1 2 /* 3 This file implements a subclass of the SeqAIJ matrix class that uses 4 the SuperLU sparse solver. 5 */ 6 7 /* 8 Defines the data structure for the base matrix type (SeqAIJ) 9 */ 10 #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/ 11 12 /* 13 SuperLU include files 14 */ 15 EXTERN_C_BEGIN 16 #if defined(PETSC_USE_COMPLEX) 17 #if defined(PETSC_USE_REAL_SINGLE) 18 #include <slu_cdefs.h> 19 #else 20 #include <slu_zdefs.h> 21 #endif 22 #else 23 #if defined(PETSC_USE_REAL_SINGLE) 24 #include <slu_sdefs.h> 25 #else 26 #include <slu_ddefs.h> 27 #endif 28 #endif 29 #include <slu_util.h> 30 EXTERN_C_END 31 32 /* 33 This is the data that defines the SuperLU factored matrix type 34 */ 35 typedef struct { 36 SuperMatrix A, L, U, B, X; 37 superlu_options_t options; 38 PetscInt *perm_c; /* column permutation vector */ 39 PetscInt *perm_r; /* row permutations from partial pivoting */ 40 PetscInt *etree; 41 PetscReal *R, *C; 42 char equed[1]; 43 PetscInt lwork; 44 void *work; 45 PetscReal rpg, rcond; 46 mem_usage_t mem_usage; 47 MatStructure flg; 48 SuperLUStat_t stat; 49 Mat A_dup; 50 PetscScalar *rhs_dup; 51 GlobalLU_t Glu; 52 PetscBool needconversion; 53 54 /* Flag to clean up (non-global) SuperLU objects during Destroy */ 55 PetscBool CleanUpSuperLU; 56 } Mat_SuperLU; 57 58 /* 59 Utility function 60 */ 61 static PetscErrorCode MatView_Info_SuperLU(Mat A, PetscViewer viewer) 62 { 63 Mat_SuperLU *lu = (Mat_SuperLU *)A->data; 64 superlu_options_t options; 65 66 PetscFunctionBegin; 67 options = lu->options; 68 69 PetscCall(PetscViewerASCIIPrintf(viewer, "SuperLU run parameters:\n")); 70 PetscCall(PetscViewerASCIIPrintf(viewer, " Equil: %s\n", (options.Equil != NO) ? "YES" : "NO")); 71 PetscCall(PetscViewerASCIIPrintf(viewer, " ColPerm: %" PetscInt_FMT "\n", options.ColPerm)); 72 PetscCall(PetscViewerASCIIPrintf(viewer, " IterRefine: %" PetscInt_FMT "\n", options.IterRefine)); 73 PetscCall(PetscViewerASCIIPrintf(viewer, " SymmetricMode: %s\n", (options.SymmetricMode != NO) ? "YES" : "NO")); 74 PetscCall(PetscViewerASCIIPrintf(viewer, " DiagPivotThresh: %g\n", options.DiagPivotThresh)); 75 PetscCall(PetscViewerASCIIPrintf(viewer, " PivotGrowth: %s\n", (options.PivotGrowth != NO) ? "YES" : "NO")); 76 PetscCall(PetscViewerASCIIPrintf(viewer, " ConditionNumber: %s\n", (options.ConditionNumber != NO) ? "YES" : "NO")); 77 PetscCall(PetscViewerASCIIPrintf(viewer, " RowPerm: %" PetscInt_FMT "\n", options.RowPerm)); 78 PetscCall(PetscViewerASCIIPrintf(viewer, " ReplaceTinyPivot: %s\n", (options.ReplaceTinyPivot != NO) ? "YES" : "NO")); 79 PetscCall(PetscViewerASCIIPrintf(viewer, " PrintStat: %s\n", (options.PrintStat != NO) ? "YES" : "NO")); 80 PetscCall(PetscViewerASCIIPrintf(viewer, " lwork: %" PetscInt_FMT "\n", lu->lwork)); 81 if (A->factortype == MAT_FACTOR_ILU) { 82 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_DropTol: %g\n", options.ILU_DropTol)); 83 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_FillTol: %g\n", options.ILU_FillTol)); 84 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_FillFactor: %g\n", options.ILU_FillFactor)); 85 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_DropRule: %" PetscInt_FMT "\n", options.ILU_DropRule)); 86 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_Norm: %" PetscInt_FMT "\n", options.ILU_Norm)); 87 PetscCall(PetscViewerASCIIPrintf(viewer, " ILU_MILU: %" PetscInt_FMT "\n", options.ILU_MILU)); 88 } 89 PetscFunctionReturn(PETSC_SUCCESS); 90 } 91 92 PetscErrorCode MatSolve_SuperLU_Private(Mat A, Vec b, Vec x) 93 { 94 Mat_SuperLU *lu = (Mat_SuperLU *)A->data; 95 const PetscScalar *barray; 96 PetscScalar *xarray; 97 PetscInt info, i, n; 98 PetscReal ferr, berr; 99 static PetscBool cite = PETSC_FALSE; 100 101 PetscFunctionBegin; 102 if (lu->lwork == -1) PetscFunctionReturn(PETSC_SUCCESS); 103 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 " 104 "pivoting},\n journal = {SIAM J. Matrix Analysis and Applications},\n year = {1999},\n volume = {20},\n number = {3},\n pages = {720-755}\n}\n", 105 &cite)); 106 107 PetscCall(VecGetLocalSize(x, &n)); 108 lu->B.ncol = 1; /* Set the number of right-hand side */ 109 if (lu->options.Equil && !lu->rhs_dup) { 110 /* superlu overwrites b when Equil is used, thus create rhs_dup to keep user's b unchanged */ 111 PetscCall(PetscMalloc1(n, &lu->rhs_dup)); 112 } 113 if (lu->options.Equil) { 114 /* Copy b into rsh_dup */ 115 PetscCall(VecGetArrayRead(b, &barray)); 116 PetscCall(PetscArraycpy(lu->rhs_dup, barray, n)); 117 PetscCall(VecRestoreArrayRead(b, &barray)); 118 barray = lu->rhs_dup; 119 } else { 120 PetscCall(VecGetArrayRead(b, &barray)); 121 } 122 PetscCall(VecGetArray(x, &xarray)); 123 124 #if defined(PETSC_USE_COMPLEX) 125 #if defined(PETSC_USE_REAL_SINGLE) 126 ((DNformat *)lu->B.Store)->nzval = (singlecomplex *)barray; 127 ((DNformat *)lu->X.Store)->nzval = (singlecomplex *)xarray; 128 #else 129 ((DNformat *)lu->B.Store)->nzval = (doublecomplex *)barray; 130 ((DNformat *)lu->X.Store)->nzval = (doublecomplex *)xarray; 131 #endif 132 #else 133 ((DNformat *)lu->B.Store)->nzval = (void *)barray; 134 ((DNformat *)lu->X.Store)->nzval = xarray; 135 #endif 136 137 lu->options.Fact = FACTORED; /* Indicate the factored form of A is supplied. */ 138 if (A->factortype == MAT_FACTOR_LU) { 139 #if defined(PETSC_USE_COMPLEX) 140 #if defined(PETSC_USE_REAL_SINGLE) 141 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)); 142 #else 143 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)); 144 #endif 145 #else 146 #if defined(PETSC_USE_REAL_SINGLE) 147 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)); 148 #else 149 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)); 150 #endif 151 #endif 152 } else if (A->factortype == MAT_FACTOR_ILU) { 153 #if defined(PETSC_USE_COMPLEX) 154 #if defined(PETSC_USE_REAL_SINGLE) 155 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)); 156 #else 157 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)); 158 #endif 159 #else 160 #if defined(PETSC_USE_REAL_SINGLE) 161 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)); 162 #else 163 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)); 164 #endif 165 #endif 166 } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Factor type not supported"); 167 if (!lu->options.Equil) PetscCall(VecRestoreArrayRead(b, &barray)); 168 PetscCall(VecRestoreArray(x, &xarray)); 169 170 if (!info || info == lu->A.ncol + 1) { 171 if (lu->options.IterRefine) { 172 PetscCall(PetscPrintf(PETSC_COMM_SELF, "Iterative Refinement:\n")); 173 PetscCall(PetscPrintf(PETSC_COMM_SELF, " %8s%8s%16s%16s\n", "rhs", "Steps", "FERR", "BERR")); 174 for (i = 0; i < 1; ++i) PetscCall(PetscPrintf(PETSC_COMM_SELF, " %8d%8d%16e%16e\n", i + 1, lu->stat.RefineSteps, ferr, berr)); 175 } 176 } else if (info > 0) { 177 if (lu->lwork == -1) { 178 PetscCall(PetscPrintf(PETSC_COMM_SELF, " ** Estimated memory: %" PetscInt_FMT " bytes\n", info - lu->A.ncol)); 179 } else { 180 PetscCall(PetscPrintf(PETSC_COMM_SELF, " Warning: gssvx() returns info %" PetscInt_FMT "\n", info)); 181 } 182 } 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); 183 184 if (lu->options.PrintStat) { 185 PetscCall(PetscPrintf(PETSC_COMM_SELF, "MatSolve__SuperLU():\n")); 186 