Definitions which are precision-neutral. More...

#include <mpi.h>
#include <stdlib.h>
#include <stdio.h>
#include <limits.h>
#include <string.h>
#include <ctype.h>
#include <math.h>
#include <stdint.h>
#include "superlu_dist_config.h"
#include "oneside.h"
#include "gpu_api_utils.h"
#include "superlu_FortranCInterface.h"
#include "superlu_FCnames.h"
#include "superlu_enum_consts.h"
#include "supermatrix.h"
#include "util_dist.h"
#include "psymbfact.h"

Include dependency graph for superlu_defs.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
struct	superlu_scope_t

struct	gridinfo_t

struct	gridinfo3d_t

struct	Glu_persist_t

struct	Glu_freeable_t

struct	pxgstrs_comm_t

struct	superlu_dist_options_t

struct	superlu_dist_mem_usage_t

struct	Ucb_indptr_t

struct	Ublock_info_t

struct	Remain_info_t

struct	etree_node

struct	superlu_pair

struct	uPanelInfo_t

struct	lPanelInfo_t

struct	packLUInfo_t

struct	HyP_t

struct	local_l_blk_info_t

struct	local_u_blk_info_t

struct	perm_array_t

struct	factStat_t

struct	d2Hreduce_t

struct	treeList_t

struct	treeTopoInfo_t

struct	gEtreeInfo_t

struct	sForest_t

struct	commRequests_t

struct	factNodelists_t

struct	msgs_t

struct	xtrsTimer_t

struct	C_Tree

Macros
#define	SUPERLU_DIST_MAJOR_VERSION 9

#define	SUPERLU_DIST_MINOR_VERSION 0

#define	SUPERLU_DIST_PATCH_VERSION 0

#define	SUPERLU_DIST_RELEASE_DATE "May 8, 2024"

#define	GPU_ACC

#define	mpi_int_t MPI_LONG_LONG_INT

#define	IFMT " %lld"

#define	SuperLU_MPI_COMPLEX MPI_C_COMPLEX

#define	SuperLU_MPI_DOUBLE_COMPLEX MPI_C_DOUBLE_COMPLEX

#define	ISORT /* NOTE: qsort() has bug on Mac */

#define	MAX_SUPER_SIZE 512 /* Sherry: moved from superlu_gpu.cu */

#define	BC_HEADER 2

#define	LB_DESCRIPTOR 2

#define	BR_HEADER 3

#define	UB_DESCRIPTOR 2

#define	BC_HEADER_NEWU 3

#define	UB_DESCRIPTOR_NEWU 2

#define	UB_DESCRIPTOR_NEWUCPP 3

#define	NBUFFERS 5

#define	NTAGS INT_MAX

#define	UjROW 10

#define	UkSUB 11

#define	UkVAL 12

#define	LkSUB 13

#define	LkVAL 14

#define	LkkDIAG 15

#define	XK_H 2 /* The header preceding each X block. */

#define	LSUM_H 2 /* The header preceding each MOD block. */

#define	GSUM 20

#define	Xk 21

#define	Yk 22

#define	LSUM 23

#define	COMM_ALL 100

#define	COMM_COLUMN 101

#define	COMM_ROW 102

#define	SUPER_LINEAR 11

#define	SUPER_BLOCK 12

#define	NO_MARKER 3

#define	IAM(comm) { int rank; MPI_Comm_rank ( comm, &rank ); rank};

#define	MYROW(iam, grid) ( (iam) / grid->npcol )

#define	MYCOL(iam, grid) ( (iam) % grid->npcol )

#define	BlockNum(i) ( supno[i] )

#define	FstBlockC(bnum) ( xsup[bnum] )

#define	SuperSize(bnum) ( xsup[bnum+1]-xsup[bnum] )

#define	LBi(bnum, grid) ( (bnum)/grid->nprow )/* Global to local block rowwise */

#define	LBj(bnum, grid) ( (bnum)/grid->npcol )/* Global to local block columnwise*/

#define	PROW(bnum, grid) ( (bnum) % grid->nprow )

#define	PCOL(bnum, grid) ( (bnum) % grid->npcol )

#define	PNUM(i, j, grid) ( (i)grid->npcol + j ) / Process number at coord(i,j) */

#define	CEILING(a, b) ( ((a)%(b)) ? ((a)/(b) + 1) : ((a)/(b)) )

#define	RHS_ITERATE(i) for (i = 0; i < nrhs; ++i)

#define	X_BLK(i) ilsum[i] * nrhs + (i+1) * XK_H

#define	XT_BLK(i) ilsumT[i] * nrhs + (i+1) * XK_H

#define	LSUM_BLK(i) ilsum[i] * nrhs + (i+1) * LSUM_H

#define	SuperLU_timer_ SuperLU_timer_dist_

#define	LOG2(x) (log10((double) x) / log10(2.0))

#define	VT_TRACEON

#define	VT_TRACEOFF

#define	SUPERLU_DIST_EXPORT

#define	MAGMA_CONST

#define	DIM_X 32

#define	DIM_Y 16

#define	DIM_XA DIM_X

#define	DIM_YA DIM_Y

#define	DIM_XB DIM_X

#define	DIM_YB DIM_Y

#define	WARP_SIZE 32

#define	NWARP DIM_X*DIM_Y/WARP_SIZE

#define	THR_M ( BLK_M / DIM_X )

#define	THR_N ( BLK_N / DIM_Y )

#define	fetch(A, m, n, bound) offs_d##A[min(n*LD##A+m, bound)]

#define	fma(A, B, C) C += (A*B)

#define	HANDLE_SIZE 8

#define	cmax(a, b) ((a) > (b) ? (a) : (b))

#define	SLU_MPI_TAG(id, num) ( (6*(num)+id) % tag_ub )

#define	__SUPERLU_ASYNC_TREE

#define	DEG_TREE 2

Typedefs
typedef int64_t	int_t

typedef int64_t	handle_t

typedef enum treePartStrat	treePartStrat

typedef struct xtrsTimer_t	xtrsTimer_t

typedef enum trtype_t	trtype_t

Enumerations
enum	treePartStrat { ND , GD }

enum	SupernodeToGridMap_t { NOT_IN_GRID , IN_GRID_ZERO , IN_GRID_AIJ }

enum	trtype_t { UPPER_TRI , LOWER_TRI }

Functions
void	superlu_gridinit (MPI_Comm, int, int, gridinfo_t *)
	All processes in the MPI communicator must call this routine. More...

void	superlu_gridmap (MPI_Comm, int, int, int[], int, gridinfo_t *)
	All processes in the MPI communicator must call this routine. More...

void	superlu_gridexit (gridinfo_t *)

void	superlu_gridinit3d (MPI_Comm Bcomm, int nprow, int npcol, int npdep, gridinfo3d_t *grid)
	All processes in the MPI communicator must call this routine. More...

void	superlu_gridmap3d (MPI_Comm, int, int, int, int[], gridinfo3d_t *)
	All processes in the MPI communicator must call this routine. On output, if a process is not in the SuperLU group, the following values are assigned to it: grid->comm = MPI_COMM_NULL grid->iam = -1. More...

void	superlu_gridexit3d (gridinfo3d_t *grid)

void	set_default_options_dist (superlu_dist_options_t *)
	Set the default values for the options argument. More...

void	print_options_dist (superlu_dist_options_t *)
	Print the options setting. More...

void	print_sp_ienv_dist (superlu_dist_options_t *)
	Print the blocking parameters. More...

void	Destroy_CompCol_Matrix_dist (SuperMatrix *)

void	Destroy_SuperNode_Matrix_dist (SuperMatrix *)

void	Destroy_SuperMatrix_Store_dist (SuperMatrix *)
	Deallocate the structure pointing to the actual storage of the matrix. More...

void	Destroy_CompCol_Permuted_dist (SuperMatrix *)
	A is of type Stype==NCP. More...

void	Destroy_CompRowLoc_Matrix_dist (SuperMatrix *)

void	Destroy_CompRow_Matrix_dist (SuperMatrix *)

void	sp_colorder (superlu_dist_options_t , SuperMatrix , int_t , int_t , SuperMatrix *)

int	sp_symetree_dist (int_t , int_t , int_t , int_t, int_t )
	Symmetric elimination tree. More...

int	sp_coletree_dist (int_t , int_t , int_t , int_t, int_t, int_t )
	Nonsymmetric elimination tree. More...

void	get_perm_c_dist (int_t, int_t, SuperMatrix , int_t )

void	get_perm_c_batch (superlu_dist_options_t options, int batchCount, handle_t SparseMatrix_handles, int **CpivPtr)
	Gets sparsity permutations for a batch of matrices. More...

void	at_plus_a_dist (const int_t, const int_t, int_t , int_t , int_t , int_t , int_t *)

void	getata_dist (const int_t m, const int_t n, const int_t nz, int_t colptr, int_t rowind, int_t atanz, int_t ata_colptr, int_t *ata_rowind)

void	get_metis_dist (int_t n, int_t bnz, int_t b_colptr, int_t b_rowind, int_t *perm_c)

void	get_colamd_dist (const int m, const int n, const int nnz, int_t colptr, int_t rowind, int_t *perm_c)

int	genmmd_dist_ (int_t , int_t , int_t a, int_t , int_t , int_t , int_t , int_t , int_t , int_t , int_t , int_t )

void	bcast_tree (void , int, MPI_Datatype, int, int, gridinfo_t , int, int *)

int_t	symbfact (superlu_dist_options_t , int, SuperMatrix , int_t , int_t , Glu_persist_t , Glu_freeable_t )

int_t	symbfact_SubInit (superlu_dist_options_t options, fact_t, void , int_t, int_t, int_t, int_t, Glu_persist_t , Glu_freeable_t )

int_t	symbfact_SubXpand (int_t, int_t, int_t, MemType, int_t , Glu_freeable_t )

int	symbfact_SubFree (Glu_freeable_t *)

void	countnz_dist (const int_t, int_t , int_t , int_t , Glu_persist_t , Glu_freeable_t *)

int64_t	fixupL_dist (const int_t, const int_t , Glu_persist_t , Glu_freeable_t *)

int_t *	TreePostorder_dist (int_t, int_t *)

float	smach_dist (const char *)

double	dmach_dist (const char *)

void *	superlu_malloc_dist (size_t)

void	superlu_free_dist (void *)

int *	int32Malloc_dist (int)

int *	int32Calloc_dist (int)

int_t *	intMalloc_dist (int_t)

int_t *	intCalloc_dist (int_t)

int	mc64id_dist (int *)

void	arrive_at_ublock (int, int_t , int_t , int , int , int , int_t, int_t, int_t , int_t , int_t , gridinfo_t *)

int_t	estimate_bigu_size (int_t, int_t *, Glu_persist_t , gridinfo_t , int_t , int_t *)

void	superlu_abort_and_exit_dist (char *)

int	sp_ienv_dist (int, superlu_dist_options_t *)

void	ifill_dist (int_t *, int_t, int_t)
	Fills an integer array with a given value. More...

void	super_stats_dist (int_t, int_t *)

void	get_diag_procs (int_t, Glu_persist_t , gridinfo_t , int_t , int_t , int_t *)

int_t	QuerySpace_dist (int_t, int_t, Glu_freeable_t , superlu_dist_mem_usage_t )

int	xerr_dist (char , int )

void	pxerr_dist (char , gridinfo_t , int_t)

void	PStatInit (SuperLUStat_t *)

void	PStatClear (SuperLUStat_t *)

void	PStatFree (SuperLUStat_t *)

void	PStatPrint (superlu_dist_options_t , SuperLUStat_t , gridinfo_t *)

void	log_memory (int64_t, SuperLUStat_t *)

void	print_memorylog (SuperLUStat_t , char )

int	superlu_dist_GetVersionNumber (int , int , int *)

void	quickSort (int_t *, int_t, int_t, int_t)

void	quickSortM (int_t *, int_t, int_t, int_t, int_t, int_t)

int_t	partition (int_t *, int_t, int_t, int_t)

int_t	partitionM (int_t *, int_t, int_t, int_t, int_t, int_t)

int	compareInt_t (void a, void b)
	Compares two integers for equality. More...

int	compareInt (void a, void b)
	Compares two integers for equality. More...

int	compareDouble (void a, void b)
	Compares two doubles for equality. More...

int	dist_checkArrayEq (void arr, int length, MPI_Datatype datatype, int src_rank, int dest_rank, MPI_Comm communicator, int(compare)(void , void ))
	Checks whether arrays at two MPI ranks are identical. More...

float	symbfact_dist (superlu_dist_options_t , int, int, SuperMatrix , int_t , int_t , int_t , int_t , Pslu_freeable_t , MPI_Comm , MPI_Comm , superlu_dist_mem_usage_t )

float	get_perm_c_parmetis (SuperMatrix , int_t , int_t , int, int, int_t , int_t , gridinfo_t , MPI_Comm *)

int_t	psymbfact_LUXpandMem (int, int_t, int_t, int_t, int_t, int, int, int, Pslu_freeable_t , Llu_symbfact_t , vtcsInfo_symbfact_t , psymbfact_stat_t )

int_t	psymbfact_LUXpand (int_t, int_t, int_t, int_t, int_t , int_t, int_t, int_t, int_t, Pslu_freeable_t , Llu_symbfact_t , vtcsInfo_symbfact_t , psymbfact_stat_t *)

int_t	psymbfact_LUXpand_RL (int_t, int_t, int_t, int_t, int_t, int_t, Pslu_freeable_t , Llu_symbfact_t , vtcsInfo_symbfact_t , psymbfact_stat_t )

int_t	psymbfact_prLUXpand (int_t, int_t, int, Llu_symbfact_t , psymbfact_stat_t )

void	isort (int_t N, int_t ARRAY1, int_t ARRAY2)

void	isort1 (int_t N, int_t *ARRAY)

void	gemm_division_cpu_gpu (superlu_dist_options_t , int , int , int , int, int, int, int *, int, int_t)

int_t	get_gpublas_nb (void)

int_t	get_num_gpu_streams (void)

int	getnGPUStreams (void)

int	get_mpi_process_per_gpu (void)

void	printGPUStats (int nsupers, SuperLUStat_t stat, gridinfo3d_t )

double	estimate_cpu_time (int m, int n, int k)

int	get_thread_per_process (void)

int_t	get_max_buffer_size (void)

int_t	get_min (int_t *, int_t)

int	compare_pair (const void , const void )

int_t	static_partition (struct superlu_pair , int_t, int_t , int_t, int_t , int_t , int)

int	get_acc_offload (superlu_dist_options_t *)

int	get_acc_solve (void)

int	get_new3dsolve (void)

int	get_new3dsolvetreecomm (void)

void	print_panel_seg_dist (int_t, int_t, int_t, int_t, int_t , int_t )
	Diagnostic print of segment info after panel_dfs(). More...

