22 #ifndef _pastix_scores_h_
23 #define _pastix_scores_h_
25 #ifndef DOXYGEN_SHOULD_SKIP_THIS
26 #define pastix_cblk_lock( cblk_ ) pastix_atomic_lock( &((cblk_)->lock) )
27 #define pastix_cblk_unlock( cblk_ ) pastix_atomic_unlock( &((cblk_)->lock) )
46 unsigned long long int seed );
207 #if defined(PRECISION_z) || defined(PRECISION_c)
328 #if defined(PRECISION_z) || defined(PRECISION_c)
403 const pastix_bcsc_t *bcsc,
407 const pastix_bcsc_t *bcsc,
409 const char *directory );
456 #if defined( PASTIX_WITH_MPI )
464 void cpucblk_smpi_rhs_fwd_progress(
const args_solve_t *enums,
483 void cpucblk_smpi_rhs_bwd_progress(
const args_solve_t *enums,
592 #if defined(PRECISION_z) || defined(PRECISION_c)
630 #if defined(PRECISION_z) || defined(PRECISION_c)
BEGIN_C_DECLS typedef int pastix_int_t
pastix_fixdbl_t core_slrorthu_fullqr(pastix_int_t M, pastix_int_t N, pastix_int_t rank, float *U, pastix_int_t ldu, float *V, pastix_int_t ldv)
Try to orthognalize the u part of the low-rank form, and update the v part accordingly using full QR.
int core_sgemdm(pastix_trans_t transA, pastix_trans_t transB, int M, int N, int K, float alpha, const float *A, int LDA, const float *B, int LDB, float beta, float *C, int LDC, const float *D, int incD, float *WORK, int LWORK)
Perform one of the following matrix-matrix operations.
int core_stqrcp(float tol, pastix_int_t maxrank, int unused, pastix_int_t nb, pastix_int_t m, pastix_int_t n, float *A, pastix_int_t lda, pastix_int_t *jpvt, float *tau, float *work, pastix_int_t lwork, float *rwork)
Compute a randomized QR factorization with truncated updates.
void core_spotrfsp(pastix_int_t n, float *A, pastix_int_t lda, pastix_int_t *nbpivots, float criterion)
Compute the block static pivoting Cholesky factorization of the matrix n-by-n A = L * L^t .
void core_ssytrfsp(pastix_int_t n, float *A, pastix_int_t lda, pastix_int_t *nbpivots, float criterion)
Compute the block static pivoting factorization of the symmetric matrix n-by-n A such that A = L * D ...
pastix_fixdbl_t core_slrorthu_cgs(pastix_int_t M1, pastix_int_t N1, pastix_int_t M2, pastix_int_t N2, pastix_int_t r1, pastix_int_t *r2ptr, pastix_int_t offx, pastix_int_t offy, float *U, pastix_int_t ldu, float *V, pastix_int_t ldv)
Try to orthognalize the U part of the low-rank form, and update the V part accordingly using CGS.
int core_sgeadd(pastix_trans_t trans, pastix_int_t M, pastix_int_t N, float alpha, const float *A, pastix_int_t LDA, float beta, float *B, pastix_int_t LDB)
Add two matrices together.
void core_sgetmo(int m, int n, const float *A, int lda, float *B, int ldb)
Transposes a m-by-n matrix out of place using an extra workspace of size m-by-n.
int core_sscalo(pastix_trans_t trans, pastix_int_t M, pastix_int_t N, const float *A, pastix_int_t lda, const float *D, pastix_int_t ldd, float *B, pastix_int_t ldb)
Scale a matrix by a diagonal out of place.
int core_spqrcp(float tol, pastix_int_t maxrank, int full_update, pastix_int_t nb, pastix_int_t m, pastix_int_t n, float *A, pastix_int_t lda, pastix_int_t *jpvt, float *tau, float *work, pastix_int_t lwork, float *rwork)
Compute a rank-reavealing QR factorization.
pastix_fixdbl_t core_slrorthu_partialqr(pastix_int_t M, pastix_int_t N, pastix_int_t r1, pastix_int_t *r2ptr, pastix_int_t offx, pastix_int_t offy, float *U, pastix_int_t ldu, float *V, pastix_int_t ldv)
Try to orthognalize the U part of the low-rank form, and update the V part accordingly using partial ...
int core_stradd(pastix_uplo_t uplo, pastix_trans_t trans, pastix_int_t M, pastix_int_t N, float alpha, const float *A, pastix_int_t LDA, float beta, float *B, pastix_int_t LDB)
Add two triangular matrices together as in PBLAS pstradd.