PetscStackCallExternalVoid("SuperLU:StatPrint", StatPrint(&lu->stat)); 187 } 188 PetscFunctionReturn(PETSC_SUCCESS); 189 } 190 191 PetscErrorCode MatSolve_SuperLU(Mat A, Vec b, Vec x) 192 { 193 Mat_SuperLU *lu = (Mat_SuperLU *)A->data; 194 trans_t oldOption; 195 196 PetscFunctionBegin; 197 if (A->factorerrortype) { 198 PetscCall(PetscInfo(A, "MatSolve is called with singular matrix factor, skip\n")); 199 PetscCall(VecSetInf(x)); 200 PetscFunctionReturn(PETSC_SUCCESS); 201 } 202 203 oldOption = lu->options.Trans; 204 lu->options.Trans = TRANS; 205 PetscCall(MatSolve_SuperLU_Private(A, b, x)); 206 lu->options.Trans = oldOption; 207 PetscFunctionReturn(PETSC_SUCCESS); 208 } 209 210 PetscErrorCode MatSolveTranspose_SuperLU(Mat A, Vec b, Vec x) 211 { 212 Mat_SuperLU *lu = (Mat_SuperLU *)A->data; 213 trans_t oldOption; 214 215 PetscFunctionBegin; 216 if (A->factorerrortype) { 217 PetscCall(PetscInfo(A, "MatSolve is called with singular matrix factor, skip\n")); 218 PetscCall(VecSetInf(x)); 219 PetscFunctionReturn(PETSC_SUCCESS); 220 } 221 222 oldOption = lu->options.Trans; 223 lu->options.Trans = NOTRANS; 224 PetscCall(MatSolve_SuperLU_Private(A, b, x)); 225 lu->options.Trans = oldOption; 226 PetscFunctionReturn(PETSC_SUCCESS); 227 } 228 229 static PetscErrorCode MatLUFactorNumeric_SuperLU(Mat F, Mat A, const MatFactorInfo *info) 230 { 231 Mat_SuperLU *lu = (Mat_SuperLU *)F->data; 232 Mat_SeqAIJ *aa; 233 PetscInt sinfo; 234 PetscReal ferr, berr; 235 NCformat *Ustore; 236 SCformat *Lstore; 237 238 PetscFunctionBegin; 239 if (lu->flg == SAME_NONZERO_PATTERN) { /* successive numerical factorization */ 240 lu->options.Fact = SamePattern; 241 /* Ref: ~SuperLU_3.0/EXAMPLE/dlinsolx2.c */ 242 Destroy_SuperMatrix_Store(&lu->A); 243 if (lu->A_dup) PetscCall(MatCopy_SeqAIJ(A, lu->A_dup, SAME_NONZERO_PATTERN)); 244 245 if (lu->lwork >= 0) { 246 PetscStackCallExternalVoid("SuperLU:Destroy_SuperNode_Matrix", Destroy_SuperNode_Matrix(&lu->L)); 247 PetscStackCallExternalVoid("SuperLU:Destroy_CompCol_Matrix", Destroy_CompCol_Matrix(&lu->U)); 248 lu->options.Fact = SamePattern; 249 } 250 } 251 252 /* Create the SuperMatrix for lu->A=A^T: 253 Since SuperLU likes column-oriented matrices,we pass it the transpose, 254 and then solve A^T X = B in MatSolve(). */ 255 if (lu->A_dup) { 256 aa = (Mat_SeqAIJ *)(lu->A_dup)->data; 257 } else { 258 aa = (Mat_SeqAIJ *)(A)->data; 259 } 260 #if defined(PETSC_USE_COMPLEX) 261 #if defined(PETSC_USE_REAL_SINGLE) 262 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)); 263 #else 264 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)); 265 #endif 266 #else 267 #if defined(PETSC_USE_REAL_SINGLE) 268 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)); 269 #else 270 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)); 271 #endif 272 #endif 273 274 /* Numerical factorization */ 275 lu->B.ncol = 0; /* Indicate not to solve the system */ 276 if (F->factortype == MAT_FACTOR_LU) { 277 #if defined(PETSC_USE_COMPLEX) 278 #if defined(PETSC_USE_REAL_SINGLE) 279 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)); 280 #else 281 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)); 282 #endif 283 #else 284 #if defined(PETSC_USE_REAL_SINGLE) 285 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)); 286 #else 287 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)); 288 #endif 289 #endif 290 } else if (F->factortype == MAT_FACTOR_ILU) { 291 /* Compute the incomplete factorization, condition number and pivot growth */ 292 #if defined(PETSC_USE_COMPLEX) 293 #if defined(PETSC_USE_REAL_SINGLE) 294 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)); 295 #else 296 