void	check_repfnz_dist (int_t, int_t, int_t, int_t *)
	Check whether repfnz[] == SLU_EMPTY after reset. More...

int_t	CheckZeroDiagonal (int_t, int_t , int_t , int_t *)

int	check_perm_dist (char what, int_t n, int_t perm)

void	PrintDouble5 (char , int_t, double )

void	PrintInt10 (char , int_t, int_t )

void	PrintInt32 (char , int, int )

int	file_PrintInt10 (FILE , char , int_t, int_t *)

int	file_PrintInt32 (FILE , char , int, int *)

int	file_PrintLong10 (FILE , char , int_t, int_t *)

void	C_RdTree_Create_nv (C_Tree tree, MPI_Comm comm, int ranks, int rank_cnt, int msgSize, char precision, int needrecvrd, int needsendrd)

void	C_RdTree_Nullify (C_Tree *tree)

yes_no_t	C_RdTree_IsRoot (C_Tree *tree)

void	C_RdTree_forwardMessageSimple (C_Tree Tree, void localBuffer, int msgSize)

void	C_RdTree_waitSendRequest (C_Tree *Tree)

void	C_BcTree_Create_nv (C_Tree tree, MPI_Comm comm, int ranks, int rank_cnt, int msgSize, char precision, int *needrecv)

void	C_BcTree_Nullify (C_Tree *tree)

yes_no_t	C_BcTree_IsRoot (C_Tree *tree)

void	C_BcTree_forwardMessageSimple (C_Tree tree, void localBuffer, int msgSize)

void	C_BcTree_waitSendRequest (C_Tree *tree)

void	DistPrint (char function_name, double value, char Units, gridinfo_t *grid)

void	DistPrint3D (char function_name, double value, char Units, gridinfo3d_t *grid3d)

void	treeImbalance3D (gridinfo3d_t grid3d, SCT_t SCT)

void	slu_SCT_printComm3D (gridinfo3d_t grid3d, SCT_t SCT)

int_t	zAllocBcast (int_t size, void *ptr, gridinfo3d_t grid3d)

int_t	zAllocBcast_gridID (int_t size, void *ptr, int_t gridID, gridinfo3d_t grid3d)

void	permCol_SymbolicFact3d (superlu_dist_options_t options, int n, SuperMatrix GA, int_t perm_c, int_t etree, Glu_persist_t Glu_persist, Glu_freeable_t Glu_freeable, SuperLUStat_t stat, superlu_dist_mem_usage_t symb_mem_usage, gridinfo3d_t *grid3d)
	This function performs the symbolic factorization on matrix PcPrA*Pc' and sets up the nonzero data structures for L & U matrices. In the process, the matrix is also ordered and its memory usage information is fetched. More...

SupernodeToGridMap_t *	createSuperGridMap (int_t nsuper, int_t maxLvl, int_t myTreeIdxs, int_t myZeroTrIdxs, int_t gNodeCount, int_t *gNodeLists)

int_t *	createSupernode2TreeMap (int_t nsupers, int_t maxLvl, int_t gNodeCount, int_t *gNodeLists)

void	allocBcastArray (void **array, int_t size, int root, MPI_Comm comm)
	Allocates and broadcasts an array in a MPI environment. More...

void	allocBcastLargeArray (void **array, int64_t size, int root, MPI_Comm comm)

int_t *	create_iperm_c_supno (int_t nsupers, superlu_dist_options_t options, Glu_persist_t Glu_persist, int_t etree, int_t Lrowind_bc_ptr, int_t Ufstnz_br_ptr, gridinfo3d_t grid3d)

gEtreeInfo_t	fillEtreeInfo (int_t nsupers, int_t setree, treeList_t treeList)

sForest_t **	compute_sForests (int_t nsupers, Glu_persist_t Glu_persist, int_t etree, gridinfo3d_t *grid3d)

int *	getBrecvTree (int_t nlb, sForest_t sforest, int bmod, gridinfo_t *grid)

int *	getBrecvTree_newsolve (int_t nlb, int_t nsupers, int supernodeMask, int bmod, gridinfo_t *grid)

int	getNrootUsolveTree (int_t nbrecvmod, sForest_t sforest, int brecv, int bmod, gridinfo_t *grid)

int	getNbrecvX (sForest_t sforest, int_t Urbs, gridinfo_t *grid)

int	getNbrecvX_newsolve (int_t nsupers, int supernodeMask, int_t Urbs, Ucb_indptr_t *Ucb_indptr, gridinfo_t grid)

int	getNrootUsolveTree_newsolve (int_t nbrecvmod, int_t nsupers, int supernodeMask, int brecv, int bmod, gridinfo_t *grid)

int_t	getNfrecvmodLeaf (int nleaf, sForest_t sforest, int frecv, int fmod, gridinfo_t *grid)

int_t	getNfrecvmod_newsolve (int nleaf, int_t nsupers, int supernodeMask, int frecv, int fmod, gridinfo_t *grid)

int *	getfrecv_newsolve (int_t nsupers, int supernodeMask, int_t nlb, int fmod, int mod_bit, gridinfo_t grid)

int *	getfrecvLeaf (sForest_t sforest, int_t nlb, int fmod, int mod_bit, gridinfo_t grid)

int	getNfrecvx_newsolve (int_t nsupers, int supernodeMask, int_t Lrowind_bc_ptr, int_t Lindval_loc_bc_ptr, gridinfo_t grid)

int	getNfrecvxLeaf (sForest_t sforest, int_t Lrowind_bc_ptr, gridinfo_t grid)

int *	getfmod_newsolve (int_t nlb, int_t nsupers, int supernodeMask, int_t Lrowind_bc_ptr, int_t Lindval_loc_bc_ptr, gridinfo_t grid)

int *	getfmodLeaf (int_t nlb, int *fmod_i)

int	getldu (int_t knsupc, int_t iklrow, int_t *usub)

int *	getBmod3d (int_t treeId, int_t nlb, sForest_t sforest, int_t xsup, int_t *Ufstnz_br_ptr, int_t supernode2treeMap, gridinfo_t *grid)

int *	getBmod3d_newsolve (int_t nlb, int_t nsupers, int supernodeMask, int_t xsup, int_t *Ufstnz_br_ptr, gridinfo_t grid)

int_t *	getPerm_c_supno (int_t nsupers, superlu_dist_options_t , int_t etree, Glu_persist_t Glu_persist, int_t Lrowind_bc_ptr, int_t Ufstnz_br_ptr, gridinfo_t )

int_t *	getPerm_c_supno_allgrid (int_t nsupers, superlu_dist_options_t options, int_t etree, Glu_persist_t Glu_persist, int_t Lrowind_bc_ptr, int_t Ufstnz_br_ptr, gridinfo3d_t grid3d)

void	slu_SCT_init (SCT_t *)

void	slu_SCT_print (gridinfo_t grid, SCT_t SCT)

void	slu_SCT_printSummary (gridinfo_t grid, SCT_t SCT)

void	slu_SCT_print3D (gridinfo3d_t grid3d, SCT_t SCT)

void	slu_SCT_free (SCT_t *)

treeList_t *	setree2list (int_t nsuper, int_t *setree)

int	free_treelist (int_t nsuper, treeList_t *treeList)

int_t	calcTreeWeight (int_t nsupers, int_t setree, treeList_t treeList, int_t *xsup)

int_t	getDescendList (int_t k, int_t dlist, treeList_t treeList)

int_t	getCommonAncestorList (int_t k, int_t alist, int_t seTree, treeList_t *treeList)

int_t	getCommonAncsCount (int_t k, treeList_t *treeList)

int_t *	getPermNodeList (int_t nnode, int_t nlist, int_t perm_c_sup, int_t *iperm_c_sup)

int_t *	getEtreeLB (int_t nnodes, int_t perm_l, int_t gTopOrder)

int_t *	getSubTreeRoots (int_t k, int_t numSubtrees, treeList_t treeList)

int_t *	merg_perms (int_t nperms, int_t nnodes, int_t *perms)

int_t *	getGlobal_iperm (int_t nsupers, int_t nperms, int_t *perms, int_t nnodes)

int_t	log2i (int_t index)

int_t *	supernodal_etree (int_t nsuper, int_t etree, int_t supno, int_t *xsup)

int_t	testSubtreeNodelist (int_t nsupers, int_t numList, int_t *nodeList, int_t nodeCount)

int_t	testListPerm (int_t nodeCount, int_t nodeList, int_t permList, int_t *gTopLevel)

int_t *	topological_ordering (int_t nsuper, int_t *setree)

int_t *	Etree_LevelBoundry (int_t perm, int_t tsort_etree, int_t nsuper)

int_t *	calculate_num_children (int_t nsuper, int_t *setree)

void	Print_EtreeLevelBoundry (int_t *Etree_LvlBdry, int_t max_level, int_t nsuper)

void	print_etree_leveled (int_t setree, int_t tsort_etree, int_t nsuper)

void	print_etree (int_t setree, int_t iperm, int_t nsuper)

int_t	printFileList (char sname, int_t nnodes, int_t dlist, int_t *setree)

int *	getLastDepBtree (int_t nsupers, treeList_t *treeList)

int_t *	getReplicatedTrees (gridinfo3d_t *grid3d)

int_t *	getGridTrees (gridinfo3d_t *grid3d)

int_t **	getNodeList (int_t maxLvl, int_t setree, int_t nnodes, int_t treeHeads, treeList_t treeList)

int_t *	calcNumNodes (int_t maxLvl, int_t treeHeads, treeList_t treeList)

int_t *	getTreeHeads (int_t maxLvl, int_t nsupers, treeList_t *treeList)

int_t *	getMyIperm (int_t nnodes, int_t nsupers, int_t *myPerm)

int_t *	getMyTopOrder (int_t nnodes, int_t myPerm, int_t myIperm, int_t *setree)

int_t *	getMyEtLims (int_t nnodes, int_t *myTopOrder)

treeTopoInfo_t	getMyTreeTopoInfo (int_t nnodes, int_t nsupers, int_t myPerm, int_t setree)

sForest_t **	getNestDissForests (int_t maxLvl, int_t nsupers, int_t setree, treeList_t treeList)

int_t **	getTreePermForest (int_t myTreeIdxs, int_t myZeroTrIdxs, sForest_t sForests, int_t perm_c_supno, int_t iperm_c_supno, gridinfo3d_t grid3d)

int_t **	getTreePermFr (int_t myTreeIdxs, sForest_t sForests, gridinfo3d_t grid3d)

int_t *	getMyNodeCountsFr (int_t maxLvl, int_t myTreeIdxs, sForest_t *sForests)

int_t **	getNodeListFr (int_t maxLvl, sForest_t **sForests)

int_t *	getNodeCountsFr (int_t maxLvl, sForest_t **sForests)

int *	getIsNodeInMyGrid (int_t nsupers, int_t maxLvl, int_t myNodeCount, int_t *treePerm)

void	printForestWeightCost (sForest_t *sForests, SCT_t SCT, gridinfo3d_t *grid3d)

sForest_t **	getGreedyLoadBalForests (int_t maxLvl, int_t nsupers, int_t setree, treeList_t treeList)

sForest_t **	getForests (int_t maxLvl, int_t nsupers, int_t setree, treeList_t treeList)

int_t	getBigUSize (superlu_dist_options_t , int_t nsupers, gridinfo_t grid, int_t **Lrowind_bc_ptr)

void	getSCUweight (int_t nsupers, treeList_t treeList, int_t xsup, int_t Lrowind_bc_ptr, int_t Ufstnz_br_ptr, gridinfo3d_t *grid3d)

void	getSCUweight_allgrid (int_t nsupers, treeList_t treeList, int_t xsup, int_t Lrowind_bc_ptr, int_t Ufstnz_br_ptr, gridinfo3d_t *grid3d)

int	Wait_LUDiagSend (int_t k, MPI_Request U_diag_blk_send_req, MPI_Request L_diag_blk_send_req, gridinfo_t grid, SCT_t SCT)

void	applyRowPerm (int_t colptr, int_t rowind, int_t *perm_r, int_t n)

int	getNsupers (int n, Glu_persist_t *Glu_persist)

int	set_tag_ub (void)

int	getNumThreads (int)

int_t	num_full_cols_U (int_t kk, int_t *Ufstnz_br_ptr, int_t xsup, gridinfo_t , int_t , int_t *)

int_t *	getFactPerm (int_t)

int_t *	getFactIperm (int_t *, int_t)

int_t	initCommRequests (commRequests_t comReqs, gridinfo_t grid)

int_t	initFactStat (int nsupers, factStat_t *factStat)

int	freeFactStat (factStat_t *factStat)

int_t	initFactNodelists (int_t, int_t, int_t, factNodelists_t *)

int	freeFactNodelists (factNodelists_t *fNlists)

int_t	initMsgs (msgs_t *msgs)

int_t	getNumLookAhead (superlu_dist_options_t *)

commRequests_t **	initCommRequestsArr (int_t mxLeafNode, int_t ldt, gridinfo_t *grid)

