1 /* 2 This file implements a subclass of the SeqAIJ matrix class that uses 3 the SuperLU sparse solver. 4 */ 5 6 /* 7 Defines the data structure for the base matrix type (SeqAIJ) 8 */ 9 #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/ 10 11 /* 12 SuperLU include files 13 */ 14 EXTERN_C_BEGIN 15 #if defined(PETSC_USE_COMPLEX) 16 #if defined(PETSC_USE_REAL_SINGLE) 17 #include <slu_cdefs.h> 18 #else 19 #include <slu_zdefs.h> 20 #endif 21 #else 22 #if defined(PETSC_USE_REAL_SINGLE) 23 #include <slu_sdefs.h> 24 #else 25 #include <slu_ddefs.h> 26 #endif 27 #endif 28 #include <slu_util.h> 29 EXTERN_C_END 30 31 /* 32 This is the data that defines the SuperLU factored matrix type 33 */ 34 typedef struct { 35 SuperMatrix A, L, U, B, X; 36 superlu_options_t options; 37 PetscInt *perm_c; /* column permutation vector */ 38 PetscInt *perm_r; /* row permutations from partial pivoting */ 39 PetscInt *etree; 40 PetscReal *R, *C; 41 char equed[1]; 42 PetscInt lwork; 43 void *work; 44 PetscReal rpg, rcond; 45 mem_usage_t mem_usage; 46 MatStructure flg; 47 SuperLUStat_t stat; 48 Mat A_dup; 49 PetscScalar *rhs_dup; 50 GlobalLU_t Glu; 51 PetscBool needconversion; 52 53 /* Flag to clean up (non-global) SuperLU objects during Destroy */ 54 PetscBool CleanUpSuperLU; 55 } Mat_SuperLU; 56 57 /* 58 Utility function 59 */ 60 static PetscErrorCode MatView_Info_SuperLU(Mat A, PetscViewer viewer) 61 { 62 Mat_SuperLU *lu = (Mat_SuperLU *)A->data; 63 superlu_options_t options; 64 65 PetscFunctionBegin; 66 options = lu->options; 67 68 PetscCall(PetscViewerASCIIPrintf(viewer, "SuperLU run parameters:\n")); 69 PetscCall(PetscViewerASCIIPrintf(viewer, " Equil: %s\n", (options.Equil != NO) ? "YES" : "NO")); 70 PetscCall(PetscViewerASCIIPrintf(viewer, " ColPerm: %" PetscInt_FMT "\n", options.ColPerm)); 71 PetscCall(PetscViewerASCIIPrintf(viewer, " IterRefine: %" PetscInt_FMT "\n", options.IterRefine)); 72 PetscCall(PetscViewerASCIIPrintf(viewer, " SymmetricMode: %s\n", (options.SymmetricMode != NO) ? "YES" : "NO")); 73 PetscCall(PetscViewerASCIIPrintf(viewer, " DiagPivotThresh: %g\n", options.DiagPivotThresh)); 74 PetscCall(PetscViewerASCIIPrintf(viewer, " PivotGrowth: %s\n", (options.PivotGrowth != NO) ? "YES" : "NO")); 75 PetscCall(PetscViewerASCIIPrintf(viewer, " ConditionNumber: %s\n", (options.ConditionNumber != NO) ? "YES" : "NO")); 76 PetscCall(PetscViewerASCIIPrintf(viewer, " RowPerm: %" PetscInt_FMT "\n", options.RowPerm)); 77 PetscCall(PetscViewerASCIIPrintf(viewer, " ReplaceTinyPivot: %s\n", (options.ReplaceTinyPivot != NO) ? "YES" : "NO")); 78 PetscCall(PetscViewerASCIIPrintf(viewer, " PrintStat: %s\n", (options.PrintStat != NO) ? "YES" : "NO")); 79 PetscCall(PetscViewerASCIIPrintf(viewer, " lwork: %" PetscInt_FMT "\n", lu->lwork)); 80 if (A->factortype == MAT_FACTOR_ILU) { 81 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_DropTol: %g\n", options.ILU_DropTol)); 82 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_FillTol: %g\n", options.ILU_FillTol)); 83 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_FillFactor: %g\n", options.ILU_FillFactor)); 84 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_DropRule: %" PetscInt_FMT "\n", options.ILU_DropRule)); 85 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_Norm: %" PetscInt_FMT "\n", options.ILU_Norm)); 86 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_MILU: %" PetscInt_FMT "\n", options.ILU_MILU)); 87 } 88 PetscFunctionReturn(PETSC_SUCCESS); 89 } 90 91 static PetscErrorCode MatSolve_SuperLU_Private(Mat A, Vec b, Vec x) 92 { 93 Mat_SuperLU *lu = (Mat_SuperLU *)A->data; 94 const PetscScalar *barray; 95 PetscScalar *xarray; 96 PetscInt info, i, n; 97 PetscReal ferr, berr; 98 static PetscBool cite = PETSC_FALSE; 99 100 