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)); 297 #endif 298 #else 299 #if defined(PETSC_USE_REAL_SINGLE) 300 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)); 301 #else 302 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)); 303 #endif 304 #endif 305 } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Factor type not supported"); 306 if (!sinfo || sinfo == lu->A.ncol + 1) { 307 if (lu->options.PivotGrowth) PetscCall(PetscPrintf(PETSC_COMM_SELF, " Recip. pivot growth = %e\n", lu->rpg)); 308 if (lu->options.ConditionNumber) PetscCall(PetscPrintf(PETSC_COMM_SELF, " Recip. condition number = %e\n", lu->rcond)); 309 } else if (sinfo > 0) { 310 if (A->erroriffailure) { 311 SETERRQ(PETSC_COMM_SELF, PETSC_ERR_MAT_LU_ZRPVT, "Zero pivot in row %" PetscInt_FMT, sinfo); 312 } else { 313 if (sinfo <= lu->A.ncol) { 314 if (lu->options.ILU_FillTol == 0.0) F->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 315 PetscCall(PetscInfo(F, "Number of zero pivots %" PetscInt_FMT ", ILU_FillTol %g\n", sinfo, lu->options.ILU_FillTol)); 316 } else if (sinfo == lu->A.ncol + 1) { 317 /* 318 U is nonsingular, but RCOND is less than machine 319 precision, meaning that the matrix is singular to 320 working precision. Nevertheless, the solution and 321 error bounds are computed because there are a number 322 of situations where the computed solution can be more 323 accurate than the value of RCOND would suggest. 324 */ 325 PetscCall(PetscInfo(F, "Matrix factor U is nonsingular, but is singular to working precision. The solution is computed. info %" PetscInt_FMT, sinfo)); 326 } else { /* sinfo > lu->A.ncol + 1 */ 327 F->factorerrortype = MAT_FACTOR_OUTMEMORY; 328 PetscCall(PetscInfo(F, "Number of bytes allocated when memory allocation fails %" PetscInt_FMT "\n", sinfo)); 329 } 330 } 331 } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_LIB, "info = %" PetscInt_FMT ", the %" PetscInt_FMT "-th argument in gssvx() had an illegal value", sinfo, -sinfo); 332 333 if (lu->options.PrintStat) { 334 PetscCall(PetscPrintf(PETSC_COMM_SELF, "MatLUFactorNumeric_SuperLU():\n")); 335 PetscStackCallExternalVoid("SuperLU:StatPrint", StatPrint(&lu->stat)); 336 Lstore = (SCformat *)lu->L.Store; 337 Ustore = (NCformat *)lu->U.Store; 338 PetscCall(PetscPrintf(PETSC_COMM_SELF, " No of nonzeros in factor L = %" PetscInt_FMT "\n", Lstore->nnz)); 339 PetscCall(PetscPrintf(PETSC_COMM_SELF, " No of nonzeros in factor U = %" PetscInt_FMT "\n", Ustore->nnz)); 340 PetscCall(PetscPrintf(PETSC_COMM_SELF, " No of nonzeros in L+U = %" PetscInt_FMT "\n", Lstore->nnz + Ustore->nnz - lu->A.ncol)); 341 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)); 342 } 343 344 lu->flg = SAME_NONZERO_PATTERN; 345 F->ops->solve = MatSolve_SuperLU; 346 F->ops->solvetranspose = MatSolveTranspose_SuperLU; 347 F->ops->matsolve = NULL; 348 PetscFunctionReturn(PETSC_SUCCESS); 349 } 350 351 static PetscErrorCode MatDestroy_SuperLU(Mat A) 352 { 353 Mat_SuperLU *lu = (Mat_SuperLU *)A->data; 354 355 PetscFunctionBegin; 356 if (lu->CleanUpSuperLU) { /* Free the SuperLU datastructures */ 357 PetscStackCallExternalVoid("SuperLU:Destroy_SuperMatrix_Store", Destroy_SuperMatrix_Store(&lu->A)); 358 PetscStackCallExternalVoid("SuperLU:Destroy_SuperMatrix_Store", Destroy_SuperMatrix_Store(&lu->B)); 359 PetscStackCallExternalVoid("SuperLU:Destroy_SuperMatrix_Store", Destroy_SuperMatrix_Store(&lu->X)); 360 PetscStackCallExternalVoid("SuperLU:StatFree", StatFree(&lu->stat)); 361 if (lu->lwork >= 0) { 362 PetscStackCallExternalVoid("SuperLU:Destroy_SuperNode_Matrix", Destroy_SuperNode_Matrix(&lu->L)); 363 