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) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_PLIB,"This should not happen"); 114 *v = aa->a; 115 if (bs == 1) { /* already in the correct format */ 116 /* though PetscInt and INT_TYPE are of the same size since they are defined differently the Intel compiler requires a cast */ 117 *r = (INT_TYPE*)aa->i; 118 *c = (INT_TYPE*)aa->j; 119 *nnz = (INT_TYPE)aa->nz; 120 *free = PETSC_FALSE; 121 } else if (reuse == MAT_INITIAL_MATRIX) { 122 PetscInt m = A->rmap->n,nz = aa->nz; 123 PetscInt *row,*col; 124 ierr = PetscMalloc2(m+1,&row,nz,&col);CHKERRQ(ierr); 125 for (i=0; i<m+1; i++) { 126 row[i] = aa->i[i]+1; 127 } 128 for (i=0; i<nz; i++) { 129 col[i] = aa->j[i]+1; 130 } 131 *r = (INT_TYPE*)row; 132 *c = (INT_TYPE*)col; 133 *nnz = (INT_TYPE)nz; 134 *free = PETSC_TRUE; 135 } 136 PetscFunctionReturn(0); 137 } 138 139 PetscErrorCode MatMKLPardiso_Convert_seqbaij(Mat A,PetscBool sym,MatReuse reuse,PetscBool *free,INT_TYPE *nnz,INT_TYPE **r,INT_TYPE **c,PetscScalar **v) 140 { 141 Mat_SeqBAIJ *aa = (Mat_SeqBAIJ*)A->data; 142 PetscInt bs = A->rmap->bs,i; 143 PetscErrorCode ierr; 144 145 PetscFunctionBegin; 146 if (!sym) { 147 *v = aa->a; 148 if (bs == 1) { /* already in the correct format */ 149 /* though PetscInt and INT_TYPE are of the same size since they are defined differently the Intel compiler requires a cast */ 150 *r = (INT_TYPE*)aa->i; 151 *c = (INT_TYPE*)aa->j; 152 *nnz = (INT_TYPE)aa->nz; 153 *free = PETSC_FALSE; 154 PetscFunctionReturn(0); 155 } else if (reuse == MAT_INITIAL_MATRIX) { 156 PetscInt m = A->rmap->n,nz = aa->nz; 157 PetscInt *row,*col; 158 ierr = PetscMalloc2(m+1,&row,nz,&col);CHKERRQ(ierr); 159 for (i=0; i<m+1; i++) { 160 row[i] = aa->i[i]+1; 161 } 162 for (i=0; i<nz; i++) { 163 col[i] = aa->j[i]+1; 164 } 165 *r = (INT_TYPE*)row; 166 *c = (INT_TYPE*)col; 167 *nnz = (INT_TYPE)nz; 168 } 169 *free = PETSC_TRUE; 170 } else { 171 SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_PLIB,"This should not happen"); 172 } 173 PetscFunctionReturn(0); 174 } 175 176 PetscErrorCode MatMKLPardiso_Convert_seqaij(Mat A,PetscBool sym,MatReuse reuse,PetscBool *free,INT_TYPE *nnz,INT_TYPE **r,INT_TYPE **c,PetscScalar **v) 177 { 178 Mat_SeqAIJ *aa = (Mat_SeqAIJ*)A->data; 179 PetscScalar *aav; 180 PetscErrorCode ierr; 181 182 PetscFunctionBegin; 183 ierr = MatSeqAIJGetArrayRead(A,(const PetscScalar**)&aav);CHKERRQ(ierr); 184 if (!sym) { /* already in the correct format */ 185 *v = aav; 186 *r = (INT_TYPE*)aa->i; 187 *c = (INT_TYPE*)aa->j; 188 *nnz = (INT_TYPE)aa->nz; 189 *free = PETSC_FALSE; 190 } else if (reuse == MAT_INITIAL_MATRIX) { /* need to get the triangular part */ 191 PetscScalar *vals,*vv; 192 PetscInt *row,*col,*jj; 193 PetscInt m = A->rmap->n,nz,i; 194 195 nz = 0; 196 for (i=0; i<m; i++) nz += aa->i[i+1] - aa->diag[i]; 197 ierr = PetscMalloc2(m+1,&row,nz,&col);CHKERRQ(ierr); 198 ierr = PetscMalloc1(nz,&vals);CHKERRQ(ierr); 199 jj = col; 200 vv = vals; 201 202 row[0] = 0; 203 for (i=0; i<m; i++) { 204 PetscInt *aj = aa->j + aa->diag[i]; 205 PetscScalar *av = aav + aa->diag[i]; 206 PetscInt rl = aa->i[i+1] - aa->diag[i],j; 207 208 for (j=0; j<rl; j++) { 209 *jj = *aj; jj++; aj++; 210 *vv = *av; vv++; av++; 211 } 212 row[i+1] = row[i] + rl; 213 } 214 *v = vals; 215 *r = (INT_TYPE*)row; 216 *c = (INT_TYPE*)col; 217 *nnz = (INT_TYPE)nz; 218 *free = PETSC_TRUE; 219 } else { 220 PetscScalar *vv; 221 PetscInt m = A->rmap->n,i; 222 223 vv = *v; 224 for (i=0; i<m; i++) { 225 PetscScalar *av = aav + aa->diag[i]; 226 PetscInt rl = aa->i[i+1] - aa->diag[i],j; 227 for (j=0; j<rl; j++) { 228 *vv = *av; vv++; av++; 229 } 230 } 231 *free = PETSC_TRUE; 232 } 233 ierr = MatSeqAIJRestoreArrayRead(A,(const PetscScalar**)&aav);CHKERRQ(ierr); 234 PetscFunctionReturn(0); 235 } 236 237 238 static PetscErrorCode MatMKLPardisoSolveSchur_Private(Mat F, PetscScalar *B, PetscScalar *X) 239 { 240 Mat_MKL_PARDISO *mpardiso = (Mat_MKL_PARDISO*)F->data; 241 Mat S,Xmat,Bmat; 242 MatFactorSchurStatus schurstatus; 243 PetscErrorCode ierr; 244 245 PetscFunctionBegin; 246 ierr = MatFactorFactorizeSchurComplement(F);CHKERRQ(ierr); 247 ierr = MatFactorGetSchurComplement(F,&S,&schurstatus);CHKERRQ(ierr); 248 if (X == B && schurstatus == MAT_FACTOR_SCHUR_INVERTED) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"X and B cannot point to the same address"); 249 ierr = MatCreateSeqDense(PETSC_COMM_SELF,mpardiso->schur_size,mpardiso->nrhs,B,&Bmat);CHKERRQ(ierr); 250 ierr = MatCreateSeqDense(PETSC_COMM_SELF,mpardiso->schur_size,mpardiso->nrhs,X,&Xmat);CHKERRQ(ierr); 251 ierr = MatSetType(Bmat,((PetscObject)S)->type_name);CHKERRQ(ierr); 252 ierr = MatSetType(Xmat,((PetscObject)S)->type_name);CHKERRQ(ierr); 253 #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 254 ierr = MatBindToCPU(Xmat,S->boundtocpu);CHKERRQ(ierr); 255 ierr = MatBindToCPU(Bmat,S->boundtocpu);CHKERRQ(ierr); 256 #endif 257 258 #if defined(PETSC_USE_COMPLEX) 259 if (mpardiso->iparm[12-1] == 1) SETERRQ(PetscObjectComm((PetscObject)F),PETSC_ERR_SUP,"Hermitian solve not implemented yet"); 260 #endif 261 262 switch (schurstatus) { 263 case MAT_FACTOR_SCHUR_FACTORED: 264 if (!mpardiso->iparm[12-1]) { 265 ierr = MatMatSolve(S,Bmat,Xmat);CHKERRQ(ierr); 266 } else { /* transpose solve */ 267 ierr = MatMatSolveTranspose(S,Bmat,Xmat);CHKERRQ(ierr); 268 } 269 break; 270 case MAT_FACTOR_SCHUR_INVERTED: 271 ierr = MatProductCreateWithMat(S,Bmat,NULL,Xmat);CHKERRQ(ierr); 272 if (!mpardiso->iparm[12-1]) { 273 ierr = MatProductSetType(Xmat,MATPRODUCT_AB);CHKERRQ(ierr); 274 } else { /* transpose solve */ 275 ierr = MatProductSetType(Xmat,MATPRODUCT_AtB);CHKERRQ(ierr); 276 } 277 ierr = MatProductSetFromOptions(Xmat);CHKERRQ(ierr); 278 ierr = MatProductNumeric(Xmat);CHKERRQ(ierr); 279 ierr = MatProductClear(Xmat);CHKERRQ(ierr); 280 break; 281 default: 282 SETERRQ1(PetscObjectComm((PetscObject)F),PETSC_ERR_SUP,"Unhandled MatFactorSchurStatus %D",F->schur_status); 283 break; 284 } 285 ierr = MatFactorRestoreSchurComplement(F,&S,schurstatus);CHKERRQ(ierr); 286 ierr = MatDestroy(&Bmat);CHKERRQ(ierr); 287 ierr = MatDestroy(&Xmat);CHKERRQ(ierr); 288 PetscFunctionReturn(0); 289 } 290 291 PetscErrorCode MatFactorSetSchurIS_MKL_PARDISO(Mat F, IS is) 292 { 293 Mat_MKL_PARDISO *mpardiso = (Mat_MKL_PARDISO*)F->data; 294 const PetscScalar *arr; 295 const PetscInt *idxs; 296 PetscInt size,i; 297 PetscMPIInt csize; 298 PetscBool sorted; 299 PetscErrorCode ierr; 300 301 PetscFunctionBegin; 302 ierr = MPI_Comm_size(PetscObjectComm((PetscObject)F),&csize);CHKERRQ(ierr); 303 if (csize > 1) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MKL_PARDISO parallel Schur complements not yet supported from PETSc"); 304 ierr = ISSorted(is,&sorted);CHKERRQ(ierr); 305 if (!sorted) { 306 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"IS for MKL_PARDISO Schur complements needs to be sorted"); 307 } 308 ierr = ISGetLocalSize(is,&size);CHKERRQ(ierr); 309 ierr = PetscFree(mpardiso->schur_work);CHKERRQ(ierr); 310 ierr = PetscBLASIntCast(PetscMax(mpardiso->n,2*size),&mpardiso->schur_work_size);CHKERRQ(ierr); 311 ierr = PetscMalloc1(mpardiso->schur_work_size,&mpardiso->schur_work);CHKERRQ(ierr); 312 ierr = MatDestroy(&F->schur);CHKERRQ(ierr); 313 ierr = MatCreateSeqDense(PETSC_COMM_SELF,size,size,NULL,&F->schur);CHKERRQ(ierr); 314 ierr = MatDenseGetArrayRead(F->schur,&arr);CHKERRQ(ierr); 315 mpardiso->schur = (PetscScalar*)arr; 316 mpardiso->schur_size = size; 317 ierr = MatDenseRestoreArrayRead(F->schur,&arr);CHKERRQ(ierr); 318 if (mpardiso->mtype == 2) { 319 ierr = MatSetOption(F->schur,MAT_SPD,PETSC_TRUE);CHKERRQ(ierr); 320 } 321 322 ierr = PetscFree(mpardiso->schur_idxs);CHKERRQ(ierr); 323 ierr = PetscMalloc1(size,&mpardiso->schur_idxs);CHKERRQ(ierr); 324 ierr = PetscArrayzero(mpardiso->perm,mpardiso->n);CHKERRQ(ierr); 325 ierr = ISGetIndices(is,&idxs);CHKERRQ(ierr); 326 ierr = PetscArraycpy(mpardiso->schur_idxs,idxs,size);CHKERRQ(ierr); 327 for (i=0;i<size;i++) mpardiso->perm[idxs[i]] = 1; 328 ierr = ISRestoreIndices(is,&idxs);CHKERRQ(ierr); 329 if (size) { /* turn on Schur switch if the set of indices is not empty */ 330 mpardiso->iparm[36-1] = 2; 331 } 332 PetscFunctionReturn(0); 333 } 334 335 PetscErrorCode MatDestroy_MKL_PARDISO(Mat A) 336 { 337 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 338 PetscErrorCode ierr; 339 340 PetscFunctionBegin; 341 if (mat_mkl_pardiso->CleanUp) { 342 mat_mkl_pardiso->phase = JOB_RELEASE_OF_ALL_MEMORY; 343 344 MKL_PARDISO (mat_mkl_pardiso->pt, 345 &mat_mkl_pardiso->maxfct, 346 &mat_mkl_pardiso->mnum, 347 &mat_mkl_pardiso->mtype, 348 &mat_mkl_pardiso->phase, 349 &mat_mkl_pardiso->n, 350 NULL, 351 NULL, 352 NULL, 353 NULL, 354 &mat_mkl_pardiso->nrhs, 355 mat_mkl_pardiso->iparm, 356 &mat_mkl_pardiso->msglvl, 357 NULL, 358 NULL, 359 &mat_mkl_pardiso->err); 360 } 361 ierr = PetscFree(mat_mkl_pardiso->perm);CHKERRQ(ierr); 362 ierr = PetscFree(mat_mkl_pardiso->schur_work);CHKERRQ(ierr); 363 ierr = PetscFree(mat_mkl_pardiso->schur_idxs);CHKERRQ(ierr); 364 if (mat_mkl_pardiso->freeaij) { 365 ierr = PetscFree2(mat_mkl_pardiso->ia,mat_mkl_pardiso->ja);CHKERRQ(ierr); 366 if (mat_mkl_pardiso->iparm[34] == 1) { 367 ierr = PetscFree(mat_mkl_pardiso->a);CHKERRQ(ierr); 368 } 369 } 370 ierr = PetscFree(A->data);CHKERRQ(ierr); 371 372 /* clear composed functions */ 373 ierr = PetscObjectComposeFunction((PetscObject)A,"MatFactorGetSolverType_C",NULL);CHKERRQ(ierr); 374 ierr = PetscObjectComposeFunction((PetscObject)A,"MatFactorSetSchurIS_C",NULL);CHKERRQ(ierr); 375 ierr = PetscObjectComposeFunction((PetscObject)A,"MatMkl_PardisoSetCntl_C",NULL);CHKERRQ(ierr); 376 PetscFunctionReturn(0); 377 } 378 379 static PetscErrorCode MatMKLPardisoScatterSchur_Private(Mat_MKL_PARDISO *mpardiso, PetscScalar *whole, PetscScalar *schur, PetscBool reduce) 380 { 381 PetscFunctionBegin; 382 if (reduce) { /* data given for the whole matrix */ 383 PetscInt i,m=0,p=0; 384 for (i=0;i<mpardiso->nrhs;i++) { 385 PetscInt j; 386 for (j=0;j<mpardiso->schur_size;j++) { 387 schur[p+j] = whole[m+mpardiso->schur_idxs[j]]; 388 } 389 m += mpardiso->n; 390 p += mpardiso->schur_size; 391 } 392 } else { /* from Schur to whole */ 393 PetscInt i,m=0,p=0; 394 for (i=0;i<mpardiso->nrhs;i++) { 395 PetscInt j; 396 for (j=0;j<mpardiso->schur_size;j++) { 397 whole[m+mpardiso->schur_idxs[j]] = schur[p+j]; 398 } 399 m += mpardiso->n; 400 p += mpardiso->schur_size; 401 } 402 } 403 PetscFunctionReturn(0); 404 } 405 406 PetscErrorCode MatSolve_MKL_PARDISO(Mat A,Vec b,Vec x) 407 { 408 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 