|
libflame
12600
|
Functions | |
| void | bl1_sgemv (trans1_t transa, conj1_t conjx, int m, int n, float *alpha, float *a, int a_rs, int a_cs, float *x, int incx, float *beta, float *y, int incy) |
| void | bl1_dgemv (trans1_t transa, conj1_t conjx, int m, int n, double *alpha, double *a, int a_rs, int a_cs, double *x, int incx, double *beta, double *y, int incy) |
| void | bl1_cgemv (trans1_t transa, conj1_t conjx, int m, int n, scomplex *alpha, scomplex *a, int a_rs, int a_cs, scomplex *x, int incx, scomplex *beta, scomplex *y, int incy) |
| void | bl1_zgemv (trans1_t transa, conj1_t conjx, int m, int n, dcomplex *alpha, dcomplex *a, int a_rs, int a_cs, dcomplex *x, int incx, dcomplex *beta, dcomplex *y, int incy) |
| void | bl1_sgemv_blas (trans1_t transa, int m, int n, float *alpha, float *a, int lda, float *x, int incx, float *beta, float *y, int incy) |
| void | bl1_dgemv_blas (trans1_t transa, int m, int n, double *alpha, double *a, int lda, double *x, int incx, double *beta, double *y, int incy) |
| void | bl1_cgemv_blas (trans1_t transa, int m, int n, scomplex *alpha, scomplex *a, int lda, scomplex *x, int incx, scomplex *beta, scomplex *y, int incy) |
| void | bl1_zgemv_blas (trans1_t transa, int m, int n, dcomplex *alpha, dcomplex *a, int lda, dcomplex *x, int incx, dcomplex *beta, dcomplex *y, int incy) |
| void bl1_cgemv | ( | trans1_t | transa, |
| conj1_t | conjx, | ||
| int | m, | ||
| int | n, | ||
| scomplex * | alpha, | ||
| scomplex * | a, | ||
| int | a_rs, | ||
| int | a_cs, | ||
| scomplex * | x, | ||
| int | incx, | ||
| scomplex * | beta, | ||
| scomplex * | y, | ||
| int | incy | ||
| ) |
References bl1_c0(), bl1_c1(), bl1_callocv(), bl1_caxpyv(), bl1_cconjv(), bl1_ccopyv(), bl1_ccreate_contigm(), bl1_cfree(), bl1_cfree_contigm(), bl1_cgemv_blas(), bl1_cscalv(), bl1_does_trans(), bl1_is_conj(), bl1_is_conjnotrans(), bl1_is_row_storage(), bl1_zero_dim2(), BLIS1_CONJUGATE, BLIS1_NO_CONJUGATE, and BLIS1_NO_TRANSPOSE.
Referenced by FLA_Accum_T_UT_fc_opc_var1(), FLA_Accum_T_UT_fr_opc_var1(), FLA_Apply_H2_UT_l_opc_var1(), FLA_Apply_H2_UT_r_opc_var1(), FLA_Apply_HUD_UT_l_opc_var1(), FLA_Bidiag_UT_u_step_ofc_var2(), FLA_Bidiag_UT_u_step_ofc_var3(), FLA_Bidiag_UT_u_step_ofc_var4(), FLA_Bidiag_UT_u_step_opc_var1(), FLA_Bidiag_UT_u_step_opc_var2(), FLA_Bidiag_UT_u_step_opc_var3(), FLA_Bidiag_UT_u_step_opc_var4(), FLA_Bidiag_UT_u_step_opc_var5(), FLA_CAQR2_UT_opc_var1(), FLA_Chol_l_opc_var2(), FLA_Chol_u_opc_var2(), FLA_Eig_gest_il_opc_var2(), FLA_Eig_gest_il_opc_var3(), FLA_Eig_gest_iu_opc_var2(), FLA_Eig_gest_iu_opc_var3(), FLA_Eig_gest_nl_opc_var2(), FLA_Eig_gest_nu_opc_var2(), FLA_Gemv_external(), FLA_Gemvc_external(), FLA_Hess_UT_step_ofc_var2(), FLA_Hess_UT_step_ofc_var3(), FLA_Hess_UT_step_ofc_var4(), FLA_Hess_UT_step_opc_var1(), FLA_Hess_UT_step_opc_var2(), FLA_Hess_UT_step_opc_var3(), FLA_Hess_UT_step_opc_var4(), FLA_Hess_UT_step_opc_var5(), FLA_LQ_UT_opc_var2(), FLA_LU_nopiv_opc_var2(), FLA_LU_nopiv_opc_var3(), FLA_LU_nopiv_opc_var4(), FLA_LU_piv_opc_var3(), FLA_LU_piv_opc_var4(), FLA_Lyap_h_opc_var2(), FLA_Lyap_h_opc_var3(), FLA_Lyap_n_opc_var2(), FLA_Lyap_n_opc_var3(), FLA_QR2_UT_opc_var1(), FLA_QR_UT_opc_var2(), FLA_Tridiag_UT_l_step_ofc_var2(), FLA_Tridiag_UT_l_step_ofc_var3(), FLA_Tridiag_UT_l_step_opc_var1(), FLA_Tridiag_UT_l_step_opc_var2(), FLA_Tridiag_UT_l_step_opc_var3(), FLA_Ttmm_l_opc_var2(), and FLA_Ttmm_u_opc_var2().
