PaStiX Handbook: Internal functions for numerical factorization kernels

Functions
static void	core_dgemmsp_1d1d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const double A, const double B, double C, double *work)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static void	core_dgemmsp_1d2d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const double A, const double B, double C)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static void	core_dgemmsp_2d2d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const double A, const double B, double C)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static pastix_fixdbl_t	core_dgemmsp_block_frfr (pastix_trans_t trans, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const double A, const double B, double *C)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_dgemmsp_block_frlr (pastix_trans_t transB, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const double A, const double B, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_dgemmsp_block_lrlr (pastix_trans_t transB, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_dgemmsp_fulllr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const double A, const double B, pastix_lrblock_t lrC, double work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static pastix_fixdbl_t	core_dgemmsp_lr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, pastix_lrblock_t lrC, double work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static pastix_fixdbl_t	core_dgemmsp_lrfr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, double C, double work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static void	core_zgemmsp_1d1d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_complex64_t A, const pastix_complex64_t B, pastix_complex64_t C, pastix_complex64_t *work)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static void	core_zgemmsp_1d2d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_complex64_t A, const pastix_complex64_t B, pastix_complex64_t C)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static void	core_zgemmsp_2d2d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_complex64_t A, const pastix_complex64_t B, pastix_complex64_t C)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static pastix_fixdbl_t	core_zgemmsp_block_frfr (pastix_trans_t trans, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const pastix_complex64_t A, const pastix_complex64_t B, pastix_complex64_t *C)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_zgemmsp_block_frlr (pastix_trans_t transB, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const pastix_complex64_t A, const pastix_complex64_t B, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_zgemmsp_block_lrlr (pastix_trans_t transB, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_zgemmsp_fulllr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_complex64_t A, const pastix_complex64_t B, pastix_lrblock_t lrC, pastix_complex64_t work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static pastix_fixdbl_t	core_zgemmsp_lr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, pastix_lrblock_t lrC, pastix_complex64_t work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static pastix_fixdbl_t	core_zgemmsp_lrfr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, pastix_complex64_t C, pastix_complex64_t work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static void	core_cgemmsp_1d1d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_complex32_t A, const pastix_complex32_t B, pastix_complex32_t C, pastix_complex32_t *work)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static void	core_cgemmsp_1d2d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_complex32_t A, const pastix_complex32_t B, pastix_complex32_t C)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static void	core_cgemmsp_2d2d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_complex32_t A, const pastix_complex32_t B, pastix_complex32_t C)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static pastix_fixdbl_t	core_cgemmsp_block_frfr (pastix_trans_t trans, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const pastix_complex32_t A, const pastix_complex32_t B, pastix_complex32_t *C)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_cgemmsp_block_frlr (pastix_trans_t transB, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const pastix_complex32_t A, const pastix_complex32_t B, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_cgemmsp_block_lrlr (pastix_trans_t transB, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_cgemmsp_fulllr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_complex32_t A, const pastix_complex32_t B, pastix_lrblock_t lrC, pastix_complex32_t work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static pastix_fixdbl_t	core_cgemmsp_lr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, pastix_lrblock_t lrC, pastix_complex32_t work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static pastix_fixdbl_t	core_cgemmsp_lrfr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, pastix_complex32_t C, pastix_complex32_t work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static void	core_sgemmsp_1d1d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const float A, const float B, float C, float *work)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static void	core_sgemmsp_1d2d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const float A, const float B, float C)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static void	core_sgemmsp_2d2d (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const float A, const float B, float C)
	Compute the updates that are generated by the transposition of one single off-diagonal block. More...

static pastix_fixdbl_t	core_sgemmsp_block_frfr (pastix_trans_t trans, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const float A, const float B, float *C)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_sgemmsp_block_frlr (pastix_trans_t transB, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const float A, const float B, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_sgemmsp_block_lrlr (pastix_trans_t transB, pastix_int_t blok_mk, pastix_int_t blok_kn, pastix_int_t blok_mn, const SolverCblk cblk, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks. More...

static pastix_fixdbl_t	core_sgemmsp_fulllr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const float A, const float B, pastix_lrblock_t lrC, float work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static pastix_fixdbl_t	core_sgemmsp_lr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, pastix_lrblock_t lrC, float work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static pastix_fixdbl_t	core_sgemmsp_lrfr (pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_lrblock_t lrA, const pastix_lrblock_t lrB, float C, float work, pastix_int_t lwork, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block. More...

static void	core_dtrsmsp_1d (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const double A, double *C)
	Apply all the trsm updates on a panel stored in 1D layout. More...

static void	core_dtrsmsp_2d (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const double A, double *C)
	Compute the updates associated to one off-diagonal block between two cblk stored in 2D. More...

static pastix_fixdbl_t	core_dtrsmsp_lr (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const pastix_lrblock_t lrA, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Computes the updates associated to one off-diagonal block between two cblk stored in low-rank format. More...

static pastix_fixdbl_t	core_dtrsmsp_2dsub (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, pastix_int_t blok_m, const double A, double *C)
	Compute the updates associated to one off-diagonal block between two cblk stored in 2D. More...

static pastix_fixdbl_t	core_dtrsmsp_lrsub (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, pastix_int_t blok_m, const pastix_lrblock_t lrA, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block between two cblk stored in low-rank format. More...

static void	core_ztrsmsp_1d (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const pastix_complex64_t A, pastix_complex64_t *C)
	Apply all the trsm updates on a panel stored in 1D layout. More...

static void	core_ztrsmsp_2d (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const pastix_complex64_t A, pastix_complex64_t *C)
	Compute the updates associated to one off-diagonal block between two cblk stored in 2D. More...

static pastix_fixdbl_t	core_ztrsmsp_lr (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const pastix_lrblock_t lrA, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Computes the updates associated to one off-diagonal block between two cblk stored in low-rank format. More...

static pastix_fixdbl_t	core_ztrsmsp_2dsub (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, pastix_int_t blok_m, const pastix_complex64_t A, pastix_complex64_t *C)
	Compute the updates associated to one off-diagonal block between two cblk stored in 2D. More...

static pastix_fixdbl_t	core_ztrsmsp_lrsub (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, pastix_int_t blok_m, const pastix_lrblock_t lrA, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block between two cblk stored in low-rank format. More...

static void	core_ctrsmsp_1d (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const pastix_complex32_t A, pastix_complex32_t *C)
	Apply all the trsm updates on a panel stored in 1D layout. More...

static void	core_ctrsmsp_2d (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const pastix_complex32_t A, pastix_complex32_t *C)
	Compute the updates associated to one off-diagonal block between two cblk stored in 2D. More...

static pastix_fixdbl_t	core_ctrsmsp_lr (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const pastix_lrblock_t lrA, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Computes the updates associated to one off-diagonal block between two cblk stored in low-rank format. More...

static pastix_fixdbl_t	core_ctrsmsp_2dsub (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, pastix_int_t blok_m, const pastix_complex32_t A, pastix_complex32_t *C)
	Compute the updates associated to one off-diagonal block between two cblk stored in 2D. More...

static pastix_fixdbl_t	core_ctrsmsp_lrsub (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, pastix_int_t blok_m, const pastix_lrblock_t lrA, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block between two cblk stored in low-rank format. More...

static void	core_strsmsp_1d (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const float A, float *C)
	Apply all the trsm updates on a panel stored in 1D layout. More...

static void	core_strsmsp_2d (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const float A, float *C)
	Compute the updates associated to one off-diagonal block between two cblk stored in 2D. More...

static pastix_fixdbl_t	core_strsmsp_lr (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, const pastix_lrblock_t lrA, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Computes the updates associated to one off-diagonal block between two cblk stored in low-rank format. More...

static pastix_fixdbl_t	core_strsmsp_2dsub (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, pastix_int_t blok_m, const float A, float *C)
	Compute the updates associated to one off-diagonal block between two cblk stored in 2D. More...

static pastix_fixdbl_t	core_strsmsp_lrsub (pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk cblk, pastix_int_t blok_m, const pastix_lrblock_t lrA, pastix_lrblock_t lrC, const pastix_lr_t lowrank)
	Compute the updates associated to one off-diagonal block between two cblk stored in low-rank format. More...

void	core_csytrfsp1d_gemm (const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_complex32_t L, pastix_complex32_t C, pastix_complex32_t work)

int	cpucblk_cpotrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the Cholesky factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_cpotrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, pastix_complex32_t *work, pastix_int_t lwork)
	Apply the updates of the cholesky factorisation of a given panel. More...

int	cpucblk_csytrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the LDL^t factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_csytrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, pastix_complex32_t *work)
	Apply the updates of the LDL^t factorisation of a given panel. More...

int	cpucblk_cgetrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the LU factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_cgetrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, pastix_complex32_t *work, pastix_int_t lwork)
	Apply the updates of the LU factorisation of a given panel. More...

void	core_ssytrfsp1d_gemm (const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const float L, float C, float work)

int	cpucblk_spotrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the Cholesky factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_spotrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, float *work, pastix_int_t lwork)
	Apply the updates of the cholesky factorisation of a given panel. More...

int	cpucblk_ssytrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the LDL^t factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_ssytrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, float *work)
	Apply the updates of the LDL^t factorisation of a given panel. More...

int	cpucblk_sgetrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the LU factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_sgetrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, float *work, pastix_int_t lwork)
	Apply the updates of the LU factorisation of a given panel. More...

void	core_zsytrfsp1d_gemm (const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const pastix_complex64_t L, pastix_complex64_t C, pastix_complex64_t work)

int	cpucblk_zpotrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the Cholesky factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_zpotrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, pastix_complex64_t *work, pastix_int_t lwork)
	Apply the updates of the cholesky factorisation of a given panel. More...

int	cpucblk_zsytrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the LDL^t factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_zsytrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, pastix_complex64_t *work)
	Apply the updates of the LDL^t factorisation of a given panel. More...

int	cpucblk_zgetrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the LU factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_zgetrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, pastix_complex64_t *work, pastix_int_t lwork)
	Apply the updates of the LU factorisation of a given panel. More...

void	core_dsytrfsp1d_gemm (const SolverCblk cblk, const SolverBlok blok, SolverCblk fcblk, const double L, double C, double work)

int	cpucblk_dpotrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the Cholesky factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_dpotrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, double *work, pastix_int_t lwork)
	Apply the updates of the cholesky factorisation of a given panel. More...

int	cpucblk_dsytrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the LDL^t factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_dsytrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, double *work)
	Apply the updates of the LDL^t factorisation of a given panel. More...

int	cpucblk_dgetrfsp1dplus (SolverMatrix solvmtx, SolverCblk cblk)
	Perform the LU factorization of a given panel and submit tasks for the subsequent updates. More...

void	cpucblk_dgetrfsp1dplus_update (SolverMatrix solvmtx, SolverBlok blok, double *work, pastix_int_t lwork)
	Apply the updates of the LU factorisation of a given panel. More...