409 PetscErrorCode ierr; 410 PetscScalar *xarray; 411 const PetscScalar *barray; 412 413 PetscFunctionBegin; 414 mat_mkl_pardiso->nrhs = 1; 415 ierr = VecGetArray(x,&xarray);CHKERRQ(ierr); 416 ierr = VecGetArrayRead(b,&barray);CHKERRQ(ierr); 417 418 if (!mat_mkl_pardiso->schur) mat_mkl_pardiso->phase = JOB_SOLVE_ITERATIVE_REFINEMENT; 419 else mat_mkl_pardiso->phase = JOB_SOLVE_FORWARD_SUBSTITUTION; 420 421 if (barray == xarray) { /* if the two vectors share the same memory */ 422 PetscScalar *work; 423 if (!mat_mkl_pardiso->schur_work) { 424 ierr = PetscMalloc1(mat_mkl_pardiso->n,&work);CHKERRQ(ierr); 425 } else { 426 work = mat_mkl_pardiso->schur_work; 427 } 428 mat_mkl_pardiso->iparm[6-1] = 1; 429 MKL_PARDISO (mat_mkl_pardiso->pt, 430 &mat_mkl_pardiso->maxfct, 431 &mat_mkl_pardiso->mnum, 432 &mat_mkl_pardiso->mtype, 433 &mat_mkl_pardiso->phase, 434 &mat_mkl_pardiso->n, 435 mat_mkl_pardiso->a, 436 mat_mkl_pardiso->ia, 437 mat_mkl_pardiso->ja, 438 NULL, 439 &mat_mkl_pardiso->nrhs, 440 mat_mkl_pardiso->iparm, 441 &mat_mkl_pardiso->msglvl, 442 (void*)xarray, 443 (void*)work, 444 &mat_mkl_pardiso->err); 445 if (!mat_mkl_pardiso->schur_work) { 446 ierr = PetscFree(work);CHKERRQ(ierr); 447 } 448 } else { 449 mat_mkl_pardiso->iparm[6-1] = 0; 450 MKL_PARDISO (mat_mkl_pardiso->pt, 451 &mat_mkl_pardiso->maxfct, 452 &mat_mkl_pardiso->mnum, 453 &mat_mkl_pardiso->mtype, 454 &mat_mkl_pardiso->phase, 455 &mat_mkl_pardiso->n, 456 mat_mkl_pardiso->a, 457 mat_mkl_pardiso->ia, 458 mat_mkl_pardiso->ja, 459 mat_mkl_pardiso->perm, 460 &mat_mkl_pardiso->nrhs, 461 mat_mkl_pardiso->iparm, 462 &mat_mkl_pardiso->msglvl, 463 (void*)barray, 464 (void*)xarray, 465 &mat_mkl_pardiso->err); 466 } 467 ierr = VecRestoreArrayRead(b,&barray);CHKERRQ(ierr); 468 469 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); 470 471 if (mat_mkl_pardiso->schur) { /* solve Schur complement and expand solution */ 472 PetscInt shift = mat_mkl_pardiso->schur_size; 473 474 /* if inverted, uses BLAS *MM subroutines, otherwise LAPACK *TRS */ 475 if (A->schur_status != MAT_FACTOR_SCHUR_INVERTED) shift = 0; 476 477 if (!mat_mkl_pardiso->solve_interior) { 478 /* solve Schur complement */ 479 ierr = MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work,PETSC_TRUE);CHKERRQ(ierr); 480 ierr = MatMKLPardisoSolveSchur_Private(A,mat_mkl_pardiso->schur_work,mat_mkl_pardiso->schur_work+shift);CHKERRQ(ierr); 481 ierr = MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work+shift,PETSC_FALSE);CHKERRQ(ierr); 482 } else { /* if we are solving for the interior problem, any value in barray[schur] forward-substituted to xarray[schur] will be neglected */ 483 PetscInt i; 484 for (i=0;i<mat_mkl_pardiso->schur_size;i++) { 485 xarray[mat_mkl_pardiso->schur_idxs[i]] = 0.; 486 } 487 } 488 489 /* expansion phase */ 490 mat_mkl_pardiso->iparm[6-1] = 1; 491 mat_mkl_pardiso->phase = JOB_SOLVE_BACKWARD_SUBSTITUTION; 492 MKL_PARDISO (mat_mkl_pardiso->pt, 493 &mat_mkl_pardiso->maxfct, 494 &mat_mkl_pardiso->mnum, 495 &mat_mkl_pardiso->mtype, 496 &mat_mkl_pardiso->phase, 497 &mat_mkl_pardiso->n, 498 mat_mkl_pardiso->a, 499 mat_mkl_pardiso->ia, 500 mat_mkl_pardiso->ja, 501 mat_mkl_pardiso->perm, 502 &mat_mkl_pardiso->nrhs, 503 mat_mkl_pardiso->iparm, 504 &mat_mkl_pardiso->msglvl, 505 (void*)xarray, 506 (void*)mat_mkl_pardiso->schur_work, /* according to the specs, the solution vector is always used */ 507 &mat_mkl_pardiso->err); 508 509 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); 510 mat_mkl_pardiso->iparm[6-1] = 0; 511 } 512 ierr = VecRestoreArray(x,&xarray);CHKERRQ(ierr); 513 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 514 PetscFunctionReturn(0); 515 } 516 517 PetscErrorCode MatSolveTranspose_MKL_PARDISO(Mat A,Vec b,Vec x) 518 { 519 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 520 PetscInt oiparm12; 521 PetscErrorCode ierr; 522 523 PetscFunctionBegin; 524 oiparm12 = mat_mkl_pardiso->iparm[12 - 1]; 525 mat_mkl_pardiso->iparm[12 - 1] = 2; 526 ierr = MatSolve_MKL_PARDISO(A,b,x);CHKERRQ(ierr); 527 mat_mkl_pardiso->iparm[12 - 1] = oiparm12; 528 PetscFunctionReturn(0); 529 } 530 531 PetscErrorCode MatMatSolve_MKL_PARDISO(Mat A,Mat B,Mat X) 532 { 533 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)(A)->data; 534 PetscErrorCode ierr; 535 const PetscScalar *barray; 536 PetscScalar *xarray; 537 PetscBool flg; 538 539 PetscFunctionBegin; 540 ierr = PetscObjectBaseTypeCompare((PetscObject)B,MATSEQDENSE,&flg);CHKERRQ(ierr); 541 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix B must be MATSEQDENSE matrix"); 542 if (X != B) { 543 ierr = PetscObjectBaseTypeCompare((PetscObject)X,MATSEQDENSE,&flg);CHKERRQ(ierr); 544 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix X must be MATSEQDENSE matrix"); 545 } 546 547 ierr = MatGetSize(B,NULL,(PetscInt*)&mat_mkl_pardiso->nrhs);CHKERRQ(ierr); 548 549 if (mat_mkl_pardiso->nrhs > 0) { 550 ierr = MatDenseGetArrayRead(B,&barray);CHKERRQ(ierr); 551 ierr = MatDenseGetArray(X,&xarray);CHKERRQ(ierr); 552 553 if (barray == xarray) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"B and X cannot share the same memory location"); 554 if (!mat_mkl_pardiso->schur) mat_mkl_pardiso->phase = JOB_SOLVE_ITERATIVE_REFINEMENT; 555 else mat_mkl_pardiso->phase = JOB_SOLVE_FORWARD_SUBSTITUTION; 556 557 MKL_PARDISO (mat_mkl_pardiso->pt, 558 &mat_mkl_pardiso->maxfct, 559 &mat_mkl_pardiso->mnum, 560 &mat_mkl_pardiso->mtype, 561 &mat_mkl_pardiso->phase, 562 &mat_mkl_pardiso->n, 563 mat_mkl_pardiso->a, 564 mat_mkl_pardiso->ia, 565 mat_mkl_pardiso->ja, 566 mat_mkl_pardiso->perm, 567 &mat_mkl_pardiso->nrhs, 568 mat_mkl_pardiso->iparm, 569 &mat_mkl_pardiso->msglvl, 570 (void*)barray, 571 (void*)xarray, 572 &mat_mkl_pardiso->err); 573 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); 574 575 ierr = MatDenseRestoreArrayRead(B,&barray);CHKERRQ(ierr); 576 if (mat_mkl_pardiso->schur) { /* solve Schur complement and expand solution */ 577 PetscScalar *o_schur_work = NULL; 578 PetscInt shift = mat_mkl_pardiso->schur_size*mat_mkl_pardiso->nrhs,scale; 579 PetscInt mem = mat_mkl_pardiso->n*mat_mkl_pardiso->nrhs; 580 581 /* allocate extra memory if it is needed */ 582 scale = 1; 583 if (A->schur_status == MAT_FACTOR_SCHUR_INVERTED) scale = 2; 584 585 mem *= scale; 586 if (mem > mat_mkl_pardiso->schur_work_size) { 587 o_schur_work = mat_mkl_pardiso->schur_work; 588 ierr = PetscMalloc1(mem,&mat_mkl_pardiso->schur_work);CHKERRQ(ierr); 589 } 590 591 /* if inverted, uses BLAS *MM subroutines, otherwise LAPACK *TRS */ 592 if (A->schur_status != MAT_FACTOR_SCHUR_INVERTED) shift = 0; 593 594 /* solve Schur complement */ 595 if (!mat_mkl_pardiso->solve_interior) { 596 ierr = MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work,PETSC_TRUE);CHKERRQ(ierr); 597 ierr = MatMKLPardisoSolveSchur_Private(A,mat_mkl_pardiso->schur_work,mat_mkl_pardiso->schur_work+shift);CHKERRQ(ierr); 598 ierr = MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work+shift,PETSC_FALSE);CHKERRQ(ierr); 599 } else { /* if we are solving for the interior problem, any value in barray[schur,n] forward-substituted to xarray[schur,n] will be neglected */ 600 PetscInt i,n,m=0; 601 for (n=0;n<mat_mkl_pardiso->nrhs;n++) { 602 for (i=0;i<mat_mkl_pardiso->schur_size;i++) { 603 xarray[mat_mkl_pardiso->schur_idxs[i]+m] = 0.; 604 } 605 m += mat_mkl_pardiso->n; 606 } 607 } 608 609 /* expansion phase */ 610 mat_mkl_pardiso->iparm[6-1] = 1; 611 mat_mkl_pardiso->phase = JOB_SOLVE_BACKWARD_SUBSTITUTION; 612 MKL_PARDISO (mat_mkl_pardiso->pt, 613 &mat_mkl_pardiso->maxfct, 614 &mat_mkl_pardiso->mnum, 615 &mat_mkl_pardiso->mtype, 616 &mat_mkl_pardiso->phase, 617 &mat_mkl_pardiso->n, 618 mat_mkl_pardiso->a, 619 mat_mkl_pardiso->ia, 620 mat_mkl_pardiso->ja, 621 mat_mkl_pardiso->perm, 622 &mat_mkl_pardiso->nrhs, 623 mat_mkl_pardiso->iparm, 624 &mat_mkl_pardiso->msglvl, 625 (void*)xarray, 626 (void*)mat_mkl_pardiso->schur_work, /* according to the specs, the solution vector is always used */ 627 &mat_mkl_pardiso->err); 628 if (o_schur_work) { /* restore original schur_work (minimal size) */ 629 ierr = PetscFree(mat_mkl_pardiso->schur_work);CHKERRQ(ierr); 630 mat_mkl_pardiso->schur_work = o_schur_work; 631 } 632 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); 633 mat_mkl_pardiso->iparm[6-1] = 0; 634 } 635 } 636 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 637 PetscFunctionReturn(0); 638 } 639 640 PetscErrorCode MatFactorNumeric_MKL_PARDISO(Mat F,Mat A,const MatFactorInfo *info) 641 { 642 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)(F)->data; 643 PetscErrorCode ierr; 644 645 PetscFunctionBegin; 646 mat_mkl_pardiso->matstruc = SAME_NONZERO_PATTERN; 647 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); 648 649 mat_mkl_pardiso->phase = JOB_NUMERICAL_FACTORIZATION; 650 MKL_PARDISO (mat_mkl_pardiso->pt, 651 &mat_mkl_pardiso->maxfct, 652 &mat_mkl_pardiso->mnum, 653 &mat_mkl_pardiso->mtype, 654 &mat_mkl_pardiso->phase, 655 &mat_mkl_pardiso->n, 656 mat_mkl_pardiso->a, 657 mat_mkl_pardiso->ia, 658 mat_mkl_pardiso->ja, 659 mat_mkl_pardiso->perm, 660 &mat_mkl_pardiso->nrhs, 661 mat_mkl_pardiso->iparm, 662 &mat_mkl_pardiso->msglvl, 663 NULL, 664 (void*)mat_mkl_pardiso->schur, 665 &mat_mkl_pardiso->err); 666 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); 667 668 /* report flops */ 669 if (mat_mkl_pardiso->iparm[18] > 0) { 670 ierr = PetscLogFlops(PetscPowRealInt(10.,6)*mat_mkl_pardiso->iparm[18]);CHKERRQ(ierr); 671 } 672 673 if (F->schur) { /* schur output from pardiso is in row major format */ 674 #if defined(PETSC_HAVE_CUDA) 675 F->schur->offloadmask = PETSC_OFFLOAD_CPU; 676 #endif 677 ierr = MatFactorRestoreSchurComplement(F,NULL,MAT_FACTOR_SCHUR_UNFACTORED);CHKERRQ(ierr); 678 ierr = MatTranspose(F->schur,MAT_INPLACE_MATRIX,&F->schur);CHKERRQ(ierr); 679 } 680 mat_mkl_pardiso->matstruc = SAME_NONZERO_PATTERN; 681 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 682 PetscFunctionReturn(0); 683 } 684 685 PetscErrorCode PetscSetMKL_PARDISOFromOptions(Mat F, Mat A) 686 { 687 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)F->data; 688 PetscErrorCode ierr; 689 PetscInt icntl,bs,threads=1; 690 PetscBool flg; 691 692 PetscFunctionBegin; 693 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)A),((PetscObject)A)->prefix,"MKL_PARDISO Options","Mat");CHKERRQ(ierr); 694 695 ierr = PetscOptionsInt("-mat_mkl_pardiso_65","Number of threads to use within PARDISO","None",threads,&threads,&flg);CHKERRQ(ierr); 696 if (flg) PetscSetMKL_PARDISOThreads((int)threads); 697 698 