PetscStackCallExternalVoid("SuperLU:Destroy_CompCol_Matrix", Destroy_CompCol_Matrix(&lu->U)); 364 } 365 } 366 PetscCall(PetscFree(lu->etree)); 367 PetscCall(PetscFree(lu->perm_r)); 368 PetscCall(PetscFree(lu->perm_c)); 369 PetscCall(PetscFree(lu->R)); 370 PetscCall(PetscFree(lu->C)); 371 PetscCall(PetscFree(lu->rhs_dup)); 372 PetscCall(MatDestroy(&lu->A_dup)); 373 PetscCall(PetscFree(A->data)); 374 375 /* clear composed functions */ 376 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatFactorGetSolverType_C", NULL)); 377 PetscCall(PetscObjectComposeFunction((PetscObject)A, "MatSuperluSetILUDropTol_C", NULL)); 378 PetscFunctionReturn(PETSC_SUCCESS); 379 } 380 381 static PetscErrorCode MatView_SuperLU(Mat A, PetscViewer viewer) 382 { 383 PetscBool iascii; 384 PetscViewerFormat format; 385 386 PetscFunctionBegin; 387 PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii)); 388 if (iascii) { 389 PetscCall(PetscViewerGetFormat(viewer, &format)); 390 if (format == PETSC_VIEWER_ASCII_INFO) PetscCall(MatView_Info_SuperLU(A, viewer)); 391 } 392 PetscFunctionReturn(PETSC_SUCCESS); 393 } 394 395 PetscErrorCode MatMatSolve_SuperLU(Mat A, Mat B, Mat X) 396 { 397 Mat_SuperLU *lu = (Mat_SuperLU *)A->data; 398 PetscBool flg; 399 400 PetscFunctionBegin; 401 PetscCall(PetscObjectTypeCompareAny((PetscObject)B, &flg, MATSEQDENSE, MATMPIDENSE, NULL)); 402 PetscCheck(flg, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Matrix B must be MATDENSE matrix"); 403 PetscCall(PetscObjectTypeCompareAny((PetscObject)X, &flg, MATSEQDENSE, MATMPIDENSE, NULL)); 404 PetscCheck(flg, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Matrix X must be MATDENSE matrix"); 405 lu->options.Trans = TRANS; 406 SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "MatMatSolve_SuperLU() is not implemented yet"); 407 PetscFunctionReturn(PETSC_SUCCESS); 408 } 409 410 static PetscErrorCode MatLUFactorSymbolic_SuperLU(Mat F, Mat A, IS r, IS c, const MatFactorInfo *info) 411 { 412 Mat_SuperLU *lu = (Mat_SuperLU *)(F->data); 413 PetscInt indx; 414 PetscBool flg, set; 415 PetscReal real_input; 416 const char *colperm[] = {"NATURAL", "MMD_ATA", "MMD_AT_PLUS_A", "COLAMD"}; /* MY_PERMC - not supported by the petsc interface yet */ 417 const char *iterrefine[] = {"NOREFINE", "SINGLE", "DOUBLE", "EXTRA"}; 418 const char *rowperm[] = {"NOROWPERM", "LargeDiag"}; /* MY_PERMC - not supported by the petsc interface yet */ 419 420 PetscFunctionBegin; 421 /* Set options to F */ 422 PetscOptionsBegin(PetscObjectComm((PetscObject)F), ((PetscObject)F)->prefix, "SuperLU Options", "Mat"); 423 PetscCall(PetscOptionsBool("-mat_superlu_equil", "Equil", "None", (PetscBool)lu->options.Equil, (PetscBool *)&lu->options.Equil, NULL)); 424 PetscCall(PetscOptionsEList("-mat_superlu_colperm", "ColPerm", "None", colperm, 4, colperm[3], &indx, &flg)); 425 if (flg) lu->options.ColPerm = (colperm_t)indx; 426 PetscCall(PetscOptionsEList("-mat_superlu_iterrefine", "IterRefine", "None", iterrefine, 4, iterrefine[0], &indx, &flg)); 427 if (flg) lu->options.IterRefine = (IterRefine_t)indx; 428 PetscCall(PetscOptionsBool("-mat_superlu_symmetricmode", "SymmetricMode", "None", (PetscBool)lu->options.SymmetricMode, &flg, &set)); 429 if (set && flg) lu->options.SymmetricMode = YES; 430 PetscCall(PetscOptionsReal("-mat_superlu_diagpivotthresh", "DiagPivotThresh", "None", lu->options.DiagPivotThresh, &real_input, &flg)); 431 if (flg) lu->options.DiagPivotThresh = (double)real_input; 432 PetscCall(PetscOptionsBool("-mat_superlu_pivotgrowth", "PivotGrowth", "None", (PetscBool)lu->options.PivotGrowth, &flg, &set)); 433 if (set && flg) lu->options.PivotGrowth = YES; 434 