SuperLU Distributed 8.2.1
Distributed memory sparse direct solver
|
Solves a lower triangular system L*X = B, with L being the lower triangular factor computed previously by PDGSTRF. More...
#include "superlu_ddefs.h"
Macros | |
#define | ISEND_IRECV |
Functions | |
int_t | pdReDistribute_B_to_X (double *B, int_t m_loc, int nrhs, int_t ldb, int_t fst_row, int_t *ilsum, double *x, ScalePermstruct_t *ScalePermstruct, Glu_persist_t *Glu_persist, gridinfo_t *grid, SOLVEstruct_t *SOLVEstruct) |
int_t | pdReDistribute_X_to_B (int_t n, double *B, int_t m_loc, int_t ldb, int_t fst_row, int_t nrhs, double *x, int_t *ilsum, ScalePermstruct_t *ScalePermstruct, Glu_persist_t *Glu_persist, gridinfo_t *grid, SOLVEstruct_t *SOLVEstruct) |
void | pdgstrsL (int_t n, LUstruct_t *LUstruct, ScalePermstruct_t *ScalePermstruct, gridinfo_t *grid, double *B, int_t m_loc, int_t fst_row, int_t ldb, int nrhs, SOLVEstruct_t *SOLVEstruct, SuperLUStat_t *stat, int *info) |
Solves a lower triangular system L*X = B, with L being the lower triangular factor computed previously by PDGSTRF.
Copyright (c) 2003, The Regents of the University of California, through Lawrence Berkeley National Laboratory (subject to receipt of any required approvals from U.S. Dept. of Energy)
All rights reserved.
The source code is distributed under BSD license, see the file License.txt at the top-level directory.
-- Distributed SuperLU routine (version 2.3) -- Lawrence Berkeley National Lab, Univ. of California Berkeley. October 15, 2008
#define ISEND_IRECV |
void pdgstrsL | ( | int_t | n, |
LUstruct_t * | LUstruct, | ||
ScalePermstruct_t * | ScalePermstruct, | ||
gridinfo_t * | grid, | ||
double * | B, | ||
int_t | m_loc, | ||
int_t | fst_row, | ||
int_t | ldb, | ||
int | nrhs, | ||
SOLVEstruct_t * | SOLVEstruct, | ||
SuperLUStat_t * | stat, | ||
int * | info | ||
) |
Purpose ======= PDGSTRSL solves a lower triangular system L*X = B, with L being the lower triangular factor computed previously by PDGSTRF. If the equilibration, and row and column permutations were performed, the LU factorization was performed for A1 where A1 = Pc*Pr*diag(R)*A*diag(C)*Pc^T = L*U and the linear system solved is A1 * Y = Pc*Pr*B1, where B was overwritten by B1 = diag(R)*B, and the permutation to B1 by Pc*Pr is applied internally in this routine. Arguments ========= n (input) int (global) The order of the system of linear equations. LUstruct (input) LUstruct_t* The distributed data structures storing L and U factors. The L and U factors are obtained from PDGSTRF for the possibly scaled and permuted matrix A. See superlu_ddefs.h for the definition of 'LUstruct_t'. A may be scaled and permuted into A1, so that A1 = Pc*Pr*diag(R)*A*diag(C)*Pc^T = L*U grid (input) gridinfo_t* The 2D process mesh. It contains the MPI communicator, the number of process rows (NPROW), the number of process columns (NPCOL), and my process rank. It is an input argument to all the parallel routines. Grid can be initialized by subroutine SUPERLU_GRIDINIT. See superlu_defs.h for the definition of 'gridinfo_t'. B (input/output) double* On entry, the distributed right-hand side matrix of the possibly equilibrated system. That is, B may be overwritten by diag(R)*B. On exit, the distributed solution matrix Y of the possibly equilibrated system if info = 0, where Y = Pc*diag(C)^(-1)*X, and X is the solution of the original system. m_loc (input) int (local) The local row dimension of matrix B. fst_row (input) int (global) The row number of B's first row in the global matrix. ldb (input) int (local) The leading dimension of matrix B. nrhs (input) int (global) Number of right-hand sides. SOLVEstruct (output) SOLVEstruct_t* (global) Contains the information for the communication during the solution phase. stat (output) SuperLUStat_t* Record the statistics about the triangular solves. See util.h for the definition of 'SuperLUStat_t'. info (output) int* = 0: successful exit < 0: if info = -i, the i-th argument had an illegal value
int_t pdReDistribute_B_to_X | ( | double * | B, |
int_t | m_loc, | ||
int | nrhs, | ||
int_t | ldb, | ||
int_t | fst_row, | ||
int_t * | ilsum, | ||
double * | x, | ||
ScalePermstruct_t * | ScalePermstruct, | ||
Glu_persist_t * | Glu_persist, | ||
gridinfo_t * | grid, | ||
SOLVEstruct_t * | SOLVEstruct | ||
) |
Purpose ======= Re-distribute B on the diagonal processes of the 2D process mesh. Note ==== This routine can only be called after the routine pxgstrs_init(), in which the structures of the send and receive buffers are set up. Arguments ========= B (input) double* The distributed right-hand side matrix of the possibly equilibrated system. m_loc (input) int (local) The local row dimension of matrix B. nrhs (input) int (global) Number of right-hand sides. ldb (input) int (local) Leading dimension of matrix B. fst_row (input) int (global) The row number of B's first row in the global matrix. ilsum (input) int* (global) Starting position of each supernode in a full array. x (output) double* The solution vector. It is valid only on the diagonal processes. ScalePermstruct (input) ScalePermstruct_t* The data structure to store the scaling and permutation vectors describing the transformations performed to the original matrix A. grid (input) gridinfo_t* The 2D process mesh. SOLVEstruct (input) SOLVEstruct_t* Contains the information for the communication during the solution phase. Return value ============
int_t pdReDistribute_X_to_B | ( | int_t | n, |
double * | B, | ||
int_t | m_loc, | ||
int_t | ldb, | ||
int_t | fst_row, | ||
int_t | nrhs, | ||
double * | x, | ||
int_t * | ilsum, | ||
ScalePermstruct_t * | ScalePermstruct, | ||
Glu_persist_t * | Glu_persist, | ||
gridinfo_t * | grid, | ||
SOLVEstruct_t * | SOLVEstruct | ||
) |
Purpose ======= Re-distribute X on the diagonal processes to B distributed on all the processes. Note ==== This routine can only be called after the routine pxgstrs_init(), in which the structures of the send and receive buffers are set up.