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); 699 if (flg) mat_mkl_pardiso->maxfct = icntl; 700 701 ierr = PetscOptionsInt("-mat_mkl_pardiso_67","Indicates the actual matrix for the solution phase","None",mat_mkl_pardiso->mnum,&icntl,&flg);CHKERRQ(ierr); 702 if (flg) mat_mkl_pardiso->mnum = icntl; 703 704 ierr = PetscOptionsInt("-mat_mkl_pardiso_68","Message level information","None",mat_mkl_pardiso->msglvl,&icntl,&flg);CHKERRQ(ierr); 705 if (flg) mat_mkl_pardiso->msglvl = icntl; 706 707 ierr = PetscOptionsInt("-mat_mkl_pardiso_69","Defines the matrix type","None",mat_mkl_pardiso->mtype,&icntl,&flg);CHKERRQ(ierr); 708 if (flg) { 709 void *pt[IPARM_SIZE]; 710 mat_mkl_pardiso->mtype = icntl; 711 icntl = mat_mkl_pardiso->iparm[34]; 712 bs = mat_mkl_pardiso->iparm[36]; 713 MKL_PARDISO_INIT(pt, &mat_mkl_pardiso->mtype, mat_mkl_pardiso->iparm); 714 #if defined(PETSC_USE_REAL_SINGLE) 715 mat_mkl_pardiso->iparm[27] = 1; 716 #else 717 mat_mkl_pardiso->iparm[27] = 0; 718 #endif 719 mat_mkl_pardiso->iparm[34] = icntl; 720 mat_mkl_pardiso->iparm[36] = bs; 721 } 722 ierr = PetscOptionsInt("-mat_mkl_pardiso_1","Use default values (if 0)","None",mat_mkl_pardiso->iparm[0],&icntl,&flg);CHKERRQ(ierr); 723 724 if (flg && icntl != 0) { 725 ierr = PetscOptionsInt("-mat_mkl_pardiso_2","Fill-in reducing ordering for the input matrix","None",mat_mkl_pardiso->iparm[1],&icntl,&flg);CHKERRQ(ierr); 726 if (flg) mat_mkl_pardiso->iparm[1] = icntl; 727 728 ierr = PetscOptionsInt("-mat_mkl_pardiso_4","Preconditioned CGS/CG","None",mat_mkl_pardiso->iparm[3],&icntl,&flg);CHKERRQ(ierr); 729 if (flg) mat_mkl_pardiso->iparm[3] = icntl; 730 731 ierr = PetscOptionsInt("-mat_mkl_pardiso_5","User permutation","None",mat_mkl_pardiso->iparm[4],&icntl,&flg);CHKERRQ(ierr); 732 if (flg) mat_mkl_pardiso->iparm[4] = icntl; 733 734 ierr = PetscOptionsInt("-mat_mkl_pardiso_6","Write solution on x","None",mat_mkl_pardiso->iparm[5],&icntl,&flg);CHKERRQ(ierr); 735 if (flg) mat_mkl_pardiso->iparm[5] = icntl; 736 737 ierr = PetscOptionsInt("-mat_mkl_pardiso_8","Iterative refinement step","None",mat_mkl_pardiso->iparm[7],&icntl,&flg);CHKERRQ(ierr); 738 if (flg) mat_mkl_pardiso->iparm[7] = icntl; 739 740 ierr = PetscOptionsInt("-mat_mkl_pardiso_10","Pivoting perturbation","None",mat_mkl_pardiso->iparm[9],&icntl,&flg);CHKERRQ(ierr); 741 if (flg) mat_mkl_pardiso->iparm[9] = icntl; 742 743 ierr = PetscOptionsInt("-mat_mkl_pardiso_11","Scaling vectors","None",mat_mkl_pardiso->iparm[10],&icntl,&flg);CHKERRQ(ierr); 744 if (flg) mat_mkl_pardiso->iparm[10] = icntl; 745 746 ierr = PetscOptionsInt("-mat_mkl_pardiso_12","Solve with transposed or conjugate transposed matrix A","None",mat_mkl_pardiso->iparm[11],&icntl,&flg);CHKERRQ(ierr); 747 if (flg) mat_mkl_pardiso->iparm[11] = icntl; 748 749 ierr = PetscOptionsInt("-mat_mkl_pardiso_13","Improved accuracy using (non-) symmetric weighted matching","None",mat_mkl_pardiso->iparm[12],&icntl,&flg);CHKERRQ(ierr); 750 if (flg) mat_mkl_pardiso->iparm[12] = icntl; 751 752 ierr = PetscOptionsInt("-mat_mkl_pardiso_18","Numbers of non-zero elements","None",mat_mkl_pardiso->iparm[17],&icntl,&flg);CHKERRQ(ierr); 753 if (flg) mat_mkl_pardiso->iparm[17] = icntl; 754 755 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); 756 if (flg) mat_mkl_pardiso->iparm[18] = icntl; 757 758 ierr = PetscOptionsInt("-mat_mkl_pardiso_21","Pivoting for symmetric indefinite matrices","None",mat_mkl_pardiso->iparm[20],&icntl,&flg);CHKERRQ(ierr); 759 if (flg) mat_mkl_pardiso->iparm[20] = icntl; 760 761 ierr = PetscOptionsInt("-mat_mkl_pardiso_24","Parallel factorization control","None",mat_mkl_pardiso->iparm[23],&icntl,&flg);CHKERRQ(ierr); 762 if (flg) mat_mkl_pardiso->iparm[23] = icntl; 763 764 ierr = PetscOptionsInt("-mat_mkl_pardiso_25","Parallel forward/backward solve control","None",mat_mkl_pardiso->iparm[24],&icntl,&flg);CHKERRQ(ierr); 765 if (flg) mat_mkl_pardiso->iparm[24] = icntl; 766 767 ierr = PetscOptionsInt("-mat_mkl_pardiso_27","Matrix checker","None",mat_mkl_pardiso->iparm[26],&icntl,&flg);CHKERRQ(ierr); 768 if (flg) mat_mkl_pardiso->iparm[26] = icntl; 769 770 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); 771 if (flg) mat_mkl_pardiso->iparm[30] = icntl; 772 773 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); 774 if (flg) mat_mkl_pardiso->iparm[33] = icntl; 775 776 ierr = PetscOptionsInt("-mat_mkl_pardiso_60","Intel MKL_PARDISO mode","None",mat_mkl_pardiso->iparm[59],&icntl,&flg);CHKERRQ(ierr); 777 if (flg) mat_mkl_pardiso->iparm[59] = icntl; 778 } 779 PetscOptionsEnd(); 780 PetscFunctionReturn(0); 781 } 782 783 PetscErrorCode MatFactorMKL_PARDISOInitialize_Private(Mat A, MatFactorType ftype, Mat_MKL_PARDISO *mat_mkl_pardiso) 784 { 785 PetscErrorCode ierr; 786 PetscInt i,bs; 787 PetscBool match; 788 789 PetscFunctionBegin; 790 for (i=0; i<IPARM_SIZE; i++) mat_mkl_pardiso->iparm[i] = 0; 791 for (i=0; i<IPARM_SIZE; i++) mat_mkl_pardiso->pt[i] = 0; 792 #if defined(PETSC_USE_REAL_SINGLE) 793 mat_mkl_pardiso->iparm[27] = 1; 794 #else 795 mat_mkl_pardiso->iparm[27] = 0; 796 #endif 797 /* Default options for both sym and unsym */ 798 mat_mkl_pardiso->iparm[ 0] = 1; /* Solver default parameters overriden with provided by iparm */ 799 mat_mkl_pardiso->iparm[ 1] = 2; /* Metis reordering */ 800 