sparsemat.h File Reference

Types and functions related to sparse matrices. More...

#include "config.h"
#include "types.h"
#include "spmv.h"
#include "context.h"
#include "densemat.h"
#include "sparsemat_src.h"
#include <stdarg.h>

Include dependency graph for sparsemat.h:

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

Go to the source code of this file.

Classes
struct	ghost_sorting_helper
	Helper for sparse matrix row sorting. More...
struct	ghost_sparsemat_rowfunc_crs_arg
struct	ghost_spmv_opts
struct	ghost_kacz_opts
struct	ghost_carp_opts
struct	ghost_sellspmv_parameters
	The parameters to identify a SELL SpMV kernel. More...
struct	ghost_cusellspmv_parameters
	The parameters to identify a CUDA SELL SpMV kernel. More...
struct	ghost_kacz_parameters
	The parameters to identify a SELL Kaczmarz kernel. More...
struct	ghost_sparsemat_traits
	Sparse matrix traits. More...
struct	ghost_sparsemat
	A sparse matrix. More...
struct	ghost_sparsemat_rowfunc_file_initargs

Macros
#define	GHOST_SPARSEMAT_SORT_GLOBAL   -1
#define	GHOST_SPARSEMAT_SORT_LOCAL   -2
#define	GHOST_SELL_CHUNKHEIGHT_ELLPACK   0
	Create only a single chunk, i.e., use the ELLPACK storage format. More...
#define	GHOST_SELL_CHUNKHEIGHT_AUTO   -1
	A chunkheight should automatically be determined. More...
#define	GHOST_SPARSEMAT_PERM_ANY   (GHOST_SPARSEMAT_PERM_ANY_LOCAL|GHOST_SPARSEMAT_PERM_ANY_GLOBAL)
	Combination of flags which apply any permutation to a ghost_sparsemat. More...
#define	GHOST_SPARSEMAT_PERM_ANY_LOCAL   (GHOST_SPARSEMAT_COLOR|GHOST_SPARSEMAT_RCM|GHOST_SPARSEMAT_BLOCKCOLOR|GHOST_SPARSEMAT_SORT_ROWS)
#define	GHOST_SPARSEMAT_PERM_ANY_GLOBAL   (GHOST_SPARSEMAT_SCOTCHIFY|GHOST_SPARSEMAT_ZOLTAN)
#define	GHOST_SCOTCH_STRAT_DEFAULT   "g"
#define	SPM_NROWS(mat)   mat->context->row_map->dim
#define	SPM_NNZ(mat)   mat->context->nnz
#define	SPM_NCOLS(mat)   mat->context->col_map->dim
#define	SPM_GNCOLS(mat)   mat->context->col_map->gdim
#define	SPM_NROWSPAD(mat)   mat->context->row_map->dimpad
#define	SPM_NCHUNKS(mat)   (mat->nchunks)

Typedefs
typedef struct ghost_sparsemat_traits	ghost_sparsemat_traits
typedef struct ghost_sparsemat	ghost_sparsemat
typedef ghost_error(*	ghost_spmv_kernel )(ghost_densemat *, ghost_sparsemat *, ghost_densemat *, ghost_spmv_opts)
typedef ghost_error(*	ghost_kacz_kernel )(ghost_densemat *, ghost_sparsemat *, ghost_densemat *, ghost_kacz_opts)
typedef ghost_error(*	ghost_kacz_shift_kernel )(ghost_densemat *, ghost_densemat *, ghost_sparsemat *, ghost_densemat *, double, double, ghost_kacz_opts)

