ierr = PetscSetMKL_PARDISOFromOptions(F,A);CHKERRQ(ierr); 847 848 /* throw away any previously computed structure */ 849 if (mat_mkl_pardiso->freeaij) { 850 ierr = PetscFree2(mat_mkl_pardiso->ia,mat_mkl_pardiso->ja);CHKERRQ(ierr); 851 if (mat_mkl_pardiso->iparm[34] == 1) { 852 ierr = PetscFree(mat_mkl_pardiso->a);CHKERRQ(ierr); 853 } 854 } 855 ierr = (*mat_mkl_pardiso->Convert)(A,mat_mkl_pardiso->needsym,MAT_INITIAL_MATRIX,&mat_mkl_pardiso->freeaij,&mat_mkl_pardiso->nz,&mat_mkl_pardiso->ia,&mat_mkl_pardiso->ja,(PetscScalar**)&mat_mkl_pardiso->a);CHKERRQ(ierr); 856 if (mat_mkl_pardiso->iparm[34] == 1) mat_mkl_pardiso->n = A->rmap->N; 857 else mat_mkl_pardiso->n = A->rmap->N/A->rmap->bs; 858 859 mat_mkl_pardiso->phase = JOB_ANALYSIS; 860 861 /* reset flops counting if requested */ 862 if (mat_mkl_pardiso->iparm[18]) mat_mkl_pardiso->iparm[18] = -1; 863 864 MKL_PARDISO (mat_mkl_pardiso->pt, 865 &mat_mkl_pardiso->maxfct, 866 &mat_mkl_pardiso->mnum, 867 &mat_mkl_pardiso->mtype, 868 &mat_mkl_pardiso->phase, 869 &mat_mkl_pardiso->n, 870 mat_mkl_pardiso->a, 871 mat_mkl_pardiso->ia, 872 mat_mkl_pardiso->ja, 873 mat_mkl_pardiso->perm, 874 &mat_mkl_pardiso->nrhs, 875 mat_mkl_pardiso->iparm, 876 &mat_mkl_pardiso->msglvl, 877 NULL, 878 NULL, 879 &mat_mkl_pardiso->err); 880 if (mat_mkl_pardiso->err < 0) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 881 882 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 883 884 if (F->factortype == MAT_FACTOR_LU) F->ops->lufactornumeric = MatFactorNumeric_MKL_PARDISO; 885 else F->ops->choleskyfactornumeric = MatFactorNumeric_MKL_PARDISO; 886 887 F->ops->solve = MatSolve_MKL_PARDISO; 888 F->ops->solvetranspose = MatSolveTranspose_MKL_PARDISO; 889 F->ops->matsolve = MatMatSolve_MKL_PARDISO; 890 PetscFunctionReturn(0); 891 } 892 893 PetscErrorCode MatLUFactorSymbolic_AIJMKL_PARDISO(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info) 894 { 895 PetscErrorCode ierr; 896 897 PetscFunctionBegin; 898 ierr = MatFactorSymbolic_AIJMKL_PARDISO_Private(F, A, info);CHKERRQ(ierr); 899 PetscFunctionReturn(0); 900 } 901 902 #if !defined(PETSC_USE_COMPLEX) 903 PetscErrorCode MatGetInertia_MKL_PARDISO(Mat F,PetscInt *nneg,PetscInt *nzero,PetscInt *npos) 904 { 905 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)F->data; 906 907 PetscFunctionBegin; 908 if (nneg) *nneg = mat_mkl_pardiso->iparm[22]; 909 if (npos) *npos = mat_mkl_pardiso->iparm[21]; 910 if (nzero) *nzero = F->rmap->N - (mat_mkl_pardiso->iparm[22] + mat_mkl_pardiso->iparm[21]); 911 PetscFunctionReturn(0); 912 } 913 #endif 914 915 PetscErrorCode MatCholeskyFactorSymbolic_AIJMKL_PARDISO(Mat F,Mat A,IS r,const MatFactorInfo *info) 916 { 917 PetscErrorCode ierr; 918 919 PetscFunctionBegin; 920 ierr = MatFactorSymbolic_AIJMKL_PARDISO_Private(F, A, info);CHKERRQ(ierr); 921 #if defined(PETSC_USE_COMPLEX) 922 F->ops->getinertia = NULL; 923 #else 924 F->ops->getinertia = MatGetInertia_MKL_PARDISO; 925 #endif 926 PetscFunctionReturn(0); 927 } 928 929 PetscErrorCode MatView_MKL_PARDISO(Mat A, PetscViewer viewer) 930 { 931 PetscErrorCode ierr; 932 PetscBool iascii; 933 PetscViewerFormat format; 934 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 935 PetscInt i; 936 937 PetscFunctionBegin; 938 if (A->ops->solve != MatSolve_MKL_PARDISO) PetscFunctionReturn(0); 939 940 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 