Detailed Description

This module contains the three terms update functions for the LDL^t and LDL^h factorizations.

Function Documentation

◆ core_dgemmsp_1d1d()

static void core_dgemmsp_1d1d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const double *	A,
		const double *	B,
		double *	C,
		double *	work
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

Both cblk involved in the computation are stored with the 1D storage: Column Major Layout with blocks interleaved.

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.
[in]	work	Temporary memory buffer that is at least equal to the height of the block B by the sum of the height of all the blocks below the block B.

See also: core_dgemmsp_1d2d; core_dgemmsp_2d2d

Definition at line 106 of file core_dgemmsp.c.

References blok_rownbr(), cblk_colnbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, pastix_int_t, PastixUCoef, and solver_cblk_s::stride.

Referenced by cpucblk_dgemmsp().

◆ core_dgemmsp_1d2d()

static void core_dgemmsp_1d2d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const double *	A,
		const double *	B,
		double *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

The cblk involved in the matrices A and B are stored with the 1D storage: Column Major Layout with blocks interleaved. The facing cblk of the atrix C, is stored with the 2D storage where each block is stored continuously one after another. (Similar to dense tile storage with variant tile size)

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

See also: core_dgemmsp_1d1d; core_dgemmsp_2d2d

Add contribution to C in fcblk: Get the first facing block of the distant panel, and the last block of the current cblk

Definition at line 264 of file core_dgemmsp.c.

References blok_rownbr(), cblk_colnbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), pastix_int_t, PastixUCoef, and solver_cblk_s::stride.

Referenced by cpucblk_dgemmsp().

◆ core_dgemmsp_2d2d()

static void core_dgemmsp_2d2d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const double *	A,
		const double *	B,
		double *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

Both cblk involved in the matrices A, B and C are stored with the 2D storage where each block is stored continuously one after another. (Similar to dense tile storage with variant tile size)

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

See also: core_dgemmsp_1d1d; core_dgemmsp_1d2d

Definition at line 404 of file core_dgemmsp.c.

Referenced by cpucblk_dgemmsp().

◆ core_dgemmsp_block_frfr()

static pastix_fixdbl_t core_dgemmsp_block_frfr	(	pastix_trans_t	trans,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const double *	A,
		const double *	B,
		double *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to performed an update:

                 C_mn = C_mn - A_mk * op(B_kn)

where A_mk is the set of blocks in cblk k, facing the diagonal block of the cblk m; B_kn is the set of blocks in cblk n facing the diagonal block of the cblk k; and C_mn is the set of blocks impacted by this update, it necessarily belongs to the set of block of the cblk n facing the diagonal block of the cblk m.

Parameters

[in]	trans	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Returns: The number of flops performed

See also: core_dgemmsp_block_frlr; core_dgemmsp_block_lrlr

Definition at line 545 of file core_dgemmsp.c.

References blok_rownbr(), cblk_colnbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, solver_blok_s::frownum, solver_blok_s::inlast, is_block_inside_fblock(), kernel_trace_start(), kernel_trace_stop(), solver_blok_s::lcblknm, pastix_int_t, and PastixKernelGEMMBlok2d2d.

Referenced by cpublok_dgemmsp().

◆ core_dgemmsp_block_frlr()

static pastix_fixdbl_t core_dgemmsp_block_frlr	(	pastix_trans_t	transB,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const double *	A,
		const double *	B,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to perform an update:

                 C_mn = C_mn - A_mk * op(B_kn)

In this routine, all the matrices are low-rank

Parameters

[in]	transB	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 712 of file core_dgemmsp.c.

References core_dlrmm_s::A, core_dlrmm_s::alpha, core_dlrmm_s::B, core_dlrmm_s::beta, blok_rownbr(), core_dlrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_dlrmm_s::Cm, core_dlrmm_s::Cn, solver_blok_s::coefind, core_dlrmm(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, solver_blok_s::frownum, solver_blok_s::inlast, is_block_inside_fblock(), core_dlrmm_s::K, kernel_trace_start(), kernel_trace_stop(), solver_blok_s::lcblknm, core_dlrmm_s::lock, core_dlrmm_s::lowrank, core_dlrmm_s::lwork, core_dlrmm_s::M, core_dlrmm_s::N, core_dlrmm_s::offx, core_dlrmm_s::offy, pastix_int_t, PastixKernelGEMMBlok2d2d, PastixKernelGEMMBlokLRLR, PastixNoTrans, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, core_dlrmm_s::transA, core_dlrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_dlrmm_s::work.

Referenced by cpublok_dgemmsp().

◆ core_dgemmsp_block_lrlr()

static pastix_fixdbl_t core_dgemmsp_block_lrlr	(	pastix_trans_t	transB,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to perform an update:

                 C_mn = C_mn - A_mk * op(B_kn)

In this routine, all the matrices are low-rank

Parameters

[in]	transB	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in,out]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 901 of file core_dgemmsp.c.

References core_dlrmm_s::A, core_dlrmm_s::alpha, core_dlrmm_s::B, core_dlrmm_s::beta, blok_rownbr(), core_dlrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_dlrmm_s::Cm, core_dlrmm_s::Cn, core_dlrmm(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, solver_blok_s::frownum, solver_blok_s::inlast, is_block_inside_fblock(), core_dlrmm_s::K, kernel_trace_start(), kernel_trace_stop(), solver_blok_s::lcblknm, core_dlrmm_s::lock, core_dlrmm_s::lowrank, core_dlrmm_s::lwork, core_dlrmm_s::M, core_dlrmm_s::N, core_dlrmm_s::offx, core_dlrmm_s::offy, pastix_int_t, PastixKernelGEMMBlok2d2d, PastixKernelGEMMBlokLRLR, PastixNoTrans, core_dlrmm_s::transA, core_dlrmm_s::transB, and core_dlrmm_s::work.

Referenced by cpublok_dgemmsp().

◆ core_dgemmsp_fulllr()

static pastix_fixdbl_t core_dgemmsp_fulllr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const double *	A,
		const double *	B,
		pastix_lrblock_t *	lrC,
		double *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	work	Temporary memory buffer.
[in]	lwork	TODO
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Add contribution to C in fcblk: Get the first facing block of the distant panel, and the last block of the current cblk

Definition at line 1074 of file core_dgemmsp.c.

References core_dlrmm_s::A, core_dlrmm_s::alpha, core_dlrmm_s::B, core_dlrmm_s::beta, blok_rownbr(), core_dlrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_dlrmm_s::Cm, core_dlrmm_s::Cn, solver_blok_s::coefind, core_dlrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_dlrmm_s::K, solver_cblk_s::lock, core_dlrmm_s::lock, core_dlrmm_s::lowrank, core_dlrmm_s::lwork, core_dlrmm_s::M, core_dlrmm_s::N, core_dlrmm_s::offx, core_dlrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, solver_cblk_s::stride, core_dlrmm_s::transA, core_dlrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_dlrmm_s::work.

Referenced by cpucblk_dgemmsp().

◆ core_dgemmsp_lr()

static pastix_fixdbl_t core_dgemmsp_lr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		pastix_lrblock_t *	lrC,
		double *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in]	lrC	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in]	work	Temporary memory buffer.
[in]	lwork	TODO
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 1228 of file core_dgemmsp.c.

References core_dlrmm_s::A, core_dlrmm_s::alpha, core_dlrmm_s::B, core_dlrmm_s::beta, blok_rownbr(), core_dlrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_dlrmm_s::Cm, core_dlrmm_s::Cn, core_dlrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_dlrmm_s::K, solver_cblk_s::lock, core_dlrmm_s::lock, core_dlrmm_s::lowrank, core_dlrmm_s::lwork, core_dlrmm_s::lwused, core_dlrmm_s::M, core_dlrmm_s::N, core_dlrmm_s::offx, core_dlrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, core_dlrmm_s::transA, core_dlrmm_s::transB, and core_dlrmm_s::work.

Referenced by cpucblk_dgemmsp().

◆ core_dgemmsp_lrfr()

static pastix_fixdbl_t core_dgemmsp_lrfr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		double *	C,
		double *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.
[in]	work	Temporary memory buffer.
[in]	lwork	The length of work. On entry, if trans = PastixTrans lwork >= max(1, KN). Otherwise lwork >= max(1, MK).
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 1375 of file core_dgemmsp.c.