PetscFunctionBegin; 101 if (lu->lwork == -1) PetscFunctionReturn(PETSC_SUCCESS); 102 PetscCall(PetscCitationsRegister("@article{superlu99,\n author = {James W. Demmel and Stanley C. Eisenstat and\n John R. Gilbert and Xiaoye S. Li and Joseph W. H. Liu},\n title = {A supernodal approach to sparse partial " 103 "pivoting},\n journal = {SIAM J. Matrix Analysis and Applications},\n year = {1999},\n volume = {20},\n number = {3},\n pages = {720-755}\n}\n", 104 &cite)); 105 106 PetscCall(VecGetLocalSize(x, &n)); 107 lu->B.ncol = 1; /* Set the number of right-hand side */ 108 if (lu->options.Equil && !lu->rhs_dup) { 109 /* superlu overwrites b when Equil is used, thus create rhs_dup to keep user's b unchanged */ 110 PetscCall(PetscMalloc1(n, &lu->rhs_dup)); 111 } 112 if (lu->options.Equil) { 113 /* Copy b into rsh_dup */ 114 PetscCall(VecGetArrayRead(b, &barray)); 115 PetscCall(PetscArraycpy(lu->rhs_dup, barray, n)); 116 PetscCall(VecRestoreArrayRead(b, &barray)); 117 barray = lu->rhs_dup; 118 } else { 119 PetscCall(VecGetArrayRead(b, &barray)); 120 } 121 PetscCall(VecGetArray(x, &xarray)); 122 123 #if defined(PETSC_USE_COMPLEX) 124 #if defined(PETSC_USE_REAL_SINGLE) 125 ((DNformat *)lu->B.Store)->nzval = (singlecomplex *)barray; 126 ((DNformat *)lu->X.Store)->nzval = (singlecomplex *)xarray; 127 #else 128 ((DNformat *)lu->B.Store)->nzval = (doublecomplex *)barray; 129 ((DNformat *)lu->X.Store)->nzval = (doublecomplex *)xarray; 130 #endif 131 #else 132 ((DNformat *)lu->B.Store)->nzval = (void *)barray; 133 ((DNformat *)lu->X.Store)->nzval = xarray; 134 #endif 135 136 lu->options.Fact = FACTORED; /* Indicate the factored form of A is supplied. */ 137 if (A->factortype == MAT_FACTOR_LU) { 138 #if defined(PETSC_USE_COMPLEX) 139 #if defined(PETSC_USE_REAL_SINGLE) 140 PetscStackCallExternalVoid("SuperLU:cgssvx", cgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 141 #else 142 PetscStackCallExternalVoid("SuperLU:zgssvx", zgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 143 #endif 144 #else 145 #if defined(PETSC_USE_REAL_SINGLE) 146 PetscStackCallExternalVoid("SuperLU:sgssvx", sgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 147 #else 148 PetscStackCallExternalVoid("SuperLU:dgssvx", dgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 149 #endif 150 #endif 151 } else if (A->factortype == MAT_FACTOR_ILU) { 152 #if defined(PETSC_USE_COMPLEX) 153 #if defined(PETSC_USE_REAL_SINGLE) 154 PetscStackCallExternalVoid("SuperLU:cgsisx", cgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 155 #else 156 PetscStackCallExternalVoid("SuperLU:zgsisx", zgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 157 #endif 158 #else 159 #if defined(PETSC_USE_REAL_SINGLE) 160 PetscStackCallExternalVoid("SuperLU:sgsisx", sgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 161 #else 162 PetscStackCallExternalVoid("SuperLU:dgsisx", dgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 163 #endif 164 #endif 165 } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Factor type not supported"); 166 if (!lu->options.Equil) PetscCall(VecRestoreArrayRead(b, &barray)); 167 PetscCall(VecRestoreArray(x, &xarray)); 168 169 if (!info || info == lu->A.ncol + 1) { 170 if (lu->options.IterRefine) { 171 PetscCall(PetscPrintf(PETSC_COMM_SELF, "Iterative Refinement:\n")); 172 PetscCall(PetscPrintf(PETSC_COMM_SELF, " %8s%8s%16s%16s\n", "rhs", "Steps", "FERR", "BERR")); 173 for (i = 0; i < 1; ++i) PetscCall(PetscPrintf(PETSC_COMM_SELF, " %8d%8d%16e%16e\n", i + 1, lu->stat.RefineSteps, ferr, berr)); 174 } 175 } else if (info > 0) { 176 if (lu->lwork == -1) { 177 PetscCall(PetscPrintf(PETSC_COMM_SELF, " ** Estimated memory: %" PetscInt_FMT " bytes\n", info - lu->A.ncol)); 178 } else { 179 