941 if (iascii) { 942 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 943 if (format == PETSC_VIEWER_ASCII_INFO) { 944 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO run parameters:\n");CHKERRQ(ierr); 945 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO phase: %d \n",mat_mkl_pardiso->phase);CHKERRQ(ierr); 946 for (i=1; i<=64; i++) { 947 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO iparm[%d]: %d \n",i, mat_mkl_pardiso->iparm[i - 1]);CHKERRQ(ierr); 948 } 949 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO maxfct: %d \n", mat_mkl_pardiso->maxfct);CHKERRQ(ierr); 950 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO mnum: %d \n", mat_mkl_pardiso->mnum);CHKERRQ(ierr); 951 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO mtype: %d \n", mat_mkl_pardiso->mtype);CHKERRQ(ierr); 952 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO n: %d \n", mat_mkl_pardiso->n);CHKERRQ(ierr); 953 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO nrhs: %d \n", mat_mkl_pardiso->nrhs);CHKERRQ(ierr); 954 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO msglvl: %d \n", mat_mkl_pardiso->msglvl);CHKERRQ(ierr); 955 } 956 } 957 PetscFunctionReturn(0); 958 } 959 960 961 PetscErrorCode MatGetInfo_MKL_PARDISO(Mat A, MatInfoType flag, MatInfo *info) 962 { 963 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)A->data; 964 965 PetscFunctionBegin; 966 info->block_size = 1.0; 967 info->nz_used = mat_mkl_pardiso->iparm[17]; 968 info->nz_allocated = mat_mkl_pardiso->iparm[17]; 969 info->nz_unneeded = 0.0; 970 info->assemblies = 0.0; 971 info->mallocs = 0.0; 972 info->memory = 0.0; 973 info->fill_ratio_given = 0; 974 info->fill_ratio_needed = 0; 975 info->factor_mallocs = 0; 976 PetscFunctionReturn(0); 977 } 978 979 PetscErrorCode MatMkl_PardisoSetCntl_MKL_PARDISO(Mat F,PetscInt icntl,PetscInt ival) 980 { 981 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)F->data; 982 983 PetscFunctionBegin; 984 if (icntl <= 64) { 985 mat_mkl_pardiso->iparm[icntl - 1] = ival; 986 } else { 987 if (icntl == 65) PetscSetMKL_PARDISOThreads(ival); 988 else if (icntl == 66) mat_mkl_pardiso->maxfct = ival; 989 else if (icntl == 67) mat_mkl_pardiso->mnum = ival; 990 else if (icntl == 68) mat_mkl_pardiso->msglvl = ival; 991 else if (icntl == 69) { 992 void *pt[IPARM_SIZE]; 993 mat_mkl_pardiso->mtype = ival; 994 MKL_PARDISO_INIT(pt, &mat_mkl_pardiso->mtype, mat_mkl_pardiso->iparm); 995 #if defined(PETSC_USE_REAL_SINGLE) 996 mat_mkl_pardiso->iparm[27] = 1; 997 #else 998 mat_mkl_pardiso->iparm[27] = 0; 999 #endif 1000 mat_mkl_pardiso->iparm[34] = 1; 1001 } else if (icntl==70) mat_mkl_pardiso->solve_interior = (PetscBool)!!ival; 1002 } 1003 PetscFunctionReturn(0); 1004 } 1005 1006 /*@ 1007 MatMkl_PardisoSetCntl - Set Mkl_Pardiso parameters 1008 1009 Logically Collective on Mat 1010 1011 Input Parameters: 1012 + F - the factored matrix obtained by calling MatGetFactor() 1013 . icntl - index of Mkl_Pardiso parameter 1014 - ival - value of Mkl_Pardiso parameter 1015 1016 Options Database: 1017 . -mat_mkl_pardiso_<icntl> <ival> 1018 1019 Level: beginner 1020 1021 References: 1022 . Mkl_Pardiso Users' Guide 1023 1024 .seealso: MatGetFactor() 1025 @*/ 1026 PetscErrorCode MatMkl_PardisoSetCntl(Mat F,PetscInt icntl,PetscInt ival) 1027 { 1028 PetscErrorCode ierr; 1029 1030 PetscFunctionBegin; 1031 ierr = PetscTryMethod(F,"MatMkl_PardisoSetCntl_C",(Mat,PetscInt,PetscInt),(F,icntl,ival));CHKERRQ(ierr); 1032 PetscFunctionReturn(0); 1033 } 1034 1035 /*MC 1036 MATSOLVERMKL_PARDISO - A matrix type providing direct solvers (LU) for 1037 sequential matrices via the external package MKL_PARDISO. 