libflame revision_anchor
Functions
blis_prototypes_util.h File Reference

(r)

Go to the source code of this file.

Functions

float bl1_s2 (void)
 
double bl1_d2 (void)
 
scomplex bl1_c2 (void)
 
dcomplex bl1_z2 (void)
 
float bl1_s1 (void)
 
double bl1_d1 (void)
 
scomplex bl1_c1 (void)
 
dcomplex bl1_z1 (void)
 
float bl1_s1h (void)
 
double bl1_d1h (void)
 
scomplex bl1_c1h (void)
 
dcomplex bl1_z1h (void)
 
float bl1_s0 (void)
 
double bl1_d0 (void)
 
scomplex bl1_c0 (void)
 
dcomplex bl1_z0 (void)
 
float bl1_sm1h (void)
 
double bl1_dm1h (void)
 
scomplex bl1_cm1h (void)
 
dcomplex bl1_zm1h (void)
 
float bl1_sm1 (void)
 
double bl1_dm1 (void)
 
scomplex bl1_cm1 (void)
 
dcomplex bl1_zm1 (void)
 
float bl1_sm2 (void)
 
double bl1_dm2 (void)
 
scomplex bl1_cm2 (void)
 
dcomplex bl1_zm2 (void)
 
void * bl1_vallocv (unsigned int n_elem, unsigned int elem_size)
 
int * bl1_iallocv (unsigned int n_elem)
 
float * bl1_sallocv (unsigned int n_elem)
 
double * bl1_dallocv (unsigned int n_elem)
 
scomplex * bl1_callocv (unsigned int n_elem)
 
dcomplex * bl1_zallocv (unsigned int n_elem)
 
void * bl1_vallocm (unsigned int m, unsigned int n, unsigned int elem_size)
 
int * bl1_iallocm (unsigned int m, unsigned int n)
 
float * bl1_sallocm (unsigned int m, unsigned int n)
 
double * bl1_dallocm (unsigned int m, unsigned int n)
 
scomplex * bl1_callocm (unsigned int m, unsigned int n)
 
dcomplex * bl1_zallocm (unsigned int m, unsigned int n)
 
void bl1_sapdiagmv (side1_t side, conj1_t conj, int m, int n, float *x, int incx, float *a, int a_rs, int a_cs)
 
void bl1_dapdiagmv (side1_t side, conj1_t conj, int m, int n, double *x, int incx, double *a, int a_rs, int a_cs)
 
void bl1_csapdiagmv (side1_t side, conj1_t conj, int m, int n, float *x, int incx, scomplex *a, int a_rs, int a_cs)
 
void bl1_capdiagmv (side1_t side, conj1_t conj, int m, int n, scomplex *x, int incx, scomplex *a, int a_rs, int a_cs)
 
void bl1_zdapdiagmv (side1_t side, conj1_t conj, int m, int n, double *x, int incx, dcomplex *a, int a_rs, int a_cs)
 
void bl1_zapdiagmv (side1_t side, conj1_t conj, int m, int n, dcomplex *x, int incx, dcomplex *a, int a_rs, int a_cs)
 
void bl1_screate_contigm (int m, int n, float *a_save, int a_rs_save, int a_cs_save, float **a, int *a_rs, int *a_cs)
 
void bl1_dcreate_contigm (int m, int n, double *a_save, int a_rs_save, int a_cs_save, double **a, int *a_rs, int *a_cs)
 
void bl1_ccreate_contigm (int m, int n, scomplex *a_save, int a_rs_save, int a_cs_save, scomplex **a, int *a_rs, int *a_cs)
 
void bl1_zcreate_contigm (int m, int n, dcomplex *a_save, int a_rs_save, int a_cs_save, dcomplex **a, int *a_rs, int *a_cs)
 
void bl1_screate_contigmt (trans1_t trans_dims, int m, int n, float *a_save, int a_rs_save, int a_cs_save, float **a, int *a_rs, int *a_cs)
 
void bl1_dcreate_contigmt (trans1_t trans_dims, int m, int n, double *a_save, int a_rs_save, int a_cs_save, double **a, int *a_rs, int *a_cs)
 
void bl1_ccreate_contigmt (trans1_t trans_dims, int m, int n, scomplex *a_save, int a_rs_save, int a_cs_save, scomplex **a, int *a_rs, int *a_cs)
 
void bl1_zcreate_contigmt (trans1_t trans_dims, int m, int n, dcomplex *a_save, int a_rs_save, int a_cs_save, dcomplex **a, int *a_rs, int *a_cs)
 
void bl1_screate_contigmr (uplo1_t uplo, int m, int n, float *a_save, int a_rs_save, int a_cs_save, float **a, int *a_rs, int *a_cs)
 
void bl1_dcreate_contigmr (uplo1_t uplo, int m, int n, double *a_save, int a_rs_save, int a_cs_save, double **a, int *a_rs, int *a_cs)
 
void bl1_ccreate_contigmr (uplo1_t uplo, int m, int n, scomplex *a_save, int a_rs_save, int a_cs_save, scomplex **a, int *a_rs, int *a_cs)
 
void bl1_zcreate_contigmr (uplo1_t uplo, int m, int n, dcomplex *a_save, int a_rs_save, int a_cs_save, dcomplex **a, int *a_rs, int *a_cs)
 
void bl1_screate_contigmsr (side1_t side, uplo1_t uplo, int m, int n, float *a_save, int a_rs_save, int a_cs_save, float **a, int *a_rs, int *a_cs)
 
void bl1_dcreate_contigmsr (side1_t side, uplo1_t uplo, int m, int n, double *a_save, int a_rs_save, int a_cs_save, double **a, int *a_rs, int *a_cs)
 
void bl1_ccreate_contigmsr (side1_t side, uplo1_t uplo, int m, int n, scomplex *a_save, int a_rs_save, int a_cs_save, scomplex **a, int *a_rs, int *a_cs)
 
void bl1_zcreate_contigmsr (side1_t side, uplo1_t uplo, int m, int n, dcomplex *a_save, int a_rs_save, int a_cs_save, dcomplex **a, int *a_rs, int *a_cs)
 
void bl1_sfree_contigm (float *a_save, int a_rs_save, int a_cs_save, float **a, int *a_rs, int *a_cs)
 
void bl1_dfree_contigm (double *a_save, int a_rs_save, int a_cs_save, double **a, int *a_rs, int *a_cs)
 
void bl1_cfree_contigm (scomplex *a_save, int a_rs_save, int a_cs_save, scomplex **a, int *a_rs, int *a_cs)
 
void bl1_zfree_contigm (dcomplex *a_save, int a_rs_save, int a_cs_save, dcomplex **a, int *a_rs, int *a_cs)
 
void bl1_sfree_saved_contigm (int m, int n, float *a_save, int a_rs_save, int a_cs_save, float **a, int *a_rs, int *a_cs)
 
void bl1_dfree_saved_contigm (int m, int n, double *a_save, int a_rs_save, int a_cs_save, double **a, int *a_rs, int *a_cs)
 
void bl1_cfree_saved_contigm (int m, int n, scomplex *a_save, int a_rs_save, int a_cs_save, scomplex **a, int *a_rs, int *a_cs)
 
void bl1_zfree_saved_contigm (int m, int n, dcomplex *a_save, int a_rs_save, int a_cs_save, dcomplex **a, int *a_rs, int *a_cs)
 
void bl1_sfree_saved_contigmr (uplo1_t uplo, int m, int n, float *a_save, int a_rs_save, int a_cs_save, float **a, int *a_rs, int *a_cs)
 
void bl1_dfree_saved_contigmr (uplo1_t uplo, int m, int n, double *a_save, int a_rs_save, int a_cs_save, double **a, int *a_rs, int *a_cs)
 
void bl1_cfree_saved_contigmr (uplo1_t uplo, int m, int n, scomplex *a_save, int a_rs_save, int a_cs_save, scomplex **a, int *a_rs, int *a_cs)
 
void bl1_zfree_saved_contigmr (uplo1_t uplo, int m, int n, dcomplex *a_save, int a_rs_save, int a_cs_save, dcomplex **a, int *a_rs, int *a_cs)
 
void bl1_sfree_saved_contigmsr (side1_t side, uplo1_t uplo, int m, int n, float *a_save, int a_rs_save, int a_cs_save, float **a, int *a_rs, int *a_cs)
 
void bl1_dfree_saved_contigmsr (side1_t side, uplo1_t uplo, int m, int n, double *a_save, int a_rs_save, int a_cs_save, double **a, int *a_rs, int *a_cs)
 
void bl1_cfree_saved_contigmsr (side1_t side, uplo1_t uplo, int m, int n, scomplex *a_save, int a_rs_save, int a_cs_save, scomplex **a, int *a_rs, int *a_cs)
 
void bl1_zfree_saved_contigmsr (side1_t side, uplo1_t uplo, int m, int n, dcomplex *a_save, int a_rs_save, int a_cs_save, dcomplex **a, int *a_rs, int *a_cs)
 
void bl1_sewinvscalv (conj1_t conj, int n, float *x, int incx, float *y, int incy)
 
void bl1_dewinvscalv (conj1_t conj, int n, double *x, int incx, double *y, int incy)
 
void bl1_csewinvscalv (conj1_t conj, int n, float *x, int incx, scomplex *y, int incy)
 
void bl1_cewinvscalv (conj1_t conj, int n, scomplex *x, int incx, scomplex *y, int incy)
 
void bl1_zdewinvscalv (conj1_t conj, int n, double *x, int incx, dcomplex *y, int incy)
 
void bl1_zewinvscalv (conj1_t conj, int n, dcomplex *x, int incx, dcomplex *y, int incy)
 
void bl1_sewinvscalmt (trans1_t trans, int m, int n, float *a, int a_rs, int a_cs, float *b, int b_rs, int b_cs)
 
void bl1_dewinvscalmt (trans1_t trans, int m, int n, double *a, int a_rs, int a_cs, double *b, int b_rs, int b_cs)
 
void bl1_csewinvscalmt (trans1_t trans, int m, int n, float *a, int a_rs, int a_cs, scomplex *b, int b_rs, int b_cs)
 
void bl1_cewinvscalmt (trans1_t trans, int m, int n, scomplex *a, int a_rs, int a_cs, scomplex *b, int b_rs, int b_cs)
 
void bl1_zdewinvscalmt (trans1_t trans, int m, int n, double *a, int a_rs, int a_cs, dcomplex *b, int b_rs, int b_cs)
 
void bl1_zewinvscalmt (trans1_t trans, int m, int n, dcomplex *a, int a_rs, int a_cs, dcomplex *b, int b_rs, int b_cs)
 
void bl1_sewscalv (conj1_t conj, int n, float *x, int incx, float *y, int incy)
 
void bl1_dewscalv (conj1_t conj, int n, double *x, int incx, double *y, int incy)
 
void bl1_csewscalv (conj1_t conj, int n, float *x, int incx, scomplex *y, int incy)
 
void bl1_cewscalv (conj1_t conj, int n, scomplex *x, int incx, scomplex *y, int incy)
 
void bl1_zdewscalv (conj1_t conj, int n, double *x, int incx, dcomplex *y, int incy)
 
void bl1_zewscalv (conj1_t conj, int n, dcomplex *x, int incx, dcomplex *y, int incy)
 
void bl1_sewscalmt (trans1_t trans, int m, int n, float *a, int a_rs, int a_cs, float *b, int b_rs, int b_cs)
 
void bl1_dewscalmt (trans1_t trans, int m, int n, double *a, int a_rs, int a_cs, double *b, int b_rs, int b_cs)
 
void bl1_csewscalmt (trans1_t trans, int m, int n, float *a, int a_rs, int a_cs, scomplex *b, int b_rs, int b_cs)
 
void bl1_cewscalmt (trans1_t trans, int m, int n, scomplex *a, int a_rs, int a_cs, scomplex *b, int b_rs, int b_cs)
 
void bl1_zdewscalmt (trans1_t trans, int m, int n, double *a, int a_rs, int a_cs, dcomplex *b, int b_rs, int b_cs)
 
void bl1_zewscalmt (trans1_t trans, int m, int n, dcomplex *a, int a_rs, int a_cs, dcomplex *b, int b_rs, int b_cs)
 
void bl1_vfree (void *p)
 
void bl1_ifree (int *p)
 
void bl1_sfree (float *p)
 
void bl1_dfree (double *p)
 
void bl1_cfree (scomplex *p)
 
void bl1_zfree (dcomplex *p)
 
void bl1_sinverts (conj1_t conj, float *alpha)
 
void bl1_dinverts (conj1_t conj, double *alpha)
 
void bl1_cinverts (conj1_t conj, scomplex *alpha)
 
void bl1_zinverts (conj1_t conj, dcomplex *alpha)
 
void bl1_sinvert2s (conj1_t conj, float *alpha, float *beta)
 
void bl1_dinvert2s (conj1_t conj, double *alpha, double *beta)
 
void bl1_cinvert2s (conj1_t conj, scomplex *alpha, scomplex *beta)
 
void bl1_zinvert2s (conj1_t conj, dcomplex *alpha, dcomplex *beta)
 
void bl1_sinvertv (conj1_t conj, int n, float *x, int incx)
 
void bl1_dinvertv (conj1_t conj, int n, double *x, int incx)
 
void bl1_cinvertv (conj1_t conj, int n, scomplex *x, int incx)
 
void bl1_zinvertv (conj1_t conj, int n, dcomplex *x, int incx)
 
void bl1_sident (int m, float *a, int a_rs, int a_cs)
 
void bl1_dident (int m, double *a, int a_rs, int a_cs)
 
void bl1_cident (int m, scomplex *a, int a_rs, int a_cs)
 
void bl1_zident (int m, dcomplex *a, int a_rs, int a_cs)
 
void bl1_smaxabsv (int n, float *x, int incx, float *maxabs)
 
void bl1_dmaxabsv (int n, double *x, int incx, double *maxabs)
 
void bl1_cmaxabsv (int n, scomplex *x, int incx, float *maxabs)
 
void bl1_zmaxabsv (int n, dcomplex *x, int incx, double *maxabs)
 
void bl1_smaxabsm (int m, int n, float *a, int a_rs, int a_cs, float *maxabs)
 
void bl1_dmaxabsm (int m, int n, double *a, int a_rs, int a_cs, double *maxabs)
 
void bl1_cmaxabsm (int m, int n, scomplex *a, int a_rs, int a_cs, float *maxabs)
 
void bl1_zmaxabsm (int m, int n, dcomplex *a, int a_rs, int a_cs, double *maxabs)
 
void bl1_smaxabsmr (uplo1_t uplo, int m, int n, float *a, int a_rs, int a_cs, float *maxabs)
 
void bl1_dmaxabsmr (uplo1_t uplo, int m, int n, double *a, int a_rs, int a_cs, double *maxabs)
 
void bl1_cmaxabsmr (uplo1_t uplo, int m, int n, scomplex *a, int a_rs, int a_cs, float *maxabs)
 
void bl1_zmaxabsmr (uplo1_t uplo, int m, int n, dcomplex *a, int a_rs, int a_cs, double *maxabs)
 
void bl1_srands (float *alpha)
 
void bl1_drands (double *alpha)
 
void bl1_crands (scomplex *alpha)
 
void bl1_zrands (dcomplex *alpha)
 
void bl1_srandv (int n, float *x, int incx)
 
void bl1_drandv (int n, double *x, int incx)
 
void bl1_crandv (int n, scomplex *x, int incx)
 
void bl1_zrandv (int n, dcomplex *x, int incx)
 
void bl1_srandm (int m, int n, float *a, int a_rs, int a_cs)
 
void bl1_drandm (int m, int n, double *a, int a_rs, int a_cs)
 
void bl1_crandm (int m, int n, scomplex *a, int a_rs, int a_cs)
 
void bl1_zrandm (int m, int n, dcomplex *a, int a_rs, int a_cs)
 
void bl1_srandmr (uplo1_t uplo, diag1_t diag, int m, int n, float *a, int a_rs, int a_cs)
 
void bl1_drandmr (uplo1_t uplo, diag1_t diag, int m, int n, double *a, int a_rs, int a_cs)
 
void bl1_crandmr (uplo1_t uplo, diag1_t diag, int m, int n, scomplex *a, int a_rs, int a_cs)
 
void bl1_zrandmr (uplo1_t uplo, diag1_t diag, int m, int n, dcomplex *a, int a_rs, int a_cs)
 
void bl1_set_contig_strides (int m, int n, int *rs, int *cs)
 
void bl1_set_dim_with_side (side1_t side, int m, int n, int *dim_new)
 
void bl1_set_dims_with_trans (trans1_t trans, int m, int n, int *m_new, int *n_new)
 
void bl1_isetv (int m, int *sigma, int *x, int incx)
 
void bl1_ssetv (int m, float *sigma, float *x, int incx)
 
void bl1_dsetv (int m, double *sigma, double *x, int incx)
 
void bl1_csetv (int m, scomplex *sigma, scomplex *x, int incx)
 
void bl1_zsetv (int m, dcomplex *sigma, dcomplex *x, int incx)
 
void bl1_isetm (int m, int n, int *sigma, int *a, int a_rs, int a_cs)
 
void bl1_ssetm (int m, int n, float *sigma, float *a, int a_rs, int a_cs)
 
void bl1_dsetm (int m, int n, double *sigma, double *a, int a_rs, int a_cs)
 
void bl1_csetm (int m, int n, scomplex *sigma, scomplex *a, int a_rs, int a_cs)
 
void bl1_zsetm (int m, int n, dcomplex *sigma, dcomplex *a, int a_rs, int a_cs)
 
void bl1_ssetmr (uplo1_t uplo, int m, int n, float *sigma, float *a, int a_rs, int a_cs)
 
void bl1_dsetmr (uplo1_t uplo, int m, int n, double *sigma, double *a, int a_rs, int a_cs)
 
void bl1_csetmr (uplo1_t uplo, int m, int n, scomplex *sigma, scomplex *a, int a_rs, int a_cs)
 
void bl1_zsetmr (uplo1_t uplo, int m, int n, dcomplex *sigma, dcomplex *a, int a_rs, int a_cs)
 
void bl1_isetdiag (int offset, int m, int n, int *sigma, int *a, int a_rs, int a_cs)
 
void bl1_ssetdiag (int offset, int m, int n, float *sigma, float *a, int a_rs, int a_cs)
 
void bl1_dsetdiag (int offset, int m, int n, double *sigma, double *a, int a_rs, int a_cs)
 
void bl1_csetdiag (int offset, int m, int n, scomplex *sigma, scomplex *a, int a_rs, int a_cs)
 
void bl1_zsetdiag (int offset, int m, int n, dcomplex *sigma, dcomplex *a, int a_rs, int a_cs)
 
void bl1_sscalediag (conj1_t conj, int offset, int m, int n, float *sigma, float *a, int a_rs, int a_cs)
 
void bl1_dscalediag (conj1_t conj, int offset, int m, int n, double *sigma, double *a, int a_rs, int a_cs)
 
void bl1_cscalediag (conj1_t conj, int offset, int m, int n, scomplex *sigma, scomplex *a, int a_rs, int a_cs)
 
void bl1_zscalediag (conj1_t conj, int offset, int m, int n, dcomplex *sigma, dcomplex *a, int a_rs, int a_cs)
 
void bl1_csscalediag (conj1_t conj, int offset, int m, int n, float *sigma, scomplex *a, int a_rs, int a_cs)
 
void bl1_zdscalediag (conj1_t conj, int offset, int m, int n, double *sigma, dcomplex *a, int a_rs, int a_cs)
 
void bl1_sshiftdiag (conj1_t conj, int offset, int m, int n, float *sigma, float *a, int a_rs, int a_cs)
 
void bl1_dshiftdiag (conj1_t conj, int offset, int m, int n, double *sigma, double *a, int a_rs, int a_cs)
 
void bl1_cshiftdiag (conj1_t conj, int offset, int m, int n, scomplex *sigma, scomplex *a, int a_rs, int a_cs)
 
void bl1_zshiftdiag (conj1_t conj, int offset, int m, int n, dcomplex *sigma, dcomplex *a, int a_rs, int a_cs)
 
void bl1_csshiftdiag (conj1_t conj, int offset, int m, int n, float *sigma, scomplex *a, int a_rs, int a_cs)
 
void bl1_zdshiftdiag (conj1_t conj, int offset, int m, int n, double *sigma, dcomplex *a, int a_rs, int a_cs)
 
void bl1_ssymmize (conj1_t conj, uplo1_t uplo, int m, float *a, int a_rs, int a_cs)
 
void bl1_dsymmize (conj1_t conj, uplo1_t uplo, int m, double *a, int a_rs, int a_cs)
 
void bl1_csymmize (conj1_t conj, uplo1_t uplo, int m, scomplex *a, int a_rs, int a_cs)
 
void bl1_zsymmize (conj1_t conj, uplo1_t uplo, int m, dcomplex *a, int a_rs, int a_cs)
 

Function Documentation

◆ bl1_c0()

scomplex bl1_c0 ( void  )
126{
127 scomplex x;
128 x.real = bl1_s0();
129 x.imag = bl1_s0();
130 return x;
131}
i
int i
Definition bl1_axmyv2.c:145
bl1_s0
float bl1_s0(void)
Definition bl1_constants.c:111
scomplex
Definition blis_type_defs.h:133
scomplex::real
float real
Definition blis_type_defs.h:134

References bl1_s0(), scomplex::imag, and scomplex::real.

Referenced by bl1_cgemm(), bl1_cgemv(), bl1_chemm(), bl1_chemv(), bl1_crandmr(), bl1_csymm(), FLA_QR_UT_form_Q_opc_var1(), and FLA_Tridiag_UT_shift_U_l_opc().

◆ bl1_c1()

scomplex bl1_c1 ( void  )
62{
63 scomplex x;
64 x.real = bl1_s1();
65 x.imag = bl1_s0();
66 return x;
67}
bl1_s1
float bl1_s1(void)
Definition bl1_constants.c:47

References bl1_s0(), bl1_s1(), scomplex::imag, and scomplex::real.

Referenced by bl1_cgemm(), bl1_cgemv(), bl1_chemm(), bl1_chemv(), bl1_cher2k(), bl1_cherk(), bl1_crandmr(), bl1_csymm(), bl1_ctrmmsx(), bl1_ctrsmsx(), FLA_Bsvd_ext_opc_var1(), FLA_Bsvd_ext_ops_var1(), FLA_Bsvd_v_opc_var1(), FLA_Bsvd_v_ops_var1(), FLA_QR_UT_form_Q_opc_var1(), and FLA_Tridiag_UT_shift_U_l_opc().

◆ bl1_c1h()

scomplex bl1_c1h ( void  )
94{
95 scomplex x;
96 x.real = bl1_s1h();
97 x.imag = bl1_s0();
98 return x;
99}
bl1_s1h
float bl1_s1h(void)
Definition bl1_constants.c:79

References bl1_s0(), bl1_s1h(), scomplex::imag, and scomplex::real.

◆ bl1_c2()

scomplex bl1_c2 ( void  )
30{
31 scomplex x;
32 x.real = bl1_s2();
33 x.imag = bl1_s0();
34 return x;
35}
bl1_s2
float bl1_s2(void)
Definition bl1_constants.c:15

References bl1_s0(), bl1_s2(), scomplex::imag, and scomplex::real.

◆ bl1_callocm()

scomplex * bl1_callocm ( unsigned int  m,
unsigned int  n 
)
41{
42 return ( scomplex* ) BLIS1_MALLOC( m * n * sizeof( scomplex ) );
43}

Referenced by bl1_ccreate_contigm(), bl1_ccreate_contigmr(), bl1_ccreate_contigmt(), bl1_cgemm(), bl1_chemm(), bl1_cher2k(), bl1_cherk(), bl1_csymm(), bl1_csyr2k(), bl1_ctrmm(), bl1_ctrmmsx(), bl1_ctrsm(), and bl1_ctrsmsx().

◆ bl1_callocv()

scomplex * bl1_callocv ( unsigned int  n_elem)
41{
42 return ( scomplex* ) BLIS1_MALLOC( n_elem * sizeof( scomplex ) );
43}

Referenced by bl1_caxpymt(), bl1_caxpysmt(), bl1_caxpyv(), bl1_cgemv(), bl1_cger(), bl1_chemv(), bl1_cher(), bl1_cher2(), bl1_csymv_blas(), bl1_csyr2_blas(), bl1_csyr_blas(), bl1_ctrmv(), bl1_ctrmvsx(), bl1_ctrsv(), and bl1_ctrsvsx().

◆ bl1_capdiagmv()

void bl1_capdiagmv ( side1_t  side,
conj1_t  conj,
int  m,
int  n,
scomplex *  x,
int  incx,
scomplex *  a,
int  a_rs,
int  a_cs 
)
179{
180 scomplex* chi;
181 scomplex* a_begin;
182 int inca, lda;
183 int n_iter;
184 int n_elem;
185 int j;
186
187 // Return early if possible.
188 if ( bl1_zero_dim2( m, n ) ) return;
189
190 // Initialize with optimal values for column-major storage.
191 inca = a_rs;
192 lda = a_cs;
193 n_iter = n;
194 n_elem = m;
195
196 // An optimization: if A is row-major, then we can proceed as if the
197 // operation were transposed (applying the diagonal values in x from the
198 // opposite side) for increased spatial locality.
199 if ( bl1_is_row_storage( a_rs, a_cs ) )
200 {
201 bl1_swap_ints( n_iter, n_elem );
202 bl1_swap_ints( lda, inca );
203 bl1_toggle_side( side );
204 }
205
206 if ( bl1_is_left( side ) )
207 {
208 for ( j = 0; j < n_iter; j++ )
209 {
210 a_begin = a + j*lda;
211
212 bl1_cewscalv( conj,
213 n_elem,
214 x, incx,
215 a_begin, inca );
216 }
217 }
218 else
219 {
220 for ( j = 0; j < n_iter; j++ )
221 {
222 a_begin = a + j*lda;
223 chi = x + j*incx;
224
225 bl1_cscalv( conj,
226 n_elem,
227 chi,
228 a_begin, inca );
229 }
230 }
231}
bl1_cscalv
void bl1_cscalv(conj1_t conj, int n, scomplex *alpha, scomplex *x, int incx)
Definition bl1_scalv.c:46
bl1_is_row_storage
int bl1_is_row_storage(int rs, int cs)
Definition bl1_is.c:95
bl1_is_left
int bl1_is_left(side1_t side)
Definition bl1_is.c:61
bl1_zero_dim2
int bl1_zero_dim2(int m, int n)
Definition bl1_is.c:118
bl1_cewscalv
void bl1_cewscalv(conj1_t conj, int n, scomplex *x, int incx, scomplex *y, int incy)
Definition bl1_ewscalv.c:58

References bl1_cewscalv(), bl1_cscalv(), bl1_is_left(), bl1_is_row_storage(), and bl1_zero_dim2().

Referenced by FLA_Apply_diag_matrix().

◆ bl1_ccreate_contigm()

void bl1_ccreate_contigm ( int  m,
int  n,
scomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
scomplex **  a,
int *  a_rs,
int *  a_cs 
)
82{
83 int m_contig, n_contig;
84
85 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
86 {
87 // Initialize dimensions assuming no transposition needed during copy.
88 m_contig = m;
89 n_contig = n;
90
91/*
92 // Transpose the dimensions of the contiguous matrix, if requested.
93 if ( bl1_does_trans( trans_copy ) )
94 {
95 m_contig = n;
96 n_contig = m;
97 }
98*/
99
100 // Allocate temporary contiguous storage for the matrix.
101 *a = bl1_callocm( m_contig, n_contig );
102
103 // Set the row and column strides for the temporary matrix.
104 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
105
106 // Initialize the contiguous matrix with the contents of the original.
107 bl1_ccopymt( BLIS1_NO_TRANSPOSE,
108 m_contig,
109 n_contig,
110 a_save, a_rs_save, a_cs_save,
111 *a, *a_rs, *a_cs );
112 }
113}
bl1_ccopymt
void bl1_ccopymt(trans1_t trans, int m, int n, scomplex *a, int a_rs, int a_cs, scomplex *b, int b_rs, int b_cs)
Definition bl1_copymt.c:215
bl1_is_gen_storage
int bl1_is_gen_storage(int rs, int cs)
Definition bl1_is.c:100
bl1_set_contig_strides
void bl1_set_contig_strides(int m, int n, int *rs, int *cs)
Definition bl1_set_contig_strides.c:13
bl1_callocm
scomplex * bl1_callocm(unsigned int m, unsigned int n)
Definition bl1_allocm.c:40
BLIS1_NO_TRANSPOSE
@ BLIS1_NO_TRANSPOSE
Definition blis_type_defs.h:54

References bl1_callocm(), bl1_ccopymt(), bl1_is_gen_storage(), bl1_set_contig_strides(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_cgemm(), bl1_cgemv(), bl1_cger(), bl1_chemm(), bl1_csymm(), bl1_ctrmm(), bl1_ctrmmsx(), bl1_ctrsm(), and bl1_ctrsmsx().

◆ bl1_ccreate_contigmr()

void bl1_ccreate_contigmr ( uplo1_t  uplo,
int  m,
int  n,
scomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
scomplex **  a,
int *  a_rs,
int *  a_cs 
)
78{
79 int m_contig, n_contig;
80
81 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
82 {
83 // Initialize dimensions assuming no transposition needed during copy.
84 m_contig = m;
85 n_contig = n;
86/*
87 // Transpose the dimensions of the contiguous matrix, if requested.
88 if ( bl1_does_trans( trans_copy ) )
89 {
90 m_contig = n;
91 n_contig = m;
92 }
93*/
94 // Allocate temporary contiguous storage for the matrix.
95 *a = bl1_callocm( m_contig, n_contig );
96
97 // Set the row and column strides for the temporary matrix.
98 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
99
100 // Initialize the contiguous matrix with the contents of the original.
101 bl1_ccopymr( uplo,
102 m_contig,
103 n_contig,
104 a_save, a_rs_save, a_cs_save,
105 *a, *a_rs, *a_cs );
106 }
107}
bl1_ccopymr
void bl1_ccopymr(uplo1_t uplo, int m, int n, scomplex *a, int a_rs, int a_cs, scomplex *b, int b_rs, int b_cs)
Definition bl1_copymr.c:139

References bl1_callocm(), bl1_ccopymr(), bl1_is_gen_storage(), and bl1_set_contig_strides().

Referenced by bl1_ccreate_contigmsr(), bl1_chemm(), bl1_chemv(), bl1_cher(), bl1_cher2(), bl1_cher2k(), bl1_cherk(), bl1_csymm(), bl1_csymv(), bl1_csyr(), bl1_csyr2(), bl1_csyr2k(), bl1_csyrk(), bl1_ctrmm(), bl1_ctrmmsx(), bl1_ctrmv(), bl1_ctrmvsx(), bl1_ctrsm(), bl1_ctrsmsx(), bl1_ctrsv(), and bl1_ctrsvsx().