References core_dlrmm_s::A, core_dlrmm_s::alpha, core_dlrmm_s::B, core_dlrmm_s::beta, blok_rownbr(), core_dlrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_dlrmm_s::Cm, core_dlrmm_s::Cn, solver_blok_s::coefind, core_dlrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_dlrmm_s::K, solver_cblk_s::lock, core_dlrmm_s::lock, core_dlrmm_s::lowrank, core_dlrmm_s::lwork, core_dlrmm_s::lwused, core_dlrmm_s::M, core_dlrmm_s::N, core_dlrmm_s::offx, core_dlrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, core_dlrmm_s::transA, core_dlrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_dlrmm_s::work.

Referenced by cpucblk_dgemmsp().

◆ core_zgemmsp_1d1d()

static void core_zgemmsp_1d1d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_complex64_t *	A,
		const pastix_complex64_t *	B,
		pastix_complex64_t *	C,
		pastix_complex64_t *	work
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

Both cblk involved in the computation are stored with the 1D storage: Column Major Layout with blocks interleaved.

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.
[in]	work	Temporary memory buffer that is at least equal to the height of the block B by the sum of the height of all the blocks below the block B.

See also: core_zgemmsp_1d2d; core_zgemmsp_2d2d

Definition at line 106 of file core_zgemmsp.c.

References blok_rownbr(), cblk_colnbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, pastix_int_t, PastixUCoef, and solver_cblk_s::stride.

Referenced by cpucblk_zgemmsp().

◆ core_zgemmsp_1d2d()

static void core_zgemmsp_1d2d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_complex64_t *	A,
		const pastix_complex64_t *	B,
		pastix_complex64_t *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

See also: core_zgemmsp_1d1d; core_zgemmsp_2d2d

Add contribution to C in fcblk: Get the first facing block of the distant panel, and the last block of the current cblk

Definition at line 264 of file core_zgemmsp.c.

Referenced by cpucblk_zgemmsp().

◆ core_zgemmsp_2d2d()

static void core_zgemmsp_2d2d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_complex64_t *	A,
		const pastix_complex64_t *	B,
		pastix_complex64_t *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

Both cblk involved in the matrices A, B and C are stored with the 2D storage where each block is stored continuously one after another. (Similar to dense tile storage with variant tile size)

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

See also: core_zgemmsp_1d1d; core_zgemmsp_1d2d

Definition at line 404 of file core_zgemmsp.c.

Referenced by cpucblk_zgemmsp().

◆ core_zgemmsp_block_frfr()

static pastix_fixdbl_t core_zgemmsp_block_frfr	(	pastix_trans_t	trans,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const pastix_complex64_t *	A,
		const pastix_complex64_t *	B,
		pastix_complex64_t *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to performed an update:

                 C_mn = C_mn - A_mk * op(B_kn)

Parameters

[in]	trans	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixConjTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Returns: The number of flops performed

See also: core_zgemmsp_block_frlr; core_zgemmsp_block_lrlr

Definition at line 545 of file core_zgemmsp.c.

Referenced by cpublok_zgemmsp().

◆ core_zgemmsp_block_frlr()

static pastix_fixdbl_t core_zgemmsp_block_frlr	(	pastix_trans_t	transB,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const pastix_complex64_t *	A,
		const pastix_complex64_t *	B,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to perform an update:

                 C_mn = C_mn - A_mk * op(B_kn)

In this routine, all the matrices are low-rank

Parameters

[in]	transB	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixConjTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 712 of file core_zgemmsp.c.

References core_zlrmm_s::A, core_zlrmm_s::alpha, core_zlrmm_s::B, core_zlrmm_s::beta, blok_rownbr(), core_zlrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_zlrmm_s::Cm, core_zlrmm_s::Cn, solver_blok_s::coefind, core_zlrmm(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, solver_blok_s::frownum, solver_blok_s::inlast, is_block_inside_fblock(), core_zlrmm_s::K, kernel_trace_start(), kernel_trace_stop(), solver_blok_s::lcblknm, core_zlrmm_s::lock, core_zlrmm_s::lowrank, core_zlrmm_s::lwork, core_zlrmm_s::M, core_zlrmm_s::N, core_zlrmm_s::offx, core_zlrmm_s::offy, pastix_int_t, PastixKernelGEMMBlok2d2d, PastixKernelGEMMBlokLRLR, PastixNoTrans, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, core_zlrmm_s::transA, core_zlrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_zlrmm_s::work.

Referenced by cpublok_zgemmsp().

◆ core_zgemmsp_block_lrlr()

static pastix_fixdbl_t core_zgemmsp_block_lrlr	(	pastix_trans_t	transB,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to perform an update:

                 C_mn = C_mn - A_mk * op(B_kn)

In this routine, all the matrices are low-rank

Parameters

[in]	transB	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixConjTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in,out]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 901 of file core_zgemmsp.c.

References core_zlrmm_s::A, core_zlrmm_s::alpha, core_zlrmm_s::B, core_zlrmm_s::beta, blok_rownbr(), core_zlrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_zlrmm_s::Cm, core_zlrmm_s::Cn, core_zlrmm(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, solver_blok_s::frownum, solver_blok_s::inlast, is_block_inside_fblock(), core_zlrmm_s::K, kernel_trace_start(), kernel_trace_stop(), solver_blok_s::lcblknm, core_zlrmm_s::lock, core_zlrmm_s::lowrank, core_zlrmm_s::lwork, core_zlrmm_s::M, core_zlrmm_s::N, core_zlrmm_s::offx, core_zlrmm_s::offy, pastix_int_t, PastixKernelGEMMBlok2d2d, PastixKernelGEMMBlokLRLR, PastixNoTrans, core_zlrmm_s::transA, core_zlrmm_s::transB, and core_zlrmm_s::work.

Referenced by cpublok_zgemmsp().

◆ core_zgemmsp_fulllr()

static pastix_fixdbl_t core_zgemmsp_fulllr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_complex64_t *	A,
		const pastix_complex64_t *	B,
		pastix_lrblock_t *	lrC,
		pastix_complex64_t *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	work	Temporary memory buffer.
[in]	lwork	TODO
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Add contribution to C in fcblk: Get the first facing block of the distant panel, and the last block of the current cblk

Definition at line 1074 of file core_zgemmsp.c.

References core_zlrmm_s::A, core_zlrmm_s::alpha, core_zlrmm_s::B, core_zlrmm_s::beta, blok_rownbr(), core_zlrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_zlrmm_s::Cm, core_zlrmm_s::Cn, solver_blok_s::coefind, core_zlrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_zlrmm_s::K, solver_cblk_s::lock, core_zlrmm_s::lock, core_zlrmm_s::lowrank, core_zlrmm_s::lwork, core_zlrmm_s::M, core_zlrmm_s::N, core_zlrmm_s::offx, core_zlrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, solver_cblk_s::stride, core_zlrmm_s::transA, core_zlrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_zlrmm_s::work.

Referenced by cpucblk_zgemmsp().

◆ core_zgemmsp_lr()

static pastix_fixdbl_t core_zgemmsp_lr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		pastix_lrblock_t *	lrC,
		pastix_complex64_t *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in]	lrC	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in]	work	Temporary memory buffer.
[in]	lwork	TODO
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 1228 of file core_zgemmsp.c.

References core_zlrmm_s::A, core_zlrmm_s::alpha, core_zlrmm_s::B, core_zlrmm_s::beta, blok_rownbr(), core_zlrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_zlrmm_s::Cm, core_zlrmm_s::Cn, core_zlrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_zlrmm_s::K, solver_cblk_s::lock, core_zlrmm_s::lock, core_zlrmm_s::lowrank, core_zlrmm_s::lwork, core_zlrmm_s::lwused, core_zlrmm_s::M, core_zlrmm_s::N, core_zlrmm_s::offx, core_zlrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, core_zlrmm_s::transA, core_zlrmm_s::transB, and core_zlrmm_s::work.

Referenced by cpucblk_zgemmsp().

◆ core_zgemmsp_lrfr()

static pastix_fixdbl_t core_zgemmsp_lrfr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		pastix_complex64_t *	C,
		pastix_complex64_t *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.
[in]	work	Temporary memory buffer.
[in]	lwork	The length of work. On entry, if trans = PastixTrans lwork >= max(1, KN). Otherwise lwork >= max(1, MK).
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 1375 of file core_zgemmsp.c.

References core_zlrmm_s::A, core_zlrmm_s::alpha, core_zlrmm_s::B, core_zlrmm_s::beta, blok_rownbr(), core_zlrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_zlrmm_s::Cm, core_zlrmm_s::Cn, solver_blok_s::coefind, core_zlrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_zlrmm_s::K, solver_cblk_s::lock, core_zlrmm_s::lock, core_zlrmm_s::lowrank, core_zlrmm_s::lwork, core_zlrmm_s::lwused, core_zlrmm_s::M, core_zlrmm_s::N, core_zlrmm_s::offx, core_zlrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, core_zlrmm_s::transA, core_zlrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_zlrmm_s::work.

Referenced by cpucblk_zgemmsp().

◆ core_cgemmsp_1d1d()

static void core_cgemmsp_1d1d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_complex32_t *	A,
		const pastix_complex32_t *	B,
		pastix_complex32_t *	C,
		pastix_complex32_t *	work
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

Both cblk involved in the computation are stored with the 1D storage: Column Major Layout with blocks interleaved.

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.
[in]	work	Temporary memory buffer that is at least equal to the height of the block B by the sum of the height of all the blocks below the block B.

See also: core_cgemmsp_1d2d; core_cgemmsp_2d2d

Definition at line 106 of file core_cgemmsp.c.

References blok_rownbr(), cblk_colnbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, pastix_int_t, PastixUCoef, and solver_cblk_s::stride.