int core_srqrcp(float tol, pastix_int_t maxrank, int refine, pastix_int_t nb, pastix_int_t m, pastix_int_t n, float *A, pastix_int_t lda, pastix_int_t *jpvt, float *tau, float *work, pastix_int_t lwork, float *rwork)
Compute a randomized QR factorization.
void core_sgetrfsp(pastix_int_t n, float *A, pastix_int_t lda, pastix_int_t *nbpivots, float criterion)
Compute the block static pivoting LU factorization of the matrix m-by-n A = L * U.
int core_srqrrt(float tol, pastix_int_t maxrank, pastix_int_t nb, pastix_int_t m, pastix_int_t n, float *A, pastix_int_t lda, float *tau, float *B, pastix_int_t ldb, float *tau_b, float *work, pastix_int_t lwork, float normA)
Compute a randomized QR factorization with rotation technique.
void core_splrnt(int m, int n, float *A, int lda, int gM, int m0, int n0, unsigned long long int seed)
Generate a random tile.
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.
void cpucblk_ssytrfsp1dplus_update(SolverMatrix *solvmtx, SolverBlok *blok, float *work)
Apply the updates of the LDL^t factorisation of a given panel.
void core_ssytrfsp1d_gemm(const SolverCblk *cblk, const SolverBlok *blok, SolverCblk *fcblk, const float *L, float *C, float *work)
int cpucblk_ssytrfsp1dplus(SolverMatrix *solvmtx, SolverCblk *cblk)
Perform the LDL^t factorization of a given panel and submit tasks for the subsequent updates.
int cpucblk_sgetrfsp1dplus(SolverMatrix *solvmtx, SolverCblk *cblk)
Perform the LU factorization of a given panel and submit tasks for the subsequent updates.
int cpucblk_spotrfsp1dplus(SolverMatrix *solvmtx, SolverCblk *cblk)
Perform the Cholesky factorization of a given panel and submit tasks for the subsequent updates.
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.
int cpucblk_sincoming_rhs_fwd_deps(int rank, const args_solve_t *enums, SolverMatrix *solvmtx, SolverCblk *cblk, pastix_rhs_t rhsb)
Wait for incoming dependencies, and return when cblk->ctrbcnt has reached 0.
void cpucblk_srequest_rhs_bwd_cleanup(const args_solve_t *enums, pastix_int_t sched, SolverMatrix *solvmtx, pastix_rhs_t rhsb)
Waitall routine for current cblk request.
pastix_fixdbl_t cpucblk_sadd(float alpha, const SolverCblk *cblkA, SolverCblk *cblkB, const void *A, void *B, float *work, pastix_int_t lwork, const pastix_lr_t *lowrank)
Add two column bloks in full rank format.
void cpucblk_srelease_rhs_bwd_deps(const args_solve_t *enums, SolverMatrix *solvmtx, pastix_rhs_t rhsb, const SolverCblk *cblk, SolverCblk *fcbk)
Release the dependencies of the given cblk after an update.
int cpucblk_spotrfsp1d(SolverMatrix *solvmtx, SolverCblk *cblk, float *work, pastix_int_t lwork)
Perform the Cholesky factorization of a given panel and apply all its updates.
void cpucblk_salloc_lr(pastix_coefside_t side, SolverCblk *cblk, int rkmax)
Allocate the cblk structure to store the coefficient.
void cpucblk_srecv_rhs_backward(const SolverMatrix *solvmtx, SolverCblk *cblk, pastix_rhs_t b)
Receive the rhs associated to a cblk->lcolidx to the remote node.
void cpucblk_srequest_cleanup(pastix_coefside_t side, pastix_int_t sched, SolverMatrix *solvmtx)
Waitall routine for current cblk request.
void cpucblk_sinit(pastix_coefside_t side, const SolverMatrix *solvmtx, const pastix_bcsc_t *bcsc, pastix_int_t itercblk, const char *directory)
Fully initialize a single cblk.
void cpucblk_sfillin(pastix_coefside_t side, const SolverMatrix *solvmtx, const pastix_bcsc_t *bcsc, pastix_int_t itercblk)
Initialize the coeftab structure from the internal bcsc.
void cpucblk_salloc_lrws(const SolverCblk *cblk, pastix_lrblock_t *lrblok, float *ws)
Initialize lrblock structure from a workspace for all blocks of the cblk associated.