Enumerations
enum	ghost_sparsemat_symmetry { GHOST_SPARSEMAT_SYMM_GENERAL = 1, GHOST_SPARSEMAT_SYMM_SYMMETRIC = 2, GHOST_SPARSEMAT_SYMM_SKEW_SYMMETRIC = 4, GHOST_SPARSEMAT_SYMM_HERMITIAN = 8 }
	Possible sparse matrix symmetries. More...
enum	ghost_kacz_direction { GHOST_KACZ_DIRECTION_UNDEFINED, GHOST_KACZ_DIRECTION_FORWARD, GHOST_KACZ_DIRECTION_BACKWARD }
enum	ghost_kacz_normalize { GHOST_KACZ_NORMALIZE_YES = 0, GHOST_KACZ_NORMALIZE_NO = 1<<0, GHOST_KACZ_NORMALIZE_SAVE = 1<<1 }
enum	ghost_kacz_mode { GHOST_KACZ_MODE_NORMAL, GHOST_KACZ_MODE_ECO, GHOST_KACZ_MODE_PERFORMANCE }
enum	ghost_sparsemat_flags {   GHOST_SPARSEMAT_DEFAULT = 0, GHOST_SPARSEMAT_HOST = 1<<0, GHOST_SPARSEMAT_DEVICE = 1<<1, GHOST_SPARSEMAT_NOT_PERMUTE_COLS = 1<<2,   GHOST_SPARSEMAT_PERMUTE = 1<<3, GHOST_SPARSEMAT_NOT_SORT_COLS = 1<<4, GHOST_SPARSEMAT_NOT_STORE_SPLIT = 1<<5, GHOST_SPARSEMAT_NOT_STORE_FULL = 1<<6,   GHOST_SPARSEMAT_SCOTCHIFY = 1<<7, GHOST_SPARSEMAT_SAVE_ORIG_COLS = 1<<8, GHOST_SPARSEMAT_COLOR = 1<<9, GHOST_SPARSEMAT_TRANSPOSE_MM = 1<<10,   GHOST_SPARSEMAT_ZOLTAN = 1<<11, GHOST_SPARSEMAT_RCM = 1<<12, GHOST_SPARSEMAT_BLOCKCOLOR = 1<<13, GHOST_SOLVER_KACZ = 1<<14,   GHOST_SPARSEMAT_SORT_ROWS = 1<<15, GHOST_SPARSEMAT_PERM_NO_DISTINCTION =1<<16, GHOST_SPARSEMAT_DIAG_FIRST =1<<17 }
	Flags to a sparse matrix. More...