PetscCall(PetscPrintf(PETSC_COMM_SELF, " Warning: gssvx() returns info %" PetscInt_FMT "\n", info)); 180 } 181 } else PetscCheck(info >= 0, PETSC_COMM_SELF, PETSC_ERR_LIB, "info = %" PetscInt_FMT ", the %" PetscInt_FMT "-th argument in gssvx() had an illegal value", info, -info); 182 183 if (lu->options.PrintStat) { 184 PetscCall(PetscPrintf(PETSC_COMM_SELF, "MatSolve__SuperLU():\n")); 185 PetscStackCallExternalVoid("SuperLU:StatPrint", StatPrint(&lu->stat)); 186 } 187 PetscFunctionReturn(PETSC_SUCCESS); 188 } 189 190 static PetscErrorCode MatSolve_SuperLU(Mat A, Vec b, Vec x) 191 { 192 Mat_SuperLU *lu = (Mat_SuperLU *)A->data; 193 trans_t oldOption; 194 195 PetscFunctionBegin; 196 if (A->factorerrortype) { 197 PetscCall(PetscInfo(A, "MatSolve is called with singular matrix factor, skip\n")); 198 PetscCall(VecSetInf(x)); 199 PetscFunctionReturn(PETSC_SUCCESS); 200 } 201 202 oldOption = lu->options.Trans; 203 lu->options.Trans = TRANS; 204 PetscCall(MatSolve_SuperLU_Private(A, b, x)); 205 lu->options.Trans = oldOption; 206 PetscFunctionReturn(PETSC_SUCCESS); 207 } 208 209 static PetscErrorCode MatSolveTranspose_SuperLU(Mat A, Vec b, Vec x) 210 { 211 Mat_SuperLU *lu = (Mat_SuperLU *)A->data; 212 trans_t oldOption; 213 214 PetscFunctionBegin; 215 if (A->factorerrortype) { 216 PetscCall(PetscInfo(A, "MatSolve is called with singular matrix factor, skip\n")); 217 PetscCall(VecSetInf(x)); 218 PetscFunctionReturn(PETSC_SUCCESS); 219 } 220 221 oldOption = lu->options.Trans; 222 lu->options.Trans = NOTRANS; 223 PetscCall(MatSolve_SuperLU_Private(A, b, x)); 224 lu->options.Trans = oldOption; 225 PetscFunctionReturn(PETSC_SUCCESS); 226 } 227 228 static PetscErrorCode MatLUFactorNumeric_SuperLU(Mat F, Mat A, const MatFactorInfo *info) 229 { 230 Mat_SuperLU *lu = (Mat_SuperLU *)F->data; 231 Mat_SeqAIJ *aa; 232 PetscInt sinfo; 233 PetscReal ferr, berr; 234 NCformat *Ustore; 235 SCformat *Lstore; 236 237 PetscFunctionBegin; 238 if (lu->flg == SAME_NONZERO_PATTERN) { /* successive numerical factorization */ 239 lu->options.Fact = SamePattern; 240 /* Ref: ~SuperLU_3.0/EXAMPLE/dlinsolx2.c */ 241 Destroy_SuperMatrix_Store(&lu->A); 242 if (lu->A_dup) PetscCall(MatCopy_SeqAIJ(A, lu->A_dup, SAME_NONZERO_PATTERN)); 243 244 if (lu->lwork >= 0) { 245 PetscStackCallExternalVoid("SuperLU:Destroy_SuperNode_Matrix", Destroy_SuperNode_Matrix(&lu->L)); 246 PetscStackCallExternalVoid("SuperLU:Destroy_CompCol_Matrix", Destroy_CompCol_Matrix(&lu->U)); 247 lu->options.Fact = SamePattern; 248 } 249 } 250 251 /* Create the SuperMatrix for lu->A=A^T: 252 Since SuperLU likes column-oriented matrices,we pass it the transpose, 253 and then solve A^T X = B in MatSolve(). */ 254 if (lu->A_dup) { 255 aa = (Mat_SeqAIJ *)(lu->A_dup)->data; 256 } else { 257 aa = (Mat_SeqAIJ *)(A)->data; 258 } 259 #if defined(PETSC_USE_COMPLEX) 260 #if defined(PETSC_USE_REAL_SINGLE) 261 PetscStackCallExternalVoid("SuperLU:cCreate_CompCol_Matrix", cCreate_CompCol_Matrix(&lu->A, A->cmap->n, A->rmap->n, aa->nz, (singlecomplex *)aa->a, aa->j, aa->i, SLU_NC, SLU_C, SLU_GE)); 262 #else 263 PetscStackCallExternalVoid("SuperLU:zCreate_CompCol_Matrix", zCreate_CompCol_Matrix(&lu->A, A->cmap->n, A->rmap->n, aa->nz, (doublecomplex *)aa->a, aa->j, aa->i, SLU_NC, SLU_Z, SLU_GE)); 264 #endif 265 #else 266 #if defined(PETSC_USE_REAL_SINGLE) 267 PetscStackCallExternalVoid("SuperLU:sCreate_CompCol_Matrix", sCreate_CompCol_Matrix(&lu->A, A->cmap->n, A->rmap->n, aa->nz, aa->a, aa->j, aa->i, SLU_NC, SLU_S, SLU_GE)); 268 #else 269 PetscStackCallExternalVoid("SuperLU:dCreate_CompCol_Matrix", dCreate_CompCol_Matrix(&lu->A, A->cmap->n, A->rmap->n, aa->nz, aa->a, aa->j, aa->i, SLU_NC, SLU_D, SLU_GE)); 270 #endif 271 #endif 272 273 /* Numerical factorization */ 274 lu->B.ncol = 