void cpucblk_sgetschur(const SolverCblk *cblk, int upper_part, float *S, pastix_int_t lds)
Extract a cblk panel of the Schur complement to a dense lapack form.
int cpucblk_ssytrfsp1d_sytrf(SolverMatrix *solvmtx, SolverCblk *cblk, void *dataL)
Computes the LDL^t factorization of the diagonal block in a panel.
int cpucblk_sincoming_rhs_bwd_deps(int rank, const args_solve_t *enums, SolverMatrix *solvmtx, SolverCblk *cblk, pastix_rhs_t rhsb)
Wait for incoming dependencies, and return when cblk->ctrbcnt has reached 0.
void cpucblk_salloc_fr(pastix_coefside_t side, SolverCblk *cblk)
Allocate the cblk structure to store the coefficient.
pastix_fixdbl_t cpucblk_sgemmsp(pastix_coefside_t sideA, pastix_trans_t trans, const SolverCblk *cblk, const SolverBlok *blok, SolverCblk *fcblk, const void *A, const void *B, void *C, float *work, pastix_int_t lwork, const pastix_lr_t *lowrank)
Compute the updates associated to one off-diagonal block.
int cpucblk_sgetrfsp1d(SolverMatrix *solvmtx, SolverCblk *cblk, float *work, pastix_int_t lwork)
Perform the LU factorization of a given panel and apply all its updates.
void cpucblk_strsmsp(pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk *cblk, const void *A, void *C, const pastix_lr_t *lowrank)
Compute the updates associated to a column of off-diagonal blocks.
void cpucblk_suncompress(pastix_coefside_t side, SolverCblk *cblk)
Uncompress a single column block from low-rank format to full-rank format.
pastix_fixdbl_t cpublok_sgemmsp(pastix_trans_t trans, const SolverCblk *cblk, SolverCblk *fcblk, pastix_int_t blok_mk, pastix_int_t blok_nk, pastix_int_t blok_mn, const void *A, const void *B, void *C, const pastix_lr_t *lowrank)
Compute the CPU gemm associated to a couple of off-diagonal blocks.
int cpucblk_sgeaddsp1d(const SolverCblk *cblk1, SolverCblk *cblk2, const float *L1, float *L2, const float *U1, float *U2)
Add two column blocks together.
void cpucblk_salloc(pastix_coefside_t side, SolverCblk *cblk)
Allocate the cblk structure to store the coefficient.
void cpucblk_srelease_rhs_fwd_deps(const args_solve_t *enums, SolverMatrix *solvmtx, pastix_rhs_t rhsb, const SolverCblk *cblk, SolverCblk *fcbk)
Release the dependencies of the given cblk after an update.
void cpucblk_ssend_rhs_backward(const SolverMatrix *solvmtx, SolverCblk *cblk, pastix_rhs_t b)
Send the rhs associated to a cblk->lcolidx to the remote node.
void cpucblk_srequest_rhs_fwd_cleanup(const args_solve_t *enums, pastix_int_t sched, SolverMatrix *solvmtx, pastix_rhs_t rhsb)
Waitall routine for current cblk request.
void cpucblk_sfree(pastix_coefside_t side, SolverCblk *cblk)
Free the cblk structure that store the coefficient.
void cpucblk_smemory(pastix_coefside_t side, const SolverMatrix *solvmtx, SolverCblk *cblk, pastix_int_t *orig, pastix_int_t *gain)
Return the memory gain of the low-rank form over the full-rank form for a single column-block.
int cpucblk_sgetrfsp1d_panel(SolverMatrix *solvmtx, SolverCblk *cblk, void *L, void *U)
Compute the LU factorization of one panel.
int cpucblk_ssytrfsp1d(SolverMatrix *solvmtx, SolverCblk *cblk, float *Dlt, float *work, pastix_int_t lwork)
Perform the LDL^t factorization of a given panel and apply all its updates.
void cpucblk_ssend_rhs_forward(const SolverMatrix *solvmtx, SolverCblk *cblk, pastix_rhs_t b)
Send the rhs associated to a cblk->lcolidx to the remote node.
int cpucblk_sincoming_deps(int mt_flag, pastix_coefside_t side, SolverMatrix *solvmtx, SolverCblk *cblk)
Wait for incoming dependencies, and return when cblk->ctrbcnt has reached 0.
void cpublok_sscalo(pastix_trans_t trans, const SolverCblk *cblk, pastix_int_t blok_m, const void *A, const void *dataD, void *dataB)
Copy the lower terms of the block with scaling for the two-terms algorithm.