int	freeCommRequestsArr (int_t mxLeafNode, commRequests_t **comReqss)

msgs_t **	initMsgsArr (int_t numLA)

int	freeMsgsArr (int_t numLA, msgs_t **msgss)

int_t	Trs2_InitUblock_info (int_t klst, int_t nb, Ublock_info_t , int_t usub, Glu_persist_t , SuperLUStat_t )

int	Cmpfunc_R_info (const void a, const void b)

int	Cmpfunc_U_info (const void a, const void b)

int	sort_R_info (Remain_info_t *Remain_info, int n)

int	sort_U_info (Ublock_info_t *Ublock_info, int n)

int	sort_R_info_elm (Remain_info_t *Remain_info, int n)

int	sort_U_info_elm (Ublock_info_t *Ublock_info, int n)

void	printTRStimer (xtrsTimer_t xtrsTimer, gridinfo3d_t grid3d)

void	initTRStimer (xtrsTimer_t xtrsTimer, gridinfo_t grid)

int_t **	getTreePerm (int_t myTreeIdxs, int_t myZeroTrIdxs, int_t nodeCount, int_t nodeList, int_t perm_c_supno, int_t iperm_c_supno, gridinfo3d_t grid3d)

int_t *	getMyNodeCounts (int_t maxLvl, int_t myTreeIdxs, int_t gNodeCount)

int_t	checkIntVector3d (int_t vec, int_t len, gridinfo3d_t grid3d)

int_t	reduceStat (PhaseType PHASE, SuperLUStat_t stat, gridinfo3d_t grid3d)

int_t	Wait_LSend (int_t k, gridinfo_t grid, int ToSendR, MPI_Request s, SCT_t *)

int_t	Wait_USend (MPI_Request , gridinfo_t , SCT_t *)

int_t	Check_LRecv (MPI_Request , int msgcnt)

int_t	Wait_UDiagBlockSend (MPI_Request , gridinfo_t , SCT_t *)

int_t	Wait_LDiagBlockSend (MPI_Request , gridinfo_t , SCT_t *)

int_t	Wait_UDiagBlock_Recv (MPI_Request , SCT_t )

int_t	Test_UDiagBlock_Recv (MPI_Request , SCT_t )

int_t	Wait_LDiagBlock_Recv (MPI_Request , SCT_t )

int_t	Test_LDiagBlock_Recv (MPI_Request , SCT_t )

int_t	LDiagBlockRecvWait (int_t k, int factored_U, MPI_Request , gridinfo_t *)

int_t	num_full_cols_U_mod (int_t kk, int_t usub, int_t xsup, gridinfo_t grid, int_t perm_u, int_t *ldu)

Variables
static const int	BC_L =1

static const int	RD_L =2

static const int	BC_U =3

static const int	RD_U =4

Detailed Description

Definitions which are precision-neutral.

The source code is distributed under BSD license, see the file License.txt at the top-level directory.

-- Distributed SuperLU routine (version 9.0) --
Lawrence Berkeley National Lab, Univ. of California Berkeley.
November 1, 2007

Modified:
    February 20, 2008
    October 11, 2014
    September 18, 2018  version 6.0
    February 8, 2019    version 6.1.1
    November 12, 2019   version 6.2.0
    October 23, 2020    version 6.4.0
    May 12, 2021        version 7.0.0
    October 5, 2021     version 7.1.0
    October 18, 2021    version 7.1.1
    December 12, 2021   version 7.2.0
    May 22, 2022        version 8.0.0
    July 5, 2022        version 8.1.0
    October 1, 2022     version 8.1.1
    November 12, 2022   version 8.1.2
    November 17, 2023   version 8.2.1
    May 8, 2024         version 9.0.0

Macro Definition Documentation

◆ __SUPERLU_ASYNC_TREE

#define __SUPERLU_ASYNC_TREE

◆ BC_HEADER

#define BC_HEADER 2

◆ BC_HEADER_NEWU

#define BC_HEADER_NEWU 3

◆ BlockNum

#define BlockNum ( i ) ( supno[i] )

◆ BR_HEADER

#define BR_HEADER 3

◆ CEILING

#define CEILING	(	a,
		b
	)	( ((a)%(b)) ? ((a)/(b) + 1) : ((a)/(b)) )

◆ cmax

#define cmax	(	a,
		b
	)	((a) > (b) ? (a) : (b))

◆ COMM_ALL

#define COMM_ALL 100

◆ COMM_COLUMN

#define COMM_COLUMN 101

◆ COMM_ROW

#define COMM_ROW 102

◆ DEG_TREE

#define DEG_TREE 2

◆ DIM_X

#define DIM_X 32

◆ DIM_XA

#define DIM_XA DIM_X

◆ DIM_XB

#define DIM_XB DIM_X

◆ DIM_Y

#define DIM_Y 16

◆ DIM_YA

#define DIM_YA DIM_Y

◆ DIM_YB

#define DIM_YB DIM_Y

◆ fetch

#define fetch	(	A,
		m,
		n,
		bound
	)	offs_d##A[min(n*LD##A+m, bound)]

◆ fma

#define fma	(	A,
		B,
		C
	)	C += (A*B)

◆ FstBlockC

#define FstBlockC ( bnum ) ( xsup[bnum] )

◆ GPU_ACC

#define GPU_ACC

◆ GSUM

#define GSUM 20

◆ HANDLE_SIZE

#define HANDLE_SIZE 8

◆ IAM

#define IAM ( comm ) { int rank; MPI_Comm_rank ( comm, &rank ); rank};

◆ IFMT

#define IFMT " %lld"

◆ ISORT

#define ISORT /* NOTE: qsort() has bug on Mac */

◆ LB_DESCRIPTOR

#define LB_DESCRIPTOR 2

◆ LBi

#define LBi	(	bnum,
		grid
	)	( (bnum)/grid->nprow )/* Global to local block rowwise */

◆ LBj

#define LBj	(	bnum,
		grid
	)	( (bnum)/grid->npcol )/* Global to local block columnwise*/

◆ LkkDIAG

#define LkkDIAG 15

◆ LkSUB

#define LkSUB 13

◆ LkVAL

#define LkVAL 14

◆ LOG2

#define LOG2 ( x ) (log10((double) x) / log10(2.0))

◆ LSUM

#define LSUM 23

◆ LSUM_BLK

#define LSUM_BLK ( i ) ilsum[i] * nrhs + (i+1) * LSUM_H

◆ LSUM_H

#define LSUM_H 2 /* The header preceding each MOD block. */

◆ MAGMA_CONST

#define MAGMA_CONST

◆ MAX_SUPER_SIZE

#define MAX_SUPER_SIZE 512 /* Sherry: moved from superlu_gpu.cu */

◆ mpi_int_t

#define mpi_int_t MPI_LONG_LONG_INT

◆ MYCOL

#define MYCOL	(	iam,
		grid
	)	( (iam) % grid->npcol )

◆ MYROW

#define MYROW	(	iam,
		grid
	)	( (iam) / grid->npcol )

◆ NBUFFERS

#define NBUFFERS 5

◆ NO_MARKER

#define NO_MARKER 3

◆ NTAGS

#define NTAGS INT_MAX

◆ NWARP

#define NWARP DIM_X*DIM_Y/WARP_SIZE

◆ PCOL

#define PCOL	(	bnum,
		grid
	)	( (bnum) % grid->npcol )

◆ PNUM

#define PNUM	(	i,
		j,
		grid
	)	( (i)grid->npcol + j ) / Process number at coord(i,j) */

◆ PROW

#define PROW	(	bnum,
		grid
	)	( (bnum) % grid->nprow )

◆ RHS_ITERATE

#define RHS_ITERATE ( i ) for (i = 0; i < nrhs; ++i)

◆ SLU_MPI_TAG

#define SLU_MPI_TAG	(	id,
		num
	)	( (6*(num)+id) % tag_ub )

◆ SUPER_BLOCK

#define SUPER_BLOCK 12

◆ SUPER_LINEAR

#define SUPER_LINEAR 11

◆ SUPERLU_DIST_EXPORT

#define SUPERLU_DIST_EXPORT

◆ SUPERLU_DIST_MAJOR_VERSION

#define SUPERLU_DIST_MAJOR_VERSION 9

◆ SUPERLU_DIST_MINOR_VERSION

#define SUPERLU_DIST_MINOR_VERSION 0

◆ SUPERLU_DIST_PATCH_VERSION

#define SUPERLU_DIST_PATCH_VERSION 0

◆ SUPERLU_DIST_RELEASE_DATE

#define SUPERLU_DIST_RELEASE_DATE "May 8, 2024"

◆ SuperLU_MPI_COMPLEX

#define SuperLU_MPI_COMPLEX MPI_C_COMPLEX

◆ SuperLU_MPI_DOUBLE_COMPLEX

#define SuperLU_MPI_DOUBLE_COMPLEX MPI_C_DOUBLE_COMPLEX

◆ SuperLU_timer_

double SuperLU_timer_ SuperLU_timer_dist_

◆ SuperSize

#define SuperSize ( bnum ) ( xsup[bnum+1]-xsup[bnum] )

◆ THR_M

#define THR_M ( BLK_M / DIM_X )

◆ THR_N

#define THR_N ( BLK_N / DIM_Y )

◆ UB_DESCRIPTOR

#define UB_DESCRIPTOR 2

◆ UB_DESCRIPTOR_NEWU

#define UB_DESCRIPTOR_NEWU 2

◆ UB_DESCRIPTOR_NEWUCPP

#define UB_DESCRIPTOR_NEWUCPP 3

◆ UjROW

#define UjROW 10

◆ UkSUB

#define UkSUB 11

◆ UkVAL

#define UkVAL 12

◆ VT_TRACEOFF

#define VT_TRACEOFF

◆ VT_TRACEON

#define VT_TRACEON

◆ WARP_SIZE

#define WARP_SIZE 32

◆ X_BLK

#define X_BLK ( i ) ilsum[i] * nrhs + (i+1) * XK_H

◆ Xk

#define Xk 21

◆ XK_H

#define XK_H 2 /* The header preceding each X block. */

◆ XT_BLK

#define XT_BLK ( i ) ilsumT[i] * nrhs + (i+1) * XK_H

◆ Yk

#define Yk 22

Typedef Documentation

◆ handle_t

typedef int64_t handle_t

◆ int_t

typedef int64_t int_t

◆ treePartStrat

typedef enum treePartStrat treePartStrat

◆ trtype_t

typedef enum trtype_t trtype_t

◆ xtrsTimer_t

typedef struct xtrsTimer_t xtrsTimer_t

Enumeration Type Documentation

◆ SupernodeToGridMap_t

enum SupernodeToGridMap_t

Enumerator
NOT_IN_GRID
IN_GRID_ZERO
IN_GRID_AIJ

◆ treePartStrat

enum treePartStrat

Enumerator
ND
GD

◆ trtype_t

enum trtype_t

Enumerator
UPPER_TRI
LOWER_TRI

Function Documentation

◆ allocBcastArray()

void allocBcastArray	(	void **	array,
		int_t	size,
		int	root,
		MPI_Comm	comm
	)

Allocates and broadcasts an array in a MPI environment.

This function sends the size from the root process to all other processes in the communicator. If the process is not the root, it receives the size from the root and allocates the array. Then, the function broadcasts the array from the root process to all other processes in the communicator.

Parameters

array	Pointer to the array to be allocated and broadcasted.
size	The size of the array.
comm	The MPI communicator.
root	The root process.

Here is the caller graph for this function:

◆ allocBcastLargeArray()

void allocBcastLargeArray	(	void **	array,
		int64_t	size,
		int	root,
		MPI_Comm	comm
	)

Here is the caller graph for this function:

◆ applyRowPerm()

void applyRowPerm	(	int_t *	colptr,
		int_t *	rowind,
		int_t *	perm_r,
		int_t	n
	)

Performs a row permutation operation on a sparse matrix (CSC format) using a user-provided permutation array.

Parameters

colptr	The column pointer array of the sparse matrix (CSC format).
rowind	The row index array of the sparse matrix (CSC format).
perm_r	The user-provided permutation array for the rows.
n	The number of columns in the sparse matrix.

Here is the caller graph for this function:

◆ arrive_at_ublock()

void arrive_at_ublock	(	int	j,
		int_t *	iukp,
		int_t *	rukp,
		int *	jb,
		int *	ljb,
		int *	nsupc,
		int_t	iukp0,
		int_t	rukp0,
		int_t *	usub,
		int_t *	perm_u,
		int_t *	xsup,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ at_plus_a_dist()

void at_plus_a_dist	(	const int_t	n,
		const int_t	nz,
		int_t *	colptr,
		int_t *	rowind,
		int_t *	bnz,
		int_t **	b_colptr,
		int_t **	b_rowind
	)

Purpose
=======

Form the structure of A'+A. A is an n-by-n matrix in column oriented
format represented by (colptr, rowind). The output A'+A is in column
oriented format (symmetrically, also row oriented), represented by
(b_colptr, b_rowind).

Here is the caller graph for this function:

◆ bcast_tree()

void bcast_tree	(	void *	buf,
		int	count,
		MPI_Datatype	dtype,
		int	root,
		int	tag,
		gridinfo_t *	grid,
		int	scope,
		int *	recvcnt
	)

Purpose
=======
  Broadcast an array of *dtype* numbers. The communication pattern
  is a tree with number of branches equal to NBRANCHES.
  The process ranks are between 0 and Np-1.

  The following two pairs of graphs give different ways of viewing the same
  algorithm.  The first pair shows the trees as they should be visualized
  when examining the algorithm.  The second pair are isomorphic graphs of
  of the first, which show the actual pattern of data movement.
  Note that a tree broadcast with NBRANCHES = 2 is isomorphic with a
  hypercube broadcast (however, it does not require the nodes be a
  power of two to work).