0; /* Indicate not to solve the system */ 275 if (F->factortype == MAT_FACTOR_LU) { 276 #if defined(PETSC_USE_COMPLEX) 277 #if defined(PETSC_USE_REAL_SINGLE) 278 PetscStackCallExternalVoid("SuperLU:cgssvx", cgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 279 #else 280 PetscStackCallExternalVoid("SuperLU:zgssvx", zgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 281 #endif 282 #else 283 #if defined(PETSC_USE_REAL_SINGLE) 284 PetscStackCallExternalVoid("SuperLU:sgssvx", sgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 285 #else 286 PetscStackCallExternalVoid("SuperLU:dgssvx", dgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 287 #endif 288 #endif 289 } else if (F->factortype == MAT_FACTOR_ILU) { 290 /* Compute the incomplete factorization, condition number and pivot growth */ 291 #if defined(PETSC_USE_COMPLEX) 292 #if defined(PETSC_USE_REAL_SINGLE) 293 PetscStackCallExternalVoid("SuperLU:cgsisx", cgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 294 #else 295 PetscStackCallExternalVoid("SuperLU:zgsisx", zgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 296 #endif 297 #else 298 #if defined(PETSC_USE_REAL_SINGLE) 299 PetscStackCallExternalVoid("SuperLU:sgsisx", sgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 300 #else 301 PetscStackCallExternalVoid("SuperLU:dgsisx", dgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 302 #endif 303 #endif 304 } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Factor type not supported"); 305 if (!sinfo || sinfo == lu->A.ncol + 1) { 306 if (lu->options.PivotGrowth) PetscCall(PetscPrintf(PETSC_COMM_SELF, " Recip. pivot growth = %e\n", lu->rpg)); 307 if (lu->options.ConditionNumber) PetscCall(PetscPrintf(PETSC_COMM_SELF, " Recip. condition number = %e\n", lu->rcond)); 308 } else if (sinfo > 0) { 309 if (A->erroriffailure) { 310 SETERRQ(PETSC_COMM_SELF, PETSC_ERR_MAT_LU_ZRPVT, "Zero pivot in row %" PetscInt_FMT, sinfo); 311 } else { 312 if (sinfo <= lu->A.ncol) { 313 if (lu->options.ILU_FillTol == 0.0) F->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 314 PetscCall(PetscInfo(F, "Number of zero pivots %" PetscInt_FMT ", ILU_FillTol %g\n", sinfo, lu->options.ILU_FillTol)); 315 } else if (sinfo == lu->A.ncol + 1) { 316 /* 317 U is nonsingular, but RCOND is less than machine 318 precision, meaning that the matrix is singular to 319 working precision. Nevertheless, the solution and 320 error bounds are computed because there are a number 321 of situations where the computed solution can be more 322 accurate than the value of RCOND would suggest. 323 */ 324 PetscCall(PetscInfo(F, "Matrix factor U is nonsingular, but is singular to working precision. The solution is computed. info %" PetscInt_FMT "\n", sinfo)); 325 } else { /* sinfo > lu->A.ncol + 1 */ 326 F->factorerrortype = MAT_FACTOR_OUTMEMORY; 327 PetscCall(PetscInfo(F, "Number of bytes allocated when memory allocation fails %" PetscInt_FMT "\n", sinfo)); 328 } 329 } 330 } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_LIB, "info = %" PetscInt_FMT ", the %" PetscInt_FMT "-th argument in gssvx() had an illegal value", sinfo, -sinfo); 331 332 if (lu->options.PrintStat) { 333 PetscCall(PetscPrintf(PETSC_COMM_SELF, "MatLUFactorNumeric_SuperLU():\n")); 334 PetscStackCallExternalVoid("SuperLU:StatPrint", StatPrint(&lu->stat)); 335 Lstore = (SCformat *)lu->L.Store; 336 Ustore = (NCformat *)lu->U.Store; 337 PetscCall(PetscPrintf(PETSC_COMM_SELF, " No of nonzeros in factor L = %" PetscInt_FMT "\n", Lstore->nnz)); 338 PetscCall(PetscPrintf(PETSC_COMM_SELF, " No of nonzeros in factor U = %" PetscInt_FMT "\n", Ustore->nnz)); 339 PetscCall(PetscPrintf(PETSC_COMM_SELF, " No of nonzeros in L+U = %" PetscInt_FMT "\n", Lstore->nnz + Ustore->nnz - lu->A.ncol)); 340 PetscCall(PetscPrintf(PETSC_COMM_SELF, " L\\U MB %.3f\ttotal MB needed %.3f\n", lu->mem_usage.for_lu / 1e6, lu->mem_usage.total_needed / 1e6)); 341 } 342 343 lu->flg = SAME_NONZERO_PATTERN; 344 F->ops->solve = MatSolve_SuperLU; 345 F->ops->solvetranspose = MatSolveTranspose_SuperLU; 346 F->ops->matsolve = NULL; 347 PetscFunctionReturn(PETSC_SUCCESS); 348 } 349 350 static PetscErrorCode MatDestroy_SuperLU(Mat A) 351 { 352 Mat_SuperLU *lu = (Mat_SuperLU *)A->data; 353 354 PetscFunctionBegin; 355 if (lu->CleanUpSuperLU) { /* Free the SuperLU datastructures */ 356 PetscStackCallExternalVoid("SuperLU:Destroy_SuperMatrix_Store", Destroy_SuperMatrix_Store(&lu->A)); 357 if (lu->lwork >= 0) { 358 PetscStackCallExternalVoid("SuperLU:Destroy_SuperNode_Matrix", Destroy_SuperNode_Matrix(&lu->L)); 359 PetscStackCallExternalVoid("SuperLU:Destroy_CompCol_Matrix", Destroy_CompCol_Matrix(&lu->U)); 360 } 361 } 362 PetscStackCallExternalVoid("SuperLU:Destroy_SuperMatrix_Store", Destroy_SuperMatrix_Store(&lu->B)); 363 PetscStackCallExternalVoid("SuperLU:Destroy_SuperMatrix_Store", Destroy_SuperMatrix_Store(&lu->X)); 364 PetscStackCallExternalVoid("SuperLU:StatFree", StatFree(&lu->stat)); 365 PetscCall(PetscFree(lu->etree)); 366 PetscCall(PetscFree(lu->perm_r)); 367 PetscCall(PetscFree(lu->perm_c)); 368 PetscCall(PetscFree(lu->R)); 369 PetscCall(PetscFree(lu->C)); 370 PetscCall(PetscFree(lu->rhs_dup)); 371 PetscCall(MatDestroy(&lu->A_dup)); 372 PetscCall(PetscFree(A->data)); 373 374 /* clear composed functions */ 375 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatFactorGetSolverType_C", NULL)); 376 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSuperluSetILUDropTol_C", NULL)); 377 PetscFunctionReturn(PETSC_SUCCESS); 378 } 379 380 static PetscErrorCode MatView_SuperLU(Mat A, PetscViewer viewer) 381 { 382 PetscBool iascii; 383 PetscViewerFormat format; 384 385 PetscFunctionBegin; 386 PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii)); 387 if (iascii) { 388 PetscCall(PetscViewerGetFormat(viewer, &format)); 389 if (format == PETSC_VIEWER_ASCII_INFO) PetscCall(MatView_Info_SuperLU(A, viewer)); 390 } 391 PetscFunctionReturn(PETSC_SUCCESS); 392 } 393 394 static PetscErrorCode MatLUFactorSymbolic_SuperLU(Mat F, Mat A, IS r, IS c, const MatFactorInfo *info) 395 { 396 Mat_SuperLU *lu = (Mat_SuperLU *)(F->data); 397 PetscInt indx; 398 PetscBool flg, set; 399 PetscReal real_input; 400 const char *colperm[] = {"NATURAL", "MMD_ATA", "MMD_AT_PLUS_A", "COLAMD"}; /* MY_PERMC - not supported by the petsc interface yet */ 401 const char *iterrefine[] = {"NOREFINE", "SINGLE", "DOUBLE", "EXTRA"}; 402 const char *rowperm[] = {"NOROWPERM", "LargeDiag"}; /* MY_PERMC - not supported by the petsc interface yet */ 403 404 PetscFunctionBegin; 405 /* Set options to F */ 406 PetscOptionsBegin(PetscObjectComm((PetscObject)F), ((PetscObject)F)->prefix, "SuperLU Options", "Mat"); 407 PetscCall(PetscOptionsBool("-mat_superlu_equil", "Equil", "None", (PetscBool)lu->options.Equil, (PetscBool *)&lu->options.Equil, NULL)); 408 PetscCall(PetscOptionsEList("-mat_superlu_colperm", "ColPerm", "None", colperm, 4, colperm[3], &indx, &flg)); 409 if (flg) lu->options.ColPerm = (colperm_t)indx; 410 PetscCall(PetscOptionsEList("-mat_superlu_iterrefine", "IterRefine", "None", iterrefine, 4, iterrefine[0], &indx, &flg)); 411 if (flg) lu->options.IterRefine = (IterRefine_t)indx; 412 PetscCall(PetscOptionsBool("-mat_superlu_symmetricmode", "SymmetricMode", "None", (PetscBool)lu->options.SymmetricMode, &flg, &set)); 413 if (set && flg) lu->options.SymmetricMode = YES; 414 PetscCall(PetscOptionsReal("-mat_superlu_diagpivotthresh", "DiagPivotThresh", "None", lu->options.DiagPivotThresh, &real_input, &flg)); 415 if (flg) lu->options.DiagPivotThresh = (double)real_input; 416 PetscCall(PetscOptionsBool("-mat_superlu_pivotgrowth", "PivotGrowth", "None", (PetscBool)lu->options.PivotGrowth, &flg, &set)); 417 if (set && flg) lu->options.PivotGrowth = YES; 418 PetscCall(PetscOptionsBool("-mat_superlu_conditionnumber", "ConditionNumber", "None", (PetscBool)lu->options.ConditionNumber, &flg, &set)); 419 if (set && flg) lu->options.ConditionNumber = YES; 420 PetscCall(PetscOptionsEList("-mat_superlu_rowperm", "rowperm", "None", rowperm, 2, rowperm[lu->options.RowPerm], &indx, &flg)); 421 if (flg) lu->options.RowPerm = (rowperm_t)indx; 422 PetscCall(PetscOptionsBool("-mat_superlu_replacetinypivot", "ReplaceTinyPivot", "None", (PetscBool)lu->options.ReplaceTinyPivot, &flg, &set)); 423 if (set && flg) lu->options.ReplaceTinyPivot = YES; 424 PetscCall(PetscOptionsBool("-mat_superlu_printstat", "PrintStat", "None", (PetscBool)lu->options.PrintStat, &flg, &set)); 425 if (set && flg) lu->options.PrintStat = YES; 426 PetscCall(PetscOptionsInt("-mat_superlu_lwork", "size of work array in bytes used by factorization", "None", lu->lwork, &lu->lwork, NULL)); 427 if (lu->lwork > 0) { 428 /* lwork is in bytes, hence PetscMalloc() is used here, not PetscMalloc1()*/ 429 PetscCall(PetscMalloc(lu->lwork, &lu->work)); 430 } else if (lu->lwork != 0 && lu->lwork != -1) { 431 PetscCall(PetscPrintf(PETSC_COMM_SELF, " Warning: lwork %" PetscInt_FMT " is not supported by SUPERLU. The default lwork=0 is used.\n", lu->lwork)); 432 lu->lwork = 0; 433 } 434 /* ilu options */ 435 PetscCall(PetscOptionsReal("-mat_superlu_ilu_droptol", "ILU_DropTol", "None", lu->options.ILU_DropTol, &real_input, &flg)); 436 if (flg) lu->options.ILU_DropTol = (double)real_input; 437 PetscCall(PetscOptionsReal("-mat_superlu_ilu_filltol", "ILU_FillTol", "None", lu->options.ILU_FillTol, &real_input, &flg)); 438 if (flg) lu->options.ILU_FillTol = (double)real_input; 439 PetscCall(PetscOptionsReal("-mat_superlu_ilu_fillfactor", "ILU_FillFactor", "None", lu->options.ILU_FillFactor, &real_input, &flg)); 440 if (flg) lu->options.ILU_FillFactor = (double)real_input; 441 PetscCall(PetscOptionsInt("-mat_superlu_ilu_droprull", "ILU_DropRule", "None", lu->options.ILU_DropRule, &lu->options.ILU_DropRule, NULL)); 442 PetscCall(PetscOptionsInt("-mat_superlu_ilu_norm", "ILU_Norm", "None", lu->options.ILU_Norm, &indx, &flg)); 443 if (flg) lu->options.ILU_Norm = (norm_t)indx; 444 PetscCall(PetscOptionsInt("-mat_superlu_ilu_milu", "ILU_MILU", "None", lu->options.ILU_MILU, &indx, &flg)); 445 if (flg) lu->options.ILU_MILU = (milu_t)indx; 446 PetscOptionsEnd(); 447 448 lu->flg = DIFFERENT_NONZERO_PATTERN; 449 lu->CleanUpSuperLU = PETSC_TRUE; 450 F->ops->lufactornumeric = MatLUFactorNumeric_SuperLU; 451 452 /* if we are here, the nonzero pattern has changed unless the user explicitly called MatLUFactorSymbolic */ 453 PetscCall(MatDestroy(&lu->A_dup)); 454 if (lu->needconversion) PetscCall(MatConvert(A, MATSEQAIJ, MAT_INITIAL_MATRIX, &lu->A_dup)); 455 if (lu->options.Equil == YES && !lu->A_dup) { /* superlu overwrites input matrix and rhs when Equil is used, thus create A_dup to keep user's A unchanged */ 456 PetscCall(MatDuplicate_SeqAIJ(A, MAT_COPY_VALUES, &lu->A_dup)); 457 } 458 PetscFunctionReturn(PETSC_SUCCESS); 459 } 460 461 static PetscErrorCode MatSuperluSetILUDropTol_SuperLU(Mat F, PetscReal dtol) 462 { 463 Mat_SuperLU *lu = (Mat_SuperLU *)F->data; 464 465 PetscFunctionBegin; 466 lu->options.ILU_DropTol = dtol; 467 PetscFunctionReturn(PETSC_SUCCESS); 468 } 469 470 /*@ 471 MatSuperluSetILUDropTol - Set SuperLU ILU drop tolerance 472 473 Logically Collective 474 475 Input Parameters: 476 + F - the factored matrix obtained by calling `MatGetFactor()` 477 - dtol - drop tolerance 478 479 Options Database Key: 480 . -mat_superlu_ilu_droptol <dtol> - the drop tolerance 481 482 Level: beginner 483 484 References: 485 . * - SuperLU Users' Guide 486 487 .