◆ bl1_ccreate_contigmsr()

void bl1_ccreate_contigmsr ( side1_t  side,
uplo1_t  uplo,
int  m,
int  n,
scomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
scomplex **  a,
int *  a_rs,
int *  a_cs 
)
46{
47 int dim_a;
48
49 // Choose the dimension of the matrix based on the side parameter.
50 if ( bl1_is_left( side ) ) dim_a = m;
51 else dim_a = n;
52
53 // Call the simple version with chosen dimensions.
54 bl1_ccreate_contigmr( uplo,
55 dim_a,
56 dim_a,
57 a_save, a_rs_save, a_cs_save,
58 a, a_rs, a_cs );
59}
bl1_ccreate_contigmr
void bl1_ccreate_contigmr(uplo1_t uplo, int m, int n, scomplex *a_save, int a_rs_save, int a_cs_save, scomplex **a, int *a_rs, int *a_cs)
Definition bl1_create_contigmr.c:77

References bl1_ccreate_contigmr(), and bl1_is_left().

◆ bl1_ccreate_contigmt()

void bl1_ccreate_contigmt ( trans1_t  trans_dims,
int  m,
int  n,
scomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
scomplex **  a,
int *  a_rs,
int *  a_cs 
)
90{
91 int m_contig, n_contig;
92
93 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
94 {
95 // Transpose the dimensions if requested.
96 if ( bl1_does_trans( trans_dims ) )
97 bl1_swap_ints( m, n );
98
99 // Initialize dimensions assuming no transposition needed during copy.
100 m_contig = m;
101 n_contig = n;
102
103/*
104 // Transpose the dimensions of the contiguous matrix, if requested.
105 if ( bl1_does_trans( trans_copy ) )
106 {
107 m_contig = n;
108 n_contig = m;
109 }
110*/
111
112 // Allocate temporary contiguous storage for the matrix.
113 *a = bl1_callocm( m_contig, n_contig );
114
115 // Set the row and column strides for the temporary matrix.
116 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
117
118 // Initialize the contiguous matrix with the contents of the original.
119 bl1_ccopymt( BLIS1_NO_TRANSPOSE,
120 m_contig,
121 n_contig,
122 a_save, a_rs_save, a_cs_save,
123 *a, *a_rs, *a_cs );
124 }
125}
bl1_does_trans
int bl1_does_trans(trans1_t trans)
Definition bl1_does.c:13

References bl1_callocm(), bl1_ccopymt(), bl1_does_trans(), bl1_is_gen_storage(), bl1_set_contig_strides(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_cgemm(), bl1_cher2k(), bl1_cherk(), bl1_csyr2k(), and bl1_csyrk().

◆ bl1_cewinvscalmt()

void bl1_cewinvscalmt ( trans1_t  trans,
int  m,
int  n,
scomplex *  a,
int  a_rs,
int  a_cs,
scomplex *  b,
int  b_rs,
int  b_cs 
)
230{
231 scomplex* a_begin;
232 scomplex* b_begin;
233 int lda, inca;
234 int ldb, incb;
235 int n_iter;
236 int n_elem;
237 int j;
238 conj1_t conj;
239
240 // Return early if possible.
241 if ( bl1_zero_dim2( m, n ) ) return;
242
243 // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
244 // gets invoked only once.
245 if ( bl1_is_vector( m, n ) )
246 {
247 // Initialize with values appropriate for vectors.
248 n_iter = 1;
249 n_elem = bl1_vector_dim( m, n );
250 lda = 1; // multiplied by zero when n_iter == 1; not needed.
251 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
252 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
253 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
254 }
255 else // matrix case
256 {
257 // Initialize with optimal values for column-major storage.
258 n_iter = n;
259 n_elem = m;
260 lda = a_cs;
261 inca = a_rs;
262 ldb = b_cs;
263 incb = b_rs;
264
265 // Handle the transposition of A.
266 if ( bl1_does_trans( trans ) )
267 {
268 bl1_swap_ints( lda, inca );
269 }
270
271 // An optimization: if B is row-major and if A is effectively row-major
272 // after a possible transposition, then let's access the matrices by rows
273 // instead of by columns for increased spatial locality.
274 if ( bl1_is_row_storage( b_rs, b_cs ) )
275 {
276 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
277 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
278 {
279 bl1_swap_ints( n_iter, n_elem );
280 bl1_swap_ints( lda, inca );
281 bl1_swap_ints( ldb, incb );
282 }
283 }
284 }
285
286 // Extract conj component from trans parameter.
287 conj = bl1_proj_trans1_to_conj( trans );
288
289 for ( j = 0; j < n_iter; j++ )
290 {
291 a_begin = a + j*lda;
292 b_begin = b + j*ldb;
293
294 bl1_cewinvscalv( conj,
295 n_elem,
296 a_begin, inca,
297 b_begin, incb );
298 }
299}
bl1_does_notrans
int bl1_does_notrans(trans1_t trans)
Definition bl1_does.c:19
bl1_is_vector
int bl1_is_vector(int m, int n)
Definition bl1_is.c:106
bl1_is_col_storage
int bl1_is_col_storage(int rs, int cs)
Definition bl1_is.c:90
bl1_proj_trans1_to_conj
conj1_t bl1_proj_trans1_to_conj(trans1_t trans)
Definition bl1_proj.c:13
bl1_vector_dim
int bl1_vector_dim(int m, int n)
Definition bl1_vector.c:13
bl1_vector_inc
int bl1_vector_inc(trans1_t trans, int m, int n, int rs, int cs)
Definition bl1_vector.c:19
bl1_cewinvscalv
void bl1_cewinvscalv(conj1_t conj, int n, scomplex *x, int incx, scomplex *y, int incy)
Definition bl1_ewinvscalv.c:58
conj1_t
conj1_t
Definition blis_type_defs.h:80

References bl1_cewinvscalv(), bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Inv_scal_elemwise().

◆ bl1_cewinvscalv()

void bl1_cewinvscalv ( conj1_t  conj,
int  n,
scomplex *  x,
int  incx,
scomplex *  y,
int  incy 
)
59{
60 scomplex* chi;
61 scomplex* psi;
62 scomplex conjchi;
63 int i;
64
65 if ( bl1_is_conj( conj ) )
66 {
67 for ( i = 0; i < n; ++i )
68 {
69 chi = x + i*incx;
70 psi = y + i*incy;
71
72 bl1_ccopyconj( chi, &conjchi );
73 bl1_cinvscals( &conjchi, psi );
74 }
75 }
76 else
77 {
78 for ( i = 0; i < n; ++i )
79 {
80 chi = x + i*incx;
81 psi = y + i*incy;
82
83 bl1_cinvscals( chi, psi );
84 }
85 }
86}
bl1_is_conj
int bl1_is_conj(conj1_t conj)
Definition bl1_is.c:42

References bl1_is_conj(), and i.

Referenced by bl1_cewinvscalmt().

◆ bl1_cewscalmt()

void bl1_cewscalmt ( trans1_t  trans,
int  m,
int  n,
scomplex *  a,
int  a_rs,
int  a_cs,
scomplex *  b,
int  b_rs,
int  b_cs 
)
230{
231 scomplex* a_begin;
232 scomplex* b_begin;
233 int lda, inca;
234 int ldb, incb;
235 int n_iter;
236 int n_elem;
237 int j;
238 conj1_t conj;
239
240 // Return early if possible.
241 if ( bl1_zero_dim2( m, n ) ) return;
242
243 // Handle cases where A and B are vectors to ensure that the underlying ewscal
244 // gets invoked only once.
245 if ( bl1_is_vector( m, n ) )
246 {
247 // Initialize with values appropriate for vectors.
248 n_iter = 1;
249 n_elem = bl1_vector_dim( m, n );
250 lda = 1; // multiplied by zero when n_iter == 1; not needed.
251 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
252 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
253 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
254 }
255 else // matrix case
256 {
257 // Initialize with optimal values for column-major storage.
258 n_iter = n;
259 n_elem = m;
260 lda = a_cs;
261 inca = a_rs;
262 ldb = b_cs;
263 incb = b_rs;
264
265 // Handle the transposition of A.
266 if ( bl1_does_trans( trans ) )
267 {
268 bl1_swap_ints( lda, inca );
269 }
270
271 // An optimization: if B is row-major and if A is effectively row-major
272 // after a possible transposition, then let's access the matrices by rows
273 // instead of by columns for increased spatial locality.
274 if ( bl1_is_row_storage( b_rs, b_cs ) )
275 {
276 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
277 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
278 {
279 bl1_swap_ints( n_iter, n_elem );
280 bl1_swap_ints( lda, inca );
281 bl1_swap_ints( ldb, incb );
282 }
283 }
284 }
285
286 // Extract conj component from trans parameter.
287 conj = bl1_proj_trans1_to_conj( trans );
288
289 for ( j = 0; j < n_iter; j++ )
290 {
291 a_begin = a + j*lda;
292 b_begin = b + j*ldb;
293
294 bl1_cewscalv( conj,
295 n_elem,
296 a_begin, inca,
297 b_begin, incb );
298 }
299}

References bl1_cewscalv(), bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Scal_elemwise().

◆ bl1_cewscalv()

void bl1_cewscalv ( conj1_t  conj,
int  n,
scomplex *  x,
int  incx,
scomplex *  y,
int  incy 
)
59{
60 scomplex* chi;
61 scomplex* psi;
62 scomplex conjchi;
63 int i;
64
65 if ( bl1_is_conj( conj ) )
66 {
67 for ( i = 0; i < n; ++i )
68 {
69 chi = x + i*incx;
70 psi = y + i*incy;
71
72 bl1_ccopyconj( chi, &conjchi );
73 bl1_cscals( &conjchi, psi );
74 }
75 }
76 else
77 {
78 for ( i = 0; i < n; ++i )
79 {
80 chi = x + i*incx;
81 psi = y + i*incy;
82
83 bl1_cscals( chi, psi );
84 }
85 }
86}

References bl1_is_conj(), and i.

Referenced by bl1_capdiagmv(), and bl1_cewscalmt().

◆ bl1_cfree()

void bl1_cfree ( scomplex *  p)
41{
42 free( ( void* ) p );
43}

Referenced by bl1_caxpymt(), bl1_caxpysmt(), bl1_caxpyv(), bl1_cfree_contigm(), bl1_cfree_saved_contigm(), bl1_cfree_saved_contigmr(), bl1_cfree_saved_contigmsr(), bl1_cgemm(), bl1_cgemv(), bl1_cger(), bl1_chemm(), bl1_chemv(), bl1_cher(), bl1_cher2(), bl1_cher2k(), bl1_cherk(), bl1_csymm(), bl1_csymv_blas(), bl1_csyr2_blas(), bl1_csyr2k(), bl1_csyr_blas(), bl1_ctrmm(), bl1_ctrmmsx(), bl1_ctrmv(), bl1_ctrmvsx(), bl1_ctrsm(), bl1_ctrsmsx(), bl1_ctrsv(), and bl1_ctrsvsx().

◆ bl1_cfree_contigm()

void bl1_cfree_contigm ( scomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
scomplex **  a,
int *  a_rs,
int *  a_cs 
)
46{
47 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
48 {
49 // Free the temporary contiguous storage for the matrix.
50 bl1_cfree( *a );
51
52 // Restore the original matrix address.
53 *a = a_save;
54
55 // Restore the original row and column strides.
56 *a_rs = a_rs_save;
57 *a_cs = a_cs_save;
58 }
59}
bl1_cfree
void bl1_cfree(scomplex *p)
Definition bl1_free.c:40

References bl1_cfree(), and bl1_is_gen_storage().

Referenced by bl1_cgemm(), bl1_cgemv(), bl1_chemm(), bl1_chemv(), bl1_cher2k(), bl1_cherk(), bl1_csymm(), bl1_csymv(), bl1_csyr2k(), bl1_csyrk(), bl1_ctrmm(), bl1_ctrmmsx(), bl1_ctrmv(), bl1_ctrmvsx(), bl1_ctrsm(), bl1_ctrsmsx(), bl1_ctrsv(), and bl1_ctrsvsx().

◆ bl1_cfree_saved_contigm()

void bl1_cfree_saved_contigm ( int  m,
int  n,
scomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
scomplex **  a,
int *  a_rs,
int *  a_cs 
)
60{
61 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
62 {
63 // Copy the contents of the temporary matrix back to the original.
64 bl1_ccopymt( BLIS1_NO_TRANSPOSE,
65 m,
66 n,
67 *a, *a_rs, *a_cs,
68 a_save, a_rs_save, a_cs_save );
69
70 // Free the temporary contiguous storage for the matrix.
71 bl1_cfree( *a );
72
73 // Restore the original matrix address.
74 *a = a_save;
75
76 // Restore the original row and column strides.
77 *a_rs = a_rs_save;
78 *a_cs = a_cs_save;
79 }
80}

References bl1_ccopymt(), bl1_cfree(), bl1_is_gen_storage(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_cgemm(), bl1_cger(), bl1_chemm(), bl1_cher(), bl1_cher2(), bl1_csymm(), bl1_csyr(), bl1_csyr2(), bl1_ctrmm(), bl1_ctrmmsx(), bl1_ctrsm(), and bl1_ctrsmsx().

◆ bl1_cfree_saved_contigmr()

void bl1_cfree_saved_contigmr ( uplo1_t  uplo,
int  m,
int  n,
scomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
scomplex **  a,
int *  a_rs,
int *  a_cs 
)
60{
61 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
62 {
63 // Copy the contents of the temporary matrix back to the original.
64 bl1_ccopymr( uplo,
65 m,
66 n,
67 *a, *a_rs, *a_cs,
68 a_save, a_rs_save, a_cs_save );
69
70 // Free the temporary contiguous storage for the matrix.
71 bl1_cfree( *a );
72
73 // Restore the original matrix address.
74 *a = a_save;
75
76 // Restore the original row and column strides.
77 *a_rs = a_rs_save;
78 *a_cs = a_cs_save;
79 }
80}

References bl1_ccopymr(), bl1_cfree(), and bl1_is_gen_storage().

Referenced by bl1_cher2k(), bl1_cherk(), bl1_csyr2k(), and bl1_csyrk().

◆ bl1_cfree_saved_contigmsr()

void bl1_cfree_saved_contigmsr ( side1_t  side,
uplo1_t  uplo,
int  m,
int  n,
scomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
scomplex **  a,
int *  a_rs,
int *  a_cs 
)
72{
73 int dim_a;
74
75 // Choose the dimension of the matrix based on the side parameter.
76 if ( bl1_is_left( side ) ) dim_a = m;
77 else dim_a = n;
78
79 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
80 {
81 // Copy the contents of the temporary matrix back to the original.
82 bl1_ccopymt( uplo,
83 dim_a,
84 dim_a,
85 *a, *a_rs, *a_cs,
86 a_save, a_rs_save, a_cs_save );
87
88 // Free the temporary contiguous storage for the matrix.
89 bl1_cfree( *a );
90
91 // Restore the original matrix address.
92 *a = a_save;
93
94 // Restore the original row and column strides.
95 *a_rs = a_rs_save;
96 *a_cs = a_cs_save;
97 }
98}

References bl1_ccopymt(), bl1_cfree(), bl1_is_gen_storage(), and bl1_is_left().

◆ bl1_cident()

void bl1_cident ( int  m,
scomplex *  a,
int  a_rs,
int  a_cs 
)
52{
53 scomplex* alpha;
54 int i, j;
55
56 for ( j = 0; j < m; ++j )
57 {
58 for ( i = 0; i < m; ++i )
59 {
60 alpha = a + i*a_rs + j*a_cs;
61
62 alpha->real = 0.0F;
63 alpha->imag = 0.0F;
64
65 if ( i == j )
66 alpha->real = 1.0F;
67 }
68 }
69}

References i, scomplex::imag, and scomplex::real.

Referenced by FLA_UDdate_UT_opc_var1().

◆ bl1_cinvert2s()

void bl1_cinvert2s ( conj1_t  conj,
scomplex *  alpha,
scomplex *  beta 
)
28{
29 float temp;
30 float s, xr_s, xi_s;
31
32 s = bl1_fmaxabs( alpha->real, alpha->imag ); \
33 xr_s = alpha->real / s;
34 xi_s = alpha->imag / s;
35 temp = xr_s * alpha->real + xi_s * alpha->imag;
36
37 beta->real = xr_s / temp;
38 beta->imag = -xi_s / temp;
39
40 if ( bl1_is_conj( conj ) )
41 bl1_cconjs( beta );
42}
temp
dcomplex temp
Definition bl1_axpyv2b.c:301
dcomplex::real
double real
Definition blis_type_defs.h:139
dcomplex::imag
double imag
Definition blis_type_defs.h:139

References bl1_is_conj(), scomplex::imag, scomplex::real, and temp.

Referenced by bl1_cinvscalm(), and bl1_cinvscalv().

◆ bl1_cinverts()

void bl1_cinverts ( conj1_t  conj,
scomplex *  alpha 
)
28{
29 float temp;
30 float s, xr_s, xi_s;
31
32 s = bl1_fmaxabs( alpha->real, alpha->imag ); \
33 xr_s = alpha->real / s;
34 xi_s = alpha->imag / s;
35 temp = xr_s * alpha->real + xi_s * alpha->imag;
36
37 alpha->real = xr_s / temp;
38 alpha->imag = -xi_s / temp;
39
40 if ( bl1_is_conj( conj ) )
41 bl1_cconjs( alpha );
42}

References bl1_is_conj(), scomplex::imag, scomplex::real, and temp.

Referenced by FLA_Trinv_ln_opc_var1(), FLA_Trinv_ln_opc_var2(), FLA_Trinv_ln_opc_var3(), FLA_Trinv_ln_opc_var4(), FLA_Trinv_un_opc_var1(), FLA_Trinv_un_opc_var2(), FLA_Trinv_un_opc_var3(), and FLA_Trinv_un_opc_var4().

◆ bl1_cinvertv()

void bl1_cinvertv ( conj1_t  conj,
int  n,
scomplex *  x,
int  incx 
)
42{
43 float one = 1.0F;
44 float temp;
45 float s, xr_s, xi_s;
46 float conjsign;
47 scomplex* chi;
48 int i;
49
50 if ( bl1_is_conj( conj ) ) conjsign = one;
51 else conjsign = -one;
52
53 for ( i = 0; i < n; ++i )
54 {
55 chi = x + i*incx;
56
57 s = bl1_fmaxabs( chi->real, chi->imag ); \
58 xr_s = chi->real / s;
59 xi_s = chi->imag / s;
60 temp = xr_s * chi->real + xi_s * chi->imag;
61
62 chi->real = xr_s / temp;
63 chi->imag = conjsign * xi_s / temp;
64 }
65}

References bl1_is_conj(), i, scomplex::imag, scomplex::real, and temp.

Referenced by FLA_Invert().

◆ bl1_cm1()

scomplex bl1_cm1 ( void  )
190{
191 scomplex x;
192 x.real = bl1_sm1();
193 x.imag = bl1_s0();
194 return x;
195}
bl1_sm1
float bl1_sm1(void)
Definition bl1_constants.c:175

References bl1_s0(), bl1_sm1(), scomplex::imag, and scomplex::real.

◆ bl1_cm1h()

scomplex bl1_cm1h ( void  )
158{
159 scomplex x;
160 x.real = bl1_sm1h();
161 x.imag = bl1_s0();
162 return x;
163}
bl1_sm1h
float bl1_sm1h(void)
Definition bl1_constants.c:143

References bl1_s0(), bl1_sm1h(), scomplex::imag, and scomplex::real.

◆ bl1_cm2()

scomplex bl1_cm2 ( void  )
222{
223 scomplex x;
224 x.real = bl1_sm2();
225 x.imag = bl1_s0();
226 return x;
227}
bl1_sm2
float bl1_sm2(void)
Definition bl1_constants.c:207

References bl1_s0(), bl1_sm2(), scomplex::imag, and scomplex::real.

◆ bl1_cmaxabsm()

void bl1_cmaxabsm ( int  m,
int  n,
scomplex *  a,
int  a_rs,
int  a_cs,
float *  maxabs 
)
104{
105 float zero = bl1_s0();
106 scomplex* a_begin;
107 float maxabs_cand;
108 float maxabs_temp;
109 int inca, lda;
110 int n_iter;
111 int n_elem;
112 int j;
113
114 // Return early if possible.
115 if ( bl1_zero_dim2( m, n ) ) { *maxabs = zero; return; }
116
117 // Initialize with optimal values for column-major storage.
118 inca = a_rs;
119 lda = a_cs;
120 n_iter = n;
121 n_elem = m;
122
123 // An optimization: if A is row-major, then let's access the matrix by
124 // rows instead of by columns for increased spatial locality.
125 if ( bl1_is_row_storage( a_rs, a_cs ) )
126 {
127 bl1_swap_ints( n_iter, n_elem );
128 bl1_swap_ints( lda, inca );
129 }
130
131 // Initialize the maximum absolute value candidate to the first element.
132 bl1_csabsval2( a, &maxabs_cand );
133
134 for ( j = 0; j < n_iter; j++ )
135 {
136 a_begin = a + j*lda;
137
138 bl1_cmaxabsv( n_elem,
139 a_begin, inca,
140 &maxabs_temp );
141
142 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
143 }
144
145 *maxabs = maxabs_cand;
146}
bl1_s0
float bl1_s0(void)
Definition bl1_constants.c:111
bl1_cmaxabsv
void bl1_cmaxabsv(int n, scomplex *x, int incx, float *maxabs)
Definition bl1_maxabsv.c:55

References bl1_cmaxabsv(), bl1_is_row_storage(), bl1_s0(), and bl1_zero_dim2().

Referenced by FLA_Max_abs_value().

◆ bl1_cmaxabsmr()

void bl1_cmaxabsmr ( uplo1_t  uplo,
int  m,
int  n,
scomplex *  a,
int  a_rs,
int  a_cs,
float *  maxabs 
)
144{
145 float zero = bl1_d0();
146 scomplex* a_begin;
147 float maxabs_cand;
148 float maxabs_temp;
149 int inca, lda;
150 int n_iter;
151 int n_elem_max;
152 int n_elem;
153 int j;
154
155 // Return early if possible.
156 if ( bl1_zero_dim2( m, n ) ) { *maxabs = zero; return; }
157
158 // Initialize with optimal values for column-major storage.
159 n_iter = n;
160 n_elem_max = m;
161 lda = a_cs;
162 inca = a_rs;
163
164 // An optimization: if A is row-major, then let's access the matrix by
165 // rows instead of by columns for increased spatial locality.
166 if ( bl1_is_row_storage( a_rs, a_cs ) )
167 {
168 bl1_swap_ints( n_iter, n_elem_max );
169 bl1_swap_ints( lda, inca );
170 bl1_toggle_uplo( uplo );
171 }
172
173 // Initialize the maximum absolute value candidate to the first element.
174 bl1_csabsval2( a, &maxabs_cand );
175
176 if ( bl1_is_upper( uplo ) )
177 {
178 for ( j = 0; j < n_iter; j++ )
179 {
180 n_elem = bl1_min( j + 1, n_elem_max );
181 a_begin = a + j*lda;
182
183 bl1_cmaxabsv( n_elem,
184 a_begin, inca,
185 &maxabs_temp );
186
187 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
188 }
189 }
190 else // if ( bl1_is_lower( uplo ) )
191 {
192 for ( j = 0; j < n_iter; j++ )
193 {
194 n_elem = bl1_max( 0, n_elem_max - j );
195 a_begin = a + j*lda + j*inca;
196
197 bl1_cmaxabsv( n_elem,
198 a_begin, inca,
199 &maxabs_temp );
200
201 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
202 }
203 }
204
205 *maxabs = maxabs_cand;
206}
bl1_is_upper
int bl1_is_upper(uplo1_t uplo)
Definition bl1_is.c:54
bl1_d0
double bl1_d0(void)
Definition bl1_constants.c:118

References bl1_cmaxabsv(), bl1_d0(), bl1_is_row_storage(), bl1_is_upper(), and bl1_zero_dim2().

Referenced by FLA_Max_abs_value_herm().

◆ bl1_cmaxabsv()

void bl1_cmaxabsv ( int  n,
scomplex *  x,
int  incx,
float *  maxabs 
)
56{
57 scomplex* chi;
58 float maxabs_cand;
59 float maxabs_temp;
60 int i;
61
62 bl1_csabsval2( x, &maxabs_cand );
63
64 for ( i = 0; i < n; ++i )
65 {
66 chi = x + i*incx;
67
68 bl1_csabsval2( chi, &maxabs_temp );
69
70 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
71 }
72
73 *maxabs = maxabs_cand;
74}

References i.

Referenced by bl1_cmaxabsm(), and bl1_cmaxabsmr().

◆ bl1_crandm()

void bl1_crandm ( int  m,
int  n,
scomplex *  a,
int  a_rs,
int  a_cs 
)
82{
83 scomplex* a_begin;
84 int inca, lda;
85 int n_iter;
86 int n_elem;
87 int j;
88
89 // Return early if possible.
90 if ( bl1_zero_dim2( m, n ) ) return;
91
92 // Initialize with optimal values for column-major storage.
93 inca = a_rs;
94 lda = a_cs;
95 n_iter = n;
96 n_elem = m;
97
98 // An optimization: if A is row-major, then let's access the matrix by
99 // rows instead of by columns for increased spatial locality.
100 if ( bl1_is_row_storage( a_rs, a_cs ) )
101 {
102 bl1_swap_ints( n_iter, n_elem );
103 bl1_swap_ints( lda, inca );
104 }
105
106 for ( j = 0; j < n_iter; j++ )
107 {
108 a_begin = a + j*lda;
109
110 bl1_crandv( n_elem,
111 a_begin, inca );
112 }
113}
bl1_crandv
void bl1_crandv(int n, scomplex *x, int incx)
Definition bl1_randv.c:39

References bl1_crandv(), bl1_is_row_storage(), and bl1_zero_dim2().

Referenced by FLA_Random_matrix().

◆ bl1_crandmr()

void bl1_crandmr ( uplo1_t  uplo,
diag1_t  diag,
int  m,
int  n,
scomplex *  a,
int  a_rs,
int  a_cs 
)
256{
257 scomplex* a_begin;
258 scomplex* ajj;
259 scomplex one;
260 scomplex zero;
261 scomplex ord;
262 int lda, inca;
263 int n_iter;
264 int n_elem_max;
265 int n_elem;
266 int j;
267
268 // Return early if possible.
269 if ( bl1_zero_dim2( m, n ) ) return;
270
271 // Initialize with optimal values for column-major storage.
272 n_iter = n;
273 n_elem_max = m;
274 lda = a_cs;
275 inca = a_rs;
276
277 // An optimization: if A is row-major, then let's access the matrix by
278 // rows instead of by columns to increase spatial locality.
279 if ( bl1_is_row_storage( a_rs, a_cs ) )
280 {
281 bl1_swap_ints( n_iter, n_elem_max );
282 bl1_swap_ints( lda, inca );
283 bl1_toggle_uplo( uplo );
284 }
285
286 // Initialize some scalars.
287 one = bl1_c1();
288 zero = bl1_c0();
289 ord = bl1_c0();
290 ord.real = ( float ) bl1_max( m, n );
291
292 if ( bl1_is_upper( uplo ) )
293 {
294 for ( j = 0; j < n_iter; j++ )
295 {
296 n_elem = bl1_min( j, n_elem_max );
297 a_begin = a + j*lda;
298
299 // Randomize super-diagonal elements.
300 bl1_crandv( n_elem,
301 a_begin, inca );
302
303 // Normalize super-diagonal elements by order of the matrix.
304 bl1_cinvscalv( BLIS1_NO_CONJUGATE,
305 n_elem,
306 &ord,
307 a_begin, inca );
308
309 // Initialize diagonal and sub-diagonal elements only if there are
310 // elements left in the column (ie: j < n_elem_max).
311 if ( j < n_elem_max )
312 {
313 ajj = a_begin + j*inca;
314
315 // Initialize diagonal element.
316 if ( bl1_is_unit_diag( diag ) ) *ajj = one;
317 else if ( bl1_is_zero_diag( diag ) ) *ajj = zero;
318 else if ( bl1_is_nonunit_diag( diag ) )
319 {
320 // We want positive diagonal elements between 1 and 2.
321 bl1_crands( ajj );
322 bl1_cabsval2( ajj, ajj );
323 bl1_cadd3( ajj, &one, ajj );
324 }
325
326 // Initialize sub-diagonal elements to zero.
327 bl1_csetv( n_elem_max - j - 1,
328 &zero,
329 ajj + inca, inca );
330 }
331 }
332 }
333 else // if ( bl1_is_lower( uplo ) )
334 {
335 for ( j = 0; j < n_iter; j++ )
336 {
337 n_elem = bl1_min( j, n_elem_max );
338 a_begin = a + j*lda;
339
340 // Initialize super-diagonal to zero.
341 bl1_csetv( n_elem,
342 &zero,
343 a_begin, inca );
344
345 // Initialize diagonal and sub-diagonal elements only if there are
346 // elements left in the column (ie: j < n_elem_max).
347 if ( j < n_elem_max )
348 {
349 ajj = a_begin + j*inca;
350
351 // Initialize diagonal element.
352 if ( bl1_is_unit_diag( diag ) ) *ajj = one;
353 else if ( bl1_is_zero_diag( diag ) ) *ajj = zero;
354 else if ( bl1_is_nonunit_diag( diag ) )
355 {
356 // We want positive diagonal elements between 1 and 2.
357 bl1_crands( ajj );
358 bl1_cabsval2( ajj, ajj );
359 bl1_cadd3( ajj, &one, ajj );
360 }
361
362 // Randomize sub-diagonal elements.
363 bl1_crandv( n_elem_max - j - 1,
364 ajj + inca, inca );
365
366 // Normalize sub-diagonal elements by order of the matrix.
367 bl1_cinvscalv( BLIS1_NO_CONJUGATE,
368 n_elem_max - j - 1,
369 &ord,
370 ajj + inca, inca );
371
372 }
373 }
374 }
375}
bl1_cinvscalv
void bl1_cinvscalv(conj1_t conj, int n, scomplex *alpha, scomplex *x, int incx)
Definition bl1_invscalv.c:52
bl1_is_unit_diag
int bl1_is_unit_diag(diag1_t diag)
Definition bl1_is.c:78
bl1_is_zero_diag
int bl1_is_zero_diag(diag1_t diag)
Definition bl1_is.c:83
bl1_is_nonunit_diag
int bl1_is_nonunit_diag(diag1_t diag)
Definition bl1_is.c:73
bl1_c1
scomplex bl1_c1(void)
Definition bl1_constants.c:61
bl1_csetv
void bl1_csetv(int m, scomplex *sigma, scomplex *x, int incx)
Definition bl1_setv.c:52
bl1_crands
void bl1_crands(scomplex *alpha)
Definition bl1_rands.c:23
bl1_c0
scomplex bl1_c0(void)
Definition bl1_constants.c:125
BLIS1_NO_CONJUGATE
@ BLIS1_NO_CONJUGATE
Definition blis_type_defs.h:81

References bl1_c0(), bl1_c1(), bl1_cinvscalv(), bl1_crands(), bl1_crandv(), bl1_csetv(), bl1_is_nonunit_diag(), bl1_is_row_storage(), bl1_is_unit_diag(), bl1_is_upper(), bl1_is_zero_diag(), bl1_zero_dim2(), BLIS1_NO_CONJUGATE, and scomplex::real.