PetscCall(PetscOptionsBool("-mat_superlu_conditionnumber", "ConditionNumber", "None", (PetscBool)lu->options.ConditionNumber, &flg, &set)); 435 if (set && flg) lu->options.ConditionNumber = YES; 436 PetscCall(PetscOptionsEList("-mat_superlu_rowperm", "rowperm", "None", rowperm, 2, rowperm[lu->options.RowPerm], &indx, &flg)); 437 if (flg) lu->options.RowPerm = (rowperm_t)indx; 438 PetscCall(PetscOptionsBool("-mat_superlu_replacetinypivot", "ReplaceTinyPivot", "None", (PetscBool)lu->options.ReplaceTinyPivot, &flg, &set)); 439 if (set && flg) lu->options.ReplaceTinyPivot = YES; 440 PetscCall(PetscOptionsBool("-mat_superlu_printstat", "PrintStat", "None", (PetscBool)lu->options.PrintStat, &flg, &set)); 441 if (set && flg) lu->options.PrintStat = YES; 442 PetscCall(PetscOptionsInt("-mat_superlu_lwork", "size of work array in bytes used by factorization", "None", lu->lwork, &lu->lwork, NULL)); 443 if (lu->lwork > 0) { 444 /* lwork is in bytes, hence PetscMalloc() is used here, not PetscMalloc1()*/ 445 PetscCall(PetscMalloc(lu->lwork, &lu->work)); 446 } else if (lu->lwork != 0 && lu->lwork != -1) { 447 PetscCall(PetscPrintf(PETSC_COMM_SELF, " Warning: lwork %" PetscInt_FMT " is not supported by SUPERLU. The default lwork=0 is used.\n", lu->lwork)); 448 lu->lwork = 0; 449 } 450 /* ilu options */ 451 PetscCall(PetscOptionsReal("-mat_superlu_ilu_droptol", "ILU_DropTol", "None", lu->options.ILU_DropTol, &real_input, &flg)); 452 if (flg) lu->options.ILU_DropTol = (double)real_input; 453 PetscCall(PetscOptionsReal("-mat_superlu_ilu_filltol", "ILU_FillTol", "None", lu->options.ILU_FillTol, &real_input, &flg)); 454 if (flg) lu->options.ILU_FillTol = (double)real_input; 455 PetscCall(PetscOptionsReal("-mat_superlu_ilu_fillfactor", "ILU_FillFactor", "None", lu->options.ILU_FillFactor, &real_input, &flg)); 456 if (flg) lu->options.ILU_FillFactor = (double)real_input; 457 PetscCall(PetscOptionsInt("-mat_superlu_ilu_droprull", "ILU_DropRule", "None", lu->options.ILU_DropRule, &lu->options.ILU_DropRule, NULL)); 458 PetscCall(PetscOptionsInt("-mat_superlu_ilu_norm", "ILU_Norm", "None", lu->options.ILU_Norm, &indx, &flg)); 459 if (flg) lu->options.ILU_Norm = (norm_t)indx; 460 PetscCall(PetscOptionsInt("-mat_superlu_ilu_milu", "ILU_MILU", "None", lu->options.ILU_MILU, &indx, &flg)); 461 if (flg) lu->options.ILU_MILU = (milu_t)indx; 462 PetscOptionsEnd(); 463 464 lu->flg = DIFFERENT_NONZERO_PATTERN; 465 lu->CleanUpSuperLU = PETSC_TRUE; 466 F->ops->lufactornumeric = MatLUFactorNumeric_SuperLU; 467 468 /* if we are here, the nonzero pattern has changed unless the user explicitly called MatLUFactorSymbolic */ 469 PetscCall(MatDestroy(&lu->A_dup)); 470 if (lu->needconversion) PetscCall(MatConvert(A, MATSEQAIJ, MAT_INITIAL_MATRIX, &lu->A_dup)); 471 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 */ 472 PetscCall(MatDuplicate_SeqAIJ(A, MAT_COPY_VALUES, &lu->A_dup)); 473 } 474 PetscFunctionReturn(PETSC_SUCCESS); 475 } 476 477 static PetscErrorCode MatSuperluSetILUDropTol_SuperLU(Mat F, PetscReal dtol) 478 { 479 Mat_SuperLU *lu = (Mat_SuperLU *)F->data; 480 481 PetscFunctionBegin; 482 lu->options.ILU_DropTol = dtol; 483 PetscFunctionReturn(PETSC_SUCCESS); 484 } 485 486 /*@ 487 MatSuperluSetILUDropTol - Set SuperLU ILU drop tolerance 488 489 Logically Collective 490 491 Input Parameters: 492 + F - the factored matrix obtained by calling `MatGetFactor()` 493 - dtol - drop tolerance 494 495 Options Database Key: 496 . -mat_superlu_ilu_droptol <dtol> - the drop tolerance 497 498 Level: beginner 499 500 References: 501 . * - SuperLU Users' Guide 502 503 .seealso: [](chapter_matrices), `Mat`, `MatGetFactor()`, `MATSOLVERSUPERLU` 