libflame  12600
Functions
bl1_trmv.c File Reference

(r12600)

Functions

void bl1_strmv (uplo1_t uplo, trans1_t trans, diag1_t diag, int m, float *a, int a_rs, int a_cs, float *x, int incx)
void bl1_dtrmv (uplo1_t uplo, trans1_t trans, diag1_t diag, int m, double *a, int a_rs, int a_cs, double *x, int incx)
void bl1_ctrmv (uplo1_t uplo, trans1_t trans, diag1_t diag, int m, scomplex *a, int a_rs, int a_cs, scomplex *x, int incx)
void bl1_ztrmv (uplo1_t uplo, trans1_t trans, diag1_t diag, int m, dcomplex *a, int a_rs, int a_cs, dcomplex *x, int incx)
void bl1_strmv_blas (uplo1_t uplo, trans1_t trans, diag1_t diag, int m, float *a, int lda, float *x, int incx)
void bl1_dtrmv_blas (uplo1_t uplo, trans1_t trans, diag1_t diag, int m, double *a, int lda, double *x, int incx)
void bl1_ctrmv_blas (uplo1_t uplo, trans1_t trans, diag1_t diag, int m, scomplex *a, int lda, scomplex *x, int incx)
void bl1_ztrmv_blas (uplo1_t uplo, trans1_t trans, diag1_t diag, int m, dcomplex *a, int lda, dcomplex *x, int incx)

Function Documentation

void bl1_ctrmv ( uplo1_t  uplo,
trans1_t  trans,
diag1_t  diag,
int  m,
scomplex a,
int  a_rs,
int  a_cs,
scomplex x,
int  incx 
)

References bl1_callocv(), bl1_ccopyv(), bl1_ccreate_contigmr(), bl1_cfree(), bl1_cfree_contigm(), bl1_ctrmv_blas(), bl1_is_conjnotrans(), bl1_is_row_storage(), bl1_zero_dim1(), and BLIS1_CONJUGATE.

Referenced by bl1_ctrmvsx(), FLA_CAQR2_UT_opc_var1(), FLA_Eig_gest_nl_opc_var1(), FLA_Eig_gest_nl_opc_var5(), FLA_Eig_gest_nu_opc_var1(), FLA_Eig_gest_nu_opc_var5(), FLA_Hess_UT_step_opc_var5(), FLA_Trinv_ln_opc_var1(), FLA_Trinv_ln_opc_var4(), FLA_Trinv_lu_opc_var1(), FLA_Trinv_lu_opc_var4(), FLA_Trinv_un_opc_var1(), FLA_Trinv_un_opc_var4(), FLA_Trinv_uu_opc_var1(), FLA_Trinv_uu_opc_var4(), FLA_Trmv_external(), FLA_Ttmm_l_opc_var3(), and FLA_Ttmm_u_opc_var3().

{
    scomplex* a_save    = a;
    int       a_rs_save = a_rs;
    int       a_cs_save = a_cs;
    scomplex* x_conj;
    int       incx_conj;
    int       lda, inca;

    // Return early if possible.
    if ( bl1_zero_dim1( m ) ) return;

    // If necessary, allocate, initialize, and use a temporary contiguous
    // copy of the matrix rather than the original matrix.
    bl1_ccreate_contigmr( uplo,
                          m,
                          m,
                          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( lda, inca );
        bl1_toggle_uplo( uplo );
        bl1_toggle_trans( trans );
    }

    // Initialize with values assuming that trans is not conjnotrans.
    x_conj    = x;
    incx_conj = incx;

    // 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) ).
    // Note: strictly speaking, we don't need to create a copy of x since
    // the operation is simpler than, say, gemv. However, we create a copy
    // anyway since in practice it performs better due to increased spatial
    // locality.
    if ( bl1_is_conjnotrans( trans ) )
    {
        x_conj    = bl1_callocv( m );
        incx_conj = 1;

        bl1_ccopyv( BLIS1_CONJUGATE,
                    m,
                    x,      incx,
                    x_conj, incx_conj );
    }

    bl1_ctrmv_blas( uplo,
                    trans,
                    diag,
                    m,
                    a,      lda,
                    x_conj, incx_conj );

    // Save the contents of and then free the temporary conjugated x vector.
    if ( bl1_is_conjnotrans( trans ) )
    {
        bl1_ccopyv( BLIS1_CONJUGATE,
                    m,
                    x_conj, incx_conj,
                    x,      incx );

        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_ctrmv_blas ( uplo1_t  uplo,
trans1_t  trans,
diag1_t  diag,
int  m,
scomplex a,
int  lda,
scomplex x,
int  incx 
)

References bl1_param_map_to_netlib_diag(), bl1_param_map_to_netlib_trans(), bl1_param_map_to_netlib_uplo(), cblas_ctrmv(), CblasColMajor, and F77_ctrmv().