Referenced by cpucblk_cgemmsp().

◆ core_cgemmsp_1d2d()

static void core_cgemmsp_1d2d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_complex32_t *	A,
		const pastix_complex32_t *	B,
		pastix_complex32_t *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

See also: core_cgemmsp_1d1d; core_cgemmsp_2d2d

Add contribution to C in fcblk: Get the first facing block of the distant panel, and the last block of the current cblk

Definition at line 264 of file core_cgemmsp.c.

Referenced by cpucblk_cgemmsp().

◆ core_cgemmsp_2d2d()

static void core_cgemmsp_2d2d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_complex32_t *	A,
		const pastix_complex32_t *	B,
		pastix_complex32_t *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

Both cblk involved in the matrices A, B and C are stored with the 2D storage where each block is stored continuously one after another. (Similar to dense tile storage with variant tile size)

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

See also: core_cgemmsp_1d1d; core_cgemmsp_1d2d

Definition at line 404 of file core_cgemmsp.c.

Referenced by cpucblk_cgemmsp().

◆ core_cgemmsp_block_frfr()

static pastix_fixdbl_t core_cgemmsp_block_frfr	(	pastix_trans_t	trans,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const pastix_complex32_t *	A,
		const pastix_complex32_t *	B,
		pastix_complex32_t *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to performed an update:

                 C_mn = C_mn - A_mk * op(B_kn)

Parameters

[in]	trans	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixConjTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Returns: The number of flops performed

See also: core_cgemmsp_block_frlr; core_cgemmsp_block_lrlr

Definition at line 545 of file core_cgemmsp.c.

Referenced by cpublok_cgemmsp().

◆ core_cgemmsp_block_frlr()

static pastix_fixdbl_t core_cgemmsp_block_frlr	(	pastix_trans_t	transB,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const pastix_complex32_t *	A,
		const pastix_complex32_t *	B,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to perform an update:

                 C_mn = C_mn - A_mk * op(B_kn)

In this routine, all the matrices are low-rank

Parameters

[in]	transB	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixConjTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 712 of file core_cgemmsp.c.

References core_clrmm_s::A, core_clrmm_s::alpha, core_clrmm_s::B, core_clrmm_s::beta, blok_rownbr(), core_clrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_clrmm_s::Cm, core_clrmm_s::Cn, solver_blok_s::coefind, core_clrmm(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, solver_blok_s::frownum, solver_blok_s::inlast, is_block_inside_fblock(), core_clrmm_s::K, kernel_trace_start(), kernel_trace_stop(), solver_blok_s::lcblknm, core_clrmm_s::lock, core_clrmm_s::lowrank, core_clrmm_s::lwork, core_clrmm_s::M, core_clrmm_s::N, core_clrmm_s::offx, core_clrmm_s::offy, pastix_int_t, PastixKernelGEMMBlok2d2d, PastixKernelGEMMBlokLRLR, PastixNoTrans, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, core_clrmm_s::transA, core_clrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_clrmm_s::work.

Referenced by cpublok_cgemmsp().

◆ core_cgemmsp_block_lrlr()

static pastix_fixdbl_t core_cgemmsp_block_lrlr	(	pastix_trans_t	transB,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to perform an update:

                 C_mn = C_mn - A_mk * op(B_kn)

In this routine, all the matrices are low-rank

Parameters

[in]	transB	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixConjTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in,out]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 901 of file core_cgemmsp.c.

References core_clrmm_s::A, core_clrmm_s::alpha, core_clrmm_s::B, core_clrmm_s::beta, blok_rownbr(), core_clrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_clrmm_s::Cm, core_clrmm_s::Cn, core_clrmm(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, solver_blok_s::frownum, solver_blok_s::inlast, is_block_inside_fblock(), core_clrmm_s::K, kernel_trace_start(), kernel_trace_stop(), solver_blok_s::lcblknm, core_clrmm_s::lock, core_clrmm_s::lowrank, core_clrmm_s::lwork, core_clrmm_s::M, core_clrmm_s::N, core_clrmm_s::offx, core_clrmm_s::offy, pastix_int_t, PastixKernelGEMMBlok2d2d, PastixKernelGEMMBlokLRLR, PastixNoTrans, core_clrmm_s::transA, core_clrmm_s::transB, and core_clrmm_s::work.

Referenced by cpublok_cgemmsp().

◆ core_cgemmsp_fulllr()

static pastix_fixdbl_t core_cgemmsp_fulllr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_complex32_t *	A,
		const pastix_complex32_t *	B,
		pastix_lrblock_t *	lrC,
		pastix_complex32_t *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Hermitian cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	work	Temporary memory buffer.
[in]	lwork	TODO
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Add contribution to C in fcblk: Get the first facing block of the distant panel, and the last block of the current cblk

Definition at line 1074 of file core_cgemmsp.c.

References core_clrmm_s::A, core_clrmm_s::alpha, core_clrmm_s::B, core_clrmm_s::beta, blok_rownbr(), core_clrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_clrmm_s::Cm, core_clrmm_s::Cn, solver_blok_s::coefind, core_clrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_clrmm_s::K, solver_cblk_s::lock, core_clrmm_s::lock, core_clrmm_s::lowrank, core_clrmm_s::lwork, core_clrmm_s::M, core_clrmm_s::N, core_clrmm_s::offx, core_clrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, solver_cblk_s::stride, core_clrmm_s::transA, core_clrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_clrmm_s::work.

Referenced by cpucblk_cgemmsp().

◆ core_cgemmsp_lr()

static pastix_fixdbl_t core_cgemmsp_lr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		pastix_lrblock_t *	lrC,
		pastix_complex32_t *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in]	lrC	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in]	work	Temporary memory buffer.
[in]	lwork	TODO
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 1228 of file core_cgemmsp.c.

References core_clrmm_s::A, core_clrmm_s::alpha, core_clrmm_s::B, core_clrmm_s::beta, blok_rownbr(), core_clrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_clrmm_s::Cm, core_clrmm_s::Cn, core_clrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_clrmm_s::K, solver_cblk_s::lock, core_clrmm_s::lock, core_clrmm_s::lowrank, core_clrmm_s::lwork, core_clrmm_s::lwused, core_clrmm_s::M, core_clrmm_s::N, core_clrmm_s::offx, core_clrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, core_clrmm_s::transA, core_clrmm_s::transB, and core_clrmm_s::work.

Referenced by cpucblk_cgemmsp().

◆ core_cgemmsp_lrfr()

static pastix_fixdbl_t core_cgemmsp_lrfr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		pastix_complex32_t *	C,
		pastix_complex32_t *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.
[in]	work	Temporary memory buffer.
[in]	lwork	The length of work. On entry, if trans = PastixTrans lwork >= max(1, KN). Otherwise lwork >= max(1, MK).
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 1375 of file core_cgemmsp.c.

References core_clrmm_s::A, core_clrmm_s::alpha, core_clrmm_s::B, core_clrmm_s::beta, blok_rownbr(), core_clrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_clrmm_s::Cm, core_clrmm_s::Cn, solver_blok_s::coefind, core_clrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_clrmm_s::K, solver_cblk_s::lock, core_clrmm_s::lock, core_clrmm_s::lowrank, core_clrmm_s::lwork, core_clrmm_s::lwused, core_clrmm_s::M, core_clrmm_s::N, core_clrmm_s::offx, core_clrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, core_clrmm_s::transA, core_clrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_clrmm_s::work.

Referenced by cpucblk_cgemmsp().

◆ core_sgemmsp_1d1d()

static void core_sgemmsp_1d1d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const float *	A,
		const float *	B,
		float *	C,
		float *	work
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

Both cblk involved in the computation are stored with the 1D storage: Column Major Layout with blocks interleaved.

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.
[in]	work	Temporary memory buffer that is at least equal to the height of the block B by the sum of the height of all the blocks below the block B.

See also: core_sgemmsp_1d2d; core_sgemmsp_2d2d

Definition at line 106 of file core_sgemmsp.c.

References blok_rownbr(), cblk_colnbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, pastix_int_t, PastixUCoef, and solver_cblk_s::stride.

Referenced by cpucblk_sgemmsp().

◆ core_sgemmsp_1d2d()

static void core_sgemmsp_1d2d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const float *	A,
		const float *	B,
		float *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

See also: core_sgemmsp_1d1d; core_sgemmsp_2d2d

Add contribution to C in fcblk: Get the first facing block of the distant panel, and the last block of the current cblk

Definition at line 264 of file core_sgemmsp.c.

Referenced by cpucblk_sgemmsp().

◆ core_sgemmsp_2d2d()

static void core_sgemmsp_2d2d	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const float *	A,
		const float *	B,
		float *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of one single off-diagonal block.

Both cblk involved in the matrices A, B and C are stored with the 2D storage where each block is stored continuously one after another. (Similar to dense tile storage with variant tile size)

All the off-diagonal block below block are multiplied by the selected block and added to the facing cblk.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C. The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

See also: core_sgemmsp_1d1d; core_sgemmsp_1d2d

Definition at line 404 of file core_sgemmsp.c.

Referenced by cpucblk_sgemmsp().

◆ core_sgemmsp_block_frfr()

static pastix_fixdbl_t core_sgemmsp_block_frfr	(	pastix_trans_t	trans,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const float *	A,
		const float *	B,
		float *	C
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to performed an update:

                 C_mn = C_mn - A_mk * op(B_kn)

Parameters

[in]	trans	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Returns: The number of flops performed

See also: core_sgemmsp_block_frlr; core_sgemmsp_block_lrlr

Definition at line 545 of file core_sgemmsp.c.

Referenced by cpublok_sgemmsp().