Referenced by FLA_Random_tri_matrix().

◆ bl1_crands()

void bl1_crands ( scomplex *  alpha)
24{
25 bl1_srands( &(alpha->real) );
26 bl1_srands( &(alpha->imag) );
27}
bl1_srands
void bl1_srands(float *alpha)
Definition bl1_rands.c:13

References bl1_srands(), scomplex::imag, and scomplex::real.

Referenced by bl1_crandmr(), and bl1_crandv().

◆ bl1_crandv()

void bl1_crandv ( int  n,
scomplex *  x,
int  incx 
)
40{
41 scomplex* chi;
42 int i;
43
44 for ( i = 0; i < n; ++i )
45 {
46 chi = x + i*incx;
47
48 bl1_crands( chi );
49 }
50}

References bl1_crands(), and i.

Referenced by bl1_crandm(), and bl1_crandmr().

◆ bl1_csapdiagmv()

void bl1_csapdiagmv ( side1_t  side,
conj1_t  conj,
int  m,
int  n,
float *  x,
int  incx,
scomplex *  a,
int  a_rs,
int  a_cs 
)
124{
125 float* chi;
126 scomplex* a_begin;
127 int inca, lda;
128 int n_iter;
129 int n_elem;
130 int j;
131
132 // Return early if possible.
133 if ( bl1_zero_dim2( m, n ) ) return;
134
135 // Initialize with optimal values for column-major storage.
136 inca = a_rs;
137 lda = a_cs;
138 n_iter = n;
139 n_elem = m;
140
141 // An optimization: if A is row-major, then we can proceed as if the
142 // operation were transposed (applying the diagonal values in x from the
143 // opposite side) for increased spatial locality.
144 if ( bl1_is_row_storage( a_rs, a_cs ) )
145 {
146 bl1_swap_ints( n_iter, n_elem );
147 bl1_swap_ints( lda, inca );
148 bl1_toggle_side( side );
149 }
150
151 if ( bl1_is_left( side ) )
152 {
153 for ( j = 0; j < n_iter; j++ )
154 {
155 a_begin = a + j*lda;
156
157 bl1_csewscalv( conj,
158 n_elem,
159 x, incx,
160 a_begin, inca );
161 }
162 }
163 else
164 {
165 for ( j = 0; j < n_iter; j++ )
166 {
167 a_begin = a + j*lda;
168 chi = x + j*incx;
169
170 bl1_csscalv( conj,
171 n_elem,
172 chi,
173 a_begin, inca );
174 }
175 }
176}
bl1_csscalv
void bl1_csscalv(conj1_t conj, int n, float *alpha, scomplex *x, int incx)
Definition bl1_scalv.c:35
bl1_csewscalv
void bl1_csewscalv(conj1_t conj, int n, float *x, int incx, scomplex *y, int incy)
Definition bl1_ewscalv.c:43

References bl1_csewscalv(), bl1_csscalv(), bl1_is_left(), bl1_is_row_storage(), and bl1_zero_dim2().

Referenced by FLA_Apply_diag_matrix().

◆ bl1_cscalediag()

void bl1_cscalediag ( conj1_t  conj,
int  offset,
int  m,
int  n,
scomplex *  sigma,
scomplex *  a,
int  a_rs,
int  a_cs 
)
100{
101 scomplex* alpha;
102 scomplex sigma_conj;
103 int i, j;
104
105 bl1_ccopys( conj, sigma, &sigma_conj );
106
107 i = j = 0;
108
109 if ( offset < 0 ) i = -offset;
110 else if ( offset > 0 ) j = offset;
111
112 while ( i < m && j < n )
113 {
114 alpha = a + i*a_rs + j*a_cs;
115
116 bl1_cscals( &sigma_conj, alpha );
117
118 ++i;
119 ++j;
120 }
121}

References i.

Referenced by FLA_Scale_diag(), and FLA_UDdate_UT_opc_var1().

◆ bl1_csetdiag()

void bl1_csetdiag ( int  offset,
int  m,
int  n,
scomplex *  sigma,
scomplex *  a,
int  a_rs,
int  a_cs 
)
77{
78 scomplex* alpha;
79 int i, j;
80
81 i = j = 0;
82
83 if ( offset < 0 ) i = -offset;
84 else if ( offset > 0 ) j = offset;
85
86 while ( i < m && j < n )
87 {
88 alpha = a + i*a_rs + j*a_cs;
89
90 alpha->real = sigma->real;
91 alpha->imag = sigma->imag;
92
93 ++i;
94 ++j;
95 }
96}

References i, scomplex::imag, and scomplex::real.

Referenced by FLA_Set_diag(), FLA_Set_offdiag(), and FLA_Triangularize().

◆ bl1_csetm()

void bl1_csetm ( int  m,
int  n,
scomplex *  sigma,
scomplex *  a,
int  a_rs,
int  a_cs 
)
62{
63 scomplex* alpha;
64 int i, j;
65
66 for ( j = 0; j < n; ++j )
67 {
68 for ( i = 0; i < m; ++i )
69 {
70 alpha = a + i*a_rs + j*a_cs;
71
72 alpha->real = sigma->real;
73 alpha->imag = sigma->imag;
74 }
75 }
76}

References i, scomplex::imag, and scomplex::real.

Referenced by FLA_Bidiag_UT_u_step_ofc_var4(), FLA_Bidiag_UT_u_step_opc_var4(), FLA_Bidiag_UT_u_step_opc_var5(), FLA_Bsvd_ext_opc_var1(), FLA_Bsvd_ext_ops_var1(), FLA_Bsvd_v_opc_var1(), FLA_Bsvd_v_ops_var1(), FLA_Hess_UT_step_ofc_var4(), FLA_Hess_UT_step_opc_var4(), FLA_Hess_UT_step_opc_var5(), FLA_Set(), FLA_Tridiag_UT_l_step_ofc_var3(), and FLA_Tridiag_UT_l_step_opc_var3().

◆ bl1_csetmr()

void bl1_csetmr ( uplo1_t  uplo,
int  m,
int  n,
scomplex *  sigma,
scomplex *  a,
int  a_rs,
int  a_cs 
)
120{
121 scomplex* a_begin;
122 int lda, inca;
123 int n_iter;
124 int n_elem_max;
125 int n_elem;
126 int j;
127
128 // Return early if possible.
129 if ( bl1_zero_dim2( m, n ) ) return;
130
131 // Initialize with optimal values for column-major storage.
132 n_iter = n;
133 n_elem_max = m;
134 lda = a_cs;
135 inca = a_rs;
136
137 // An optimization: if A is row-major, then let's access the matrix by
138 // rows instead of by columns to increase spatial locality.
139 if ( bl1_is_row_storage( a_rs, a_cs ) )
140 {
141 bl1_swap_ints( n_iter, n_elem_max );
142 bl1_swap_ints( lda, inca );
143 bl1_toggle_uplo( uplo );
144 }
145
146 if ( bl1_is_upper( uplo ) )
147 {
148 for ( j = 0; j < n_iter; j++ )
149 {
150 n_elem = bl1_min( j, n_elem_max );
151 a_begin = a + j*lda;
152
153 bl1_csetv( n_elem,
154 sigma,
155 a_begin, inca );
156 }
157 }
158 else // if ( bl1_is_lower( uplo ) )
159 {
160 for ( j = 0; j < n_iter; j++ )
161 {
162 n_elem = bl1_max( 0, n_elem_max - j - 1 );
163 a_begin = a + j*lda + (j + 1)*inca;
164
165 bl1_csetv( n_elem,
166 sigma,
167 a_begin, inca );
168 }
169 }
170}

References bl1_csetv(), bl1_is_row_storage(), bl1_is_upper(), and bl1_zero_dim2().

Referenced by FLA_Setr(), and FLA_Triangularize().

◆ bl1_csetv()

void bl1_csetv ( int  m,
scomplex *  sigma,
scomplex *  x,
int  incx 
)
53{
54 scomplex* chi;
55 int i;
56
57 for ( i = 0; i < n; ++i )
58 {
59 chi = x + i*incx;
60
61 chi->real = sigma->real;
62 chi->imag = sigma->imag;
63 }
64}

References i, scomplex::imag, and scomplex::real.

Referenced by bl1_crandmr(), bl1_csetmr(), FLA_Bidiag_UT_u_step_ofc_var4(), FLA_Bidiag_UT_u_step_opc_var4(), FLA_Fused_Ahx_Ax_opc_var1(), FLA_Fused_Ahx_Axpy_Ax_opc_var1(), FLA_Fused_Gerc2_Ahx_Ax_opc_var1(), FLA_Fused_Gerc2_Ahx_Axpy_Ax_opc_var1(), FLA_Fused_Her2_Ax_l_opc_var1(), FLA_Tridiag_UT_l_realify_opt(), FLA_Tridiag_UT_realify_subdiagonal_opt(), FLA_Tridiag_UT_shift_U_l_opc(), and FLA_Tridiag_UT_u_realify_opt().

◆ bl1_csewinvscalmt()

void bl1_csewinvscalmt ( trans1_t  trans,
int  m,
int  n,
float *  a,
int  a_rs,
int  a_cs,
scomplex *  b,
int  b_rs,
int  b_cs 
)
158{
159 float* a_begin;
160 scomplex* b_begin;
161 int lda, inca;
162 int ldb, incb;
163 int n_iter;
164 int n_elem;
165 int j;
166 conj1_t conj;
167
168 // Return early if possible.
169 if ( bl1_zero_dim2( m, n ) ) return;
170
171 // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
172 // gets invoked only once.
173 if ( bl1_is_vector( m, n ) )
174 {
175 // Initialize with values appropriate for vectors.
176 n_iter = 1;
177 n_elem = bl1_vector_dim( m, n );
178 lda = 1; // multiplied by zero when n_iter == 1; not needed.
179 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
180 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
181 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
182 }
183 else // matrix case
184 {
185 // Initialize with optimal values for column-major storage.
186 n_iter = n;
187 n_elem = m;
188 lda = a_cs;
189 inca = a_rs;
190 ldb = b_cs;
191 incb = b_rs;
192
193 // Handle the transposition of A.
194 if ( bl1_does_trans( trans ) )
195 {
196 bl1_swap_ints( lda, inca );
197 }
198
199 // An optimization: if B is row-major and if A is effectively row-major
200 // after a possible transposition, then let's access the matrices by rows
201 // instead of by columns for increased spatial locality.
202 if ( bl1_is_row_storage( b_rs, b_cs ) )
203 {
204 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
205 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
206 {
207 bl1_swap_ints( n_iter, n_elem );
208 bl1_swap_ints( lda, inca );
209 bl1_swap_ints( ldb, incb );
210 }
211 }
212 }
213
214 // Extract conj component from trans parameter.
215 conj = bl1_proj_trans1_to_conj( trans );
216
217 for ( j = 0; j < n_iter; j++ )
218 {
219 a_begin = a + j*lda;
220 b_begin = b + j*ldb;
221
222 bl1_csewinvscalv( conj,
223 n_elem,
224 a_begin, inca,
225 b_begin, incb );
226 }
227}
bl1_csewinvscalv
void bl1_csewinvscalv(conj1_t conj, int n, float *x, int incx, scomplex *y, int incy)
Definition bl1_ewinvscalv.c:43

References bl1_csewinvscalv(), bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

◆ bl1_csewinvscalv()

void bl1_csewinvscalv ( conj1_t  conj,
int  n,
float *  x,
int  incx,
scomplex *  y,
int  incy 
)
44{
45 float* chi;
46 scomplex* psi;
47 int i;
48
49 for ( i = 0; i < n; ++i )
50 {
51 chi = x + i*incx;
52 psi = y + i*incy;
53
54 bl1_csinvscals( chi, psi );
55 }
56}

References i.

Referenced by bl1_csewinvscalmt().

◆ bl1_csewscalmt()

void bl1_csewscalmt ( trans1_t  trans,
int  m,
int  n,
float *  a,
int  a_rs,
int  a_cs,
scomplex *  b,
int  b_rs,
int  b_cs 
)
158{
159 float* a_begin;
160 scomplex* b_begin;
161 int lda, inca;
162 int ldb, incb;
163 int n_iter;
164 int n_elem;
165 int j;
166 conj1_t conj;
167
168 // Return early if possible.
169 if ( bl1_zero_dim2( m, n ) ) return;
170
171 // Handle cases where A and B are vectors to ensure that the underlying ewscal
172 // gets invoked only once.
173 if ( bl1_is_vector( m, n ) )
174 {
175 // Initialize with values appropriate for vectors.
176 n_iter = 1;
177 n_elem = bl1_vector_dim( m, n );
178 lda = 1; // multiplied by zero when n_iter == 1; not needed.
179 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
180 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
181 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
182 }
183 else // matrix case
184 {
185 // Initialize with optimal values for column-major storage.
186 n_iter = n;
187 n_elem = m;
188 lda = a_cs;
189 inca = a_rs;
190 ldb = b_cs;
191 incb = b_rs;
192
193 // Handle the transposition of A.
194 if ( bl1_does_trans( trans ) )
195 {
196 bl1_swap_ints( lda, inca );
197 }
198
199 // An optimization: if B is row-major and if A is effectively row-major
200 // after a possible transposition, then let's access the matrices by rows
201 // instead of by columns for increased spatial locality.
202 if ( bl1_is_row_storage( b_rs, b_cs ) )
203 {
204 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
205 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
206 {
207 bl1_swap_ints( n_iter, n_elem );
208 bl1_swap_ints( lda, inca );
209 bl1_swap_ints( ldb, incb );
210 }
211 }
212 }
213
214 // Extract conj component from trans parameter.
215 conj = bl1_proj_trans1_to_conj( trans );
216
217 for ( j = 0; j < n_iter; j++ )
218 {
219 a_begin = a + j*lda;
220 b_begin = b + j*ldb;
221
222 bl1_csewscalv( conj,
223 n_elem,
224 a_begin, inca,
225 b_begin, incb );
226 }
227}

References bl1_csewscalv(), bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

◆ bl1_csewscalv()

void bl1_csewscalv ( conj1_t  conj,
int  n,
float *  x,
int  incx,
scomplex *  y,
int  incy 
)
44{
45 float* chi;
46 scomplex* psi;
47 int i;
48
49 for ( i = 0; i < n; ++i )
50 {
51 chi = x + i*incx;
52 psi = y + i*incy;
53
54 bl1_csscals( chi, psi );
55 }
56}

References i.

Referenced by bl1_csapdiagmv(), and bl1_csewscalmt().

◆ bl1_cshiftdiag()

void bl1_cshiftdiag ( conj1_t  conj,
int  offset,
int  m,
int  n,
scomplex *  sigma,
scomplex *  a,
int  a_rs,
int  a_cs 
)
98{
99 scomplex* alpha;
100 scomplex sigma_conj;
101 int i, j;
102
103 bl1_ccopys( conj, sigma, &sigma_conj );
104
105 i = j = 0;
106
107 if ( offset < 0 ) i = -offset;
108 else if ( offset > 0 ) j = offset;
109
110 while ( i < m && j < n )
111 {
112 alpha = a + i*a_rs + j*a_cs;
113
114 alpha->real += sigma_conj.real;
115 alpha->imag += sigma_conj.imag;
116
117 ++i;
118 ++j;
119 }
120}

References i, scomplex::imag, and scomplex::real.

Referenced by FLA_Lyap_h_opc_var1(), FLA_Lyap_h_opc_var2(), FLA_Lyap_h_opc_var3(), FLA_Lyap_h_opc_var4(), FLA_Lyap_n_opc_var1(), FLA_Lyap_n_opc_var2(), FLA_Lyap_n_opc_var3(), FLA_Lyap_n_opc_var4(), and FLA_Shift_diag().

◆ bl1_csscalediag()

void bl1_csscalediag ( conj1_t  conj,
int  offset,
int  m,
int  n,
float *  sigma,
scomplex *  a,
int  a_rs,
int  a_cs 
)
56{
57 scomplex* alpha;
58 int i, j;
59
60 i = j = 0;
61
62 if ( offset < 0 ) i = -offset;
63 else if ( offset > 0 ) j = offset;
64
65 while ( i < m && j < n )
66 {
67 alpha = a + i*a_rs + j*a_cs;
68
69 alpha->real *= *sigma;
70 alpha->imag *= *sigma;
71
72 ++i;
73 ++j;
74 }
75}

References i, scomplex::imag, and scomplex::real.

Referenced by FLA_Scale_diag().

◆ bl1_csshiftdiag()

void bl1_csshiftdiag ( conj1_t  conj,
int  offset,
int  m,
int  n,
float *  sigma,
scomplex *  a,
int  a_rs,
int  a_cs 
)
56{
57 scomplex* alpha;
58 int i, j;
59
60 i = j = 0;
61
62 if ( offset < 0 ) i = -offset;
63 else if ( offset > 0 ) j = offset;
64
65 while ( i < m && j < n )
66 {
67 alpha = a + i*a_rs + j*a_cs;
68
69 alpha->real += *sigma;
70
71 ++i;
72 ++j;
73 }
74}

References i, and scomplex::real.

Referenced by FLA_Shift_diag().

◆ bl1_csymmize()

void bl1_csymmize ( conj1_t  conj,
uplo1_t  uplo,
int  m,
scomplex *  a,
int  a_rs,
int  a_cs 
)
236{
237 scomplex* a_src;
238 scomplex* a_dst;
239 scomplex* a_jj;
240 int rs_src, cs_src, inc_src;
241 int rs_dst, cs_dst, inc_dst;
242 int n_iter;
243 int j;
244
245 // Return early if possible.
246 if ( bl1_zero_dim1( m ) ) return;
247
248 // Assume A is square.
249 n_iter = m;
250
251 // Initialize with appropriate values based on storage.
252 if ( bl1_is_col_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
253 {
254 cs_src = 1;
255 rs_src = 0;
256 inc_src = a_cs;
257 cs_dst = a_cs;
258 rs_dst = 0;
259 inc_dst = 1;
260 }
261 else if ( bl1_is_col_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
262 {
263 cs_src = a_cs;
264 rs_src = 0;
265 inc_src = 1;
266 cs_dst = 1;
267 rs_dst = 0;
268 inc_dst = a_cs;
269 }
270 else if ( bl1_is_row_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
271 {
272 cs_src = 0;
273 rs_src = a_rs;
274 inc_src = 1;
275 cs_dst = 0;
276 rs_dst = 1;
277 inc_dst = a_rs;
278 }
279 else if ( bl1_is_row_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
280 {
281 cs_src = 0;
282 rs_src = 1;
283 inc_src = a_rs;
284 cs_dst = 0;
285 rs_dst = a_rs;
286 inc_dst = 1;
287 }
288 else if ( bl1_is_gen_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
289 {
290 // General stride with column-major tilt looks similar to column-major.
291 // General stride with row-major tilt looks similar to row-major.
292 if ( a_rs < a_cs )
293 {
294 cs_src = 1 * a_rs;
295 rs_src = 0;
296 inc_src = a_cs;
297 cs_dst = a_cs;
298 rs_dst = 0;
299 inc_dst = 1 * a_rs;
300 }
301 else // if ( a_rs > a_cs )
302 {
303 cs_src = 0;
304 rs_src = a_rs;
305 inc_src = 1 * a_cs;
306 cs_dst = 0;
307 rs_dst = 1 * a_cs;
308 inc_dst = a_rs;
309 }
310 }
311 else // if ( bl1_is_gen_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
312 {
313 // General stride with column-major tilt looks similar to column-major.
314 // General stride with row-major tilt looks similar to row-major.
315 if ( a_rs < a_cs )
316 {
317 cs_src = a_cs;
318 rs_src = 0;
319 inc_src = 1 * a_rs;
320 cs_dst = 1 * a_rs;
321 rs_dst = 0;
322 inc_dst = a_cs;
323 }
324 else // if ( a_rs > a_cs )
325 {
326 cs_src = 0;
327 rs_src = 1 * a_cs;
328 inc_src = a_rs;
329 cs_dst = 0;
330 rs_dst = a_rs;
331 inc_dst = 1 * a_cs;
332 }
333 }
334
335 for ( j = 0; j < n_iter; j++ )
336 {
337 a_src = a + j*cs_src + j*rs_src;
338 a_dst = a + j*cs_dst + j*rs_dst;
339
340 bl1_ccopyv( conj,
341 j,
342 a_src, inc_src,
343 a_dst, inc_dst );
344
345 if ( bl1_is_conj( conj ) )
346 {
347 a_jj = a + j*a_rs + j*a_cs;
348 a_jj->imag = bl1_s0();
349 }
350 }
351}
bl1_ccopyv
void bl1_ccopyv(conj1_t conj, int m, scomplex *x, int incx, scomplex *y, int incy)
Definition bl1_copyv.c:49
bl1_is_lower
int bl1_is_lower(uplo1_t uplo)
Definition bl1_is.c:49
bl1_zero_dim1
int bl1_zero_dim1(int m)
Definition bl1_is.c:113

References bl1_ccopyv(), bl1_is_col_storage(), bl1_is_conj(), bl1_is_gen_storage(), bl1_is_lower(), bl1_is_row_storage(), bl1_is_upper(), bl1_s0(), bl1_zero_dim1(), and scomplex::imag.

Referenced by FLA_Hermitianize(), and FLA_Symmetrize().

◆ bl1_d0()

double bl1_d0 ( void  )
119{
120 double x;
121 x = 0.0;
122 return x;
123}

Referenced by bl1_cmaxabsmr(), bl1_dgemm(), bl1_dmaxabsm(), bl1_dmaxabsmr(), bl1_drandmr(), bl1_dsymm(), bl1_z0(), bl1_z1(), bl1_z1h(), bl1_z2(), bl1_zher2k(), bl1_zherk(), bl1_zm1(), bl1_zm1h(), bl1_zm2(), bl1_zmaxabsm(), bl1_zmaxabsmr(), bl1_zsymmize(), FLA_Apply_G_rf_asd_var1(), FLA_Apply_G_rf_asd_var2(), FLA_Apply_G_rf_asd_var3(), FLA_Apply_G_rf_asd_var3b(), FLA_Apply_G_rf_asd_var6(), FLA_Apply_G_rf_asd_var6b(), FLA_Apply_G_rf_asd_var9(), FLA_Apply_G_rf_asd_var9b(), FLA_Apply_G_rf_asz_var1(), FLA_Apply_G_rf_asz_var2(), FLA_Apply_G_rf_asz_var3(), FLA_Apply_G_rf_asz_var6(), FLA_Apply_G_rf_asz_var9(), FLA_Apply_G_rf_opd_var1(), FLA_Apply_G_rf_opd_var2(), FLA_Apply_G_rf_opd_var3(), FLA_Apply_G_rf_opd_var6(), FLA_Apply_G_rf_opd_var9(), FLA_Apply_G_rf_opz_var1(), FLA_Apply_G_rf_opz_var2(), FLA_Apply_G_rf_opz_var3(), FLA_Apply_G_rf_opz_var6(), FLA_Apply_G_rf_opz_var9(), FLA_Bsvd_compute_tol_thresh_opd(), FLA_Bsvd_ext_opd_var1(), FLA_Bsvd_ext_opz_var1(), FLA_Bsvd_find_submatrix_opd(), FLA_Bsvd_v_opd_var1(), FLA_Bsvd_v_opd_var2(), FLA_Bsvd_v_opz_var1(), FLA_Bsvd_v_opz_var2(), FLA_Fused_Ahx_Ax_opd_var1(), FLA_Fused_Ahx_Axpy_Ax_opd_var1(), FLA_Fused_Gerc2_Ahx_Ax_opd_var1(), FLA_Fused_Gerc2_Ahx_Axpy_Ax_opd_var1(), FLA_Fused_Uhu_Yhu_Zhu_opd_var1(), FLA_Fused_UYx_ZVx_opd_var1(), FLA_Fused_UZhu_ZUhu_opd_var1(), FLA_Pythag2_opd(), FLA_Pythag3_opd(), FLA_Tevd_find_submatrix_opd(), FLA_Tevd_v_opd_var2(), FLA_Tevd_v_opz_var2(), and FLA_Tridiag_UT_shift_U_l_opd().

◆ bl1_d1()

double bl1_d1 ( void  )
55{
56 double x;
57 x = 1.0;
58 return x;
59}

Referenced by bl1_dgemm(), bl1_drandmr(), bl1_dsymm(), bl1_dtrmmsx(), bl1_dtrsmsx(), bl1_z1(), FLA_Apply_G_rf_asd_var1(), FLA_Apply_G_rf_asd_var2(), FLA_Apply_G_rf_asd_var3(), FLA_Apply_G_rf_asd_var3b(), FLA_Apply_G_rf_asd_var6(), FLA_Apply_G_rf_asd_var6b(), FLA_Apply_G_rf_asd_var9(), FLA_Apply_G_rf_asd_var9b(), FLA_Apply_G_rf_asz_var1(), FLA_Apply_G_rf_asz_var2(), FLA_Apply_G_rf_asz_var3(), FLA_Apply_G_rf_asz_var6(), FLA_Apply_G_rf_asz_var9(), FLA_Apply_G_rf_opd_var1(), FLA_Apply_G_rf_opd_var2(), FLA_Apply_G_rf_opd_var3(), FLA_Apply_G_rf_opd_var6(), FLA_Apply_G_rf_opd_var9(), FLA_Apply_G_rf_opz_var1(), FLA_Apply_G_rf_opz_var2(), FLA_Apply_G_rf_opz_var3(), FLA_Apply_G_rf_opz_var6(), FLA_Apply_G_rf_opz_var9(), FLA_Bsvd_francis_v_opd_var1(), FLA_Bsvd_sinval_v_opd_var1(), FLA_Bsvd_v_opd_var2(), FLA_Bsvd_v_opz_var2(), FLA_Pythag2_opd(), FLA_QR_UT_form_Q_opd_var1(), FLA_QR_UT_form_Q_ops_var1(), FLA_Tevd_compute_scaling_opd(), FLA_Tevd_n_opz_var1(), FLA_Tevd_v_opd_var2(), FLA_Tevd_v_opz_var2(), and FLA_Tridiag_UT_shift_U_l_opd().

◆ bl1_d1h()

double bl1_d1h ( void  )
87{
88 double x;
89 x = 0.5;
90 return x;
91}

Referenced by bl1_z1h().

◆ bl1_d2()

double bl1_d2 ( void  )
23{
24 double x;
25 x = 2.0;
26 return x;
27}

Referenced by bl1_z2().

◆ bl1_dallocm()

double * bl1_dallocm ( unsigned int  m,
unsigned int  n 
)
36{
37 return ( double* ) BLIS1_MALLOC( m * n * sizeof( double ) );
38}

Referenced by bl1_dcreate_contigm(), bl1_dcreate_contigmr(), bl1_dcreate_contigmt(), bl1_dgemm(), bl1_dsymm(), bl1_dsyr2k(), bl1_dtrmmsx(), and bl1_dtrsmsx().

◆ bl1_dallocv()

double * bl1_dallocv ( unsigned int  n_elem)
36{
37 return ( double* ) BLIS1_MALLOC( n_elem * sizeof( double ) );
38}

Referenced by bl1_dtrmvsx(), and bl1_dtrsvsx().

◆ bl1_dapdiagmv()

void bl1_dapdiagmv ( side1_t  side,
conj1_t  conj,
int  m,
int  n,
double *  x,
int  incx,
double *  a,
int  a_rs,
int  a_cs 
)
69{
70 double* chi;
71 double* a_begin;
72 int inca, lda;
73 int n_iter;
74 int n_elem;
75 int j;
76
77 // Return early if possible.
78 if ( bl1_zero_dim2( m, n ) ) return;
79
80 // Initialize with optimal values for column-major storage.
81 inca = a_rs;
82 lda = a_cs;
83 n_iter = n;
84 n_elem = m;
85
86 // An optimization: if A is row-major, then we can proceed as if the
87 // operation were transposed (applying the diagonal values in x from the
88 // opposite side) for increased spatial locality.
89 if ( bl1_is_row_storage( a_rs, a_cs ) )
90 {
91 bl1_swap_ints( n_iter, n_elem );
92 bl1_swap_ints( lda, inca );
93 bl1_toggle_side( side );
94 }
95
96 if ( bl1_is_left( side ) )
97 {
98 for ( j = 0; j < n_iter; j++ )
99 {
100 a_begin = a + j*lda;
101
102 bl1_dewscalv( conj,
103 n_elem,
104 x, incx,
105 a_begin, inca );
106 }
107 }
108 else
109 {
110 for ( j = 0; j < n_iter; j++ )
111 {
112 a_begin = a + j*lda;
113 chi = x + j*incx;
114
115 bl1_dscalv( conj,
116 n_elem,
117 chi,
118 a_begin, inca );
119 }
120 }
121}
bl1_dscalv
void bl1_dscalv(conj1_t conj, int n, double *alpha, double *x, int incx)
Definition bl1_scalv.c:24
bl1_dewscalv
void bl1_dewscalv(conj1_t conj, int n, double *x, int incx, double *y, int incy)
Definition bl1_ewscalv.c:28

References bl1_dewscalv(), bl1_dscalv(), bl1_is_left(), bl1_is_row_storage(), and bl1_zero_dim2().

Referenced by FLA_Apply_diag_matrix().

◆ bl1_dcreate_contigm()

void bl1_dcreate_contigm ( int  m,
int  n,
double *  a_save,
int  a_rs_save,
int  a_cs_save,
double **  a,
int *  a_rs,
int *  a_cs 
)
48{
49 int m_contig, n_contig;
50
51 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
52 {
53 // Initialize dimensions assuming no transposition needed during copy.
54 m_contig = m;
55 n_contig = n;
56
57/*
58 // Transpose the dimensions of the contiguous matrix, if requested.
59 if ( bl1_does_trans( trans_copy ) )
60 {
61 m_contig = n;
62 n_contig = m;
63 }
64*/
65
66 // Allocate temporary contiguous storage for the matrix.
67 *a = bl1_dallocm( m_contig, n_contig );
68
69 // Set the row and column strides for the temporary matrix.
70 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
71
72 // Initialize the contiguous matrix with the contents of the original.
73 bl1_dcopymt( BLIS1_NO_TRANSPOSE,
74 m_contig,
75 n_contig,
76 a_save, a_rs_save, a_cs_save,
77 *a, *a_rs, *a_cs );
78 }
79}
bl1_dcopymt
void bl1_dcopymt(trans1_t trans, int m, int n, double *a, int a_rs, int a_cs, double *b, int b_rs, int b_cs)
Definition bl1_copymt.c:148
bl1_dallocm
double * bl1_dallocm(unsigned int m, unsigned int n)
Definition bl1_allocm.c:35

References bl1_dallocm(), bl1_dcopymt(), bl1_is_gen_storage(), bl1_set_contig_strides(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_dgemm(), bl1_dgemv(), bl1_dger(), bl1_dsymm(), bl1_dtrmm(), bl1_dtrmmsx(), bl1_dtrsm(), and bl1_dtrsmsx().

