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 * Internal data structure. 58 * For more information check mkl_pardiso manual. 59 */ 60 typedef struct { 61 62 /* Configuration vector*/ 63 INT_TYPE iparm[IPARM_SIZE]; 64 65 /* 66 * Internal mkl_pardiso memory location. 67 * After the first call to mkl_pardiso do not modify pt, as that could cause a serious memory leak. 68 */ 69 void *pt[IPARM_SIZE]; 70 71 /* Basic mkl_pardiso info*/ 72 INT_TYPE phase, maxfct, mnum, mtype, n, nrhs, msglvl, err; 73 74 /* Matrix structure*/ 75 void *a; 76 INT_TYPE *ia, *ja; 77 78 /* Number of non-zero elements*/ 79 INT_TYPE nz; 80 81 /* Row permutaton vector*/ 82 INT_TYPE *perm; 83 84 /* Define if matrix preserves sparse structure.*/ 85 MatStructure matstruc; 86 87 PetscBool needsym; 88 PetscBool freeaij; 89 90 /* Schur complement */ 91 PetscScalar *schur; 92 PetscInt schur_size; 93 PetscInt *schur_idxs; 94 PetscScalar *schur_work; 95 PetscBLASInt schur_work_size; 96 PetscBool solve_interior; 97 98 /* True if mkl_pardiso function have been used.*/ 99 PetscBool CleanUp; 100 101 /* Conversion to a format suitable for MKL */ 102 PetscErrorCode (*Convert)(Mat, PetscBool, MatReuse, PetscBool*, INT_TYPE*, INT_TYPE**, INT_TYPE**, PetscScalar**); 103 } Mat_MKL_PARDISO; 104 105 PetscErrorCode MatMKLPardiso_Convert_seqsbaij(Mat A,PetscBool sym,MatReuse reuse,PetscBool *free,INT_TYPE *nnz,INT_TYPE **r,INT_TYPE **c,PetscScalar **v) 106 { 107 Mat_SeqSBAIJ *aa = (Mat_SeqSBAIJ*)A->data; 108 PetscInt bs = A->rmap->bs,i; 109 110 PetscFunctionBegin; 111 PetscCheck(sym,PetscObjectComm((PetscObject)A),PETSC_ERR_PLIB,"This should not happen"); 112 *v = aa->a; 113 if (bs == 1) { /* already in the correct format */ 114 /* though PetscInt and INT_TYPE are of the same size since they are defined differently the Intel compiler requires a cast */ 115 *r = (INT_TYPE*)aa->i; 116 *c = (INT_TYPE*)aa->j; 117 *nnz = (INT_TYPE)aa->nz; 118 *free = PETSC_FALSE; 119 } else if (reuse == MAT_INITIAL_MATRIX) { 120 PetscInt m = A->rmap->n,nz = aa->nz; 121 PetscInt *row,*col; 122 PetscCall(PetscMalloc2(m+1,&row,nz,&col)); 123 for (i=0; i<m+1; i++) { 124 row[i] = aa->i[i]+1; 125 } 126 for (i=0; i<nz; i++) { 127 col[i] = aa->j[i]+1; 128 } 129 *r = (INT_TYPE*)row; 130 *c = (INT_TYPE*)col; 131 *nnz = (INT_TYPE)nz; 132 *free = PETSC_TRUE; 133 } 134 PetscFunctionReturn(0); 135 } 136 137 PetscErrorCode MatMKLPardiso_Convert_seqbaij(Mat A,PetscBool sym,MatReuse reuse,PetscBool *free,INT_TYPE *nnz,INT_TYPE **r,INT_TYPE **c,PetscScalar **v) 138 { 139 Mat_SeqBAIJ *aa = (Mat_SeqBAIJ*)A->data; 140 PetscInt bs = A->rmap->bs,i; 141 142 PetscFunctionBegin; 143 if (!sym) { 144 *v = aa->a; 145 if (bs == 1) { /* already in the correct format */ 146 /* though PetscInt and INT_TYPE are of the same size since they are defined differently the Intel compiler requires a cast */ 147 *r = (INT_TYPE*)aa->i; 148 *c = (INT_TYPE*)aa->j; 149 *nnz = (INT_TYPE)aa->nz; 150 *free = PETSC_FALSE; 151 PetscFunctionReturn(0); 152 } else if (reuse == MAT_INITIAL_MATRIX) { 153 PetscInt m = A->rmap->n,nz = aa->nz; 154 PetscInt *row,*col; 155 PetscCall(PetscMalloc2(m+1,&row,nz,&col)); 156 for (i=0; i<m+1; i++) { 157 row[i] = aa->i[i]+1; 158 } 159 for (i=0; i<nz; i++) { 160 col[i] = aa->j[i]+1; 161 } 162 *r = (INT_TYPE*)row; 163 *c = (INT_TYPE*)col; 164 *nnz = (INT_TYPE)nz; 165 } 166 *free = PETSC_TRUE; 167 } else { 168 SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_PLIB,"This should not happen"); 169 } 170 PetscFunctionReturn(0); 171 } 172 173 PetscErrorCode MatMKLPardiso_Convert_seqaij(Mat A,PetscBool sym,MatReuse reuse,PetscBool *free,INT_TYPE *nnz,INT_TYPE **r,INT_TYPE **c,PetscScalar **v) 174 { 175 Mat_SeqAIJ *aa = (Mat_SeqAIJ*)A->data; 176 PetscScalar *aav; 177 178 PetscFunctionBegin; 179 PetscCall(MatSeqAIJGetArrayRead(A,(const PetscScalar**)&aav)); 180 if (!sym) { /* already in the correct format */ 181 *v = aav; 182 *r = (INT_TYPE*)aa->i; 183 *c = (INT_TYPE*)aa->j; 184 *nnz = (INT_TYPE)aa->nz; 185 *free = PETSC_FALSE; 186 } else if (reuse == MAT_INITIAL_MATRIX) { /* need to get the triangular part */ 187 PetscScalar *vals,*vv; 188 PetscInt *row,*col,*jj; 189 PetscInt m = A->rmap->n,nz,i; 190 191 nz = 0; 192 for (i=0; i<m; i++) nz += aa->i[i+1] - aa->diag[i]; 193 PetscCall(PetscMalloc2(m+1,&row,nz,&col)); 194 PetscCall(PetscMalloc1(nz,&vals)); 195 jj = col; 196 vv = vals; 197 198 row[0] = 0; 199 for (i=0; i<m; i++) { 200 PetscInt *aj = aa->j + aa->diag[i]; 201 PetscScalar *av = aav + aa->diag[i]; 202 PetscInt rl = aa->i[i+1] - aa->diag[i],j; 203 204 for (j=0; j<rl; j++) { 205 *jj = *aj; jj++; aj++; 206 *vv = *av; vv++; av++; 207 } 208 row[i+1] = row[i] + rl; 209 } 210 *v = vals; 211 *r = (INT_TYPE*)row; 212 *c = (INT_TYPE*)col; 213 *nnz = (INT_TYPE)nz; 214 *free = PETSC_TRUE; 215 } else { 216 PetscScalar *vv; 217 PetscInt m = A->rmap->n,i; 218 219 vv = *v; 220 for (i=0; i<m; i++) { 221 PetscScalar *av = aav + aa->diag[i]; 222 PetscInt rl = aa->i[i+1] - aa->diag[i],j; 223 for (j=0; j<rl; j++) { 224 *vv = *av; vv++; av++; 225 } 226 } 227 *free = PETSC_TRUE; 228 } 229 PetscCall(MatSeqAIJRestoreArrayRead(A,(const PetscScalar**)&aav)); 230 PetscFunctionReturn(0); 231 } 232 233 static PetscErrorCode MatMKLPardisoSolveSchur_Private(Mat F, PetscScalar *B, PetscScalar *X) 234 { 235 Mat_MKL_PARDISO *mpardiso = (Mat_MKL_PARDISO*)F->data; 236 Mat S,Xmat,Bmat; 237 MatFactorSchurStatus