core_cpxtrfsp.c File Reference

#include "common.h"
#include "cblas.h"
#include "blend/solver.h"
#include "pastix_ccores.h"
#include "kernels_trace.h"

Go to the source code of this file.

Functions
static void	core_cpxtf2sp (pastix_int_t n, pastix_complex32_t *A, pastix_int_t lda, pastix_int_t *nbpivots, float criterion)
	Compute the sequential static pivoting LL^t factorization of the matrix n-by-n A = L * L^t . More...
void	core_cpxtrfsp (pastix_int_t n, pastix_complex32_t *A, pastix_int_t lda, pastix_int_t *nbpivots, float criterion)
	Compute the block static pivoting LL^t factorization of the matrix n-by-n A = L * L^t . More...
int	cpucblk_cpxtrfsp1d_pxtrf (SolverMatrix *solvmtx, SolverCblk *cblk, void *dataL)
	Compute the LL^t factorization of the diagonal block in a panel. More...
int	cpucblk_cpxtrfsp1d_panel (SolverMatrix *solvmtx, SolverCblk *cblk, void *L)
	Compute the LL^t factorization of one panel. More...
int	cpucblk_cpxtrfsp1d (SolverMatrix *solvmtx, SolverCblk *cblk, pastix_complex32_t *work, pastix_int_t lwork)
	Perform the LL^t factorization of a given panel and apply all its updates. More...
int	cpucblk_cpxtrfsp1dplus (SolverMatrix *solvmtx, SolverCblk *cblk)
	Perform the LL^t factorization of a given panel. More...
void	cpucblk_cpxtrfsp1dplus_update (SolverMatrix *solvmtx, SolverBlok *blok, pastix_complex32_t *work, pastix_int_t lwork)
	Apply the updates of the LL^t factorisation of a given panel. More...

Detailed Description

PaStiX kernel routines for LL^t factorization.

Copyright

2011-2023 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria, Univ. Bordeaux. All rights reserved.

Version

6.3.2

Author

Mathieu Faverge

Pierre Ramet

Xavier Lacoste

Date

2023-12-11

Generated arithmetic file

from /builds/solverstack/pastix/kernels/core_zpxtrfsp.c, normal z -> c, Wed Dec 13 12:09:12 2023

Definition in file core_cpxtrfsp.c.

Function Documentation

core_cpxtf2sp()

static void core_cpxtf2sp ( pastix_int_t  n, pastix_complex32_t *  A, pastix_int_t  lda, pastix_int_t *  nbpivots, float  criterion  )

inlinestatic

Compute the sequential static pivoting LL^t factorization of the matrix n-by-n A = L * L^t .

Parameters

[in]	n	The number of rows and columns of the matrix A.
[in,out]	A	The matrix A to factorize with LL^t factorization. The matrix is of size lda -by- n.
[in]	lda	The leading dimension of the matrix A.
[in,out]	nbpivots	Pointer to the number of piovting operations made during factorization. It is updated during this call
[in]	criterion	Threshold use for static pivoting. If diagonal value is under this threshold, its value is replaced by the threshold and the number of pivots is incremented.

Warning

This routine will fail if it discovers a null or negative value on the diagonal during factorization.

Definition at line 66 of file core_cpxtrfsp.c.

References pastix_int_t.

Referenced by core_cpxtrfsp().

cpucblk_cpxtrfsp1d_pxtrf()

int cpucblk_cpxtrfsp1d_pxtrf	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk,
		void *	dataL
	)

Compute the LL^t factorization of the diagonal block in a panel.

Warning

This routine will fail if it discovers a null or negative value on the diagonal during factorization.

Parameters

[in]	solvmtx	Solver Matrix structure of the problem
[in]	cblk	Pointer to the structure representing the panel to factorize in the cblktab array. Next column blok must be accessible through cblk[1].
[in,out]	dataL	The pointer to the correct representation of the lower part of the data. coeftab if the block is in full rank. Must be of size cblk.stride -by- cblk.width. pastix_lr_block if the block is compressed.