504 @*/ 505 PetscErrorCode MatSuperluSetILUDropTol(Mat F, PetscReal dtol) 506 { 507 PetscFunctionBegin; 508 PetscValidHeaderSpecific(F, MAT_CLASSID, 1); 509 PetscValidLogicalCollectiveReal(F, dtol, 2); 510 PetscTryMethod(F, "MatSuperluSetILUDropTol_C", (Mat, PetscReal), (F, dtol)); 511 PetscFunctionReturn(PETSC_SUCCESS); 512 } 513 514 PetscErrorCode MatFactorGetSolverType_seqaij_superlu(Mat A, MatSolverType *type) 515 { 516 PetscFunctionBegin; 517 *type = MATSOLVERSUPERLU; 518 PetscFunctionReturn(PETSC_SUCCESS); 519 } 520 521 /*MC 522 MATSOLVERSUPERLU = "superlu" - A solver package providing solvers LU and ILU for sequential matrices 523 via the external package SuperLU. 524 525 Use `./configure --download-superlu` to have PETSc installed with SuperLU 526 527 Use `-pc_type lu` `-pc_factor_mat_solver_type superlu` to use this direct solver 528 529 Options Database Keys: 530 + -mat_superlu_equil <FALSE> - Equil (None) 531 . -mat_superlu_colperm <COLAMD> - (choose one of) `NATURAL`, `MMD_ATA MMD_AT_PLUS_A`, `COLAMD` 532 . -mat_superlu_iterrefine <NOREFINE> - (choose one of) `NOREFINE`, `SINGLE`, `DOUBLE`, `EXTRA` 533 . -mat_superlu_symmetricmode: <FALSE> - SymmetricMode (None) 534 . -mat_superlu_diagpivotthresh <1> - DiagPivotThresh (None) 535 . -mat_superlu_pivotgrowth <FALSE> - PivotGrowth (None) 536 . -mat_superlu_conditionnumber <FALSE> - ConditionNumber (None) 537 . -mat_superlu_rowperm <NOROWPERM> - (choose one of) `NOROWPERM`, `LargeDiag` 538 . -mat_superlu_replacetinypivot <FALSE> - ReplaceTinyPivot (None) 539 . -mat_superlu_printstat <FALSE> - PrintStat (None) 540 . -mat_superlu_lwork <0> - size of work array in bytes used by factorization (None) 541 . -mat_superlu_ilu_droptol <0> - ILU_DropTol (None) 542 . -mat_superlu_ilu_filltol <0> - ILU_FillTol (None) 543 . -mat_superlu_ilu_fillfactor <0> - ILU_FillFactor (None) 544 . -mat_superlu_ilu_droprull <0> - ILU_DropRule (None) 545 . -mat_superlu_ilu_norm <0> - ILU_Norm (None) 546 - -mat_superlu_ilu_milu <0> - ILU_MILU (None) 547 548 Level: beginner 549 550 Notes: 551 Do not confuse this with `MATSOLVERSUPERLU_DIST` which is for parallel sparse solves 552 553 Cannot use ordering provided by PETSc, provides its own. 554 555 .seealso: [](chapter_matrices), `Mat`, `PCLU`, `PCILU`, `MATSOLVERSUPERLU_DIST`, `MATSOLVERMUMPS`, `PCFactorSetMatSolverType()`, `MatSolverType` 556 M*/ 557 558 static PetscErrorCode MatGetFactor_seqaij_superlu(Mat A, MatFactorType ftype, Mat *F) 559 { 560 Mat B; 561 Mat_SuperLU *lu; 562 PetscInt m = A->rmap->n, n = A->cmap->n; 563 564 PetscFunctionBegin; 565 PetscCall(MatCreate(PetscObjectComm((PetscObject)A), &B)); 566 PetscCall(MatSetSizes(B, A->rmap->n, A->cmap->n, PETSC_DETERMINE, PETSC_DETERMINE)); 567 PetscCall(PetscStrallocpy("superlu", &((PetscObject)B)->type_name)); 568 PetscCall(MatSetUp(B)); 569 B->trivialsymbolic = PETSC_TRUE; 570 if (ftype == MAT_FACTOR_LU || ftype == MAT_FACTOR_ILU) { 571 B->ops->lufactorsymbolic = MatLUFactorSymbolic_SuperLU; 572 B->ops->ilufactorsymbolic = MatLUFactorSymbolic_SuperLU; 573 } else SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Factor type not supported"); 574 575 PetscCall(PetscFree(B->solvertype)); 576 PetscCall(PetscStrallocpy(MATSOLVERSUPERLU, &B->solvertype)); 577 578 B->ops->getinfo = MatGetInfo_External; 579 B->ops->destroy = MatDestroy_SuperLU; 580 B->ops->view = MatView_SuperLU; 581 B->factortype = ftype; 582 B->assembled = PETSC_TRUE; /* required by -ksp_view */ 583 B->preallocated = PETSC_TRUE; 584 585 PetscCall(PetscNew(&lu)); 586 587 if (ftype == MAT_FACTOR_LU) { 588 set_default_options(&lu->options); 589 /* Comments from SuperLU_4.0/SRC/dgssvx.c: 590 "Whether or not the system will be equilibrated depends on the 591 scaling of the matrix A, but if equilibration is used, A is 592 overwritten by diag(R)*A*diag(C) and B by diag(R)*B 593 (if options->Trans=NOTRANS) or diag(C)*B (if options->Trans = TRANS or CONJ)." 