Functions
ghost_error	ghost_sparsemat_create (ghost_sparsemat **mat, ghost_context *ctx, ghost_sparsemat_traits *traits, int nTraits)
	Create a sparse matrix. More...
ghost_error	ghost_sparsemat_info_string (char **str, ghost_sparsemat *matrix)
	Create a string holding information about the sparsemat. More...
ghost_error	ghost_sparsemat_nnz (ghost_gidx *nnz, ghost_sparsemat *mat)
	Obtain the global number of nonzero elements of a sparse matrix. More...
ghost_error	ghost_sparsemat_nrows (ghost_gidx *nrows, ghost_sparsemat *mat)
	Obtain the global number of rows of a sparse matrix. More...
const char *	ghost_sparsemat_symmetry_string (ghost_sparsemat_symmetry symmetry)
	Convert the matrix' symmetry information to a string. More...
bool	ghost_sparsemat_symmetry_valid (ghost_sparsemat_symmetry symmetry)
	Check if the symmetry information of a sparse matrix is valid. More...
ghost_error	ghost_sparsemat_perm_scotch (ghost_context *ctx, ghost_sparsemat *mat)
	Create a matrix permutation based on (PT-)SCOTCH. More...
ghost_error	ghost_sparsemat_perm_sort (ghost_context *ctx, ghost_sparsemat *mat, ghost_lidx scope)
	Create a matrix permutation based on row length sorting within a given scope. More...
ghost_error	ghost_sparsemat_perm_spmp (ghost_context *ctx, ghost_sparsemat *mat)
ghost_error	ghost_sparsemat_perm_color (ghost_context *ctx, ghost_sparsemat *mat)
	Create a matrix permutation based on 2-way coloring using ColPack. More...
ghost_error	ghost_sparsemat_blockColor (ghost_context *ctx, ghost_sparsemat *mat)
ghost_error	ghost_sparsemat_perm_zoltan (ghost_context *ctx, ghost_sparsemat *mat)
ghost_error	ghost_sparsemat_fromfile_common (ghost_sparsemat *mat, char *matrixPath, ghost_lidx **rpt)
	Common function for matrix creation from a file. More...
ghost_error	ghost_sparsematofile_header (ghost_sparsemat *mat, char *path)
	Write the sparse matrix header to a binary CRS file. More...
ghost_error	ghost_sparsemat_registerrow (ghost_sparsemat *mat, ghost_gidx row, ghost_gidx *col, ghost_lidx rowlen, ghost_lidx stride)
	Store matrix information like bandwidth and nonzero distribution for a given matrix row. More...
ghost_error	ghost_sparsemat_registerrow_finalize (ghost_sparsemat *mat)
	Finalize the storing of matrix information like bandwidth and nonzero distribution. More...
void	ghost_sparsemat_destroy (ghost_sparsemat *mat)
	Destroy a sparsemat and free all memory. More...
ghost_error	ghost_sell_spmv_selector (ghost_densemat *lhs, ghost_sparsemat *mat, ghost_densemat *rhs, ghost_spmv_opts traits)
	Select and call the right SELL SpMV kernel. More...
ghost_error	ghost_cu_sell_spmv_selector (ghost_densemat *lhs, ghost_sparsemat *mat, ghost_densemat *rhs, ghost_spmv_opts traits)
	Select and call the right CUDA SELL SpMV kernel. More...
ghost_error	ghost_sell_stringify_selector (ghost_sparsemat *mat, char **str, int dense)
	Select and call the right SELL stringification function. More...
ghost_error	ghost_sparsemat_string (char **str, ghost_sparsemat *mat, int dense)
	Creates a string of the sparsemat's contents. More...
ghost_error	ghost_sparsemat_init_rowfunc (ghost_sparsemat *mat, ghost_sparsemat_src_rowfunc *src, ghost_mpi_comm mpicomm, double weight)
	Initializes a sparsemat from a row-based callback function. More...
ghost_error	ghost_sparsemat_init_bin (ghost_sparsemat *mat, char *path, ghost_mpi_comm mpicomm, double weight)
	Initializes a sparsemat from a binary CRS file. More...
ghost_error	ghost_sparsemat_init_mm (ghost_sparsemat *mat, char *path, ghost_mpi_comm mpicomm, double weight)
	Initializes a sparsemat from a Matrix Market file. More...
ghost_error	ghost_sparsemat_init_crs (ghost_sparsemat *mat, ghost_gidx offs, ghost_lidx n, ghost_gidx *col, void *val, ghost_lidx *rpt, ghost_mpi_comm mpicomm, double weight)
	Initializes a sparsemat from local CRS data. More...
ghost_error	ghost_sparsemat_to_bin (ghost_sparsemat *mat, char *path)
	Write a matrix to a binary CRS file. More...
size_t	ghost_sparsemat_bytesize (ghost_sparsemat *mat)
	Get the entire memory footprint of the matrix. More...
ghost_error	ghost_cu_sell1_spmv_selector (ghost_densemat *lhs_in, ghost_sparsemat *mat, ghost_densemat *rhs_in, ghost_spmv_opts traits)
ghost_error	ghost_sell_kacz_selector (ghost_densemat *x, ghost_sparsemat *mat, ghost_densemat *b, ghost_kacz_opts opts)
	Select and call the right SELL KACZ kernel. More...
ghost_error	ghost_sell_kacz_shift_selector (ghost_densemat *x_real, ghost_densemat *x_imag, ghost_sparsemat *mat, ghost_densemat *b, double sigma_r, double sigma_i, ghost_kacz_opts opts)
	Select and call the right SELL KACZ kernel with complex shifts. More...
ghost_error	ghost_kacz (ghost_densemat *x, ghost_sparsemat *mat, ghost_densemat *b, ghost_kacz_opts opts)
	Perform a Kaczmarz sweep with the SELL matrix. More...
ghost_error	ghost_kacz_mc (ghost_densemat *x, ghost_sparsemat *mat, ghost_densemat *b, ghost_kacz_opts opts)
ghost_error	ghost_kacz_rb (ghost_densemat *x, ghost_sparsemat *mat, ghost_densemat *b, ghost_kacz_opts opts)
ghost_error	ghost_kacz_bmc (ghost_densemat *x, ghost_sparsemat *mat, ghost_densemat *b, ghost_kacz_opts opts)
ghost_error	ghost_kacz_rb_with_shift (ghost_densemat *x, ghost_sparsemat *mat, ghost_densemat *b, double *shift_r, ghost_kacz_opts opts)
ghost_error	ghost_carp (ghost_sparsemat *mat, ghost_densemat *x, ghost_densemat *b, ghost_carp_opts opts)
ghost_error	checker (ghost_sparsemat *mat)
ghost_error	split_transition (ghost_sparsemat *mat)
ghost_error	ghost_carp_init (ghost_sparsemat *mat, ghost_densemat *b, ghost_carp_opts *opts)
	Initialize CARP. More...
ghost_error	ghost_carp_perf_init (ghost_sparsemat *mat, ghost_carp_opts *opts)
	Finds optimum parameters for CARP. More...
void	ghost_carp_destroy (ghost_carp_opts *opts)
	Destroy(Finalize) CARP. More...
ghost_error	kacz_analyze_print (ghost_sparsemat *mat)
	Prints the row distribution details of KACZ. More...
ghost_error	ghost_sparsemat_to_mm (char *path, ghost_sparsemat *mat)
ghost_error	ghost_context_comm_init (ghost_context *ctx, ghost_gidx *col_orig, ghost_sparsemat *mat, ghost_lidx *col, ghost_lidx *nhalo)
	Assemble communication information in the given context. More...
ghost_error	ghost_sparsemat_perm_global_cols (ghost_gidx *cols, ghost_lidx ncols, ghost_context *context)
static int	ghost_sparsemat_rowfunc_crs (ghost_gidx row, ghost_lidx *rowlen, ghost_gidx *col, void *val, void *crsdata)
ghost_error	calculate_bw (ghost_sparsemat *mat, void *matrixSource, ghost_sparsemat_src srcType)
ghost_error	set_kacz_ratio (ghost_sparsemat *mat, void *matrixSource, ghost_sparsemat_src srcType)

