1 2 /* -------------------------------------------------------------------- 3 4 This file implements a subclass of the SeqAIJ matrix class that uses 5 the SuperLU sparse solver. 6 */ 7 8 /* 9 Defines the data structure for the base matrix type (SeqAIJ) 10 */ 11 #include <../src/mat/impls/aij/seq/aij.h> /*I "petscmat.h" I*/ 12 13 /* 14 SuperLU include files 15 */ 16 EXTERN_C_BEGIN 17 #if defined(PETSC_USE_COMPLEX) 18 #if defined(PETSC_USE_REAL_SINGLE) 19 #include <slu_cdefs.h> 20 #else 21 #include <slu_zdefs.h> 22 #endif 23 #else 24 #if defined(PETSC_USE_REAL_SINGLE) 25 #include <slu_sdefs.h> 26 #else 27 #include <slu_ddefs.h> 28 #endif 29 #endif 30 #include <slu_util.h> 31 EXTERN_C_END 32 33 /* 34 This is the data that defines the SuperLU factored matrix type 35 */ 36 typedef struct { 37 SuperMatrix A,L,U,B,X; 38 superlu_options_t options; 39 PetscInt *perm_c; /* column permutation vector */ 40 PetscInt *perm_r; /* row permutations from partial pivoting */ 41 PetscInt *etree; 42 PetscReal *R, *C; 43 char equed[1]; 44 PetscInt lwork; 45 void *work; 46 PetscReal rpg, rcond; 47 mem_usage_t mem_usage; 48 MatStructure flg; 49 SuperLUStat_t stat; 50 Mat A_dup; 51 PetscScalar *rhs_dup; 52 GlobalLU_t Glu; 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 MatFactorInfo_SuperLU(Mat A,PetscViewer viewer) 62 { 63 Mat_SuperLU *lu= (Mat_SuperLU*)A->data; 64 PetscErrorCode ierr; 65 superlu_options_t options; 66 67 PetscFunctionBegin; 68 options = lu->options; 69 70 ierr = PetscViewerASCIIPrintf(viewer,"SuperLU run parameters:\n");CHKERRQ(ierr); 71 ierr = PetscViewerASCIIPrintf(viewer," Equil: %s\n",(options.Equil != NO) ? "YES" : "NO");CHKERRQ(ierr); 72 ierr = PetscViewerASCIIPrintf(viewer," ColPerm: %D\n",options.ColPerm);CHKERRQ(ierr); 73 ierr = PetscViewerASCIIPrintf(viewer," IterRefine: %D\n",options.IterRefine);CHKERRQ(ierr); 74 ierr = PetscViewerASCIIPrintf(viewer," SymmetricMode: %s\n",(options.SymmetricMode != NO) ? "YES" : "NO");CHKERRQ(ierr); 75 ierr = PetscViewerASCIIPrintf(viewer," DiagPivotThresh: %g\n",options.DiagPivotThresh);CHKERRQ(ierr); 76 ierr = PetscViewerASCIIPrintf(viewer," PivotGrowth: %s\n",(options.PivotGrowth != NO) ? "YES" : "NO");CHKERRQ(ierr); 77 ierr = PetscViewerASCIIPrintf(viewer," ConditionNumber: %s\n",(options.ConditionNumber != NO) ? "YES" : "NO");CHKERRQ(ierr); 78 ierr = PetscViewerASCIIPrintf(viewer," RowPerm: %D\n",options.RowPerm);CHKERRQ(ierr); 79 ierr = PetscViewerASCIIPrintf(viewer," ReplaceTinyPivot: %s\n",(options.ReplaceTinyPivot != NO) ? "YES" : "NO");CHKERRQ(ierr); 80 ierr = PetscViewerASCIIPrintf(viewer," PrintStat: %s\n",(options.PrintStat != NO) ? "YES" : "NO");CHKERRQ(ierr); 81 ierr = PetscViewerASCIIPrintf(viewer," lwork: %D\n",lu->lwork);CHKERRQ(ierr); 82 if (A->factortype == MAT_FACTOR_ILU) { 83 ierr = PetscViewerASCIIPrintf(viewer," ILU_DropTol: %g\n",options.ILU_DropTol);CHKERRQ(ierr); 84 ierr = PetscViewerASCIIPrintf(viewer," ILU_FillTol: %g\n",options.ILU_FillTol);CHKERRQ(ierr); 85 ierr = PetscViewerASCIIPrintf(viewer," ILU_FillFactor: %g\n",options.ILU_FillFactor);CHKERRQ(ierr); 86 ierr = PetscViewerASCIIPrintf(viewer," ILU_DropRule: %D\n",options.ILU_DropRule);CHKERRQ(ierr); 87 ierr = PetscViewerASCIIPrintf(viewer," ILU_Norm: %D\n",options.ILU_Norm);CHKERRQ(ierr); 88 ierr = PetscViewerASCIIPrintf(viewer," ILU_MILU: %D\n",options.ILU_MILU);CHKERRQ(ierr); 89 } 90 PetscFunctionReturn(0); 91 } 92 93 PetscErrorCode MatSolve_SuperLU_Private(Mat A,Vec b,Vec x) 94 { 95 Mat_SuperLU *lu = (Mat_SuperLU*)A->data; 96 const PetscScalar *barray; 97 PetscScalar *xarray; 98 PetscErrorCode ierr; 99 PetscInt info,i,n; 100 PetscReal ferr,berr; 101 static PetscBool cite = PETSC_FALSE; 102 103 PetscFunctionBegin; 104 if (lu->lwork == -1) PetscFunctionReturn(0); 105 ierr = 