◆ bl1_dcreate_contigmr()

void bl1_dcreate_contigmr ( uplo1_t  uplo,
int  m,
int  n,
double *  a_save,
int  a_rs_save,
int  a_cs_save,
double **  a,
int *  a_rs,
int *  a_cs 
)
46{
47 int m_contig, n_contig;
48
49 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
50 {
51 // Initialize dimensions assuming no transposition needed during copy.
52 m_contig = m;
53 n_contig = n;
54/*
55 // Transpose the dimensions of the contiguous matrix, if requested.
56 if ( bl1_does_trans( trans_copy ) )
57 {
58 m_contig = n;
59 n_contig = m;
60 }
61*/
62 // Allocate temporary contiguous storage for the matrix.
63 *a = bl1_dallocm( m_contig, n_contig );
64
65 // Set the row and column strides for the temporary matrix.
66 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
67
68 // Initialize the contiguous matrix with the contents of the original.
69 bl1_dcopymr( uplo,
70 m_contig,
71 n_contig,
72 a_save, a_rs_save, a_cs_save,
73 *a, *a_rs, *a_cs );
74 }
75}
bl1_dcopymr
void bl1_dcopymr(uplo1_t uplo, int m, int n, double *a, int a_rs, int a_cs, double *b, int b_rs, int b_cs)
Definition bl1_copymr.c:76

References bl1_dallocm(), bl1_dcopymr(), bl1_is_gen_storage(), and bl1_set_contig_strides().

Referenced by bl1_dcreate_contigmsr(), bl1_dsymm(), bl1_dsymv(), bl1_dsyr(), bl1_dsyr2(), bl1_dsyr2k(), bl1_dsyrk(), bl1_dtrmm(), bl1_dtrmmsx(), bl1_dtrmv(), bl1_dtrmvsx(), bl1_dtrsm(), bl1_dtrsmsx(), bl1_dtrsv(), and bl1_dtrsvsx().

◆ bl1_dcreate_contigmsr()

void bl1_dcreate_contigmsr ( side1_t  side,
uplo1_t  uplo,
int  m,
int  n,
double *  a_save,
int  a_rs_save,
int  a_cs_save,
double **  a,
int *  a_rs,
int *  a_cs 
)
30{
31 int dim_a;
32
33 // Choose the dimension of the matrix based on the side parameter.
34 if ( bl1_is_left( side ) ) dim_a = m;
35 else dim_a = n;
36
37 // Call the simple version with chosen dimensions.
38 bl1_dcreate_contigmr( uplo,
39 dim_a,
40 dim_a,
41 a_save, a_rs_save, a_cs_save,
42 a, a_rs, a_cs );
43}
bl1_dcreate_contigmr
void bl1_dcreate_contigmr(uplo1_t uplo, int m, int n, double *a_save, int a_rs_save, int a_cs_save, double **a, int *a_rs, int *a_cs)
Definition bl1_create_contigmr.c:45

References bl1_dcreate_contigmr(), and bl1_is_left().

◆ bl1_dcreate_contigmt()

void bl1_dcreate_contigmt ( trans1_t  trans_dims,
int  m,
int  n,
double *  a_save,
int  a_rs_save,
int  a_cs_save,
double **  a,
int *  a_rs,
int *  a_cs 
)
52{
53 int m_contig, n_contig;
54
55 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
56 {
57 // Transpose the dimensions if requested.
58 if ( bl1_does_trans( trans_dims ) )
59 bl1_swap_ints( m, n );
60
61 // Initialize dimensions assuming no transposition needed during copy.
62 m_contig = m;
63 n_contig = n;
64
65/*
66 // Transpose the dimensions of the contiguous matrix, if requested.
67 if ( bl1_does_trans( trans_copy ) )
68 {
69 m_contig = n;
70 n_contig = m;
71 }
72*/
73
74 // Allocate temporary contiguous storage for the matrix.
75 *a = bl1_dallocm( m_contig, n_contig );
76
77 // Set the row and column strides for the temporary matrix.
78 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
79
80 // Initialize the contiguous matrix with the contents of the original.
81 bl1_dcopymt( BLIS1_NO_TRANSPOSE,
82 m_contig,
83 n_contig,
84 a_save, a_rs_save, a_cs_save,
85 *a, *a_rs, *a_cs );
86 }
87}

References bl1_dallocm(), bl1_dcopymt(), bl1_does_trans(), bl1_is_gen_storage(), bl1_set_contig_strides(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_dgemm(), bl1_dsyr2k(), and bl1_dsyrk().

◆ bl1_dewinvscalmt()

void bl1_dewinvscalmt ( trans1_t  trans,
int  m,
int  n,
double *  a,
int  a_rs,
int  a_cs,
double *  b,
int  b_rs,
int  b_cs 
)
86{
87 double* a_begin;
88 double* b_begin;
89 int lda, inca;
90 int ldb, incb;
91 int n_iter;
92 int n_elem;
93 int j;
94 conj1_t conj;
95
96 // Return early if possible.
97 if ( bl1_zero_dim2( m, n ) ) return;
98
99 // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
100 // gets invoked only once.
101 if ( bl1_is_vector( m, n ) )
102 {
103 // Initialize with values appropriate for vectors.
104 n_iter = 1;
105 n_elem = bl1_vector_dim( m, n );
106 lda = 1; // multiplied by zero when n_iter == 1; not needed.
107 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
108 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
109 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
110 }
111 else // matrix case
112 {
113 // Initialize with optimal values for column-major storage.
114 n_iter = n;
115 n_elem = m;
116 lda = a_cs;
117 inca = a_rs;
118 ldb = b_cs;
119 incb = b_rs;
120
121 // Handle the transposition of A.
122 if ( bl1_does_trans( trans ) )
123 {
124 bl1_swap_ints( lda, inca );
125 }
126
127 // An optimization: if B is row-major and if A is effectively row-major
128 // after a possible transposition, then let's access the matrices by rows
129 // instead of by columns for increased spatial locality.
130 if ( bl1_is_row_storage( b_rs, b_cs ) )
131 {
132 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
133 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
134 {
135 bl1_swap_ints( n_iter, n_elem );
136 bl1_swap_ints( lda, inca );
137 bl1_swap_ints( ldb, incb );
138 }
139 }
140 }
141
142 // Extract conj component from trans parameter.
143 conj = bl1_proj_trans1_to_conj( trans );
144
145 for ( j = 0; j < n_iter; j++ )
146 {
147 a_begin = a + j*lda;
148 b_begin = b + j*ldb;
149
150 bl1_dewinvscalv( conj,
151 n_elem,
152 a_begin, inca,
153 b_begin, incb );
154 }
155}
bl1_dewinvscalv
void bl1_dewinvscalv(conj1_t conj, int n, double *x, int incx, double *y, int incy)
Definition bl1_ewinvscalv.c:28

References bl1_dewinvscalv(), bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Inv_scal_elemwise().

◆ bl1_dewinvscalv()

void bl1_dewinvscalv ( conj1_t  conj,
int  n,
double *  x,
int  incx,
double *  y,
int  incy 
)
29{
30 double* chi;
31 double* psi;
32 int i;
33
34 for ( i = 0; i < n; ++i )
35 {
36 chi = x + i*incx;
37 psi = y + i*incy;
38
39 bl1_dinvscals( chi, psi );
40 }
41}

References i.

Referenced by bl1_dewinvscalmt().

◆ bl1_dewscalmt()

void bl1_dewscalmt ( trans1_t  trans,
int  m,
int  n,
double *  a,
int  a_rs,
int  a_cs,
double *  b,
int  b_rs,
int  b_cs 
)
86{
87 double* a_begin;
88 double* b_begin;
89 int lda, inca;
90 int ldb, incb;
91 int n_iter;
92 int n_elem;
93 int j;
94 conj1_t conj;
95
96 // Return early if possible.
97 if ( bl1_zero_dim2( m, n ) ) return;
98
99 // Handle cases where A and B are vectors to ensure that the underlying ewscal
100 // gets invoked only once.
101 if ( bl1_is_vector( m, n ) )
102 {
103 // Initialize with values appropriate for vectors.
104 n_iter = 1;
105 n_elem = bl1_vector_dim( m, n );
106 lda = 1; // multiplied by zero when n_iter == 1; not needed.
107 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
108 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
109 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
110 }
111 else // matrix case
112 {
113 // Initialize with optimal values for column-major storage.
114 n_iter = n;
115 n_elem = m;
116 lda = a_cs;
117 inca = a_rs;
118 ldb = b_cs;
119 incb = b_rs;
120
121 // Handle the transposition of A.
122 if ( bl1_does_trans( trans ) )
123 {
124 bl1_swap_ints( lda, inca );
125 }
126
127 // An optimization: if B is row-major and if A is effectively row-major
128 // after a possible transposition, then let's access the matrices by rows
129 // instead of by columns for increased spatial locality.
130 if ( bl1_is_row_storage( b_rs, b_cs ) )
131 {
132 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
133 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
134 {
135 bl1_swap_ints( n_iter, n_elem );
136 bl1_swap_ints( lda, inca );
137 bl1_swap_ints( ldb, incb );
138 }
139 }
140 }
141
142 // Extract conj component from trans parameter.
143 conj = bl1_proj_trans1_to_conj( trans );
144
145 for ( j = 0; j < n_iter; j++ )
146 {
147 a_begin = a + j*lda;
148 b_begin = b + j*ldb;
149
150 bl1_dewscalv( conj,
151 n_elem,
152 a_begin, inca,
153 b_begin, incb );
154 }
155}

References bl1_dewscalv(), bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Scal_elemwise().

◆ bl1_dewscalv()

void bl1_dewscalv ( conj1_t  conj,
int  n,
double *  x,
int  incx,
double *  y,
int  incy 
)
29{
30 double* chi;
31 double* psi;
32 int i;
33
34 for ( i = 0; i < n; ++i )
35 {
36 chi = x + i*incx;
37 psi = y + i*incy;
38
39 bl1_dscals( chi, psi );
40 }
41}

References i.

Referenced by bl1_dapdiagmv(), and bl1_dewscalmt().

◆ bl1_dfree()

void bl1_dfree ( double *  p)
36{
37 free( ( void* ) p );
38}

Referenced by bl1_dfree_contigm(), bl1_dfree_saved_contigm(), bl1_dfree_saved_contigmr(), bl1_dfree_saved_contigmsr(), bl1_dgemm(), bl1_dsymm(), bl1_dsyr2k(), bl1_dtrmmsx(), bl1_dtrmvsx(), bl1_dtrsmsx(), and bl1_dtrsvsx().

◆ bl1_dfree_contigm()

void bl1_dfree_contigm ( double *  a_save,
int  a_rs_save,
int  a_cs_save,
double **  a,
int *  a_rs,
int *  a_cs 
)
30{
31 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
32 {
33 // Free the temporary contiguous storage for the matrix.
34 bl1_dfree( *a );
35
36 // Restore the original matrix address.
37 *a = a_save;
38
39 // Restore the original row and column strides.
40 *a_rs = a_rs_save;
41 *a_cs = a_cs_save;
42 }
43}
bl1_dfree
void bl1_dfree(double *p)
Definition bl1_free.c:35

References bl1_dfree(), and bl1_is_gen_storage().

Referenced by bl1_dgemm(), bl1_dgemv(), bl1_dsymm(), bl1_dsymv(), bl1_dsyr2k(), bl1_dsyrk(), bl1_dtrmm(), bl1_dtrmmsx(), bl1_dtrmv(), bl1_dtrmvsx(), bl1_dtrsm(), bl1_dtrsmsx(), bl1_dtrsv(), and bl1_dtrsvsx().

◆ bl1_dfree_saved_contigm()

void bl1_dfree_saved_contigm ( int  m,
int  n,
double *  a_save,
int  a_rs_save,
int  a_cs_save,
double **  a,
int *  a_rs,
int *  a_cs 
)
37{
38 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
39 {
40 // Copy the contents of the temporary matrix back to the original.
41 bl1_dcopymt( BLIS1_NO_TRANSPOSE,
42 m,
43 n,
44 *a, *a_rs, *a_cs,
45 a_save, a_rs_save, a_cs_save );
46
47 // Free the temporary contiguous storage for the matrix.
48 bl1_dfree( *a );
49
50 // Restore the original matrix address.
51 *a = a_save;
52
53 // Restore the original row and column strides.
54 *a_rs = a_rs_save;
55 *a_cs = a_cs_save;
56 }
57}

References bl1_dcopymt(), bl1_dfree(), bl1_is_gen_storage(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_dgemm(), bl1_dger(), bl1_dsymm(), bl1_dsyr(), bl1_dsyr2(), bl1_dtrmm(), bl1_dtrmmsx(), bl1_dtrsm(), and bl1_dtrsmsx().

◆ bl1_dfree_saved_contigmr()

void bl1_dfree_saved_contigmr ( uplo1_t  uplo,
int  m,
int  n,
double *  a_save,
int  a_rs_save,
int  a_cs_save,
double **  a,
int *  a_rs,
int *  a_cs 
)
37{
38 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
39 {
40 // Copy the contents of the temporary matrix back to the original.
41 bl1_dcopymr( uplo,
42 m,
43 n,
44 *a, *a_rs, *a_cs,
45 a_save, a_rs_save, a_cs_save );
46
47 // Free the temporary contiguous storage for the matrix.
48 bl1_dfree( *a );
49
50 // Restore the original matrix address.
51 *a = a_save;
52
53 // Restore the original row and column strides.
54 *a_rs = a_rs_save;
55 *a_cs = a_cs_save;
56 }
57}

References bl1_dcopymr(), bl1_dfree(), and bl1_is_gen_storage().

Referenced by bl1_dsyr2k(), and bl1_dsyrk().

◆ bl1_dfree_saved_contigmsr()

void bl1_dfree_saved_contigmsr ( side1_t  side,
uplo1_t  uplo,
int  m,
int  n,
double *  a_save,
int  a_rs_save,
int  a_cs_save,
double **  a,
int *  a_rs,
int *  a_cs 
)
43{
44 int dim_a;
45
46 // Choose the dimension of the matrix based on the side parameter.
47 if ( bl1_is_left( side ) ) dim_a = m;
48 else dim_a = n;
49
50 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
51 {
52 // Copy the contents of the temporary matrix back to the original.
53 bl1_dcopymt( uplo,
54 dim_a,
55 dim_a,
56 *a, *a_rs, *a_cs,
57 a_save, a_rs_save, a_cs_save );
58
59 // Free the temporary contiguous storage for the matrix.
60 bl1_dfree( *a );
61
62 // Restore the original matrix address.
63 *a = a_save;
64
65 // Restore the original row and column strides.
66 *a_rs = a_rs_save;
67 *a_cs = a_cs_save;
68 }
69}

References bl1_dcopymt(), bl1_dfree(), bl1_is_gen_storage(), and bl1_is_left().

◆ bl1_dident()

void bl1_dident ( int  m,
double *  a,
int  a_rs,
int  a_cs 
)
33{
34 double* alpha;
35 int i, j;
36
37 for ( j = 0; j < m; ++j )
38 {
39 for ( i = 0; i < m; ++i )
40 {
41 alpha = a + i*a_rs + j*a_cs;
42
43 *alpha = 0.0;
44
45 if ( i == j )
46 *alpha = 1.0;
47 }
48 }
49}

References i.

Referenced by FLA_Bsvd_v_opd_var2(), FLA_Bsvd_v_opz_var2(), FLA_Tevd_v_opd_var2(), FLA_Tevd_v_opz_var2(), and FLA_UDdate_UT_opd_var1().

◆ bl1_dinvert2s()

void bl1_dinvert2s ( conj1_t  conj,
double *  alpha,
double *  beta 
)
21{
22 double one = 1.0;
23
24 *beta = one / *alpha;
25}

Referenced by bl1_dinvscalm(), and bl1_zdinvscalm().

◆ bl1_dinverts()

void bl1_dinverts ( conj1_t  conj,
double *  alpha 
)
21{
22 double one = 1.0;
23
24 *alpha = one / *alpha;
25}

Referenced by FLA_Trinv_ln_opd_var1(), FLA_Trinv_ln_opd_var2(), FLA_Trinv_ln_opd_var3(), FLA_Trinv_ln_opd_var4(), FLA_Trinv_un_opd_var1(), FLA_Trinv_un_opd_var2(), FLA_Trinv_un_opd_var3(), and FLA_Trinv_un_opd_var4().

◆ bl1_dinvertv()

void bl1_dinvertv ( conj1_t  conj,
int  n,
double *  x,
int  incx 
)
28{
29 double one = 1.0;
30 double* chi;
31 int i;
32
33 for ( i = 0; i < n; ++i )
34 {
35 chi = x + i*incx;
36
37 *chi = one / *chi;
38 }
39}

References i.

Referenced by FLA_Invert().

◆ bl1_dm1()

double bl1_dm1 ( void  )
183{
184 double x;
185 x = -1.0;
186 return x;
187}

Referenced by bl1_zconjm(), bl1_zconjmr(), bl1_zconjv(), bl1_zm1(), FLA_Bsvd_ext_opd_var1(), FLA_Bsvd_ext_opz_var1(), FLA_Bsvd_v_opd_var1(), FLA_Bsvd_v_opd_var2(), FLA_Bsvd_v_opz_var1(), FLA_Bsvd_v_opz_var2(), FLA_Fused_Ahx_Axpy_Ax_opd_var1(), and FLA_Fused_Gerc2_Ahx_Axpy_Ax_opd_var1().

◆ bl1_dm1h()

double bl1_dm1h ( void  )
151{
152 double x;
153 x = -0.5;
154 return x;
155}

Referenced by bl1_zm1h().

◆ bl1_dm2()

double bl1_dm2 ( void  )
215{
216 double x;
217 x = -2.0;
218 return x;
219}

Referenced by bl1_zm2().

◆ bl1_dmaxabsm()

void bl1_dmaxabsm ( int  m,
int  n,
double *  a,
int  a_rs,
int  a_cs,
double *  maxabs 
)
59{
60 double zero = bl1_d0();
61 double* a_begin;
62 double maxabs_cand;
63 double maxabs_temp;
64 int inca, lda;
65 int n_iter;
66 int n_elem;
67 int j;
68
69 // Return early if possible.
70 if ( bl1_zero_dim2( m, n ) ) { *maxabs = zero; return; }
71
72 // Initialize with optimal values for column-major storage.
73 inca = a_rs;
74 lda = a_cs;
75 n_iter = n;
76 n_elem = m;
77
78 // An optimization: if A is row-major, then let's access the matrix by
79 // rows instead of by columns for increased spatial locality.
80 if ( bl1_is_row_storage( a_rs, a_cs ) )
81 {
82 bl1_swap_ints( n_iter, n_elem );
83 bl1_swap_ints( lda, inca );
84 }
85
86 // Initialize the maximum absolute value candidate to the first element.
87 bl1_dabsval2( a, &maxabs_cand );
88
89 for ( j = 0; j < n_iter; j++ )
90 {
91 a_begin = a + j*lda;
92
93 bl1_dmaxabsv( n_elem,
94 a_begin, inca,
95 &maxabs_temp );
96
97 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
98 }
99
100 *maxabs = maxabs_cand;
101}
bl1_dmaxabsv
void bl1_dmaxabsv(int n, double *x, int incx, double *maxabs)
Definition bl1_maxabsv.c:34

References bl1_d0(), bl1_dmaxabsv(), bl1_is_row_storage(), and bl1_zero_dim2().

Referenced by FLA_Max_abs_value().

◆ bl1_dmaxabsmr()

void bl1_dmaxabsmr ( uplo1_t  uplo,
int  m,
int  n,
double *  a,
int  a_rs,
int  a_cs,
double *  maxabs 
)
79{
80 double zero = bl1_d0();
81 double* a_begin;
82 double maxabs_cand;
83 double maxabs_temp;
84 int inca, lda;
85 int n_iter;
86 int n_elem_max;
87 int n_elem;
88 int j;
89
90 // Return early if possible.
91 if ( bl1_zero_dim2( m, n ) ) { *maxabs = zero; return; }
92
93 // Initialize with optimal values for column-major storage.
94 n_iter = n;
95 n_elem_max = m;
96 lda = a_cs;
97 inca = a_rs;
98
99 // An optimization: if A is row-major, then let's access the matrix by
100 // rows instead of by columns for increased spatial locality.
101 if ( bl1_is_row_storage( a_rs, a_cs ) )
102 {
103 bl1_swap_ints( n_iter, n_elem_max );
104 bl1_swap_ints( lda, inca );
105 bl1_toggle_uplo( uplo );
106 }
107
108 // Initialize the maximum absolute value candidate to the first element.
109 bl1_dabsval2( a, &maxabs_cand );
110
111 if ( bl1_is_upper( uplo ) )
112 {
113 for ( j = 0; j < n_iter; j++ )
114 {
115 n_elem = bl1_min( j + 1, n_elem_max );
116 a_begin = a + j*lda;
117
118 bl1_dmaxabsv( n_elem,
119 a_begin, inca,
120 &maxabs_temp );
121
122 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
123 }
124 }
125 else // if ( bl1_is_lower( uplo ) )
126 {
127 for ( j = 0; j < n_iter; j++ )
128 {
129 n_elem = bl1_max( 0, n_elem_max - j );
130 a_begin = a + j*lda + j*inca;
131
132 bl1_dmaxabsv( n_elem,
133 a_begin, inca,
134 &maxabs_temp );
135
136 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
137 }
138 }
139
140 *maxabs = maxabs_cand;
141}

References bl1_d0(), bl1_dmaxabsv(), bl1_is_row_storage(), bl1_is_upper(), and bl1_zero_dim2().

Referenced by FLA_Max_abs_value_herm().

◆ bl1_dmaxabsv()

void bl1_dmaxabsv ( int  n,
double *  x,
int  incx,
double *  maxabs 
)
35{
36 double* chi;
37 double maxabs_cand;
38 double maxabs_temp;
39 int i;
40
41 bl1_dabsval2( x, &maxabs_cand );
42
43 for ( i = 0; i < n; ++i )
44 {
45 chi = x + i*incx;
46
47 bl1_dabsval2( chi, &maxabs_temp );
48
49 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
50 }
51
52 *maxabs = maxabs_cand;
53}

References i.

Referenced by bl1_dmaxabsm(), and bl1_dmaxabsmr().

◆ bl1_drandm()

void bl1_drandm ( int  m,
int  n,
double *  a,
int  a_rs,
int  a_cs 
)
48{
49 double* a_begin;
50 int inca, lda;
51 int n_iter;
52 int n_elem;
53 int j;
54
55 // Return early if possible.
56 if ( bl1_zero_dim2( m, n ) ) return;
57
58 // Initialize with optimal values for column-major storage.
59 inca = a_rs;
60 lda = a_cs;
61 n_iter = n;
62 n_elem = m;
63
64 // An optimization: if A is row-major, then let's access the matrix by
65 // rows instead of by columns for increased spatial locality.
66 if ( bl1_is_row_storage( a_rs, a_cs ) )
67 {
68 bl1_swap_ints( n_iter, n_elem );
69 bl1_swap_ints( lda, inca );
70 }
71
72 for ( j = 0; j < n_iter; j++ )
73 {
74 a_begin = a + j*lda;
75
76 bl1_drandv( n_elem,
77 a_begin, inca );
78 }
79}
bl1_drandv
void bl1_drandv(int n, double *x, int incx)
Definition bl1_randv.c:26

References bl1_drandv(), bl1_is_row_storage(), and bl1_zero_dim2().

Referenced by FLA_Random_matrix().

◆ bl1_drandmr()

void bl1_drandmr ( uplo1_t  uplo,
diag1_t  diag,
int  m,
int  n,
double *  a,
int  a_rs,
int  a_cs 
)
135{
136 double* a_begin;
137 double* ajj;
138 double one;
139 double zero;
140 double ord;
141 int lda, inca;
142 int n_iter;
143 int n_elem_max;
144 int n_elem;
145 int j;
146
147 // Return early if possible.
148 if ( bl1_zero_dim2( m, n ) ) return;
149
150 // Initialize with optimal values for column-major storage.
151 n_iter = n;
152 n_elem_max = m;
153 lda = a_cs;
154 inca = a_rs;
155
156 // An optimization: if A is row-major, then let's access the matrix by
157 // rows instead of by columns to increase spatial locality.
158 if ( bl1_is_row_storage( a_rs, a_cs ) )
159 {
160 bl1_swap_ints( n_iter, n_elem_max );
161 bl1_swap_ints( lda, inca );
162 bl1_toggle_uplo( uplo );
163 }
164
165 // Initialize some scalars.
166 one = bl1_d1();
167 zero = bl1_d0();
168 ord = ( double ) bl1_max( m, n );
169
170 if ( bl1_is_upper( uplo ) )
171 {
172 for ( j = 0; j < n_iter; j++ )
173 {
174 n_elem = bl1_min( j, n_elem_max );
175 a_begin = a + j*lda;
176
177 // Randomize super-diagonal elements.
178 bl1_drandv( n_elem,
179 a_begin, inca );
180
181 // Normalize super-diagonal elements by order of the matrix.
182 bl1_dinvscalv( BLIS1_NO_CONJUGATE,
183 n_elem,
184 &ord,
185 a_begin, inca );
186
187 // Initialize diagonal and sub-diagonal elements only if there are
188 // elements left in the column (ie: j < n_elem_max).
189 if ( j < n_elem_max )
190 {
191 ajj = a_begin + j*inca;
192
193 // Initialize diagonal element.
194 if ( bl1_is_unit_diag( diag ) ) *ajj = one;
195 else if ( bl1_is_zero_diag( diag ) ) *ajj = zero;
196 else if ( bl1_is_nonunit_diag( diag ) )
197 {
198 // We want positive diagonal elements between 1 and 2.
199 bl1_drands( ajj );
200 bl1_dabsval2( ajj, ajj );
201 bl1_dadd3( ajj, &one, ajj );
202 }
203
204 // Initialize sub-diagonal elements to zero.
205 bl1_dsetv( n_elem_max - j - 1,
206 &zero,
207 ajj + inca, inca );
208 }
209 }
210 }
211 else // if ( bl1_is_lower( uplo ) )
212 {
213 for ( j = 0; j < n_iter; j++ )
214 {
215 n_elem = bl1_min( j, n_elem_max );
216 a_begin = a + j*lda;
217
218 // Initialize super-diagonal to zero.
219 bl1_dsetv( n_elem,
220 &zero,
221 a_begin, inca );
222
223 // Initialize diagonal and sub-diagonal elements only if there are
224 // elements left in the column (ie: j < n_elem_max).
225 if ( j < n_elem_max )
226 {
227 ajj = a_begin + j*inca;
228
229 // Initialize diagonal element.
230 if ( bl1_is_unit_diag( diag ) ) *ajj = one;
231 else if ( bl1_is_zero_diag( diag ) ) *ajj = zero;
232 else if ( bl1_is_nonunit_diag( diag ) )
233 {
234 // We want positive diagonal elements between 1 and 2.
235 bl1_drands( ajj );
236 bl1_dabsval2( ajj, ajj );
237 bl1_dadd3( ajj, &one, ajj );
238 }
239
240 // Randomize sub-diagonal elements.
241 bl1_drandv( n_elem_max - j - 1,
242 ajj + inca, inca );
243
244 // Normalize sub-diagonal elements by order of the matrix.
245 bl1_dinvscalv( BLIS1_NO_CONJUGATE,
246 n_elem_max - j - 1,
247 &ord,
248 ajj + inca, inca );
249
250 }
251 }
252 }
253}
bl1_dinvscalv
void bl1_dinvscalv(conj1_t conj, int n, double *alpha, double *x, int incx)
Definition bl1_invscalv.c:26
bl1_dsetv
void bl1_dsetv(int m, double *sigma, double *x, int incx)
Definition bl1_setv.c:39
bl1_drands
void bl1_drands(double *alpha)
Definition bl1_rands.c:18
bl1_d1
double bl1_d1(void)
Definition bl1_constants.c:54

References bl1_d0(), bl1_d1(), bl1_dinvscalv(), bl1_drands(), bl1_drandv(), bl1_dsetv(), bl1_is_nonunit_diag(), bl1_is_row_storage(), bl1_is_unit_diag(), bl1_is_upper(), bl1_is_zero_diag(), bl1_zero_dim2(), and BLIS1_NO_CONJUGATE.

Referenced by FLA_Random_tri_matrix().

◆ bl1_drands()

void bl1_drands ( double *  alpha)
19{
20 *alpha = ( ( double ) rand() / ( ( double ) RAND_MAX / 2.0 ) ) - 1.0;
21}

Referenced by bl1_drandmr(), bl1_drandv(), and bl1_zrands().

◆ bl1_drandv()

void bl1_drandv ( int  n,
double *  x,
int  incx 
)
27{
28 double* chi;
29 int i;
30
31 for ( i = 0; i < n; ++i )
32 {
33 chi = x + i*incx;
34
35 bl1_drands( chi );
36 }
37}

References bl1_drands(), and i.

Referenced by bl1_drandm(), and bl1_drandmr().

◆ bl1_dscalediag()

void bl1_dscalediag ( conj1_t  conj,
int  offset,
int  m,
int  n,
double *  sigma,
double *  a,
int  a_rs,
int  a_cs 
)
35{
36 double* alpha;
37 int i, j;
38
39 i = j = 0;
40
41 if ( offset < 0 ) i = -offset;
42 else if ( offset > 0 ) j = offset;
43
44 while ( i < m && j < n )
45 {
46 alpha = a + i*a_rs + j*a_cs;
47
48 *alpha *= *sigma;
49
50 ++i;
51 ++j;
52 }
53}

References i.

Referenced by FLA_Scale_diag(), and FLA_UDdate_UT_opd_var1().

◆ bl1_dsetdiag()

void bl1_dsetdiag ( int  offset,
int  m,
int  n,
double *  sigma,
double *  a,
int  a_rs,
int  a_cs 
)
56{
57 double* alpha;
58 int i, j;
59
60 i = j = 0;
61
62 if ( offset < 0 ) i = -offset;
63 else if ( offset > 0 ) j = offset;
64
65 while ( i < m && j < n )
66 {
67 alpha = a + i*a_rs + j*a_cs;
68
69 *alpha = *sigma;
70
71 ++i;
72 ++j;
73 }
74}

References i.

