8.1
general documentation
cs_param_sles.h File Reference

Structure and routines handling the SLES settings stored inside a cs_param_sles_t structure. More...

#include "cs_param_types.h"
+ Include dependency graph for cs_param_sles.h:

Go to the source code of this file.

Data Structures

struct  cs_param_sles_cvg_t
 Set of parameters to check the convergence (or the divergence) of an iterative process (tolerances or max. number of iterations) More...
 
struct  cs_param_sles_mumps_t
 Set of parameters to specify additional options to MUMPS For more advanced settings, one has to use the cs_user_sles_mumps_hook function. Please also refer to the MUMPS user guide for more details. More...
 
struct  cs_param_sles_t
 Structure storing all metadata related to the resolution of a linear system with an iterative solver. More...
 
struct  cs_param_sles_saddle_t
 Structure storing all metadata related to the resolution of a saddle-point linear system. More...
 

Enumerations

enum  cs_param_sles_facto_type_t { CS_PARAM_SLES_FACTO_LU , CS_PARAM_SLES_FACTO_LDLT_SYM , CS_PARAM_SLES_FACTO_LDLT_SPD , CS_PARAM_SLES_N_FACTO_TYPES }
 type of factorization to consider when using the MUMPS solver to solve a linear system More...
 
enum  cs_param_sles_analysis_algo_t {
  CS_PARAM_SLES_ANALYSIS_AMD , CS_PARAM_SLES_ANALYSIS_QAMD , CS_PARAM_SLES_ANALYSIS_PORD , CS_PARAM_SLES_ANALYSIS_SCOTCH ,
  CS_PARAM_SLES_ANALYSIS_PTSCOTCH , CS_PARAM_SLES_ANALYSIS_METIS , CS_PARAM_SLES_ANALYSIS_PARMETIS , CS_PARAM_SLES_ANALYSIS_AUTO ,
  CS_PARAM_SLES_N_ANALYSIS_ALGOS
}
 Type of algorithm to consider when using the MUMPS solver to perform the analysis step (renumbering and graph manipulation). Please refer to the MUMPS user guide for more details about the following algorithms. AMD, QAMD and PORD are available with MUMPS without any prerequesite. More...
 

Functions

static bool cs_param_sles_is_mumps_set (cs_param_itsol_type_t solver)
 Find if a MUMPS-related solver is set or not. More...
 
cs_param_sles_saddle_tcs_param_sles_saddle_create (void)
 Create a cs_param_sles_saddle_t structure and assign a minimalist default settings. More...
 
void cs_param_sles_saddle_init_schur (const char *basename, cs_param_sles_saddle_t *saddlep)
 Initialize a cs_param_sles_t structure for the Schur approximation nested inside a ref cs_param_sles_saddle_t structure. By default, this member is not allocated. Do nothing if the related structure is already allocated. More...
 
void cs_param_sles_saddle_copy (const cs_param_sles_saddle_t *ref, cs_param_sles_saddle_t *dest)
 Copy a cs_param_sles_saddle_t structure from ref to dest. More...
 
void cs_param_sles_saddle_free (cs_param_sles_saddle_t **p_saddlep)
 Free the structure storing the parameter settings for a saddle-point system. More...
 
cs_param_sles_tcs_param_sles_create (int field_id, const char *system_name)
 Create a cs_param_sles_t structure and assign a default settings. More...
 
void cs_param_sles_free (cs_param_sles_t **p_slesp)
 Free a cs_param_sles_t structure. More...
 
void cs_param_sles_log (cs_param_sles_t *slesp)
 Log information related to the linear settings stored in the structure. More...
 
void cs_param_sles_mumps (cs_param_sles_t *slesp, bool is_single, cs_param_sles_facto_type_t facto_type)
 Set the main memebers of a cs_param_sles_mumps_t structure. This structure is allocated if needed. Other members are kept to their values. More...
 
void cs_param_sles_mumps_advanced (cs_param_sles_t *slesp, cs_param_sles_analysis_algo_t analysis_algo, int block_analysis, double mem_coef, double blr_threshold, int ir_steps, bool advanced_optim)
 Set the main memebers of a cs_param_sles_mumps_t structure. This structure is allocated if needed. Other members are kept to their values. Please refer to the MUMPS user guide for more details about the following advanced options. More...
 
void cs_param_sles_copy_from (cs_param_sles_t *src, cs_param_sles_t *dst)
 Copy a cs_param_sles_t structure from src to dst. More...
 
int cs_param_sles_set (bool use_field_id, cs_param_sles_t *slesp)
 Define cs_sles_t structure in accordance with the settings of a cs_param_sles_t structure (SLES = Sparse Linear Equation Solver) More...
 