594 We set 'options.Equil = NO' as default because additional space is needed for it. 595 */ 596 lu->options.Equil = NO; 597 } else if (ftype == MAT_FACTOR_ILU) { 598 /* Set the default input options of ilu: */ 599 PetscStackCallExternalVoid("SuperLU:ilu_set_default_options", ilu_set_default_options(&lu->options)); 600 } 601 lu->options.PrintStat = NO; 602 603 /* Initialize the statistics variables. */ 604 PetscStackCallExternalVoid("SuperLU:StatInit", StatInit(&lu->stat)); 605 lu->lwork = 0; /* allocate space internally by system malloc */ 606 607 /* Allocate spaces (notice sizes are for the transpose) */ 608 PetscCall(PetscMalloc1(m, &lu->etree)); 609 PetscCall(PetscMalloc1(n, &lu->perm_r)); 610 PetscCall(PetscMalloc1(m, &lu->perm_c)); 611 PetscCall(PetscMalloc1(n, &lu->R)); 612 PetscCall(PetscMalloc1(m, &lu->C)); 613 614 /* create rhs and solution x without allocate space for .Store */ 615 #if defined(PETSC_USE_COMPLEX) 616 #if defined(PETSC_USE_REAL_SINGLE) 617 PetscStackCallExternalVoid("SuperLU:cCreate_Dense_Matrix(", cCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE)); 618 PetscStackCallExternalVoid("SuperLU:cCreate_Dense_Matrix(", cCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE)); 619 #else 620 PetscStackCallExternalVoid("SuperLU:zCreate_Dense_Matrix", zCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE)); 621 PetscStackCallExternalVoid("SuperLU:zCreate_Dense_Matrix", zCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE)); 622 #endif 623 #else 624 #if defined(PETSC_USE_REAL_SINGLE) 625 PetscStackCallExternalVoid("SuperLU:sCreate_Dense_Matrix", sCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE)); 626 PetscStackCallExternalVoid("SuperLU:sCreate_Dense_Matrix", sCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE)); 627 #else 628 PetscStackCallExternalVoid("SuperLU:dCreate_Dense_Matrix", dCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE)); 629 PetscStackCallExternalVoid("SuperLU:dCreate_Dense_Matrix", dCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE)); 630 #endif 631 #endif 632 633 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatFactorGetSolverType_C", MatFactorGetSolverType_seqaij_superlu)); 634 PetscCall(PetscObjectComposeFunction((PetscObject)B, "MatSuperluSetILUDropTol_C", MatSuperluSetILUDropTol_SuperLU)); 635 B->data = lu; 636 637 *F = B; 638 PetscFunctionReturn(PETSC_SUCCESS); 639 } 640 641 static PetscErrorCode MatGetFactor_seqsell_superlu(Mat A, MatFactorType ftype, Mat *F) 642 { 643 Mat_SuperLU *lu; 644 645 PetscFunctionBegin; 646 PetscCall(MatGetFactor_seqaij_superlu(A, ftype, F)); 647 lu = (Mat_SuperLU *)((*F)->data); 648 lu->needconversion = PETSC_TRUE; 649 PetscFunctionReturn(PETSC_SUCCESS); 650 } 651 652 PETSC_EXTERN PetscErrorCode MatSolverTypeRegister_SuperLU(void) 653 { 654 PetscFunctionBegin; 655 PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQAIJ, MAT_FACTOR_LU, MatGetFactor_seqaij_superlu)); 656 PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQAIJ, MAT_FACTOR_ILU, MatGetFactor_seqaij_superlu)); 657 PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQSELL, MAT_FACTOR_LU, MatGetFactor_seqsell_superlu)); 658 PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU, MATSEQSELL, MAT_FACTOR_ILU, MatGetFactor_seqsell_superlu)); 659 PetscFunctionReturn(PETSC_SUCCESS); 660 } 661