   TREE BROADCAST, NBRANCHES = 2     *    TREE BROADCAST, NBRANCHES = 3

    root=2
i=4   &______________                *
      |              \               *       root=2
i=2   &______         &______        * i=3     &______________________
      |      \        |      \       *         |          \           \
i=1   &__     &__     &__     &__    * i=1     &______     &______     &__
      |  \    |  \    |  \    |  \   *         |  \   \    |  \   \    |  \
      2   3   4   5   6   7   0   1  *         2   3   4   5   6   7   0   1


         ISOMORPHIC GRAPHS OF ABOVE, SHOWN IN MORE FAMILIAR TERMS:

               2                                           2
      _________|_________                       ___________|____________
     /         |         \                     /           |      |     \
    6          4          3                   5            0      3      4
   / \         |                             / \           |
  0   7        5                            6   7          1
  |
  1


Arguments
=========

scope

◆ C_BcTree_Create_nv()

void C_BcTree_Create_nv	(	C_Tree *	tree,
		MPI_Comm	comm,
		int *	ranks,
		int	rank_cnt,
		int	msgSize,
		char	precision,
		int *	needrecv
	)

Here is the caller graph for this function:

◆ C_BcTree_forwardMessageSimple()

void C_BcTree_forwardMessageSimple	(	C_Tree *	tree,
		void *	localBuffer,
		int	msgSize
	)

Here is the caller graph for this function:

◆ C_BcTree_IsRoot()

yes_no_t C_BcTree_IsRoot ( C_Tree * tree )

Here is the caller graph for this function:

◆ C_BcTree_Nullify()

void C_BcTree_Nullify ( C_Tree * tree )

Here is the caller graph for this function:

◆ C_BcTree_waitSendRequest()

void C_BcTree_waitSendRequest ( C_Tree * tree )

Here is the caller graph for this function:

◆ C_RdTree_Create_nv()

void C_RdTree_Create_nv	(	C_Tree *	tree,
		MPI_Comm	comm,
		int *	ranks,
		int	rank_cnt,
		int	msgSize,
		char	precision,
		int *	needrecvrd,
		int *	needsendrd
	)

Here is the caller graph for this function:

◆ C_RdTree_forwardMessageSimple()

void C_RdTree_forwardMessageSimple	(	C_Tree *	Tree,
		void *	localBuffer,
		int	msgSize
	)

Here is the caller graph for this function:

◆ C_RdTree_IsRoot()

yes_no_t C_RdTree_IsRoot ( C_Tree * tree )

Here is the caller graph for this function:

◆ C_RdTree_Nullify()

void C_RdTree_Nullify ( C_Tree * tree )

Here is the caller graph for this function:

◆ C_RdTree_waitSendRequest()

void C_RdTree_waitSendRequest ( C_Tree * Tree )

Here is the caller graph for this function:

◆ calcNumNodes()

int_t * calcNumNodes	(	int_t	maxLvl,
		int_t *	treeHeads,
		treeList_t *	treeList
	)

Here is the caller graph for this function:

◆ calcTreeWeight()

int_t calcTreeWeight	(	int_t	nsupers,
		int_t *	setree,
		treeList_t *	treeList,
		int_t *	xsup
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ calculate_num_children()

int_t * calculate_num_children	(	int_t	nsuper,
		int_t *	setree
	)

Here is the call graph for this function:

◆ Check_LRecv()

int_t Check_LRecv	(	MPI_Request *	recv_req,
		int *	msgcnt
	)

◆ check_perm_dist()

int check_perm_dist	(	char *	what,
		int_t	n,
		int_t *	perm
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ check_repfnz_dist()

void check_repfnz_dist	(	int_t	n,
		int_t	w,
		int_t	jcol,
		int_t *	repfnz
	)

Check whether repfnz[] == SLU_EMPTY after reset.

◆ checkIntVector3d()

int_t checkIntVector3d	(	int_t *	vec,
		int_t	len,
		gridinfo3d_t *	grid3d
	)

Here is the call graph for this function:

◆ CheckZeroDiagonal()

int_t CheckZeroDiagonal	(	int_t	n,
		int_t *	rowind,
		int_t *	colbeg,
		int_t *	colcnt
	)

◆ Cmpfunc_R_info()

int Cmpfunc_R_info	(	const void *	a,
		const void *	b
	)

Here is the caller graph for this function:

◆ Cmpfunc_U_info()

int Cmpfunc_U_info	(	const void *	a,
		const void *	b
	)

Here is the caller graph for this function:

◆ compare_pair()

int compare_pair	(	const void *	a,
		const void *	b
	)

Here is the caller graph for this function:

◆ compareDouble()

int compareDouble	(	void *	a,
		void *	b
	)

Compares two doubles for equality.

Parameters

a	Void pointer to the first double
b	Void pointer to the second double

Returns: int Returns 0 if the doubles are equal, 1 otherwise

◆ compareInt()

int compareInt	(	void *	a,
		void *	b
	)

Compares two integers for equality.

Parameters

a	Void pointer to the first integer
b	Void pointer to the second integer

Returns: int Returns 0 if the integers are equal, 1 otherwise

◆ compareInt_t()

int compareInt_t	(	void *	a,
		void *	b
	)

Compares two integers for equality.

Parameters

a	Void pointer to the first integer
b	Void pointer to the second integer

Returns: int Returns 0 if the integers are equal, 1 otherwise

◆ compute_sForests()

sForest_t ** compute_sForests	(	int_t	nsupers,
		Glu_persist_t *	Glu_persist,
		int_t *	etree,
		gridinfo3d_t *	grid3d
	)

Here is the call graph for this function:

◆ countnz_dist()

void countnz_dist	(	const int_t	n,
		int_t *	xprune,
		int_t *	nnzL,
		int_t *	nnzU,
		Glu_persist_t *	Glu_persist,
		Glu_freeable_t *	Glu_freeable
	)

Count the total number of nonzeros in factors L and U,  and in the 
symmetrically reduced L.

Here is the caller graph for this function:

◆ create_iperm_c_supno()

int_t * create_iperm_c_supno	(	int_t	nsupers,
		superlu_dist_options_t *	options,
		Glu_persist_t *	Glu_persist,
		int_t *	etree,
		int_t **	Lrowind_bc_ptr,
		int_t **	Ufstnz_br_ptr,
		gridinfo3d_t *	grid3d
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ createSuperGridMap()

SupernodeToGridMap_t * createSuperGridMap	(	int_t	nsuper,
		int_t	maxLvl,
		int_t *	myTreeIdxs,
		int_t *	myZeroTrIdxs,
		int_t *	gNodeCount,
		int_t **	gNodeLists
	)

Here is the caller graph for this function:

◆ createSupernode2TreeMap()

int_t * createSupernode2TreeMap	(	int_t	nsupers,
		int_t	maxLvl,
		int_t *	gNodeCount,
		int_t **	gNodeLists
	)

Here is the caller graph for this function:

◆ Destroy_CompCol_Matrix_dist()

void Destroy_CompCol_Matrix_dist ( SuperMatrix * A )

Here is the caller graph for this function:

◆ Destroy_CompCol_Permuted_dist()

void Destroy_CompCol_Permuted_dist ( SuperMatrix * A )

A is of type Stype==NCP.

Here is the caller graph for this function:

◆ Destroy_CompRow_Matrix_dist()

void Destroy_CompRow_Matrix_dist ( SuperMatrix * A )

◆ Destroy_CompRowLoc_Matrix_dist()

void Destroy_CompRowLoc_Matrix_dist ( SuperMatrix * A )

Here is the caller graph for this function:

◆ Destroy_SuperMatrix_Store_dist()

void Destroy_SuperMatrix_Store_dist ( SuperMatrix * A )

Deallocate the structure pointing to the actual storage of the matrix.

Here is the caller graph for this function:

◆ Destroy_SuperNode_Matrix_dist()

void Destroy_SuperNode_Matrix_dist ( SuperMatrix * A )

◆ dist_checkArrayEq()

int dist_checkArrayEq	(	void *	arr,
		int	length,
		MPI_Datatype	datatype,
		int	src_rank,
		int	dest_rank,
		MPI_Comm	communicator,
		int()(void , void *)	compare
	)

Checks whether arrays at two MPI ranks are identical.

This function is used to check if a copy of an array at two different MPI ranks are the same. It uses MPI_Send and MPI_Recv to transfer data between ranks, then compares the arrays.

Parameters

arr	Void pointer to the array to be compared
length	The length of the array
datatype	MPI_Datatype of the array elements
src_rank	The source rank that has the original array
dest_rank	The destination rank that has the copied array
communicator	The MPI_Comm communicator that includes both ranks
compare	A function pointer to the function used to compare elements. Should take two void pointers and return 0 if they are equal and a non-zero value otherwise.

Returns: int Returns 0 if arrays are identical, 1 otherwise.

◆ DistPrint()

void DistPrint	(	char *	function_name,
		double	value,
		char *	Units,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ DistPrint3D()

void DistPrint3D	(	char *	function_name,
		double	value,
		char *	Units,
		gridinfo3d_t *	grid3d
	)

Here is the caller graph for this function:

◆ dmach_dist()

double dmach_dist ( const char * cmach )

Here is the caller graph for this function:

◆ estimate_bigu_size()

int_t estimate_bigu_size	(	int_t	nsupers,
		int_t **	Ufstnz_br_ptr,
		Glu_persist_t *	Glu_persist,
		gridinfo_t *	grid,
		int_t *	perm_u,
		int_t *	max_ncols
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ estimate_cpu_time()

double estimate_cpu_time	(	int	m,
		int	n,
		int	k
	)

Here is the caller graph for this function:

◆ Etree_LevelBoundry()

int_t * Etree_LevelBoundry	(	int_t *	perm,
		int_t *	tsort_etree,
		int_t	nsuper
	)

Here is the call graph for this function:

◆ file_PrintInt10()

int file_PrintInt10	(	FILE *	fp,
		char *	name,
		int_t	len,
		int_t *	x
	)

Here is the caller graph for this function:

◆ file_PrintInt32()

int file_PrintInt32	(	FILE *	fp,
		char *	name,
		int	len,
		int *	x
	)

Here is the caller graph for this function:

◆ file_PrintLong10()

int file_PrintLong10	(	FILE *	,
		char *	,
		int_t	,
		int_t *
	)

◆ fillEtreeInfo()

gEtreeInfo_t fillEtreeInfo	(	int_t	nsupers,
		int_t *	setree,
		treeList_t *	treeList
	)

Here is the caller graph for this function:

◆ fixupL_dist()

int64_t fixupL_dist	(	const int_t	n,
		const int_t *	perm_r,
		Glu_persist_t *	Glu_persist,
		Glu_freeable_t *	Glu_freeable
	)

Fix up the data storage lsub[] for L-subscripts. It removes the subscript
sets for structural pruning,    and applies permuation to the remaining
subscripts.

Return value:
  number of entries in lsub[], which includes the size of the pruned graph,
  which is interspersed in the supernodal graph in the lsub[] array.

Fix up the data storage lsub for L-subscripts. It removes the subscript
sets for structural pruning,    and applies permuation to the remaining
subscripts.

Here is the caller graph for this function:

◆ free_treelist()

int free_treelist	(	int_t	nsuper,
		treeList_t *	treeList
	)

Here is the caller graph for this function:

◆ freeCommRequestsArr()

int freeCommRequestsArr	(	int_t	mxLeafNode,
		commRequests_t **	comReqss
	)

Here is the caller graph for this function:

◆ freeFactNodelists()

int freeFactNodelists ( factNodelists_t * fNlists )

Here is the caller graph for this function:

◆ freeFactStat()

int freeFactStat ( factStat_t * factStat )

Here is the caller graph for this function:

◆ freeMsgsArr()

int freeMsgsArr	(	int_t	numLA,
		msgs_t **	msgss
	)

Here is the caller graph for this function:

◆ gemm_division_cpu_gpu()

void gemm_division_cpu_gpu	(	superlu_dist_options_t *	options,
		int *	num_streams_used,
		int *	stream_end_col,
		int *	ncpu_blks,
		int	nbrow,
		int	ldu,
		int	nstreams,
		int *	full_u_cols,
		int	num_blks,
		int_t	gemmBufferSize
	)

Here is the call graph for this function:

◆ genmmd_dist_()

int genmmd_dist_	(	int_t *	neqns,
		int_t *	xadj,
		int_t *	a,
		int_t *	invp,
		int_t *	perm,
		int_t *	delta,
		int_t *	dhead,
		int_t *	qsize,
		int_t *	llist,
		int_t *	marker,
		int_t *	maxint,
		int_t *	nofsub
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ get_acc_offload()

int get_acc_offload ( superlu_dist_options_t * options )

Here is the call graph for this function:

Here is the caller graph for this function:

◆ get_acc_solve()

int get_acc_solve ( void )

Here is the call graph for this function:

Here is the caller graph for this function:

◆ get_colamd_dist()

void get_colamd_dist	(	const int	m,
		const int	n,
		const int	nnz,
		int_t *	colptr,
		int_t *	rowind,
		int_t *	perm_c
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ get_diag_procs()

void get_diag_procs	(	int_t	n,
		Glu_persist_t *	Glu_persist,
		gridinfo_t *	grid,
		int_t *	num_diag_procs,
		int_t **	diag_procs,
		int_t **	diag_len
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ get_gpublas_nb()

int_t get_gpublas_nb ( void )

Here is the caller graph for this function:

◆ get_max_buffer_size()

int_t get_max_buffer_size ( void )

◆ get_metis_dist()

void get_metis_dist	(	int_t	n,
		int_t	bnz,
		int_t *	b_colptr,
		int_t *	b_rowind,
		int_t *	perm_c
	)

Here is the caller graph for this function:

◆ get_min()

int_t get_min	(	int_t *	sums,
		int_t	nprocs
	)

Here is the caller graph for this function:

◆ get_mpi_process_per_gpu()

int get_mpi_process_per_gpu ( void )

Here is the caller graph for this function:

◆ get_new3dsolve()

int get_new3dsolve ( void )

Here is the caller graph for this function:

◆ get_new3dsolvetreecomm()

int get_new3dsolvetreecomm ( void )

Here is the caller graph for this function:

◆ get_num_gpu_streams()

int_t get_num_gpu_streams ( void )

Here is the caller graph for this function:

◆ get_perm_c_batch()

void get_perm_c_batch	(	superlu_dist_options_t *	options,
		int	batchCount,
		handle_t *	SparseMatrix_handles,
		int **	CpivPtr
	)

Gets sparsity permutations for a batch of matrices.