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 pivoting},\n journal = {SIAM J. Matrix Analysis and Applications},\n year = {1999},\n volume = {20},\n number = {3},\n pages = {720-755}\n}\n",&cite);CHKERRQ(ierr); 106 107 ierr = VecGetLocalSize(x,&n);CHKERRQ(ierr); 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 ierr = PetscMalloc1(n,&lu->rhs_dup);CHKERRQ(ierr); 112 } 113 if (lu->options.Equil) { 114 /* Copy b into rsh_dup */ 115 ierr = VecGetArrayRead(b,&barray);CHKERRQ(ierr); 116 ierr = PetscMemcpy(lu->rhs_dup,barray,n*sizeof(PetscScalar));CHKERRQ(ierr); 117 ierr = VecRestoreArrayRead(b,&barray);CHKERRQ(ierr); 118 barray = lu->rhs_dup; 119 } else { 120 ierr = VecGetArrayRead(b,&barray);CHKERRQ(ierr); 121 } 122 ierr = VecGetArray(x,&xarray);CHKERRQ(ierr); 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 PetscStackCall("SuperLU:cgssvx",cgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 142 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, 143 &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 144 #else 145 PetscStackCall("SuperLU:zgssvx",zgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 146 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, 147 &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 148 #endif 149 #else 150 #if defined(PETSC_USE_REAL_SINGLE) 151 PetscStackCall("SuperLU:sgssvx",sgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 152 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, 153 &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 154 #else 155 PetscStackCall("SuperLU:dgssvx",dgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 156 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, 157 &lu->Glu,&lu->mem_usage, &lu->stat, &info)); 158 #endif 159 #endif 160 } else if (A->factortype == MAT_FACTOR_ILU) { 161 #if defined(PETSC_USE_COMPLEX) 162 #if defined(PETSC_USE_REAL_SINGLE) 163 PetscStackCall("SuperLU:cgsisx",cgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 164 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, 165 &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 166 #else 167 PetscStackCall("SuperLU:zgsisx",zgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 168 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, 169 &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 170 #endif 171 #else 172 #if defined(PETSC_USE_REAL_SINGLE) 173 PetscStackCall("SuperLU:sgsisx",sgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 174 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, 175 &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 176 #else 177 PetscStackCall("SuperLU:dgsisx",dgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 178 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, 179 &lu->Glu, &lu->mem_usage, &lu->stat, &info)); 180 #endif 181 #endif 182 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Factor type not supported"); 183 if (!lu->options.Equil) { 184 ierr = VecRestoreArrayRead(b,&barray);CHKERRQ(ierr); 185 } 186 ierr = VecRestoreArray(x,&xarray);CHKERRQ(ierr); 187 188 if (!info || info == lu->A.ncol+1) { 189 if (lu->options.IterRefine) { 190 ierr = PetscPrintf(PETSC_COMM_SELF,"Iterative Refinement:\n"); 191 ierr = PetscPrintf(PETSC_COMM_SELF," %8s%8s%16s%16s\n", "rhs", "Steps", "FERR", "BERR"); 192 for (i = 0; i < 1; ++i) { 193 ierr = PetscPrintf(PETSC_COMM_SELF," %8d%8d%16e%16e\n", i+1, lu->stat.RefineSteps, ferr, berr); 194 } 195 } 196 } else if (info > 0) { 197 if (lu->lwork == -1) { 198 ierr = PetscPrintf(PETSC_COMM_SELF," ** Estimated memory: %D bytes\n", info - lu->A.ncol); 199 } else { 200 ierr = PetscPrintf(PETSC_COMM_SELF," Warning: gssvx() returns info %D\n",info); 201 } 202 } else if (info < 0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB, "info = %D, the %D-th argument in gssvx() had an illegal value", info,-info); 203 204 if (lu->options.PrintStat) { 205 ierr = PetscPrintf(PETSC_COMM_SELF,"MatSolve__SuperLU():\n"); 206 PetscStackCall("SuperLU:StatPrint",StatPrint(&lu->stat)); 207 } 208 PetscFunctionReturn(0); 209 } 210 211 PetscErrorCode MatSolve_SuperLU(Mat A,Vec b,Vec x) 212 { 213 Mat_SuperLU *lu = (Mat_SuperLU*)A->data; 214 PetscErrorCode ierr; 215 216 PetscFunctionBegin; 217 if (A->factorerrortype) { 218 ierr = PetscInfo(A,"MatSolve is called with singular matrix factor, skip\n");CHKERRQ(ierr); 219 ierr = VecSetInf(x);CHKERRQ(ierr); 220 PetscFunctionReturn(0); 221 } 222 223 lu->options.Trans = TRANS; 224 ierr = MatSolve_SuperLU_Private(A,b,x);CHKERRQ(ierr); 225 PetscFunctionReturn(0); 226 } 227 228 PetscErrorCode MatSolveTranspose_SuperLU(Mat A,Vec b,Vec x) 229 { 230 Mat_SuperLU *lu = (Mat_SuperLU*)A->data; 231 PetscErrorCode ierr; 232 233 PetscFunctionBegin; 234 if (A->factorerrortype) { 235 ierr = PetscInfo(A,"MatSolve is called with singular matrix factor, skip\n");CHKERRQ(ierr); 236 ierr = VecSetInf(x);CHKERRQ(ierr); 237 PetscFunctionReturn(0); 238 } 239 240 lu->options.Trans = NOTRANS; 241 ierr = MatSolve_SuperLU_Private(A,b,x);CHKERRQ(ierr); 242 PetscFunctionReturn(0); 243 } 244 245 static PetscErrorCode MatLUFactorNumeric_SuperLU(Mat F,Mat A,const MatFactorInfo *info) 246 { 247 Mat_SuperLU *lu = (Mat_SuperLU*)F->data; 248 Mat_SeqAIJ *aa; 249 PetscErrorCode ierr; 250 PetscInt sinfo; 251 PetscReal ferr, berr; 252 NCformat *Ustore; 253 SCformat *Lstore; 254 255 PetscFunctionBegin; 256 if (lu->flg == SAME_NONZERO_PATTERN) { /* successing numerical factorization */ 257 lu->options.Fact = SamePattern; 258 /* Ref: ~SuperLU_3.0/EXAMPLE/dlinsolx2.c */ 259 Destroy_SuperMatrix_Store(&lu->A); 260 if (lu->options.Equil) { 261 ierr = MatCopy_SeqAIJ(A,lu->A_dup,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 262 } 263 if (lu->lwork >= 0) { 264 PetscStackCall("SuperLU:Destroy_SuperNode_Matrix",Destroy_SuperNode_Matrix(&lu->L)); 265 PetscStackCall("SuperLU:Destroy_CompCol_Matrix",Destroy_CompCol_Matrix(&lu->U)); 266 lu->options.Fact = SamePattern; 267 } 268 } 269 270 /* Create the SuperMatrix for lu->A=A^T: 271 Since SuperLU likes column-oriented matrices,we pass it the transpose, 272 and then solve A^T X = B in MatSolve(). */ 273 if (lu->options.Equil) { 274 aa = (Mat_SeqAIJ*)(lu->A_dup)->data; 275 } else { 276 aa = (Mat_SeqAIJ*)(A)->data; 277 } 278 #if defined(PETSC_USE_COMPLEX) 279 #if defined(PETSC_USE_REAL_SINGLE) 280 PetscStackCall("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)); 281 #else 282 PetscStackCall("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)); 283 #endif 284 #else 285 #if defined(PETSC_USE_REAL_SINGLE) 286 PetscStackCall("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)); 287 #else 288 PetscStackCall("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)); 289 #endif 290 #endif 291 292 /* Numerical factorization */ 293 lu->B.ncol = 0; /* Indicate not to solve the system */ 294 if (F->factortype == MAT_FACTOR_LU) { 295 #if defined(PETSC_USE_COMPLEX) 296 #if defined(PETSC_USE_REAL_SINGLE) 297 PetscStackCall("SuperLU:cgssvx",cgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 298 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, 299 &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 300 #else 301 PetscStackCall("SuperLU:zgssvx",zgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 302 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, 303 &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 304 #endif 305 #else 306 #if defined(PETSC_USE_REAL_SINGLE) 307 PetscStackCall("SuperLU:sgssvx",sgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 308 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, 