1038 1039 Works with MATSEQAIJ matrices 1040 1041 Use -pc_type lu -pc_factor_mat_solver_type mkl_pardiso to use this direct solver 1042 1043 Options Database Keys: 1044 + -mat_mkl_pardiso_65 - Number of threads to use within MKL_PARDISO 1045 . -mat_mkl_pardiso_66 - Maximum number of factors with identical sparsity structure that must be kept in memory at the same time 1046 . -mat_mkl_pardiso_67 - Indicates the actual matrix for the solution phase 1047 . -mat_mkl_pardiso_68 - Message level information 1048 . -mat_mkl_pardiso_69 - Defines the matrix type. IMPORTANT: When you set this flag, iparm parameters are going to be set to the default ones for the matrix type 1049 . -mat_mkl_pardiso_1 - Use default values 1050 . -mat_mkl_pardiso_2 - Fill-in reducing ordering for the input matrix 1051 . -mat_mkl_pardiso_4 - Preconditioned CGS/CG 1052 . -mat_mkl_pardiso_5 - User permutation 1053 . -mat_mkl_pardiso_6 - Write solution on x 1054 . -mat_mkl_pardiso_8 - Iterative refinement step 1055 . -mat_mkl_pardiso_10 - Pivoting perturbation 1056 . -mat_mkl_pardiso_11 - Scaling vectors 1057 . -mat_mkl_pardiso_12 - Solve with transposed or conjugate transposed matrix A 1058 . -mat_mkl_pardiso_13 - Improved accuracy using (non-) symmetric weighted matching 1059 . -mat_mkl_pardiso_18 - Numbers of non-zero elements 1060 . -mat_mkl_pardiso_19 - Report number of floating point operations 1061 . -mat_mkl_pardiso_21 - Pivoting for symmetric indefinite matrices 1062 . -mat_mkl_pardiso_24 - Parallel factorization control 1063 . -mat_mkl_pardiso_25 - Parallel forward/backward solve control 1064 . -mat_mkl_pardiso_27 - Matrix checker 1065 . -mat_mkl_pardiso_31 - Partial solve and computing selected components of the solution vectors 1066 . -mat_mkl_pardiso_34 - Optimal number of threads for conditional numerical reproducibility (CNR) mode 1067 - -mat_mkl_pardiso_60 - Intel MKL_PARDISO mode 1068 1069 Level: beginner 1070 1071 For more information please check mkl_pardiso manual 1072 1073 .seealso: PCFactorSetMatSolverType(), MatSolverType 1074 1075 M*/ 1076 static PetscErrorCode MatFactorGetSolverType_mkl_pardiso(Mat A, MatSolverType *type) 1077 { 1078 PetscFunctionBegin; 1079 *type = MATSOLVERMKL_PARDISO; 1080 PetscFunctionReturn(0); 1081 } 1082 1083 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_mkl_pardiso(Mat A,MatFactorType ftype,Mat *F) 1084 { 1085 Mat B; 1086 PetscErrorCode ierr; 1087 Mat_MKL_PARDISO *mat_mkl_pardiso; 1088 PetscBool isSeqAIJ,isSeqBAIJ,isSeqSBAIJ; 1089 1090 PetscFunctionBegin; 1091 ierr = PetscObjectBaseTypeCompare((PetscObject)A,MATSEQAIJ,&isSeqAIJ);CHKERRQ(ierr); 1092 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQBAIJ,&isSeqBAIJ);CHKERRQ(ierr); 1093 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQSBAIJ,&isSeqSBAIJ);CHKERRQ(ierr); 1094 ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr); 1095 ierr = MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N);CHKERRQ(ierr); 1096 ierr = PetscStrallocpy("mkl_pardiso",&((PetscObject)B)->type_name);CHKERRQ(ierr); 