Referenced by FLA_Set_diag(), FLA_Set_offdiag(), and FLA_Triangularize().

◆ bl1_dsetm()

void bl1_dsetm ( int  m,
int  n,
double *  sigma,
double *  a,
int  a_rs,
int  a_cs 
)
46{
47 double* alpha;
48 int i, j;
49
50 for ( j = 0; j < n; ++j )
51 {
52 for ( i = 0; i < m; ++i )
53 {
54 alpha = a + i*a_rs + j*a_cs;
55
56 *alpha = *sigma;
57 }
58 }
59}

References i.

Referenced by FLA_Bidiag_UT_u_step_ofd_var4(), FLA_Bidiag_UT_u_step_opd_var4(), FLA_Bidiag_UT_u_step_opd_var5(), FLA_Hess_UT_step_ofd_var4(), FLA_Hess_UT_step_opd_var4(), FLA_Hess_UT_step_opd_var5(), FLA_Set(), FLA_Tridiag_UT_l_step_ofd_var3(), and FLA_Tridiag_UT_l_step_opd_var3().

◆ bl1_dsetmr()

void bl1_dsetmr ( uplo1_t  uplo,
int  m,
int  n,
double *  sigma,
double *  a,
int  a_rs,
int  a_cs 
)
67{
68 double* a_begin;
69 int lda, inca;
70 int n_iter;
71 int n_elem_max;
72 int n_elem;
73 int j;
74
75 // Return early if possible.
76 if ( bl1_zero_dim2( m, n ) ) return;
77
78 // Initialize with optimal values for column-major storage.
79 n_iter = n;
80 n_elem_max = m;
81 lda = a_cs;
82 inca = a_rs;
83
84 // An optimization: if A is row-major, then let's access the matrix by
85 // rows instead of by columns to increase spatial locality.
86 if ( bl1_is_row_storage( a_rs, a_cs ) )
87 {
88 bl1_swap_ints( n_iter, n_elem_max );
89 bl1_swap_ints( lda, inca );
90 bl1_toggle_uplo( uplo );
91 }
92
93 if ( bl1_is_upper( uplo ) )
94 {
95 for ( j = 0; j < n_iter; j++ )
96 {
97 n_elem = bl1_min( j, n_elem_max );
98 a_begin = a + j*lda;
99
100 bl1_dsetv( n_elem,
101 sigma,
102 a_begin, inca );
103 }
104 }
105 else // if ( bl1_is_lower( uplo ) )
106 {
107 for ( j = 0; j < n_iter; j++ )
108 {
109 n_elem = bl1_max( 0, n_elem_max - j - 1 );
110 a_begin = a + j*lda + (j + 1)*inca;
111
112 bl1_dsetv( n_elem,
113 sigma,
114 a_begin, inca );
115 }
116 }
117}

References bl1_dsetv(), bl1_is_row_storage(), bl1_is_upper(), and bl1_zero_dim2().

Referenced by FLA_Setr(), and FLA_Triangularize().

◆ bl1_dsetv()

void bl1_dsetv ( int  m,
double *  sigma,
double *  x,
int  incx 
)
40{
41 double* chi;
42 int i;
43
44 for ( i = 0; i < n; ++i )
45 {
46 chi = x + i*incx;
47
48 *chi = *sigma;
49 }
50}

References i.

Referenced by bl1_drandmr(), bl1_dsetmr(), FLA_Bidiag_UT_l_realify_opt(), FLA_Bidiag_UT_realify_diagonals_opt(), FLA_Bidiag_UT_u_realify_opt(), FLA_Bidiag_UT_u_step_ofd_var4(), FLA_Bidiag_UT_u_step_opd_var4(), FLA_Fused_Ahx_Ax_opd_var1(), FLA_Fused_Ahx_Axpy_Ax_opd_var1(), FLA_Fused_Gerc2_Ahx_Ax_opd_var1(), FLA_Fused_Gerc2_Ahx_Axpy_Ax_opd_var1(), FLA_Fused_Her2_Ax_l_opd_var1(), FLA_Obj_extract_imag_part(), FLA_Tridiag_UT_l_realify_opt(), FLA_Tridiag_UT_realify_subdiagonal_opt(), FLA_Tridiag_UT_shift_U_l_opd(), and FLA_Tridiag_UT_u_realify_opt().

◆ bl1_dshiftdiag()

void bl1_dshiftdiag ( conj1_t  conj,
int  offset,
int  m,
int  n,
double *  sigma,
double *  a,
int  a_rs,
int  a_cs 
)
35{
36 double* alpha;
37 int i, j;
38
39 i = j = 0;
40
41 if ( offset < 0 ) i = -offset;
42 else if ( offset > 0 ) j = offset;
43
44 while ( i < m && j < n )
45 {
46 alpha = a + i*a_rs + j*a_cs;
47
48 *alpha += *sigma;
49
50 ++i;
51 ++j;
52 }
53}

References i.

Referenced by FLA_Lyap_h_opd_var1(), FLA_Lyap_h_opd_var2(), FLA_Lyap_h_opd_var3(), FLA_Lyap_h_opd_var4(), FLA_Lyap_n_opd_var1(), FLA_Lyap_n_opd_var2(), FLA_Lyap_n_opd_var3(), FLA_Lyap_n_opd_var4(), and FLA_Shift_diag().

◆ bl1_dsymmize()

void bl1_dsymmize ( conj1_t  conj,
uplo1_t  uplo,
int  m,
double *  a,
int  a_rs,
int  a_cs 
)
125{
126 double* a_src;
127 double* a_dst;
128 int rs_src, cs_src, inc_src;
129 int rs_dst, cs_dst, inc_dst;
130 int n_iter;
131 int j;
132
133 // Return early if possible.
134 if ( bl1_zero_dim1( m ) ) return;
135
136 // Assume A is square.
137 n_iter = m;
138
139 // Initialize with appropriate values based on storage.
140 if ( bl1_is_col_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
141 {
142 cs_src = 1;
143 rs_src = 0;
144 inc_src = a_cs;
145 cs_dst = a_cs;
146 rs_dst = 0;
147 inc_dst = 1;
148 }
149 else if ( bl1_is_col_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
150 {
151 cs_src = a_cs;
152 rs_src = 0;
153 inc_src = 1;
154 cs_dst = 1;
155 rs_dst = 0;
156 inc_dst = a_cs;
157 }
158 else if ( bl1_is_row_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
159 {
160 cs_src = 0;
161 rs_src = a_rs;
162 inc_src = 1;
163 cs_dst = 0;
164 rs_dst = 1;
165 inc_dst = a_rs;
166 }
167 else if ( bl1_is_row_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
168 {
169 cs_src = 0;
170 rs_src = 1;
171 inc_src = a_rs;
172 cs_dst = 0;
173 rs_dst = a_rs;
174 inc_dst = 1;
175 }
176 else if ( bl1_is_gen_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
177 {
178 // General stride with column-major tilt looks similar to column-major.
179 // General stride with row-major tilt looks similar to row-major.
180 if ( a_rs < a_cs )
181 {
182 cs_src = 1 * a_rs;
183 rs_src = 0;
184 inc_src = a_cs;
185 cs_dst = a_cs;
186 rs_dst = 0;
187 inc_dst = 1 * a_rs;
188 }
189 else // if ( a_rs > a_cs )
190 {
191 cs_src = 0;
192 rs_src = a_rs;
193 inc_src = 1 * a_cs;
194 cs_dst = 0;
195 rs_dst = 1 * a_cs;
196 inc_dst = a_rs;
197 }
198 }
199 else // if ( bl1_is_gen_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
200 {
201 // General stride with column-major tilt looks similar to column-major.
202 // General stride with row-major tilt looks similar to row-major.
203 if ( a_rs < a_cs )
204 {
205 cs_src = a_cs;
206 rs_src = 0;
207 inc_src = 1 * a_rs;
208 cs_dst = 1 * a_rs;
209 rs_dst = 0;
210 inc_dst = a_cs;
211 }
212 else // if ( a_rs > a_cs )
213 {
214 cs_src = 0;
215 rs_src = 1 * a_cs;
216 inc_src = a_rs;
217 cs_dst = 0;
218 rs_dst = a_rs;
219 inc_dst = 1 * a_cs;
220 }
221 }
222
223 for ( j = 0; j < n_iter; j++ )
224 {
225 a_src = a + j*cs_src + j*rs_src;
226 a_dst = a + j*cs_dst + j*rs_dst;
227
228 bl1_dcopyv( conj,
229 j,
230 a_src, inc_src,
231 a_dst, inc_dst );
232 }
233}
bl1_dcopyv
void bl1_dcopyv(conj1_t conj, int m, double *x, int incx, double *y, int incy)
Definition bl1_copyv.c:42

References bl1_dcopyv(), bl1_is_col_storage(), bl1_is_gen_storage(), bl1_is_lower(), bl1_is_row_storage(), bl1_is_upper(), and bl1_zero_dim1().

Referenced by FLA_Hermitianize(), and FLA_Symmetrize().

◆ bl1_iallocm()

int * bl1_iallocm ( unsigned int  m,
unsigned int  n 
)
26{
27 return ( int* ) BLIS1_MALLOC( m * n * sizeof( int ) );
28}

◆ bl1_iallocv()

int * bl1_iallocv ( unsigned int  n_elem)
26{
27 return ( int* ) BLIS1_MALLOC( n_elem * sizeof( int ) );
28}

◆ bl1_ifree()

void bl1_ifree ( int *  p)
26{
27 free( ( int* ) p );
28}

◆ bl1_isetdiag()

void bl1_isetdiag ( int  offset,
int  m,
int  n,
int *  sigma,
int *  a,
int  a_rs,
int  a_cs 
)
14{
15 int* alpha;
16 int i, j;
17
18 i = j = 0;
19
20 if ( offset < 0 ) i = -offset;
21 else if ( offset > 0 ) j = offset;
22
23 while ( i < m && j < n )
24 {
25 alpha = a + i*a_rs + j*a_cs;
26
27 *alpha = *sigma;
28
29 ++i;
30 ++j;
31 }
32}

References i.

Referenced by FLA_Set_diag(), and FLA_Set_offdiag().

◆ bl1_isetm()

void bl1_isetm ( int  m,
int  n,
int *  sigma,
int *  a,
int  a_rs,
int  a_cs 
)
14{
15 int* alpha;
16 int i, j;
17
18 for ( j = 0; j < n; ++j )
19 {
20 for ( i = 0; i < m; ++i )
21 {
22 alpha = a + i*a_rs + j*a_cs;
23
24 *alpha = *sigma;
25 }
26 }
27}

References i.

Referenced by FLA_Set().

◆ bl1_isetv()

void bl1_isetv ( int  m,
int *  sigma,
int *  x,
int  incx 
)
14{
15 int* chi;
16 int i;
17
18 for ( i = 0; i < n; ++i )
19 {
20 chi = x + i*incx;
21
22 *chi = *sigma;
23 }
24}

References i.

◆ bl1_s0()

float bl1_s0 ( void  )
112{
113 float x;
114 x = 0.0F;
115 return x;
116}

Referenced by bl1_c0(), bl1_c1(), bl1_c1h(), bl1_c2(), bl1_cher2k(), bl1_cherk(), bl1_cm1(), bl1_cm1h(), bl1_cm2(), bl1_cmaxabsm(), bl1_csymmize(), bl1_sgemm(), bl1_smaxabsm(), bl1_smaxabsmr(), bl1_srandmr(), bl1_ssymm(), FLA_Apply_G_rf_asc_var1(), FLA_Apply_G_rf_asc_var2(), FLA_Apply_G_rf_asc_var3(), FLA_Apply_G_rf_asc_var6(), FLA_Apply_G_rf_asc_var9(), FLA_Apply_G_rf_ass_var1(), FLA_Apply_G_rf_ass_var2(), FLA_Apply_G_rf_ass_var3(), FLA_Apply_G_rf_ass_var6(), FLA_Apply_G_rf_ass_var9(), FLA_Apply_G_rf_opc_var1(), FLA_Apply_G_rf_opc_var2(), FLA_Apply_G_rf_opc_var3(), FLA_Apply_G_rf_opc_var6(), FLA_Apply_G_rf_opc_var9(), FLA_Apply_G_rf_ops_var1(), FLA_Apply_G_rf_ops_var2(), FLA_Apply_G_rf_ops_var3(), FLA_Apply_G_rf_ops_var6(), FLA_Apply_G_rf_ops_var9(), FLA_Bsvd_compute_tol_thresh_ops(), FLA_Bsvd_ext_opc_var1(), FLA_Bsvd_ext_ops_var1(), FLA_Bsvd_find_submatrix_ops(), FLA_Bsvd_v_opc_var1(), FLA_Bsvd_v_ops_var1(), FLA_Pythag2_ops(), FLA_Pythag3_ops(), and FLA_Tridiag_UT_shift_U_l_ops().

◆ bl1_s1()

float bl1_s1 ( void  )
48{
49 float x;
50 x = 1.0F;
51 return x;
52}

Referenced by bl1_c1(), bl1_sgemm(), bl1_srandmr(), bl1_ssymm(), bl1_strmmsx(), bl1_strsmsx(), FLA_Apply_G_rf_asc_var1(), FLA_Apply_G_rf_asc_var2(), FLA_Apply_G_rf_asc_var3(), FLA_Apply_G_rf_asc_var6(), FLA_Apply_G_rf_asc_var9(), FLA_Apply_G_rf_ass_var1(), FLA_Apply_G_rf_ass_var2(), FLA_Apply_G_rf_ass_var3(), FLA_Apply_G_rf_ass_var6(), FLA_Apply_G_rf_ass_var9(), FLA_Apply_G_rf_opc_var1(), FLA_Apply_G_rf_opc_var2(), FLA_Apply_G_rf_opc_var3(), FLA_Apply_G_rf_opc_var6(), FLA_Apply_G_rf_opc_var9(), FLA_Apply_G_rf_ops_var1(), FLA_Apply_G_rf_ops_var2(), FLA_Apply_G_rf_ops_var3(), FLA_Apply_G_rf_ops_var6(), FLA_Apply_G_rf_ops_var9(), FLA_Bsvd_francis_v_ops_var1(), FLA_Bsvd_sinval_v_ops_var1(), FLA_Pythag2_ops(), FLA_Tevd_compute_scaling_ops(), and FLA_Tridiag_UT_shift_U_l_ops().

◆ bl1_s1h()

float bl1_s1h ( void  )
80{
81 float x;
82 x = 0.5F;
83 return x;
84}

Referenced by bl1_c1h().

◆ bl1_s2()

float bl1_s2 ( void  )
16{
17 float x;
18 x = 2.0F;
19 return x;
20}

Referenced by bl1_c2().

◆ bl1_sallocm()

float * bl1_sallocm ( unsigned int  m,
unsigned int  n 
)
31{
32 return ( float* ) BLIS1_MALLOC( m * n * sizeof( float ) );
33}

Referenced by bl1_screate_contigm(), bl1_screate_contigmr(), bl1_screate_contigmt(), bl1_sgemm(), bl1_ssymm(), bl1_ssyr2k(), bl1_strmmsx(), and bl1_strsmsx().

◆ bl1_sallocv()

float * bl1_sallocv ( unsigned int  n_elem)
31{
32 return ( float* ) BLIS1_MALLOC( n_elem * sizeof( float ) );
33}

Referenced by bl1_strmvsx(), and bl1_strsvsx().

◆ bl1_sapdiagmv()

void bl1_sapdiagmv ( side1_t  side,
conj1_t  conj,
int  m,
int  n,
float *  x,
int  incx,
float *  a,
int  a_rs,
int  a_cs 
)
14{
15 float* chi;
16 float* a_begin;
17 int inca, lda;
18 int n_iter;
19 int n_elem;
20 int j;
21
22 // Return early if possible.
23 if ( bl1_zero_dim2( m, n ) ) return;
24
25 // Initialize with optimal values for column-major storage.
26 inca = a_rs;
27 lda = a_cs;
28 n_iter = n;
29 n_elem = m;
30
31 // An optimization: if A is row-major, then we can proceed as if the
32 // operation were transposed (applying the diagonal values in x from the
33 // opposite side) for increased spatial locality.
34 if ( bl1_is_row_storage( a_rs, a_cs ) )
35 {
36 bl1_swap_ints( n_iter, n_elem );
37 bl1_swap_ints( lda, inca );
38 bl1_toggle_side( side );
39 }
40
41 if ( bl1_is_left( side ) )
42 {
43 for ( j = 0; j < n_iter; j++ )
44 {
45 a_begin = a + j*lda;
46
47 bl1_sewscalv( conj,
48 n_elem,
49 x, incx,
50 a_begin, inca );
51 }
52 }
53 else
54 {
55 for ( j = 0; j < n_iter; j++ )
56 {
57 a_begin = a + j*lda;
58 chi = x + j*incx;
59
60 bl1_sscalv( conj,
61 n_elem,
62 chi,
63 a_begin, inca );
64 }
65 }
66}
bl1_sscalv
void bl1_sscalv(conj1_t conj, int n, float *alpha, float *x, int incx)
Definition bl1_scalv.c:13
bl1_sewscalv
void bl1_sewscalv(conj1_t conj, int n, float *x, int incx, float *y, int incy)
Definition bl1_ewscalv.c:13

References bl1_is_left(), bl1_is_row_storage(), bl1_sewscalv(), bl1_sscalv(), and bl1_zero_dim2().

Referenced by FLA_Apply_diag_matrix().

◆ bl1_screate_contigm()

void bl1_screate_contigm ( int  m,
int  n,
float *  a_save,
int  a_rs_save,
int  a_cs_save,
float **  a,
int *  a_rs,
int *  a_cs 
)
14{
15 int m_contig, n_contig;
16
17 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
18 {
19 // Initialize dimensions assuming no transposition needed during copy.
20 m_contig = m;
21 n_contig = n;
22
23/*
24 // Transpose the dimensions of the contiguous matrix, if requested.
25 if ( bl1_does_trans( trans_copy ) )
26 {
27 m_contig = n;
28 n_contig = m;
29 }
30*/
31
32 // Allocate temporary contiguous storage for the matrix.
33 *a = bl1_sallocm( m_contig, n_contig );
34
35 // Set the row and column strides for the temporary matrix.
36 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
37
38 // Initialize the contiguous matrix with the contents of the original.
39 bl1_scopymt( BLIS1_NO_TRANSPOSE,
40 m_contig,
41 n_contig,
42 a_save, a_rs_save, a_cs_save,
43 *a, *a_rs, *a_cs );
44 }
45}
bl1_scopymt
void bl1_scopymt(trans1_t trans, int m, int n, float *a, int a_rs, int a_cs, float *b, int b_rs, int b_cs)
Definition bl1_copymt.c:81
bl1_sallocm
float * bl1_sallocm(unsigned int m, unsigned int n)
Definition bl1_allocm.c:30

References bl1_is_gen_storage(), bl1_sallocm(), bl1_scopymt(), bl1_set_contig_strides(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_sgemm(), bl1_sgemv(), bl1_sger(), bl1_ssymm(), bl1_strmm(), bl1_strmmsx(), bl1_strsm(), and bl1_strsmsx().

◆ bl1_screate_contigmr()

void bl1_screate_contigmr ( uplo1_t  uplo,
int  m,
int  n,
float *  a_save,
int  a_rs_save,
int  a_cs_save,
float **  a,
int *  a_rs,
int *  a_cs 
)
14{
15 int m_contig, n_contig;
16
17 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
18 {
19 // Initialize dimensions assuming no transposition needed during copy.
20 m_contig = m;
21 n_contig = n;
22/*
23 // Transpose the dimensions of the contiguous matrix, if requested.
24 if ( bl1_does_trans( trans_copy ) )
25 {
26 m_contig = n;
27 n_contig = m;
28 }
29*/
30 // Allocate temporary contiguous storage for the matrix.
31 *a = bl1_sallocm( m_contig, n_contig );
32
33 // Set the row and column strides for the temporary matrix.
34 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
35
36 // Initialize the contiguous matrix with the contents of the original.
37 bl1_scopymr( uplo,
38 m_contig,
39 n_contig,
40 a_save, a_rs_save, a_cs_save,
41 *a, *a_rs, *a_cs );
42 }
43}
bl1_scopymr
void bl1_scopymr(uplo1_t uplo, int m, int n, float *a, int a_rs, int a_cs, float *b, int b_rs, int b_cs)
Definition bl1_copymr.c:13

References bl1_is_gen_storage(), bl1_sallocm(), bl1_scopymr(), and bl1_set_contig_strides().

Referenced by bl1_screate_contigmsr(), bl1_ssymm(), bl1_ssymv(), bl1_ssyr(), bl1_ssyr2(), bl1_ssyr2k(), bl1_ssyrk(), bl1_strmm(), bl1_strmmsx(), bl1_strmv(), bl1_strmvsx(), bl1_strsm(), bl1_strsmsx(), bl1_strsv(), and bl1_strsvsx().

◆ bl1_screate_contigmsr()

void bl1_screate_contigmsr ( side1_t  side,
uplo1_t  uplo,
int  m,
int  n,
float *  a_save,
int  a_rs_save,
int  a_cs_save,
float **  a,
int *  a_rs,
int *  a_cs 
)
14{
15 int dim_a;
16
17 // Choose the dimension of the matrix based on the side parameter.
18 if ( bl1_is_left( side ) ) dim_a = m;
19 else dim_a = n;
20
21 // Call the simple version with chosen dimensions.
22 bl1_screate_contigmr( uplo,
23 dim_a,
24 dim_a,
25 a_save, a_rs_save, a_cs_save,
26 a, a_rs, a_cs );
27}
bl1_screate_contigmr
void bl1_screate_contigmr(uplo1_t uplo, int m, int n, float *a_save, int a_rs_save, int a_cs_save, float **a, int *a_rs, int *a_cs)
Definition bl1_create_contigmr.c:13

References bl1_is_left(), and bl1_screate_contigmr().

◆ bl1_screate_contigmt()

void bl1_screate_contigmt ( trans1_t  trans_dims,
int  m,
int  n,
float *  a_save,
int  a_rs_save,
int  a_cs_save,
float **  a,
int *  a_rs,
int *  a_cs 
)
14{
15 int m_contig, n_contig;
16
17 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
18 {
19 // Transpose the dimensions if requested.
20 if ( bl1_does_trans( trans_dims ) )
21 bl1_swap_ints( m, n );
22
23 // Initialize dimensions assuming no transposition needed during copy.
24 m_contig = m;
25 n_contig = n;
26
27/*
28 // Transpose the dimensions of the contiguous matrix, if requested.
29 if ( bl1_does_trans( trans_copy ) )
30 {
31 m_contig = n;
32 n_contig = m;
33 }
34*/
35
36 // Allocate temporary contiguous storage for the matrix.
37 *a = bl1_sallocm( m_contig, n_contig );
38
39 // Set the row and column strides for the temporary matrix.
40 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
41
42 // Initialize the contiguous matrix with the contents of the original.
43 bl1_scopymt( BLIS1_NO_TRANSPOSE,
44 m_contig,
45 n_contig,
46 a_save, a_rs_save, a_cs_save,
47 *a, *a_rs, *a_cs );
48 }
49}

References bl1_does_trans(), bl1_is_gen_storage(), bl1_sallocm(), bl1_scopymt(), bl1_set_contig_strides(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_sgemm(), bl1_ssyr2k(), and bl1_ssyrk().

◆ bl1_set_contig_strides()

void bl1_set_contig_strides ( int  m,
int  n,
int *  rs,
int *  cs 
)
14{
15 // Default to column-major order.
16 *rs = 1;
17 *cs = m;
18
19 // Handle special cases first.
20 // Check the strides, and modify them if needed.
21 if ( *rs == 1 && *cs == 1 )
22 {
23 // If both strides are unit, we are probably trying to create a
24 // 1-by-n matrix in column-major order, or an m-by-1 matrix in
25 // row-major order. We have decided to "reserve" the case where
26 // rs == cs == 1 for scalars only, as having unit strides can
27 // upset the BLAS error checking when attempting to induce a
28 // row-major operation.
29 if ( m > 1 && n == 1 )
30 {
31 // Set the column stride to indicate that this is an m-by-1
32 // matrix (or vector) stored in column-major order. This is
33 // necessary because, in some cases, we have to satisfy error
34 // checking in the underlying BLAS library, which expects the
35 // leading dimension to be set to at least m, even if it will
36 // never be used for indexing since there is only one column
37 // of data. Note that rs is already set to 1.
38 *cs = m;
39 }
40 else if ( m == 1 && 1 < n )
41 {
42 // Set the row stride to indicate that this is a 1-by-n matrix
43 // stored in row-major order. Note that cs is already set to 1.
44 *rs = n;
45 }
46 else
47 {
48 // If m == n == 1, then we are dealing with a scalar. Since rs
49 // and cs do not exceed m and n, we don't have to do anything.
50 }
51 }
52}

Referenced by bl1_ccreate_contigm(), bl1_ccreate_contigmr(), bl1_ccreate_contigmt(), bl1_dcreate_contigm(), bl1_dcreate_contigmr(), bl1_dcreate_contigmt(), bl1_screate_contigm(), bl1_screate_contigmr(), bl1_screate_contigmt(), bl1_zcreate_contigm(), bl1_zcreate_contigmr(), and bl1_zcreate_contigmt().

◆ bl1_set_dim_with_side()

void bl1_set_dim_with_side ( side1_t  side,
int  m,
int  n,
int *  dim_new 
)
28{
29 if ( bl1_is_left( side ) )
30 {
31 *dim_new = m;
32 }
33 else // if ( bl1_is_right( side ) )
34 {
35 *dim_new = n;
36 }
37}

References bl1_is_left().

Referenced by bl1_chemm(), bl1_csymm(), bl1_ctrmm(), bl1_ctrmmsx(), bl1_ctrsm(), bl1_ctrsmsx(), bl1_dsymm(), bl1_dtrmm(), bl1_dtrmmsx(), bl1_dtrsm(), bl1_dtrsmsx(), bl1_ssymm(), bl1_strmm(), bl1_strmmsx(), bl1_strsm(), bl1_strsmsx(), bl1_zhemm(), bl1_zsymm(), bl1_ztrmm(), bl1_ztrmmsx(), bl1_ztrsm(), and bl1_ztrsmsx().

◆ bl1_set_dims_with_trans()

void bl1_set_dims_with_trans ( trans1_t  trans,
int  m,
int  n,
int *  m_new,
int *  n_new 
)
14{
15 if ( bl1_does_trans( trans ) )
16 {
17 *m_new = n;
18 *n_new = m;
19 }
20 else
21 {
22 *m_new = m;
23 *n_new = n;
24 }
25}

References bl1_does_trans().

Referenced by bl1_cher2k(), bl1_csyr2k(), bl1_dsyr2k(), bl1_ssyr2k(), bl1_zher2k(), and bl1_zsyr2k().

◆ bl1_sewinvscalmt()

void bl1_sewinvscalmt ( trans1_t  trans,
int  m,
int  n,
float *  a,
int  a_rs,
int  a_cs,
float *  b,
int  b_rs,
int  b_cs 
)
14{
15 float* a_begin;
16 float* b_begin;
17 int lda, inca;
18 int ldb, incb;
19 int n_iter;
20 int n_elem;
21 int j;
22 conj1_t conj;
23
24 // Return early if possible.
25 if ( bl1_zero_dim2( m, n ) ) return;
26
27 // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
28 // gets invoked only once.
29 if ( bl1_is_vector( m, n ) )
30 {
31 // Initialize with values appropriate for vectors.
32 n_iter = 1;
33 n_elem = bl1_vector_dim( m, n );
34 lda = 1; // multiplied by zero when n_iter == 1; not needed.
35 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
36 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
37 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
38 }
39 else // matrix case
40 {
41 // Initialize with optimal values for column-major storage.
42 n_iter = n;
43 n_elem = m;
44 lda = a_cs;
45 inca = a_rs;
46 ldb = b_cs;
47 incb = b_rs;
48
49 // Handle the transposition of A.
50 if ( bl1_does_trans( trans ) )
51 {
52 bl1_swap_ints( lda, inca );
53 }
54
55 // An optimization: if B is row-major and if A is effectively row-major
56 // after a possible transposition, then let's access the matrices by rows
57 // instead of by columns for increased spatial locality.
58 if ( bl1_is_row_storage( b_rs, b_cs ) )
59 {
60 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
61 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
62 {
63 bl1_swap_ints( n_iter, n_elem );
64 bl1_swap_ints( lda, inca );
65 bl1_swap_ints( ldb, incb );
66 }
67 }
68 }
69
70 // Extract conj component from trans parameter.
71 conj = bl1_proj_trans1_to_conj( trans );
72
73 for ( j = 0; j < n_iter; j++ )
74 {
75 a_begin = a + j*lda;
76 b_begin = b + j*ldb;
77
78 bl1_sewinvscalv( conj,
79 n_elem,
80 a_begin, inca,
81 b_begin, incb );
82 }
83}
bl1_sewinvscalv
void bl1_sewinvscalv(conj1_t conj, int n, float *x, int incx, float *y, int incy)
Definition bl1_ewinvscalv.c:13

References bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_sewinvscalv(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Inv_scal_elemwise().

◆ bl1_sewinvscalv()

void bl1_sewinvscalv ( conj1_t  conj,
int  n,
float *  x,
int  incx,
float *  y,
int  incy 
)
14{
15 float* chi;
16 float* psi;
17 int i;
18
19 for ( i = 0; i < n; ++i )
20 {
21 chi = x + i*incx;
22 psi = y + i*incy;
23
24 bl1_sinvscals( chi, psi );
25 }
26}

References i.

Referenced by bl1_sewinvscalmt().

◆ bl1_sewscalmt()

void bl1_sewscalmt ( trans1_t  trans,
int  m,
int  n,
float *  a,
int  a_rs,
int  a_cs,
float *  b,
int  b_rs,
int  b_cs 
)
14{
15 float* a_begin;
16 float* b_begin;
17 int lda, inca;
18 int ldb, incb;
19 int n_iter;
20 int n_elem;
21 int j;
22 conj1_t conj;
23
24 // Return early if possible.
25 if ( bl1_zero_dim2( m, n ) ) return;
26
27 // Handle cases where A and B are vectors to ensure that the underlying ewscal
28 // gets invoked only once.
29 if ( bl1_is_vector( m, n ) )
30 {
31 // Initialize with values appropriate for vectors.
32 n_iter = 1;
33 n_elem = bl1_vector_dim( m, n );
34 lda = 1; // multiplied by zero when n_iter == 1; not needed.
35 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
36 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
37 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
38 }
39 else // matrix case
40 {
41 // Initialize with optimal values for column-major storage.
42 n_iter = n;
43 n_elem = m;
44 lda = a_cs;
45 inca = a_rs;
46 ldb = b_cs;
47 incb = b_rs;
48
49 // Handle the transposition of A.
50 if ( bl1_does_trans( trans ) )
51 {
52 bl1_swap_ints( lda, inca );
53 }
54
55 // An optimization: if B is row-major and if A is effectively row-major
56 // after a possible transposition, then let's access the matrices by rows
57 // instead of by columns for increased spatial locality.
58 if ( bl1_is_row_storage( b_rs, b_cs ) )
59 {
60 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
61 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
62 {
63 bl1_swap_ints( n_iter, n_elem );
64 bl1_swap_ints( lda, inca );
65 bl1_swap_ints( ldb, incb );
66 }
67 }
68 }
69
70 // Extract conj component from trans parameter.
71 conj = bl1_proj_trans1_to_conj( trans );
72
73 for ( j = 0; j < n_iter; j++ )
74 {
75 a_begin = a + j*lda;
76 b_begin = b + j*ldb;
77
78 bl1_sewscalv( conj,
79 n_elem,
80 a_begin, inca,
81 b_begin, incb );
82 }
83}

References bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_sewscalv(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Scal_elemwise().