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MATSOLVERSUPERLU` 488 @*/ 489 PetscErrorCode MatSuperluSetILUDropTol(Mat F, PetscReal dtol) 490 { 491 PetscFunctionBegin; 492 PetscValidHeaderSpecific(F, MAT_CLASSID, 1); 493 PetscValidLogicalCollectiveReal(F, dtol, 2); 494 PetscTryMethod(F, "MatSuperluSetILUDropTol_C", (Mat, PetscReal), (F, dtol)); 495 PetscFunctionReturn(PETSC_SUCCESS); 496 } 497 498 static PetscErrorCode MatFactorGetSolverType_seqaij_superlu(Mat A, MatSolverType *type) 499 { 500 PetscFunctionBegin; 501 *type = MATSOLVERSUPERLU; 502 PetscFunctionReturn(PETSC_SUCCESS); 503 } 504 505 /*MC 506 MATSOLVERSUPERLU = "superlu" - A solver package providing solvers LU and ILU for sequential matrices 507 via the external package SuperLU. 508 509 Use `./configure --download-superlu` to have PETSc installed with SuperLU 510 511 Use `-pc_type lu` `-pc_factor_mat_solver_type superlu` to use this direct solver 512 513 Options Database Keys: 514 + -mat_superlu_equil <FALSE> - Equil (None) 515 . -mat_superlu_colperm <COLAMD> - (choose one of) `NATURAL`, `MMD_ATA MMD_AT_PLUS_A`, `COLAMD` 516 . -mat_superlu_iterrefine <NOREFINE> - (choose one of) `NOREFINE`, `SINGLE`, `DOUBLE`, `EXTRA` 517 . -mat_superlu_symmetricmode: <FALSE> - SymmetricMode (None) 518 . -mat_superlu_diagpivotthresh <1> - DiagPivotThresh (None) 519 . -mat_superlu_pivotgrowth <FALSE> - PivotGrowth (None) 520 . -mat_superlu_conditionnumber <FALSE> - ConditionNumber (None) 521 . -mat_superlu_rowperm <NOROWPERM> - (choose one of) `NOROWPERM`, `LargeDiag` 522 . -mat_superlu_replacetinypivot <FALSE> - ReplaceTinyPivot (None) 523 . -mat_superlu_printstat <FALSE> - PrintStat (None) 524 . -mat_superlu_lwork <0> - size of work array in bytes used by factorization (None) 525 . -mat_superlu_ilu_droptol <0> - ILU_DropTol (None) 526 . -mat_superlu_ilu_filltol <0> - ILU_FillTol (None) 527 . -mat_superlu_ilu_fillfactor <0> - ILU_FillFactor (None) 528 . -mat_superlu_ilu_droprull <0> - ILU_DropRule (None) 529 . -mat_superlu_ilu_norm <0> - ILU_Norm (None) 530 - -mat_superlu_ilu_milu <0> - ILU_MILU (None) 531 532 Level: beginner 533 534 Notes: 535 Do not confuse this with `MATSOLVERSUPERLU_DIST` which is for parallel sparse solves 536 537 Cannot use ordering provided by PETSc, provides its own. 538 539 .seealso: [](ch_matrices), `Mat`, `PCLU`, `PCILU`, `MATSOLVERSUPERLU_DIST`, `MATSOLVERMUMPS`, `PCFactorSetMatSolverType()`, `MatSolverType` 540 M*/ 541 542 static PetscErrorCode MatGetFactor_seqaij_superlu(Mat A, MatFactorType ftype, Mat *F) 543 { 544 Mat B; 545 Mat_SuperLU *lu; 546 PetscInt m = A->rmap->n, n = A->cmap->n; 547 548 PetscFunctionBegin; 549 PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B)); 550 PetscCall(MatSetSizes(B, A->rmap->n, A->cmap->n, PETSC_DETERMINE, PETSC_DETERMINE)); 551 PetscCall(PetscStrallocpy("superlu", &((PetscObject)B)->type_name)); 552 PetscCall(MatSetUp(B)); 553 B->trivialsymbolic = PETSC_TRUE; 554 if (ftype == MAT_FACTOR_LU || ftype == MAT_FACTOR_ILU) { 555 B->ops->lufactorsymbolic = MatLUFactorSymbolic_SuperLU; 556 B->ops->ilufactorsymbolic = MatLUFactorSymbolic_SuperLU; 557 } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Factor type not supported"); 558 559 PetscCall(PetscFree(B->solvertype)); 560 PetscCall(PetscStrallocpy(MATSOLVERSUPERLU, &B->solvertype)); 561 562 B->ops->getinfo = MatGetInfo_External; 563 B->ops->destroy = MatDestroy_SuperLU; 564 B->ops->view = MatView_SuperLU; 565 B->factortype = ftype; 566 B->assembled = PETSC_TRUE; /* required by -ksp_view */ 567 B->preallocated = PETSC_TRUE; 568 569 PetscCall(PetscNew(&lu)); 570 571 if (ftype == MAT_FACTOR_LU) { 572 set_default_options(&lu->options); 573 /* Comments from SuperLU_4.0/SRC/dgssvx.c: 574 "Whether or not the system will be equilibrated depends on the 575 scaling of the matrix A, but if equilibration is used, A is 576 overwritten by diag(R)*A*diag(C) and B by diag(R)*B 577 (if options->Trans=NOTRANS) or diag(C)*B (if options->Trans = TRANS or CONJ)." 