Variables
const ghost_spmv_opts	GHOST_SPMV_OPTS_INITIALIZER
const ghost_kacz_opts	GHOST_KACZ_OPTS_INITIALIZER
const ghost_carp_opts	GHOST_CARP_OPTS_INITIALIZER
const ghost_sparsemat_traits	GHOST_SPARSEMAT_TRAITS_INITIALIZER
	Initialize sparse matrix traits with default values. More...

Detailed Description

Types and functions related to sparse matrices.

Author

Moritz Kreutzer morit.nosp@m.z.kr.nosp@m.eutze.nosp@m.r@fa.nosp@m.u.de

Macro Definition Documentation

#define GHOST_SCOTCH_STRAT_DEFAULT   "g"

#define GHOST_SELL_CHUNKHEIGHT_AUTO   -1

A chunkheight should automatically be determined.

#define GHOST_SELL_CHUNKHEIGHT_ELLPACK   0

Create only a single chunk, i.e., use the ELLPACK storage format.

#define GHOST_SPARSEMAT_PERM_ANY   (GHOST_SPARSEMAT_PERM_ANY_LOCAL|GHOST_SPARSEMAT_PERM_ANY_GLOBAL)

Combination of flags which apply any permutation to a ghost_sparsemat.

#define GHOST_SPARSEMAT_PERM_ANY_GLOBAL   (GHOST_SPARSEMAT_SCOTCHIFY|GHOST_SPARSEMAT_ZOLTAN)

#define GHOST_SPARSEMAT_PERM_ANY_LOCAL   (GHOST_SPARSEMAT_COLOR|GHOST_SPARSEMAT_RCM|GHOST_SPARSEMAT_BLOCKCOLOR|GHOST_SPARSEMAT_SORT_ROWS)

#define GHOST_SPARSEMAT_SORT_GLOBAL   -1

#define GHOST_SPARSEMAT_SORT_LOCAL   -2

#define SPM_GNCOLS

(

mat

)

mat->context->col_map->gdim

#define SPM_NCHUNKS

(

mat

)

(mat->nchunks)