Parameters

  
    [in] options solver options

    [in] batchCount number of matrices in the batch

    [in] SparseMatrix_handles pointers to the matrices in the batch, each pointing to the actual stoage in CSC format
    On entry, the original matrices, may be overwritten by A1 <- Pr*diag(R)*A*diag(C) from dequil_batch() and dpivot_batch()

    [out] CpivPtr pointers to column permutation vectors for each matrix, each of size n

Here is the call graph for this function:

Here is the caller graph for this function:

◆ get_perm_c_dist()

void get_perm_c_dist	(	int_t	pnum,
		int_t	ispec,
		SuperMatrix *	A,
		int_t *	perm_c
	)

Purpose
=======

GET_PERM_C_DIST obtains a permutation matrix Pc, by applying the multiple
minimum degree ordering code by Joseph Liu to matrix A'*A or A+A',
or using approximate minimum degree column ordering by Davis et. al.
The LU factorization of A*Pc tends to have less fill than the LU 
factorization of A.

Arguments
=========

ispec   (input) colperm_t
        Specifies what type of column permutation to use to reduce fill.
        = NATURAL: natural ordering (i.e., Pc = I)
        = MMD_AT_PLUS_A: minimum degree ordering on structure of A'+A
        = MMD_ATA: minimum degree ordering on structure of A'*A
        = METIS_AT_PLUS_A: MeTis on A'+A

A       (input) SuperMatrix*
        Matrix A in A*X=B, of dimension (A->nrow, A->ncol). The number
        of the linear equations is A->nrow. Currently, the type of A 
        can be: Stype = SLU_NC; Dtype = SLU_D; Mtype = SLU_GE.
        In the future, more general A can be handled.

perm_c  (output) int*
    Column permutation vector of size A->ncol, which defines the 
        permutation matrix Pc; perm_c[i] = j means column i of A is 
        in position j in A*Pc.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ get_perm_c_parmetis()

float get_perm_c_parmetis	(	SuperMatrix *	A,
		int_t *	perm_r,
		int_t *	perm_c,
		int	nprocs_i,
		int	noDomains,
		int_t **	sizes,
		int_t **	fstVtxSep,
		gridinfo_t *	grid,
		MPI_Comm *	metis_comm
	)

Purpose
=======

GET_PERM_C_PARMETIS obtains a permutation matrix Pc, by applying a
graph partitioning algorithm to the symmetrized graph A+A'.  The
multilevel graph partitioning algorithm used is the
ParMETIS_V3_NodeND routine available in the parallel graph
partitioning package parMETIS.  

The number of independent sub-domains noDomains computed by this
algorithm has to be a power of 2.  Hence noDomains is the larger
number power of 2 that is smaller than nprocs_i, where nprocs_i = nprow
* npcol is the number of processors used in SuperLU_DIST.

Arguments
=========

A       (input) SuperMatrix*
        Matrix A in A*X=B, of dimension (A->nrow, A->ncol). The number
        of the linear equations is A->nrow.  Matrix A is distributed
        in NRformat_loc format.

perm_r  (input) int_t*
        Row permutation vector of size A->nrow, which defines the 
        permutation matrix Pr; perm_r[i] = j means row i of A is in 
        position j in Pr*A.

perm_c  (output) int_t*
    Column permutation vector of size A->ncol, which defines the 
        permutation matrix Pc; perm_c[i] = j means column i of A is 
        in position j in A*Pc.

nprocs_i (input) int*
        Number of processors the input matrix is distributed on in a block
        row format.  It corresponds to number of processors used in
        SuperLU_DIST.

noDomains (input) int*, must be power of 2
        Number of independent domains to be computed by the graph
        partitioning algorithm.  ( noDomains <= nprocs_i )

sizes   (output) int_t**, of size 2 * noDomains
        Returns pointer to an array containing the number of nodes
        for each sub-domain and each separator.  Separators are stored 
        from left to right.
        Memory for the array is allocated in this routine.

fstVtxSep (output) int_t**, of size 2 * noDomains
        Returns pointer to an array containing first node for each
        sub-domain and each separator.
        Memory for the array is allocated in this routine.

Return value
============
  < 0, number of bytes allocated on return from the symbolic factorization.
  > 0, number of bytes allocated when out of memory.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ get_thread_per_process()

int get_thread_per_process ( void )

Here is the caller graph for this function:

◆ getata_dist()

void getata_dist	(	const int_t	m,
		const int_t	n,
		const int_t	nz,
		int_t *	colptr,
		int_t *	rowind,
		int_t *	atanz,
		int_t **	ata_colptr,
		int_t **	ata_rowind
	)

Purpose
=======

Form the structure of A'*A. A is an m-by-n matrix in column oriented
format represented by (colptr, rowind). The output A'*A is in column
oriented format (symmetrically, also row oriented), represented by
(ata_colptr, ata_rowind).

This routine is modified from GETATA routine by Tim Davis.
The complexity of this algorithm is: SUM_{i=1,m} r(i)^2,
i.e., the sum of the square of the row counts.

Questions
=========
    o  Do I need to withhold the *dense* rows?
    o  How do I know the number of nonzeros in A'*A?

Here is the caller graph for this function:

◆ getBigUSize()

int_t getBigUSize	(	superlu_dist_options_t *	options,
		int_t	nsupers,
		gridinfo_t *	grid,
		int_t **	Lrowind_bc_ptr
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getBmod3d()

int * getBmod3d	(	int_t	treeId,
		int_t	nlb,
		sForest_t *	sforest,
		int_t *	xsup,
		int_t **	Ufstnz_br_ptr,
		int_t *	supernode2treeMap,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getBmod3d_newsolve()

int * getBmod3d_newsolve	(	int_t	nlb,
		int_t	nsupers,
		int *	supernodeMask,
		int_t *	xsup,
		int_t **	Ufstnz_br_ptr,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getBrecvTree()

int * getBrecvTree	(	int_t	nlb,
		sForest_t *	sforest,
		int *	bmod,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getBrecvTree_newsolve()

int * getBrecvTree_newsolve	(	int_t	nlb,
		int_t	nsupers,
		int *	supernodeMask,
		int *	bmod,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getCommonAncestorList()

int_t getCommonAncestorList	(	int_t	k,
		int_t *	alist,
		int_t *	seTree,
		treeList_t *	treeList
	)

Here is the caller graph for this function:

◆ getCommonAncsCount()

int_t getCommonAncsCount	(	int_t	k,
		treeList_t *	treeList
	)

Here is the caller graph for this function:

◆ getDescendList()

int_t getDescendList	(	int_t	k,
		int_t *	dlist,
		treeList_t *	treeList
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getEtreeLB()

int_t * getEtreeLB	(	int_t	nnodes,
		int_t *	perm_l,
		int_t *	gTopOrder
	)

Here is the call graph for this function:

◆ getFactIperm()

int_t * getFactIperm	(	int_t *	perm,
		int_t	nsupers
	)

Here is the caller graph for this function:

◆ getFactPerm()

int_t * getFactPerm ( int_t nsupers )

◆ getfmod_newsolve()

int * getfmod_newsolve	(	int_t	nlb,
		int_t	nsupers,
		int *	supernodeMask,
		int_t **	Lrowind_bc_ptr,
		int_t **	Lindval_loc_bc_ptr,
		gridinfo_t *	grid
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getfmodLeaf()

int * getfmodLeaf	(	int_t	nlb,
		int *	fmod_i
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getForests()

sForest_t ** getForests	(	int_t	maxLvl,
		int_t	nsupers,
		int_t *	setree,
		treeList_t *	treeList
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getfrecv_newsolve()

int * getfrecv_newsolve	(	int_t	nsupers,
		int *	supernodeMask,
		int_t	nlb,
		int *	fmod,
		int *	mod_bit,
		gridinfo_t *	grid
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getfrecvLeaf()

int * getfrecvLeaf	(	sForest_t *	sforest,
		int_t	nlb,
		int *	fmod,
		int *	mod_bit,
		gridinfo_t *	grid
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getGlobal_iperm()

int_t * getGlobal_iperm	(	int_t	nsupers,
		int_t	nperms,
		int_t **	perms,
		int_t *	nnodes
	)

◆ getGreedyLoadBalForests()

sForest_t ** getGreedyLoadBalForests	(	int_t	maxLvl,
		int_t	nsupers,
		int_t *	setree,
		treeList_t *	treeList
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getGridTrees()

int_t * getGridTrees ( gridinfo3d_t * grid3d )

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getIsNodeInMyGrid()

int * getIsNodeInMyGrid	(	int_t	nsupers,
		int_t	maxLvl,
		int_t *	myNodeCount,
		int_t **	treePerm
	)

Here is the caller graph for this function:

◆ getLastDepBtree()

int * getLastDepBtree	(	int_t	nsupers,
		treeList_t *	treeList
	)

◆ getldu()

int getldu	(	int_t	knsupc,
		int_t	iklrow,
		int_t *	usub
	)

Here is the caller graph for this function:

◆ getMyEtLims()

int_t * getMyEtLims	(	int_t	nnodes,
		int_t *	myTopOrder
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getMyIperm()

int_t * getMyIperm	(	int_t	nnodes,
		int_t	nsupers,
		int_t *	myPerm
	)

Here is the caller graph for this function:

◆ getMyNodeCounts()

int_t * getMyNodeCounts	(	int_t	maxLvl,
		int_t *	myTreeIdxs,
		int_t *	gNodeCount
	)

◆ getMyNodeCountsFr()

int_t * getMyNodeCountsFr	(	int_t	maxLvl,
		int_t *	myTreeIdxs,
		sForest_t **	sForests
	)

Here is the caller graph for this function:

◆ getMyTopOrder()

int_t * getMyTopOrder	(	int_t	nnodes,
		int_t *	myPerm,
		int_t *	myIperm,
		int_t *	setree
	)

Here is the caller graph for this function:

◆ getMyTreeTopoInfo()

treeTopoInfo_t getMyTreeTopoInfo	(	int_t	nnodes,
		int_t	nsupers,
		int_t *	myPerm,
		int_t *	setree
	)

Here is the call graph for this function:

◆ getNbrecvX()

int getNbrecvX	(	sForest_t *	sforest,
		int_t *	Urbs,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getNbrecvX_newsolve()

int getNbrecvX_newsolve	(	int_t	nsupers,
		int *	supernodeMask,
		int_t *	Urbs,
		Ucb_indptr_t **	Ucb_indptr,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getNestDissForests()

sForest_t ** getNestDissForests	(	int_t	maxLvl,
		int_t	nsupers,
		int_t *	setree,
		treeList_t *	treeList
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getNfrecvmod_newsolve()

int_t getNfrecvmod_newsolve	(	int *	nleaf,
		int_t	nsupers,
		int *	supernodeMask,
		int *	frecv,
		int *	fmod,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getNfrecvmodLeaf()

int_t getNfrecvmodLeaf	(	int *	nleaf,
		sForest_t *	sforest,
		int *	frecv,
		int *	fmod,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getNfrecvx_newsolve()

int getNfrecvx_newsolve	(	int_t	nsupers,
		int *	supernodeMask,
		int_t **	Lrowind_bc_ptr,
		int_t **	Lindval_loc_bc_ptr,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getNfrecvxLeaf()

int getNfrecvxLeaf	(	sForest_t *	sforest,
		int_t **	Lrowind_bc_ptr,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getnGPUStreams()

int getnGPUStreams ( void )

◆ getNodeCountsFr()

int_t * getNodeCountsFr	(	int_t	maxLvl,
		sForest_t **	sForests
	)

Here is the caller graph for this function:

◆ getNodeList()

int_t ** getNodeList	(	int_t	maxLvl,
		int_t *	setree,
		int_t *	nnodes,
		int_t *	treeHeads,
		treeList_t *	treeList
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getNodeListFr()

int_t ** getNodeListFr	(	int_t	maxLvl,
		sForest_t **	sForests
	)

Here is the caller graph for this function:

◆ getNrootUsolveTree()

int getNrootUsolveTree	(	int_t *	nbrecvmod,
		sForest_t *	sforest,
		int *	brecv,
		int *	bmod,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getNrootUsolveTree_newsolve()

int getNrootUsolveTree_newsolve	(	int_t *	nbrecvmod,
		int_t	nsupers,
		int *	supernodeMask,
		int *	brecv,
		int *	bmod,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ getNsupers()

int getNsupers	(	int	n,
		Glu_persist_t *	Glu_persist
	)

◆ getNumLookAhead()

int_t getNumLookAhead ( superlu_dist_options_t * options )

Here is the caller graph for this function:

◆ getNumThreads()

int getNumThreads ( int iam )

Here is the caller graph for this function:

◆ getPerm_c_supno()

int_t * getPerm_c_supno	(	int_t	nsupers,
		superlu_dist_options_t *	options,
		int_t *	etree,
		Glu_persist_t *	Glu_persist,
		int_t **	Lrowind_bc_ptr,
		int_t **	Ufstnz_br_ptr,
		gridinfo_t *	grid
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getPerm_c_supno_allgrid()

int_t * getPerm_c_supno_allgrid	(	int_t	nsupers,
		superlu_dist_options_t *	options,
		int_t *	etree,
		Glu_persist_t *	Glu_persist,
		int_t **	Lrowind_bc_ptr,
		int_t **	Ufstnz_br_ptr,
		gridinfo3d_t *	grid3d
	)

Here is the call graph for this function:

◆ getPermNodeList()

int_t * getPermNodeList	(	int_t	nnode,
		int_t *	nlist,
		int_t *	perm_c_sup,
		int_t *	iperm_c_sup
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getReplicatedTrees()

int_t * getReplicatedTrees ( gridinfo3d_t * grid3d )

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getSCUweight()

void getSCUweight	(	int_t	nsupers,
		treeList_t *	treeList,
		int_t *	xsup,
		int_t **	Lrowind_bc_ptr,
		int_t **	Ufstnz_br_ptr,
		gridinfo3d_t *	grid3d
	)

Here is the call graph for this function:

◆ getSCUweight_allgrid()

void getSCUweight_allgrid	(	int_t	nsupers,
		treeList_t *	treeList,
		int_t *	xsup,
		int_t **	Lrowind_bc_ptr,
		int_t **	Ufstnz_br_ptr,
		gridinfo3d_t *	grid3d
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getSubTreeRoots()

int_t * getSubTreeRoots	(	int_t	k,
		int_t *	numSubtrees,
		treeList_t *	treeList
	)

Here is the caller graph for this function:

◆ getTreeHeads()

int_t * getTreeHeads	(	int_t	maxLvl,
		int_t	nsupers,
		treeList_t *	treeList
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ getTreePerm()

int_t ** getTreePerm	(	int_t *	myTreeIdxs,
		int_t *	myZeroTrIdxs,
		int_t *	nodeCount,
		int_t **	nodeList,
		int_t *	perm_c_supno,
		int_t *	iperm_c_supno,
		gridinfo3d_t *	grid3d
	)

Here is the call graph for this function:

◆ getTreePermForest()

int_t ** getTreePermForest	(	int_t *	myTreeIdxs,
		int_t *	myZeroTrIdxs,
		sForest_t *	sForests,
		int_t *	perm_c_supno,
		int_t *	iperm_c_supno,
		gridinfo3d_t *	grid3d
	)

◆ getTreePermFr()

int_t ** getTreePermFr	(	int_t *	myTreeIdxs,
		sForest_t **	sForests,
		gridinfo3d_t *	grid3d
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ ifill_dist()

void ifill_dist	(	int_t *	a,
		int_t	alen,
		int_t	ival
	)

Fills an integer array with a given value.

Here is the caller graph for this function:

◆ initCommRequests()

int_t initCommRequests	(	commRequests_t *	comReqs,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ initCommRequestsArr()

commRequests_t ** initCommRequestsArr	(	int_t	mxLeafNode,
		int_t	ldt,
		gridinfo_t *	grid
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ initFactNodelists()

int_t initFactNodelists	(	int_t	ldt,
		int_t	num_threads,
		int_t	nsupers,
		factNodelists_t *	fNlists
	)

Here is the caller graph for this function:

◆ initFactStat()

int_t initFactStat	(	int	nsupers,
		factStat_t *	factStat
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ initMsgs()

int_t initMsgs ( msgs_t * msgs )

Here is the caller graph for this function:

◆ initMsgsArr()

msgs_t ** initMsgsArr ( int_t numLA )

Here is the call graph for this function:

Here is the caller graph for this function:

◆ initTRStimer()

void initTRStimer	(	xtrsTimer_t *	xtrsTimer,
		gridinfo_t *	grid
	)

Here is the caller graph for this function:

◆ int32Calloc_dist()

int * int32Calloc_dist ( int n )

Here is the caller graph for this function:

◆ int32Malloc_dist()

int * int32Malloc_dist ( int n )

Here is the caller graph for this function:

◆ intCalloc_dist()

int_t * intCalloc_dist ( int_t n )

◆ intMalloc_dist()

int_t * intMalloc_dist ( int_t n )

◆ isort()

void isort	(	int_t	N,
		int_t *	ARRAY1,
		int_t *	ARRAY2
	)

Here is the caller graph for this function:

◆ isort1()

void isort1	(	int_t	N,
		int_t *	ARRAY
	)

Here is the caller graph for this function:

◆ LDiagBlockRecvWait()

int_t LDiagBlockRecvWait	(	int_t	k,
		int *	factored_U,
		MPI_Request *	L_diag_blk_recv_req,
		gridinfo_t *	grid
	)

◆ log2i()

int_t log2i ( int_t index )

◆ log_memory()

void log_memory	(	int64_t	cur_bytes,
		SuperLUStat_t *	stat
	)

Here is the caller graph for this function:

◆ mc64id_dist()

int mc64id_dist ( int * icntl )

Here is the caller graph for this function:

◆ merg_perms()

int_t * merg_perms	(	int_t	nperms,
		int_t *	nnodes,
		int_t **	perms
	)

Here is the call graph for this function:

◆ num_full_cols_U()

int_t num_full_cols_U	(	int_t	kk,
		int_t **	Ufstnz_br_ptr,
		int_t *	xsup,
		gridinfo_t *	grid,
		int_t *	perm_u,
		int_t *	ldu
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ num_full_cols_U_mod()

int_t num_full_cols_U_mod	(	int_t	kk,
		int_t *	usub,
		int_t *	xsup,
		gridinfo_t *	grid,
		int_t *	perm_u,
		int_t *	ldu
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ partition()

int_t partition	(	int_t *	a,
		int_t	l,
		int_t	r,
		int_t	dir
	)

Here is the caller graph for this function:

◆ partitionM()

int_t partitionM	(	int_t *	a,
		int_t	l,
		int_t	r,
		int_t	lda,
		int_t	dir,
		int_t	dims
	)

Here is the caller graph for this function:

◆ permCol_SymbolicFact3d()

void permCol_SymbolicFact3d	(	superlu_dist_options_t *	options,
		int	n,
		SuperMatrix *	GA,
		int_t *	perm_c,
		int_t *	etree,
		Glu_persist_t *	Glu_persist,
		Glu_freeable_t *	Glu_freeable,
		SuperLUStat_t *	stat,
		superlu_dist_mem_usage_t *	symb_mem_usage,
		gridinfo3d_t *	grid3d
	)

This function performs the symbolic factorization on matrix Pc*Pr*A*Pc' and sets up the nonzero data structures for L & U matrices. In the process, the matrix is also ordered and its memory usage information is fetched.

Parameters

options	The options for the SuperLU distribution.
n	Dimension of the global matrix A.
GA	A pointer to the global matrix A.
perm_c	The column permutation vector.
etree	The elimination tree of PcPrA*Pc'.
Glu_persist	Pointer to the structure which tracks the symbolic factorization information.
Glu_freeable	Pointer to the structure which tracks the space used to store L/U data structures.
stat	Information on program execution.
grid3d	The 3D process grid.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ print_etree()

void print_etree	(	int_t *	setree,
		int_t *	iperm,
		int_t	nsuper
	)

◆ print_etree_leveled()

void print_etree_leveled	(	int_t *	setree,
		int_t *	tsort_etree,
		int_t	nsuper
	)

◆ Print_EtreeLevelBoundry()

void Print_EtreeLevelBoundry	(	int_t *	Etree_LvlBdry,
		int_t	max_level,
		int_t	nsuper
	)

◆ print_memorylog()

void print_memorylog	(	SuperLUStat_t *	stat,
		char *	msg
	)

Here is the caller graph for this function:

◆ print_options_dist()

void print_options_dist ( superlu_dist_options_t * options )

Print the options setting.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ print_panel_seg_dist()

void print_panel_seg_dist	(	int_t	n,
		int_t	w,
		int_t	jcol,
		int_t	nseg,
		int_t *	segrep,
		int_t *	repfnz
	)

Diagnostic print of segment info after panel_dfs().

◆ print_sp_ienv_dist()

void print_sp_ienv_dist ( superlu_dist_options_t * options )

Print the blocking parameters.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ PrintDouble5()

void PrintDouble5	(	char *	,
		int_t	,
		double *
	)

◆ printFileList()

int_t printFileList	(	char *	sname,
		int_t	nnodes,
		int_t *	dlist,
		int_t *	setree
	)

◆ printForestWeightCost()

void printForestWeightCost	(	sForest_t **	sForests,
		SCT_t *	SCT,
		gridinfo3d_t *	grid3d
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ printGPUStats()

void printGPUStats	(	int	nsupers,
		SuperLUStat_t *	stat,
		gridinfo3d_t *
	)

Here is the caller graph for this function:

◆ PrintInt10()

void PrintInt10	(	char *	name,
		int_t	len,
		int_t *	x
	)

◆ PrintInt32()

void PrintInt32	(	char *	name,
		int	len,
		int *	x
	)

Here is the caller graph for this function:

◆ printTRStimer()

void printTRStimer	(	xtrsTimer_t *	xtrsTimer,
		gridinfo3d_t *	grid3d
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ PStatClear()

void PStatClear ( SuperLUStat_t * stat )

Here is the caller graph for this function:

◆ PStatFree()

void PStatFree ( SuperLUStat_t * stat )

Here is the caller graph for this function:

◆ PStatInit()

void PStatInit ( SuperLUStat_t * stat )

Here is the caller graph for this function:

◆ PStatPrint()

void PStatPrint	(	superlu_dist_options_t *	options,
		SuperLUStat_t *	stat,
		gridinfo_t *	grid
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ psymbfact_LUXpand()

int_t psymbfact_LUXpand	(	int_t	iam,
		int_t	n,
		int_t	fstVtxLvl_loc,
		int_t	vtxXp,
		int_t *	p_next,
		int_t	min_new_len,
		int_t	mem_type,
		int_t	rout_type,
		int_t	free_prev_mem,
		Pslu_freeable_t *	Pslu_freeable,
		Llu_symbfact_t *	Llu_symbfact,
		vtcsInfo_symbfact_t *	VInfo,
		psymbfact_stat_t *	PS
	)

Expand the data structures for L and U during the factorization.
Return value: SUCCES_RET - successful return
              ERROR_RET - error due to a memory alocation failure

Sherry: this function is used in the upper separator tree above the domains. It does not call 'expand()'

Here is the call graph for this function:

Here is the caller graph for this function:

◆ psymbfact_LUXpand_RL()

int_t psymbfact_LUXpand_RL	(	int_t	iam,
		int_t	n,
		int_t	vtxXp,
		int_t	next,
		int_t	len_texp,
		int_t	mem_type,
		Pslu_freeable_t *	Pslu_freeable,
		Llu_symbfact_t *	Llu_symbfact,
		vtcsInfo_symbfact_t *	VInfo,
		psymbfact_stat_t *	PS
	)

Expand the data structures for L and U during the factorization.
Return value:   0 - successful return
              > 0 - number of bytes allocated when run out of space

Sherry: this function calls psymbfact_LUXpandMem().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ psymbfact_LUXpandMem()

int_t psymbfact_LUXpandMem	(	int	iam,
		int_t	n,
		int_t	vtxXp,
		int_t	next,
		int_t	min_new_len,
		int	mem_type,
		int	rout_type,
		int	free_prev_mem,
		Pslu_freeable_t *	Pslu_freeable,
		Llu_symbfact_t *	Llu_symbfact,
		vtcsInfo_symbfact_t *	VInfo,
		psymbfact_stat_t *	PS
	)

Expand the data structures for L and U during the factorization.
Return value:   0 - successful return
              > 0 - number of bytes allocated when run out of space

Sherry: this function is used inside the domains.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ psymbfact_prLUXpand()

int_t psymbfact_prLUXpand	(	int_t	iam,
		int_t	min_new_len,
		int	mem_type,
		Llu_symbfact_t *	Llu_symbfact,
		psymbfact_stat_t *	PS
	)

Expand the data structures for L and U pruned during the factorization.
Return value: SUCCES_RET - successful return
              ERROR_RET - error when run out of space

Sherry: this function calls 'expand()' directly.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ pxerr_dist()

void pxerr_dist	(	char *	srname,
		gridinfo_t *	grid,
		int_t	info
	)

◆ QuerySpace_dist()

int_t QuerySpace_dist	(	int_t	n,
		int_t	lsub_size,
		Glu_freeable_t *	Glu_freeable,
		superlu_dist_mem_usage_t *	mem_usage
	)

mem_usage consists of the following fields:
for_lu (float)
     The amount of space used in bytes for the L\U data structures.
total (float)
     The amount of space needed in bytes to perform factorization.
expansions (int)
     Number of memory expansions during the LU factorization.

Here is the caller graph for this function:

◆ quickSort()

void quickSort	(	int_t *	a,
		int_t	l,
		int_t	r,
		int_t	dir
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ quickSortM()

void quickSortM	(	int_t *	a,
		int_t	l,
		int_t	r,
		int_t	lda,
		int_t	dir,
		int_t	dims
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ reduceStat()

int_t reduceStat	(	PhaseType	PHASE,
		SuperLUStat_t *	stat,
		gridinfo3d_t *	grid3d
	)

reduce the states from all the two grids before prinitng it out See the defenition of enum PhaseType in superlu_enum_const.h

Here is the caller graph for this function:

◆ set_default_options_dist()

void set_default_options_dist ( superlu_dist_options_t * options )

Set the default values for the options argument.