pastix_fixdbl_t cpublok_strsmsp(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 void *A, void *C, const pastix_lr_t *lowrank)
Compute the updates associated to one off-diagonal block.
void cpucblk_sscalo(pastix_trans_t trans, const SolverCblk *cblk, void *dataL, void *dataLD)
Copy the L term with scaling for the two-terms algorithm.
int cpucblk_spotrfsp1d_potrf(SolverMatrix *solvmtx, SolverCblk *cblk, void *dataL)
Compute the Cholesky factorization of the diagonal block in a panel.
void cpucblk_srecv_rhs_forward(const SolverMatrix *solvmtx, SolverCblk *cblk, float *work, pastix_rhs_t b)
Receive the rhs associated to a cblk->lcolidx to the remote node.
int cpucblk_sdiff(pastix_coefside_t side, const SolverCblk *cblkA, SolverCblk *cblkB)
Compare two column blocks in full-rank format.
pastix_fixdbl_t cpublok_sadd(float alpha, const SolverCblk *cblkA, SolverCblk *cblkB, pastix_int_t blokA_m, pastix_int_t blokB_m, const void *A, void *B, float *work, pastix_int_t lwork, const pastix_lr_t *lowrank)
Add two bloks.
int cpucblk_sgetrfsp1d_getrf(SolverMatrix *solvmtx, SolverCblk *cblk, void *L, void *U)
Compute the LU factorization of the diagonal block in a panel.
int cpucblk_spotrfsp1d_panel(SolverMatrix *solvmtx, SolverCblk *cblk, void *L)
Compute the Cholesky factorization of one panel.
pastix_fixdbl_t cpublok_scompress(const pastix_lr_t *lowrank, pastix_int_t M, pastix_int_t N, pastix_lrblock_t *blok)
Compress a single block from full-rank to low-rank format.
int cpucblk_ssytrfsp1d_panel(SolverMatrix *solvmtx, SolverCblk *cblk, void *L, void *DLt)
Compute the LDL^t factorization of one panel.
pastix_int_t cpucblk_scompress(const SolverMatrix *solvmtx, pastix_coefside_t side, int max_ilulvl, SolverCblk *cblk)
Compress a single column block from full-rank to low-rank format.
void cpucblk_srelease_deps(pastix_coefside_t side, SolverMatrix *solvmtx, const SolverCblk *cblk, SolverCblk *fcbk)
Release the dependencies of the given cblk after an update.
void cpucblk_sdump(pastix_coefside_t side, const SolverCblk *cblk, FILE *stream)
Dump a single column block into a FILE in a human readale format.
Structure to define the type of function to use for the low-rank kernels and their parameters.
The block low-rank structure to hold a matrix in low-rank form.
void solve_blok_strsm(pastix_side_t side, pastix_uplo_t uplo, pastix_trans_t trans, pastix_diag_t diag, const SolverCblk *cblk, int nrhs, const void *dataA, float *b, int ldb)
Apply a solve trsm update related to a diagonal block of the matrix A.
void solve_cblk_sdiag(const SolverCblk *cblk, const void *dataA, int nrhs, float *b, int ldb, float *work)
Apply the diagonal solve related to one cblk to all the right hand side.
void solve_cblk_strsmsp_backward(const args_solve_t *enums, SolverMatrix *datacode, SolverCblk *cblk, pastix_rhs_t b)
Apply a backward solve related to one cblk to all the right hand side.
void solve_cblk_strsmsp_forward(const args_solve_t *enums, SolverMatrix *datacode, const SolverCblk *cblk, pastix_rhs_t b)
Apply a forward solve related to one cblk to all the right hand side.
void solve_blok_sgemm(pastix_side_t side, pastix_trans_t trans, pastix_int_t nrhs, const SolverCblk *cblk, const SolverBlok *blok, SolverCblk *fcbk, const void *dataA, const float *B, pastix_int_t ldb, float *C, pastix_int_t ldc)
Apply a solve gemm update related to a single block of the matrix A.
enum pastix_diag_e pastix_diag_t
Diagonal.
enum pastix_uplo_e pastix_uplo_t
Upper/Lower part.
enum pastix_side_e pastix_side_t
Side of the operation.
enum pastix_trans_e pastix_trans_t
Transpostion.
enum pastix_coefside_e pastix_coefside_t
Data blocks used in the kernel.
Main PaStiX RHS structure.
Solver column block structure.
Solver column block structure.