#define SPM_NCOLS

(

mat

)

mat->context->col_map->dim

#define SPM_NNZ

(

mat

)

mat->context->nnz

#define SPM_NROWS

(

mat

)

mat->context->row_map->dim

#define SPM_NROWSPAD

(

mat

)

mat->context->row_map->dimpad

Typedef Documentation

typedef ghost_error(* ghost_kacz_kernel)(ghost_densemat *, ghost_sparsemat *, ghost_densemat *, ghost_kacz_opts)

typedef ghost_error(* ghost_kacz_shift_kernel)(ghost_densemat *, ghost_densemat *, ghost_sparsemat *, ghost_densemat *, double, double, ghost_kacz_opts)

typedef struct ghost_sparsemat ghost_sparsemat

typedef struct ghost_sparsemat_traits ghost_sparsemat_traits

typedef ghost_error(* ghost_spmv_kernel)(ghost_densemat *, ghost_sparsemat *, ghost_densemat *, ghost_spmv_opts)

Enumeration Type Documentation

enum ghost_kacz_direction

Enumerator
GHOST_KACZ_DIRECTION_UNDEFINED
GHOST_KACZ_DIRECTION_FORWARD
GHOST_KACZ_DIRECTION_BACKWARD

enum ghost_kacz_mode

Enumerator
GHOST_KACZ_MODE_NORMAL
GHOST_KACZ_MODE_ECO
GHOST_KACZ_MODE_PERFORMANCE

enum ghost_kacz_normalize

Enumerator
GHOST_KACZ_NORMALIZE_YES
GHOST_KACZ_NORMALIZE_NO
GHOST_KACZ_NORMALIZE_SAVE

enum ghost_sparsemat_flags

Flags to a sparse matrix.

Enumerator
GHOST_SPARSEMAT_DEFAULT	A default sparse matrix.
GHOST_SPARSEMAT_HOST	Matrix is stored on host.
GHOST_SPARSEMAT_DEVICE	Matrix is store on device.
GHOST_SPARSEMAT_NOT_PERMUTE_COLS	If the matrix rows have been re-ordered, do NOT permute the column indices accordingly.
GHOST_SPARSEMAT_PERMUTE	The matrix rows should be re-ordered in a certain way (defined in the traits).
GHOST_SPARSEMAT_NOT_SORT_COLS	Do not sort the matrix cols wrt. memory location.
GHOST_SPARSEMAT_NOT_STORE_SPLIT	Do NOT store the local and remote part of the matrix.
GHOST_SPARSEMAT_NOT_STORE_FULL	Do NOT store the full matrix (local and remote combined).
GHOST_SPARSEMAT_SCOTCHIFY	Reduce the matrix bandwidth with PT-Scotch.
GHOST_SPARSEMAT_SAVE_ORIG_COLS	Save the un-compressed original columns of a distributed matrix.
GHOST_SPARSEMAT_COLOR	Create a matrix permutation reflecting a distance-2-coloring.
GHOST_SPARSEMAT_TRANSPOSE_MM	If the matrix comes from a matrix market file, transpose it on read-in. If this is implemented for other rowfuncs, the _MM may get removed in the future.
GHOST_SPARSEMAT_ZOLTAN	Re-order the matrix globally using Zoltan hypergraph partitioning.
GHOST_SPARSEMAT_RCM	Re-order the local part of the matrix using parallel RCM re-ordering.
GHOST_SPARSEMAT_BLOCKCOLOR	Re-order the local part of the matrix using a block coloring.
GHOST_SOLVER_KACZ	SETS the sparsematrix permutation as needed by the KACZ solver depending on the bandwidth of the matrix.
GHOST_SPARSEMAT_SORT_ROWS	Sort matrix rows according to their length (SELL-C-Sigma sorting)
GHOST_SPARSEMAT_PERM_NO_DISTINCTION	Does not make a distinction between local and remote entries if set; this might lead to higher communication time.
GHOST_SPARSEMAT_DIAG_FIRST	Store the diagonal entry first in each row. Store an explicit zero if the diagonal is zero.

enum ghost_sparsemat_symmetry

Possible sparse matrix symmetries.

Enumerator
GHOST_SPARSEMAT_SYMM_GENERAL	Non-symmetric (general) matrix.
GHOST_SPARSEMAT_SYMM_SYMMETRIC	Symmetric matrix.
GHOST_SPARSEMAT_SYMM_SKEW_SYMMETRIC	Skew-symmetric matrix.
GHOST_SPARSEMAT_SYMM_HERMITIAN	Hermitian matrix.