mat_mkl_pardiso->iparm[ 5] = 0; /* Write solution into x */ 801 mat_mkl_pardiso->iparm[ 7] = 0; /* Max number of iterative refinement steps */ 802 mat_mkl_pardiso->iparm[17] = -1; /* Output: Number of nonzeros in the factor LU */ 803 mat_mkl_pardiso->iparm[18] = -1; /* Output: Mflops for LU factorization */ 804 #if 0 805 mat_mkl_pardiso->iparm[23] = 1; /* Parallel factorization control*/ 806 #endif 807 ierr = PetscObjectTypeCompareAny((PetscObject)A,&match,MATSEQBAIJ,MATSEQSBAIJ,"");CHKERRQ(ierr); 808 ierr = MatGetBlockSize(A,&bs);CHKERRQ(ierr); 809 if (!match || bs == 1) { 810 mat_mkl_pardiso->iparm[34] = 1; /* Cluster Sparse Solver use C-style indexing for ia and ja arrays */ 811 mat_mkl_pardiso->n = A->rmap->N; 812 } else { 813 mat_mkl_pardiso->iparm[34] = 0; /* Cluster Sparse Solver use Fortran-style indexing for ia and ja arrays */ 814 mat_mkl_pardiso->iparm[36] = bs; 815 mat_mkl_pardiso->n = A->rmap->N/bs; 816 } 817 mat_mkl_pardiso->iparm[39] = 0; /* Input: matrix/rhs/solution stored on master */ 818 819 mat_mkl_pardiso->CleanUp = PETSC_FALSE; 820 mat_mkl_pardiso->maxfct = 1; /* Maximum number of numerical factorizations. */ 821 mat_mkl_pardiso->mnum = 1; /* Which factorization to use. */ 822 mat_mkl_pardiso->msglvl = 0; /* 0: do not print 1: Print statistical information in file */ 823 mat_mkl_pardiso->phase = -1; 824 mat_mkl_pardiso->err = 0; 825 826 mat_mkl_pardiso->nrhs = 1; 827 mat_mkl_pardiso->err = 0; 828 mat_mkl_pardiso->phase = -1; 829 830 if (ftype == MAT_FACTOR_LU) { 831 mat_mkl_pardiso->iparm[ 9] = 13; /* Perturb the pivot elements with 1E-13 */ 832 mat_mkl_pardiso->iparm[10] = 1; /* Use nonsymmetric permutation and scaling MPS */ 833 mat_mkl_pardiso->iparm[12] = 1; /* Switch on Maximum Weighted Matching algorithm (default for non-symmetric) */ 834 } else { 835 mat_mkl_pardiso->iparm[ 9] = 8; /* Perturb the pivot elements with 1E-8 */ 836 mat_mkl_pardiso->iparm[10] = 0; /* Use nonsymmetric permutation and scaling MPS */ 837 mat_mkl_pardiso->iparm[12] = 1; /* Switch on Maximum Weighted Matching algorithm (default for non-symmetric) */ 838 #if defined(PETSC_USE_DEBUG) 839 mat_mkl_pardiso->iparm[26] = 1; /* Matrix checker */ 840 #endif 841 } 842 ierr = PetscMalloc1(A->rmap->N*sizeof(INT_TYPE), &mat_mkl_pardiso->perm);CHKERRQ(ierr); 843 for (i=0; i<A->rmap->N; i++) { 844 mat_mkl_pardiso->perm[i] = 0; 845 } 846 mat_mkl_pardiso->schur_size = 0; 847 PetscFunctionReturn(0); 848 } 849 850 PetscErrorCode MatFactorSymbolic_AIJMKL_PARDISO_Private(Mat F,Mat A,const MatFactorInfo *info) 851 { 852 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)F->data; 853 PetscErrorCode ierr; 854 855 PetscFunctionBegin; 856 mat_mkl_pardiso->matstruc = DIFFERENT_NONZERO_PATTERN; 857 ierr = PetscSetMKL_PARDISOFromOptions(F,A);CHKERRQ(ierr); 858 859 /* throw away any previously computed structure */ 860 if (mat_mkl_pardiso->freeaij) { 861 ierr = PetscFree2(mat_mkl_pardiso->ia,mat_mkl_pardiso->ja);CHKERRQ(ierr); 862 if (mat_mkl_pardiso->iparm[34] == 1) { 863 ierr = PetscFree(mat_mkl_pardiso->a);CHKERRQ(ierr); 864 } 865 } 866 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); 867 if (mat_mkl_pardiso->iparm[34] == 1) mat_mkl_pardiso->n = A->rmap->N; 868 else mat_mkl_pardiso->n = A->rmap->N/A->rmap->bs; 869 870 mat_mkl_pardiso->phase = JOB_ANALYSIS; 871 872 /* reset flops counting if requested */ 873 if (mat_mkl_pardiso->iparm[18]) mat_mkl_pardiso->iparm[18] = -1; 874 875 MKL_PARDISO (mat_mkl_pardiso->pt, 876 &mat_mkl_pardiso->maxfct, 877 &mat_mkl_pardiso->mnum, 878 &mat_mkl_pardiso->mtype, 879 &mat_mkl_pardiso->phase, 880 &mat_mkl_pardiso->n, 881 mat_mkl_pardiso->a, 882 mat_mkl_pardiso->ia, 883 mat_mkl_pardiso->ja, 884 mat_mkl_pardiso->perm, 885 &mat_mkl_pardiso->nrhs, 886 mat_mkl_pardiso->iparm, 887 &mat_mkl_pardiso->msglvl, 888 NULL, 889 NULL, 890 &mat_mkl_pardiso->err); 891 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); 892 893 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 894 895 if (F->factortype == MAT_FACTOR_LU) F->ops->lufactornumeric = MatFactorNumeric_MKL_PARDISO; 896 else F->ops->choleskyfactornumeric = MatFactorNumeric_MKL_PARDISO; 897 898 F->ops->solve = MatSolve_MKL_PARDISO; 899 F->ops->solvetranspose = MatSolveTranspose_MKL_PARDISO; 900 F->ops->matsolve = MatMatSolve_MKL_PARDISO; 901 PetscFunctionReturn(0); 902 } 903 904 PetscErrorCode MatLUFactorSymbolic_AIJMKL_PARDISO(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info) 905 { 906 PetscErrorCode ierr; 907 908 PetscFunctionBegin; 909 ierr = MatFactorSymbolic_AIJMKL_PARDISO_Private(F, A, info);CHKERRQ(ierr); 910 PetscFunctionReturn(0); 911 } 912 913 #if !defined(PETSC_USE_COMPLEX) 914 PetscErrorCode MatGetInertia_MKL_PARDISO(Mat F,PetscInt *nneg,PetscInt *nzero,PetscInt *npos) 915 { 916 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)F->data; 917 918 PetscFunctionBegin; 919 if (nneg) *nneg = mat_mkl_pardiso->iparm[22]; 920 if (npos) *npos = mat_mkl_pardiso->iparm[21]; 921 if (nzero) *nzero = F->rmap->N - (mat_mkl_pardiso->iparm[22] + mat_mkl_pardiso->iparm[21]); 922 PetscFunctionReturn(0); 923 } 924 #endif 925 926 PetscErrorCode MatCholeskyFactorSymbolic_AIJMKL_PARDISO(Mat F,Mat A,IS r,const MatFactorInfo *info) 927 { 928 PetscErrorCode ierr; 929 930 PetscFunctionBegin; 931 ierr = MatFactorSymbolic_AIJMKL_PARDISO_Private(F, A, info);CHKERRQ(ierr); 932 #if defined(PETSC_USE_COMPLEX) 933 F->ops->getinertia = NULL; 934 #else 935 F->ops->getinertia = MatGetInertia_MKL_PARDISO; 936 #endif 937 PetscFunctionReturn(0); 938 } 939 940 PetscErrorCode MatView_MKL_PARDISO(Mat A, PetscViewer viewer) 941 { 942 PetscErrorCode ierr; 943 PetscBool iascii; 944 PetscViewerFormat format; 945 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 946 PetscInt i; 947 948 PetscFunctionBegin; 949 if (A->ops->solve != MatSolve_MKL_PARDISO) PetscFunctionReturn(0); 950 951 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 952 if (iascii) { 953 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 954 if (format == PETSC_VIEWER_ASCII_INFO) { 955 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO run parameters:\n");CHKERRQ(ierr); 956 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO phase: %d \n",mat_mkl_pardiso->phase);CHKERRQ(ierr); 957 for (i=1; i<=64; i++) { 958 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO iparm[%d]: %d \n",i, mat_mkl_pardiso->iparm[i - 1]);CHKERRQ(ierr); 959 } 960 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO maxfct: %d \n", mat_mkl_pardiso->maxfct);CHKERRQ(ierr); 961 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO mnum: %d \n", mat_mkl_pardiso->mnum);CHKERRQ(ierr); 962 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO mtype: %d \n", mat_mkl_pardiso->mtype);CHKERRQ(ierr); 963 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO n: %d \n", mat_mkl_pardiso->n);CHKERRQ(ierr); 964 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO nrhs: %d \n", mat_mkl_pardiso->nrhs);CHKERRQ(ierr); 965 ierr = PetscViewerASCIIPrintf(viewer,"MKL_PARDISO msglvl: %d \n", mat_mkl_pardiso->msglvl);CHKERRQ(ierr); 966 } 967 } 968 PetscFunctionReturn(0); 969 } 970 971 972 PetscErrorCode MatGetInfo_MKL_PARDISO(Mat A, MatInfoType flag, MatInfo *info) 973 { 974 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)A->data; 975 976 PetscFunctionBegin; 977 info->block_size = 1.0; 978 info->nz_used = mat_mkl_pardiso->iparm[17]; 979 info->nz_allocated = mat_mkl_pardiso->iparm[17]; 980 info->nz_unneeded = 0.0; 981 info->assemblies = 0.0; 982 info->mallocs = 0.0; 983 info->memory = 0.0; 984 info->fill_ratio_given = 0; 985 info->fill_ratio_needed = 0; 986 info->factor_mallocs = 0; 987 PetscFunctionReturn(0); 988 } 989 990 PetscErrorCode MatMkl_PardisoSetCntl_MKL_PARDISO(Mat F,PetscInt icntl,PetscInt ival) 991 { 992 PetscInt backup,bs; 993 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)F->data; 994 995 PetscFunctionBegin; 996 if (icntl <= 64) { 997 mat_mkl_pardiso->iparm[icntl - 1] = ival; 998 } else { 999 if (icntl == 65) PetscSetMKL_PARDISOThreads(ival); 1000 else if (icntl == 66) mat_mkl_pardiso->maxfct = ival; 1001 else if (icntl == 67) mat_mkl_pardiso->mnum = ival; 1002 else if (icntl == 68) mat_mkl_pardiso->msglvl = ival; 1003 else if (icntl == 69) { 1004 void *pt[IPARM_SIZE]; 1005 backup = mat_mkl_pardiso->iparm[34]; 1006 bs = mat_mkl_pardiso->iparm[36]; 1007 mat_mkl_pardiso->mtype = ival; 1008 MKL_PARDISO_INIT(pt, &mat_mkl_pardiso->mtype, mat_mkl_pardiso->iparm); 1009 #if defined(PETSC_USE_REAL_SINGLE) 1010 mat_mkl_pardiso->iparm[27] = 1; 1011 #else 1012 mat_mkl_pardiso->iparm[27] = 0; 1013 #endif 1014 mat_mkl_pardiso->iparm[34] = backup; 1015 mat_mkl_pardiso->iparm[36] = bs; 1016 } else if (icntl==70) mat_mkl_pardiso->solve_interior = (PetscBool)!!ival; 1017 } 1018 PetscFunctionReturn(0); 1019 } 1020 1021 /*@ 1022 MatMkl_PardisoSetCntl - Set Mkl_Pardiso parameters 1023 1024 Logically Collective on Mat 1025 1026 Input Parameters: 1027 + F - the factored matrix obtained by calling MatGetFactor() 1028 . icntl - index of Mkl_Pardiso parameter 1029 - ival - value of Mkl_Pardiso parameter 1030 1031 Options Database: 1032 . -mat_mkl_pardiso_<icntl> <ival> 1033 1034 Level: beginner 1035 1036 References: 1037 . Mkl_Pardiso Users' Guide 1038 1039 .seealso: MatGetFactor() 1040 @*/ 1041 PetscErrorCode MatMkl_PardisoSetCntl(Mat F,PetscInt icntl,PetscInt ival) 1042 { 1043 PetscErrorCode ierr; 1044 1045 PetscFunctionBegin; 1046 ierr = PetscTryMethod(F,"MatMkl_PardisoSetCntl_C",(Mat,PetscInt,PetscInt),(F,icntl,ival));CHKERRQ(ierr); 1047 PetscFunctionReturn(0); 1048 } 1049 1050 /*MC 1051 MATSOLVERMKL_PARDISO - A matrix type providing direct solvers (LU) for 1052 sequential matrices via the external package MKL_PARDISO. 1053 1054 Works with MATSEQAIJ matrices 1055 1056 Use -pc_type lu -pc_factor_mat_solver_type mkl_pardiso to use this direct solver 1057 1058 Options Database Keys: 1059 + -mat_mkl_pardiso_65 - Number of threads to use within MKL_PARDISO 1060 . -mat_mkl_pardiso_66 - Maximum number of factors with identical sparsity structure that must be kept in memory at the same time 1061 . -mat_mkl_pardiso_67 - Indicates the actual matrix for the solution phase 1062 . -mat_mkl_pardiso_68 - Message level information 1063 . -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 1064 . -mat_mkl_pardiso_1 - Use default values 1065 . -mat_mkl_pardiso_2 - Fill-in reducing ordering for the input matrix 1066 . -mat_mkl_pardiso_4 - Preconditioned CGS/CG 1067 . -mat_mkl_pardiso_5 - User permutation 1068 . -mat_mkl_pardiso_6 - Write solution on x 1069 . -mat_mkl_pardiso_8 - Iterative refinement step 1070 . -mat_mkl_pardiso_10 - Pivoting perturbation 1071 . -mat_mkl_pardiso_11 - Scaling vectors 1072 . -mat_mkl_pardiso_12 - Solve with transposed or conjugate transposed matrix A 1073 . -mat_mkl_pardiso_13 - Improved accuracy using (non-) symmetric weighted matching 1074 . -mat_mkl_pardiso_18 - Numbers of non-zero elements 1075 . -mat_mkl_pardiso_19 - Report number of floating point operations 1076 . -mat_mkl_pardiso_21 - Pivoting for symmetric indefinite matrices 1077 . -mat_mkl_pardiso_24 - Parallel factorization control 1078 . -mat_mkl_pardiso_25 - Parallel forward/backward solve control 1079 . -mat_mkl_pardiso_27 - Matrix checker 1080 . -mat_mkl_pardiso_31 - Partial solve and computing selected components of the solution vectors 1081 . -mat_mkl_pardiso_34 - Optimal number of threads for conditional numerical reproducibility (CNR) mode 1082 - -mat_mkl_pardiso_60 - Intel MKL_PARDISO mode 1083 1084 Level: beginner 1085 1086 For more information please check mkl_pardiso manual 1087 1088 .seealso: PCFactorSetMatSolverType(), MatSolverType 1089 1090 M*/ 1091 static PetscErrorCode MatFactorGetSolverType_mkl_pardiso(Mat A, MatSolverType *type) 1092 { 1093 PetscFunctionBegin; 1094 *type = MATSOLVERMKL_PARDISO; 1095 PetscFunctionReturn(0); 1096 } 1097 1098 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_mkl_pardiso(Mat A,MatFactorType ftype,Mat *F) 1099 { 1100 Mat B; 1101 PetscErrorCode ierr; 1102 Mat_MKL_PARDISO *mat_mkl_pardiso; 1103 PetscBool isSeqAIJ,isSeqBAIJ,isSeqSBAIJ; 1104 1105 PetscFunctionBegin; 1106 ierr = PetscObjectBaseTypeCompare((PetscObject)A,MATSEQAIJ,&isSeqAIJ);CHKERRQ(ierr); 1107 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQBAIJ,&isSeqBAIJ);CHKERRQ(ierr); 1108 ierr = PetscObjectTypeCompare((PetscObject)A,MATSEQSBAIJ,&isSeqSBAIJ);CHKERRQ(ierr); 1109 ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr); 1110 ierr = MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N);CHKERRQ(ierr); 1111 ierr = PetscStrallocpy("mkl_pardiso",&((PetscObject)B)->type_name);CHKERRQ(ierr); 1112 ierr = MatSetUp(B);CHKERRQ(ierr); 1113 1114 ierr = PetscNewLog(B,&mat_mkl_pardiso);CHKERRQ(ierr); 1115 B->data = mat_mkl_pardiso; 1116 1117 ierr = MatFactorMKL_PARDISOInitialize_Private(A, ftype, mat_mkl_pardiso);CHKERRQ(ierr); 1118 if (ftype == MAT_FACTOR_LU) { 1119 B->ops->lufactorsymbolic = MatLUFactorSymbolic_AIJMKL_PARDISO; 1120 B->factortype = MAT_FACTOR_LU; 1121 mat_mkl_pardiso->needsym = PETSC_FALSE; 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) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO LU factor with SEQSBAIJ format! Use MAT_FACTOR_CHOLESKY instead"); 1125 else SETERRQ1(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO LU with %s format",((PetscObject)A)->type_name); 1126 #if defined(PETSC_USE_COMPLEX) 1127 mat_mkl_pardiso->mtype = 13; 1128 #else 1129 mat_mkl_pardiso->mtype = 11; 1130 #endif 1131 } else { 1132 B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_AIJMKL_PARDISO; 1133 B->factortype = MAT_FACTOR_CHOLESKY; 1134 if (isSeqAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqaij; 1135 else if (isSeqBAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqbaij; 1136 else if (isSeqSBAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqsbaij; 1137 else SETERRQ1(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO CHOLESKY with %s format",((PetscObject)A)->type_name); 1138 1139 mat_mkl_pardiso->needsym = PETSC_TRUE; 1140 #if !defined(PETSC_USE_COMPLEX) 1141 if (A->spd_set && A->spd) mat_mkl_pardiso->mtype = 2; 1142 else mat_mkl_pardiso->mtype = -2; 1143 #else 1144 mat_mkl_pardiso->mtype = 6; 1145 if (A->hermitian) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO CHOLESKY with Hermitian matrices! Use MAT_FACTOR_LU instead"); 1146 #endif 1147 } 1148 B->ops->destroy = MatDestroy_MKL_PARDISO; 1149 B->ops->view = MatView_MKL_PARDISO; 1150 B->ops->getinfo = MatGetInfo_MKL_PARDISO; 1151 B->factortype = ftype; 1152 B->assembled = PETSC_TRUE; 1153 1154 ierr = PetscFree(B->solvertype);CHKERRQ(ierr); 1155 ierr = PetscStrallocpy(MATSOLVERMKL_PARDISO,&B->solvertype);CHKERRQ(ierr); 1156 1157 ierr = PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverType_C",MatFactorGetSolverType_mkl_pardiso);CHKERRQ(ierr); 1158 ierr = PetscObjectComposeFunction((PetscObject)B,"MatFactorSetSchurIS_C",MatFactorSetSchurIS_MKL_PARDISO);CHKERRQ(ierr); 1159 ierr = PetscObjectComposeFunction((PetscObject)B,"MatMkl_PardisoSetCntl_C",MatMkl_PardisoSetCntl_MKL_PARDISO);CHKERRQ(ierr); 1160 1161 *F = B; 1162 PetscFunctionReturn(0); 1163 } 1164 1165 PETSC_EXTERN PetscErrorCode MatSolverTypeRegister_MKL_Pardiso(void) 1166 { 1167 PetscErrorCode ierr; 1168 1169 PetscFunctionBegin; 1170 ierr = MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQAIJ,MAT_FACTOR_LU,MatGetFactor_aij_mkl_pardiso);CHKERRQ(ierr); 1171 ierr = MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_aij_mkl_pardiso);CHKERRQ(ierr); 1172 ierr = MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQBAIJ,MAT_FACTOR_LU,MatGetFactor_aij_mkl_pardiso);CHKERRQ(ierr); 1173 ierr = MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQSBAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_aij_mkl_pardiso);CHKERRQ(ierr); 1174 PetscFunctionReturn(0); 1175 } 1176 1177