/* Provides an interface to the SuperLU_DIST sparse solver */ #include <../src/mat/impls/aij/seq/aij.h> #include <../src/mat/impls/aij/mpi/mpiaij.h> #include EXTERN_C_BEGIN #if defined(PETSC_USE_COMPLEX) #define CASTDOUBLECOMPLEX (doublecomplex*) #define CASTDOUBLECOMPLEXSTAR (doublecomplex**) #include #define LUstructInit zLUstructInit #define ScalePermstructInit zScalePermstructInit #define ScalePermstructFree zScalePermstructFree #define LUstructFree zLUstructFree #define Destroy_LU zDestroy_LU #define ScalePermstruct_t zScalePermstruct_t #define LUstruct_t zLUstruct_t #define SOLVEstruct_t zSOLVEstruct_t #define SolveFinalize zSolveFinalize #define pGetDiagU pzGetDiagU #define pgssvx pzgssvx #define allocateA_dist zallocateA_dist #define Create_CompRowLoc_Matrix_dist zCreate_CompRowLoc_Matrix_dist #define SLU SLU_Z #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) #define DeAllocLlu_3d zDeAllocLlu_3d #define DeAllocGlu_3d zDeAllocGlu_3d #define Destroy_A3d_gathered_on_2d zDestroy_A3d_gathered_on_2d #define pgssvx3d pzgssvx3d #endif #elif defined(PETSC_USE_REAL_SINGLE) #define CASTDOUBLECOMPLEX #define CASTDOUBLECOMPLEXSTAR #include #define LUstructInit sLUstructInit #define ScalePermstructInit sScalePermstructInit #define ScalePermstructFree sScalePermstructFree #define LUstructFree sLUstructFree #define Destroy_LU sDestroy_LU #define ScalePermstruct_t sScalePermstruct_t #define LUstruct_t sLUstruct_t #define SOLVEstruct_t sSOLVEstruct_t #define SolveFinalize sSolveFinalize #define pGetDiagU psGetDiagU #define pgssvx psgssvx #define allocateA_dist sallocateA_dist #define Create_CompRowLoc_Matrix_dist sCreate_CompRowLoc_Matrix_dist #define SLU SLU_S #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) #define DeAllocLlu_3d sDeAllocLlu_3d #define DeAllocGlu_3d sDeAllocGlu_3d #define Destroy_A3d_gathered_on_2d sDestroy_A3d_gathered_on_2d #define pgssvx3d psgssvx3d #endif #else #define CASTDOUBLECOMPLEX #define CASTDOUBLECOMPLEXSTAR #include #define LUstructInit dLUstructInit #define ScalePermstructInit dScalePermstructInit #define ScalePermstructFree dScalePermstructFree #define LUstructFree dLUstructFree #define Destroy_LU dDestroy_LU #define ScalePermstruct_t dScalePermstruct_t #define LUstruct_t dLUstruct_t #define SOLVEstruct_t dSOLVEstruct_t #define SolveFinalize dSolveFinalize #define pGetDiagU pdGetDiagU #define pgssvx pdgssvx #define allocateA_dist dallocateA_dist #define Create_CompRowLoc_Matrix_dist dCreate_CompRowLoc_Matrix_dist #define SLU SLU_D #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) #define DeAllocLlu_3d dDeAllocLlu_3d #define DeAllocGlu_3d dDeAllocGlu_3d #define Destroy_A3d_gathered_on_2d dDestroy_A3d_gathered_on_2d #define pgssvx3d pdgssvx3d #endif #endif EXTERN_C_END typedef struct { int_t nprow,npcol,*row,*col; gridinfo_t grid; #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) PetscBool use3d; int_t npdep; /* replication factor, must be power of two */ gridinfo3d_t grid3d; #endif superlu_dist_options_t options; SuperMatrix A_sup; ScalePermstruct_t ScalePermstruct; LUstruct_t LUstruct; int StatPrint; SOLVEstruct_t SOLVEstruct; fact_t FactPattern; MPI_Comm comm_superlu; PetscScalar *val; PetscBool matsolve_iscalled,matmatsolve_iscalled; PetscBool CleanUpSuperLU_Dist; /* Flag to clean up (non-global) SuperLU objects during Destroy */ } Mat_SuperLU_DIST; PetscErrorCode MatSuperluDistGetDiagU_SuperLU_DIST(Mat F,PetscScalar *diagU) { Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)F->data; PetscFunctionBegin; PetscStackCall("SuperLU_DIST:pGetDiagU",pGetDiagU(F->rmap->N,&lu->LUstruct,&lu->grid,CASTDOUBLECOMPLEX diagU)); PetscFunctionReturn(0); } PetscErrorCode MatSuperluDistGetDiagU(Mat F,PetscScalar *diagU) { PetscFunctionBegin; PetscValidHeaderSpecific(F,MAT_CLASSID,1); PetscTryMethod(F,"MatSuperluDistGetDiagU_C",(Mat,PetscScalar*),(F,diagU)); PetscFunctionReturn(0); } /* This allows reusing the Superlu_DIST communicator and grid when only a single SuperLU_DIST matrix is used at a time */ typedef struct { MPI_Comm comm; PetscBool busy; gridinfo_t grid; #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) PetscBool use3d; gridinfo3d_t grid3d; #endif } PetscSuperLU_DIST; static PetscMPIInt Petsc_Superlu_dist_keyval = MPI_KEYVAL_INVALID; PETSC_EXTERN PetscMPIInt MPIAPI Petsc_Superlu_dist_keyval_Delete_Fn(MPI_Comm comm,PetscMPIInt keyval,void *attr_val,void *extra_state) { PetscSuperLU_DIST *context = (PetscSuperLU_DIST *) attr_val; PetscFunctionBegin; if (keyval != Petsc_Superlu_dist_keyval) SETERRMPI(PETSC_COMM_SELF,PETSC_ERR_ARG_CORRUPT,"Unexpected keyval"); PetscCall(PetscInfo(NULL,"Removing Petsc_Superlu_dist_keyval attribute from communicator that is being freed\n")); #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) if (context->use3d) { PetscStackCall("SuperLU_DIST:superlu_gridexit3d",superlu_gridexit3d(&context->grid3d)); } else #endif PetscStackCall("SuperLU_DIST:superlu_gridexit",superlu_gridexit(&context->grid)); PetscCallMPI(MPI_Comm_free(&context->comm)); PetscCall(PetscFree(context)); PetscFunctionReturn(MPI_SUCCESS); } /* Performs MPI_Comm_free_keyval() on Petsc_Superlu_dist_keyval but keeps the global variable for users who do not destroy all PETSc objects before PetscFinalize(). The value Petsc_Superlu_dist_keyval is retained so that Petsc_Superlu_dist_keyval_Delete_Fn() can still check that the keyval associated with the MPI communicator is correct when the MPI communicator is destroyed. This is called in PetscFinalize() */ static PetscErrorCode Petsc_Superlu_dist_keyval_free(void) { PetscMPIInt Petsc_Superlu_dist_keyval_temp = Petsc_Superlu_dist_keyval; PetscFunctionBegin; PetscCall(PetscInfo(NULL,"Freeing Petsc_Superlu_dist_keyval\n")); PetscCallMPI(MPI_Comm_free_keyval(&Petsc_Superlu_dist_keyval_temp)); PetscFunctionReturn(0); } static PetscErrorCode MatDestroy_SuperLU_DIST(Mat A) { Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)A->data; PetscFunctionBegin; if (lu->CleanUpSuperLU_Dist) { /* Deallocate SuperLU_DIST storage */ PetscStackCall("SuperLU_DIST:Destroy_CompRowLoc_Matrix_dist",Destroy_CompRowLoc_Matrix_dist(&lu->A_sup)); if (lu->options.SolveInitialized) { PetscStackCall("SuperLU_DIST:SolveFinalize",SolveFinalize(&lu->options, &lu->SOLVEstruct)); } #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) if (lu->use3d) { if (lu->grid3d.zscp.Iam == 0) { PetscStackCall("SuperLU_DIST:Destroy_LU",Destroy_LU(A->cmap->N, &lu->grid3d.grid2d, &lu->LUstruct)); } else { PetscStackCall("SuperLU_DIST:DeAllocLlu_3d",DeAllocLlu_3d(lu->A_sup.ncol, &lu->LUstruct, &lu->grid3d)); PetscStackCall("SuperLU_DIST:DeAllocGlu_3d",DeAllocGlu_3d(&lu->LUstruct)); } PetscStackCall("SuperLU_DIST:Destroy_A3d_gathered_on_2d",Destroy_A3d_gathered_on_2d(&lu->SOLVEstruct, &lu->grid3d)); } else #endif PetscStackCall("SuperLU_DIST:Destroy_LU",Destroy_LU(A->cmap->N, &lu->grid, &lu->LUstruct)); PetscStackCall("SuperLU_DIST:ScalePermstructFree",ScalePermstructFree(&lu->ScalePermstruct)); PetscStackCall("SuperLU_DIST:LUstructFree",LUstructFree(&lu->LUstruct)); /* Release the SuperLU_DIST process grid only if the matrix has its own copy, that is it is not in the communicator context */ if (lu->comm_superlu) { #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) if (lu->use3d) { PetscStackCall("SuperLU_DIST:superlu_gridexit3d",superlu_gridexit3d(&lu->grid3d)); } else #endif PetscStackCall("SuperLU_DIST:superlu_gridexit",superlu_gridexit(&lu->grid)); } } /* * We always need to release the communicator that was created in MatGetFactor_aij_superlu_dist. * lu->CleanUpSuperLU_Dist was turned on in MatLUFactorSymbolic_SuperLU_DIST. There are some use * cases where we only create a matrix but do not solve mat. In these cases, lu->CleanUpSuperLU_Dist * is off, and the communicator was not released or marked as "not busy " in the old code. * Here we try to release comm regardless. */ if (lu->comm_superlu) { PetscCall(PetscCommRestoreComm(PetscObjectComm((PetscObject)A),&lu->comm_superlu)); } else { PetscSuperLU_DIST *context; MPI_Comm comm; PetscMPIInt flg; PetscCall(PetscObjectGetComm((PetscObject)A,&comm)); PetscCallMPI(MPI_Comm_get_attr(comm,Petsc_Superlu_dist_keyval,&context,&flg)); PetscCheck(flg,PETSC_COMM_SELF,PETSC_ERR_PLIB,"Communicator does not have expected Petsc_Superlu_dist_keyval attribute"); context->busy = PETSC_FALSE; } PetscCall(PetscFree(A->data)); /* clear composed functions */ PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatFactorGetSolverType_C",NULL)); PetscCall(PetscObjectComposeFunction((PetscObject)A,"MatSuperluDistGetDiagU_C",NULL)); PetscFunctionReturn(0); } static PetscErrorCode MatSolve_SuperLU_DIST(Mat A,Vec b_mpi,Vec x) { Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)A->data; PetscInt m=A->rmap->n; SuperLUStat_t stat; PetscReal berr[1]; PetscScalar *bptr = NULL; int info; /* SuperLU_Dist info code is ALWAYS an int, even with long long indices */ static PetscBool cite = PETSC_FALSE; PetscFunctionBegin; PetscCheck(lu->options.Fact == FACTORED,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"SuperLU_DIST options.Fact must equal FACTORED"); PetscCall(PetscCitationsRegister("@article{lidemmel03,\n author = {Xiaoye S. Li and James W. Demmel},\n title = {{SuperLU_DIST}: A Scalable Distributed-Memory Sparse Direct\n Solver for Unsymmetric Linear Systems},\n journal = {ACM Trans. Mathematical Software},\n volume = {29},\n number = {2},\n pages = {110-140},\n year = 2003\n}\n",&cite)); if (lu->options.SolveInitialized && !lu->matsolve_iscalled) { /* see comments in MatMatSolve() */ PetscStackCall("SuperLU_DIST:SolveFinalize",SolveFinalize(&lu->options, &lu->SOLVEstruct)); lu->options.SolveInitialized = NO; } PetscCall(VecCopy(b_mpi,x)); PetscCall(VecGetArray(x,&bptr)); PetscStackCall("SuperLU_DIST:PStatInit",PStatInit(&stat)); /* Initialize the statistics variables. */ #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) && !PetscDefined(MISSING_GETLINE) if (lu->use3d) PetscStackCall("SuperLU_DIST:pgssvx3d",pgssvx3d(&lu->options,&lu->A_sup,&lu->ScalePermstruct,CASTDOUBLECOMPLEX bptr,m,1,&lu->grid3d,&lu->LUstruct,&lu->SOLVEstruct,berr,&stat,&info)); else #endif PetscStackCall("SuperLU_DIST:pgssvx",pgssvx(&lu->options,&lu->A_sup,&lu->ScalePermstruct,CASTDOUBLECOMPLEX bptr,m,1,&lu->grid,&lu->LUstruct,&lu->SOLVEstruct,berr,&stat,&info)); PetscCheck(!info,PETSC_COMM_SELF,PETSC_ERR_LIB,"pdgssvx fails, info: %d",info); if (lu->options.PrintStat) PetscStackCall("SuperLU_DIST:PStatPrint",PStatPrint(&lu->options, &stat, &lu->grid)); /* Print the statistics. */ PetscStackCall("SuperLU_DIST:PStatFree",PStatFree(&stat)); PetscCall(VecRestoreArray(x,&bptr)); lu->matsolve_iscalled = PETSC_TRUE; lu->matmatsolve_iscalled = PETSC_FALSE; PetscFunctionReturn(0); } static PetscErrorCode MatMatSolve_SuperLU_DIST(Mat A,Mat B_mpi,Mat X) { Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)A->data; PetscInt m = A->rmap->n,nrhs; SuperLUStat_t stat; PetscReal berr[1]; PetscScalar *bptr; int info; /* SuperLU_Dist info code is ALWAYS an int, even with long long indices */ PetscBool flg; PetscFunctionBegin; PetscCheck(lu->options.Fact == FACTORED,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"SuperLU_DIST options.Fact must equal FACTORED"); PetscCall(PetscObjectTypeCompareAny((PetscObject)B_mpi,&flg,MATSEQDENSE,MATMPIDENSE,NULL)); PetscCheck(flg,PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix B must be MATDENSE matrix"); if (X != B_mpi) { PetscCall(PetscObjectTypeCompareAny((PetscObject)X,&flg,MATSEQDENSE,MATMPIDENSE,NULL)); PetscCheck(flg,PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Matrix X must be MATDENSE matrix"); } if (lu->options.SolveInitialized && !lu->matmatsolve_iscalled) { /* communication pattern of SOLVEstruct is unlikely created for matmatsolve, thus destroy it and create a new SOLVEstruct. Otherwise it may result in memory corruption or incorrect solution See src/mat/tests/ex125.c */ PetscStackCall("SuperLU_DIST:SolveFinalize",SolveFinalize(&lu->options, &lu->SOLVEstruct)); lu->options.SolveInitialized = NO; } if (X != B_mpi) { PetscCall(MatCopy(B_mpi,X,SAME_NONZERO_PATTERN)); } PetscCall(MatGetSize(B_mpi,NULL,&nrhs)); PetscStackCall("SuperLU_DIST:PStatInit",PStatInit(&stat)); /* Initialize the statistics variables. */ PetscCall(MatDenseGetArray(X,&bptr)); #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) && !PetscDefined(MISSING_GETLINE) if (lu->use3d) PetscStackCall("SuperLU_DIST:pgssvx3d",pgssvx3d(&lu->options,&lu->A_sup,&lu->ScalePermstruct,CASTDOUBLECOMPLEX bptr,m,nrhs,&lu->grid3d,&lu->LUstruct,&lu->SOLVEstruct,berr,&stat,&info)); else #endif PetscStackCall("SuperLU_DIST:pgssvx",pgssvx(&lu->options,&lu->A_sup,&lu->ScalePermstruct,CASTDOUBLECOMPLEX bptr,m,nrhs,&lu->grid,&lu->LUstruct,&lu->SOLVEstruct,berr,&stat,&info)); PetscCheck(!info,PETSC_COMM_SELF,PETSC_ERR_LIB,"pdgssvx fails, info: %d",info); PetscCall(MatDenseRestoreArray(X,&bptr)); if (lu->options.PrintStat) PetscStackCall("SuperLU_DIST:PStatPrint",PStatPrint(&lu->options, &stat, &lu->grid)); /* Print the statistics. */ PetscStackCall("SuperLU_DIST:PStatFree",PStatFree(&stat)); lu->matsolve_iscalled = PETSC_FALSE; lu->matmatsolve_iscalled = PETSC_TRUE; PetscFunctionReturn(0); } /* input: F: numeric Cholesky factor output: nneg: total number of negative pivots nzero: total number of zero pivots npos: (global dimension of F) - nneg - nzero */ static PetscErrorCode