{
scomplex* a_save = a;
int a_rs_save = a_rs;
int a_cs_save = a_cs;
scomplex zero = bl1_c0();
scomplex one = bl1_c1();
scomplex* x_conj;
scomplex* ax;
int lda, inca;
int n_x;
int incx_conj;
int incax;
// Return early if possible.
if ( bl1_zero_dim2( m, n ) )
{
int n_elem;
if ( bl1_does_trans( transa ) ) n_elem = n;
else n_elem = m;
bl1_cscalv( BLIS1_NO_CONJUGATE,
n_elem,
beta,
y, incy );
return;
}
// If necessary, allocate, initialize, and use a temporary contiguous
// copy of the matrix rather than the original matrix.
bl1_ccreate_contigm( m,
n,
a_save, a_rs_save, a_cs_save,
&a, &a_rs, &a_cs );
// Initialize with values assuming column-major storage.
lda = a_cs;
inca = a_rs;
// If A is a row-major matrix, then we can use the underlying column-major
// BLAS implementation by fiddling with the parameters.
if ( bl1_is_row_storage( a_rs, a_cs ) )
{
bl1_swap_ints( m, n );
bl1_swap_ints( lda, inca );
bl1_toggle_trans( transa );
}
// Initialize with values assuming no conjugation of x.
x_conj = x;
incx_conj = incx;
// We need a temporary vector for the cases when x is conjugated, and
// also for the cases where A is conjugated.
if ( bl1_is_conj( conjx ) || bl1_is_conjnotrans( transa ) )
{
if ( bl1_does_trans( transa ) ) n_x = m;
else n_x = n;
x_conj = bl1_callocv( n_x );
incx_conj = 1;
bl1_ccopyv( conjx,
n_x,
x, incx,
x_conj, incx_conj );
}
// We want to handle the conjnotrans case, but without explicitly
// conjugating A. To do so, we leverage the fact that computing the
// product conj(A) * x is equivalent to computing conj( A * conj(x) ).
if ( bl1_is_conjnotrans( transa ) )
{
// We need a temporary vector for the product A * conj(x), which is
// conformal to y. We know we are not transposing, so y is length m.
ax = bl1_callocv( m );
incax = 1;
// Start by conjugating the contents of the temporary copy of x.
bl1_cconjv( n,
x_conj, incx_conj );
// Compute A * conj(x) where x is the temporary copy of x created above.
bl1_cgemv_blas( BLIS1_NO_TRANSPOSE,
m,
n,
&one,
a, lda,
x_conj, incx_conj,
&zero,
ax, incax );
// Scale y by beta.
bl1_cscalv( BLIS1_NO_CONJUGATE,
m,
beta,
y, incy );
// And finally, accumulate alpha * conj( A * conj(x) ) into y.
bl1_caxpyv( BLIS1_CONJUGATE,
m,
alpha,
ax, incax,
y, incy);
// Free the temporary vector for Ax.
bl1_cfree( ax );
}
else // notrans, trans, or conjtrans
{
bl1_cgemv_blas( transa,
m,
n,
alpha,
a, lda,
x_conj, incx_conj,
beta,
y, incy );
}
// Free the temporary conjugated x vector.