309 &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 310 #else 311 PetscStackCall("SuperLU:dgssvx",dgssvx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 312 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, &ferr, &berr, 313 &lu->Glu,&lu->mem_usage, &lu->stat, &sinfo)); 314 #endif 315 #endif 316 } else if (F->factortype == MAT_FACTOR_ILU) { 317 /* Compute the incomplete factorization, condition number and pivot growth */ 318 #if defined(PETSC_USE_COMPLEX) 319 #if defined(PETSC_USE_REAL_SINGLE) 320 PetscStackCall("SuperLU:cgsisx",cgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r,lu->etree, lu->equed, lu->R, lu->C, 321 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, 322 &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 323 #else 324 PetscStackCall("SuperLU:zgsisx",zgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r,lu->etree, lu->equed, lu->R, lu->C, 325 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, 326 &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 327 #endif 328 #else 329 #if defined(PETSC_USE_REAL_SINGLE) 330 PetscStackCall("SuperLU:sgsisx",sgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 331 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, 332 &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 333 #else 334 PetscStackCall("SuperLU:dgsisx",dgsisx(&lu->options, &lu->A, lu->perm_c, lu->perm_r, lu->etree, lu->equed, lu->R, lu->C, 335 &lu->L, &lu->U, lu->work, lu->lwork, &lu->B, &lu->X, &lu->rpg, &lu->rcond, 336 &lu->Glu, &lu->mem_usage, &lu->stat, &sinfo)); 337 #endif 338 #endif 339 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Factor type not supported"); 340 if (!sinfo || sinfo == lu->A.ncol+1) { 341 if (lu->options.PivotGrowth) { 342 ierr = PetscPrintf(PETSC_COMM_SELF," Recip. pivot growth = %e\n", lu->rpg);CHKERRQ(ierr); 343 } 344 if (lu->options.ConditionNumber) { 345 ierr = PetscPrintf(PETSC_COMM_SELF," Recip. condition number = %e\n", lu->rcond);CHKERRQ(ierr); 346 } 347 } else if (sinfo > 0) { 348 if (A->erroriffailure) { 349 SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot in row %D",sinfo); 350 } else { 351 if (sinfo <= lu->A.ncol) { 352 if (lu->options.ILU_FillTol == 0.0) { 353 F->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; 354 } 355 ierr = PetscInfo2(F,"Number of zero pivots %D, ILU_FillTol %g\n",sinfo,lu->options.ILU_FillTol);CHKERRQ(ierr); 356 } else if (sinfo == lu->A.ncol + 1) { 357 /* 358 U is nonsingular, but RCOND is less than machine 359 precision, meaning that the matrix is singular to 360 working precision. Nevertheless, the solution and 361 error bounds are computed because there are a number 362 of situations where the computed solution can be more 363 accurate than the value of RCOND would suggest. 364 */ 365 ierr = PetscInfo1(F,"Matrix factor U is nonsingular, but is singular to working precision. The solution is computed. info %D",sinfo);CHKERRQ(ierr); 366 } else { /* sinfo > lu->A.ncol + 1 */ 367 F->factorerrortype = MAT_FACTOR_OUTMEMORY; 368 ierr = PetscInfo1(F,"Number of bytes allocated when memory allocation fails %D\n",sinfo);CHKERRQ(ierr); 369 } 370 } 371 } else SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_LIB, "info = %D, the %D-th argument in gssvx() had an illegal value", sinfo,-sinfo); 372 373 if (lu->options.PrintStat) { 374 ierr = PetscPrintf(PETSC_COMM_SELF,"MatLUFactorNumeric_SuperLU():\n");CHKERRQ(ierr); 375 PetscStackCall("SuperLU:StatPrint",StatPrint(&lu->stat)); 376 Lstore = (SCformat*) lu->L.Store; 377 Ustore = (NCformat*) lu->U.Store; 378 ierr = PetscPrintf(PETSC_COMM_SELF," No of nonzeros in factor L = %D\n", Lstore->nnz); 379 ierr = PetscPrintf(PETSC_COMM_SELF," No of nonzeros in factor U = %D\n", Ustore->nnz); 380 ierr = PetscPrintf(PETSC_COMM_SELF," No of nonzeros in L+U = %D\n", Lstore->nnz + Ustore->nnz - lu->A.ncol); 381 ierr = PetscPrintf(PETSC_COMM_SELF," L\\U MB %.3f\ttotal MB needed %.3f\n", 382 lu->mem_usage.for_lu/1e6, lu->mem_usage.total_needed/1e6); 383 } 384 385 lu->flg = SAME_NONZERO_PATTERN; 386 F->ops->solve = MatSolve_SuperLU; 387 F->ops->solvetranspose = MatSolveTranspose_SuperLU; 388 F->ops->matsolve = NULL; 389 PetscFunctionReturn(0); 390 } 391 392 static PetscErrorCode MatDestroy_SuperLU(Mat A) 393 { 394 PetscErrorCode ierr; 395 Mat_SuperLU *lu=(Mat_SuperLU*)A->data; 396 397 PetscFunctionBegin; 398 if (lu->CleanUpSuperLU) { /* Free the SuperLU datastructures */ 399 PetscStackCall("SuperLU:Destroy_SuperMatrix_Store",Destroy_SuperMatrix_Store(&lu->A)); 400 PetscStackCall("SuperLU:Destroy_SuperMatrix_Store",Destroy_SuperMatrix_Store(&lu->B)); 401 PetscStackCall("SuperLU:Destroy_SuperMatrix_Store",Destroy_SuperMatrix_Store(&lu->X)); 402 PetscStackCall("SuperLU:StatFree",StatFree(&lu->stat)); 403 if (lu->lwork >= 0) { 404 PetscStackCall("SuperLU:Destroy_SuperNode_Matrix",Destroy_SuperNode_Matrix(&lu->L)); 405 PetscStackCall("SuperLU:Destroy_CompCol_Matrix",Destroy_CompCol_Matrix(&lu->U)); 406 } 407 } 408 ierr = PetscFree(lu->etree);CHKERRQ(ierr); 409 ierr = PetscFree(lu->perm_r);CHKERRQ(ierr); 410 ierr = PetscFree(lu->perm_c);CHKERRQ(ierr); 411 ierr = PetscFree(lu->R);CHKERRQ(ierr); 412 ierr = PetscFree(lu->C);CHKERRQ(ierr); 413 ierr = PetscFree(lu->rhs_dup);CHKERRQ(ierr); 414 ierr = MatDestroy(&lu->A_dup);CHKERRQ(ierr); 415 ierr = PetscFree(A->data);CHKERRQ(ierr); 416 417 /* clear composed functions */ 418 ierr = PetscObjectComposeFunction((PetscObject)A,"MatFactorGetSolverPackage_C",NULL);CHKERRQ(ierr); 419 ierr = PetscObjectComposeFunction((PetscObject)A,"MatSuperluSetILUDropTol_C",NULL);CHKERRQ(ierr); 420 PetscFunctionReturn(0); 421 } 422 423 static PetscErrorCode MatView_SuperLU(Mat A,PetscViewer viewer) 424 { 425 PetscErrorCode ierr; 426 PetscBool iascii; 427 PetscViewerFormat format; 428 429 PetscFunctionBegin; 430 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 431 if (iascii) { 432 ierr = PetscViewerGetFormat(viewer,&format);CHKERRQ(ierr); 433 if (format == PETSC_VIEWER_ASCII_INFO) { 434 ierr = MatFactorInfo_SuperLU(A,viewer);CHKERRQ(ierr); 435 } 436 } 437 PetscFunctionReturn(0); 438 } 439 440 PetscErrorCode MatMatSolve_SuperLU(Mat A,Mat B,Mat X) 441 { 442 Mat_SuperLU *lu = (Mat_SuperLU*)A->data; 443 PetscBool flg; 444 PetscErrorCode ierr; 445 446 PetscFunctionBegin; 447 ierr = PetscObjectTypeCompareAny((PetscObject)B,&flg,MATSEQDENSE,MATMPIDENSE,NULL);CHKERRQ(ierr); 448 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix B must be MATDENSE matrix"); 449 ierr = PetscObjectTypeCompareAny((PetscObject)X,&flg,MATSEQDENSE,MATMPIDENSE,NULL);CHKERRQ(ierr); 450 if (!flg) SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix X must be MATDENSE matrix"); 451 lu->options.Trans = TRANS; 452 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"MatMatSolve_SuperLU() is not implemented yet"); 453 PetscFunctionReturn(0); 454 } 455 456 /* 457 Note the r permutation is ignored 458 */ 459 static PetscErrorCode MatLUFactorSymbolic_SuperLU(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info) 460 { 461 Mat_SuperLU *lu = (Mat_SuperLU*)(F->data); 462 463 PetscFunctionBegin; 464 lu->flg = DIFFERENT_NONZERO_PATTERN; 465 lu->CleanUpSuperLU = PETSC_TRUE; 466 F->ops->lufactornumeric = MatLUFactorNumeric_SuperLU; 467 PetscFunctionReturn(0); 468 } 469 470 static PetscErrorCode MatSuperluSetILUDropTol_SuperLU(Mat F,PetscReal dtol) 471 { 472 Mat_SuperLU *lu= (Mat_SuperLU*)F->data; 473 474 PetscFunctionBegin; 475 lu->options.ILU_DropTol = dtol; 476 PetscFunctionReturn(0); 477 } 478 479 /*@ 480 MatSuperluSetILUDropTol - Set SuperLU ILU drop tolerance 481 Logically Collective on Mat 482 483 Input Parameters: 484 + F - the factored matrix obtained by calling MatGetFactor() from PETSc-SuperLU interface 485 - dtol - drop tolerance 486 487 Options Database: 488 . -mat_superlu_ilu_droptol <dtol> 489 490 Level: beginner 491 492 References: 493 . SuperLU Users' Guide 494 495 .seealso: MatGetFactor() 496 @*/ 497 PetscErrorCode MatSuperluSetILUDropTol(Mat F,PetscReal dtol) 498 { 499 PetscErrorCode ierr; 500 501 PetscFunctionBegin; 502 PetscValidHeaderSpecific(F,MAT_CLASSID,1); 503 PetscValidLogicalCollectiveReal(F,dtol,2); 504 ierr = PetscTryMethod(F,"MatSuperluSetILUDropTol_C",(Mat,PetscReal),(F,dtol));CHKERRQ(ierr); 505 PetscFunctionReturn(0); 506 } 507 508 PetscErrorCode MatFactorGetSolverPackage_seqaij_superlu(Mat A,const MatSolverPackage *type) 509 { 510 PetscFunctionBegin; 511 *type = MATSOLVERSUPERLU; 512 PetscFunctionReturn(0); 513 } 514 515 /*MC 516 MATSOLVERSUPERLU = "superlu" - A solver package providing solvers LU and ILU for sequential matrices 517 via the external package SuperLU. 518 519 Use ./configure --download-superlu to have PETSc installed with SuperLU 520 521 Use -pc_type lu -pc_factor_mat_solver_package superlu to us this direct solver 522 523 Options Database Keys: 524 + -mat_superlu_equil <FALSE> - Equil (None) 525 . -mat_superlu_colperm <COLAMD> - (choose one of) NATURAL MMD_ATA MMD_AT_PLUS_A COLAMD 526 . -mat_superlu_iterrefine <NOREFINE> - (choose one of) NOREFINE SINGLE DOUBLE EXTRA 527 . -mat_superlu_symmetricmode: <FALSE> - SymmetricMode (None) 528 . -mat_superlu_diagpivotthresh <1> - DiagPivotThresh (None) 529 . -mat_superlu_pivotgrowth <FALSE> - PivotGrowth (None) 530 . -mat_superlu_conditionnumber <FALSE> - ConditionNumber (None) 531 . -mat_superlu_rowperm <NOROWPERM> - (choose one of) NOROWPERM LargeDiag 532 . -mat_superlu_replacetinypivot <FALSE> - ReplaceTinyPivot (None) 533 . -mat_superlu_printstat <FALSE> - PrintStat (None) 534 . -mat_superlu_lwork <0> - size of work array in bytes used by factorization (None) 535 . -mat_superlu_ilu_droptol <0> - ILU_DropTol (None) 536 . -mat_superlu_ilu_filltol <0> - ILU_FillTol (None) 537 . -mat_superlu_ilu_fillfactor <0> - ILU_FillFactor (None) 538 . -mat_superlu_ilu_droprull <0> - ILU_DropRule (None) 539 . -mat_superlu_ilu_norm <0> - ILU_Norm (None) 540 - -mat_superlu_ilu_milu <0> - ILU_MILU (None) 541 542 Notes: Do not confuse this with MATSOLVERSUPERLU_DIST which is for parallel sparse solves 543 544 Level: beginner 545 546 .seealso: PCLU, PCILU, MATSOLVERSUPERLU_DIST, MATSOLVERMUMPS, PCFactorSetMatSolverPackage(), MatSolverPackage 547 M*/ 548 549 static PetscErrorCode MatGetFactor_seqaij_superlu(Mat A,MatFactorType ftype,Mat *F) 550 { 551 Mat B; 552 Mat_SuperLU *lu; 553 PetscErrorCode ierr; 554 PetscInt indx,m=A->rmap->n,n=A->cmap->n; 555 PetscBool flg,set; 556 PetscReal real_input; 557 const char *colperm[] ={"NATURAL","MMD_ATA","MMD_AT_PLUS_A","COLAMD"}; /* MY_PERMC - not supported by the petsc interface yet */ 558 const char *iterrefine[]={"NOREFINE", "SINGLE", "DOUBLE", "EXTRA"}; 559 const char *rowperm[] ={"NOROWPERM", "LargeDiag"}; /* MY_PERMC - not supported by the petsc interface yet */ 560 561 PetscFunctionBegin; 562 ierr = MatCreate(PetscObjectComm((PetscObject)A),&B);CHKERRQ(ierr); 563 ierr = MatSetSizes(B,A->rmap->n,A->cmap->n,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr); 564 ierr = PetscStrallocpy("superlu",&((PetscObject)B)->type_name);CHKERRQ(ierr); 565 ierr = MatSetUp(B);CHKERRQ(ierr); 566 if (ftype == MAT_FACTOR_LU || ftype == MAT_FACTOR_ILU) { 567 B->ops->lufactorsymbolic = MatLUFactorSymbolic_SuperLU; 568 B->ops->ilufactorsymbolic = MatLUFactorSymbolic_SuperLU; 569 } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Factor