schurstatus; 238 239 PetscFunctionBegin; 240 PetscCall(MatFactorGetSchurComplement(F,&S,&schurstatus)); 241 PetscCheck(X != B || schurstatus != MAT_FACTOR_SCHUR_INVERTED,PETSC_COMM_SELF,PETSC_ERR_SUP,"X and B cannot point to the same address"); 242 PetscCall(MatCreateSeqDense(PETSC_COMM_SELF,mpardiso->schur_size,mpardiso->nrhs,B,&Bmat)); 243 PetscCall(MatCreateSeqDense(PETSC_COMM_SELF,mpardiso->schur_size,mpardiso->nrhs,X,&Xmat)); 244 PetscCall(MatSetType(Bmat,((PetscObject)S)->type_name)); 245 PetscCall(MatSetType(Xmat,((PetscObject)S)->type_name)); 246 #if defined(PETSC_HAVE_VIENNACL) || defined(PETSC_HAVE_CUDA) 247 PetscCall(MatBindToCPU(Xmat,S->boundtocpu)); 248 PetscCall(MatBindToCPU(Bmat,S->boundtocpu)); 249 #endif 250 251 #if defined(PETSC_USE_COMPLEX) 252 PetscCheck(mpardiso->iparm[12-1] != 1,PetscObjectComm((PetscObject)F),PETSC_ERR_SUP,"Hermitian solve not implemented yet"); 253 #endif 254 255 switch (schurstatus) { 256 case MAT_FACTOR_SCHUR_FACTORED: 257 if (!mpardiso->iparm[12-1]) { 258 PetscCall(MatMatSolve(S,Bmat,Xmat)); 259 } else { /* transpose solve */ 260 PetscCall(MatMatSolveTranspose(S,Bmat,Xmat)); 261 } 262 break; 263 case MAT_FACTOR_SCHUR_INVERTED: 264 PetscCall(MatProductCreateWithMat(S,Bmat,NULL,Xmat)); 265 if (!mpardiso->iparm[12-1]) { 266 PetscCall(MatProductSetType(Xmat,MATPRODUCT_AB)); 267 } else { /* transpose solve */ 268 PetscCall(MatProductSetType(Xmat,MATPRODUCT_AtB)); 269 } 270 PetscCall(MatProductSetFromOptions(Xmat)); 271 PetscCall(MatProductSymbolic(Xmat)); 272 PetscCall(MatProductNumeric(Xmat)); 273 PetscCall(MatProductClear(Xmat)); 274 break; 275 default: 276 SETERRQ(PetscObjectComm((PetscObject)F),PETSC_ERR_SUP,"Unhandled MatFactorSchurStatus %" PetscInt_FMT,F->schur_status); 277 break; 278 } 279 PetscCall(MatFactorRestoreSchurComplement(F,&S,schurstatus)); 280 PetscCall(MatDestroy(&Bmat)); 281 PetscCall(MatDestroy(&Xmat)); 282 PetscFunctionReturn(0); 283 } 284 285 PetscErrorCode MatFactorSetSchurIS_MKL_PARDISO(Mat F, IS is) 286 { 287 Mat_MKL_PARDISO *mpardiso = (Mat_MKL_PARDISO*)F->data; 288 const PetscScalar *arr; 289 const PetscInt *idxs; 290 PetscInt size,i; 291 PetscMPIInt csize; 292 PetscBool sorted; 293 294 PetscFunctionBegin; 295 PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)F),&csize)); 296 PetscCheck(csize <= 1,PETSC_COMM_SELF,PETSC_ERR_SUP,"MKL_PARDISO parallel Schur complements not yet supported from PETSc"); 297 PetscCall(ISSorted(is,&sorted)); 298 PetscCheck(sorted,PETSC_COMM_SELF,PETSC_ERR_SUP,"IS for MKL_PARDISO Schur complements needs to be sorted"); 299 PetscCall(ISGetLocalSize(is,&size)); 300 PetscCall(PetscFree(mpardiso->schur_work)); 301 PetscCall(PetscBLASIntCast(PetscMax(mpardiso->n,2*size),&mpardiso->schur_work_size)); 302 PetscCall(PetscMalloc1(mpardiso->schur_work_size,&mpardiso->schur_work)); 303 PetscCall(MatDestroy(&F->schur)); 304 PetscCall(MatCreateSeqDense(PETSC_COMM_SELF,size,size,NULL,&F->schur)); 305 PetscCall(MatDenseGetArrayRead(F->schur,&arr)); 306 mpardiso->schur = (PetscScalar*)arr; 307 mpardiso->schur_size = size; 308 PetscCall(MatDenseRestoreArrayRead(F->schur,&arr)); 309 if (mpardiso->mtype == 2) { 310 PetscCall(MatSetOption(F->schur,MAT_SPD,PETSC_TRUE)); 311 } 312 313 PetscCall(PetscFree(mpardiso->schur_idxs)); 314 PetscCall(PetscMalloc1(size,&mpardiso->schur_idxs)); 315 PetscCall(PetscArrayzero(mpardiso->perm,mpardiso->n)); 316 PetscCall(ISGetIndices(is,&idxs)); 317 PetscCall(PetscArraycpy(mpardiso->schur_idxs,idxs,size)); 318 for (i=0;i<size;i++) mpardiso->perm[idxs[i]] = 1; 319 PetscCall(ISRestoreIndices(is,&idxs)); 320 if (size) { /* turn on Schur switch if the set of indices is not empty */ 321 mpardiso->iparm[36-1] = 2; 322 } 323 PetscFunctionReturn(0); 324 } 325 326 PetscErrorCode MatDestroy_MKL_PARDISO(Mat A) 327 { 328 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 329 330 PetscFunctionBegin; 331 if (mat_mkl_pardiso->CleanUp) { 332 mat_mkl_pardiso->phase = JOB_RELEASE_OF_ALL_MEMORY; 333 334 MKL_PARDISO (mat_mkl_pardiso->pt, 335 &mat_mkl_pardiso->maxfct, 336 &mat_mkl_pardiso->mnum, 337 &mat_mkl_pardiso->mtype, 338 &mat_mkl_pardiso->phase, 339 &mat_mkl_pardiso->n, 340 NULL, 341 NULL, 342 NULL, 343 NULL, 344 &mat_mkl_pardiso->nrhs, 345 mat_mkl_pardiso->iparm, 346 &mat_mkl_pardiso->msglvl, 347 NULL, 348 NULL, 349 &mat_mkl_pardiso->err); 350 } 351 PetscCall(PetscFree(mat_mkl_pardiso->perm)); 352 PetscCall(PetscFree(mat_mkl_pardiso->schur_work)); 353 PetscCall(PetscFree(mat_mkl_pardiso->schur_idxs)); 354 if (mat_mkl_pardiso->freeaij) { 355 PetscCall(PetscFree2(mat_mkl_pardiso->ia,mat_mkl_pardiso->ja)); 356 if (mat_mkl_pardiso->iparm[34] == 1) { 357 PetscCall(PetscFree(mat_mkl_pardiso->a)); 358 } 359 } 360 PetscCall(PetscFree(A->data)); 361 362 /* clear composed functions */ 363 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatFactorGetSolverType_C",NULL)); 364 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatFactorSetSchurIS_C",NULL)); 365 PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatMkl_PardisoSetCntl_C",NULL)); 366 PetscFunctionReturn(0); 367 } 368 369 static PetscErrorCode MatMKLPardisoScatterSchur_Private(Mat_MKL_PARDISO *mpardiso, PetscScalar *whole, PetscScalar *schur, PetscBool reduce) 370 { 371 PetscFunctionBegin; 372 if (reduce) { /* data given for the whole matrix */ 373 PetscInt i,m=0,p=0; 374 for (i=0;i<mpardiso->nrhs;i++) { 375 PetscInt j; 376 for (j=0;j<mpardiso->schur_size;j++) { 377 schur[p+j] = whole[m+mpardiso->schur_idxs[j]]; 378 } 379 m += mpardiso->n; 380 p += mpardiso->schur_size; 381 } 382 } else { /* from Schur to whole */ 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 whole[m+mpardiso->schur_idxs[j]] = schur[p+j]; 388 } 389 m += mpardiso->n; 390 p += mpardiso->schur_size; 391 } 392 } 393 PetscFunctionReturn(0); 394 } 395 396 PetscErrorCode MatSolve_MKL_PARDISO(Mat A,Vec b,Vec x) 397 { 398 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 399 PetscScalar *xarray; 400 const PetscScalar *barray; 401 402 PetscFunctionBegin; 403 mat_mkl_pardiso->nrhs = 1; 404 PetscCall(VecGetArrayWrite(x,&xarray)); 405 PetscCall(VecGetArrayRead(b,&barray)); 406 407 if (!mat_mkl_pardiso->schur) mat_mkl_pardiso->phase = JOB_SOLVE_ITERATIVE_REFINEMENT; 408 else mat_mkl_pardiso->phase = JOB_SOLVE_FORWARD_SUBSTITUTION; 409 410 if (barray == xarray) { /* if the two vectors share the same memory */ 411 PetscScalar *work; 412 if (!mat_mkl_pardiso->schur_work) { 413 PetscCall(PetscMalloc1(mat_mkl_pardiso->n,&work)); 414 } else { 415 work = mat_mkl_pardiso->schur_work; 416 } 417 mat_mkl_pardiso->iparm[6-1] = 1; 418 MKL_PARDISO (mat_mkl_pardiso->pt, 419 &mat_mkl_pardiso->maxfct, 420 &mat_mkl_pardiso->mnum, 421 &mat_mkl_pardiso->mtype, 422 &mat_mkl_pardiso->phase, 423 &mat_mkl_pardiso->n, 424 mat_mkl_pardiso->a, 425 mat_mkl_pardiso->ia, 426 mat_mkl_pardiso->ja, 427 NULL, 428 &mat_mkl_pardiso->nrhs, 429 mat_mkl_pardiso->iparm, 430 &mat_mkl_pardiso->msglvl, 431 (void*)xarray, 432 (void*)work, 433 &mat_mkl_pardiso->err); 434 if (!mat_mkl_pardiso->schur_work) { 435 PetscCall(PetscFree(work)); 436 } 437 } else { 438 mat_mkl_pardiso->iparm[6-1] = 0; 439 MKL_PARDISO (mat_mkl_pardiso->pt, 440 &mat_mkl_pardiso->maxfct, 441 &mat_mkl_pardiso->mnum, 442 &mat_mkl_pardiso->mtype, 443 &mat_mkl_pardiso->phase, 444 &mat_mkl_pardiso->n, 445 mat_mkl_pardiso->a, 446 mat_mkl_pardiso->ia, 447 mat_mkl_pardiso->ja, 448 mat_mkl_pardiso->perm, 449 &mat_mkl_pardiso->nrhs, 450 mat_mkl_pardiso->iparm, 451 &mat_mkl_pardiso->msglvl, 452 (void*)barray, 453 (void*)xarray, 454 &mat_mkl_pardiso->err); 455 } 456 PetscCall(VecRestoreArrayRead(b,&barray)); 457 458 PetscCheck(mat_mkl_pardiso->err >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 459 460 if (mat_mkl_pardiso->schur) { /* solve Schur complement and expand solution */ 461 if (!mat_mkl_pardiso->solve_interior) { 462 PetscInt shift = mat_mkl_pardiso->schur_size; 463 464 PetscCall(MatFactorFactorizeSchurComplement(A)); 465 /* if inverted, uses BLAS *MM subroutines, otherwise LAPACK *TRS */ 466 if (A->schur_status != MAT_FACTOR_SCHUR_INVERTED) shift = 0; 467 468 /* solve Schur complement */ 469 PetscCall(MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work,PETSC_TRUE)); 470 PetscCall(MatMKLPardisoSolveSchur_Private(A,mat_mkl_pardiso->schur_work,mat_mkl_pardiso->schur_work+shift)); 471 PetscCall(MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work+shift,PETSC_FALSE)); 472 } else { /* if we are solving for the interior problem, any value in barray[schur] forward-substituted to xarray[schur] will be neglected */ 473 PetscInt i; 474 for (i=0;i<mat_mkl_pardiso->schur_size;i++) { 475 xarray[mat_mkl_pardiso->schur_idxs[i]] = 0.; 476 } 477 } 478 479 /* expansion phase */ 480 mat_mkl_pardiso->iparm[6-1] = 1; 481 mat_mkl_pardiso->phase = JOB_SOLVE_BACKWARD_SUBSTITUTION; 482 MKL_PARDISO (mat_mkl_pardiso->pt, 483 &mat_mkl_pardiso->maxfct, 484 &mat_mkl_pardiso->mnum, 485 &mat_mkl_pardiso->mtype, 486 &mat_mkl_pardiso->phase, 487 &mat_mkl_pardiso->n, 488 mat_mkl_pardiso->a, 489 mat_mkl_pardiso->ia, 490 mat_mkl_pardiso->ja, 491 mat_mkl_pardiso->perm, 492 &mat_mkl_pardiso->nrhs, 493 mat_mkl_pardiso->iparm, 494 &mat_mkl_pardiso->msglvl, 495 (void*)xarray, 496 (void*)mat_mkl_pardiso->schur_work, /* according to the specs, the solution vector is always used */ 497 &mat_mkl_pardiso->err); 498 499 PetscCheck(mat_mkl_pardiso->err >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 500 mat_mkl_pardiso->iparm[6-1] = 0; 501 } 502 PetscCall(VecRestoreArrayWrite(x,&xarray)); 503 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 504 PetscFunctionReturn(0); 505 } 506 507 PetscErrorCode MatSolveTranspose_MKL_PARDISO(Mat A,Vec b,Vec x) 508 { 509 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 510 PetscInt oiparm12; 511 512 PetscFunctionBegin; 513 oiparm12 = mat_mkl_pardiso->iparm[12 - 1]; 514 mat_mkl_pardiso->iparm[12 - 1] = 2; 515 PetscCall(MatSolve_MKL_PARDISO(A,b,x)); 516 mat_mkl_pardiso->iparm[12 - 1] = oiparm12; 517 PetscFunctionReturn(0); 518 } 519 520 PetscErrorCode MatMatSolve_MKL_PARDISO(Mat A,Mat B,Mat X) 521 { 522 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)(A)->data; 523 const PetscScalar *barray; 524 PetscScalar *xarray; 525 PetscBool flg; 526 527 PetscFunctionBegin; 528 PetscCall(PetscObjectBaseTypeCompare((PetscObject)B,MATSEQDENSE,&flg)); 529 PetscCheck(flg,PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix B must be MATSEQDENSE matrix"); 530 if (X != B) { 531 PetscCall(PetscObjectBaseTypeCompare((PetscObject)X,MATSEQDENSE,&flg)); 532 PetscCheck(flg,PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix X must be MATSEQDENSE matrix"); 533 } 534 535 PetscCall(MatGetSize(B,NULL,(PetscInt*)&mat_mkl_pardiso->nrhs)); 536 537 if (mat_mkl_pardiso->nrhs > 0) { 538 PetscCall(MatDenseGetArrayRead(B,&barray)); 539 