◆ core_sgemmsp_block_frlr()

static pastix_fixdbl_t core_sgemmsp_block_frlr	(	pastix_trans_t	transB,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const float *	A,
		const float *	B,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to perform an update:

                 C_mn = C_mn - A_mk * op(B_kn)

In this routine, all the matrices are low-rank

Parameters

[in]	transB	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 712 of file core_sgemmsp.c.

References core_slrmm_s::A, core_slrmm_s::alpha, core_slrmm_s::B, core_slrmm_s::beta, blok_rownbr(), core_slrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_slrmm_s::Cm, core_slrmm_s::Cn, solver_blok_s::coefind, core_slrmm(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, solver_blok_s::frownum, solver_blok_s::inlast, is_block_inside_fblock(), core_slrmm_s::K, kernel_trace_start(), kernel_trace_stop(), solver_blok_s::lcblknm, core_slrmm_s::lock, core_slrmm_s::lowrank, core_slrmm_s::lwork, core_slrmm_s::M, core_slrmm_s::N, core_slrmm_s::offx, core_slrmm_s::offy, pastix_int_t, PastixKernelGEMMBlok2d2d, PastixKernelGEMMBlokLRLR, PastixNoTrans, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, core_slrmm_s::transA, core_slrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_slrmm_s::work.

Referenced by cpublok_sgemmsp().

◆ core_sgemmsp_block_lrlr()

static pastix_fixdbl_t core_sgemmsp_block_lrlr	(	pastix_trans_t	transB,
		pastix_int_t	blok_mk,
		pastix_int_t	blok_kn,
		pastix_int_t	blok_mn,
		const SolverCblk *	cblk,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates that are generated by the transposition of all the blocks facing a common diagonal block, by another similar set of blocks.

This is used to perform an update:

                 C_mn = C_mn - A_mk * op(B_kn)

In this routine, all the matrices are low-rank

Parameters

[in]	transB	Specify the transposition used for the B matrices. It has to be either PastixTrans or PastixTrans.
[in]	blok_mk	Index of the first off-diagonal block in cblk, that is used for A.
[in]	blok_kn	Index of the first off-diagonal block in cblk, that is used for B.
[in]	blok_mn	Index of the first off-diagonal block in fcblk, that is used for C.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in,out]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 901 of file core_sgemmsp.c.

References core_slrmm_s::A, core_slrmm_s::alpha, core_slrmm_s::B, core_slrmm_s::beta, blok_rownbr(), core_slrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_slrmm_s::Cm, core_slrmm_s::Cn, core_slrmm(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, solver_blok_s::frownum, solver_blok_s::inlast, is_block_inside_fblock(), core_slrmm_s::K, kernel_trace_start(), kernel_trace_stop(), solver_blok_s::lcblknm, core_slrmm_s::lock, core_slrmm_s::lowrank, core_slrmm_s::lwork, core_slrmm_s::M, core_slrmm_s::N, core_slrmm_s::offx, core_slrmm_s::offy, pastix_int_t, PastixKernelGEMMBlok2d2d, PastixKernelGEMMBlokLRLR, PastixNoTrans, core_slrmm_s::transA, core_slrmm_s::transB, and core_slrmm_s::work.

Referenced by cpublok_sgemmsp().

◆ core_sgemmsp_fulllr()

static pastix_fixdbl_t core_sgemmsp_fulllr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const float *	A,
		const float *	B,
		pastix_lrblock_t *	lrC,
		float *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	B	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel, if Symmetric/Symmetric cases or if upper part is computed; cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	work	Temporary memory buffer.
[in]	lwork	TODO
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Add contribution to C in fcblk: Get the first facing block of the distant panel, and the last block of the current cblk

Definition at line 1074 of file core_sgemmsp.c.

References core_slrmm_s::A, core_slrmm_s::alpha, core_slrmm_s::B, core_slrmm_s::beta, blok_rownbr(), core_slrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_slrmm_s::Cm, core_slrmm_s::Cn, solver_blok_s::coefind, core_slrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_slrmm_s::K, solver_cblk_s::lock, core_slrmm_s::lock, core_slrmm_s::lowrank, core_slrmm_s::lwork, core_slrmm_s::M, core_slrmm_s::N, core_slrmm_s::offx, core_slrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, solver_cblk_s::stride, core_slrmm_s::transA, core_slrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_slrmm_s::work.

Referenced by cpucblk_sgemmsp().

◆ core_sgemmsp_lr()

static pastix_fixdbl_t core_sgemmsp_lr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		pastix_lrblock_t *	lrC,
		float *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in]	lrC	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in]	work	Temporary memory buffer.
[in]	lwork	TODO
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 1228 of file core_sgemmsp.c.

References core_slrmm_s::A, core_slrmm_s::alpha, core_slrmm_s::B, core_slrmm_s::beta, blok_rownbr(), core_slrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_slrmm_s::Cm, core_slrmm_s::Cn, core_slrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_slrmm_s::K, solver_cblk_s::lock, core_slrmm_s::lock, core_slrmm_s::lowrank, core_slrmm_s::lwork, core_slrmm_s::lwused, core_slrmm_s::M, core_slrmm_s::N, core_slrmm_s::offx, core_slrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, core_slrmm_s::transA, core_slrmm_s::transB, and core_slrmm_s::work.

Referenced by cpucblk_sgemmsp().

◆ core_sgemmsp_lrfr()

static pastix_fixdbl_t core_sgemmsp_lrfr	(	pastix_coefside_t	sideA,
		pastix_trans_t	trans,
		const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_lrblock_t *	lrA,
		const pastix_lrblock_t *	lrB,
		float *	C,
		float *	work,
		pastix_int_t	lwork,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block.

Parameters

[in]	sideA	Specify if A and C belong to the lower part, or to the upper part. If sideA == PastixLCoef, the contribution of: (block .. (cblk[1].fblokptr-1)) -by- block is computed and added to C, otherwise the contribution: (block+1 .. (cblk[1].fblokptr-1)) -by- block is computed and added to C.
[in]	trans	Specify the transposition used for the B matrix. It has to be either PastixTrans or PastixTrans.
[in]	cblk	The cblk structure to which block belongs to. The A and B pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok	The block from which we compute the contributions.
[in,out]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. The C pointer must be one of the coeftab from this fcblk. Next column blok must be accessible through fcblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrB	Pointer to the low-rank representation of the block B. Must be followed by the low-rank representation of the following blocks.
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.
[in]	work	Temporary memory buffer.
[in]	lwork	The length of work. On entry, if trans = PastixTrans lwork >= max(1, KN). Otherwise lwork >= max(1, MK).
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 1375 of file core_sgemmsp.c.

References core_slrmm_s::A, core_slrmm_s::alpha, core_slrmm_s::B, core_slrmm_s::beta, blok_rownbr(), core_slrmm_s::C, cblk_colnbr(), solver_cblk_s::cblktype, core_slrmm_s::Cm, core_slrmm_s::Cn, solver_blok_s::coefind, core_slrmm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), core_slrmm_s::K, solver_cblk_s::lock, core_slrmm_s::lock, core_slrmm_s::lowrank, core_slrmm_s::lwork, core_slrmm_s::lwused, core_slrmm_s::M, core_slrmm_s::N, core_slrmm_s::offx, core_slrmm_s::offy, pastix_int_t, PastixNoTrans, PastixUCoef, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, core_slrmm_s::transA, core_slrmm_s::transB, pastix_lrblock_s::u, pastix_lrblock_s::v, and core_slrmm_s::work.

Referenced by cpucblk_sgemmsp().

◆ core_dtrsmsp_1d()

static void core_dtrsmsp_1d	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const double *	A,
		double *	C
	)

inlinestatic

Apply all the trsm updates on a panel stored in 1D layout.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Definition at line 72 of file core_dtrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_cblk_s::lcolnum, pastix_int_t, and solver_cblk_s::stride.

◆ core_dtrsmsp_2d()

static void core_dtrsmsp_2d	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const double *	A,
		double *	C
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in 2D.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Definition at line 149 of file core_dtrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_cblk_s::lcolnum, and pastix_int_t.

◆ core_dtrsmsp_lr()

static pastix_fixdbl_t core_dtrsmsp_lr	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const pastix_lrblock_t *	lrA,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Computes the updates associated to one off-diagonal block between two cblk stored in low-rank format.

Parameters

[in]	side	Specify whether the off-diagonal blocks appear on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the off-diagonal blocks are upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the off-diagonal blocks. It has to be either PastixTrans or PastixTrans.
[in]	diag	Specify if the off-diagonal blocks are unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 233 of file core_dtrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, pastix_lr_s::compress_min_height, pastix_lr_s::compress_min_width, pastix_lr_s::compress_preselect, cpublok_dcompress(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, pastix_lr_s::ilu_lvl, solver_blok_s::iluklvl, solver_cblk_s::lcolnum, pastix_int_t, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, and pastix_lrblock_s::u.

◆ core_dtrsmsp_2dsub()

static pastix_fixdbl_t core_dtrsmsp_2dsub	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		pastix_int_t	blok_m,
		const double *	A,
		double *	C
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in 2D.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok_m	Index of the first off-diagonal block in cblk that is solved. The TRSM is also applied to all the folowing blocks which are facing the same diagonal block
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Return values

TODO

Definition at line 452 of file core_dtrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, solver_blok_s::inlast, kernel_trace_start(), kernel_trace_stop(), solver_cblk_s::lcolnum, pastix_int_t, and PastixKernelTRSMBlok2d.

Referenced by cpublok_dtrsmsp().