Referenced by bl1_ctrmv().

{
#ifdef BLIS1_ENABLE_CBLAS_INTERFACES
    enum CBLAS_ORDER cblas_order = CblasColMajor;
    enum CBLAS_UPLO  cblas_uplo;
    enum CBLAS_TRANSPOSE cblas_trans;
    enum CBLAS_DIAG  cblas_diag;

    bl1_param_map_to_netlib_uplo( uplo, &cblas_uplo );
    bl1_param_map_to_netlib_trans( trans, &cblas_trans );
    bl1_param_map_to_netlib_diag( diag, &cblas_diag );

    cblas_ctrmv( cblas_order,
                 cblas_uplo,
                 cblas_trans,
                 cblas_diag,
                 m,
                 a, lda,
                 x, incx );
#else
    char blas_uplo;
    char blas_trans;
    char blas_diag;

    bl1_param_map_to_netlib_uplo( uplo, &blas_uplo );
    bl1_param_map_to_netlib_trans( trans, &blas_trans );
    bl1_param_map_to_netlib_diag( diag, &blas_diag );

    F77_ctrmv( &blas_uplo,
               &blas_trans,
               &blas_diag,
               &m,
               a, &lda,
               x, &incx );
#endif
}
void bl1_dtrmv ( uplo1_t  uplo,
trans1_t  trans,
diag1_t  diag,
int  m,
double *  a,
int  a_rs,
int  a_cs,
double *  x,
int  incx 
)

References bl1_dcreate_contigmr(), bl1_dfree_contigm(), bl1_dtrmv_blas(), bl1_is_row_storage(), and bl1_zero_dim1().

Referenced by bl1_dtrmvsx(), FLA_CAQR2_UT_opd_var1(), FLA_Eig_gest_nl_opd_var1(), FLA_Eig_gest_nl_opd_var5(), FLA_Eig_gest_nu_opd_var1(), FLA_Eig_gest_nu_opd_var5(), FLA_Hess_UT_step_opd_var5(), FLA_Trinv_ln_opd_var1(), FLA_Trinv_ln_opd_var4(), FLA_Trinv_lu_opd_var1(), FLA_Trinv_lu_opd_var4(), FLA_Trinv_un_opd_var1(), FLA_Trinv_un_opd_var4(), FLA_Trinv_uu_opd_var1(), FLA_Trinv_uu_opd_var4(), FLA_Trmv_external(), FLA_Ttmm_l_opd_var3(), and FLA_Ttmm_u_opd_var3().

{
    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_dim1( m ) ) return;

    // If necessary, allocate, initialize, and use a temporary contiguous
    // copy of the matrix rather than the original matrix.
    bl1_dcreate_contigmr( uplo,
                          m,
                          m,
                          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( lda, inca );
        bl1_toggle_uplo( uplo );
        bl1_toggle_trans( trans );
    }

    bl1_dtrmv_blas( uplo,
                    trans,
                    diag,
                    m,
                    a, lda,
                    x, incx );

    // Free the temporary contiguous matrix.
    bl1_dfree_contigm( a_save, a_rs_save, a_cs_save,
                       &a,     &a_rs,     &a_cs );
}
void bl1_dtrmv_blas ( uplo1_t  uplo,
trans1_t  trans,
diag1_t  diag,
int  m,
double *  a,
int  lda,
double *  x,
int  incx 
)

References bl1_param_map_to_netlib_diag(), bl1_param_map_to_netlib_trans(), bl1_param_map_to_netlib_uplo(), cblas_dtrmv(), CblasColMajor, and F77_dtrmv().

Referenced by bl1_dtrmv().