PetscCall(MatDenseGetArrayWrite(X,&xarray)); 540 541 PetscCheck(barray != xarray,PETSC_COMM_SELF,PETSC_ERR_SUP,"B and X cannot share the same memory location"); 542 if (!mat_mkl_pardiso->schur) mat_mkl_pardiso->phase = JOB_SOLVE_ITERATIVE_REFINEMENT; 543 else mat_mkl_pardiso->phase = JOB_SOLVE_FORWARD_SUBSTITUTION; 544 545 MKL_PARDISO (mat_mkl_pardiso->pt, 546 &mat_mkl_pardiso->maxfct, 547 &mat_mkl_pardiso->mnum, 548 &mat_mkl_pardiso->mtype, 549 &mat_mkl_pardiso->phase, 550 &mat_mkl_pardiso->n, 551 mat_mkl_pardiso->a, 552 mat_mkl_pardiso->ia, 553 mat_mkl_pardiso->ja, 554 mat_mkl_pardiso->perm, 555 &mat_mkl_pardiso->nrhs, 556 mat_mkl_pardiso->iparm, 557 &mat_mkl_pardiso->msglvl, 558 (void*)barray, 559 (void*)xarray, 560 &mat_mkl_pardiso->err); 561 PetscCheck(mat_mkl_pardiso->err >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 562 563 PetscCall(MatDenseRestoreArrayRead(B,&barray)); 564 if (mat_mkl_pardiso->schur) { /* solve Schur complement and expand solution */ 565 PetscScalar *o_schur_work = NULL; 566 567 /* solve Schur complement */ 568 if (!mat_mkl_pardiso->solve_interior) { 569 PetscInt shift = mat_mkl_pardiso->schur_size*mat_mkl_pardiso->nrhs,scale; 570 PetscInt mem = mat_mkl_pardiso->n*mat_mkl_pardiso->nrhs; 571 572 PetscCall(MatFactorFactorizeSchurComplement(A)); 573 /* allocate extra memory if it is needed */ 574 scale = 1; 575 if (A->schur_status == MAT_FACTOR_SCHUR_INVERTED) scale = 2; 576 mem *= scale; 577 if (mem > mat_mkl_pardiso->schur_work_size) { 578 o_schur_work = mat_mkl_pardiso->schur_work; 579 PetscCall(PetscMalloc1(mem,&mat_mkl_pardiso->schur_work)); 580 } 581 /* if inverted, uses BLAS *MM subroutines, otherwise LAPACK *TRS */ 582 if (A->schur_status != MAT_FACTOR_SCHUR_INVERTED) shift = 0; 583 PetscCall(MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work,PETSC_TRUE)); 584 PetscCall(MatMKLPardisoSolveSchur_Private(A,mat_mkl_pardiso->schur_work,mat_mkl_pardiso->schur_work+shift)); 585 PetscCall(MatMKLPardisoScatterSchur_Private(mat_mkl_pardiso,xarray,mat_mkl_pardiso->schur_work+shift,PETSC_FALSE)); 586 } else { /* if we are solving for the interior problem, any value in barray[schur,n] forward-substituted to xarray[schur,n] will be neglected */ 587 PetscInt i,n,m=0; 588 for (n=0;n<mat_mkl_pardiso->nrhs;n++) { 589 for (i=0;i<mat_mkl_pardiso->schur_size;i++) { 590 xarray[mat_mkl_pardiso->schur_idxs[i]+m] = 0.; 591 } 592 m += mat_mkl_pardiso->n; 593 } 594 } 595 596 /* expansion phase */ 597 mat_mkl_pardiso->iparm[6-1] = 1; 598 mat_mkl_pardiso->phase = JOB_SOLVE_BACKWARD_SUBSTITUTION; 599 MKL_PARDISO (mat_mkl_pardiso->pt, 600 &mat_mkl_pardiso->maxfct, 601 &mat_mkl_pardiso->mnum, 602 &mat_mkl_pardiso->mtype, 603 &mat_mkl_pardiso->phase, 604 &mat_mkl_pardiso->n, 605 mat_mkl_pardiso->a, 606 mat_mkl_pardiso->ia, 607 mat_mkl_pardiso->ja, 608 mat_mkl_pardiso->perm, 609 &mat_mkl_pardiso->nrhs, 610 mat_mkl_pardiso->iparm, 611 &mat_mkl_pardiso->msglvl, 612 (void*)xarray, 613 (void*)mat_mkl_pardiso->schur_work, /* according to the specs, the solution vector is always used */ 614 &mat_mkl_pardiso->err); 615 if (o_schur_work) { /* restore original schur_work (minimal size) */ 616 PetscCall(PetscFree(mat_mkl_pardiso->schur_work)); 617 mat_mkl_pardiso->schur_work = o_schur_work; 618 } 619 PetscCheck(mat_mkl_pardiso->err >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 620 mat_mkl_pardiso->iparm[6-1] = 0; 621 } 622 PetscCall(MatDenseRestoreArrayWrite(X,&xarray)); 623 } 624 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 625 PetscFunctionReturn(0); 626 } 627 628 PetscErrorCode MatFactorNumeric_MKL_PARDISO(Mat F,Mat A,const MatFactorInfo *info) 629 { 630 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)(F)->data; 631 632 PetscFunctionBegin; 633 mat_mkl_pardiso->matstruc = SAME_NONZERO_PATTERN; 634 PetscCall((*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)); 635 636 mat_mkl_pardiso->phase = JOB_NUMERICAL_FACTORIZATION; 637 MKL_PARDISO (mat_mkl_pardiso->pt, 638 &mat_mkl_pardiso->maxfct, 639 &mat_mkl_pardiso->mnum, 640 &mat_mkl_pardiso->mtype, 641 &mat_mkl_pardiso->phase, 642 &mat_mkl_pardiso->n, 643 mat_mkl_pardiso->a, 644 mat_mkl_pardiso->ia, 645 mat_mkl_pardiso->ja, 646 mat_mkl_pardiso->perm, 647 &mat_mkl_pardiso->nrhs, 648 mat_mkl_pardiso->iparm, 649 &mat_mkl_pardiso->msglvl, 650 NULL, 651 (void*)mat_mkl_pardiso->schur, 652 &mat_mkl_pardiso->err); 653 PetscCheck(mat_mkl_pardiso->err >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 654 655 /* report flops */ 656 if (mat_mkl_pardiso->iparm[18] > 0) { 657 PetscCall(PetscLogFlops(PetscPowRealInt(10.,6)*mat_mkl_pardiso->iparm[18])); 658 } 659 660 if (F->schur) { /* schur output from pardiso is in row major format */ 661 #if defined(PETSC_HAVE_CUDA) 662 F->schur->offloadmask = PETSC_OFFLOAD_CPU; 663 #endif 664 PetscCall(MatFactorRestoreSchurComplement(F,NULL,MAT_FACTOR_SCHUR_UNFACTORED)); 665 PetscCall(MatTranspose(F->schur,MAT_INPLACE_MATRIX,&F->schur)); 666 } 667 mat_mkl_pardiso->matstruc = SAME_NONZERO_PATTERN; 668 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 669 PetscFunctionReturn(0); 670 } 671 672 PetscErrorCode MatSetFromOptions_MKL_PARDISO(Mat F, Mat A) 673 { 674 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)F->data; 675 PetscInt icntl,bs,threads=1; 676 PetscBool flg; 677 678 PetscFunctionBegin; 679 PetscOptionsBegin(PetscObjectComm((PetscObject)F),((PetscObject)F)->prefix,"MKL_PARDISO