MatGetInertia_SuperLU_DIST(Mat F,PetscInt *nneg,PetscInt *nzero,PetscInt *npos) { Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)F->data; PetscScalar *diagU=NULL; PetscInt M,i,neg=0,zero=0,pos=0; PetscReal r; PetscFunctionBegin; PetscCheck(F->assembled,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Matrix factor F is not assembled"); PetscCheck(lu->options.RowPerm == NOROWPERM,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Must set NOROWPERM"); PetscCall(MatGetSize(F,&M,NULL)); PetscCall(PetscMalloc1(M,&diagU)); PetscCall(MatSuperluDistGetDiagU(F,diagU)); for (i=0; i -PETSC_MACHINE_EPSILON && r < PETSC_MACHINE_EPSILON,PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"diagU[%" PetscInt_FMT "]=%g + i %g is non-real",i,(double)PetscRealPart(diagU[i]),(double)(r*10.0)); r = PetscRealPart(diagU[i]); #else r = diagU[i]; #endif if (r > 0) { pos++; } else if (r < 0) { neg++; } else zero++; } PetscCall(PetscFree(diagU)); if (nneg) *nneg = neg; if (nzero) *nzero = zero; if (npos) *npos = pos; PetscFunctionReturn(0); } static PetscErrorCode MatLUFactorNumeric_SuperLU_DIST(Mat F,Mat A,const MatFactorInfo *info) { Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)F->data; Mat Aloc; const PetscScalar *av; const PetscInt *ai = NULL,*aj = NULL; PetscInt nz,dummy; int sinfo; /* SuperLU_Dist info flag is always an int even with long long indices */ SuperLUStat_t stat; PetscReal *berr = 0; PetscBool ismpiaij,isseqaij,flg; PetscFunctionBegin; PetscCall(PetscObjectBaseTypeCompare((PetscObject)A,MATSEQAIJ,&isseqaij)); PetscCall(PetscObjectBaseTypeCompare((PetscObject)A,MATMPIAIJ,&ismpiaij)); if (ismpiaij) { PetscCall(MatMPIAIJGetLocalMat(A,MAT_INITIAL_MATRIX,&Aloc)); } else if (isseqaij) { PetscCall(PetscObjectReference((PetscObject)A)); Aloc = A; } else SETERRQ(PetscObjectComm((PetscObject)A),PETSC_ERR_SUP,"Not for type %s",((PetscObject)A)->type_name); PetscCall(MatGetRowIJ(Aloc,0,PETSC_FALSE,PETSC_FALSE,&dummy,&ai,&aj,&flg)); PetscCheck(flg,PETSC_COMM_SELF,PETSC_ERR_SUP,"GetRowIJ failed"); PetscCall(MatSeqAIJGetArrayRead(Aloc,&av)); nz = ai[Aloc->rmap->n]; /* Allocations for A_sup */ if (lu->options.Fact == DOFACT) { /* first numeric factorization */ PetscStackCall("SuperLU_DIST:allocateA_dist",allocateA_dist(Aloc->rmap->n, nz, CASTDOUBLECOMPLEXSTAR &lu->val, &lu->col, &lu->row)); } else { /* successive numeric factorization, sparsity pattern and perm_c are reused. */ if (lu->FactPattern == SamePattern_SameRowPerm) { lu->options.Fact = SamePattern_SameRowPerm; /* matrix has similar numerical values */ } else if (lu->FactPattern == SamePattern) { #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) if (lu->use3d) { if (lu->grid3d.zscp.Iam == 0) { PetscStackCall("SuperLU_DIST:Destroy_LU",Destroy_LU(A->cmap->N, &lu->grid3d.grid2d, &lu->LUstruct)); PetscStackCall("SuperLU_DIST:SolveFinalize",SolveFinalize(&lu->options, &lu->SOLVEstruct)); } else { PetscStackCall("SuperLU_DIST:DeAllocLlu_3d",DeAllocLlu_3d(lu->A_sup.ncol, &lu->LUstruct, &lu->grid3d)); PetscStackCall("SuperLU_DIST:DeAllocGlu_3d",DeAllocGlu_3d(&lu->LUstruct)); } } else #endif PetscStackCall("SuperLU_DIST:Destroy_LU",Destroy_LU(A->rmap->N, &lu->grid, &lu->LUstruct)); lu->options.Fact = SamePattern; } else if (lu->FactPattern == DOFACT) { PetscStackCall("SuperLU_DIST:Destroy_CompRowLoc_Matrix_dist",Destroy_CompRowLoc_Matrix_dist(&lu->A_sup)); PetscStackCall("SuperLU_DIST:Destroy_LU",Destroy_LU(A->rmap->N, &lu->grid, &lu->LUstruct)); lu->options.Fact = DOFACT; PetscStackCall("SuperLU_DIST:allocateA_dist",allocateA_dist(Aloc->rmap->n, nz, CASTDOUBLECOMPLEXSTAR &lu->val, &lu->col, &lu->row)); } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"options.Fact must be one of SamePattern SamePattern_SameRowPerm DOFACT"); } /* Copy AIJ matrix to superlu_dist matrix */ PetscCall(PetscArraycpy(lu->row,ai,Aloc->rmap->n+1)); PetscCall(PetscArraycpy(lu->col,aj,nz)); PetscCall(PetscArraycpy(lu->val,av,nz)); PetscCall(MatRestoreRowIJ(Aloc,0,PETSC_FALSE,PETSC_FALSE,&dummy,&ai,&aj,&flg)); PetscCheck(flg,PETSC_COMM_SELF,PETSC_ERR_SUP,"RestoreRowIJ failed"); PetscCall(MatSeqAIJRestoreArrayRead(Aloc,&av)); PetscCall(MatDestroy(&Aloc)); /* Create and setup A_sup */ if (lu->options.Fact == DOFACT) { PetscStackCall("SuperLU_DIST:Create_CompRowLoc_Matrix_dist",Create_CompRowLoc_Matrix_dist(&lu->A_sup, A->rmap->N, A->cmap->N, nz, A->rmap->n, A->rmap->rstart, CASTDOUBLECOMPLEX lu->val, lu->col, lu->row, SLU_NR_loc, SLU, SLU_GE)); } /* Factor