Here is the caller graph for this function:

◆ set_tag_ub()

int set_tag_ub ( void )

Here is the caller graph for this function:

◆ setree2list()

treeList_t * setree2list	(	int_t	nsuper,
		int_t *	setree
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ slu_SCT_free()

void slu_SCT_free ( SCT_t * SCT )

Here is the caller graph for this function:

◆ slu_SCT_init()

void slu_SCT_init ( SCT_t * SCT )

Here is the call graph for this function:

Here is the caller graph for this function:

◆ slu_SCT_print()

void slu_SCT_print	(	gridinfo_t *	grid,
		SCT_t *	SCT
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ slu_SCT_print3D()

void slu_SCT_print3D	(	gridinfo3d_t *	grid3d,
		SCT_t *	SCT
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ slu_SCT_printComm3D()

void slu_SCT_printComm3D	(	gridinfo3d_t *	grid3d,
		SCT_t *	SCT
	)

Here is the caller graph for this function:

◆ slu_SCT_printSummary()

void slu_SCT_printSummary	(	gridinfo_t *	grid,
		SCT_t *	SCT
	)

◆ smach_dist()

float smach_dist ( const char * cmach )

Here is the caller graph for this function:

◆ sort_R_info()

int sort_R_info	(	Remain_info_t *	Remain_info,
		int	n
	)

Here is the call graph for this function:

◆ sort_R_info_elm()

int sort_R_info_elm	(	Remain_info_t *	Remain_info,
		int	n
	)

Here is the call graph for this function:

◆ sort_U_info()

int sort_U_info	(	Ublock_info_t *	Ublock_info,
		int	n
	)

Here is the call graph for this function:

◆ sort_U_info_elm()

int sort_U_info_elm	(	Ublock_info_t *	Ublock_info,
		int	n
	)

Here is the call graph for this function:

◆ sp_coletree_dist()

int sp_coletree_dist	(	int_t *	acolst,
		int_t *	acolend,
		int_t *	arow,
		int_t	nr,
		int_t	nc,
		int_t *	parent
	)

Nonsymmetric elimination tree.

     Find the elimination tree for A'*A.
     This uses something similar to Liu's algorithm. 
     It runs in time O(nz(A)*log n) and does not form A'*A.

     Input:
       Sparse matrix A.  Numeric values are ignored, so any
       explicit zeros are treated as nonzero.
     Output:
       Integer array of parents representing the elimination
       tree of the symbolic product A'*A.  Each vertex is a
       column of A, and nc means a root of the elimination forest.

     John R. Gilbert, Xerox, 10 Dec 1990
     Based on code by JRG dated 1987, 1988, and 1990.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ sp_colorder()

void sp_colorder	(	superlu_dist_options_t *	options,
		SuperMatrix *	A,
		int_t *	perm_c,
		int_t *	etree,
		SuperMatrix *	AC
	)

Purpose
=======

sp_colorder() permutes the columns of the original matrix. It performs
the following steps:

   1. Apply column permutation perm_c[] to A's column pointers to form AC;

   2. If options->Fact = DOFACT, then
      (1) Compute column elimination tree etree[] of AC'AC;
      (2) Post order etree[] to get a postordered elimination tree etree[],
          and a postorder permutation post[];
      (3) Apply post[] permutation to columns of AC;
      (4) Overwrite perm_c[] with the product perm_c * post.

Arguments
=========

options (input) superlu_dist_options_t*
        Specifies whether or not the elimination tree will be re-used.
        If options->Fact == DOFACT, this means first time factor A, 
        etree is computed and output.
        Otherwise, re-factor A, etree is input, unchanged on exit.

A       (input) SuperMatrix*
        Matrix A in A*X=B, of dimension (A->nrow, A->ncol). The number
        of the linear equations is A->nrow. Currently, the type of A can be:
        Stype = SLU_NC or SLU_NCP; Dtype = SLU__D; Mtype = SLU_GE.
        In the future, more general A can be handled.

perm_c  (input/output) int*
    Column permutation vector of size A->ncol, which defines the 
        permutation matrix Pc; perm_c[i] = j means column i of A is 
        in position j in A*Pc.
        If options->Fact == DOFACT, perm_c is both input and output.
        On output, it is changed according to a postorder of etree.
        Otherwise, perm_c is input.

etree   (input/output) int*
        Elimination tree of Pc*(A'+A)*Pc', dimension A->ncol.
        If options->Fact == DOFACT, etree is an output argument,
        otherwise it is an input argument.
        Note: etree is a vector of parent pointers for a forest whose
        vertices are the integers 0 to A->ncol-1; etree[root]==A->ncol.

AC      (output) SuperMatrix*
        The resulting matrix after applied the column permutation
        perm_c[] to matrix A. The type of AC can be:
        Stype = SLU_NCP; Dtype = A->Dtype; Mtype = SLU_GE.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ sp_ienv_dist()

int sp_ienv_dist	(	int	ispec,
		superlu_dist_options_t *	options
	)

Purpose

sp_ienv_dist() is inquired to choose machine-dependent integer parameters for the local environment. See ISPEC for a description of the parameters.

This version provides a set of parameters which should give good,
but not optimal, performance on many of the currently available
computers. Users are encouraged to set the environment variable to change the tuning parameters for their particular machines.

Arguments

ISPEC (input) int Specifies the parameter to be returned as the value of SP_IENV_DIST.
= 1: the panel size w; a panel consists of w consecutive columns of matrix A in the process of Gaussian elimination. The best value depends on machine's cache characters. = 2: the relaxation parameter relax; if the number of nodes (columns) in a subtree of the elimination tree is less than relax, this subtree is considered as one supernode, regardless of the their row structures. = 3: the maximum size for a supernode, which must be greater than or equal to relaxation parameter (see case 2); = 4: the minimum row dimension for 2-D blocking to be used; = 5: the minimum column dimension for 2-D blocking to be used; = 6: the estimated fills factor for the adjacency structures of L and U, compared with A; = 7: the minimum value of the product M*N*K for a GEMM call worth being offloaded to accelerator (e.g., GPU, Xeon Phi). = 8: the maximum buffer size on GPU that can hold the "dC" matrix in the GEMM call for the Schur complement update. If this is too small, the Schur complement update will be done in multiple partitions, may be slower. = 9: number of GPU streams = 10: whether to offload computations to GPU or not = 11: whether to offload triangular solve to GPU or not

options (input) superlu_dist_options_t* The structure defines the input parameters to control how the LU decomposition the solves are performed.

(SP_IENV_DIST) (output) int >= 0: the value of the parameter specified by ISPEC
< 0: if SP_IENV_DIST = -k, the k-th argument had an illegal value.

Purpose

sp_ienv_dist() is inquired to choose machine-dependent integer parameters for the local environment. See ISPEC for a description of the parameters.

This version provides a set of parameters which should give good,
but not optimal, performance on many of the currently available
computers. Users are encouraged to set the environment variable to change the tuning parameters for their particular machines.

Arguments

ISPEC (input) int Specifies the parameter to be returned as the value of SP_IENV_DIST.
= 1: the panel size w; a panel consists of w consecutive columns of matrix A in the process of Gaussian elimination. The best value depends on machine's cache characters. = 2: the relaxation parameter relax; if the number of nodes (columns) in a subtree of the elimination tree is less than relax, this subtree is considered as one supernode, regardless of the their row structures. = 3: the maximum size for a supernode, which must be greater than or equal to relaxation parameter (see case 2); = 4: the minimum row dimension for 2-D blocking to be used; = 5: the minimum column dimension for 2-D blocking to be used; = 6: the estimated fills factor for the adjacency structures of L and U, compared with A; = 7: the minimum value of the product M*N*K for a GEMM call worth being offloaded to accelerator (e.g., GPU, Xeon Phi). = 8: the maximum buffer size on GPU that can hold the "dC" matrix in the GEMM call for the Schur complement update. If this is too small, the Schur complement update will be done in multiple partitions, may be slower. = 9: number of GPU streams = 10: whether to offload work to GPU or not

options (input) superlu_dist_options_t* The structure defines the input parameters to control how the LU decomposition the solves are performed.

(SP_IENV_DIST) (output) int >= 0: the value of the parameter specified by ISPEC
< 0: if SP_IENV_DIST = -k, the k-th argument had an illegal value.

Here is the call graph for this function:

◆ sp_symetree_dist()

int sp_symetree_dist	(	int_t *	acolst,
		int_t *	acolend,
		int_t *	arow,
		int_t	n,
		int_t *	parent
	)

Symmetric elimination tree.

     p = spsymetree (A);

     Find the elimination tree for symmetric matrix A.
     This uses Liu's algorithm, and runs in time O(nz*log n).

     Input:
       Square sparse matrix A.  No check is made for symmetry;
       elements below and on the diagonal are ignored.
       Numeric values are ignored, so any explicit zeros are 
       treated as nonzero.
     Output:
       Integer array of parents representing the etree, with n
       meaning a root of the elimination forest.
     Note:  
       This routine uses only the upper triangle, while sparse
       Cholesky (as in spchol.c) uses only the lower.  Matlab's
       dense Cholesky uses only the upper.  This routine could
       be modified to use the lower triangle either by transposing
       the matrix or by traversing it by rows with auxiliary
       pointer and link arrays.

     John R. Gilbert, Xerox, 10 Dec 1990
     Based on code by JRG dated 1987, 1988, and 1990.
     Modified by X.S. Li, November 1999.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ static_partition()

int_t static_partition	(	struct superlu_pair *	work_load,
		int_t	nwl,
		int_t *	partition,
		int_t	ldp,
		int_t *	sums,
		int_t *	counts,
		int	nprocs
	)

Here is the call graph for this function:

◆ super_stats_dist()

void super_stats_dist	(	int_t	nsuper,
		int_t *	xsup
	)

Here is the call graph for this function:

◆ superlu_abort_and_exit_dist()

void superlu_abort_and_exit_dist ( char * msg )

◆ superlu_dist_GetVersionNumber()

int superlu_dist_GetVersionNumber	(	int *	major,
		int *	minor,
		int *	bugfix
	)

Here is the caller graph for this function:

◆ superlu_free_dist()

void superlu_free_dist ( void * addr )

◆ superlu_gridexit()

void superlu_gridexit ( gridinfo_t * grid )

Here is the caller graph for this function:

◆ superlu_gridexit3d()

void superlu_gridexit3d ( gridinfo3d_t * grid )

Here is the caller graph for this function:

◆ superlu_gridinit()

void superlu_gridinit	(	MPI_Comm	Bcomm,
		int	nprow,
		int	npcol,
		gridinfo_t *	grid
	)

All processes in the MPI communicator must call this routine.

On output, if a process is not in the SuperLU group, the following values are assigned to it: grid->comm = MPI_COMM_NULL grid->iam = -1

Here is the call graph for this function:

Here is the caller graph for this function:

◆ superlu_gridinit3d()

void superlu_gridinit3d	(	MPI_Comm	Bcomm,
		int	nprow,
		int	npcol,
		int	npdep,
		gridinfo3d_t *	grid
	)

All processes in the MPI communicator must call this routine.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ superlu_gridmap()

void superlu_gridmap	(	MPI_Comm	Bcomm,
		int	nprow,
		int	npcol,
		int	usermap[],
		int	ldumap,
		gridinfo_t *	grid
	)

All processes in the MPI communicator must call this routine.

On output, if a process is not in the SuperLU group, the following values are assigned to it: grid->comm = MPI_COMM_NULL grid->iam = -1

Here is the caller graph for this function:

◆ superlu_gridmap3d()

void superlu_gridmap3d	(	MPI_Comm	Bcomm,
		int	nprow,
		int	npcol,
		int	npdep,
		int	usermap[],
		gridinfo3d_t *	grid
	)

All processes in the MPI communicator must call this routine. On output, if a process is not in the SuperLU group, the following values are assigned to it: grid->comm = MPI_COMM_NULL grid->iam = -1.

Here is the caller graph for this function:

◆ superlu_malloc_dist()

void * superlu_malloc_dist ( size_t size )

◆ supernodal_etree()

int_t * supernodal_etree	(	int_t	nsuper,
		int_t *	etree,
		int_t *	supno,
		int_t *	xsup
	)

Returns Supernodal Elimination Tree

Parameters

nsuper	Number of Supernodes
etree	Scalar elimination tree
supno	Vertex to supernode mapping
xsup	Supernodal boundaries

Returns: Supernodal elimination tree

Here is the call graph for this function:

Here is the caller graph for this function:

◆ symbfact()

int_t symbfact	(	superlu_dist_options_t *	options,
		int	pnum,
		SuperMatrix *	A,
		int_t *	perm_c,
		int_t *	etree,
		Glu_persist_t *	Glu_persist,
		Glu_freeable_t *	Glu_freeable
	)

Purpose
=======
  symbfact() performs a symbolic factorization on matrix A and sets up 
  the nonzero data structures which are suitable for supernodal Gaussian
  elimination with no pivoting (GENP). This routine features:
       o depth-first search (DFS)
       o supernodes
       o symmetric structure pruning

Return value
============
  < 0, number of bytes needed for LSUB.
  = 0, matrix dimension is 1.
  > 0, number of bytes allocated when out of memory.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ symbfact_dist()

float symbfact_dist	(	superlu_dist_options_t *	options,
		int	nprocs_num,
		int	nprocs_symb,
		SuperMatrix *	A,
		int_t *	perm_c,
		int_t *	perm_r,
		int_t *	sizes,
		int_t *	fstVtxSep,
		Pslu_freeable_t *	Pslu_freeable,
		MPI_Comm *	num_comm,
		MPI_Comm *	symb_comm,
		superlu_dist_mem_usage_t *	symb_mem_usage
	)

 
Purpose
=======
  symbfact_dist() performs symbolic factorization of matrix A suitable
  for performing the supernodal Gaussian elimination with no pivoting (GEPP). 
  This routine computes the structure of one column of L and one row of U 
  at a time.  It uses:
       o distributed input matrix
       o supernodes
       o symmetric structure pruning


Arguments
=========

nprocs_num (input) int
        Number of processors SuperLU_DIST is executed on, and the input 
        matrix is distributed on.

nprocs_symb (input) int
        Number of processors on which the symbolic factorization is
        performed.  It is equal to the number of independent domains
        idenfied in the graph partitioning algorithm executed
        previously and has to be a power of 2.  It corresponds to
        number of leaves in the separator tree.

A       (input) SuperMatrix*
        Matrix A in A*X=B, of dimension (A->nrow, A->ncol). The
        number of the linear equations is A->nrow.  Matrix A is
        distributed in NRformat_loc format.
        Matrix A is not yet permuted by perm_c.

perm_c  (input) int_t*
    Column permutation vector of size A->ncol, which defines the 
        permutation matrix Pc; perm_c[i] = j means column i of A is 
        in position j in A*Pc.

perm_r  (input) int_t*
    Row permutation vector of size A->nrow, which defines the 
        permutation matrix Pr; perm_r[i] = j means column i of A is 
        in position j in Pr*A.

sizes   (input) int_t*
        Contains the number of vertices in each separator.

fstVtxSep (input) int_t*
        Contains first vertex for each separator.