1097 ierr = MatSetUp(B);CHKERRQ(ierr); 1098 1099 ierr = PetscNewLog(B,&mat_mkl_pardiso);CHKERRQ(ierr); 1100 B->data = mat_mkl_pardiso; 1101 1102 ierr = MatFactorMKL_PARDISOInitialize_Private(A, ftype, mat_mkl_pardiso);CHKERRQ(ierr); 1103 if (ftype == MAT_FACTOR_LU) { 1104 B->ops->lufactorsymbolic = MatLUFactorSymbolic_AIJMKL_PARDISO; 1105 B->factortype = MAT_FACTOR_LU; 1106 mat_mkl_pardiso->needsym = PETSC_FALSE; 1107 if (isSeqAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqaij; 1108 else if (isSeqBAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqbaij; 1109 else if (isSeqSBAIJ) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO LU factor with SEQSBAIJ format! Use MAT_FACTOR_CHOLESKY instead"); 1110 else SETERRQ1(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO LU with %s format",((PetscObject)A)->type_name); 1111 #if defined(PETSC_USE_COMPLEX) 1112 mat_mkl_pardiso->mtype = 13; 1113 #else 1114 mat_mkl_pardiso->mtype = 11; 1115 #endif 1116 } else { 1117 #if defined(PETSC_USE_COMPLEX) 1118 SETERRQ1(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO CHOLESKY with complex scalars! Use MAT_FACTOR_LU instead",((PetscObject)A)->type_name); 1119 #endif 1120 B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_AIJMKL_PARDISO; 1121 B->factortype = MAT_FACTOR_CHOLESKY; 1122 if (isSeqAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqaij; 1123 else if (isSeqBAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqbaij; 1124 else if (isSeqSBAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqsbaij; 1125 else SETERRQ1(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO CHOLESKY with %s format",((PetscObject)A)->type_name); 1126 1127 mat_mkl_pardiso->needsym = PETSC_TRUE; 1128 if (A->spd_set && A->spd) mat_mkl_pardiso->mtype = 2; 1129 else mat_mkl_pardiso->mtype = -2; 1130 } 1131 B->ops->destroy = MatDestroy_MKL_PARDISO; 1132 B->ops->view = MatView_MKL_PARDISO; 1133 B->ops->getinfo = MatGetInfo_MKL_PARDISO; 1134 B->factortype = ftype; 1135 B->assembled = PETSC_TRUE; 1136 1137 ierr = PetscFree(B->solvertype);CHKERRQ(ierr); 1138 ierr = PetscStrallocpy(MATSOLVERMKL_PARDISO,&B->solvertype);CHKERRQ(ierr); 1139 1140 ierr = PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverType_C",MatFactorGetSolverType_mkl_pardiso);CHKERRQ(ierr); 1141 ierr = PetscObjectComposeFunction((PetscObject)B,"MatFactorSetSchurIS_C",MatFactorSetSchurIS_MKL_PARDISO);CHKERRQ(ierr); 1142 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMkl_PardisoSetCntl_C",MatMkl_PardisoSetCntl_MKL_PARDISO);CHKERRQ(ierr); 1143 1144 *F = B; 1145 PetscFunctionReturn(0); 1146 } 1147 1148 PETSC_EXTERN PetscErrorCode MatSolverTypeRegister_MKL_Pardiso(void) 1149 { 1150 PetscErrorCode ierr; 1151 1152 PetscFunctionBegin; 1153 ierr = MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQAIJ,MAT_FACTOR_LU,MatGetFactor_aij_mkl_pardiso);CHKERRQ(ierr); 1154 ierr = MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_aij_mkl_pardiso);CHKERRQ(ierr); 1155 ierr = MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQBAIJ,MAT_FACTOR_LU,MatGetFactor_aij_mkl_pardiso);CHKERRQ(ierr); 1156 ierr = MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQSBAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_aij_mkl_pardiso);CHKERRQ(ierr); 1157 PetscFunctionReturn(0); 1158 } 1159 1160