the matrix. */ PetscStackCall("SuperLU_DIST:PStatInit",PStatInit(&stat)); /* Initialize the statistics variables. */ #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) && !PetscDefined(MISSING_GETLINE) if (lu->use3d) { PetscStackCall("SuperLU_DIST:pgssvx3d",pgssvx3d(&lu->options, &lu->A_sup, &lu->ScalePermstruct, 0, A->rmap->n, 0, &lu->grid3d, &lu->LUstruct, &lu->SOLVEstruct, berr, &stat, &sinfo)); } else #endif PetscStackCall("SuperLU_DIST:pgssvx",pgssvx(&lu->options, &lu->A_sup, &lu->ScalePermstruct, 0, A->rmap->n, 0, &lu->grid, &lu->LUstruct, &lu->SOLVEstruct, berr, &stat, &sinfo)); if (sinfo > 0) { PetscCheck(!A->erroriffailure,PETSC_COMM_SELF,PETSC_ERR_MAT_LU_ZRPVT,"Zero pivot in row %d",sinfo); if (sinfo <= lu->A_sup.ncol) { F->factorerrortype = MAT_FACTOR_NUMERIC_ZEROPIVOT; PetscCall(PetscInfo(F,"U(i,i) is exactly zero, i= %d\n",sinfo)); } else if (sinfo > lu->A_sup.ncol) { /* number of bytes allocated when memory allocation failure occurred, plus A->ncol. */ F->factorerrortype = MAT_FACTOR_OUTMEMORY; PetscCall(PetscInfo(F,"Number of bytes allocated when memory allocation fails %d\n",sinfo)); } } else PetscCheck(sinfo >= 0,PETSC_COMM_SELF,PETSC_ERR_LIB, "info = %d, argument in p*gssvx() had an illegal value", sinfo); if (lu->options.PrintStat) { PetscStackCall("SuperLU_DIST:PStatPrint",PStatPrint(&lu->options, &stat, &lu->grid)); /* Print the statistics. */ } PetscStackCall("SuperLU_DIST:PStatFree",PStatFree(&stat)); F->assembled = PETSC_TRUE; F->preallocated = PETSC_TRUE; lu->options.Fact = FACTORED; /* The factored form of A is supplied. Local option used by this func. only */ PetscFunctionReturn(0); } /* Note the Petsc r and c permutations are ignored */ static PetscErrorCode MatLUFactorSymbolic_SuperLU_DIST(Mat F,Mat A,IS r,IS c,const MatFactorInfo *info) { Mat_SuperLU_DIST *lu = (Mat_SuperLU_DIST*)F->data; PetscInt M = A->rmap->N,N = A->cmap->N,indx; PetscMPIInt size,mpiflg; PetscBool flg,set; const char *colperm[] = {"NATURAL","MMD_AT_PLUS_A","MMD_ATA","METIS_AT_PLUS_A","PARMETIS"}; const char *rowperm[] = {"NOROWPERM","LargeDiag_MC64","LargeDiag_AWPM","MY_PERMR"}; const char *factPattern[] = {"SamePattern","SamePattern_SameRowPerm","DOFACT"}; MPI_Comm comm; PetscSuperLU_DIST *context = NULL; PetscFunctionBegin; /* Set options to F */ PetscCall(PetscObjectGetComm((PetscObject)F,&comm)); PetscCallMPI(MPI_Comm_size(comm,&size)); PetscOptionsBegin(PetscObjectComm((PetscObject)F),((PetscObject)F)->prefix,"SuperLU_Dist Options","Mat"); PetscCall(PetscOptionsBool("-mat_superlu_dist_equil","Equilibrate matrix","None",lu->options.Equil ? PETSC_TRUE : PETSC_FALSE,&flg,&set)); if (set && !flg) lu->options.Equil = NO; PetscCall(PetscOptionsEList("-mat_superlu_dist_rowperm","Row permutation","None",rowperm,4,rowperm[1],&indx,&flg)); if (flg) { switch (indx) { case 0: lu->options.RowPerm = NOROWPERM; break; case 1: lu->options.RowPerm = LargeDiag_MC64; break; case 2: lu->options.RowPerm = LargeDiag_AWPM; break; case 3: lu->options.RowPerm = MY_PERMR; break; default: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown row permutation"); } } PetscCall(PetscOptionsEList("-mat_superlu_dist_colperm","Column permutation","None",colperm,5,colperm[3],&indx,&flg)); if (flg) { switch (indx) { case 0: lu->options.ColPerm = NATURAL; break; case 1: lu->options.ColPerm = MMD_AT_PLUS_A; break; case 2: lu->options.ColPerm = MMD_ATA; break; case 3: lu->options.ColPerm = METIS_AT_PLUS_A; break; case 4: lu->options.ColPerm = PARMETIS; /* only works for np>1 */ break; default: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown column permutation"); } } lu->options.ReplaceTinyPivot = NO; PetscCall(PetscOptionsBool("-mat_superlu_dist_replacetinypivot","Replace tiny pivots","None",lu->options.ReplaceTinyPivot ? PETSC_TRUE : PETSC_FALSE,&flg,&set)); if (set && flg) lu->options.ReplaceTinyPivot = YES; lu->options.ParSymbFact = NO; PetscCall(PetscOptionsBool("-mat_superlu_dist_parsymbfact","Parallel symbolic factorization","None",PETSC_FALSE,&flg,&set)); if (set && flg && size>1) { #if defined(PETSC_HAVE_PARMETIS) lu->options.ParSymbFact = YES; lu->options.ColPerm = PARMETIS; /* in v2.2, PARMETIS is forced for ParSymbFact regardless of user ordering setting */ #else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"parsymbfact needs PARMETIS"); #endif } lu->FactPattern = SamePattern; PetscCall(PetscOptionsEList("-mat_superlu_dist_fact","Sparsity pattern for repeated matrix