if ( bl1_is_conj( conjx ) || bl1_is_conjnotrans( transa ) )
bl1_cfree( x_conj );
// Free the temporary contiguous matrix.
bl1_cfree_contigm( a_save, a_rs_save, a_cs_save,
&a, &a_rs, &a_cs );
}
| void bl1_cgemv_blas | ( | trans1_t | transa, |
| int | m, | ||
| int | n, | ||
| scomplex * | alpha, | ||
| scomplex * | a, | ||
| int | lda, | ||
| scomplex * | x, | ||
| int | incx, | ||
| scomplex * | beta, | ||
| scomplex * | y, | ||
| int | incy | ||
| ) |
References bl1_param_map_to_netlib_trans(), cblas_cgemv(), CblasColMajor, and F77_cgemv().
Referenced by bl1_cgemv().
{
#ifdef BLIS1_ENABLE_CBLAS_INTERFACES
enum CBLAS_ORDER cblas_order = CblasColMajor;
enum CBLAS_TRANSPOSE cblas_transa;
bl1_param_map_to_netlib_trans( transa, &cblas_transa );
cblas_cgemv( cblas_order,
cblas_transa,
m,
n,
alpha,
a, lda,
x, incx,
beta,
y, incy );
#else
char blas_transa;
bl1_param_map_to_netlib_trans( transa, &blas_transa );
F77_cgemv( &blas_transa,
&m,
&n,
alpha,
a, &lda,
x, &incx,
beta,
y, &incy );
#endif
}
| void bl1_dgemv | ( | trans1_t | transa, |
| conj1_t | conjx, | ||
| int | m, | ||
| int | n, | ||
| double * | alpha, | ||
| double * | a, | ||
| int | a_rs, | ||
| int | a_cs, | ||
| double * | x, | ||
| int | incx, | ||
| double * | beta, | ||
| double * | y, | ||
| int | incy | ||
| ) |
References bl1_dcreate_contigm(), bl1_dfree_contigm(), bl1_dgemv_blas(), bl1_does_trans(), bl1_dscalv(), bl1_is_row_storage(), bl1_zero_dim2(), and BLIS1_NO_CONJUGATE.
Referenced by FLA_Accum_T_UT_fc_opd_var1(), FLA_Accum_T_UT_fr_opd_var1(), FLA_Apply_H2_UT_l_opd_var1(), FLA_Apply_H2_UT_r_opd_var1(), FLA_Apply_HUD_UT_l_opd_var1(), FLA_Bidiag_UT_u_step_ofd_var2(), FLA_Bidiag_UT_u_step_ofd_var3(), FLA_Bidiag_UT_u_step_ofd_var4(), FLA_Bidiag_UT_u_step_opd_var1(), FLA_Bidiag_UT_u_step_opd_var2(), FLA_Bidiag_UT_u_step_opd_var3(), FLA_Bidiag_UT_u_step_opd_var4(), FLA_Bidiag_UT_u_step_opd_var5(), FLA_CAQR2_UT_opd_var1(), FLA_Chol_l_opd_var2(), FLA_Chol_u_opd_var2(), FLA_Eig_gest_il_opd_var2(), FLA_Eig_gest_il_opd_var3(), FLA_Eig_gest_iu_opd_var2(), FLA_Eig_gest_iu_opd_var3(), FLA_Eig_gest_nl_opd_var2(), FLA_Eig_gest_nu_opd_var2(), FLA_Gemv_external(), FLA_Gemvc_external(), FLA_Hess_UT_step_ofd_var2(), FLA_Hess_UT_step_ofd_var3(), FLA_Hess_UT_step_ofd_var4(), FLA_Hess_UT_step_opd_var1(), FLA_Hess_UT_step_opd_var2(), FLA_Hess_UT_step_opd_var3(), FLA_Hess_UT_step_opd_var4(), FLA_Hess_UT_step_opd_var5(), FLA_LQ_UT_opd_var2(), FLA_LU_nopiv_opd_var2(), FLA_LU_nopiv_opd_var3(), FLA_LU_nopiv_opd_var4(), FLA_LU_piv_opd_var3(), FLA_LU_piv_opd_var4(), FLA_Lyap_h_opd_var2(), FLA_Lyap_h_opd_var3(), FLA_Lyap_n_opd_var2(), FLA_Lyap_n_opd_var3(), FLA_QR2_UT_opd_var1(), FLA_QR_UT_opd_var2(), FLA_Tridiag_UT_l_step_ofd_var2(), FLA_Tridiag_UT_l_step_ofd_var3(), FLA_Tridiag_UT_l_step_opd_var1(), FLA_Tridiag_UT_l_step_opd_var2(), FLA_Tridiag_UT_l_step_opd_var3(), FLA_Ttmm_l_opd_var2(), and FLA_Ttmm_u_opd_var2().