◆ bl1_sewscalv()

void bl1_sewscalv ( conj1_t  conj,
int  n,
float *  x,
int  incx,
float *  y,
int  incy 
)
14{
15 float* chi;
16 float* psi;
17 int i;
18
19 for ( i = 0; i < n; ++i )
20 {
21 chi = x + i*incx;
22 psi = y + i*incy;
23
24 bl1_sscals( chi, psi );
25 }
26}

References i.

Referenced by bl1_sapdiagmv(), and bl1_sewscalmt().

◆ bl1_sfree()

void bl1_sfree ( float *  p)
31{
32 free( ( void* ) p );
33}

Referenced by bl1_sfree_contigm(), bl1_sfree_saved_contigm(), bl1_sfree_saved_contigmr(), bl1_sfree_saved_contigmsr(), bl1_sgemm(), bl1_ssymm(), bl1_ssyr2k(), bl1_strmmsx(), bl1_strmvsx(), bl1_strsmsx(), and bl1_strsvsx().

◆ bl1_sfree_contigm()

void bl1_sfree_contigm ( float *  a_save,
int  a_rs_save,
int  a_cs_save,
float **  a,
int *  a_rs,
int *  a_cs 
)
14{
15 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
16 {
17 // Free the temporary contiguous storage for the matrix.
18 bl1_sfree( *a );
19
20 // Restore the original matrix address.
21 *a = a_save;
22
23 // Restore the original row and column strides.
24 *a_rs = a_rs_save;
25 *a_cs = a_cs_save;
26 }
27}
bl1_sfree
void bl1_sfree(float *p)
Definition bl1_free.c:30

References bl1_is_gen_storage(), and bl1_sfree().

Referenced by bl1_sgemm(), bl1_sgemv(), bl1_ssymm(), bl1_ssymv(), bl1_ssyr2k(), bl1_ssyrk(), bl1_strmm(), bl1_strmmsx(), bl1_strmv(), bl1_strmvsx(), bl1_strsm(), bl1_strsmsx(), bl1_strsv(), and bl1_strsvsx().

◆ bl1_sfree_saved_contigm()

void bl1_sfree_saved_contigm ( int  m,
int  n,
float *  a_save,
int  a_rs_save,
int  a_cs_save,
float **  a,
int *  a_rs,
int *  a_cs 
)
14{
15 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
16 {
17 // Copy the contents of the temporary matrix back to the original.
18 bl1_scopymt( BLIS1_NO_TRANSPOSE,
19 m,
20 n,
21 *a, *a_rs, *a_cs,
22 a_save, a_rs_save, a_cs_save );
23
24 // Free the temporary contiguous storage for the matrix.
25 bl1_sfree( *a );
26
27 // Restore the original matrix address.
28 *a = a_save;
29
30 // Restore the original row and column strides.
31 *a_rs = a_rs_save;
32 *a_cs = a_cs_save;
33 }
34}

References bl1_is_gen_storage(), bl1_scopymt(), bl1_sfree(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_sgemm(), bl1_sger(), bl1_ssymm(), bl1_ssyr(), bl1_ssyr2(), bl1_strmm(), bl1_strmmsx(), bl1_strsm(), and bl1_strsmsx().

◆ bl1_sfree_saved_contigmr()

void bl1_sfree_saved_contigmr ( uplo1_t  uplo,
int  m,
int  n,
float *  a_save,
int  a_rs_save,
int  a_cs_save,
float **  a,
int *  a_rs,
int *  a_cs 
)
14{
15 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
16 {
17 // Copy the contents of the temporary matrix back to the original.
18 bl1_scopymr( uplo,
19 m,
20 n,
21 *a, *a_rs, *a_cs,
22 a_save, a_rs_save, a_cs_save );
23
24 // Free the temporary contiguous storage for the matrix.
25 bl1_sfree( *a );
26
27 // Restore the original matrix address.
28 *a = a_save;
29
30 // Restore the original row and column strides.
31 *a_rs = a_rs_save;
32 *a_cs = a_cs_save;
33 }
34}

References bl1_is_gen_storage(), bl1_scopymr(), and bl1_sfree().

Referenced by bl1_ssyr2k(), and bl1_ssyrk().

◆ bl1_sfree_saved_contigmsr()

void bl1_sfree_saved_contigmsr ( side1_t  side,
uplo1_t  uplo,
int  m,
int  n,
float *  a_save,
int  a_rs_save,
int  a_cs_save,
float **  a,
int *  a_rs,
int *  a_cs 
)
14{
15 int dim_a;
16
17 // Choose the dimension of the matrix based on the side parameter.
18 if ( bl1_is_left( side ) ) dim_a = m;
19 else dim_a = n;
20
21 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
22 {
23 // Copy the contents of the temporary matrix back to the original.
24 bl1_scopymt( uplo,
25 dim_a,
26 dim_a,
27 *a, *a_rs, *a_cs,
28 a_save, a_rs_save, a_cs_save );
29
30 // Free the temporary contiguous storage for the matrix.
31 bl1_sfree( *a );
32
33 // Restore the original matrix address.
34 *a = a_save;
35
36 // Restore the original row and column strides.
37 *a_rs = a_rs_save;
38 *a_cs = a_cs_save;
39 }
40}

References bl1_is_gen_storage(), bl1_is_left(), bl1_scopymt(), and bl1_sfree().

◆ bl1_sident()

void bl1_sident ( int  m,
float *  a,
int  a_rs,
int  a_cs 
)
14{
15 float* alpha;
16 int i, j;
17
18 for ( j = 0; j < m; ++j )
19 {
20 for ( i = 0; i < m; ++i )
21 {
22 alpha = a + i*a_rs + j*a_cs;
23
24 *alpha = 0.0F;
25
26 if ( i == j )
27 *alpha = 1.0F;
28 }
29 }
30}

References i.

Referenced by FLA_UDdate_UT_ops_var1().

◆ bl1_sinvert2s()

void bl1_sinvert2s ( conj1_t  conj,
float *  alpha,
float *  beta 
)
14{
15 float one = 1.0F;
16
17 *beta = one / *alpha;
18}

Referenced by bl1_csinvscalm(), and bl1_sinvscalm().

◆ bl1_sinverts()

void bl1_sinverts ( conj1_t  conj,
float *  alpha 
)
14{
15 float one = 1.0F;
16
17 *alpha = one / *alpha;
18}

Referenced by FLA_Trinv_ln_ops_var1(), FLA_Trinv_ln_ops_var2(), FLA_Trinv_ln_ops_var3(), FLA_Trinv_ln_ops_var4(), FLA_Trinv_un_ops_var1(), FLA_Trinv_un_ops_var2(), FLA_Trinv_un_ops_var3(), and FLA_Trinv_un_ops_var4().

◆ bl1_sinvertv()

void bl1_sinvertv ( conj1_t  conj,
int  n,
float *  x,
int  incx 
)
14{
15 float one = 1.0F;
16 float* chi;
17 int i;
18
19 for ( i = 0; i < n; ++i )
20 {
21 chi = x + i*incx;
22
23 *chi = one / *chi;
24 }
25}

References i.

Referenced by FLA_Invert().

◆ bl1_sm1()

float bl1_sm1 ( void  )
176{
177 float x;
178 x = -1.0F;
179 return x;
180}

Referenced by bl1_cconjm(), bl1_cconjmr(), bl1_cconjv(), bl1_cm1(), FLA_Bsvd_ext_opc_var1(), FLA_Bsvd_ext_ops_var1(), FLA_Bsvd_v_opc_var1(), and FLA_Bsvd_v_ops_var1().

◆ bl1_sm1h()

float bl1_sm1h ( void  )
144{
145 float x;
146 x = -0.5F;
147 return x;
148}

Referenced by bl1_cm1h().

◆ bl1_sm2()

float bl1_sm2 ( void  )
208{
209 float x;
210 x = -2.0F;
211 return x;
212}

Referenced by bl1_cm2().

◆ bl1_smaxabsm()

void bl1_smaxabsm ( int  m,
int  n,
float *  a,
int  a_rs,
int  a_cs,
float *  maxabs 
)
14{
15 float zero = bl1_s0();
16 float* a_begin;
17 float maxabs_cand;
18 float maxabs_temp;
19 int inca, lda;
20 int n_iter;
21 int n_elem;
22 int j;
23
24 // Return early if possible.
25 if ( bl1_zero_dim2( m, n ) ) { *maxabs = zero; return; }
26
27 // Initialize with optimal values for column-major storage.
28 inca = a_rs;
29 lda = a_cs;
30 n_iter = n;
31 n_elem = m;
32
33 // An optimization: if A is row-major, then let's access the matrix by
34 // rows instead of by columns for increased spatial locality.
35 if ( bl1_is_row_storage( a_rs, a_cs ) )
36 {
37 bl1_swap_ints( n_iter, n_elem );
38 bl1_swap_ints( lda, inca );
39 }
40
41 // Initialize the maximum absolute value candidate to the first element.
42 bl1_sabsval2( a, &maxabs_cand );
43
44 for ( j = 0; j < n_iter; j++ )
45 {
46 a_begin = a + j*lda;
47
48 bl1_smaxabsv( n_elem,
49 a_begin, inca,
50 &maxabs_temp );
51
52 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
53 }
54
55 *maxabs = maxabs_cand;
56}
bl1_smaxabsv
void bl1_smaxabsv(int n, float *x, int incx, float *maxabs)
Definition bl1_maxabsv.c:13

References bl1_is_row_storage(), bl1_s0(), bl1_smaxabsv(), and bl1_zero_dim2().

Referenced by FLA_Max_abs_value().

◆ bl1_smaxabsmr()

void bl1_smaxabsmr ( uplo1_t  uplo,
int  m,
int  n,
float *  a,
int  a_rs,
int  a_cs,
float *  maxabs 
)
14{
15 float zero = bl1_s0();
16 float* a_begin;
17 float maxabs_cand;
18 float maxabs_temp;
19 int inca, lda;
20 int n_iter;
21 int n_elem_max;
22 int n_elem;
23 int j;
24
25 // Return early if possible.
26 if ( bl1_zero_dim2( m, n ) ) { *maxabs = zero; return; }
27
28 // Initialize with optimal values for column-major storage.
29 n_iter = n;
30 n_elem_max = m;
31 lda = a_cs;
32 inca = a_rs;
33
34 // An optimization: if A is row-major, then let's access the matrix by
35 // rows instead of by columns for increased spatial locality.
36 if ( bl1_is_row_storage( a_rs, a_cs ) )
37 {
38 bl1_swap_ints( n_iter, n_elem_max );
39 bl1_swap_ints( lda, inca );
40 bl1_toggle_uplo( uplo );
41 }
42
43 // Initialize the maximum absolute value candidate to the first element.
44 bl1_sabsval2( a, &maxabs_cand );
45
46 if ( bl1_is_upper( uplo ) )
47 {
48 for ( j = 0; j < n_iter; j++ )
49 {
50 n_elem = bl1_min( j + 1, n_elem_max );
51 a_begin = a + j*lda;
52
53 bl1_smaxabsv( n_elem,
54 a_begin, inca,
55 &maxabs_temp );
56
57 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
58 }
59 }
60 else // if ( bl1_is_lower( uplo ) )
61 {
62 for ( j = 0; j < n_iter; j++ )
63 {
64 n_elem = bl1_max( 0, n_elem_max - j );
65 a_begin = a + j*lda + j*inca;
66
67 bl1_smaxabsv( n_elem,
68 a_begin, inca,
69 &maxabs_temp );
70
71 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
72 }
73 }
74
75 *maxabs = maxabs_cand;
76}

References bl1_is_row_storage(), bl1_is_upper(), bl1_s0(), bl1_smaxabsv(), and bl1_zero_dim2().

Referenced by FLA_Max_abs_value_herm().

◆ bl1_smaxabsv()

void bl1_smaxabsv ( int  n,
float *  x,
int  incx,
float *  maxabs 
)
14{
15 float* chi;
16 float maxabs_cand;
17 float maxabs_temp;
18 int i;
19
20 bl1_sabsval2( x, &maxabs_cand );
21
22 for ( i = 0; i < n; ++i )
23 {
24 chi = x + i*incx;
25
26 bl1_sabsval2( chi, &maxabs_temp );
27
28 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
29 }
30
31 *maxabs = maxabs_cand;
32}

References i.

Referenced by bl1_smaxabsm(), and bl1_smaxabsmr().

◆ bl1_srandm()

void bl1_srandm ( int  m,
int  n,
float *  a,
int  a_rs,
int  a_cs 
)
14{
15 float* a_begin;
16 int inca, lda;
17 int n_iter;
18 int n_elem;
19 int j;
20
21 // Return early if possible.
22 if ( bl1_zero_dim2( m, n ) ) return;
23
24 // Initialize with optimal values for column-major storage.
25 inca = a_rs;
26 lda = a_cs;
27 n_iter = n;
28 n_elem = m;
29
30 // An optimization: if A is row-major, then let's access the matrix by
31 // rows instead of by columns for increased spatial locality.
32 if ( bl1_is_row_storage( a_rs, a_cs ) )
33 {
34 bl1_swap_ints( n_iter, n_elem );
35 bl1_swap_ints( lda, inca );
36 }
37
38 for ( j = 0; j < n_iter; j++ )
39 {
40 a_begin = a + j*lda;
41
42 bl1_srandv( n_elem,
43 a_begin, inca );
44 }
45}
bl1_srandv
void bl1_srandv(int n, float *x, int incx)
Definition bl1_randv.c:13

References bl1_is_row_storage(), bl1_srandv(), and bl1_zero_dim2().

Referenced by FLA_Random_matrix().

◆ bl1_srandmr()

void bl1_srandmr ( uplo1_t  uplo,
diag1_t  diag,
int  m,
int  n,
float *  a,
int  a_rs,
int  a_cs 
)
14{
15 float* a_begin;
16 float* ajj;
17 float one;
18 float zero;
19 float ord;
20 int lda, inca;
21 int n_iter;
22 int n_elem_max;
23 int n_elem;
24 int j;
25
26 // Return early if possible.
27 if ( bl1_zero_dim2( m, n ) ) return;
28
29 // Initialize with optimal values for column-major storage.
30 n_iter = n;
31 n_elem_max = m;
32 lda = a_cs;
33 inca = a_rs;
34
35 // An optimization: if A is row-major, then let's access the matrix by
36 // rows instead of by columns to increase spatial locality.
37 if ( bl1_is_row_storage( a_rs, a_cs ) )
38 {
39 bl1_swap_ints( n_iter, n_elem_max );
40 bl1_swap_ints( lda, inca );
41 bl1_toggle_uplo( uplo );
42 }
43
44 // Initialize some scalars.
45 one = bl1_s1();
46 zero = bl1_s0();
47 ord = ( float ) bl1_max( m, n );
48
49 if ( bl1_is_upper( uplo ) )
50 {
51 for ( j = 0; j < n_iter; j++ )
52 {
53 n_elem = bl1_min( j, n_elem_max );
54 a_begin = a + j*lda;
55
56 // Randomize super-diagonal elements.
57 bl1_srandv( n_elem,
58 a_begin, inca );
59
60 // Normalize super-diagonal elements by order of the matrix.
61 bl1_sinvscalv( BLIS1_NO_CONJUGATE,
62 n_elem,
63 &ord,
64 a_begin, inca );
65
66 // Initialize diagonal and sub-diagonal elements only if there are
67 // elements left in the column (ie: j < n_elem_max).
68 if ( j < n_elem_max )
69 {
70 ajj = a_begin + j*inca;
71
72 // Initialize diagonal element.
73 if ( bl1_is_unit_diag( diag ) ) *ajj = one;
74 else if ( bl1_is_zero_diag( diag ) ) *ajj = zero;
75 else if ( bl1_is_nonunit_diag( diag ) )
76 {
77 // We want positive diagonal elements between 1 and 2.
78 bl1_srands( ajj );
79 bl1_sabsval2( ajj, ajj );
80 bl1_sadd3( ajj, &one, ajj );
81 }
82
83 // Initialize sub-diagonal elements to zero.
84 bl1_ssetv( n_elem_max - j - 1,
85 &zero,
86 ajj + inca, inca );
87 }
88 }
89 }
90 else // if ( bl1_is_lower( uplo ) )
91 {
92 for ( j = 0; j < n_iter; j++ )
93 {
94 n_elem = bl1_min( j, n_elem_max );
95 a_begin = a + j*lda;
96
97 // Initialize super-diagonal to zero.
98 bl1_ssetv( n_elem,
99 &zero,
100 a_begin, inca );
101
102 // Initialize diagonal and sub-diagonal elements only if there are
103 // elements left in the column (ie: j < n_elem_max).
104 if ( j < n_elem_max )
105 {
106 ajj = a_begin + j*inca;
107
108 // Initialize diagonal element.
109 if ( bl1_is_unit_diag( diag ) ) *ajj = one;
110 else if ( bl1_is_zero_diag( diag ) ) *ajj = zero;
111 else if ( bl1_is_nonunit_diag( diag ) )
112 {
113 // We want positive diagonal elements between 1 and 2.
114 bl1_srands( ajj );
115 bl1_sabsval2( ajj, ajj );
116 bl1_sadd3( ajj, &one, ajj );
117 }
118
119 // Randomize sub-diagonal elements.
120 bl1_srandv( n_elem_max - j - 1,
121 ajj + inca, inca );
122
123 // Normalize sub-diagonal elements by order of the matrix.
124 bl1_sinvscalv( BLIS1_NO_CONJUGATE,
125 n_elem_max - j - 1,
126 &ord,
127 ajj + inca, inca );
128
129 }
130 }
131 }
132}
bl1_sinvscalv
void bl1_sinvscalv(conj1_t conj, int n, float *alpha, float *x, int incx)
Definition bl1_invscalv.c:13
bl1_srands
void bl1_srands(float *alpha)
Definition bl1_rands.c:13
bl1_ssetv
void bl1_ssetv(int m, float *sigma, float *x, int incx)
Definition bl1_setv.c:26
bl1_s1
float bl1_s1(void)
Definition bl1_constants.c:47

References bl1_is_nonunit_diag(), bl1_is_row_storage(), bl1_is_unit_diag(), bl1_is_upper(), bl1_is_zero_diag(), bl1_s0(), bl1_s1(), bl1_sinvscalv(), bl1_srands(), bl1_srandv(), bl1_ssetv(), bl1_zero_dim2(), and BLIS1_NO_CONJUGATE.

Referenced by FLA_Random_tri_matrix().

◆ bl1_srands()

void bl1_srands ( float *  alpha)
14{
15 *alpha = ( float ) ( ( double ) rand() / ( ( double ) RAND_MAX / 2.0F ) ) - 1.0F;
16}

Referenced by bl1_crands(), bl1_srandmr(), and bl1_srandv().

◆ bl1_srandv()

void bl1_srandv ( int  n,
float *  x,
int  incx 
)
14{
15 float* chi;
16 int i;
17
18 for ( i = 0; i < n; ++i )
19 {
20 chi = x + i*incx;
21
22 bl1_srands( chi );
23 }
24}

References bl1_srands(), and i.

Referenced by bl1_srandm(), and bl1_srandmr().

◆ bl1_sscalediag()

void bl1_sscalediag ( conj1_t  conj,
int  offset,
int  m,
int  n,
float *  sigma,
float *  a,
int  a_rs,
int  a_cs 
)
14{
15 float* alpha;
16 int i, j;
17
18 i = j = 0;
19
20 if ( offset < 0 ) i = -offset;
21 else if ( offset > 0 ) j = offset;
22
23 while ( i < m && j < n )
24 {
25 alpha = a + i*a_rs + j*a_cs;
26
27 *alpha *= *sigma;
28
29 ++i;
30 ++j;
31 }
32}

References i.

Referenced by FLA_Scale_diag(), and FLA_UDdate_UT_ops_var1().

◆ bl1_ssetdiag()

void bl1_ssetdiag ( int  offset,
int  m,
int  n,
float *  sigma,
float *  a,
int  a_rs,
int  a_cs 
)
35{
36 float* alpha;
37 int i, j;
38
39 i = j = 0;
40
41 if ( offset < 0 ) i = -offset;
42 else if ( offset > 0 ) j = offset;
43
44 while ( i < m && j < n )
45 {
46 alpha = a + i*a_rs + j*a_cs;
47
48 *alpha = *sigma;
49
50 ++i;
51 ++j;
52 }
53}

References i.

Referenced by FLA_Set_diag(), FLA_Set_offdiag(), and FLA_Triangularize().

◆ bl1_ssetm()

void bl1_ssetm ( int  m,
int  n,
float *  sigma,
float *  a,
int  a_rs,
int  a_cs 
)
30{
31 float* alpha;
32 int i, j;
33
34 for ( j = 0; j < n; ++j )
35 {
36 for ( i = 0; i < m; ++i )
37 {
38 alpha = a + i*a_rs + j*a_cs;
39
40 *alpha = *sigma;
41 }
42 }
43}

References i.

Referenced by FLA_Bidiag_UT_u_step_ofs_var4(), FLA_Bidiag_UT_u_step_ops_var4(), FLA_Bidiag_UT_u_step_ops_var5(), FLA_Hess_UT_step_ofs_var4(), FLA_Hess_UT_step_ops_var4(), FLA_Hess_UT_step_ops_var5(), FLA_Set(), FLA_Tridiag_UT_l_step_ofs_var3(), and FLA_Tridiag_UT_l_step_ops_var3().

◆ bl1_ssetmr()

void bl1_ssetmr ( uplo1_t  uplo,
int  m,
int  n,
float *  sigma,
float *  a,
int  a_rs,
int  a_cs 
)
14{
15 float* a_begin;
16 int lda, inca;
17 int n_iter;
18 int n_elem_max;
19 int n_elem;
20 int j;
21
22 // Return early if possible.
23 if ( bl1_zero_dim2( m, n ) ) return;
24
25 // Initialize with optimal values for column-major storage.
26 n_iter = n;
27 n_elem_max = m;
28 lda = a_cs;
29 inca = a_rs;
30
31 // An optimization: if A is row-major, then let's access the matrix by
32 // rows instead of by columns to increase spatial locality.
33 if ( bl1_is_row_storage( a_rs, a_cs ) )
34 {
35 bl1_swap_ints( n_iter, n_elem_max );
36 bl1_swap_ints( lda, inca );
37 bl1_toggle_uplo( uplo );
38 }
39
40 if ( bl1_is_upper( uplo ) )
41 {
42 for ( j = 0; j < n_iter; j++ )
43 {
44 n_elem = bl1_min( j, n_elem_max );
45 a_begin = a + j*lda;
46
47 bl1_ssetv( n_elem,
48 sigma,
49 a_begin, inca );
50 }
51 }
52 else // if ( bl1_is_lower( uplo ) )
53 {
54 for ( j = 0; j < n_iter; j++ )
55 {
56 n_elem = bl1_max( 0, n_elem_max - j - 1 );
57 a_begin = a + j*lda + (j + 1)*inca;
58
59 bl1_ssetv( n_elem,
60 sigma,
61 a_begin, inca );
62 }
63 }
64}

References bl1_is_row_storage(), bl1_is_upper(), bl1_ssetv(), and bl1_zero_dim2().

Referenced by FLA_Setr(), and FLA_Triangularize().

◆ bl1_ssetv()

void bl1_ssetv ( int  m,
float *  sigma,
float *  x,
int  incx 
)
27{
28 float* chi;
29 int i;
30
31 for ( i = 0; i < n; ++i )
32 {
33 chi = x + i*incx;
34
35 *chi = *sigma;
36 }
37}

References i.

Referenced by bl1_srandmr(), bl1_ssetmr(), FLA_Bidiag_UT_l_realify_opt(), FLA_Bidiag_UT_realify_diagonals_opt(), FLA_Bidiag_UT_u_realify_opt(), FLA_Bidiag_UT_u_step_ofs_var4(), FLA_Bidiag_UT_u_step_ops_var4(), FLA_Fused_Ahx_Ax_ops_var1(), FLA_Fused_Ahx_Axpy_Ax_ops_var1(), FLA_Fused_Gerc2_Ahx_Ax_ops_var1(), FLA_Fused_Gerc2_Ahx_Axpy_Ax_ops_var1(), FLA_Fused_Her2_Ax_l_ops_var1(), FLA_Obj_extract_imag_part(), FLA_Tridiag_UT_l_realify_opt(), FLA_Tridiag_UT_realify_subdiagonal_opt(), FLA_Tridiag_UT_shift_U_l_ops(), and FLA_Tridiag_UT_u_realify_opt().

◆ bl1_sshiftdiag()

void bl1_sshiftdiag ( conj1_t  conj,
int  offset,
int  m,
int  n,
float *  sigma,
float *  a,
int  a_rs,
int  a_cs 
)
14{
15 float* alpha;
16 int i, j;
17
18 i = j = 0;
19
20 if ( offset < 0 ) i = -offset;
21 else if ( offset > 0 ) j = offset;
22
23 while ( i < m && j < n )
24 {
25 alpha = a + i*a_rs + j*a_cs;
26
27 *alpha += *sigma;
28
29 ++i;
30 ++j;
31 }
32}

References i.

Referenced by FLA_Lyap_h_ops_var1(), FLA_Lyap_h_ops_var2(), FLA_Lyap_h_ops_var3(), FLA_Lyap_h_ops_var4(), FLA_Lyap_n_ops_var1(), FLA_Lyap_n_ops_var2(), FLA_Lyap_n_ops_var3(), FLA_Lyap_n_ops_var4(), and FLA_Shift_diag().

◆ bl1_ssymmize()

void bl1_ssymmize ( conj1_t  conj,
uplo1_t  uplo,
int  m,
float *  a,
int  a_rs,
int  a_cs 
)
14{
15 float* a_src;
16 float* a_dst;
17 int rs_src, cs_src, inc_src;
18 int rs_dst, cs_dst, inc_dst;
19 int n_iter;
20 int j;
21
22 // Return early if possible.
23 if ( bl1_zero_dim1( m ) ) return;
24
25 // Assume A is square.
26 n_iter = m;
27
28 // Initialize with appropriate values based on storage.
29 if ( bl1_is_col_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
30 {
31 cs_src = 1;
32 rs_src = 0;
33 inc_src = a_cs;
34 cs_dst = a_cs;
35 rs_dst = 0;
36 inc_dst = 1;
37 }
38 else if ( bl1_is_col_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
39 {
40 cs_src = a_cs;
41 rs_src = 0;
42 inc_src = 1;
43 cs_dst = 1;
44 rs_dst = 0;
45 inc_dst = a_cs;
46 }
47 else if ( bl1_is_row_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
48 {
49 cs_src = 0;
50 rs_src = a_rs;
51 inc_src = 1;
52 cs_dst = 0;
53 rs_dst = 1;
54 inc_dst = a_rs;
55 }
56 else if ( bl1_is_row_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
57 {
58 cs_src = 0;
59 rs_src = 1;
60 inc_src = a_rs;
61 cs_dst = 0;
62 rs_dst = a_rs;
63 inc_dst = 1;
64 }
65 else if ( bl1_is_gen_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
66 {
67 // General stride with column-major tilt looks similar to column-major.
68 // General stride with row-major tilt looks similar to row-major.
69 if ( a_rs < a_cs )
70 {
71 cs_src = 1 * a_rs;
72 rs_src = 0;
73 inc_src = a_cs;
74 cs_dst = a_cs;
75 rs_dst = 0;
76 inc_dst = 1 * a_rs;
77 }
78 else // if ( a_rs > a_cs )
79 {
80 cs_src = 0;
81 rs_src = a_rs;
82 inc_src = 1 * a_cs;
83 cs_dst = 0;
84 rs_dst = 1 * a_cs;
85 inc_dst = a_rs;
86 }
87 }
88 else // if ( bl1_is_gen_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
89 {
90 // General stride with column-major tilt looks similar to column-major.
91 // General stride with row-major tilt looks similar to row-major.
92 if ( a_rs < a_cs )
93 {
94 cs_src = a_cs;
95 rs_src = 0;
96 inc_src = 1 * a_rs;
97 cs_dst = 1 * a_rs;
98 rs_dst = 0;
99 inc_dst = a_cs;
100 }
101 else // if ( a_rs > a_cs )
102 {
103 cs_src = 0;
104 rs_src = 1 * a_cs;
105 inc_src = a_rs;
106 cs_dst = 0;
107 rs_dst = a_rs;
108 inc_dst = 1 * a_cs;
109 }
110 }
111
112 for ( j = 0; j < n_iter; j++ )
113 {
114 a_src = a + j*cs_src + j*rs_src;
115 a_dst = a + j*cs_dst + j*rs_dst;
116
117 bl1_scopyv( conj,
118 j,
119 a_src, inc_src,
120 a_dst, inc_dst );
121 }
122}
bl1_scopyv
void bl1_scopyv(conj1_t conj, int m, float *x, int incx, float *y, int incy)
Definition bl1_copyv.c:35

References bl1_is_col_storage(), bl1_is_gen_storage(), bl1_is_lower(), bl1_is_row_storage(), bl1_is_upper(), bl1_scopyv(), and bl1_zero_dim1().

Referenced by FLA_Hermitianize(), and FLA_Symmetrize().

◆ bl1_vallocm()

void * bl1_vallocm ( unsigned int  m,
unsigned int  n,
unsigned int  elem_size 
)
21{
22 return ( void* ) BLIS1_MALLOC( m * n * elem_size );
23}

◆ bl1_vallocv()

void * bl1_vallocv ( unsigned int  n_elem,
unsigned int  elem_size 
)
21{
22 return ( void* ) BLIS1_MALLOC( n_elem * elem_size );
23}

◆ bl1_vfree()

void bl1_vfree ( void *  p)
21{
22 free( ( void* ) p );
23}

◆ bl1_z0()

dcomplex bl1_z0 ( void  )
134{
135 dcomplex x;
136 x.real = bl1_d0();
137 x.imag = bl1_d0();
138 return x;
139}
bl1_d0
double bl1_d0(void)
Definition bl1_constants.c:118
dcomplex
Definition blis_type_defs.h:138

References bl1_d0(), dcomplex::imag, and dcomplex::real.