factorization","None",factPattern,3,factPattern[0],&indx,&flg)); if (flg) { switch (indx) { case 0: lu->FactPattern = SamePattern; break; case 1: lu->FactPattern = SamePattern_SameRowPerm; break; case 2: lu->FactPattern = DOFACT; break; } } lu->options.IterRefine = NOREFINE; PetscCall(PetscOptionsBool("-mat_superlu_dist_iterrefine","Use iterative refinement","None",lu->options.IterRefine == NOREFINE ? PETSC_FALSE : PETSC_TRUE ,&flg,&set)); if (set) { if (flg) lu->options.IterRefine = SLU_DOUBLE; else lu->options.IterRefine = NOREFINE; } if (PetscLogPrintInfo) lu->options.PrintStat = YES; else lu->options.PrintStat = NO; PetscCall(PetscOptionsBool("-mat_superlu_dist_statprint","Print factorization information","None",(PetscBool)lu->options.PrintStat,(PetscBool*)&lu->options.PrintStat,NULL)); /* Additional options for special cases */ if (Petsc_Superlu_dist_keyval == MPI_KEYVAL_INVALID) { PetscCallMPI(MPI_Comm_create_keyval(MPI_COMM_NULL_COPY_FN,Petsc_Superlu_dist_keyval_Delete_Fn,&Petsc_Superlu_dist_keyval,(void*)0)); PetscCall(PetscRegisterFinalize(Petsc_Superlu_dist_keyval_free)); } PetscCallMPI(MPI_Comm_get_attr(comm,Petsc_Superlu_dist_keyval,&context,&mpiflg)); if (!mpiflg || context->busy) { /* additional options */ if (!mpiflg) { PetscCall(PetscNew(&context)); context->busy = PETSC_TRUE; PetscCallMPI(MPI_Comm_dup(comm,&context->comm)); PetscCallMPI(MPI_Comm_set_attr(comm,Petsc_Superlu_dist_keyval,context)); } else { PetscCall(PetscCommGetComm(PetscObjectComm((PetscObject)A),&lu->comm_superlu)); } /* Default number of process columns and rows */ lu->nprow = (int_t) (0.5 + PetscSqrtReal((PetscReal)size)); if (!lu->nprow) lu->nprow = 1; while (lu->nprow > 0) { lu->npcol = (int_t) (size/lu->nprow); if (size == lu->nprow * lu->npcol) break; lu->nprow--; } #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) lu->use3d = PETSC_FALSE; lu->npdep = 1; #endif #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) PetscCall(PetscOptionsBool("-mat_superlu_dist_3d","Use SuperLU_DIST 3D distribution","None",lu->use3d,&lu->use3d,NULL)); PetscCheck(!PetscDefined(MISSING_GETLINE) || !lu->use3d,PetscObjectComm((PetscObject)A),PETSC_ERR_SUP_SYS,"-mat_superlu_dist_3d requires a system with a getline() implementation"); if (lu->use3d) { PetscInt t; PetscCall(PetscOptionsInt("-mat_superlu_dist_d","Number of z entries in processor partition","None",lu->npdep,(PetscInt*)&lu->npdep,NULL)); t = (PetscInt) PetscLog2Real((PetscReal)lu->npdep); PetscCheck(PetscPowInt(2,t) == lu->npdep,PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_OUTOFRANGE,"-mat_superlu_dist_d %lld must be a power of 2",(long long)lu->npdep); if (lu->npdep > 1) { lu->nprow = (int_t) (0.5 + PetscSqrtReal((PetscReal)(size/lu->npdep))); if (!lu->nprow) lu->nprow = 1; while (lu->nprow > 0) { lu->npcol = (int_t) (size/(lu->npdep*lu->nprow)); if (size == lu->nprow * lu->npcol * lu->npdep) break; lu->nprow--; } } } #endif PetscCall(PetscOptionsInt("-mat_superlu_dist_r","Number rows in processor partition","None",lu->nprow,(PetscInt*)&lu->nprow,NULL)); PetscCall(PetscOptionsInt("-mat_superlu_dist_c","Number columns in processor partition","None",lu->npcol,(PetscInt*)&lu->npcol,NULL)); #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) PetscCheck(size == lu->nprow*lu->npcol*lu->npdep,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Number of processes %d must equal to nprow %lld * npcol %lld * npdep %lld",size,(long long)lu->nprow,(long long)lu->npcol,(long long)lu->npdep); #else PetscCheck(size == lu->nprow*lu->npcol,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Number of processes %d must equal to nprow %lld * npcol %lld",size,(long long)lu->nprow,(long long)lu->npcol); #endif /* end of adding additional options */ #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) if (lu->use3d) { PetscStackCall("SuperLU_DIST:superlu_gridinit3d",superlu_gridinit3d(context ? context->comm : lu->comm_superlu, lu->nprow, lu->npcol,lu->npdep, &lu->grid3d)); if (context) {context->grid3d = lu->grid3d; context->use3d = lu->use3d;} } else { #endif PetscStackCall("SuperLU_DIST:superlu_gridinit",superlu_gridinit(context ? context->comm : lu->comm_superlu, lu->nprow, lu->npcol, &lu->grid)); if (context) context->grid = lu->grid; #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) } #endif PetscCall(PetscInfo(NULL,"Duplicating a communicator for SuperLU_DIST and calling superlu_gridinit()\n")); if (mpiflg) { PetscCall(PetscInfo(NULL,"Communicator