{
double* a_save = a;
int a_rs_save = a_rs;
int a_cs_save = a_cs;
int lda, inca;
// Return early if possible.
if ( bl1_zero_dim2( m, n ) )
{
int n_elem;
if ( bl1_does_trans( transa ) ) n_elem = n;
else n_elem = m;
bl1_dscalv( BLIS1_NO_CONJUGATE,
n_elem,
beta,
y, incy );
return;
}
// If necessary, allocate, initialize, and use a temporary contiguous
// copy of the matrix rather than the original matrix.
bl1_dcreate_contigm( m,
n,
a_save, a_rs_save, a_cs_save,
&a, &a_rs, &a_cs );
// Initialize with values assuming column-major storage.
lda = a_cs;
inca = a_rs;
// If A is a row-major matrix, then we can use the underlying column-major
// BLAS implementation by fiddling with the parameters.
if ( bl1_is_row_storage( a_rs, a_cs ) )
{
bl1_swap_ints( m, n );
bl1_swap_ints( lda, inca );
bl1_toggle_trans( transa );
}
bl1_dgemv_blas( transa,
m,
n,
alpha,
a, lda,
x, incx,
beta,
y, incy );
// Free the temporary contiguous matrix.
bl1_dfree_contigm( a_save, a_rs_save, a_cs_save,
&a, &a_rs, &a_cs );
}
| void bl1_dgemv_blas | ( | trans1_t | transa, |
| int | m, | ||
| int | n, | ||
| double * | alpha, | ||
| double * | a, | ||
| int | lda, | ||
| double * | x, | ||
| int | incx, | ||
| double * | beta, | ||
| double * | y, | ||
| int | incy | ||
| ) |
References bl1_param_map_to_netlib_trans(), cblas_dgemv(), CblasColMajor, and F77_dgemv().
Referenced by bl1_dgemv().
{
#ifdef BLIS1_ENABLE_CBLAS_INTERFACES
enum CBLAS_ORDER cblas_order = CblasColMajor;
enum CBLAS_TRANSPOSE cblas_transa;
bl1_param_map_to_netlib_trans( transa, &cblas_transa );
cblas_dgemv( cblas_order,
cblas_transa,
m,
n,
*alpha,
a, lda,
x, incx,
*beta,
y, incy );
#else
char blas_transa;
bl1_param_map_to_netlib_trans( transa, &blas_transa );
F77_dgemv( &blas_transa,
&m,
&n,
alpha,
a, &lda,
x, &incx,
beta,
y, &incy );
#endif
}
| void bl1_sgemv | ( | trans1_t | transa, |
| conj1_t | conjx, | ||
| int | m, | ||
| int | n, | ||
| float * | alpha, | ||
| float * | a, | ||
| int | a_rs, | ||
| int | a_cs, | ||
| float * | x, | ||
| int | incx, | ||
| float * | beta, | ||
| float * | y, | ||
| int | incy | ||
| ) |
References bl1_does_trans(), bl1_is_row_storage(), bl1_screate_contigm(), bl1_sfree_contigm(), bl1_sgemv_blas(), bl1_sscalv(), bl1_zero_dim2(), and BLIS1_NO_CONJUGATE.