Options","Mat"); 680 681 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_65","Suggested number of threads to use within PARDISO","None",threads,&threads,&flg)); 682 if (flg) PetscSetMKL_PARDISOThreads((int)threads); 683 684 PetscCall(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)); 685 if (flg) mat_mkl_pardiso->maxfct = icntl; 686 687 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_67","Indicates the actual matrix for the solution phase","None",mat_mkl_pardiso->mnum,&icntl,&flg)); 688 if (flg) mat_mkl_pardiso->mnum = icntl; 689 690 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_68","Message level information","None",mat_mkl_pardiso->msglvl,&icntl,&flg)); 691 if (flg) mat_mkl_pardiso->msglvl = icntl; 692 693 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_69","Defines the matrix type","None",mat_mkl_pardiso->mtype,&icntl,&flg)); 694 if (flg) { 695 void *pt[IPARM_SIZE]; 696 mat_mkl_pardiso->mtype = icntl; 697 icntl = mat_mkl_pardiso->iparm[34]; 698 bs = mat_mkl_pardiso->iparm[36]; 699 MKL_PARDISO_INIT(pt, &mat_mkl_pardiso->mtype, mat_mkl_pardiso->iparm); 700 #if defined(PETSC_USE_REAL_SINGLE) 701 mat_mkl_pardiso->iparm[27] = 1; 702 #else 703 mat_mkl_pardiso->iparm[27] = 0; 704 #endif 705 mat_mkl_pardiso->iparm[34] = icntl; 706 mat_mkl_pardiso->iparm[36] = bs; 707 } 708 709 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_1","Use default values (if 0)","None",mat_mkl_pardiso->iparm[0],&icntl,&flg)); 710 if (flg) mat_mkl_pardiso->iparm[0] = icntl; 711 712 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_2","Fill-in reducing ordering for the input matrix","None",mat_mkl_pardiso->iparm[1],&icntl,&flg)); 713 if (flg) mat_mkl_pardiso->iparm[1] = icntl; 714 715 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_4","Preconditioned CGS/CG","None",mat_mkl_pardiso->iparm[3],&icntl,&flg)); 716 if (flg) mat_mkl_pardiso->iparm[3] = icntl; 717 718 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_5","User permutation","None",mat_mkl_pardiso->iparm[4],&icntl,&flg)); 719 if (flg) mat_mkl_pardiso->iparm[4] = icntl; 720 721 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_6","Write solution on x","None",mat_mkl_pardiso->iparm[5],&icntl,&flg)); 722 if (flg) mat_mkl_pardiso->iparm[5] = icntl; 723 724 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_8","Iterative refinement step","None",mat_mkl_pardiso->iparm[7],&icntl,&flg)); 725 if (flg) mat_mkl_pardiso->iparm[7] = icntl; 726 727 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_10","Pivoting perturbation","None",mat_mkl_pardiso->iparm[9],&icntl,&flg)); 728 if (flg) mat_mkl_pardiso->iparm[9] = icntl; 729 730 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_11","Scaling vectors","None",mat_mkl_pardiso->iparm[10],&icntl,&flg)); 731 if (flg) mat_mkl_pardiso->iparm[10] = icntl; 732 733 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_12","Solve with transposed or conjugate transposed matrix A","None",mat_mkl_pardiso->iparm[11],&icntl,&flg)); 734 if (flg) mat_mkl_pardiso->iparm[11] = icntl; 735 736 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_13","Improved accuracy using (non-) symmetric weighted matching","None",mat_mkl_pardiso->iparm[12],&icntl,&flg)); 737 if (flg) mat_mkl_pardiso->iparm[12] = icntl; 738 739 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_18","Numbers of non-zero elements","None",mat_mkl_pardiso->iparm[17],&icntl,&flg)); 740 if (flg) mat_mkl_pardiso->iparm[17] = icntl; 741 742 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_19","Report number of floating point operations (0 to disable)","None",mat_mkl_pardiso->iparm[18],&icntl,&flg)); 743 if (flg) mat_mkl_pardiso->iparm[18] = icntl; 744 745 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_21","Pivoting for symmetric indefinite matrices","None",mat_mkl_pardiso->iparm[20],&icntl,&flg)); 746 if (flg) mat_mkl_pardiso->iparm[20] = icntl; 747 748 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_24","Parallel factorization control","None",mat_mkl_pardiso->iparm[23],&icntl,&flg)); 749 if (flg) mat_mkl_pardiso->iparm[23] = icntl; 750 751 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_25","Parallel forward/backward solve control","None",mat_mkl_pardiso->iparm[24],&icntl,&flg)); 752 if (flg) mat_mkl_pardiso->iparm[24] = icntl; 753 754 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_27","Matrix checker","None",mat_mkl_pardiso->iparm[26],&icntl,&flg)); 755 if (flg) mat_mkl_pardiso->iparm[26] = icntl; 756 757 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_31","Partial solve and computing selected components of the solution vectors","None",mat_mkl_pardiso->iparm[30],&icntl,&flg)); 758 if (flg) mat_mkl_pardiso->iparm[30] = icntl; 759 760 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_34","Optimal number of threads for conditional numerical reproducibility (CNR) mode","None",mat_mkl_pardiso->iparm[33],&icntl,&flg)); 761 if (flg) mat_mkl_pardiso->iparm[33] = icntl; 762 763 PetscCall(PetscOptionsInt("-mat_mkl_pardiso_60","Intel MKL_PARDISO mode","None",mat_mkl_pardiso->iparm[59],&icntl,&flg)); 764 if (flg) mat_mkl_pardiso->iparm[59] = icntl; 765 PetscOptionsEnd(); 766 PetscFunctionReturn(0); 767 } 768 769 PetscErrorCode MatFactorMKL_PARDISOInitialize_Private(Mat A, MatFactorType ftype, Mat_MKL_PARDISO *mat_mkl_pardiso) 770 { 771 PetscInt i,bs; 772 PetscBool match; 773 774 