Referenced by bl1_zgemm(), bl1_zgemv(), bl1_zhemm(), bl1_zhemv(), bl1_zrandmr(), bl1_zsymm(), FLA_Fused_Ahx_Ax_opz_var1(), FLA_Fused_Ahx_Axpy_Ax_opz_var1(), FLA_Fused_Gerc2_Ahx_Ax_opz_var1(), FLA_Fused_Gerc2_Ahx_Axpy_Ax_opz_var1(), FLA_Fused_Her2_Ax_l_opz_var1(), FLA_Fused_Uhu_Yhu_Zhu_opz_var1(), FLA_Fused_UYx_ZVx_opz_var1(), FLA_QR_UT_form_Q_opz_var1(), and FLA_Tridiag_UT_shift_U_l_opz().

◆ bl1_z1()

dcomplex bl1_z1 ( void  )
70{
71 dcomplex x;
72 x.real = bl1_d1();
73 x.imag = bl1_d0();
74 return x;
75}
bl1_d1
double bl1_d1(void)
Definition bl1_constants.c:54

References bl1_d0(), bl1_d1(), dcomplex::imag, and dcomplex::real.

Referenced by bl1_zgemm(), bl1_zgemv(), bl1_zhemm(), bl1_zhemv(), bl1_zher2k(), bl1_zherk(), bl1_zrandmr(), bl1_zsymm(), bl1_ztrmmsx(), bl1_ztrsmsx(), FLA_Bsvd_ext_opd_var1(), FLA_Bsvd_ext_opz_var1(), FLA_Bsvd_v_opd_var1(), FLA_Bsvd_v_opd_var2(), FLA_Bsvd_v_opz_var1(), FLA_Bsvd_v_opz_var2(), FLA_QR_UT_form_Q_opz_var1(), FLA_Tevd_n_opz_var1(), FLA_Tevd_v_opd_var1(), FLA_Tevd_v_opd_var2(), FLA_Tevd_v_opz_var1(), FLA_Tevd_v_opz_var2(), and FLA_Tridiag_UT_shift_U_l_opz().

◆ bl1_z1h()

dcomplex bl1_z1h ( void  )
102{
103 dcomplex x;
104 x.real = bl1_d1h();
105 x.imag = bl1_d0();
106 return x;
107}
bl1_d1h
double bl1_d1h(void)
Definition bl1_constants.c:86

References bl1_d0(), bl1_d1h(), dcomplex::imag, and dcomplex::real.

◆ bl1_z2()

dcomplex bl1_z2 ( void  )
38{
39 dcomplex x;
40 x.real = bl1_d2();
41 x.imag = bl1_d0();
42 return x;
43}
bl1_d2
double bl1_d2(void)
Definition bl1_constants.c:22

References bl1_d0(), bl1_d2(), dcomplex::imag, and dcomplex::real.

◆ bl1_zallocm()

dcomplex * bl1_zallocm ( unsigned int  m,
unsigned int  n 
)
46{
47 return ( dcomplex* ) BLIS1_MALLOC( m * n * sizeof( dcomplex ) );
48}

Referenced by bl1_zcreate_contigm(), bl1_zcreate_contigmr(), bl1_zcreate_contigmt(), bl1_zgemm(), bl1_zhemm(), bl1_zher2k(), bl1_zherk(), bl1_zsymm(), bl1_zsyr2k(), bl1_ztrmm(), bl1_ztrmmsx(), bl1_ztrsm(), and bl1_ztrsmsx().

◆ bl1_zallocv()

dcomplex * bl1_zallocv ( unsigned int  n_elem)
46{
47 return ( dcomplex* ) BLIS1_MALLOC( n_elem * sizeof( dcomplex ) );
48}

Referenced by bl1_zaxpymt(), bl1_zaxpysmt(), bl1_zaxpyv(), bl1_zgemv(), bl1_zger(), bl1_zhemv(), bl1_zher(), bl1_zher2(), bl1_zsymv_blas(), bl1_zsyr2_blas(), bl1_zsyr_blas(), bl1_ztrmv(), bl1_ztrmvsx(), bl1_ztrsv(), and bl1_ztrsvsx().

◆ bl1_zapdiagmv()

void bl1_zapdiagmv ( side1_t  side,
conj1_t  conj,
int  m,
int  n,
dcomplex *  x,
int  incx,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
289{
290 dcomplex* chi;
291 dcomplex* a_begin;
292 int inca, lda;
293 int n_iter;
294 int n_elem;
295 int j;
296
297 // Return early if possible.
298 if ( bl1_zero_dim2( m, n ) ) return;
299
300 // Initialize with optimal values for column-major storage.
301 inca = a_rs;
302 lda = a_cs;
303 n_iter = n;
304 n_elem = m;
305
306 // An optimization: if A is row-major, then we can proceed as if the
307 // operation were transposed (applying the diagonal values in x from the
308 // opposite side) for increased spatial locality.
309 if ( bl1_is_row_storage( a_rs, a_cs ) )
310 {
311 bl1_swap_ints( n_iter, n_elem );
312 bl1_swap_ints( lda, inca );
313 bl1_toggle_side( side );
314 }
315
316 if ( bl1_is_left( side ) )
317 {
318 for ( j = 0; j < n_iter; j++ )
319 {
320 a_begin = a + j*lda;
321
322 bl1_zewscalv( conj,
323 n_elem,
324 x, incx,
325 a_begin, inca );
326 }
327 }
328 else
329 {
330 for ( j = 0; j < n_iter; j++ )
331 {
332 a_begin = a + j*lda;
333 chi = x + j*incx;
334
335 bl1_zscalv( conj,
336 n_elem,
337 chi,
338 a_begin, inca );
339 }
340 }
341}
bl1_zscalv
void bl1_zscalv(conj1_t conj, int n, dcomplex *alpha, dcomplex *x, int incx)
Definition bl1_scalv.c:72
bl1_zewscalv
void bl1_zewscalv(conj1_t conj, int n, dcomplex *x, int incx, dcomplex *y, int incy)
Definition bl1_ewscalv.c:103

References bl1_is_left(), bl1_is_row_storage(), bl1_zero_dim2(), bl1_zewscalv(), and bl1_zscalv().

Referenced by FLA_Apply_diag_matrix().

◆ bl1_zcreate_contigm()

void bl1_zcreate_contigm ( int  m,
int  n,
dcomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
dcomplex **  a,
int *  a_rs,
int *  a_cs 
)
116{
117 int m_contig, n_contig;
118
119 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
120 {
121 // Initialize dimensions assuming no transposition needed during copy.
122 m_contig = m;
123 n_contig = n;
124
125/*
126 // Transpose the dimensions of the contiguous matrix, if requested.
127 if ( bl1_does_trans( trans_copy ) )
128 {
129 m_contig = n;
130 n_contig = m;
131 }
132*/
133
134 // Allocate temporary contiguous storage for the matrix.
135 *a = bl1_zallocm( m_contig, n_contig );
136
137 // Set the row and column strides for the temporary matrix.
138 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
139
140 // Initialize the contiguous matrix with the contents of the original.
141 bl1_zcopymt( BLIS1_NO_TRANSPOSE,
142 m_contig,
143 n_contig,
144 a_save, a_rs_save, a_cs_save,
145 *a, *a_rs, *a_cs );
146 }
147}
bl1_zcopymt
void bl1_zcopymt(trans1_t trans, int m, int n, dcomplex *a, int a_rs, int a_cs, dcomplex *b, int b_rs, int b_cs)
Definition bl1_copymt.c:286
bl1_zallocm
dcomplex * bl1_zallocm(unsigned int m, unsigned int n)
Definition bl1_allocm.c:45

References bl1_is_gen_storage(), bl1_set_contig_strides(), bl1_zallocm(), bl1_zcopymt(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_zgemm(), bl1_zgemv(), bl1_zger(), bl1_zhemm(), bl1_zsymm(), bl1_ztrmm(), bl1_ztrmmsx(), bl1_ztrsm(), and bl1_ztrsmsx().

◆ bl1_zcreate_contigmr()

void bl1_zcreate_contigmr ( uplo1_t  uplo,
int  m,
int  n,
dcomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
dcomplex **  a,
int *  a_rs,
int *  a_cs 
)
110{
111 int m_contig, n_contig;
112
113 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
114 {
115 // Initialize dimensions assuming no transposition needed during copy.
116 m_contig = m;
117 n_contig = n;
118/*
119 // Transpose the dimensions of the contiguous matrix, if requested.
120 if ( bl1_does_trans( trans_copy ) )
121 {
122 m_contig = n;
123 n_contig = m;
124 }
125*/
126 // Allocate temporary contiguous storage for the matrix.
127 *a = bl1_zallocm( m_contig, n_contig );
128
129 // Set the row and column strides for the temporary matrix.
130 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
131
132 // Initialize the contiguous matrix with the contents of the original.
133 bl1_zcopymr( uplo,
134 m_contig,
135 n_contig,
136 a_save, a_rs_save, a_cs_save,
137 *a, *a_rs, *a_cs );
138 }
139}
bl1_zcopymr
void bl1_zcopymr(uplo1_t uplo, int m, int n, dcomplex *a, int a_rs, int a_cs, dcomplex *b, int b_rs, int b_cs)
Definition bl1_copymr.c:202

References bl1_is_gen_storage(), bl1_set_contig_strides(), bl1_zallocm(), and bl1_zcopymr().

Referenced by bl1_zcreate_contigmsr(), bl1_zhemm(), bl1_zhemv(), bl1_zher(), bl1_zher2(), bl1_zher2k(), bl1_zherk(), bl1_zsymm(), bl1_zsymv(), bl1_zsyr(), bl1_zsyr2(), bl1_zsyr2k(), bl1_zsyrk(), bl1_ztrmm(), bl1_ztrmmsx(), bl1_ztrmv(), bl1_ztrmvsx(), bl1_ztrsm(), bl1_ztrsmsx(), bl1_ztrsv(), and bl1_ztrsvsx().

◆ bl1_zcreate_contigmsr()

void bl1_zcreate_contigmsr ( side1_t  side,
uplo1_t  uplo,
int  m,
int  n,
dcomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
dcomplex **  a,
int *  a_rs,
int *  a_cs 
)
62{
63 int dim_a;
64
65 // Choose the dimension of the matrix based on the side parameter.
66 if ( bl1_is_left( side ) ) dim_a = m;
67 else dim_a = n;
68
69 // Call the simple version with chosen dimensions.
70 bl1_zcreate_contigmr( uplo,
71 dim_a,
72 dim_a,
73 a_save, a_rs_save, a_cs_save,
74 a, a_rs, a_cs );
75}
bl1_zcreate_contigmr
void bl1_zcreate_contigmr(uplo1_t uplo, int m, int n, dcomplex *a_save, int a_rs_save, int a_cs_save, dcomplex **a, int *a_rs, int *a_cs)
Definition bl1_create_contigmr.c:109

References bl1_is_left(), and bl1_zcreate_contigmr().

◆ bl1_zcreate_contigmt()

void bl1_zcreate_contigmt ( trans1_t  trans_dims,
int  m,
int  n,
dcomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
dcomplex **  a,
int *  a_rs,
int *  a_cs 
)
128{
129 int m_contig, n_contig;
130
131 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
132 {
133 // Transpose the dimensions if requested.
134 if ( bl1_does_trans( trans_dims ) )
135 bl1_swap_ints( m, n );
136
137 // Initialize dimensions assuming no transposition needed during copy.
138 m_contig = m;
139 n_contig = n;
140
141/*
142 // Transpose the dimensions of the contiguous matrix, if requested.
143 if ( bl1_does_trans( trans_copy ) )
144 {
145 m_contig = n;
146 n_contig = m;
147 }
148*/
149
150 // Allocate temporary contiguous storage for the matrix.
151 *a = bl1_zallocm( m_contig, n_contig );
152
153 // Set the row and column strides for the temporary matrix.
154 bl1_set_contig_strides( m_contig, n_contig, a_rs, a_cs );
155
156 // Initialize the contiguous matrix with the contents of the original.
157 bl1_zcopymt( BLIS1_NO_TRANSPOSE,
158 m_contig,
159 n_contig,
160 a_save, a_rs_save, a_cs_save,
161 *a, *a_rs, *a_cs );
162 }
163}

References bl1_does_trans(), bl1_is_gen_storage(), bl1_set_contig_strides(), bl1_zallocm(), bl1_zcopymt(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_zgemm(), bl1_zher2k(), bl1_zherk(), bl1_zsyr2k(), and bl1_zsyrk().

◆ bl1_zdapdiagmv()

void bl1_zdapdiagmv ( side1_t  side,
conj1_t  conj,
int  m,
int  n,
double *  x,
int  incx,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
234{
235 double* chi;
236 dcomplex* a_begin;
237 int inca, lda;
238 int n_iter;
239 int n_elem;
240 int j;
241
242 // Return early if possible.
243 if ( bl1_zero_dim2( m, n ) ) return;
244
245 // Initialize with optimal values for column-major storage.
246 inca = a_rs;
247 lda = a_cs;
248 n_iter = n;
249 n_elem = m;
250
251 // An optimization: if A is row-major, then we can proceed as if the
252 // operation were transposed (applying the diagonal values in x from the
253 // opposite side) for increased spatial locality.
254 if ( bl1_is_row_storage( a_rs, a_cs ) )
255 {
256 bl1_swap_ints( n_iter, n_elem );
257 bl1_swap_ints( lda, inca );
258 bl1_toggle_side( side );
259 }
260
261 if ( bl1_is_left( side ) )
262 {
263 for ( j = 0; j < n_iter; j++ )
264 {
265 a_begin = a + j*lda;
266
267 bl1_zdewscalv( conj,
268 n_elem,
269 x, incx,
270 a_begin, inca );
271 }
272 }
273 else
274 {
275 for ( j = 0; j < n_iter; j++ )
276 {
277 a_begin = a + j*lda;
278 chi = x + j*incx;
279
280 bl1_zdscalv( conj,
281 n_elem,
282 chi,
283 a_begin, inca );
284 }
285 }
286}
bl1_zdscalv
void bl1_zdscalv(conj1_t conj, int n, double *alpha, dcomplex *x, int incx)
Definition bl1_scalv.c:61
bl1_zdewscalv
void bl1_zdewscalv(conj1_t conj, int n, double *x, int incx, dcomplex *y, int incy)
Definition bl1_ewscalv.c:88

References bl1_is_left(), bl1_is_row_storage(), bl1_zdewscalv(), bl1_zdscalv(), and bl1_zero_dim2().

Referenced by FLA_Apply_diag_matrix().

◆ bl1_zdewinvscalmt()

void bl1_zdewinvscalmt ( trans1_t  trans,
int  m,
int  n,
double *  a,
int  a_rs,
int  a_cs,
dcomplex *  b,
int  b_rs,
int  b_cs 
)
302{
303 double* a_begin;
304 dcomplex* b_begin;
305 int lda, inca;
306 int ldb, incb;
307 int n_iter;
308 int n_elem;
309 int j;
310 conj1_t conj;
311
312 // Return early if possible.
313 if ( bl1_zero_dim2( m, n ) ) return;
314
315 // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
316 // gets invoked only once.
317 if ( bl1_is_vector( m, n ) )
318 {
319 // Initialize with values appropriate for vectors.
320 n_iter = 1;
321 n_elem = bl1_vector_dim( m, n );
322 lda = 1; // multiplied by zero when n_iter == 1; not needed.
323 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
324 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
325 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
326 }
327 else // matrix case
328 {
329 // Initialize with optimal values for column-major storage.
330 n_iter = n;
331 n_elem = m;
332 lda = a_cs;
333 inca = a_rs;
334 ldb = b_cs;
335 incb = b_rs;
336
337 // Handle the transposition of A.
338 if ( bl1_does_trans( trans ) )
339 {
340 bl1_swap_ints( lda, inca );
341 }
342
343 // An optimization: if B is row-major and if A is effectively row-major
344 // after a possible transposition, then let's access the matrices by rows
345 // instead of by columns for increased spatial locality.
346 if ( bl1_is_row_storage( b_rs, b_cs ) )
347 {
348 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
349 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
350 {
351 bl1_swap_ints( n_iter, n_elem );
352 bl1_swap_ints( lda, inca );
353 bl1_swap_ints( ldb, incb );
354 }
355 }
356 }
357
358 // Extract conj component from trans parameter.
359 conj = bl1_proj_trans1_to_conj( trans );
360
361 for ( j = 0; j < n_iter; j++ )
362 {
363 a_begin = a + j*lda;
364 b_begin = b + j*ldb;
365
366 bl1_zdewinvscalv( conj,
367 n_elem,
368 a_begin, inca,
369 b_begin, incb );
370 }
371}
bl1_zdewinvscalv
void bl1_zdewinvscalv(conj1_t conj, int n, double *x, int incx, dcomplex *y, int incy)
Definition bl1_ewinvscalv.c:88

References bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_vector_dim(), bl1_vector_inc(), bl1_zdewinvscalv(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

◆ bl1_zdewinvscalv()

void bl1_zdewinvscalv ( conj1_t  conj,
int  n,
double *  x,
int  incx,
dcomplex *  y,
int  incy 
)
89{
90 double* chi;
91 dcomplex* psi;
92 int i;
93
94 for ( i = 0; i < n; ++i )
95 {
96 chi = x + i*incx;
97 psi = y + i*incy;
98
99 bl1_zdinvscals( chi, psi );
100 }
101}

References i.

Referenced by bl1_zdewinvscalmt().

◆ bl1_zdewscalmt()

void bl1_zdewscalmt ( trans1_t  trans,
int  m,
int  n,
double *  a,
int  a_rs,
int  a_cs,
dcomplex *  b,
int  b_rs,
int  b_cs 
)
302{
303 double* a_begin;
304 dcomplex* b_begin;
305 int lda, inca;
306 int ldb, incb;
307 int n_iter;
308 int n_elem;
309 int j;
310 conj1_t conj;
311
312 // Return early if possible.
313 if ( bl1_zero_dim2( m, n ) ) return;
314
315 // Handle cases where A and B are vectors to ensure that the underlying ewscal
316 // gets invoked only once.
317 if ( bl1_is_vector( m, n ) )
318 {
319 // Initialize with values appropriate for vectors.
320 n_iter = 1;
321 n_elem = bl1_vector_dim( m, n );
322 lda = 1; // multiplied by zero when n_iter == 1; not needed.
323 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
324 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
325 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
326 }
327 else // matrix case
328 {
329 // Initialize with optimal values for column-major storage.
330 n_iter = n;
331 n_elem = m;
332 lda = a_cs;
333 inca = a_rs;
334 ldb = b_cs;
335 incb = b_rs;
336
337 // Handle the transposition of A.
338 if ( bl1_does_trans( trans ) )
339 {
340 bl1_swap_ints( lda, inca );
341 }
342
343 // An optimization: if B is row-major and if A is effectively row-major
344 // after a possible transposition, then let's access the matrices by rows
345 // instead of by columns for increased spatial locality.
346 if ( bl1_is_row_storage( b_rs, b_cs ) )
347 {
348 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
349 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
350 {
351 bl1_swap_ints( n_iter, n_elem );
352 bl1_swap_ints( lda, inca );
353 bl1_swap_ints( ldb, incb );
354 }
355 }
356 }
357
358 // Extract conj component from trans parameter.
359 conj = bl1_proj_trans1_to_conj( trans );
360
361 for ( j = 0; j < n_iter; j++ )
362 {
363 a_begin = a + j*lda;
364 b_begin = b + j*ldb;
365
366 bl1_zdewscalv( conj,
367 n_elem,
368 a_begin, inca,
369 b_begin, incb );
370 }
371}

References bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_vector_dim(), bl1_vector_inc(), bl1_zdewscalv(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

◆ bl1_zdewscalv()

void bl1_zdewscalv ( conj1_t  conj,
int  n,
double *  x,
int  incx,
dcomplex *  y,
int  incy 
)
89{
90 double* chi;
91 dcomplex* psi;
92 int i;
93
94 for ( i = 0; i < n; ++i )
95 {
96 chi = x + i*incx;
97 psi = y + i*incy;
98
99 bl1_zdscals( chi, psi );
100 }
101}

References i.

Referenced by bl1_zdapdiagmv(), and bl1_zdewscalmt().

◆ bl1_zdscalediag()

void bl1_zdscalediag ( conj1_t  conj,
int  offset,
int  m,
int  n,
double *  sigma,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
78{
79 dcomplex* alpha;
80 int i, j;
81
82 i = j = 0;
83
84 if ( offset < 0 ) i = -offset;
85 else if ( offset > 0 ) j = offset;
86
87 while ( i < m && j < n )
88 {
89 alpha = a + i*a_rs + j*a_cs;
90
91 alpha->real *= *sigma;
92 alpha->imag *= *sigma;
93
94 ++i;
95 ++j;
96 }
97}

References i, dcomplex::imag, and dcomplex::real.

Referenced by FLA_Scale_diag().

◆ bl1_zdshiftdiag()

void bl1_zdshiftdiag ( conj1_t  conj,
int  offset,
int  m,
int  n,
double *  sigma,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
77{
78 dcomplex* alpha;
79 int i, j;
80
81 i = j = 0;
82
83 if ( offset < 0 ) i = -offset;
84 else if ( offset > 0 ) j = offset;
85
86 while ( i < m && j < n )
87 {
88 alpha = a + i*a_rs + j*a_cs;
89
90 alpha->real += *sigma;
91
92 ++i;
93 ++j;
94 }
95}

References i, and dcomplex::real.

Referenced by FLA_Shift_diag().

◆ bl1_zewinvscalmt()

void bl1_zewinvscalmt ( trans1_t  trans,
int  m,
int  n,
dcomplex *  a,
int  a_rs,
int  a_cs,
dcomplex *  b,
int  b_rs,
int  b_cs 
)
374{
375 dcomplex* a_begin;
376 dcomplex* b_begin;
377 int lda, inca;
378 int ldb, incb;
379 int n_iter;
380 int n_elem;
381 int j;
382 conj1_t conj;
383
384 // Return early if possible.
385 if ( bl1_zero_dim2( m, n ) ) return;
386
387 // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
388 // gets invoked only once.
389 if ( bl1_is_vector( m, n ) )
390 {
391 // Initialize with values appropriate for vectors.
392 n_iter = 1;
393 n_elem = bl1_vector_dim( m, n );
394 lda = 1; // multiplied by zero when n_iter == 1; not needed.
395 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
396 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
397 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
398 }
399 else // matrix case
400 {
401 // Initialize with optimal values for column-major storage.
402 n_iter = n;
403 n_elem = m;
404 lda = a_cs;
405 inca = a_rs;
406 ldb = b_cs;
407 incb = b_rs;
408
409 // Handle the transposition of A.
410 if ( bl1_does_trans( trans ) )
411 {
412 bl1_swap_ints( lda, inca );
413 }
414
415 // An optimization: if B is row-major and if A is effectively row-major
416 // after a possible transposition, then let's access the matrices by rows
417 // instead of by columns for increased spatial locality.
418 if ( bl1_is_row_storage( b_rs, b_cs ) )
419 {
420 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
421 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
422 {
423 bl1_swap_ints( n_iter, n_elem );
424 bl1_swap_ints( lda, inca );
425 bl1_swap_ints( ldb, incb );
426 }
427 }
428 }
429
430 // Extract conj component from trans parameter.
431 conj = bl1_proj_trans1_to_conj( trans );
432
433 for ( j = 0; j < n_iter; j++ )
434 {
435 a_begin = a + j*lda;
436 b_begin = b + j*ldb;
437
438 bl1_zewinvscalv( conj,
439 n_elem,
440 a_begin, inca,
441 b_begin, incb );
442 }
443}
bl1_zewinvscalv
void bl1_zewinvscalv(conj1_t conj, int n, dcomplex *x, int incx, dcomplex *y, int incy)
Definition bl1_ewinvscalv.c:103

References bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), bl1_zewinvscalv(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Inv_scal_elemwise().

◆ bl1_zewinvscalv()

void bl1_zewinvscalv ( conj1_t  conj,
int  n,
dcomplex *  x,
int  incx,
dcomplex *  y,
int  incy 
)
104{
105 dcomplex* chi;
106 dcomplex* psi;
107 dcomplex conjchi;
108 int i;
109
110 if ( bl1_is_conj( conj ) )
111 {
112 for ( i = 0; i < n; ++i )
113 {
114 chi = x + i*incx;
115 psi = y + i*incy;
116
117 bl1_zcopyconj( chi, &conjchi );
118 bl1_zinvscals( &conjchi, psi );
119 }
120 }
121 else
122 {
123 for ( i = 0; i < n; ++i )
124 {
125 chi = x + i*incx;
126 psi = y + i*incy;
127
128 bl1_zinvscals( chi, psi );
129 }
130 }
131}

References bl1_is_conj(), and i.

Referenced by bl1_zewinvscalmt().

◆ bl1_zewscalmt()

void bl1_zewscalmt ( trans1_t  trans,
int  m,
int  n,
dcomplex *  a,
int  a_rs,
int  a_cs,
dcomplex *  b,
int  b_rs,
int  b_cs 
)
374{
375 dcomplex* a_begin;
376 dcomplex* b_begin;
377 int lda, inca;
378 int ldb, incb;
379 int n_iter;
380 int n_elem;
381 int j;
382 conj1_t conj;
383
384 // Return early if possible.
385 if ( bl1_zero_dim2( m, n ) ) return;
386
387 // Handle cases where A and B are vectors to ensure that the underlying ewscal
388 // gets invoked only once.
389 if ( bl1_is_vector( m, n ) )
390 {
391 // Initialize with values appropriate for vectors.
392 n_iter = 1;
393 n_elem = bl1_vector_dim( m, n );
394 lda = 1; // multiplied by zero when n_iter == 1; not needed.
395 inca = bl1_vector_inc( trans, m, n, a_rs, a_cs );
396 ldb = 1; // multiplied by zero when n_iter == 1; not needed.
397 incb = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
398 }
399 else // matrix case
400 {
401 // Initialize with optimal values for column-major storage.
402 n_iter = n;
403 n_elem = m;
404 lda = a_cs;
405 inca = a_rs;
406 ldb = b_cs;
407 incb = b_rs;
408
409 // Handle the transposition of A.
410 if ( bl1_does_trans( trans ) )
411 {
412 bl1_swap_ints( lda, inca );
413 }
414
415 // An optimization: if B is row-major and if A is effectively row-major
416 // after a possible transposition, then let's access the matrices by rows
417 // instead of by columns for increased spatial locality.
418 if ( bl1_is_row_storage( b_rs, b_cs ) )
419 {
420 if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
421 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
422 {
423 bl1_swap_ints( n_iter, n_elem );
424 bl1_swap_ints( lda, inca );
425 bl1_swap_ints( ldb, incb );
426 }
427 }
428 }
429
430 // Extract conj component from trans parameter.
431 conj = bl1_proj_trans1_to_conj( trans );
432
433 for ( j = 0; j < n_iter; j++ )
434 {
435 a_begin = a + j*lda;
436 b_begin = b + j*ldb;
437
438 bl1_zewscalv( conj,
439 n_elem,
440 a_begin, inca,
441 b_begin, incb );
442 }
443}

References bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_proj_trans1_to_conj(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), bl1_zewscalv(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Scal_elemwise().

◆ bl1_zewscalv()

void bl1_zewscalv ( conj1_t  conj,
int  n,
dcomplex *  x,
int  incx,
dcomplex *  y,
int  incy 
)
104{
105 dcomplex* chi;
106 dcomplex* psi;
107 dcomplex conjchi;
108 int i;
109
110 if ( bl1_is_conj( conj ) )
111 {
112 for ( i = 0; i < n; ++i )
113 {
114 chi = x + i*incx;
115 psi = y + i*incy;
116
117 bl1_zcopyconj( chi, &conjchi );
118 bl1_zscals( &conjchi, psi );
119 }
120 }
121 else
122 {
123 for ( i = 0; i < n; ++i )
124 {
125 chi = x + i*incx;
126 psi = y + i*incy;
127
128 bl1_zscals( chi, psi );
129 }
130 }
131}
bl1_zscals
bl1_zscals(beta, rho_yz)

References bl1_is_conj(), bl1_zscals(), and i.

Referenced by bl1_zapdiagmv(), and bl1_zewscalmt().

◆ bl1_zfree()

void bl1_zfree ( dcomplex *  p)
46{
47 free( ( void* ) p );
48}

Referenced by bl1_zaxpymt(), bl1_zaxpysmt(), bl1_zaxpyv(), bl1_zfree_contigm(), bl1_zfree_saved_contigm(), bl1_zfree_saved_contigmr(), bl1_zfree_saved_contigmsr(), bl1_zgemm(), bl1_zgemv(), bl1_zger(), bl1_zhemm(), bl1_zhemv(), bl1_zher(), bl1_zher2(), bl1_zher2k(), bl1_zherk(), bl1_zsymm(), bl1_zsymv_blas(), bl1_zsyr2_blas(), bl1_zsyr2k(), bl1_zsyr_blas(), bl1_ztrmm(), bl1_ztrmmsx(), bl1_ztrmv(), bl1_ztrmvsx(), bl1_ztrsm(), bl1_ztrsmsx(), bl1_ztrsv(), and bl1_ztrsvsx().

◆ bl1_zfree_contigm()

void bl1_zfree_contigm ( dcomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
dcomplex **  a,
int *  a_rs,
int *  a_cs 
)
62{
63 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
64 {
65 // Free the temporary contiguous storage for the matrix.
66 bl1_zfree( *a );
67
68 // Restore the original matrix address.
69 *a = a_save;
70
71 // Restore the original row and column strides.
72 *a_rs = a_rs_save;
73 *a_cs = a_cs_save;
74 }
75}
bl1_zfree
void bl1_zfree(dcomplex *p)
Definition bl1_free.c:45

References bl1_is_gen_storage(), and bl1_zfree().

Referenced by bl1_zgemm(), bl1_zgemv(), bl1_zhemm(), bl1_zhemv(), bl1_zher2k(), bl1_zherk(), bl1_zsymm(), bl1_zsymv(), bl1_zsyr2k(), bl1_zsyrk(), bl1_ztrmm(), bl1_ztrmmsx(), bl1_ztrmv(), bl1_ztrmvsx(), bl1_ztrsm(), bl1_ztrsmsx(), bl1_ztrsv(), and bl1_ztrsvsx().