Referenced by FLA_Accum_T_UT_fc_ops_var1(), FLA_Accum_T_UT_fr_ops_var1(), FLA_Apply_H2_UT_l_ops_var1(), FLA_Apply_H2_UT_r_ops_var1(), FLA_Apply_HUD_UT_l_ops_var1(), FLA_Bidiag_UT_u_step_ofs_var2(), FLA_Bidiag_UT_u_step_ofs_var3(), FLA_Bidiag_UT_u_step_ofs_var4(), FLA_Bidiag_UT_u_step_ops_var1(), FLA_Bidiag_UT_u_step_ops_var2(), FLA_Bidiag_UT_u_step_ops_var3(), FLA_Bidiag_UT_u_step_ops_var4(), FLA_Bidiag_UT_u_step_ops_var5(), FLA_CAQR2_UT_ops_var1(), FLA_Chol_l_ops_var2(), FLA_Chol_u_ops_var2(), FLA_Eig_gest_il_ops_var2(), FLA_Eig_gest_il_ops_var3(), FLA_Eig_gest_iu_ops_var2(), FLA_Eig_gest_iu_ops_var3(), FLA_Eig_gest_nl_ops_var2(), FLA_Eig_gest_nu_ops_var2(), FLA_Gemv_external(), FLA_Gemvc_external(), FLA_Hess_UT_step_ofs_var2(), FLA_Hess_UT_step_ofs_var3(), FLA_Hess_UT_step_ofs_var4(), FLA_Hess_UT_step_ops_var1(), FLA_Hess_UT_step_ops_var2(), FLA_Hess_UT_step_ops_var3(), FLA_Hess_UT_step_ops_var4(), FLA_Hess_UT_step_ops_var5(), FLA_LQ_UT_ops_var2(), FLA_LU_nopiv_ops_var2(), FLA_LU_nopiv_ops_var3(), FLA_LU_nopiv_ops_var4(), FLA_LU_piv_ops_var3(), FLA_LU_piv_ops_var4(), FLA_Lyap_h_ops_var2(), FLA_Lyap_h_ops_var3(), FLA_Lyap_n_ops_var2(), FLA_Lyap_n_ops_var3(), FLA_QR2_UT_ops_var1(), FLA_QR_UT_ops_var2(), FLA_Tridiag_UT_l_step_ofs_var2(), FLA_Tridiag_UT_l_step_ofs_var3(), FLA_Tridiag_UT_l_step_ops_var1(), FLA_Tridiag_UT_l_step_ops_var2(), FLA_Tridiag_UT_l_step_ops_var3(), FLA_Ttmm_l_ops_var2(), and FLA_Ttmm_u_ops_var2().
{
float* a_save = a;
int a_rs_save = a_rs;
int a_cs_save = a_cs;
int lda, inca;
// Return early if possible.
if ( bl1_zero_dim2( m, n ) )
{
int n_elem;
if ( bl1_does_trans( transa ) ) n_elem = n;
else n_elem = m;
bl1_sscalv( BLIS1_NO_CONJUGATE,
n_elem,
beta,
y, incy );
return;
}
// If necessary, allocate, initialize, and use a temporary contiguous
// copy of the matrix rather than the original matrix.
bl1_screate_contigm( m,
n,
a_save, a_rs_save, a_cs_save,
&a, &a_rs, &a_cs );
// Initialize with values assuming column-major storage.
lda = a_cs;
inca = a_rs;
// If A is a row-major matrix, then we can use the underlying column-major
// BLAS implementation by fiddling with the parameters.
if ( bl1_is_row_storage( a_rs, a_cs ) )
{
bl1_swap_ints( m, n );
bl1_swap_ints( lda, inca );
bl1_toggle_trans( transa );
}
bl1_sgemv_blas( transa,
m,
n,
alpha,
a, lda,
x, incx,
beta,
y, incy );
// Free the temporary contiguous matrix.
bl1_sfree_contigm( a_save, a_rs_save, a_cs_save,
&a, &a_rs, &a_cs );
}
| void bl1_sgemv_blas | ( | trans1_t | transa, |
| int | m, | ||
| int | n, | ||
| float * | alpha, | ||
| float * | a, | ||
| int | lda, | ||
| float * | x, | ||
| int | incx, | ||
| float * | beta, | ||
| float * | y, | ||
| int | incy | ||
| ) |
References bl1_param_map_to_netlib_trans(), cblas_sgemv(), CblasColMajor, and F77_sgemv().
Referenced by bl1_sgemv().