type not supported"); 570 571 ierr = PetscFree(B->solvertype);CHKERRQ(ierr); 572 ierr = PetscStrallocpy(MATSOLVERSUPERLU,&B->solvertype);CHKERRQ(ierr); 573 574 B->ops->getinfo = MatGetInfo_External; 575 B->ops->destroy = MatDestroy_SuperLU; 576 B->ops->view = MatView_SuperLU; 577 B->factortype = ftype; 578 B->assembled = PETSC_TRUE; /* required by -ksp_view */ 579 B->preallocated = PETSC_TRUE; 580 581 ierr = PetscNewLog(B,&lu);CHKERRQ(ierr); 582 583 if (ftype == MAT_FACTOR_LU) { 584 set_default_options(&lu->options); 585 /* Comments from SuperLU_4.0/SRC/dgssvx.c: 586 "Whether or not the system will be equilibrated depends on the 587 scaling of the matrix A, but if equilibration is used, A is 588 overwritten by diag(R)*A*diag(C) and B by diag(R)*B 589 (if options->Trans=NOTRANS) or diag(C)*B (if options->Trans = TRANS or CONJ)." 590 We set 'options.Equil = NO' as default because additional space is needed for it. 591 */ 592 lu->options.Equil = NO; 593 } else if (ftype == MAT_FACTOR_ILU) { 594 /* Set the default input options of ilu: */ 595 PetscStackCall("SuperLU:ilu_set_default_options",ilu_set_default_options(&lu->options)); 596 } 597 lu->options.PrintStat = NO; 598 599 /* Initialize the statistics variables. */ 600 PetscStackCall("SuperLU:StatInit",StatInit(&lu->stat)); 601 lu->lwork = 0; /* allocate space internally by system malloc */ 602 603 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)A),((PetscObject)A)->prefix,"SuperLU Options","Mat");CHKERRQ(ierr); 604 ierr = PetscOptionsBool("-mat_superlu_equil","Equil","None",(PetscBool)lu->options.Equil,(PetscBool*)&lu->options.Equil,NULL);CHKERRQ(ierr); 605 ierr = PetscOptionsEList("-mat_superlu_colperm","ColPerm","None",colperm,4,colperm[3],&indx,&flg);CHKERRQ(ierr); 606 if (flg) lu->options.ColPerm = (colperm_t)indx; 607 ierr = PetscOptionsEList("-mat_superlu_iterrefine","IterRefine","None",iterrefine,4,iterrefine[0],&indx,&flg);CHKERRQ(ierr); 608 if (flg) lu->options.IterRefine = (IterRefine_t)indx; 609 ierr = PetscOptionsBool("-mat_superlu_symmetricmode","SymmetricMode","None",(PetscBool)lu->options.SymmetricMode,&flg,&set);CHKERRQ(ierr); 610 if (set && flg) lu->options.SymmetricMode = YES; 611 ierr = PetscOptionsReal("-mat_superlu_diagpivotthresh","DiagPivotThresh","None",lu->options.DiagPivotThresh,&real_input,&flg);CHKERRQ(ierr); 612 if (flg) lu->options.DiagPivotThresh = (double) real_input; 613 ierr = PetscOptionsBool("-mat_superlu_pivotgrowth","PivotGrowth","None",(PetscBool)lu->options.PivotGrowth,&flg,&set);CHKERRQ(ierr); 614 if (set && flg) lu->options.PivotGrowth = YES; 615 ierr = PetscOptionsBool("-mat_superlu_conditionnumber","ConditionNumber","None",(PetscBool)lu->options.ConditionNumber,&flg,&set);CHKERRQ(ierr); 616 if (set && flg) lu->options.ConditionNumber = YES; 617 ierr = PetscOptionsEList("-mat_superlu_rowperm","rowperm","None",rowperm,2,rowperm[lu->options.RowPerm],&indx,&flg);CHKERRQ(ierr); 618 if (flg) lu->options.RowPerm = (rowperm_t)indx; 619 ierr = PetscOptionsBool("-mat_superlu_replacetinypivot","ReplaceTinyPivot","None",(PetscBool)lu->options.ReplaceTinyPivot,&flg,&set);CHKERRQ(ierr); 620 if (set && flg) lu->options.ReplaceTinyPivot = YES; 621 ierr = PetscOptionsBool("-mat_superlu_printstat","PrintStat","None",(PetscBool)lu->options.PrintStat,&flg,&set);CHKERRQ(ierr); 622 if (set && flg) lu->options.PrintStat = YES; 623 ierr = PetscOptionsInt("-mat_superlu_lwork","size of work array in bytes used by factorization","None",lu->lwork,&lu->lwork,NULL);CHKERRQ(ierr); 624 if (lu->lwork > 0) { 625 /* lwork is in bytes, hence PetscMalloc() is used here, not PetscMalloc1()*/ 626 ierr = PetscMalloc(lu->lwork,&lu->work);CHKERRQ(ierr); 627 } else if (lu->lwork != 0 && lu->lwork != -1) { 628 ierr = PetscPrintf(PETSC_COMM_SELF," Warning: lwork %D is not supported by SUPERLU. The default lwork=0 is used.