◆ core_dtrsmsp_lrsub()

static pastix_fixdbl_t core_dtrsmsp_lrsub	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		pastix_int_t	blok_m,
		const pastix_lrblock_t *	lrA,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in low-rank format.

Parameters

[in]	side	Specify whether the off-diagonal blocks appear on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the off-diagonal blocks are upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the off-diagonal blocks. It has to be either PastixTrans or PastixTrans.
[in]	diag	Specify if the off-diagonal blocks are unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok_m	Index of the first off-diagonal block in cblk that is solved. The TRSM is also applied to all the folowing blocks which are facing the same diagonal block
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Return values

TODO

Definition at line 554 of file core_dtrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, pastix_lr_s::compress_min_height, pastix_lr_s::compress_min_width, pastix_lr_s::compress_preselect, cpublok_dcompress(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, pastix_lr_s::ilu_lvl, solver_blok_s::iluklvl, solver_blok_s::inlast, kernel_trace_start(), kernel_trace_stop(), solver_cblk_s::lcolnum, pastix_int_t, PastixKernelTRSMBlokLR, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, pastix_lrblock_s::u, and pastix_lrblock_s::v.

Referenced by cpublok_dtrsmsp().

◆ core_ztrsmsp_1d()

static void core_ztrsmsp_1d	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const pastix_complex64_t *	A,
		pastix_complex64_t *	C
	)

inlinestatic

Apply all the trsm updates on a panel stored in 1D layout.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Definition at line 72 of file core_ztrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_cblk_s::lcolnum, pastix_int_t, and solver_cblk_s::stride.

◆ core_ztrsmsp_2d()

static void core_ztrsmsp_2d	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const pastix_complex64_t *	A,
		pastix_complex64_t *	C
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in 2D.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Definition at line 149 of file core_ztrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_cblk_s::lcolnum, and pastix_int_t.

◆ core_ztrsmsp_lr()

static pastix_fixdbl_t core_ztrsmsp_lr	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const pastix_lrblock_t *	lrA,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Computes the updates associated to one off-diagonal block between two cblk stored in low-rank format.

Parameters

[in]	side	Specify whether the off-diagonal blocks appear on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the off-diagonal blocks are upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the off-diagonal blocks. It has to be either PastixTrans or PastixConjTrans.
[in]	diag	Specify if the off-diagonal blocks are unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 233 of file core_ztrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, pastix_lr_s::compress_min_height, pastix_lr_s::compress_min_width, pastix_lr_s::compress_preselect, cpublok_zcompress(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, pastix_lr_s::ilu_lvl, solver_blok_s::iluklvl, solver_cblk_s::lcolnum, pastix_int_t, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, and pastix_lrblock_s::u.

◆ core_ztrsmsp_2dsub()

static pastix_fixdbl_t core_ztrsmsp_2dsub	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		pastix_int_t	blok_m,
		const pastix_complex64_t *	A,
		pastix_complex64_t *	C
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in 2D.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok_m	Index of the first off-diagonal block in cblk that is solved. The TRSM is also applied to all the folowing blocks which are facing the same diagonal block
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Return values

TODO

Definition at line 452 of file core_ztrsmsp.c.

Referenced by cpublok_ztrsmsp().

◆ core_ztrsmsp_lrsub()

static pastix_fixdbl_t core_ztrsmsp_lrsub	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		pastix_int_t	blok_m,
		const pastix_lrblock_t *	lrA,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in low-rank format.

Parameters

[in]	side	Specify whether the off-diagonal blocks appear on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the off-diagonal blocks are upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the off-diagonal blocks. It has to be either PastixTrans or PastixConjTrans.
[in]	diag	Specify if the off-diagonal blocks are unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok_m	Index of the first off-diagonal block in cblk that is solved. The TRSM is also applied to all the folowing blocks which are facing the same diagonal block
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Return values

TODO

Definition at line 554 of file core_ztrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, pastix_lr_s::compress_min_height, pastix_lr_s::compress_min_width, pastix_lr_s::compress_preselect, cpublok_zcompress(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, pastix_lr_s::ilu_lvl, solver_blok_s::iluklvl, solver_blok_s::inlast, kernel_trace_start(), kernel_trace_stop(), solver_cblk_s::lcolnum, pastix_int_t, PastixKernelTRSMBlokLR, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, pastix_lrblock_s::u, and pastix_lrblock_s::v.

Referenced by cpublok_ztrsmsp().

◆ core_ctrsmsp_1d()

static void core_ctrsmsp_1d	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const pastix_complex32_t *	A,
		pastix_complex32_t *	C
	)

inlinestatic

Apply all the trsm updates on a panel stored in 1D layout.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Definition at line 72 of file core_ctrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_cblk_s::lcolnum, pastix_int_t, and solver_cblk_s::stride.

◆ core_ctrsmsp_2d()

static void core_ctrsmsp_2d	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const pastix_complex32_t *	A,
		pastix_complex32_t *	C
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in 2D.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Definition at line 149 of file core_ctrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_cblk_s::lcolnum, and pastix_int_t.

◆ core_ctrsmsp_lr()

static pastix_fixdbl_t core_ctrsmsp_lr	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const pastix_lrblock_t *	lrA,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Computes the updates associated to one off-diagonal block between two cblk stored in low-rank format.

Parameters

[in]	side	Specify whether the off-diagonal blocks appear on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the off-diagonal blocks are upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the off-diagonal blocks. It has to be either PastixTrans or PastixConjTrans.
[in]	diag	Specify if the off-diagonal blocks are unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 233 of file core_ctrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, pastix_lr_s::compress_min_height, pastix_lr_s::compress_min_width, pastix_lr_s::compress_preselect, cpublok_ccompress(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, pastix_lr_s::ilu_lvl, solver_blok_s::iluklvl, solver_cblk_s::lcolnum, pastix_int_t, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, and pastix_lrblock_s::u.

◆ core_ctrsmsp_2dsub()

static pastix_fixdbl_t core_ctrsmsp_2dsub	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		pastix_int_t	blok_m,
		const pastix_complex32_t *	A,
		pastix_complex32_t *	C
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in 2D.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixConjTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok_m	Index of the first off-diagonal block in cblk that is solved. The TRSM is also applied to all the folowing blocks which are facing the same diagonal block
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Return values

TODO

Definition at line 452 of file core_ctrsmsp.c.

Referenced by cpublok_ctrsmsp().

◆ core_ctrsmsp_lrsub()

static pastix_fixdbl_t core_ctrsmsp_lrsub	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		pastix_int_t	blok_m,
		const pastix_lrblock_t *	lrA,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in low-rank format.

Parameters

[in]	side	Specify whether the off-diagonal blocks appear on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the off-diagonal blocks are upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the off-diagonal blocks. It has to be either PastixTrans or PastixConjTrans.
[in]	diag	Specify if the off-diagonal blocks are unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok_m	Index of the first off-diagonal block in cblk that is solved. The TRSM is also applied to all the folowing blocks which are facing the same diagonal block
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Return values

TODO

Definition at line 554 of file core_ctrsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, pastix_lr_s::compress_min_height, pastix_lr_s::compress_min_width, pastix_lr_s::compress_preselect, cpublok_ccompress(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, pastix_lr_s::ilu_lvl, solver_blok_s::iluklvl, solver_blok_s::inlast, kernel_trace_start(), kernel_trace_stop(), solver_cblk_s::lcolnum, pastix_int_t, PastixKernelTRSMBlokLR, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, pastix_lrblock_s::u, and pastix_lrblock_s::v.

Referenced by cpublok_ctrsmsp().

◆ core_strsmsp_1d()

static void core_strsmsp_1d	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const float *	A,
		float *	C
	)

inlinestatic

Apply all the trsm updates on a panel stored in 1D layout.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Definition at line 72 of file core_strsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_cblk_s::lcolnum, pastix_int_t, and solver_cblk_s::stride.

◆ core_strsmsp_2d()

static void core_strsmsp_2d	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const float *	A,
		float *	C
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in 2D.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Definition at line 149 of file core_strsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_cblk_s::lcolnum, and pastix_int_t.

◆ core_strsmsp_lr()

static pastix_fixdbl_t core_strsmsp_lr	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		const pastix_lrblock_t *	lrA,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Computes the updates associated to one off-diagonal block between two cblk stored in low-rank format.

Parameters

[in]	side	Specify whether the off-diagonal blocks appear on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the off-diagonal blocks are upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the off-diagonal blocks. It has to be either PastixTrans or PastixTrans.
[in]	diag	Specify if the off-diagonal blocks are unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Returns: The number of flops performed

Definition at line 233 of file core_strsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, pastix_lr_s::compress_min_height, pastix_lr_s::compress_min_width, pastix_lr_s::compress_preselect, cpublok_scompress(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, pastix_lr_s::ilu_lvl, solver_blok_s::iluklvl, solver_cblk_s::lcolnum, pastix_int_t, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, and pastix_lrblock_s::u.

◆ core_strsmsp_2dsub()

static pastix_fixdbl_t core_strsmsp_2dsub	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		pastix_int_t	blok_m,
		const float *	A,
		float *	C
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in 2D.

Parameters

[in]	side	Specify whether the A matrix appears on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the A matrix is upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the A matrix. It has to be either PastixTrans or PastixTrans.
[in]	diag	Specify if the A matrix is unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok_m	Index of the first off-diagonal block in cblk that is solved. The TRSM is also applied to all the folowing blocks which are facing the same diagonal block
[in]	A	The pointer to the coeftab of the cblk.lcoeftab matrix storing the coefficients of the panel when the Lower part is computed, cblk.ucoeftab otherwise. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the fcblk.lcoeftab if the lower part is computed, fcblk.ucoeftab otherwise.

Return values

TODO

Definition at line 452 of file core_strsmsp.c.