578 We set 'options.Equil = NO' as default because additional space is needed for it. 579 */ 580 lu->options.Equil = NO; 581 } else if (ftype == MAT_FACTOR_ILU) { 582 /* Set the default input options of ilu: */ 583 PetscStackCallExternalVoid("SuperLU:ilu_set_default_options", ilu_set_default_options(&lu->options)); 584 } 585 lu->options.PrintStat = NO; 586 587 /* Initialize the statistics variables. */ 588 PetscStackCallExternalVoid("SuperLU:StatInit", StatInit(&lu->stat)); 589 lu->lwork = 0; /* allocate space internally by system malloc */ 590 591 /* Allocate spaces (notice sizes are for the transpose) */ 592 PetscCall(PetscMalloc1(m, &lu->etree)); 593 PetscCall(PetscMalloc1(n, &lu->perm_r)); 594 PetscCall(PetscMalloc1(m, &lu->perm_c)); 595 PetscCall(PetscMalloc1(n, &lu->R)); 596 PetscCall(PetscMalloc1(m, &lu->C)); 597 598 /* create rhs and solution x without allocate space for .Store */ 599 #if defined(PETSC_USE_COMPLEX) 600 #if defined(PETSC_USE_REAL_SINGLE) 601 PetscStackCallExternalVoid("SuperLU:cCreate_Dense_Matrix(", cCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE)); 602 PetscStackCallExternalVoid("SuperLU:cCreate_Dense_Matrix(", cCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE)); 603 #else 604 PetscStackCallExternalVoid("SuperLU:zCreate_Dense_Matrix", zCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE)); 605 PetscStackCallExternalVoid("SuperLU:zCreate_Dense_Matrix", zCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE)); 606 #endif 607 #else 608 #if defined(PETSC_USE_REAL_SINGLE) 609 PetscStackCallExternalVoid("SuperLU:sCreate_Dense_Matrix", sCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE)); 610 PetscStackCallExternalVoid("SuperLU:sCreate_Dense_Matrix", sCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE)); 611 #else 612 PetscStackCallExternalVoid("SuperLU:dCreate_Dense_Matrix", dCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE)); 613 PetscStackCallExternalVoid("SuperLU:dCreate_Dense_Matrix", dCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE)); 614 #endif 615 #endif 616 617 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatFactorGetSolverType_C", MatFactorGetSolverType_seqaij_superlu)); 618 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSuperluSetILUDropTol_C", MatSuperluSetILUDropTol_SuperLU)); 619 B->data = lu; 620 621 *F = B; 622 PetscFunctionReturn(PETSC_SUCCESS); 623 } 624 625 static PetscErrorCode MatGetFactor_seqsell_superlu(Mat A, MatFactorType ftype, Mat *F) 626 { 627 Mat_SuperLU *lu; 628 629 PetscFunctionBegin; 630 PetscCall(MatGetFactor_seqaij_superlu(A, ftype, F)); 631 lu = (Mat_SuperLU *)((*F)->data); 632 lu->needconversion = PETSC_TRUE; 633 PetscFunctionReturn(PETSC_SUCCESS); 634 } 635 636 PETSC_EXTERN PetscErrorCode MatSolverTypeRegister_SuperLU(void) 637 { 638 PetscFunctionBegin; 639 PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQAIJ, MAT_FACTOR_LU, MatGetFactor_seqaij_superlu)); 640 PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQAIJ, MAT_FACTOR_ILU, MatGetFactor_seqaij_superlu)); 641 PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQSELL, MAT_FACTOR_LU, MatGetFactor_seqsell_superlu)); 642 PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQSELL, MAT_FACTOR_ILU, MatGetFactor_seqsell_superlu)); 643 PetscFunctionReturn(PETSC_SUCCESS); 644 } 645