{
#ifdef BLIS1_ENABLE_CBLAS_INTERFACES
enum CBLAS_ORDER cblas_order = CblasColMajor;
enum CBLAS_TRANSPOSE cblas_transa;
bl1_param_map_to_netlib_trans( transa, &cblas_transa );
cblas_sgemv( cblas_order,
cblas_transa,
m,
n,
*alpha,
a, lda,
x, incx,
*beta,
y, incy );
#else
char blas_transa;
bl1_param_map_to_netlib_trans( transa, &blas_transa );
F77_sgemv( &blas_transa,
&m,
&n,
alpha,
a, &lda,
x, &incx,
beta,
y, &incy );
#endif
}
| void bl1_zgemv | ( | trans1_t | transa, |
| conj1_t | conjx, | ||
| int | m, | ||
| int | n, | ||
| dcomplex * | alpha, | ||
| dcomplex * | a, | ||
| int | a_rs, | ||
| int | a_cs, | ||
| dcomplex * | x, | ||
| int | incx, | ||
| dcomplex * | beta, | ||
| dcomplex * | y, | ||
| int | incy | ||
| ) |
References bl1_does_trans(), bl1_is_conj(), bl1_is_conjnotrans(), bl1_is_row_storage(), bl1_z0(), bl1_z1(), bl1_zallocv(), bl1_zaxpyv(), bl1_zconjv(), bl1_zcopyv(), bl1_zcreate_contigm(), bl1_zero_dim2(), bl1_zfree(), bl1_zfree_contigm(), bl1_zgemv_blas(), bl1_zscalv(), BLIS1_CONJUGATE, BLIS1_NO_CONJUGATE, and BLIS1_NO_TRANSPOSE.
Referenced by FLA_Accum_T_UT_fc_opz_var1(), FLA_Accum_T_UT_fr_opz_var1(), FLA_Apply_H2_UT_l_opz_var1(), FLA_Apply_H2_UT_r_opz_var1(), FLA_Apply_HUD_UT_l_opz_var1(), FLA_Bidiag_UT_u_step_ofz_var2(), FLA_Bidiag_UT_u_step_ofz_var3(), FLA_Bidiag_UT_u_step_ofz_var4(), FLA_Bidiag_UT_u_step_opz_var1(), FLA_Bidiag_UT_u_step_opz_var2(), FLA_Bidiag_UT_u_step_opz_var3(), FLA_Bidiag_UT_u_step_opz_var4(), FLA_Bidiag_UT_u_step_opz_var5(), FLA_CAQR2_UT_opz_var1(), FLA_Chol_l_opz_var2(), FLA_Chol_u_opz_var2(), FLA_Eig_gest_il_opz_var2(), FLA_Eig_gest_il_opz_var3(), FLA_Eig_gest_iu_opz_var2(), FLA_Eig_gest_iu_opz_var3(), FLA_Eig_gest_nl_opz_var2(), FLA_Eig_gest_nu_opz_var2(), FLA_Gemv_external(), FLA_Gemvc_external(), FLA_Hess_UT_step_ofz_var2(), FLA_Hess_UT_step_ofz_var3(), FLA_Hess_UT_step_ofz_var4(), FLA_Hess_UT_step_opz_var1(), FLA_Hess_UT_step_opz_var2(), FLA_Hess_UT_step_opz_var3(), FLA_Hess_UT_step_opz_var4(), FLA_Hess_UT_step_opz_var5(), FLA_LQ_UT_opz_var2(), FLA_LU_nopiv_opz_var2(), FLA_LU_nopiv_opz_var3(), FLA_LU_nopiv_opz_var4(), FLA_LU_piv_opz_var3(), FLA_LU_piv_opz_var4(), FLA_Lyap_h_opz_var2(), FLA_Lyap_h_opz_var3(), FLA_Lyap_n_opz_var2(), FLA_Lyap_n_opz_var3(), FLA_QR2_UT_opz_var1(), FLA_QR_UT_opz_var2(), FLA_Tridiag_UT_l_step_ofz_var2(), FLA_Tridiag_UT_l_step_ofz_var3(), FLA_Tridiag_UT_l_step_opz_var1(), FLA_Tridiag_UT_l_step_opz_var2(), FLA_Tridiag_UT_l_step_opz_var3(), FLA_Ttmm_l_opz_var2(), and FLA_Ttmm_u_opz_var2().