void cs_param_sles_update_cvg_settings (bool use_field_id, const cs_param_sles_t *slesp)
 Update the settings associated to a cs_sles_t structure and apply those defined in the given cs_param_sles_t structure. This function is used only when a first setup has been performed. More...
 
cs_param_sles_class_t cs_param_sles_get_class_from_amg (cs_param_amg_type_t amg_type)
 Retrieve the related solver class from the amg type. More...
 
cs_param_sles_class_t cs_param_sles_check_class (cs_param_sles_class_t wanted_class)
 Check the availability of a solver library and return the requested one if this is possible or an alternative or CS_PARAM_SLES_N_CLASSES if no alternative is available. More...
 
void cs_param_sles_check_amg (cs_param_sles_t *slesp)
 Check if the setting related to the AMG is consistent with the solver class. More...
 

Detailed Description

Structure and routines handling the SLES settings stored inside a cs_param_sles_t structure.

Enumeration Type Documentation

◆ cs_param_sles_analysis_algo_t

Type of algorithm to consider when using the MUMPS solver to perform the analysis step (renumbering and graph manipulation). Please refer to the MUMPS user guide for more details about the following algorithms. AMD, QAMD and PORD are available with MUMPS without any prerequesite.

Enumerator
CS_PARAM_SLES_ANALYSIS_AMD 

AMD is a sequential algorithm which is well-suited for 2D problem (for 3D problems it induces a higher memeory consumption).

CS_PARAM_SLES_ANALYSIS_QAMD 

QAMD is a sequential algorithm which is well-suited for 2D problem (for 3D problems it induces a higher memeory consumption).

CS_PARAM_SLES_ANALYSIS_PORD 

PORD is a sequential algorithm which is a good trade-off when MUMPS is installed with no prerequisite such as METIS or Scotch.

CS_PARAM_SLES_ANALYSIS_SCOTCH 

SCOTCH is a sequential algorithm which delivers the very good performance with 3D meshes, generally, better than PORD and not as good as METIS

CS_PARAM_SLES_ANALYSIS_PTSCOTCH 

PTSCOTCH is a parallel version of the sequential SCOTCH algorithm

CS_PARAM_SLES_ANALYSIS_METIS 

METIS is a sequential algorithm which delivers the best performance in case of 2D meshes.

CS_PARAM_SLES_ANALYSIS_PARMETIS 

PARMETIS is a parallel version of the sequential METIS algorithm

CS_PARAM_SLES_ANALYSIS_AUTO 

MUMPS decides what is the best choice among available algorithms. This is the default choice.

CS_PARAM_SLES_N_ANALYSIS_ALGOS 

◆ cs_param_sles_facto_type_t

type of factorization to consider when using the MUMPS solver to solve a linear system

Enumerator
CS_PARAM_SLES_FACTO_LU 

LU factorization is the most generic factorization available with MUMPS. It can handle general matrices (block and/or unsymmetric matrices)

CS_PARAM_SLES_FACTO_LDLT_SYM 

This factorization is a Cholesky factorization (L.D.Lt) for general symmetric matrices.

CS_PARAM_SLES_FACTO_LDLT_SPD 

This factorization is devoted to SPD matrices and corresponds to a Cholesky factorization. This is more specific and thus more efficient than CS_PARAM_SLES_FACTO_LDLT_SYM.

CS_PARAM_SLES_N_FACTO_TYPES 

Function Documentation

◆ cs_param_sles_check_amg()

void cs_param_sles_check_amg ( cs_param_sles_t slesp)

Check if the setting related to the AMG is consistent with the solver class.

Parameters
[in,out]slesppointer to a cs_pparam_sles_t structure

◆ cs_param_sles_check_class()

cs_param_sles_class_t cs_param_sles_check_class ( cs_param_sles_class_t  wanted_class)

Check the availability of a solver library and return the requested one if this is possible or an alternative or CS_PARAM_SLES_N_CLASSES if no alternative is available.

Parameters
[in]wanted_classrequested class of solvers
Returns
the available solver class related to the requested class

◆ cs_param_sles_copy_from()

void cs_param_sles_copy_from ( cs_param_sles_t src,
cs_param_sles_t dst 
)

Copy a cs_param_sles_t structure from src to dst.

Parameters
[in]srcreference cs_param_sles_t structure to copy
[in,out]dstcopy of the reference at exit

◆ cs_param_sles_create()

cs_param_sles_t* cs_param_sles_create ( int  field_id,
const char *  system_name 
)

Create a cs_param_sles_t structure and assign a default settings.

Parameters
[in]field_idid related to to the variable field or -1
[in]system_namename of the system to solve or NULL
Returns
a pointer to a cs_param_sles_t stucture

◆ cs_param_sles_free()

void cs_param_sles_free ( cs_param_sles_t **  p_slesp)

Free a cs_param_sles_t structure.