Referenced by cpublok_strsmsp().

◆ core_strsmsp_lrsub()

static pastix_fixdbl_t core_strsmsp_lrsub	(	pastix_side_t	side,
		pastix_uplo_t	uplo,
		pastix_trans_t	trans,
		pastix_diag_t	diag,
		const SolverCblk *	cblk,
		pastix_int_t	blok_m,
		const pastix_lrblock_t *	lrA,
		pastix_lrblock_t *	lrC,
		const pastix_lr_t *	lowrank
	)

inlinestatic

Compute the updates associated to one off-diagonal block between two cblk stored in low-rank format.

Parameters

[in]	side	Specify whether the off-diagonal blocks appear on the left or right in the equation. It has to be either PastixLeft or PastixRight.
[in]	uplo	Specify whether the off-diagonal blocks are upper or lower triangular. It has to be either PastixUpper or PastixLower.
[in]	trans	Specify the transposition used for the off-diagonal blocks. It has to be either PastixTrans or PastixTrans.
[in]	diag	Specify if the off-diagonal blocks are unit triangular. It has to be either PastixUnit or PastixNonUnit.
[in]	cblk	The cblk structure to which block belongs to. The A and C pointers must be the coeftab of this column block. Next column blok must be accessible through cblk[1].
[in]	blok_m	Index of the first off-diagonal block in cblk that is solved. The TRSM is also applied to all the folowing blocks which are facing the same diagonal block
[in]	lrA	Pointer to the low-rank representation of the block A. Must be followed by the low-rank representation of the following blocks.
[in,out]	lrC	Pointer to the low-rank representation of the block C. Must be followed by the low-rank representation of the following blocks.
[in]	lowrank	The structure with low-rank parameters.

Return values

TODO

Definition at line 554 of file core_strsmsp.c.

References blok_rownbr(), solver_cblk_s::cblktype, pastix_lr_s::compress_min_height, pastix_lr_s::compress_min_width, pastix_lr_s::compress_preselect, cpublok_scompress(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_cblk_s::fcolnum, pastix_lr_s::ilu_lvl, solver_blok_s::iluklvl, solver_blok_s::inlast, kernel_trace_start(), kernel_trace_stop(), solver_cblk_s::lcolnum, pastix_int_t, PastixKernelTRSMBlokLR, pastix_lrblock_s::rk, pastix_lrblock_s::rkmax, pastix_lrblock_s::u, and pastix_lrblock_s::v.

Referenced by cpublok_strsmsp().

◆ core_csytrfsp1d_gemm()

void core_csytrfsp1d_gemm	(	const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_complex32_t *	L,
		pastix_complex32_t *	C,
		pastix_complex32_t *	work
	)

core_csytrfsp1d_gemm - Computes the LDL^t factorization of one panel and apply all the trsm updates to this panel.

Parameters

[in]	cblk	The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	blok	The pointer to the data structure that describes the blok from which we compute the contributions.
[in]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. Next column blok must be accessible through fcblk[1].
[in,out]	L	The pointer to the matrix storing the coefficients of the panel. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the matrix storing the coefficients of the target.
[in,out]	work	Temporary buffer used in core_cgemdm().

Definition at line 316 of file core_csytrfsp.c.

References blok_rownbr(), cblk_colnbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, core_cgemdm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), pastix_int_t, PASTIX_SUCCESS, PastixConjTrans, PastixNoTrans, and solver_cblk_s::stride.

Referenced by cpucblk_csytrfsp1d(), and cpucblk_csytrfsp1dplus_update().

◆ cpucblk_cpotrfsp1dplus()

int cpucblk_cpotrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the Cholesky factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 391 of file core_cpotrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_cpotrfsp1d_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_cpotrf_dynamic().

◆ cpucblk_cpotrfsp1dplus_update()

void cpucblk_cpotrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		pastix_complex32_t *	work,
		pastix_int_t	lwork
	)

Apply the updates of the cholesky 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 444 of file core_cpotrfsp.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, PastixConjTrans, and PastixLCoef.

Referenced by thread_cpotrf_dynamic().

◆ cpucblk_csytrfsp1dplus()

int cpucblk_csytrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the LDL^t factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 566 of file core_csytrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_csytrfsp1d_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_csytrf_dynamic().

◆ cpucblk_csytrfsp1dplus_update()

void cpucblk_csytrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		pastix_complex32_t *	work
	)

Apply the updates of the LDL^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 619 of file core_csytrfsp.c.

References cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, core_csytrfsp1d_gemm(), cpucblk_calloc(), cpucblk_crelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_blok_s::lcblknm, solver_cblk_s::lcoeftab, and PastixLCoef.

Referenced by thread_csytrf_dynamic().

◆ cpucblk_cgetrfsp1dplus()

int cpucblk_cgetrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the LU factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 421 of file core_cgetrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), cblk_getdataU(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_cgetrfsp1d_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_cgetrf_dynamic().

◆ cpucblk_cgetrfsp1dplus_update()

void cpucblk_cgetrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		pastix_complex32_t *	work,
		pastix_int_t	lwork
	)

Apply the updates of the LU 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 475 of file core_cgetrfsp.c.

References cblk_getdataL(), cblk_getdataU(), 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, PastixLUCoef, PastixTrans, and PastixUCoef.

Referenced by thread_cgetrf_dynamic().

◆ core_ssytrfsp1d_gemm()

void core_ssytrfsp1d_gemm	(	const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const float *	L,
		float *	C,
		float *	work
	)

core_ssytrfsp1d_gemm - Computes the LDL^t factorization of one panel and apply all the trsm updates to this panel.

Parameters

[in]	cblk	The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	blok	The pointer to the data structure that describes the blok from which we compute the contributions.
[in]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. Next column blok must be accessible through fcblk[1].
[in,out]	L	The pointer to the matrix storing the coefficients of the panel. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the matrix storing the coefficients of the target.
[in,out]	work	Temporary buffer used in core_sgemdm().

Definition at line 316 of file core_ssytrfsp.c.

References blok_rownbr(), cblk_colnbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, core_sgemdm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), pastix_int_t, PASTIX_SUCCESS, PastixNoTrans, PastixTrans, and solver_cblk_s::stride.

Referenced by cpucblk_ssytrfsp1d(), and cpucblk_ssytrfsp1dplus_update().

◆ cpucblk_spotrfsp1dplus()

int cpucblk_spotrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the Cholesky factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 391 of file core_spotrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_spotrfsp1d_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_spotrf_dynamic().

◆ cpucblk_spotrfsp1dplus_update()

void cpucblk_spotrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		float *	work,
		pastix_int_t	lwork
	)

Apply the updates of the cholesky 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 444 of file core_spotrfsp.c.

References cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_salloc(), cpucblk_sgemmsp(), cpucblk_srelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_blok_s::lcblknm, solver_matrix_s::lowrank, PastixLCoef, and PastixTrans.

Referenced by thread_spotrf_dynamic().

◆ cpucblk_ssytrfsp1dplus()

int cpucblk_ssytrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the LDL^t factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 566 of file core_ssytrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_ssytrfsp1d_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_ssytrf_dynamic().

◆ cpucblk_ssytrfsp1dplus_update()

void cpucblk_ssytrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		float *	work
	)

Apply the updates of the LDL^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 619 of file core_ssytrfsp.c.

References cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, core_ssytrfsp1d_gemm(), cpucblk_salloc(), cpucblk_srelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_blok_s::lcblknm, solver_cblk_s::lcoeftab, and PastixLCoef.

Referenced by thread_ssytrf_dynamic().

◆ cpucblk_sgetrfsp1dplus()

int cpucblk_sgetrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the LU factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 421 of file core_sgetrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), cblk_getdataU(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_sgetrfsp1d_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_sgetrf_dynamic().

◆ cpucblk_sgetrfsp1dplus_update()

void cpucblk_sgetrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		float *	work,
		pastix_int_t	lwork
	)

Apply the updates of the LU 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 475 of file core_sgetrfsp.c.

References cblk_getdataL(), cblk_getdataU(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_salloc(), cpucblk_sgemmsp(), cpucblk_srelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_blok_s::lcblknm, solver_matrix_s::lowrank, PastixLCoef, PastixLUCoef, PastixTrans, and PastixUCoef.

Referenced by thread_sgetrf_dynamic().

◆ core_zsytrfsp1d_gemm()

void core_zsytrfsp1d_gemm	(	const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const pastix_complex64_t *	L,
		pastix_complex64_t *	C,
		pastix_complex64_t *	work
	)

core_zsytrfsp1d_gemm - Computes the LDL^t factorization of one panel and apply all the trsm updates to this panel.

Parameters

[in]	cblk	The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	blok	The pointer to the data structure that describes the blok from which we compute the contributions.
[in]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. Next column blok must be accessible through fcblk[1].
[in,out]	L	The pointer to the matrix storing the coefficients of the panel. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the matrix storing the coefficients of the target.
[in,out]	work	Temporary buffer used in core_zgemdm().

Definition at line 316 of file core_zsytrfsp.c.

References blok_rownbr(), cblk_colnbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, core_zgemdm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), pastix_int_t, PASTIX_SUCCESS, PastixConjTrans, PastixNoTrans, and solver_cblk_s::stride.

Referenced by cpucblk_zsytrfsp1d(), and cpucblk_zsytrfsp1dplus_update().

◆ cpucblk_zpotrfsp1dplus()

int cpucblk_zpotrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the Cholesky factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 391 of file core_zpotrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_zpotrfsp1d_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_zpotrf_dynamic().