{
dcomplex* a_save = a;
int a_rs_save = a_rs;
int a_cs_save = a_cs;
dcomplex zero = bl1_z0();
dcomplex one = bl1_z1();
dcomplex* x_conj;
dcomplex* ax;
int lda, inca;
int n_x;
int incx_conj;
int incax;
// Return early if possible.
if ( bl1_zero_dim2( m, n ) )
{
int n_elem;
if ( bl1_does_trans( transa ) ) n_elem = n;
else n_elem = m;
bl1_zscalv( BLIS1_NO_CONJUGATE,
n_elem,
beta,
y, incy );
return;
}
// If necessary, allocate, initialize, and use a temporary contiguous
// copy of the matrix rather than the original matrix.
bl1_zcreate_contigm( m,
n,
a_save, a_rs_save, a_cs_save,
&a, &a_rs, &a_cs );
// Initialize with values assuming column-major storage.
lda = a_cs;
inca = a_rs;
// If A is a row-major matrix, then we can use the underlying column-major
// BLAS implementation by fiddling with the parameters.
if ( bl1_is_row_storage( a_rs, a_cs ) )
{
bl1_swap_ints( m, n );
bl1_swap_ints( lda, inca );
bl1_toggle_trans( transa );
}
// Initialize with values assuming no conjugation of x.
x_conj = x;
incx_conj = incx;
// We need a temporary vector for the cases when x is conjugated, and
// also for the cases where A is conjugated.
if ( bl1_is_conj( conjx ) || bl1_is_conjnotrans( transa ) )
{
if ( bl1_does_trans( transa ) ) n_x = m;
else n_x = n;
x_conj = bl1_zallocv( n_x );
incx_conj = 1;
bl1_zcopyv( conjx,
n_x,
x, incx,
x_conj, incx_conj );
}
// We want to handle the conjnotrans case, but without explicitly
// conjugating A. To do so, we leverage the fact that computing the
// product conj(A) * x is equivalent to computing conj( A * conj(x) ).
if ( bl1_is_conjnotrans( transa ) )
{
// We need a temporary vector for the product A * conj(x), which is
// conformal to y. We know we are not transposing, so y is length m.
ax = bl1_zallocv( m );
incax = 1;
// Start by conjugating the contents of the temporary copy of x.
bl1_zconjv( n,
x_conj, incx_conj );
// Compute A * conj(x) where x is the temporary copy of x created above.
bl1_zgemv_blas( BLIS1_NO_TRANSPOSE,
m,
n,
&one,
a, lda,
x_conj, incx_conj,
&zero,
ax, incax );
// Scale y by beta.
bl1_zscalv( BLIS1_NO_CONJUGATE,
m,
beta,
y, incy );
// And finally, accumulate alpha * conj( A * conj(x) ) into y.
bl1_zaxpyv( BLIS1_CONJUGATE,
m,
alpha,
ax, incax,
y, incy);
// Free the temporary vector for Ax.
bl1_zfree( ax );
}
else // notrans, trans, or conjtrans
{
bl1_zgemv_blas( transa,
m,
n,
alpha,
a, lda,
x_conj, incx_conj,
beta,
y, incy );
}
// Free the temporary conjugated x vector.
if ( bl1_is_conj( conjx ) || bl1_is_conjnotrans( transa ) )
bl1_zfree( x_conj );
// Free the temporary contiguous matrix.
bl1_zfree_contigm( a_save, a_rs_save, a_cs_save,
&a, &a_rs, &a_cs );
}
| void bl1_zgemv_blas | ( | trans1_t | transa, |
| int | m, | ||
| int | n, | ||
| dcomplex * | alpha, | ||
| dcomplex * | a, | ||
| int | lda, | ||
| dcomplex * | x, | ||
| int | incx, | ||
| dcomplex * | beta, | ||
| dcomplex * | y, | ||
| int | incy | ||
| ) |
References bl1_param_map_to_netlib_trans(), cblas_zgemv(), CblasColMajor, and F77_zgemv().
Referenced by bl1_zgemv().
{
#ifdef BLIS1_ENABLE_CBLAS_INTERFACES
enum CBLAS_ORDER cblas_order = CblasColMajor;
enum CBLAS_TRANSPOSE cblas_transa;
bl1_param_map_to_netlib_trans( transa, &cblas_transa );
cblas_zgemv( cblas_order,
cblas_transa,
m,
n,
alpha,
a, lda,
x, incx,
beta,
y, incy );
#else
char blas_transa;
bl1_param_map_to_netlib_trans( transa, &blas_transa );
F77_zgemv( &blas_transa,
&m,
&n,
alpha,
a, &lda,
x, &incx,
beta,
y, &incy );
#endif
}
1.7.6.1