Parameters
[in,out]slesppointer to a \cs_param_sles_t structure to free

◆ cs_param_sles_get_class_from_amg()

cs_param_sles_class_t cs_param_sles_get_class_from_amg ( cs_param_amg_type_t  amg_type)

Retrieve the related solver class from the amg type.

Parameters
[in]amg_typetype of AMG to consider
Returns
the related solver class or CS_PARAM_SLES_CLASS_CS

◆ cs_param_sles_is_mumps_set()

static bool cs_param_sles_is_mumps_set ( cs_param_itsol_type_t  solver)
inlinestatic

Find if a MUMPS-related solver is set or not.

Parameters
[in]solvertype of solver
Returns
true or false

◆ cs_param_sles_log()

void cs_param_sles_log ( cs_param_sles_t slesp)

Log information related to the linear settings stored in the structure.

Parameters
[in]slesppointer to a cs_param_sles_log

◆ cs_param_sles_mumps()

void cs_param_sles_mumps ( cs_param_sles_t slesp,
bool  is_single,
cs_param_sles_facto_type_t  facto_type 
)

Set the main memebers of a cs_param_sles_mumps_t structure. This structure is allocated if needed. Other members are kept to their values.

Parameters
[in,out]slesppointer to a cs_param_sles_t structure
[in]is_singlesingle-precision or double-precision
[in]facto_typetype of factorization to consider

◆ cs_param_sles_mumps_advanced()

void cs_param_sles_mumps_advanced ( cs_param_sles_t slesp,
cs_param_sles_analysis_algo_t  analysis_algo,
int  block_analysis,
double  mem_coef,
double  blr_threshold,
int  ir_steps,
bool  advanced_optim 
)

Set the main memebers of a cs_param_sles_mumps_t structure. This structure is allocated if needed. Other members are kept to their values. Please refer to the MUMPS user guide for more details about the following advanced options.

Parameters
[in,out]slesppointer to a cs_param_sles_t structure
[in]analysis_algoalgorithm used for the analysis step
[in]block_analysis< 0: fixed block size; 0: nothing
[in]mem_coefpercentage increase in the memory workspace
[in]blr_thresholdAccuracy in BLR compression (< 0: not used)
[in]ir_steps0: No, otherwise number of iterations
[in]advanced_optimactivate advanced optimization (MPI/openMP)

◆ cs_param_sles_saddle_copy()

void cs_param_sles_saddle_copy ( const cs_param_sles_saddle_t ref,
cs_param_sles_saddle_t dest 
)

Copy a cs_param_sles_saddle_t structure from ref to dest.

Parameters
[in]refreference structure to be copied
[in,out]destdestination structure

◆ cs_param_sles_saddle_create()

cs_param_sles_saddle_t* cs_param_sles_saddle_create ( void  )

Create a cs_param_sles_saddle_t structure and assign a minimalist default settings.

Returns
a pointer to the new cs_param_sles_saddle_t structure

◆ cs_param_sles_saddle_free()

void cs_param_sles_saddle_free ( cs_param_sles_saddle_t **  p_saddlep)

Free the structure storing the parameter settings for a saddle-point system.

Parameters
[in,out]p_saddlepdouble pointer to the structure to free

◆ cs_param_sles_saddle_init_schur()

void cs_param_sles_saddle_init_schur ( const char *  basename,
cs_param_sles_saddle_t saddlep 
)

Initialize a cs_param_sles_t structure for the Schur approximation nested inside a ref cs_param_sles_saddle_t structure. By default, this member is not allocated. Do nothing if the related structure is already allocated.

Parameters
[in]basenameprefix for the naming of the Schur system
[in,out]saddleppointer to the structure to update

◆ cs_param_sles_set()

int cs_param_sles_set ( bool  use_field_id,
cs_param_sles_t slesp 
)

Define cs_sles_t structure in accordance with the settings of a cs_param_sles_t structure (SLES = Sparse Linear Equation Solver)

Parameters
[in]use_field_idif false use system name to define a SLES
[in,out]slesppointer to a cs_param_sles_t structure
Returns
an error code (-1 if a problem is encountered, 0 otherwise)

◆ cs_param_sles_update_cvg_settings()

void cs_param_sles_update_cvg_settings ( bool  use_field_id,
const cs_param_sles_t slesp 
)

Update the settings associated to a cs_sles_t structure and apply those defined in the given cs_param_sles_t structure. This function is used only when a first setup has been performed.

One modifies only some specific options like the max. number of iterations or the relative tolerance

Parameters
[in]use_field_idif false use a name to retrieve the cs_sles_t struc.
[in]slesppointer to a cs_param_sles_t structure