Function Documentation

ghost_error calculate_bw	(	ghost_sparsemat *	mat,
		void *	matrixSource,
		ghost_sparsemat_src	srcType
	)

ghost_error checker

(

ghost_sparsemat *

mat

)

Here is the call graph for this function:

ghost_error ghost_carp	(	ghost_sparsemat *	mat,
		ghost_densemat *	x,
		ghost_densemat *	b,
		ghost_carp_opts	opts
	)

Here is the call graph for this function:

void ghost_carp_destroy

(

ghost_carp_opts *

opts

)

Destroy(Finalize) CARP.

Parameters

opts	CARP options

ghost_error ghost_carp_init	(	ghost_sparsemat *	mat,
		ghost_densemat *	b,
		ghost_carp_opts *	opts
	)

Initialize CARP.

Parameters

[in]	mat	The sparsematrix
[in]	b	The rhs
	opts	CARP Options.

Here is the call graph for this function:

ghost_error ghost_carp_perf_init	(	ghost_sparsemat *	mat,
		ghost_carp_opts *	opts
	)

Finds optimum parameters for CARP.

Parameters

[in]	mat	The sparsematrix
	opts	Options.

Here is the call graph for this function:

ghost_error ghost_context_comm_init	(	ghost_context *	ctx,
		ghost_gidx *	col_orig,
		ghost_sparsemat *	mat,
		ghost_lidx *	col,
		ghost_lidx *	nhalo
	)

Assemble communication information in the given context.

Parameters

[in,out]	ctx	The context.
[in]	col_orig	The original column indices of the sparse matrix which is bound to the context.
[out]	col	The compressed column indices of the sparse matrix which is bound to the context.
[out]	nhalo	The number of halo elements.

Returns

GHOST_SUCCESS on success or an error indicator.

The following fields of ghost_context are being filled in this function: wishes, wishlist, dues, duelist, hput_pos, wishpartners, nwishpartners, duepartners, nduepartners. Additionally, the columns in col_orig are being compressed and stored in col.

Here is the call graph for this function:

ghost_error ghost_cu_sell1_spmv_selector	(	ghost_densemat *	lhs_in,
		ghost_sparsemat *	mat,
		ghost_densemat *	rhs_in,
		ghost_spmv_opts	traits
	)

ghost_error ghost_cu_sell_spmv_selector	(	ghost_densemat *	lhs,
		ghost_sparsemat *	mat,
		ghost_densemat *	rhs,
		ghost_spmv_opts	traits
	)

Select and call the right CUDA SELL SpMV kernel.

Parameters

mat	The matrix.
lhs	The result densemat.
rhs	The input densemat.
traits	The SpMV traits.

Returns

GHOST_SUCCESS on success or an error indicator.

Here is the call graph for this function:

ghost_error ghost_kacz	(	ghost_densemat *	x,
		ghost_sparsemat *	mat,
		ghost_densemat *	b,
		ghost_kacz_opts	opts
	)

Perform a Kaczmarz sweep with the SELL matrix.

Parameters

x	Output densemat.
mat	The matrix.
b	Input densemat.
opts	Options.

Returns

GHOST_SUCCESS on success or an error indicator.

Here is the call graph for this function:

ghost_error ghost_kacz_bmc	(	ghost_densemat *	x,
		ghost_sparsemat *	mat,
		ghost_densemat *	b,
		ghost_kacz_opts	opts
	)

Here is the call graph for this function:

ghost_error ghost_kacz_mc	(	ghost_densemat *	x,
		ghost_sparsemat *	mat,
		ghost_densemat *	b,
		ghost_kacz_opts	opts
	)

Here is the call graph for this function:

ghost_error ghost_kacz_rb	(	ghost_densemat *	x,
		ghost_sparsemat *	mat,
		ghost_densemat *	b,
		ghost_kacz_opts	opts
	)

Here is the call graph for this function:

ghost_error ghost_kacz_rb_with_shift	(	ghost_densemat *	x,
		ghost_sparsemat *	mat,
		ghost_densemat *	b,
		double *	shift_r,
		ghost_kacz_opts	opts
	)

Here is the call graph for this function:

ghost_error ghost_sell_kacz_selector	(	ghost_densemat *	x,
		ghost_sparsemat *	mat,
		ghost_densemat *	b,
		ghost_kacz_opts	opts
	)

Select and call the right SELL KACZ kernel.