\n",lu->lwork); 629 lu->lwork = 0; 630 } 631 /* ilu options */ 632 ierr = PetscOptionsReal("-mat_superlu_ilu_droptol","ILU_DropTol","None",lu->options.ILU_DropTol,&real_input,&flg);CHKERRQ(ierr); 633 if (flg) lu->options.ILU_DropTol = (double) real_input; 634 ierr = PetscOptionsReal("-mat_superlu_ilu_filltol","ILU_FillTol","None",lu->options.ILU_FillTol,&real_input,&flg);CHKERRQ(ierr); 635 if (flg) lu->options.ILU_FillTol = (double) real_input; 636 ierr = PetscOptionsReal("-mat_superlu_ilu_fillfactor","ILU_FillFactor","None",lu->options.ILU_FillFactor,&real_input,&flg);CHKERRQ(ierr); 637 if (flg) lu->options.ILU_FillFactor = (double) real_input; 638 ierr = PetscOptionsInt("-mat_superlu_ilu_droprull","ILU_DropRule","None",lu->options.ILU_DropRule,&lu->options.ILU_DropRule,NULL);CHKERRQ(ierr); 639 ierr = PetscOptionsInt("-mat_superlu_ilu_norm","ILU_Norm","None",lu->options.ILU_Norm,&indx,&flg);CHKERRQ(ierr); 640 if (flg) lu->options.ILU_Norm = (norm_t)indx; 641 ierr = PetscOptionsInt("-mat_superlu_ilu_milu","ILU_MILU","None",lu->options.ILU_MILU,&indx,&flg);CHKERRQ(ierr); 642 if (flg) lu->options.ILU_MILU = (milu_t)indx; 643 ierr = PetscOptionsEnd();CHKERRQ(ierr); 644 if (lu->options.Equil == YES) { 645 /* superlu overwrites input matrix and rhs when Equil is used, thus create A_dup to keep user's A unchanged */ 646 ierr = MatDuplicate_SeqAIJ(A,MAT_COPY_VALUES,&lu->A_dup);CHKERRQ(ierr); 647 } 648 649 /* Allocate spaces (notice sizes are for the transpose) */ 650 ierr = PetscMalloc1(m,&lu->etree);CHKERRQ(ierr); 651 ierr = PetscMalloc1(n,&lu->perm_r);CHKERRQ(ierr); 652 ierr = PetscMalloc1(m,&lu->perm_c);CHKERRQ(ierr); 653 ierr = PetscMalloc1(n,&lu->R);CHKERRQ(ierr); 654 ierr = PetscMalloc1(m,&lu->C);CHKERRQ(ierr); 655 656 /* create rhs and solution x without allocate space for .Store */ 657 #if defined(PETSC_USE_COMPLEX) 658 #if defined(PETSC_USE_REAL_SINGLE) 659 PetscStackCall("SuperLU:cCreate_Dense_Matrix(",cCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE)); 660 PetscStackCall("SuperLU:cCreate_Dense_Matrix(",cCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_C, SLU_GE)); 661 #else 662 PetscStackCall("SuperLU:zCreate_Dense_Matrix",zCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE)); 663 PetscStackCall("SuperLU:zCreate_Dense_Matrix",zCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_Z, SLU_GE)); 664 #endif 665 #else 666 #if defined(PETSC_USE_REAL_SINGLE) 667 PetscStackCall("SuperLU:sCreate_Dense_Matrix",sCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE)); 668 PetscStackCall("SuperLU:sCreate_Dense_Matrix",sCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_S, SLU_GE)); 669 #else 670 PetscStackCall("SuperLU:dCreate_Dense_Matrix",dCreate_Dense_Matrix(&lu->B, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE)); 671 PetscStackCall("SuperLU:dCreate_Dense_Matrix",dCreate_Dense_Matrix(&lu->X, m, 1, NULL, m, SLU_DN, SLU_D, SLU_GE)); 672 #endif 673 #endif 674 675 ierr = PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverPackage_C",MatFactorGetSolverPackage_seqaij_superlu);CHKERRQ(ierr); 676 ierr = PetscObjectComposeFunction((PetscObject)B,"MatSuperluSetILUDropTol_C",MatSuperluSetILUDropTol_SuperLU);CHKERRQ(ierr); 677 B->data = lu; 678 679 *F = B; 680 PetscFunctionReturn(0); 681 } 682 683 PETSC_EXTERN PetscErrorCode MatSolverPackageRegister_SuperLU(void) 684 { 685 PetscErrorCode ierr; 686 687 PetscFunctionBegin; 688 ierr = MatSolverPackageRegister(MATSOLVERSUPERLU,MATSEQAIJ, MAT_FACTOR_LU,MatGetFactor_seqaij_superlu);CHKERRQ(ierr); 689 ierr = MatSolverPackageRegister(MATSOLVERSUPERLU,MATSEQAIJ, MAT_FACTOR_ILU,MatGetFactor_seqaij_superlu);CHKERRQ(ierr); 690 PetscFunctionReturn(0); 691 } 692