{
#ifdef BLIS1_ENABLE_CBLAS_INTERFACES
    enum CBLAS_ORDER cblas_order = CblasColMajor;
    enum CBLAS_UPLO  cblas_uplo;
    enum CBLAS_TRANSPOSE cblas_trans;
    enum CBLAS_DIAG  cblas_diag;

    bl1_param_map_to_netlib_uplo( uplo, &cblas_uplo );
    bl1_param_map_to_netlib_trans( trans, &cblas_trans );
    bl1_param_map_to_netlib_diag( diag, &cblas_diag );

    cblas_dtrmv( cblas_order,
                 cblas_uplo,
                 cblas_trans,
                 cblas_diag,
                 m,
                 a, lda,
                 x, incx );
#else
    char blas_uplo;
    char blas_trans;
    char blas_diag;

    bl1_param_map_to_netlib_uplo( uplo, &blas_uplo );
    bl1_param_map_to_netlib_trans( trans, &blas_trans );
    bl1_param_map_to_netlib_diag( diag, &blas_diag );

    F77_dtrmv( &blas_uplo,
               &blas_trans,
               &blas_diag,
               &m,
               a, &lda,
               x, &incx );
#endif
}
void bl1_strmv ( uplo1_t  uplo,
trans1_t  trans,
diag1_t  diag,
int  m,
float *  a,
int  a_rs,
int  a_cs,
float *  x,
int  incx 
)

References bl1_is_row_storage(), bl1_screate_contigmr(), bl1_sfree_contigm(), bl1_strmv_blas(), and bl1_zero_dim1().

Referenced by bl1_strmvsx(), FLA_CAQR2_UT_ops_var1(), FLA_Eig_gest_nl_ops_var1(), FLA_Eig_gest_nl_ops_var5(), FLA_Eig_gest_nu_ops_var1(), FLA_Eig_gest_nu_ops_var5(), FLA_Hess_UT_step_ops_var5(), FLA_Trinv_ln_ops_var1(), FLA_Trinv_ln_ops_var4(), FLA_Trinv_lu_ops_var1(), FLA_Trinv_lu_ops_var4(), FLA_Trinv_un_ops_var1(), FLA_Trinv_un_ops_var4(), FLA_Trinv_uu_ops_var1(), FLA_Trinv_uu_ops_var4(), FLA_Trmv_external(), FLA_Ttmm_l_ops_var3(), and FLA_Ttmm_u_ops_var3().

{
    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_dim1( m ) ) return;

    // If necessary, allocate, initialize, and use a temporary contiguous
    // copy of the matrix rather than the original matrix.
    bl1_screate_contigmr( uplo,
                          m,
                          m,
                          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( lda, inca );
        bl1_toggle_uplo( uplo );
        bl1_toggle_trans( trans );
    }

    bl1_strmv_blas( uplo,
                    trans,
                    diag,
                    m,
                    a, lda,
                    x, incx );

    // Free the temporary contiguous matrix.
    bl1_sfree_contigm( a_save, a_rs_save, a_cs_save,
                       &a,     &a_rs,     &a_cs );
}
void bl1_strmv_blas ( uplo1_t  uplo,
trans1_t  trans,
diag1_t  diag,
int  m,
float *  a,
int  lda,
float *  x,
int  incx 
)

References bl1_param_map_to_netlib_diag(), bl1_param_map_to_netlib_trans(), bl1_param_map_to_netlib_uplo(), cblas_strmv(), CblasColMajor, and F77_strmv().

Referenced by bl1_strmv().

{
#ifdef BLIS1_ENABLE_CBLAS_INTERFACES
    enum CBLAS_ORDER cblas_order = CblasColMajor;
    enum CBLAS_UPLO  cblas_uplo;
    enum CBLAS_TRANSPOSE cblas_trans;
    enum CBLAS_DIAG  cblas_diag;

    bl1_param_map_to_netlib_uplo( uplo, &cblas_uplo );
    bl1_param_map_to_netlib_trans( trans, &cblas_trans );
    bl1_param_map_to_netlib_diag( diag, &cblas_diag );

    cblas_strmv( cblas_order,
                 cblas_uplo,
                 cblas_trans,
                 cblas_diag,
                 m,
                 a, lda,
                 x, incx );
#else
    char blas_uplo;
    char blas_trans;
    char blas_diag;

    bl1_param_map_to_netlib_uplo( uplo, &blas_uplo );
    bl1_param_map_to_netlib_trans( trans, &blas_trans );
    bl1_param_map_to_netlib_diag( diag, &blas_diag );

    F77_strmv( &blas_uplo,
               &blas_trans,
               &blas_diag,
               &m,
               a, &lda,
               x, &incx );
#endif
}
void bl1_ztrmv ( uplo1_t  uplo,
trans1_t  trans,
diag1_t  diag,
int  m,
dcomplex a,
int  a_rs,
int  a_cs,
dcomplex x,
int  incx 
)

References bl1_is_conjnotrans(), bl1_is_row_storage(), bl1_zallocv(), bl1_zcopyv(), bl1_zcreate_contigmr(), bl1_zero_dim1(), bl1_zfree(), bl1_zfree_contigm(), bl1_ztrmv_blas(), and BLIS1_CONJUGATE.