PetscFunctionBegin; 775 for (i=0; i<IPARM_SIZE; i++) mat_mkl_pardiso->iparm[i] = 0; 776 for (i=0; i<IPARM_SIZE; i++) mat_mkl_pardiso->pt[i] = 0; 777 #if defined(PETSC_USE_REAL_SINGLE) 778 mat_mkl_pardiso->iparm[27] = 1; 779 #else 780 mat_mkl_pardiso->iparm[27] = 0; 781 #endif 782 /* Default options for both sym and unsym */ 783 mat_mkl_pardiso->iparm[ 0] = 1; /* Solver default parameters overriden with provided by iparm */ 784 mat_mkl_pardiso->iparm[ 1] = 2; /* Metis reordering */ 785 mat_mkl_pardiso->iparm[ 5] = 0; /* Write solution into x */ 786 mat_mkl_pardiso->iparm[ 7] = 0; /* Max number of iterative refinement steps */ 787 mat_mkl_pardiso->iparm[17] = -1; /* Output: Number of nonzeros in the factor LU */ 788 mat_mkl_pardiso->iparm[18] = -1; /* Output: Mflops for LU factorization */ 789 #if 0 790 mat_mkl_pardiso->iparm[23] = 1; /* Parallel factorization control*/ 791 #endif 792 PetscCall(PetscObjectTypeCompareAny((PetscObject)A,&match,MATSEQBAIJ,MATSEQSBAIJ,"")); 793 PetscCall(MatGetBlockSize(A,&bs)); 794 if (!match || bs == 1) { 795 mat_mkl_pardiso->iparm[34] = 1; /* Cluster Sparse Solver use C-style indexing for ia and ja arrays */ 796 mat_mkl_pardiso->n = A->rmap->N; 797 } else { 798 mat_mkl_pardiso->iparm[34] = 0; /* Cluster Sparse Solver use Fortran-style indexing for ia and ja arrays */ 799 mat_mkl_pardiso->iparm[36] = bs; 800 mat_mkl_pardiso->n = A->rmap->N/bs; 801 } 802 mat_mkl_pardiso->iparm[39] = 0; /* Input: matrix/rhs/solution stored on rank-0 */ 803 804 mat_mkl_pardiso->CleanUp = PETSC_FALSE; 805 mat_mkl_pardiso->maxfct = 1; /* Maximum number of numerical factorizations. */ 806 mat_mkl_pardiso->mnum = 1; /* Which factorization to use. */ 807 mat_mkl_pardiso->msglvl = 0; /* 0: do not print 1: Print statistical information in file */ 808 mat_mkl_pardiso->phase = -1; 809 mat_mkl_pardiso->err = 0; 810 811 mat_mkl_pardiso->nrhs = 1; 812 mat_mkl_pardiso->err = 0; 813 mat_mkl_pardiso->phase = -1; 814 815 if (ftype == MAT_FACTOR_LU) { 816 mat_mkl_pardiso->iparm[ 9] = 13; /* Perturb the pivot elements with 1E-13 */ 817 mat_mkl_pardiso->iparm[10] = 1; /* Use nonsymmetric permutation and scaling MPS */ 818 mat_mkl_pardiso->iparm[12] = 1; /* Switch on Maximum Weighted Matching algorithm (default for non-symmetric) */ 819 } else { 820 mat_mkl_pardiso->iparm[ 9] = 8; /* Perturb the pivot elements with 1E-8 */ 821 mat_mkl_pardiso->iparm[10] = 0; /* Use nonsymmetric permutation and scaling MPS */ 822 mat_mkl_pardiso->iparm[12] = 1; /* Switch on Maximum Weighted Matching algorithm (default for non-symmetric) */ 823 #if defined(PETSC_USE_DEBUG) 824 mat_mkl_pardiso->iparm[26] = 1; /* Matrix checker */ 825 #endif 826 } 827 PetscCall(PetscCalloc1(A->rmap->N*sizeof(INT_TYPE), &mat_mkl_pardiso->perm)); 828 mat_mkl_pardiso->schur_size = 0; 829 PetscFunctionReturn(0); 830 } 831 832 PetscErrorCode MatFactorSymbolic_AIJMKL_PARDISO_Private(Mat F,Mat A,const MatFactorInfo *info) 833 { 834 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)F->data; 835 836 PetscFunctionBegin; 837 mat_mkl_pardiso->matstruc = DIFFERENT_NONZERO_PATTERN; 838 PetscCall(MatSetFromOptions_MKL_PARDISO(F,A)); 839 /* throw away any previously computed structure */ 840 if (mat_mkl_pardiso->freeaij) { 841 PetscCall(PetscFree2(mat_mkl_pardiso->ia,mat_mkl_pardiso->ja)); 842 if (mat_mkl_pardiso->iparm[34] == 1) { 843 PetscCall(PetscFree(mat_mkl_pardiso->a)); 844 } 845 } 846 PetscCall((*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)); 847 if (mat_mkl_pardiso->iparm[34] == 1) mat_mkl_pardiso->n = A->rmap->N; 848 else mat_mkl_pardiso->n = A->rmap->N/A->rmap->bs; 849 850 mat_mkl_pardiso->phase = JOB_ANALYSIS; 851 852 /* reset flops counting if requested */ 853 if (mat_mkl_pardiso->iparm[18]) mat_mkl_pardiso->iparm[18] = -1; 854 855 MKL_PARDISO (mat_mkl_pardiso->pt, 856 &mat_mkl_pardiso->maxfct, 857 &mat_mkl_pardiso->mnum, 858 &mat_mkl_pardiso->mtype, 859 &mat_mkl_pardiso->phase, 860 &mat_mkl_pardiso->n, 861 mat_mkl_pardiso->a, 862 mat_mkl_pardiso->ia, 863 mat_mkl_pardiso->ja, 864 mat_mkl_pardiso->perm, 865 &mat_mkl_pardiso->nrhs, 866 mat_mkl_pardiso->iparm, 867 &mat_mkl_pardiso->msglvl, 868 NULL, 869 NULL, 870 &mat_mkl_pardiso->err); 871 PetscCheck(mat_mkl_pardiso->err >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB,"Error reported by MKL_PARDISO: err=%d. Please check manual",mat_mkl_pardiso->err); 872 873 mat_mkl_pardiso->CleanUp = PETSC_TRUE; 874 875 if (F->factortype == MAT_FACTOR_LU) F->ops->lufactornumeric = MatFactorNumeric_MKL_PARDISO; 876 else F->ops->choleskyfactornumeric = MatFactorNumeric_MKL_PARDISO; 877 878 F->ops->solve = MatSolve_MKL_PARDISO; 879 F->ops->solvetranspose = MatSolveTranspose_MKL_PARDISO; 880 F->ops->matsolve = MatMatSolve_MKL_PARDISO; 881 PetscFunctionReturn(0); 882 } 883 884 PetscErrorCode MatLUFactorSymbolic_AIJMKL_PARDISO(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info) 885 { 886 PetscFunctionBegin; 887 PetscCall(MatFactorSymbolic_AIJMKL_PARDISO_Private(F, A, info)); 888 PetscFunctionReturn(0); 889 } 890 891 #if !defined(PETSC_USE_COMPLEX) 892 PetscErrorCode MatGetInertia_MKL_PARDISO(Mat F,PetscInt *nneg,PetscInt *nzero,PetscInt *npos) 893 { 894 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)F->data; 895 896 PetscFunctionBegin; 897 if (nneg) *nneg = mat_mkl_pardiso->iparm[22]; 898 if (npos) *npos = mat_mkl_pardiso->iparm[21]; 899 if (nzero) *nzero = F->rmap->N - (mat_mkl_pardiso->iparm[22] + mat_mkl_pardiso->iparm[21]); 900 PetscFunctionReturn(0); 901 } 902 #endif 903 904 PetscErrorCode MatCholeskyFactorSymbolic_AIJMKL_PARDISO(Mat F,Mat A,IS r,const MatFactorInfo *info) 905 { 906 PetscFunctionBegin; 907 PetscCall(MatFactorSymbolic_AIJMKL_PARDISO_Private(F, A, info)); 908 #if defined(PETSC_USE_COMPLEX) 909 F->ops->getinertia = NULL; 910 #else 911 F->ops->getinertia = MatGetInertia_MKL_PARDISO; 912 #endif 913 PetscFunctionReturn(0); 914 } 915 916 PetscErrorCode MatView_MKL_PARDISO(Mat A, PetscViewer viewer) 917 { 918 PetscBool iascii; 919 PetscViewerFormat format; 920 Mat_MKL_PARDISO *mat_mkl_pardiso=(Mat_MKL_PARDISO*)A->data; 921 PetscInt i; 922 923 PetscFunctionBegin; 924 if (A->ops->solve != MatSolve_MKL_PARDISO) PetscFunctionReturn(0); 925 926 PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii)); 927 if (iascii) { 928 PetscCall(PetscViewerGetFormat(viewer,&format)); 929 if (format == PETSC_VIEWER_ASCII_INFO) { 930 PetscCall(PetscViewerASCIIPrintf(viewer,"MKL_PARDISO run parameters:\n")); 931 PetscCall(PetscViewerASCIIPrintf(viewer,"MKL_PARDISO phase: %d \n",mat_mkl_pardiso->phase)); 932 for (i=1; i<=64; i++) { 933 PetscCall(PetscViewerASCIIPrintf(viewer,"MKL_PARDISO iparm[%d]: %d \n",i, mat_mkl_pardiso->iparm[i - 1])); 934 } 935 PetscCall(PetscViewerASCIIPrintf(viewer,"MKL_PARDISO maxfct: %d \n", mat_mkl_pardiso->maxfct)); 936 PetscCall(PetscViewerASCIIPrintf(viewer,"MKL_PARDISO mnum: %d \n", mat_mkl_pardiso->mnum)); 937 PetscCall(PetscViewerASCIIPrintf(viewer,"MKL_PARDISO mtype: %d \n", mat_mkl_pardiso->mtype)); 938 PetscCall(PetscViewerASCIIPrintf(viewer,"MKL_PARDISO n: %d \n", mat_mkl_pardiso->n)); 939 PetscCall(PetscViewerASCIIPrintf(viewer,"MKL_PARDISO nrhs: %d \n", mat_mkl_pardiso->nrhs)); 940 PetscCall(PetscViewerASCIIPrintf(viewer,"MKL_PARDISO msglvl: %d \n", mat_mkl_pardiso->msglvl)); 941 } 942 } 943 PetscFunctionReturn(0); 944 } 945 946 PetscErrorCode MatGetInfo_MKL_PARDISO(Mat A, MatInfoType flag, MatInfo *info) 947 { 948 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)A->data; 949 950 PetscFunctionBegin; 951 info->block_size = 1.0; 952 info->nz_used = mat_mkl_pardiso->iparm[17]; 953 info->nz_allocated = mat_mkl_pardiso->iparm[17]; 954 info->nz_unneeded = 0.0; 955 info->assemblies = 0.0; 956 info->mallocs = 0.0; 957 info->memory = 0.0; 958 info->fill_ratio_given = 0; 959 info->fill_ratio_needed = 0; 960 info->factor_mallocs = 0; 961 PetscFunctionReturn(0); 962 } 963 964 PetscErrorCode MatMkl_PardisoSetCntl_MKL_PARDISO(Mat F,PetscInt icntl,PetscInt ival) 965 { 966 PetscInt backup,bs; 967 Mat_MKL_PARDISO *mat_mkl_pardiso = (Mat_MKL_PARDISO*)F->data; 968 969 PetscFunctionBegin; 970 if (icntl <= 64) { 971 mat_mkl_pardiso->iparm[icntl - 1] = ival; 972 } else { 973 if (icntl == 65) PetscSetMKL_PARDISOThreads(ival); 974 else if (icntl == 66) mat_mkl_pardiso->maxfct = ival; 975 else if (icntl == 67) mat_mkl_pardiso->mnum = ival; 976 else if (icntl == 68) mat_mkl_pardiso->msglvl = ival; 977 else if (icntl == 69) { 978 void *pt[IPARM_SIZE]; 979 backup = mat_mkl_pardiso->iparm[34]; 980 bs = mat_mkl_pardiso->iparm[36]; 981 mat_mkl_pardiso->mtype = ival; 982 MKL_PARDISO_INIT(pt, &mat_mkl_pardiso->mtype, mat_mkl_pardiso->iparm); 983 #if defined(PETSC_USE_REAL_SINGLE) 984 mat_mkl_pardiso->iparm[27] = 1; 985 #else 986 mat_mkl_pardiso->iparm[27] = 0; 987 #endif 988 mat_mkl_pardiso->iparm[34] = backup; 989 mat_mkl_pardiso->iparm[36] = bs; 990 } else if (icntl==70) mat_mkl_pardiso->solve_interior = (PetscBool)!!ival; 991 } 992 PetscFunctionReturn(0); 993 } 994 995 /*@ 996 MatMkl_PardisoSetCntl - Set Mkl_Pardiso parameters 997 998 Logically Collective on Mat 999 1000 Input Parameters: 1001 + F - the factored matrix obtained by calling MatGetFactor() 1002 . icntl - index of Mkl_Pardiso parameter 1003 - ival - value of Mkl_Pardiso parameter 1004 1005 Options Database: 1006 . -mat_mkl_pardiso_<icntl> <ival> - change the option numbered icntl to the value ival 1007 1008 Level: beginner 1009 1010 References: 1011 . * - Mkl_Pardiso Users' Guide 1012 1013 .seealso: `MatGetFactor()` 1014 @*/ 1015 PetscErrorCode MatMkl_PardisoSetCntl(Mat F,PetscInt icntl,PetscInt ival) 1016 { 1017 PetscFunctionBegin; 1018 PetscTryMethod(F,"MatMkl_PardisoSetCntl_C",(Mat,PetscInt,PetscInt),(F,icntl,ival)); 1019 PetscFunctionReturn(0); 1020 } 1021 1022 /*MC 1023 MATSOLVERMKL_PARDISO - A matrix type providing direct solvers (LU) for 1024 sequential matrices via the external package MKL_PARDISO. 1025 1026 Works with MATSEQAIJ matrices 1027 1028 Use -pc_type lu -pc_factor_mat_solver_type mkl_pardiso to use this direct solver 1029 1030 Options Database Keys: 1031 + -mat_mkl_pardiso_65 - Suggested number of threads to use within MKL_PARDISO 1032 . -mat_mkl_pardiso_66 - Maximum number of factors with identical sparsity structure that must be kept in memory at the same time 1033 . -mat_mkl_pardiso_67 - Indicates the actual matrix for the solution phase 1034 . -mat_mkl_pardiso_68 - Message level information, use 1 to get detailed information on the solver options 1035 . -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 1036 . -mat_mkl_pardiso_1 - Use default values 1037 . -mat_mkl_pardiso_2 - Fill-in reducing ordering for the input matrix 1038 . -mat_mkl_pardiso_4 - Preconditioned CGS/CG 1039 . -mat_mkl_pardiso_5 - User permutation 1040 . -mat_mkl_pardiso_6 - Write solution on x 1041 . -mat_mkl_pardiso_8 - Iterative refinement step 1042 . -mat_mkl_pardiso_10 - Pivoting perturbation 1043 . -mat_mkl_pardiso_11 - Scaling vectors 1044 . -mat_mkl_pardiso_12 - Solve with transposed or conjugate transposed matrix A 1045 . -mat_mkl_pardiso_13 - Improved accuracy using (non-) symmetric weighted matching 1046 . -mat_mkl_pardiso_18 - Numbers of non-zero elements 1047 . -mat_mkl_pardiso_19 - Report number of floating point operations 1048 . -mat_mkl_pardiso_21 - Pivoting for symmetric indefinite matrices 1049 . -mat_mkl_pardiso_24 - Parallel factorization control 1050 . -mat_mkl_pardiso_25 - Parallel forward/backward solve control 1051 . -mat_mkl_pardiso_27 - Matrix checker 1052 . -mat_mkl_pardiso_31 - Partial solve and computing selected components of the solution vectors 1053 . -mat_mkl_pardiso_34 - Optimal number of threads for conditional numerical reproducibility (CNR) mode 1054 - -mat_mkl_pardiso_60 - Intel MKL_PARDISO mode 1055 1056 Level: beginner 1057 1058 Notes: 1059 Use -mat_mkl_pardiso_68 1 to display the number of threads the solver is using. MKL does not provide a way to directly access this 1060 information. 1061 1062 For more information on the options check the MKL_Pardiso manual 1063 1064 .seealso: `PCFactorSetMatSolverType()`, `MatSolverType` 1065 1066 M*/ 1067 static PetscErrorCode MatFactorGetSolverType_mkl_pardiso(Mat A, MatSolverType *type) 1068 { 1069 PetscFunctionBegin; 1070 *type = MATSOLVERMKL_PARDISO; 1071 PetscFunctionReturn(0); 1072 } 1073 1074 PETSC_EXTERN PetscErrorCode MatGetFactor_aij_mkl_pardiso(Mat A,MatFactorType ftype,Mat *F) 1075 { 1076 Mat B; 1077 Mat_MKL_PARDISO *mat_mkl_pardiso; 1078 PetscBool isSeqAIJ,isSeqBAIJ,isSeqSBAIJ; 1079 1080 PetscFunctionBegin; 1081 PetscCall(PetscObjectBaseTypeCompare((PetscObject)A,MATSEQAIJ,&isSeqAIJ)); 1082 PetscCall(PetscObjectTypeCompare((PetscObject)A,MATSEQBAIJ,&isSeqBAIJ)); 1083 PetscCall(PetscObjectTypeCompare((PetscObject)A,MATSEQSBAIJ,&isSeqSBAIJ)); 1084 PetscCall(MatCreate(PetscObjectComm((PetscObject)A),&B)); 1085 PetscCall(MatSetSizes(B,A->rmap->n,A->cmap->n,A->rmap->N,A->cmap->N)); 1086 PetscCall(PetscStrallocpy("mkl_pardiso",&((PetscObject)B)->type_name)); 1087 PetscCall(MatSetUp(B)); 1088 1089 PetscCall(PetscNewLog(B,&mat_mkl_pardiso)); 1090 B->data = mat_mkl_pardiso; 1091 1092 PetscCall(MatFactorMKL_PARDISOInitialize_Private(A, ftype, mat_mkl_pardiso)); 1093 if (ftype == MAT_FACTOR_LU) { 1094 B->ops->lufactorsymbolic = MatLUFactorSymbolic_AIJMKL_PARDISO; 1095 B->factortype = MAT_FACTOR_LU; 1096 mat_mkl_pardiso->needsym = PETSC_FALSE; 1097 if (isSeqAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqaij; 1098 else if (isSeqBAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqbaij; 1099 else { 1100 PetscCheck(!isSeqSBAIJ,PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO LU factor with SEQSBAIJ format! Use MAT_FACTOR_CHOLESKY instead"); 1101 SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO LU with %s format",((PetscObject)A)->type_name); 1102 } 1103 #if defined(PETSC_USE_COMPLEX) 1104 mat_mkl_pardiso->mtype = 13; 1105 #else 1106 mat_mkl_pardiso->mtype = 11; 1107 #endif 1108 } else { 1109 B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_AIJMKL_PARDISO; 1110 B->factortype = MAT_FACTOR_CHOLESKY; 1111 if (isSeqAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqaij; 1112 else if (isSeqBAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqbaij; 1113 else if (isSeqSBAIJ) mat_mkl_pardiso->Convert = MatMKLPardiso_Convert_seqsbaij; 1114 else SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO CHOLESKY with %s format",((PetscObject)A)->type_name); 1115 1116 mat_mkl_pardiso->needsym = PETSC_TRUE; 1117 #if !defined(PETSC_USE_COMPLEX) 1118 if (A->spd == PETSC_BOOL3_TRUE) mat_mkl_pardiso->mtype = 2; 1119 else mat_mkl_pardiso->mtype = -2; 1120 #else 1121 mat_mkl_pardiso->mtype = 6; 1122 PetscCheck(A->hermitian != PETSC_BOOL3_TRUE,PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"No support for PARDISO CHOLESKY with Hermitian matrices! Use MAT_FACTOR_LU instead"); 1123 #endif 1124 } 1125 B->ops->destroy = MatDestroy_MKL_PARDISO; 1126 B->ops->view = MatView_MKL_PARDISO; 1127 B->ops->getinfo = MatGetInfo_MKL_PARDISO; 1128 B->factortype = ftype; 1129 B->assembled = PETSC_TRUE; 1130 1131 PetscCall(PetscFree(B->solvertype)); 1132 PetscCall(PetscStrallocpy(MATSOLVERMKL_PARDISO,&B->solvertype)); 1133 1134 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverType_C",MatFactorGetSolverType_mkl_pardiso)); 1135 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorSetSchurIS_C",MatFactorSetSchurIS_MKL_PARDISO)); 1136 PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatMkl_PardisoSetCntl_C",MatMkl_PardisoSetCntl_MKL_PARDISO)); 1137 1138 *F = B; 1139 PetscFunctionReturn(0); 1140 } 1141 1142 PETSC_EXTERN PetscErrorCode MatSolverTypeRegister_MKL_Pardiso(void) 1143 { 1144 PetscFunctionBegin; 1145 PetscCall(MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQAIJ,MAT_FACTOR_LU,MatGetFactor_aij_mkl_pardiso)); 1146 PetscCall(MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_aij_mkl_pardiso)); 1147 PetscCall(MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQBAIJ,MAT_FACTOR_LU,MatGetFactor_aij_mkl_pardiso)); 1148 PetscCall(MatSolverTypeRegister(MATSOLVERMKL_PARDISO,MATSEQSBAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_aij_mkl_pardiso)); 1149 PetscFunctionReturn(0); 1150 } 1151