◆ bl1_zfree_saved_contigm()

void bl1_zfree_saved_contigm ( int  m,
int  n,
dcomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
dcomplex **  a,
int *  a_rs,
int *  a_cs 
)
83{
84 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
85 {
86 // Copy the contents of the temporary matrix back to the original.
87 bl1_zcopymt( BLIS1_NO_TRANSPOSE,
88 m,
89 n,
90 *a, *a_rs, *a_cs,
91 a_save, a_rs_save, a_cs_save );
92
93 // Free the temporary contiguous storage for the matrix.
94 bl1_zfree( *a );
95
96 // Restore the original matrix address.
97 *a = a_save;
98
99 // Restore the original row and column strides.
100 *a_rs = a_rs_save;
101 *a_cs = a_cs_save;
102 }
103}

References bl1_is_gen_storage(), bl1_zcopymt(), bl1_zfree(), and BLIS1_NO_TRANSPOSE.

Referenced by bl1_zgemm(), bl1_zger(), bl1_zhemm(), bl1_zher(), bl1_zher2(), bl1_zsymm(), bl1_zsyr(), bl1_zsyr2(), bl1_ztrmm(), bl1_ztrmmsx(), bl1_ztrsm(), and bl1_ztrsmsx().

◆ bl1_zfree_saved_contigmr()

void bl1_zfree_saved_contigmr ( uplo1_t  uplo,
int  m,
int  n,
dcomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
dcomplex **  a,
int *  a_rs,
int *  a_cs 
)
83{
84 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
85 {
86 // Copy the contents of the temporary matrix back to the original.
87 bl1_zcopymr( uplo,
88 m,
89 n,
90 *a, *a_rs, *a_cs,
91 a_save, a_rs_save, a_cs_save );
92
93 // Free the temporary contiguous storage for the matrix.
94 bl1_zfree( *a );
95
96 // Restore the original matrix address.
97 *a = a_save;
98
99 // Restore the original row and column strides.
100 *a_rs = a_rs_save;
101 *a_cs = a_cs_save;
102 }
103}

References bl1_is_gen_storage(), bl1_zcopymr(), and bl1_zfree().

Referenced by bl1_zher2k(), bl1_zherk(), bl1_zsyr2k(), and bl1_zsyrk().

◆ bl1_zfree_saved_contigmsr()

void bl1_zfree_saved_contigmsr ( side1_t  side,
uplo1_t  uplo,
int  m,
int  n,
dcomplex *  a_save,
int  a_rs_save,
int  a_cs_save,
dcomplex **  a,
int *  a_rs,
int *  a_cs 
)
101{
102 int dim_a;
103
104 // Choose the dimension of the matrix based on the side parameter.
105 if ( bl1_is_left( side ) ) dim_a = m;
106 else dim_a = n;
107
108 if ( bl1_is_gen_storage( a_rs_save, a_cs_save ) )
109 {
110 // Copy the contents of the temporary matrix back to the original.
111 bl1_zcopymr( uplo,
112 dim_a,
113 dim_a,
114 *a, *a_rs, *a_cs,
115 a_save, a_rs_save, a_cs_save );
116
117 // Free the temporary contiguous storage for the matrix.
118 bl1_zfree( *a );
119
120 // Restore the original matrix address.
121 *a = a_save;
122
123 // Restore the original row and column strides.
124 *a_rs = a_rs_save;
125 *a_cs = a_cs_save;
126 }
127}

References bl1_is_gen_storage(), bl1_is_left(), bl1_zcopymr(), and bl1_zfree().

◆ bl1_zident()

void bl1_zident ( int  m,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
72{
73 dcomplex* alpha;
74 int i, j;
75
76 for ( j = 0; j < m; ++j )
77 {
78 for ( i = 0; i < m; ++i )
79 {
80 alpha = a + i*a_rs + j*a_cs;
81
82 alpha->real = 0.0;
83 alpha->imag = 0.0;
84
85 if ( i == j )
86 alpha->real = 1.0;
87 }
88 }
89}

References i, dcomplex::imag, and dcomplex::real.

Referenced by FLA_UDdate_UT_opz_var1().

◆ bl1_zinvert2s()

void bl1_zinvert2s ( conj1_t  conj,
dcomplex *  alpha,
dcomplex *  beta 
)
45{
46 double temp;
47 double s, xr_s, xi_s;
48
49 s = bl1_fmaxabs( alpha->real, alpha->imag ); \
50 xr_s = alpha->real / s;
51 xi_s = alpha->imag / s;
52 temp = xr_s * alpha->real + xi_s * alpha->imag;
53
54 beta->real = xr_s / temp;
55 beta->imag = -xi_s / temp;
56
57 if ( bl1_is_conj( conj ) )
58 bl1_zconjs( beta );
59}

References bl1_is_conj(), dcomplex::imag, dcomplex::real, and temp.

Referenced by bl1_zinvscalm(), and bl1_zinvscalv().

◆ bl1_zinverts()

void bl1_zinverts ( conj1_t  conj,
dcomplex *  alpha 
)
45{
46 double temp;
47 double s, xr_s, xi_s;
48
49 s = bl1_fmaxabs( alpha->real, alpha->imag ); \
50 xr_s = alpha->real / s;
51 xi_s = alpha->imag / s;
52 temp = xr_s * alpha->real + xi_s * alpha->imag;
53
54 alpha->real = xr_s / temp;
55 alpha->imag = -xi_s / temp;
56
57 if ( bl1_is_conj( conj ) )
58 bl1_zconjs( alpha );
59}

References bl1_is_conj(), dcomplex::imag, dcomplex::real, and temp.

Referenced by FLA_Trinv_ln_opz_var1(), FLA_Trinv_ln_opz_var2(), FLA_Trinv_ln_opz_var3(), FLA_Trinv_ln_opz_var4(), FLA_Trinv_un_opz_var1(), FLA_Trinv_un_opz_var2(), FLA_Trinv_un_opz_var3(), and FLA_Trinv_un_opz_var4().

◆ bl1_zinvertv()

void bl1_zinvertv ( conj1_t  conj,
int  n,
dcomplex *  x,
int  incx 
)
68{
69 double one = 1.0;
70 double temp;
71 double s, xr_s, xi_s;
72 double conjsign;
73 dcomplex* chi;
74 int i;
75
76 if ( bl1_is_conj( conj ) ) conjsign = one;
77 else conjsign = -one;
78
79 for ( i = 0; i < n; ++i )
80 {
81 chi = x + i*incx;
82
83 s = bl1_fmaxabs( chi->real, chi->imag ); \
84 xr_s = chi->real / s;
85 xi_s = chi->imag / s;
86 temp = xr_s * chi->real + xi_s * chi->imag;
87
88 chi->real = xr_s / temp;
89 chi->imag = conjsign * xi_s / temp;
90 }
91}

References bl1_is_conj(), i, dcomplex::imag, dcomplex::real, and temp.

Referenced by FLA_Invert().

◆ bl1_zm1()

dcomplex bl1_zm1 ( void  )
198{
199 dcomplex x;
200 x.real = bl1_dm1();
201 x.imag = bl1_d0();
202 return x;
203}
bl1_dm1
double bl1_dm1(void)
Definition bl1_constants.c:182

References bl1_d0(), bl1_dm1(), dcomplex::imag, and dcomplex::real.

Referenced by FLA_Fused_Ahx_Axpy_Ax_opz_var1(), and FLA_Fused_Gerc2_Ahx_Axpy_Ax_opz_var1().

◆ bl1_zm1h()

dcomplex bl1_zm1h ( void  )
166{
167 dcomplex x;
168 x.real = bl1_dm1h();
169 x.imag = bl1_d0();
170 return x;
171}
bl1_dm1h
double bl1_dm1h(void)
Definition bl1_constants.c:150

References bl1_d0(), bl1_dm1h(), dcomplex::imag, and dcomplex::real.

◆ bl1_zm2()

dcomplex bl1_zm2 ( void  )
230{
231 dcomplex x;
232 x.real = bl1_dm2();
233 x.imag = bl1_d0();
234 return x;
235}
bl1_dm2
double bl1_dm2(void)
Definition bl1_constants.c:214

References bl1_d0(), bl1_dm2(), dcomplex::imag, and dcomplex::real.

◆ bl1_zmaxabsm()

void bl1_zmaxabsm ( int  m,
int  n,
dcomplex *  a,
int  a_rs,
int  a_cs,
double *  maxabs 
)
149{
150 double zero = bl1_d0();
151 dcomplex* a_begin;
152 double maxabs_cand;
153 double maxabs_temp;
154 int inca, lda;
155 int n_iter;
156 int n_elem;
157 int j;
158
159 // Return early if possible.
160 if ( bl1_zero_dim2( m, n ) ) { *maxabs = zero; return; }
161
162 // Initialize with optimal values for column-major storage.
163 inca = a_rs;
164 lda = a_cs;
165 n_iter = n;
166 n_elem = m;
167
168 // An optimization: if A is row-major, then let's access the matrix by
169 // rows instead of by columns for increased spatial locality.
170 if ( bl1_is_row_storage( a_rs, a_cs ) )
171 {
172 bl1_swap_ints( n_iter, n_elem );
173 bl1_swap_ints( lda, inca );
174 }
175
176 // Initialize the maximum absolute value candidate to the first element.
177 bl1_zdabsval2( a, &maxabs_cand );
178
179 for ( j = 0; j < n_iter; j++ )
180 {
181 a_begin = a + j*lda;
182
183 bl1_zmaxabsv( n_elem,
184 a_begin, inca,
185 &maxabs_temp );
186
187 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
188 }
189
190 *maxabs = maxabs_cand;
191}
bl1_zmaxabsv
void bl1_zmaxabsv(int n, dcomplex *x, int incx, double *maxabs)
Definition bl1_maxabsv.c:76

References bl1_d0(), bl1_is_row_storage(), bl1_zero_dim2(), and bl1_zmaxabsv().

Referenced by FLA_Max_abs_value().

◆ bl1_zmaxabsmr()

void bl1_zmaxabsmr ( uplo1_t  uplo,
int  m,
int  n,
dcomplex *  a,
int  a_rs,
int  a_cs,
double *  maxabs 
)
209{
210 double zero = bl1_d0();
211 dcomplex* a_begin;
212 double maxabs_cand;
213 double maxabs_temp;
214 int inca, lda;
215 int n_iter;
216 int n_elem_max;
217 int n_elem;
218 int j;
219
220 // Return early if possible.
221 if ( bl1_zero_dim2( m, n ) ) { *maxabs = zero; return; }
222
223 // Initialize with optimal values for column-major storage.
224 n_iter = n;
225 n_elem_max = m;
226 lda = a_cs;
227 inca = a_rs;
228
229 // An optimization: if A is row-major, then let's access the matrix by
230 // rows instead of by columns for increased spatial locality.
231 if ( bl1_is_row_storage( a_rs, a_cs ) )
232 {
233 bl1_swap_ints( n_iter, n_elem_max );
234 bl1_swap_ints( lda, inca );
235 bl1_toggle_uplo( uplo );
236 }
237
238 // Initialize the maximum absolute value candidate to the first element.
239 bl1_zdabsval2( a, &maxabs_cand );
240
241 if ( bl1_is_upper( uplo ) )
242 {
243 for ( j = 0; j < n_iter; j++ )
244 {
245 n_elem = bl1_min( j + 1, n_elem_max );
246 a_begin = a + j*lda;
247
248 bl1_zmaxabsv( n_elem,
249 a_begin, inca,
250 &maxabs_temp );
251
252 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
253 }
254 }
255 else // if ( bl1_is_lower( uplo ) )
256 {
257 for ( j = 0; j < n_iter; j++ )
258 {
259 n_elem = bl1_max( 0, n_elem_max - j );
260 a_begin = a + j*lda + j*inca;
261
262 bl1_zmaxabsv( n_elem,
263 a_begin, inca,
264 &maxabs_temp );
265
266 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
267 }
268 }
269
270 *maxabs = maxabs_cand;
271}

References bl1_d0(), bl1_is_row_storage(), bl1_is_upper(), bl1_zero_dim2(), and bl1_zmaxabsv().

Referenced by FLA_Max_abs_value_herm().

◆ bl1_zmaxabsv()

void bl1_zmaxabsv ( int  n,
dcomplex *  x,
int  incx,
double *  maxabs 
)
77{
78 dcomplex* chi;
79 double maxabs_cand;
80 double maxabs_temp;
81 int i;
82
83 bl1_zdabsval2( x, &maxabs_cand );
84
85 for ( i = 0; i < n; ++i )
86 {
87 chi = x + i*incx;
88
89 bl1_zdabsval2( chi, &maxabs_temp );
90
91 if ( maxabs_temp > maxabs_cand ) maxabs_cand = maxabs_temp;
92 }
93
94 *maxabs = maxabs_cand;
95}

References i.

Referenced by bl1_zmaxabsm(), and bl1_zmaxabsmr().

◆ bl1_zrandm()

void bl1_zrandm ( int  m,
int  n,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
116{
117 dcomplex* a_begin;
118 int inca, lda;
119 int n_iter;
120 int n_elem;
121 int j;
122
123 // Return early if possible.
124 if ( bl1_zero_dim2( m, n ) ) return;
125
126 // Initialize with optimal values for column-major storage.
127 inca = a_rs;
128 lda = a_cs;
129 n_iter = n;
130 n_elem = m;
131
132 // An optimization: if A is row-major, then let's access the matrix by
133 // rows instead of by columns for increased spatial locality.
134 if ( bl1_is_row_storage( a_rs, a_cs ) )
135 {
136 bl1_swap_ints( n_iter, n_elem );
137 bl1_swap_ints( lda, inca );
138 }
139
140 for ( j = 0; j < n_iter; j++ )
141 {
142 a_begin = a + j*lda;
143
144 bl1_zrandv( n_elem,
145 a_begin, inca );
146 }
147}
bl1_zrandv
void bl1_zrandv(int n, dcomplex *x, int incx)
Definition bl1_randv.c:52

References bl1_is_row_storage(), bl1_zero_dim2(), and bl1_zrandv().

Referenced by FLA_Random_matrix().

◆ bl1_zrandmr()

void bl1_zrandmr ( uplo1_t  uplo,
diag1_t  diag,
int  m,
int  n,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
378{
379 dcomplex* a_begin;
380 dcomplex* ajj;
381 dcomplex one;
382 dcomplex zero;
383 dcomplex ord;
384 int lda, inca;
385 int n_iter;
386 int n_elem_max;
387 int n_elem;
388 int j;
389
390 // Return early if possible.
391 if ( bl1_zero_dim2( m, n ) ) return;
392
393 // Initialize with optimal values for column-major storage.
394 n_iter = n;
395 n_elem_max = m;
396 lda = a_cs;
397 inca = a_rs;
398
399 // An optimization: if A is row-major, then let's access the matrix by
400 // rows instead of by columns to increase spatial locality.
401 if ( bl1_is_row_storage( a_rs, a_cs ) )
402 {
403 bl1_swap_ints( n_iter, n_elem_max );
404 bl1_swap_ints( lda, inca );
405 bl1_toggle_uplo( uplo );
406 }
407
408 // Initialize some scalars.
409 one = bl1_z1();
410 zero = bl1_z0();
411 ord = bl1_z0();
412 ord.real = ( double ) bl1_max( m, n );
413
414 if ( bl1_is_upper( uplo ) )
415 {
416 for ( j = 0; j < n_iter; j++ )
417 {
418 n_elem = bl1_min( j, n_elem_max );
419 a_begin = a + j*lda;
420
421 // Randomize super-diagonal elements.
422 bl1_zrandv( n_elem,
423 a_begin, inca );
424
425 // Normalize super-diagonal elements by order of the matrix.
426 bl1_zinvscalv( BLIS1_NO_CONJUGATE,
427 n_elem,
428 &ord,
429 a_begin, inca );
430
431 // Initialize diagonal and sub-diagonal elements only if there are
432 // elements left in the column (ie: j < n_elem_max).
433 if ( j < n_elem_max )
434 {
435 ajj = a_begin + j*inca;
436
437 // Initialize diagonal element.
438 if ( bl1_is_unit_diag( diag ) ) *ajj = one;
439 else if ( bl1_is_zero_diag( diag ) ) *ajj = zero;
440 else if ( bl1_is_nonunit_diag( diag ) )
441 {
442 // We want positive diagonal elements between 1 and 2.
443 bl1_zrands( ajj );
444 bl1_zabsval2( ajj, ajj );
445 bl1_zadd3( ajj, &one, ajj );
446 }
447
448 // Initialize sub-diagonal elements to zero.
449 bl1_zsetv( n_elem_max - j - 1,
450 &zero,
451 ajj + inca, inca );
452 }
453 }
454 }
455 else // if ( bl1_is_lower( uplo ) )
456 {
457 for ( j = 0; j < n_iter; j++ )
458 {
459 n_elem = bl1_min( j, n_elem_max );
460 a_begin = a + j*lda;
461
462 // Initialize super-diagonal to zero.
463 bl1_zsetv( n_elem,
464 &zero,
465 a_begin, inca );
466
467 // Initialize diagonal and sub-diagonal elements only if there are
468 // elements left in the column (ie: j < n_elem_max).
469 if ( j < n_elem_max )
470 {
471 ajj = a_begin + j*inca;
472
473 // Initialize diagonal element.
474 if ( bl1_is_unit_diag( diag ) ) *ajj = one;
475 else if ( bl1_is_zero_diag( diag ) ) *ajj = zero;
476 else if ( bl1_is_nonunit_diag( diag ) )
477 {
478 // We want positive diagonal elements between 1 and 2.
479 bl1_zrands( ajj );
480 bl1_zabsval2( ajj, ajj );
481 bl1_zadd3( ajj, &one, ajj );
482 }
483
484 // Randomize sub-diagonal elements.
485 bl1_zrandv( n_elem_max - j - 1,
486 ajj + inca, inca );
487
488 // Normalize sub-diagonal elements by order of the matrix.
489 bl1_zinvscalv( BLIS1_NO_CONJUGATE,
490 n_elem_max - j - 1,
491 &ord,
492 ajj + inca, inca );
493
494 }
495 }
496 }
497}
bl1_zinvscalv
void bl1_zinvscalv(conj1_t conj, int n, dcomplex *alpha, dcomplex *x, int incx)
Definition bl1_invscalv.c:78
bl1_z0
dcomplex bl1_z0(void)
Definition bl1_constants.c:133
bl1_zsetv
void bl1_zsetv(int m, dcomplex *sigma, dcomplex *x, int incx)
Definition bl1_setv.c:66
bl1_z1
dcomplex bl1_z1(void)
Definition bl1_constants.c:69
bl1_zrands
void bl1_zrands(dcomplex *alpha)
Definition bl1_rands.c:29

References bl1_is_nonunit_diag(), bl1_is_row_storage(), bl1_is_unit_diag(), bl1_is_upper(), bl1_is_zero_diag(), bl1_z0(), bl1_z1(), bl1_zero_dim2(), bl1_zinvscalv(), bl1_zrands(), bl1_zrandv(), bl1_zsetv(), BLIS1_NO_CONJUGATE, and dcomplex::real.

Referenced by FLA_Random_tri_matrix().

◆ bl1_zrands()

void bl1_zrands ( dcomplex *  alpha)
30{
31 bl1_drands( &(alpha->real) );
32 bl1_drands( &(alpha->imag) );
33}
bl1_drands
void bl1_drands(double *alpha)
Definition bl1_rands.c:18

References bl1_drands(), dcomplex::imag, and dcomplex::real.

Referenced by bl1_zrandmr(), and bl1_zrandv().

◆ bl1_zrandv()

void bl1_zrandv ( int  n,
dcomplex *  x,
int  incx 
)
53{
54 dcomplex* chi;
55 int i;
56
57 for ( i = 0; i < n; ++i )
58 {
59 chi = x + i*incx;
60
61 bl1_zrands( chi );
62 }
63}

References bl1_zrands(), and i.

Referenced by bl1_zrandm(), and bl1_zrandmr().

◆ bl1_zscalediag()

void bl1_zscalediag ( conj1_t  conj,
int  offset,
int  m,
int  n,
dcomplex *  sigma,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
124{
125 dcomplex* alpha;
126 dcomplex sigma_conj;
127 int i, j;
128
129 bl1_zcopys( conj, sigma, &sigma_conj );
130
131 i = j = 0;
132
133 if ( offset < 0 ) i = -offset;
134 else if ( offset > 0 ) j = offset;
135
136 while ( i < m && j < n )
137 {
138 alpha = a + i*a_rs + j*a_cs;
139
140 bl1_zscals( &sigma_conj, alpha );
141
142 ++i;
143 ++j;
144 }
145}

References bl1_zscals(), and i.

Referenced by FLA_Scale_diag(), and FLA_UDdate_UT_opz_var1().

◆ bl1_zsetdiag()

void bl1_zsetdiag ( int  offset,
int  m,
int  n,
dcomplex *  sigma,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
99{
100 dcomplex* alpha;
101 int i, j;
102
103 i = j = 0;
104
105 if ( offset < 0 ) i = -offset;
106 else if ( offset > 0 ) j = offset;
107
108 while ( i < m && j < n )
109 {
110 alpha = a + i*a_rs + j*a_cs;
111
112 alpha->real = sigma->real;
113 alpha->imag = sigma->imag;
114
115 ++i;
116 ++j;
117 }
118}

References i, dcomplex::imag, and dcomplex::real.

Referenced by FLA_Set_diag(), FLA_Set_offdiag(), and FLA_Triangularize().

◆ bl1_zsetm()

void bl1_zsetm ( int  m,
int  n,
dcomplex *  sigma,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
79{
80 dcomplex* alpha;
81 int i, j;
82
83 for ( j = 0; j < n; ++j )
84 {
85 for ( i = 0; i < m; ++i )
86 {
87 alpha = a + i*a_rs + j*a_cs;
88
89 alpha->real = sigma->real;
90 alpha->imag = sigma->imag;
91 }
92 }
93}

References i, dcomplex::imag, and dcomplex::real.

Referenced by FLA_Bidiag_UT_u_step_ofz_var4(), FLA_Bidiag_UT_u_step_opz_var4(), FLA_Bidiag_UT_u_step_opz_var5(), FLA_Bsvd_ext_opd_var1(), FLA_Bsvd_ext_opz_var1(), FLA_Bsvd_v_opd_var1(), FLA_Bsvd_v_opd_var2(), FLA_Bsvd_v_opz_var1(), FLA_Bsvd_v_opz_var2(), FLA_Hess_UT_step_ofz_var4(), FLA_Hess_UT_step_opz_var4(), FLA_Hess_UT_step_opz_var5(), FLA_Set(), FLA_Tevd_n_opz_var1(), FLA_Tevd_v_opd_var1(), FLA_Tevd_v_opd_var2(), FLA_Tevd_v_opz_var1(), FLA_Tevd_v_opz_var2(), FLA_Tridiag_UT_l_step_ofz_var3(), and FLA_Tridiag_UT_l_step_opz_var3().

◆ bl1_zsetmr()

void bl1_zsetmr ( uplo1_t  uplo,
int  m,
int  n,
dcomplex *  sigma,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
173{
174 dcomplex* a_begin;
175 int lda, inca;
176 int n_iter;
177 int n_elem_max;
178 int n_elem;
179 int j;
180
181 // Return early if possible.
182 if ( bl1_zero_dim2( m, n ) ) return;
183
184 // Initialize with optimal values for column-major storage.
185 n_iter = n;
186 n_elem_max = m;
187 lda = a_cs;
188 inca = a_rs;
189
190 // An optimization: if A is row-major, then let's access the matrix by
191 // rows instead of by columns to increase spatial locality.
192 if ( bl1_is_row_storage( a_rs, a_cs ) )
193 {
194 bl1_swap_ints( n_iter, n_elem_max );
195 bl1_swap_ints( lda, inca );
196 bl1_toggle_uplo( uplo );
197 }
198
199 if ( bl1_is_upper( uplo ) )
200 {
201 for ( j = 0; j < n_iter; j++ )
202 {
203 n_elem = bl1_min( j, n_elem_max );
204 a_begin = a + j*lda;
205
206 bl1_zsetv( n_elem,
207 sigma,
208 a_begin, inca );
209 }
210 }
211 else // if ( bl1_is_lower( uplo ) )
212 {
213 for ( j = 0; j < n_iter; j++ )
214 {
215 n_elem = bl1_max( 0, n_elem_max - j - 1 );
216 a_begin = a + j*lda + (j + 1)*inca;
217
218 bl1_zsetv( n_elem,
219 sigma,
220 a_begin, inca );
221 }
222 }
223}

References bl1_is_row_storage(), bl1_is_upper(), bl1_zero_dim2(), and bl1_zsetv().

Referenced by FLA_Setr(), and FLA_Triangularize().

◆ bl1_zsetv()

void bl1_zsetv ( int  m,
dcomplex *  sigma,
dcomplex *  x,
int  incx 
)
67{
68 dcomplex* chi;
69 int i;
70
71 for ( i = 0; i < n; ++i )
72 {
73 chi = x + i*incx;
74
75 chi->real = sigma->real;
76 chi->imag = sigma->imag;
77 }
78}

References i, dcomplex::imag, and dcomplex::real.

Referenced by bl1_zrandmr(), bl1_zsetmr(), FLA_Bidiag_UT_u_step_ofz_var4(), FLA_Bidiag_UT_u_step_opz_var4(), FLA_Fused_Ahx_Ax_opz_var1(), FLA_Fused_Ahx_Axpy_Ax_opz_var1(), FLA_Fused_Gerc2_Ahx_Ax_opz_var1(), FLA_Fused_Gerc2_Ahx_Axpy_Ax_opz_var1(), FLA_Fused_Her2_Ax_l_opz_var1(), FLA_Tridiag_UT_l_realify_opt(), FLA_Tridiag_UT_realify_subdiagonal_opt(), FLA_Tridiag_UT_shift_U_l_opz(), and FLA_Tridiag_UT_u_realify_opt().

◆ bl1_zshiftdiag()

void bl1_zshiftdiag ( conj1_t  conj,
int  offset,
int  m,
int  n,
dcomplex *  sigma,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
123{
124 dcomplex* alpha;
125 dcomplex sigma_conj;
126 int i, j;
127
128 bl1_zcopys( conj, sigma, &sigma_conj );
129
130 i = j = 0;
131
132 if ( offset < 0 ) i = -offset;
133 else if ( offset > 0 ) j = offset;
134
135 while ( i < m && j < n )
136 {
137 alpha = a + i*a_rs + j*a_cs;
138
139 alpha->real += sigma_conj.real;
140 alpha->imag += sigma_conj.imag;
141
142 ++i;
143 ++j;
144 }
145}

References i, dcomplex::imag, and dcomplex::real.

Referenced by FLA_Lyap_h_opz_var1(), FLA_Lyap_h_opz_var2(), FLA_Lyap_h_opz_var3(), FLA_Lyap_h_opz_var4(), FLA_Lyap_n_opz_var1(), FLA_Lyap_n_opz_var2(), FLA_Lyap_n_opz_var3(), FLA_Lyap_n_opz_var4(), and FLA_Shift_diag().

◆ bl1_zsymmize()

void bl1_zsymmize ( conj1_t  conj,
uplo1_t  uplo,
int  m,
dcomplex *  a,
int  a_rs,
int  a_cs 
)
354{
355 dcomplex* a_src;
356 dcomplex* a_dst;
357 dcomplex* a_jj;
358 int rs_src, cs_src, inc_src;
359 int rs_dst, cs_dst, inc_dst;
360 int n_iter;
361 int j;
362
363 // Return early if possible.
364 if ( bl1_zero_dim1( m ) ) return;
365
366 // Assume A is square.
367 n_iter = m;
368
369 // Initialize with appropriate values based on storage.
370 if ( bl1_is_col_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
371 {
372 cs_src = 1;
373 rs_src = 0;
374 inc_src = a_cs;
375 cs_dst = a_cs;
376 rs_dst = 0;
377 inc_dst = 1;
378 }
379 else if ( bl1_is_col_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
380 {
381 cs_src = a_cs;
382 rs_src = 0;
383 inc_src = 1;
384 cs_dst = 1;
385 rs_dst = 0;
386 inc_dst = a_cs;
387 }
388 else if ( bl1_is_row_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
389 {
390 cs_src = 0;
391 rs_src = a_rs;
392 inc_src = 1;
393 cs_dst = 0;
394 rs_dst = 1;
395 inc_dst = a_rs;
396 }
397 else if ( bl1_is_row_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
398 {
399 cs_src = 0;
400 rs_src = 1;
401 inc_src = a_rs;
402 cs_dst = 0;
403 rs_dst = a_rs;
404 inc_dst = 1;
405 }
406 else if ( bl1_is_gen_storage( a_rs, a_cs ) && bl1_is_lower( uplo ) )
407 {
408 // General stride with column-major tilt looks similar to column-major.
409 // General stride with row-major tilt looks similar to row-major.
410 if ( a_rs < a_cs )
411 {
412 cs_src = 1 * a_rs;
413 rs_src = 0;
414 inc_src = a_cs;
415 cs_dst = a_cs;
416 rs_dst = 0;
417 inc_dst = 1 * a_rs;
418 }
419 else // if ( a_rs > a_cs )
420 {
421 cs_src = 0;
422 rs_src = a_rs;
423 inc_src = 1 * a_cs;
424 cs_dst = 0;
425 rs_dst = 1 * a_cs;
426 inc_dst = a_rs;
427 }
428 }
429 else // if ( bl1_is_gen_storage( a_rs, a_cs ) && bl1_is_upper( uplo ) )
430 {
431 // General stride with column-major tilt looks similar to column-major.
432 // General stride with row-major tilt looks similar to row-major.
433 if ( a_rs < a_cs )
434 {
435 cs_src = a_cs;
436 rs_src = 0;
437 inc_src = 1 * a_rs;
438 cs_dst = 1 * a_rs;
439 rs_dst = 0;
440 inc_dst = a_cs;
441 }
442 else // if ( a_rs > a_cs )
443 {
444 cs_src = 0;
445 rs_src = 1 * a_cs;
446 inc_src = a_rs;
447 cs_dst = 0;
448 rs_dst = a_rs;
449 inc_dst = 1 * a_cs;
450 }
451 }
452
453 for ( j = 0; j < n_iter; j++ )
454 {
455 a_src = a + j*cs_src + j*rs_src;
456 a_dst = a + j*cs_dst + j*rs_dst;
457
458 bl1_zcopyv( conj,
459 j,
460 a_src, inc_src,
461 a_dst, inc_dst );
462
463 if ( bl1_is_conj( conj ) )
464 {
465 a_jj = a + j*a_rs + j*a_cs;
466 a_jj->imag = bl1_d0();
467 }
468 }
469}
bl1_zcopyv
void bl1_zcopyv(conj1_t conj, int m, dcomplex *x, int incx, dcomplex *y, int incy)
Definition bl1_copyv.c:63

References bl1_d0(), bl1_is_col_storage(), bl1_is_conj(), bl1_is_gen_storage(), bl1_is_lower(), bl1_is_row_storage(), bl1_is_upper(), bl1_zcopyv(), bl1_zero_dim1(), and dcomplex::imag.

Referenced by FLA_Hermitianize(), and FLA_Symmetrize().

Generated on Thu Sep 23 2021 17:46:17 for libflame by doxygen 1.9.8