◆ cpucblk_zpotrfsp1dplus_update()

void cpucblk_zpotrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		pastix_complex64_t *	work,
		pastix_int_t	lwork
	)

Apply the updates of the cholesky 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 444 of file core_zpotrfsp.c.

References cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_zalloc(), cpucblk_zgemmsp(), cpucblk_zrelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_blok_s::lcblknm, solver_matrix_s::lowrank, PastixConjTrans, and PastixLCoef.

Referenced by thread_zpotrf_dynamic().

◆ cpucblk_zsytrfsp1dplus()

int cpucblk_zsytrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the LDL^t factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 566 of file core_zsytrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_zsytrfsp1d_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_zsytrf_dynamic().

◆ cpucblk_zsytrfsp1dplus_update()

void cpucblk_zsytrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		pastix_complex64_t *	work
	)

Apply the updates of the LDL^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 619 of file core_zsytrfsp.c.

References cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, core_zsytrfsp1d_gemm(), cpucblk_zalloc(), cpucblk_zrelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_blok_s::lcblknm, solver_cblk_s::lcoeftab, and PastixLCoef.

Referenced by thread_zsytrf_dynamic().

◆ cpucblk_zgetrfsp1dplus()

int cpucblk_zgetrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the LU factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 421 of file core_zgetrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), cblk_getdataU(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_zgetrfsp1d_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_zgetrf_dynamic().

◆ cpucblk_zgetrfsp1dplus_update()

void cpucblk_zgetrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		pastix_complex64_t *	work,
		pastix_int_t	lwork
	)

Apply the updates of the LU 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 475 of file core_zgetrfsp.c.

References cblk_getdataL(), cblk_getdataU(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_zalloc(), cpucblk_zgemmsp(), cpucblk_zrelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_blok_s::lcblknm, solver_matrix_s::lowrank, PastixLCoef, PastixLUCoef, PastixTrans, and PastixUCoef.

Referenced by thread_zgetrf_dynamic().

◆ core_dsytrfsp1d_gemm()

void core_dsytrfsp1d_gemm	(	const SolverCblk *	cblk,
		const SolverBlok *	blok,
		SolverCblk *	fcblk,
		const double *	L,
		double *	C,
		double *	work
	)

core_dsytrfsp1d_gemm - Computes the LDL^t factorization of one panel and apply all the trsm updates to this panel.

Parameters

[in]	cblk	The pointer to the data structure that describes the panel from which we compute the contributions. Next column blok must be accessible through cblk[1].
[in]	blok	The pointer to the data structure that describes the blok from which we compute the contributions.
[in]	fcblk	The pointer to the data structure that describes the panel on which we compute the contributions. Next column blok must be accessible through fcblk[1].
[in,out]	L	The pointer to the matrix storing the coefficients of the panel. Must be of size cblk.stride -by- cblk.width
[in,out]	C	The pointer to the matrix storing the coefficients of the target.
[in,out]	work	Temporary buffer used in core_dgemdm().

Definition at line 316 of file core_dsytrfsp.c.

References blok_rownbr(), cblk_colnbr(), solver_cblk_s::cblktype, solver_blok_s::coefind, core_dgemdm(), solver_cblk_s::fblokptr, solver_cblk_s::fcolnum, solver_blok_s::frownum, is_block_inside_fblock(), pastix_int_t, PASTIX_SUCCESS, PastixNoTrans, PastixTrans, and solver_cblk_s::stride.

Referenced by cpucblk_dsytrfsp1d(), and cpucblk_dsytrfsp1dplus_update().

◆ cpucblk_dpotrfsp1dplus()

int cpucblk_dpotrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the Cholesky factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 391 of file core_dpotrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_dpotrfsp1d_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_dpotrf_dynamic().

◆ cpucblk_dpotrfsp1dplus_update()

void cpucblk_dpotrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		double *	work,
		pastix_int_t	lwork
	)

Apply the updates of the cholesky 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 444 of file core_dpotrfsp.c.

References cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_dalloc(), cpucblk_dgemmsp(), cpucblk_drelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_blok_s::lcblknm, solver_matrix_s::lowrank, PastixLCoef, and PastixTrans.

Referenced by thread_dpotrf_dynamic().

◆ cpucblk_dsytrfsp1dplus()

int cpucblk_dsytrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the LDL^t factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 566 of file core_dsytrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_dsytrfsp1d_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_dsytrf_dynamic().

◆ cpucblk_dsytrfsp1dplus_update()

void cpucblk_dsytrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		double *	work
	)

Apply the updates of the LDL^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 619 of file core_dsytrfsp.c.

References cblk_getdataL(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, core_dsytrfsp1d_gemm(), cpucblk_dalloc(), cpucblk_drelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_blok_s::lcblknm, solver_cblk_s::lcoeftab, and PastixLCoef.

Referenced by thread_dsytrf_dynamic().

◆ cpucblk_dgetrfsp1dplus()

int cpucblk_dgetrfsp1dplus	(	SolverMatrix *	solvmtx,
		SolverCblk *	cblk
	)

Perform the LU factorization of a given panel and submit tasks for the subsequent 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].

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

Definition at line 421 of file core_dgetrfsp.c.

References solver_matrix_s::bloktab, cblk_getdataL(), cblk_getdataU(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_dgetrfsp1d_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_dgetrf_dynamic().

◆ cpucblk_dgetrfsp1dplus_update()

void cpucblk_dgetrfsp1dplus_update	(	SolverMatrix *	solvmtx,
		SolverBlok *	blok,
		double *	work,
		pastix_int_t	lwork
	)

Apply the updates of the LU 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 475 of file core_dgetrfsp.c.

References cblk_getdataL(), cblk_getdataU(), solver_matrix_s::cblktab, solver_cblk_s::cblktype, cpucblk_dalloc(), cpucblk_dgemmsp(), cpucblk_drelease_deps(), solver_cblk_s::fblokptr, solver_blok_s::fcblknm, solver_blok_s::lcblknm, solver_matrix_s::lowrank, PastixLCoef, PastixLUCoef, PastixTrans, and PastixUCoef.

Referenced by thread_dgetrf_dynamic().

Functions

Detailed Description

Function Documentation

◆ core_dgemmsp_1d1d()

◆ core_dgemmsp_1d2d()

◆ core_dgemmsp_2d2d()

◆ core_dgemmsp_block_frfr()

◆ core_dgemmsp_block_frlr()

◆ core_dgemmsp_block_lrlr()

◆ core_dgemmsp_fulllr()

◆ core_dgemmsp_lr()

◆ core_dgemmsp_lrfr()

◆ core_zgemmsp_1d1d()

◆ core_zgemmsp_1d2d()

◆ core_zgemmsp_2d2d()

◆ core_zgemmsp_block_frfr()

◆ core_zgemmsp_block_frlr()

◆ core_zgemmsp_block_lrlr()

◆ core_zgemmsp_fulllr()

◆ core_zgemmsp_lr()

◆ core_zgemmsp_lrfr()

◆ core_cgemmsp_1d1d()

◆ core_cgemmsp_1d2d()

◆ core_cgemmsp_2d2d()

◆ core_cgemmsp_block_frfr()

◆ core_cgemmsp_block_frlr()

◆ core_cgemmsp_block_lrlr()

◆ core_cgemmsp_fulllr()

◆ core_cgemmsp_lr()

◆ core_cgemmsp_lrfr()

◆ core_sgemmsp_1d1d()

◆ core_sgemmsp_1d2d()

◆ core_sgemmsp_2d2d()

◆ core_sgemmsp_block_frfr()

◆ core_sgemmsp_block_frlr()

◆ core_sgemmsp_block_lrlr()

◆ core_sgemmsp_fulllr()

◆ core_sgemmsp_lr()

◆ core_sgemmsp_lrfr()

◆ core_dtrsmsp_1d()

◆ core_dtrsmsp_2d()

◆ core_dtrsmsp_lr()

◆ core_dtrsmsp_2dsub()

◆ core_dtrsmsp_lrsub()

◆ core_ztrsmsp_1d()

◆ core_ztrsmsp_2d()

◆ core_ztrsmsp_lr()

◆ core_ztrsmsp_2dsub()

◆ core_ztrsmsp_lrsub()

◆ core_ctrsmsp_1d()

◆ core_ctrsmsp_2d()

◆ core_ctrsmsp_lr()

◆ core_ctrsmsp_2dsub()

◆ core_ctrsmsp_lrsub()

◆ core_strsmsp_1d()

◆ core_strsmsp_2d()

◆ core_strsmsp_lr()

◆ core_strsmsp_2dsub()

◆ core_strsmsp_lrsub()

◆ core_csytrfsp1d_gemm()

◆ cpucblk_cpotrfsp1dplus()

◆ cpucblk_cpotrfsp1dplus_update()

◆ cpucblk_csytrfsp1dplus()

◆ cpucblk_csytrfsp1dplus_update()

◆ cpucblk_cgetrfsp1dplus()

◆ cpucblk_cgetrfsp1dplus_update()

◆ core_ssytrfsp1d_gemm()

◆ cpucblk_spotrfsp1dplus()

◆ cpucblk_spotrfsp1dplus_update()

◆ cpucblk_ssytrfsp1dplus()

◆ cpucblk_ssytrfsp1dplus_update()

◆ cpucblk_sgetrfsp1dplus()

◆ cpucblk_sgetrfsp1dplus_update()

◆ core_zsytrfsp1d_gemm()

◆ cpucblk_zpotrfsp1dplus()

◆ cpucblk_zpotrfsp1dplus_update()

◆ cpucblk_zsytrfsp1dplus()

◆ cpucblk_zsytrfsp1dplus_update()

◆ cpucblk_zgetrfsp1dplus()

◆ cpucblk_zgetrfsp1dplus_update()