Referenced by bl1_ztrmvsx(), FLA_CAQR2_UT_opz_var1(), FLA_Eig_gest_nl_opz_var1(), FLA_Eig_gest_nl_opz_var5(), FLA_Eig_gest_nu_opz_var1(), FLA_Eig_gest_nu_opz_var5(), FLA_Hess_UT_step_opz_var5(), FLA_Trinv_ln_opz_var1(), FLA_Trinv_ln_opz_var4(), FLA_Trinv_lu_opz_var1(), FLA_Trinv_lu_opz_var4(), FLA_Trinv_un_opz_var1(), FLA_Trinv_un_opz_var4(), FLA_Trinv_uu_opz_var1(), FLA_Trinv_uu_opz_var4(), FLA_Trmv_external(), FLA_Ttmm_l_opz_var3(), and FLA_Ttmm_u_opz_var3().

{
    dcomplex* a_save    = a;
    int       a_rs_save = a_rs;
    int       a_cs_save = a_cs;
    dcomplex* x_conj;
    int       incx_conj;
    int       lda, inca;

    // Return early if possible.
    if ( bl1_zero_dim1( m ) ) return;

    // If necessary, allocate, initialize, and use a temporary contiguous
    // copy of the matrix rather than the original matrix.
    bl1_zcreate_contigmr( uplo,
                          m,
                          m,
                          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( lda, inca );
        bl1_toggle_uplo( uplo );
        bl1_toggle_trans( trans );
    }

    // Initialize with values assuming that trans is not conjnotrans.
    x_conj    = x;
    incx_conj = incx;

    // 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) ).
    // Note: strictly speaking, we don't need to create a copy of x since
    // the operation is simpler than, say, gemv. However, we create a copy
    // anyway since in practice it performs better due to increased spatial
    // locality.
    if ( bl1_is_conjnotrans( trans ) )
    {
        x_conj    = bl1_zallocv( m );
        incx_conj = 1;

        bl1_zcopyv( BLIS1_CONJUGATE,
                    m,
                    x,      incx,
                    x_conj, incx_conj );
    }

    bl1_ztrmv_blas( uplo,
                    trans,
                    diag,
                    m,
                    a,      lda,
                    x_conj, incx_conj );

    // Save the contents of and then free the temporary conjugated x vector.
    if ( bl1_is_conjnotrans( trans ) )
    {
        bl1_zcopyv( BLIS1_CONJUGATE,
                    m,
                    x_conj, incx_conj,
                    x,      incx );

        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_ztrmv_blas ( uplo1_t  uplo,
trans1_t  trans,
diag1_t  diag,
int  m,
dcomplex a,
int  lda,
dcomplex x,
int  incx 
)

References bl1_param_map_to_netlib_diag(), bl1_param_map_to_netlib_trans(), bl1_param_map_to_netlib_uplo(), cblas_ztrmv(), CblasColMajor, and F77_ztrmv().

Referenced by bl1_ztrmv().

{
#ifdef BLIS1_ENABLE_CBLAS_INTERFACES
    enum CBLAS_ORDER cblas_order = CblasColMajor;
    enum CBLAS_UPLO  cblas_uplo;
    enum CBLAS_TRANSPOSE cblas_trans;
    enum CBLAS_DIAG  cblas_diag;

    bl1_param_map_to_netlib_uplo( uplo, &cblas_uplo );
    bl1_param_map_to_netlib_trans( trans, &cblas_trans );
    bl1_param_map_to_netlib_diag( diag, &cblas_diag );

    cblas_ztrmv( cblas_order,
                 cblas_uplo,
                 cblas_trans,
                 cblas_diag,
                 m,
                 a, lda,
                 x, incx );
#else
    char blas_uplo;
    char blas_trans;
    char blas_diag;

    bl1_param_map_to_netlib_uplo( uplo, &blas_uplo );
    bl1_param_map_to_netlib_trans( trans, &blas_trans );
    bl1_param_map_to_netlib_diag( diag, &blas_diag );

    F77_ztrmv( &blas_uplo,
               &blas_trans,
               &blas_diag,
               &m,
               a, &lda,
               x, &incx );
#endif
}