Parameters

mat	The matrix.
x	The result densemat.
b	The input densemat.
opts	The kacz Options.

Returns

GHOST_SUCCESS on success or an error indicator.

ghost_error ghost_sell_kacz_shift_selector	(	ghost_densemat *	x_real,
		ghost_densemat *	x_imag,
		ghost_sparsemat *	mat,
		ghost_densemat *	b,
		double	sigma_r,
		double	sigma_i,
		ghost_kacz_opts	opts
	)

Select and call the right SELL KACZ kernel with complex shifts.

Parameters

mat	The matrix.
x_real	(densemat) The real part of result.
x_imag	(densemat) The imaginary part of result.
b	The input densemat.
sigma_r	The real part of shift
sigma_i	The imaginary part of shift
opts	The kacz Options.

Returns

GHOST_SUCCESS on success or an error indicator.

ghost_error ghost_sell_spmv_selector	(	ghost_densemat *	lhs,
		ghost_sparsemat *	mat,
		ghost_densemat *	rhs,
		ghost_spmv_opts	traits
	)

Select and call the right SELL SpMV kernel.

Parameters

mat	The matrix.
lhs	The result densemat.
rhs	The input densemat.
traits	The SpMV traits.

Returns

GHOST_SUCCESS on success or an error indicator.

Here is the call graph for this function:

ghost_error ghost_sell_stringify_selector	(	ghost_sparsemat *	mat,
		char **	str,
		int	dense
	)

Select and call the right SELL stringification function.

Parameters

mat	The matrix.
str	Where to store the string.
dense	Print in a dense or sparse manner.

Returns

GHOST_SUCCESS on success or an error indicator.

ghost_error ghost_sparsemat_blockColor	(	ghost_context *	ctx,
		ghost_sparsemat *	mat
	)

Here is the call graph for this function:

size_t ghost_sparsemat_bytesize

(

ghost_sparsemat *

mat

)

Get the entire memory footprint of the matrix.

Parameters

mat	The matrix.

Returns

The memory footprint of the matrix in bytes or zero if the matrix is not valid.

ghost_error ghost_sparsemat_fromfile_common	(	ghost_sparsemat *	mat,
		char *	matrixPath,
		ghost_lidx **	rpt
	)

Common function for matrix creation from a file.

This function does work which is independent of the storage format and it should be called at the beginning of a sparse matrix' fromFile function. This function also reads the row pointer from the file and stores it into the parameter rpt.

Parameters

[in]	mat	The sparse matrix.
[in]	matrixPath	The matrix file path.
[out]	rpt	Where to store the row pointer information which may be needed afterwards.

Returns

GHOST_SUCCESS on success or an error indicator.

ghost_error ghost_sparsemat_nnz	(	ghost_gidx *	nnz,
		ghost_sparsemat *	mat
	)

Obtain the global number of nonzero elements of a sparse matrix.

Parameters

[out]	nnz	Where to store the result.
[in]	mat	The sparse matrix.

Returns

GHOST_SUCCESS on success or an error indicator.

ghost_error ghost_sparsemat_nrows	(	ghost_gidx *	nrows,
		ghost_sparsemat *	mat
	)

Obtain the global number of rows of a sparse matrix.

Parameters

[out]	nrows	Where to store the result.
[in]	mat	The sparse matrix.

Returns

GHOST_SUCCESS on success or an error indicator.

ghost_error ghost_sparsemat_perm_color	(	ghost_context *	ctx,
		ghost_sparsemat *	mat
	)

Create a matrix permutation based on 2-way coloring using ColPack.

Parameters

[out]	ctx	The context in which to store the permutations and color information.
[in]	mat	The unpermuted SELL-1-1 source sparse matrix.

Returns

GHOST_SUCCESS on success or an error indicator.