attribute already in use so not saving communicator and SuperLU_DIST grid in communicator attribute \n")); } else { PetscCall(PetscInfo(NULL,"Storing communicator and SuperLU_DIST grid in communicator attribute\n")); } } else { /* (mpiflg && !context->busy) */ PetscCall(PetscInfo(NULL,"Reusing communicator and superlu_gridinit() for SuperLU_DIST from communicator attribute.")); context->busy = PETSC_TRUE; lu->grid = context->grid; } PetscOptionsEnd(); /* Initialize ScalePermstruct and LUstruct. */ PetscStackCall("SuperLU_DIST:ScalePermstructInit",ScalePermstructInit(M, N, &lu->ScalePermstruct)); PetscStackCall("SuperLU_DIST:LUstructInit",LUstructInit(N, &lu->LUstruct)); F->ops->lufactornumeric = MatLUFactorNumeric_SuperLU_DIST; F->ops->solve = MatSolve_SuperLU_DIST; F->ops->matsolve = MatMatSolve_SuperLU_DIST; F->ops->getinertia = NULL; if (A->symmetric || A->hermitian) F->ops->getinertia = MatGetInertia_SuperLU_DIST; lu->CleanUpSuperLU_Dist = PETSC_TRUE; PetscFunctionReturn(0); } static PetscErrorCode MatCholeskyFactorSymbolic_SuperLU_DIST(Mat F,Mat A,IS r,const MatFactorInfo *info) { PetscFunctionBegin; PetscCall(MatLUFactorSymbolic_SuperLU_DIST(F,A,r,r,info)); F->ops->choleskyfactornumeric = MatLUFactorNumeric_SuperLU_DIST; PetscFunctionReturn(0); } static PetscErrorCode MatFactorGetSolverType_aij_superlu_dist(Mat A,MatSolverType *type) { PetscFunctionBegin; *type = MATSOLVERSUPERLU_DIST; PetscFunctionReturn(0); } static PetscErrorCode MatView_Info_SuperLU_DIST(Mat A,PetscViewer viewer) { Mat_SuperLU_DIST *lu=(Mat_SuperLU_DIST*)A->data; superlu_dist_options_t options; PetscFunctionBegin; /* check if matrix is superlu_dist type */ if (A->ops->solve != MatSolve_SuperLU_DIST) PetscFunctionReturn(0); options = lu->options; PetscCall(PetscViewerASCIIPrintf(viewer,"SuperLU_DIST run parameters:\n")); /* would love to use superlu 'IFMT' macro but it looks like it's inconsistently applied, the * format spec for int64_t is set to %d for whatever reason */ PetscCall(PetscViewerASCIIPrintf(viewer," Process grid nprow %lld x npcol %lld \n",(long long)lu->nprow,(long long)lu->npcol)); #if PETSC_PKG_SUPERLU_DIST_VERSION_GE(7,2,0) if (lu->use3d) { PetscCall(PetscViewerASCIIPrintf(viewer," Using 3d decomposition with npdep %lld \n",(long long)lu->npdep)); } #endif PetscCall(PetscViewerASCIIPrintf(viewer," Equilibrate matrix %s \n",PetscBools[options.Equil != NO])); PetscCall(PetscViewerASCIIPrintf(viewer," Replace tiny pivots %s \n",PetscBools[options.ReplaceTinyPivot != NO])); PetscCall(PetscViewerASCIIPrintf(viewer," Use iterative refinement %s \n",PetscBools[options.IterRefine == SLU_DOUBLE])); PetscCall(PetscViewerASCIIPrintf(viewer," Processors in row %lld col partition %lld \n",(long long)lu->nprow,(long long)lu->npcol)); switch (options.RowPerm) { case NOROWPERM: PetscCall(PetscViewerASCIIPrintf(viewer," Row permutation NOROWPERM\n")); break; case LargeDiag_MC64: PetscCall(PetscViewerASCIIPrintf(viewer," Row permutation LargeDiag_MC64\n")); break; case LargeDiag_AWPM: PetscCall(PetscViewerASCIIPrintf(viewer," Row permutation LargeDiag_AWPM\n")); break; case MY_PERMR: PetscCall(PetscViewerASCIIPrintf(viewer," Row permutation MY_PERMR\n")); break; default: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown column permutation"); } switch (options.ColPerm) { case NATURAL: PetscCall(PetscViewerASCIIPrintf(viewer," Column permutation NATURAL\n")); break; case MMD_AT_PLUS_A: PetscCall(PetscViewerASCIIPrintf(viewer," Column permutation MMD_AT_PLUS_A\n")); break; case MMD_ATA: PetscCall(PetscViewerASCIIPrintf(viewer," Column permutation MMD_ATA\n")); break; /* Even though this is called METIS, the SuperLU_DIST code sets this by default if PARMETIS is defined, not METIS */ case METIS_AT_PLUS_A: PetscCall(PetscViewerASCIIPrintf(viewer," Column permutation METIS_AT_PLUS_A\n")); break; case PARMETIS: PetscCall(PetscViewerASCIIPrintf(viewer," Column permutation PARMETIS\n")); break; default: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown column permutation"); } PetscCall(PetscViewerASCIIPrintf(viewer," Parallel symbolic factorization %s \n",PetscBools[options.ParSymbFact != NO])); if (lu->FactPattern == SamePattern) { PetscCall(PetscViewerASCIIPrintf(viewer," Repeated factorization SamePattern\n")); } else if (lu->FactPattern == SamePattern_SameRowPerm) { PetscCall(PetscViewerASCIIPrintf(viewer," Repeated factorization SamePattern_SameRowPerm\n")); } else if (lu->FactPattern == DOFACT) { PetscCall(PetscViewerASCIIPrintf(viewer," Repeated factorization DOFACT\n")); } else { SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Unknown factorization pattern"); } PetscFunctionReturn(0); } static PetscErrorCode MatView_SuperLU_DIST(Mat A,PetscViewer viewer) { PetscBool iascii; PetscViewerFormat format; PetscFunctionBegin; PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii)); if (iascii) { PetscCall(PetscViewerGetFormat(viewer,&format)); if (format == PETSC_VIEWER_ASCII_INFO) { PetscCall(MatView_Info_SuperLU_DIST(A,viewer)); } } PetscFunctionReturn(0); } static PetscErrorCode MatGetFactor_aij_superlu_dist(Mat A,MatFactorType ftype,Mat *F) { Mat B; Mat_SuperLU_DIST *lu; PetscInt M=A->rmap->N,N=A->cmap->N; PetscMPIInt size; superlu_dist_options_t options; PetscFunctionBegin; /* Create the factorization matrix */ PetscCall(MatCreate(PetscObjectComm((PetscObject)A),&B)); PetscCall(MatSetSizes(B,A->rmap->n,A->cmap->n,M,N)); PetscCall(PetscStrallocpy("superlu_dist",&((PetscObject)B)->type_name)); PetscCall(MatSetUp(B)); B->ops->getinfo = MatGetInfo_External; B->ops->view = MatView_SuperLU_DIST; B->ops->destroy = MatDestroy_SuperLU_DIST; /* Set the default input options: options.Fact = DOFACT; options.Equil = YES; options.ParSymbFact = NO; options.ColPerm = METIS_AT_PLUS_A; options.RowPerm = LargeDiag_MC64; options.ReplaceTinyPivot = YES; options.IterRefine = DOUBLE; options.Trans = NOTRANS; options.SolveInitialized = NO; -hold the communication pattern used MatSolve() and MatMatSolve() options.RefineInitialized = NO; options.PrintStat = YES; options.SymPattern = NO; */ set_default_options_dist(&options); B->trivialsymbolic = PETSC_TRUE; if (ftype == MAT_FACTOR_LU) { B->factortype = MAT_FACTOR_LU; B->ops->lufactorsymbolic = MatLUFactorSymbolic_SuperLU_DIST; } else { B->factortype = MAT_FACTOR_CHOLESKY; B->ops->choleskyfactorsymbolic = MatCholeskyFactorSymbolic_SuperLU_DIST; options.SymPattern = YES; } /* set solvertype */ PetscCall(PetscFree(B->solvertype)); PetscCall(PetscStrallocpy(MATSOLVERSUPERLU_DIST,&B->solvertype)); PetscCall(PetscNewLog(B,&lu)); B->data = lu; PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)A),&size)); lu->options = options; lu->options.Fact = DOFACT; lu->matsolve_iscalled = PETSC_FALSE; lu->matmatsolve_iscalled = PETSC_FALSE; PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatFactorGetSolverType_C",MatFactorGetSolverType_aij_superlu_dist)); PetscCall(PetscObjectComposeFunction((PetscObject)B,"MatSuperluDistGetDiagU_C",MatSuperluDistGetDiagU_SuperLU_DIST)); *F = B; PetscFunctionReturn(0); } PETSC_EXTERN PetscErrorCode MatSolverTypeRegister_SuperLU_DIST(void) { PetscFunctionBegin; PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU_DIST,MATMPIAIJ,MAT_FACTOR_LU,MatGetFactor_aij_superlu_dist)); PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU_DIST,MATSEQAIJ,MAT_FACTOR_LU,MatGetFactor_aij_superlu_dist)); PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU_DIST,MATMPIAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_aij_superlu_dist)); PetscCall(MatSolverTypeRegister(MATSOLVERSUPERLU_DIST,MATSEQAIJ,MAT_FACTOR_CHOLESKY,MatGetFactor_aij_superlu_dist)); PetscFunctionReturn(0); } /*MC MATSOLVERSUPERLU_DIST - Parallel direct solver package for LU factorization Use ./configure --download-superlu_dist --download-parmetis --download-metis --download-ptscotch to have PETSc installed with SuperLU_DIST Use -pc_type lu -pc_factor_mat_solver_type superlu_dist to use this direct solver Works with AIJ matrices Options Database Keys: + -mat_superlu_dist_r - number of rows in processor partition . -mat_superlu_dist_c - number of columns in processor partition . -mat_superlu_dist_3d - use 3d partition, requires SuperLU_DIST 7.2 or later . -mat_superlu_dist_d - depth in 3d partition (valid only if -mat_superlu_dist_3d) is provided . -mat_superlu_dist_equil - equilibrate the matrix . -mat_superlu_dist_rowperm - row permutation . -mat_superlu_dist_colperm - column permutation . -mat_superlu_dist_replacetinypivot - replace tiny pivots . -mat_superlu_dist_fact - (choose one of) SamePattern SamePattern_SameRowPerm DOFACT . -mat_superlu_dist_iterrefine - use iterative refinement - -mat_superlu_dist_statprint - print factorization information Notes: If PETSc was configured with --with-cuda than this solver will automatically use the GPUs. Level: beginner .seealso: `PCLU` .seealso: `PCFactorSetMatSolverType()`, `MatSolverType` M*/