Returns

The number of static pivoting performed during the diagonal block factorization.

Definition at line 212 of file core_cpxtrfsp.c.

References solver_cblk_s::cblktype, solver_matrix_s::diagthreshold, solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, kernel_trace_start(), solver_cblk_s::lcolnum, solver_blok_s::lrownum, pastix_int_t, PastixKernelPXTRF, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, solver_cblk_s::stride, and pastix_lrblock_s::u.

Referenced by cpucblk_cpxtrfsp1d_panel(), and fct_blok_cpxtrfsp_cpu().

cpucblk_cpxtrfsp1d_panel()

int cpucblk_cpxtrfsp1d_panel	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk,
		void *	L
	)

Compute the LL^t factorization of one panel.

Parameters

[in]	solvmtx	Solver Matrix structure of the problem
[in]	cblk	Pointer to the structure representing the panel to factorize in the cblktab array. Next column blok must be accessible through cblk[1].
[in,out]	L	The pointer to the correct representation of the lower part of the data. coeftab if the block is in full rank. Must be of size cblk.stride -by- cblk.width. pastix_lr_block if the block is compressed.

Returns

The number of static pivoting during factorization of the diagonal block.

Definition at line 284 of file core_cpxtrfsp.c.

References cpucblk_cpxtrfsp1d_pxtrf(), cpucblk_ctrsmsp(), solver_matrix_s::lowrank, pastix_int_t, PastixLower, PastixNonUnit, PastixRight, and PastixTrans.

Referenced by cpucblk_cpxtrfsp1d(), and cpucblk_cpxtrfsp1dplus().

cpucblk_cpxtrfsp1d()

int cpucblk_cpxtrfsp1d	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk,
		pastix_complex32_t *	work,
		pastix_int_t	lwork
	)

Perform the LL^t factorization of a given panel and apply all its updates.

Parameters

[in]	solvmtx	Solver Matrix structure of the problem
[in]	cblk	Pointer to the structure representing the panel to factorize in the cblktab array. Next column blok must be accessible through cblk[1].
[in]	work	Temporary memory buffer.
[in]	lwork	Temporary workspace dimension.

Returns

The number of static pivoting during factorization of the diagonal block.

Definition at line 326 of file core_cpxtrfsp.c.

References cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_calloc(), cpucblk_cgemmsp(), cpucblk_cpxtrfsp1d_panel(), cpucblk_crelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_matrix_s::lowrank, pastix_int_t, PastixLCoef, and PastixTrans.

Referenced by sequential_cpxtrf(), thread_cpxtrf_dynamic(), and thread_cpxtrf_static().

cpucblk_cpxtrfsp1dplus()

int cpucblk_cpxtrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the LL^t factorization of a given panel.

Parameters

[in]	solvmtx	Solver Matrix structure of the problem
[in]	cblk	Pointer to the structure representing the panel to factorize in the cblktab array. Next column blok must be accessible through cblk[1].

Returns

The number of static pivoting during factorization of the diagonal block.

Definition at line 382 of file core_cpxtrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_cpxtrfsp1d_panel(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, pastix_int_t, pqueuePush1(), solver_cblk_s::priority, and solver_cblk_s::threadid.

Referenced by thread_cpxtrf_dynamic().

cpucblk_cpxtrfsp1dplus_update()

void cpucblk_cpxtrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		pastix_complex32_t *	work,
		pastix_int_t	lwork
	)

Apply the updates of the LL^t factorisation of a given panel.

Parameters

[in]	solvmtx	Solver Matrix structure of the problem
[in]	blok	Pointer to the blok where the update start.
[in]	work	Temporary memory buffer.
[in]	lwork	Temporary workspace dimension.

Definition at line 435 of file core_cpxtrfsp.c.

References cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_calloc(), cpucblk_cgemmsp(), cpucblk_crelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_blok_s::lcblknm, solver_matrix_s::lowrank, PastixLCoef, and PastixTrans.

Referenced by thread_cpxtrf_dynamic().