Here is the call graph for this function:

ghost_error ghost_sparsemat_perm_global_cols	(	ghost_gidx *	cols,
		ghost_lidx	ncols,
		ghost_context *	context
	)

Here is the call graph for this function:

ghost_error ghost_sparsemat_perm_scotch	(	ghost_context *	ctx,
		ghost_sparsemat *	mat
	)

Create a matrix permutation based on (PT-)SCOTCH.

Parameters

[out]	ctx	The context in which to store the permutations.
[in]	mat	The unpermuted SELL-1-1 source sparse matrix.

Returns

GHOST_SUCCESS on success or an error indicator.

Here is the call graph for this function:

ghost_error ghost_sparsemat_perm_sort	(	ghost_context *	ctx,
		ghost_sparsemat *	mat,
		ghost_lidx	scope
	)

Create a matrix permutation based on row length sorting within a given scope.

Parameters

[out]	ctx	The context in which to store the permutations.
[in]	mat	The unpermuted SELL-1-1 source sparse matrix.
[in]	scope	The sorting scope.

Returns

GHOST_SUCCESS on success or an error indicator.

Here is the call graph for this function:

ghost_error ghost_sparsemat_perm_spmp	(	ghost_context *	ctx,
		ghost_sparsemat *	mat
	)

Here is the call graph for this function:

ghost_error ghost_sparsemat_perm_zoltan	(	ghost_context *	ctx,
		ghost_sparsemat *	mat
	)

Here is the call graph for this function:

ghost_error ghost_sparsemat_registerrow	(	ghost_sparsemat *	mat,
		ghost_gidx	row,
		ghost_gidx *	col,
		ghost_lidx	rowlen,
		ghost_lidx	stride
	)

Store matrix information like bandwidth and nonzero distribution for a given matrix row.

This function should be called at matrix creation for each row.

Parameters

[out]	mat	The matrix.
[in]	row	The row index.
[in]	col	The column indices of the row.
[in]	rowlen	The length of the row.
[in]	stride	The stride for the parameter col.

Returns

GHOST_SUCCESS on success or an error indicator.

ghost_error ghost_sparsemat_registerrow_finalize

(

ghost_sparsemat *

mat

)

Finalize the storing of matrix information like bandwidth and nonzero distribution.

This function should be after matrix creation. It finalizes the processing of information obtained in ghost_sparsemat_registerrow.

Parameters

[out]

mat

The matrix.

Returns

GHOST_SUCCESS on success or an error indicator.

Here is the call graph for this function:

static int ghost_sparsemat_rowfunc_crs ( ghost_gidx  row, ghost_lidx *  rowlen, ghost_gidx *  col, void *  val, void *  crsdata  )

inlinestatic

bool ghost_sparsemat_symmetry_valid

(

ghost_sparsemat_symmetry

symmetry

)

Check if the symmetry information of a sparse matrix is valid.

Parameters

[in]

symmetry

The symmetry information.

Returns

True if it is valid, false otherwise.

ghost_error ghost_sparsemat_to_bin	(	ghost_sparsemat *	mat,
		char *	path
	)

Write a matrix to a binary CRS file.

Parameters

mat	The matrix.
path	Path of the file.

ghost_error ghost_sparsemat_to_mm	(	char *	path,
		ghost_sparsemat *	mat
	)

Here is the call graph for this function:

ghost_error kacz_analyze_print

(

ghost_sparsemat *

mat

)

Prints the row distribution details of KACZ.

Parameters

[in]

mat

The sparsematrix

Here is the call graph for this function:

ghost_error set_kacz_ratio	(	ghost_sparsemat *	mat,
		void *	matrixSource,
		ghost_sparsemat_src	srcType
	)

ghost_error split_transition

(

ghost_sparsemat *

mat

)

Here is the call graph for this function:

Variable Documentation

const ghost_carp_opts GHOST_CARP_OPTS_INITIALIZER

const ghost_kacz_opts GHOST_KACZ_OPTS_INITIALIZER

const ghost_sparsemat_traits GHOST_SPARSEMAT_TRAITS_INITIALIZER

Initialize sparse matrix traits with default values.

const ghost_spmv_opts GHOST_SPMV_OPTS_INITIALIZER