Pslu_freeable (output) Pslu_freeable_t*
        Returns the local L and U structure, and global to local
        information on the indexing of the vertices.  Contains all
        the information necessary for performing the data
        distribution towards the numeric factorization.

num_comm (input) MPI_Comm*
        Communicator for numerical factorization 

symb_comm (input) MPI_Comm*
        Communicator for symbolic factorization 

symb_mem_usage (input) superlu_dist_mem_usage_t *
        Statistics on memory usage.

Return value
============
  < 0, number of bytes allocated on return from the symbolic factorization.
  > 0, number of bytes allocated when out of memory.

Sketch of the algorithm
=======================

 Distrbute the vertices on the processors using a subtree to
 subcube algorithm.

 Redistribute the structure of the input matrix A according to the
 subtree to subcube computed previously for the symbolic
 factorization routine.  This implies in particular a distribution
 from nprocs_num processors to nprocs_symb processors.

 Perform symbolic factorization guided by the separator tree provided by
 a graph partitioning algorithm.  The symbolic factorization uses a 
 combined left-looking, right-looking approach.

 
Purpose
=======
  symbfact_dist() performs symbolic factorization of matrix A suitable
  for performing the supernodal Gaussian elimination with no pivoting (GEPP). 
  This routine computes the structure of one column of L and one row of U 
  at a time.  It uses:
       o distributed input matrix
       o supernodes
       o symmetric structure pruning


Arguments
=========

nprocs_num (input) int
        Number of processors SuperLU_DIST is executed on, and the input 
        matrix is distributed on.

nprocs_symb (input) int
        Number of processors on which the symbolic factorization is
        performed.  It is equal to the number of independent domains
        idenfied in the graph partitioning algorithm executed
        previously and has to be a power of 2.  It corresponds to
        number of leaves in the separator tree.

A       (input) SuperMatrix*
        Matrix A in A*X=B, of dimension (A->nrow, A->ncol). The
        number of the linear equations is A->nrow.  Matrix A is
        distributed in NRformat_loc format.
        Matrix A is not yet permuted by perm_c.

perm_c  (input) int_t*
    Column permutation vector of size A->ncol, which defines the 
        permutation matrix Pc; perm_c[i] = j means column i of A is 
        in position j in A*Pc.

perm_r  (input) int_t*
    Row permutation vector of size A->nrow, which defines the 
        permutation matrix Pr; perm_r[i] = j means column i of A is 
        in position j in Pr*A.

sizes   (input) int_t*
        Contains the number of vertices in each separator.

fstVtxSep (input) int_t*
        Contains first vertex for each separator.

Pslu_freeable (output) Pslu_freeable_t*
        Returns the local L and U structure, and global to local
        information on the indexing of the vertices.  Contains all
        the information necessary for performing the data
        distribution towards the numeric factorization.

num_comm (input) MPI_Comm*
        Communicator for numerical factorization 

symb_comm (input) MPI_Comm*
        Communicator for symbolic factorization 

symb_mem_usage (input) superlu_dist_mem_usage_t *
        Statistics on memory usage.

Return value
============
  < 0, number of bytes allocated on return from the symbolic factorization.
  > 0, number of bytes allocated when out of memory.

Sketch of the algorithm
=======================

 Distrbute the vertices on the processors using a subtree to
 subcube algorithm.

 Redistribute the structure of the input matrix A according to the
 subtree to subcube computed previously for the symbolic
 factorization routine.  This implies in particular a distribution
 from nprocs_num processors to nprocs_symb processors.

 Perform symbolic factorization guided by the separator tree provided by
 a graph partitioning algorithm.  The symbolic factorization uses a 
 combined left-looking, right-looking approach.

 
Purpose
=======
  symbfact_dist() performs symbolic factorization of matrix A suitable
  for performing the supernodal Gaussian elimination with no pivoting (GEPP). 
  This routine computes the structure of one column of L and one row of U 
  at a time.  It uses:
       o distributed input matrix
       o supernodes
       o symmetric structure pruning


Arguments
=========

nprocs_num (input) int
        Number of processors SuperLU_DIST is executed on, and the input 
        matrix is distributed on.

nprocs_symb (input) int
        Number of processors on which the symbolic factorization is
        performed.  It is equal to the number of independent domains
        idenfied in the graph partitioning algorithm executed
        previously and has to be a power of 2.  It corresponds to
        number of leaves in the separator tree.

A       (input) SuperMatrix*
        Matrix A in A*X=B, of dimension (A->nrow, A->ncol). The
        number of the linear equations is A->nrow.  Matrix A is
        distributed in NRformat_loc format.
        Matrix A is not yet permuted by perm_c.

perm_c  (input) int_t*
    Column permutation vector of size A->ncol, which defines the 
        permutation matrix Pc; perm_c[i] = j means column i of A is 
        in position j in A*Pc.

perm_r  (input) int_t*
    Row permutation vector of size A->nrow, which defines the 
        permutation matrix Pr; perm_r[i] = j means column i of A is 
        in position j in Pr*A.

sizes   (input) int_t*
        Contains the number of vertices in each separator.

fstVtxSep (input) int_t*
        Contains first vertex for each separator.

Pslu_freeable (output) Pslu_freeable_t*
        Returns the local L and U structure, and global to local
        information on the indexing of the vertices.  Contains all
        the information necessary for performing the data
        distribution towards the numeric factorization.

num_comm (input) MPI_Comm*
        Communicator for numerical factorization 

symb_comm (input) MPI_Comm*
        Communicator for symbolic factorization 

symb_mem_usage (input) superlu_dist_mem_usage_t *
        Statistics on memory usage.

Return value
============
  < 0, number of bytes allocated on return from the symbolic factorization.
  > 0, number of bytes allocated when out of memory.

Sketch of the algorithm
=======================

 Distrbute the vertices on the processors using a subtree to
 subcube algorithm.

 Redistribute the structure of the input matrix A according to the
 subtree to subcube computed previously for the symbolic
 factorization routine.  This implies in particular a distribution
 from nprocs_num processors to nprocs_symb processors.

 Perform symbolic factorization guided by the separator tree provided by
 a graph partitioning algorithm.  The symbolic factorization uses a 
 combined left-looking, right-looking approach.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ symbfact_SubFree()

int symbfact_SubFree ( Glu_freeable_t * Glu_freeable )

Deallocate storage of the data structures common to symbolic
factorization routines.

Here is the caller graph for this function:

◆ symbfact_SubInit()

int_t symbfact_SubInit	(	superlu_dist_options_t *	options,
		fact_t	fact,
		void *	work,
		int_t	lwork,
		int_t	m,
		int_t	n,
		int_t	annz,
		Glu_persist_t *	Glu_persist,
		Glu_freeable_t *	Glu_freeable
	)

Allocate storage for the data structures common to symbolic factorization
routines. For those unpredictable size, make a guess as FILL * nnz(A).
Return value:
    If lwork = -1, return the estimated amount of space required, plus n;
    otherwise, return the amount of space actually allocated when
    memory allocation failure occurred.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ symbfact_SubXpand()

int_t symbfact_SubXpand	(	int_t	n,
		int_t	jcol,
		int_t	next,
		MemType	mem_type,
		int_t *	maxlen,
		Glu_freeable_t *	Glu_freeable
	)

Expand the data structures for L and U during the factorization.
Return value:   0 - successful return
              > 0 - number of bytes allocated when run out of space

Here is the call graph for this function:

Here is the caller graph for this function:

◆ Test_LDiagBlock_Recv()

int_t Test_LDiagBlock_Recv	(	MPI_Request *	request,
		SCT_t *	SCT
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ Test_UDiagBlock_Recv()

int_t Test_UDiagBlock_Recv	(	MPI_Request *	request,
		SCT_t *	SCT
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ testListPerm()

int_t testListPerm	(	int_t	nodeCount,
		int_t *	nodeList,
		int_t *	permList,
		int_t *	gTopLevel
	)

Here is the call graph for this function:

◆ testSubtreeNodelist()

int_t testSubtreeNodelist	(	int_t	nsupers,
		int_t	numList,
		int_t **	nodeList,
		int_t *	nodeCount
	)

Here is the call graph for this function:

◆ topological_ordering()

int_t * topological_ordering	(	int_t	nsuper,
		int_t *	setree
	)

Here is the call graph for this function:

◆ treeImbalance3D()

void treeImbalance3D	(	gridinfo3d_t *	grid3d,
		SCT_t *	SCT
	)

Here is the call graph for this function:

◆ TreePostorder_dist()

int_t * TreePostorder_dist	(	int_t	n,
		int_t *	parent
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ Trs2_InitUblock_info()

int_t Trs2_InitUblock_info	(	int_t	klst,
		int_t	nb,
		Ublock_info_t *	Ublock_info,
		int_t *	usub,
		Glu_persist_t *	Glu_persist,
		SuperLUStat_t *	stat
	)

Here is the caller graph for this function:

◆ Wait_LDiagBlock_Recv()

int_t Wait_LDiagBlock_Recv	(	MPI_Request *	request,
		SCT_t *	SCT
	)

Here is the call graph for this function:

◆ Wait_LDiagBlockSend()

int_t Wait_LDiagBlockSend	(	MPI_Request *	L_diag_blk_send_req,
		gridinfo_t *	grid,
		SCT_t *	SCT
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ Wait_LSend()

int_t Wait_LSend	(	int_t	k,
		gridinfo_t *	grid,
		int **	ToSendR,
		MPI_Request *	s,
		SCT_t *	SCT
	)

Here is the call graph for this function:

◆ Wait_LUDiagSend()

int Wait_LUDiagSend	(	int_t	k,
		MPI_Request *	U_diag_blk_send_req,
		MPI_Request *	L_diag_blk_send_req,
		gridinfo_t *	grid,
		SCT_t *	SCT
	)

Here is the call graph for this function:

◆ Wait_UDiagBlock_Recv()

int_t Wait_UDiagBlock_Recv	(	MPI_Request *	request,
		SCT_t *	SCT
	)

Here is the call graph for this function:

◆ Wait_UDiagBlockSend()

int_t Wait_UDiagBlockSend	(	MPI_Request *	U_diag_blk_send_req,
		gridinfo_t *	grid,
		SCT_t *	SCT
	)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ Wait_USend()

int_t Wait_USend	(	MPI_Request *	send_req,
		gridinfo_t *	grid,
		SCT_t *	SCT
	)

Here is the call graph for this function:

◆ xerr_dist()

int xerr_dist	(	char *	srname,
		int *	info
	)

Here is the caller graph for this function:

◆ zAllocBcast()

int_t zAllocBcast	(	int_t	size,
		void **	ptr,
		gridinfo3d_t *	grid3d
	)

◆ zAllocBcast_gridID()

int_t zAllocBcast_gridID	(	int_t	size,
		void **	ptr,
		int_t	gridID,
		gridinfo3d_t *	grid3d
	)

Here is the caller graph for this function:

Variable Documentation

◆ BC_L

const int BC_L =1

static

◆ BC_U

const int BC_U =3

static

◆ RD_L

const int RD_L =2

static

◆ RD_U

const int RD_U =4

static

[in]	options	solver options
[in]	batchCount	number of matrices in the batch
[in]	SparseMatrix_handles	pointers to the matrices in the batch, each pointing to the actual stoage in CSC format On entry, the original matrices, may be overwritten by A1 <- Prdiag(R)A*diag(C) from dequil_batch() and dpivot_batch()
[out]	CpivPtr	pointers to column permutation vectors for each matrix, each of size n

Classes

Macros

Typedefs

Enumerations

Functions

Variables

Detailed Description

Macro Definition Documentation

◆ __SUPERLU_ASYNC_TREE

◆ BC_HEADER

◆ BC_HEADER_NEWU

◆ BlockNum

◆ BR_HEADER

◆ CEILING

◆ cmax

◆ COMM_ALL

◆ COMM_COLUMN

◆ COMM_ROW

◆ DEG_TREE

◆ DIM_X

◆ DIM_XA

◆ DIM_XB

◆ DIM_Y

◆ DIM_YA

◆ DIM_YB

◆ fetch

◆ fma

◆ FstBlockC

◆ GPU_ACC

◆ GSUM

◆ HANDLE_SIZE

◆ IAM

◆ IFMT

◆ ISORT

◆ LB_DESCRIPTOR

◆ LBi

◆ LBj

◆ LkkDIAG

◆ LkSUB

◆ LkVAL

◆ LOG2

◆ LSUM

◆ LSUM_BLK

◆ LSUM_H

◆ MAGMA_CONST

◆ MAX_SUPER_SIZE

◆ mpi_int_t

◆ MYCOL

◆ MYROW

◆ NBUFFERS

◆ NO_MARKER

◆ NTAGS

◆ NWARP

◆ PCOL

◆ PNUM

◆ PROW

◆ RHS_ITERATE

◆ SLU_MPI_TAG

◆ SUPER_BLOCK

◆ SUPER_LINEAR

◆ SUPERLU_DIST_EXPORT

◆ SUPERLU_DIST_MAJOR_VERSION

◆ SUPERLU_DIST_MINOR_VERSION

◆ SUPERLU_DIST_PATCH_VERSION

◆ SUPERLU_DIST_RELEASE_DATE

◆ SuperLU_MPI_COMPLEX

◆ SuperLU_MPI_DOUBLE_COMPLEX

◆ SuperLU_timer_

◆ SuperSize

◆ THR_M

◆ THR_N

◆ UB_DESCRIPTOR

◆ UB_DESCRIPTOR_NEWU

◆ UB_DESCRIPTOR_NEWUCPP

◆ UjROW

◆ UkSUB

◆ UkVAL

◆ VT_TRACEOFF

◆ VT_TRACEON

◆ WARP_SIZE