bl1_invscalm.c File Reference

(r12600)

Functions
void	bl1_sinvscalm (conj1_t conj, int m, int n, float *alpha, float *a, int a_rs, int a_cs)
void	bl1_dinvscalm (conj1_t conj, int m, int n, double *alpha, double *a, int a_rs, int a_cs)
void	bl1_csinvscalm (conj1_t conj, int m, int n, float *alpha, scomplex *a, int a_rs, int a_cs)
void	bl1_cinvscalm (conj1_t conj, int m, int n, scomplex *alpha, scomplex *a, int a_rs, int a_cs)
void	bl1_zdinvscalm (conj1_t conj, int m, int n, double *alpha, dcomplex *a, int a_rs, int a_cs)
void	bl1_zinvscalm (conj1_t conj, int m, int n, dcomplex *alpha, dcomplex *a, int a_rs, int a_cs)

---

Function Documentation

void bl1_cinvscalm	(	conj1_t	conj,
		int	m,
		int	n,
		scomplex *	alpha,
		scomplex *	a,
		int	a_rs,
		int	a_cs
	)

References bl1_cinvert2s(), bl1_cscal(), bl1_is_row_storage(), bl1_is_vector(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Inv_scal_external(), and FLA_Inv_scalc_external().

{
    scomplex  alpha_inv;
    scomplex* a_begin;
    int       lda, inca;
    int       n_iter;
    int       n_elem;
    int       j;

    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
    if ( bl1_ceq1( alpha ) ) return;

    // Handle cases where A is a vector to ensure that the underlying axpy
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for a vector.
        n_iter = 1;
        n_elem = bl1_vector_dim( m, n );
        lda    = 1; // multiplied by zero when n_iter == 1; not needed.
        inca   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, a_rs, a_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;

        // An optimization: if A is row-major, then let's access the matrix
        // by rows instead of by columns to increase spatial locality.
        if ( bl1_is_row_storage( a_rs, a_cs ) )
        {
            bl1_swap_ints( n_iter, n_elem );
            bl1_swap_ints( lda, inca );
        }
    }

    bl1_cinvert2s( conj, alpha, &alpha_inv );

    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;

        bl1_cscal( n_elem,
                   &alpha_inv,
                   a_begin, inca );
    }
}

void bl1_csinvscalm	(	conj1_t	conj,
		int	m,
		int	n,
		float *	alpha,
		scomplex *	a,
		int	a_rs,
		int	a_cs
	)

References bl1_csscal(), bl1_is_row_storage(), bl1_is_vector(), bl1_sinvert2s(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Inv_scal_external(), and FLA_Inv_scalc_external().

{
    float     alpha_inv;
    scomplex* a_begin;
    int       lda, inca;
    int       n_iter;
    int       n_elem;
    int       j;

    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
    if ( bl1_seq1( alpha ) ) return;

    // Handle cases where A is a vector to ensure that the underlying axpy
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for a vector.
        n_iter = 1;
        n_elem = bl1_vector_dim( m, n );
        lda    = 1; // multiplied by zero when n_iter == 1; not needed.
        inca   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, a_rs, a_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;

        // An optimization: if A is row-major, then let's access the matrix
        // by rows instead of by columns to increase spatial locality.
        if ( bl1_is_row_storage( a_rs, a_cs ) )
        {
            bl1_swap_ints( n_iter, n_elem );
            bl1_swap_ints( lda, inca );
        }
    }

    bl1_sinvert2s( conj, alpha, &alpha_inv );

    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;

        bl1_csscal( n_elem,
                    &alpha_inv,
                    a_begin, inca );
    }
}

void bl1_dinvscalm	(	conj1_t	conj,
		int	m,
		int	n,
		double *	alpha,
		double *	a,
		int	a_rs,
		int	a_cs
	)

References bl1_dinvert2s(), bl1_dscal(), bl1_is_row_storage(), bl1_is_vector(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Inv_scal_external(), and FLA_Inv_scalc_external().

{
    double    alpha_inv;
    double*   a_begin;
    int       lda, inca;
    int       n_iter;
    int       n_elem;
    int       j;

    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
    if ( bl1_deq1( alpha ) ) return;

    // Handle cases where A is a vector to ensure that the underlying axpy
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for a vector.
        n_iter = 1;
        n_elem = bl1_vector_dim( m, n );
        lda    = 1; // multiplied by zero when n_iter == 1; not needed.
        inca   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, a_rs, a_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;

        // An optimization: if A is row-major, then let's access the matrix
        // by rows instead of by columns to increase spatial locality.
        if ( bl1_is_row_storage( a_rs, a_cs ) )
        {
            bl1_swap_ints( n_iter, n_elem );
            bl1_swap_ints( lda, inca );
        }
    }

    bl1_dinvert2s( conj, alpha, &alpha_inv );

    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;

        bl1_dscal( n_elem,
                   &alpha_inv,
                   a_begin, inca );
    }
}

void bl1_sinvscalm	(	conj1_t	conj,
		int	m,
		int	n,
		float *	alpha,
		float *	a,
		int	a_rs,
		int	a_cs
	)

References bl1_is_row_storage(), bl1_is_vector(), bl1_sinvert2s(), bl1_sscal(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Inv_scal_external(), and FLA_Inv_scalc_external().

{
    float     alpha_inv;
    float*    a_begin;
    int       lda, inca;
    int       n_iter;
    int       n_elem;
    int       j;

    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
    if ( bl1_seq1( alpha ) ) return;

    // Handle cases where A is a vector to ensure that the underlying axpy
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for a vector.
        n_iter = 1;
        n_elem = bl1_vector_dim( m, n );
        lda    = 1; // multiplied by zero when n_iter == 1; not needed.
        inca   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, a_rs, a_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;

        // An optimization: if A is row-major, then let's access the matrix
        // by rows instead of by columns to increase spatial locality.
        if ( bl1_is_row_storage( a_rs, a_cs ) )
        {
            bl1_swap_ints( n_iter, n_elem );
            bl1_swap_ints( lda, inca );
        }
    }

    bl1_sinvert2s( conj, alpha, &alpha_inv );

    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;

        bl1_sscal( n_elem,
                   &alpha_inv,
                   a_begin, inca );
    }
}

void bl1_zdinvscalm	(	conj1_t	conj,
		int	m,
		int	n,
		double *	alpha,
		dcomplex *	a,
		int	a_rs,
		int	a_cs
	)

References bl1_dinvert2s(), bl1_is_row_storage(), bl1_is_vector(), bl1_vector_dim(), bl1_vector_inc(), bl1_zdscal(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Inv_scal_external(), and FLA_Inv_scalc_external().

{
    double    alpha_inv;
    dcomplex* a_begin;
    int       lda, inca;
    int       n_iter;
    int       n_elem;
    int       j;

    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
    if ( bl1_deq1( alpha ) ) return;

    // Handle cases where A is a vector to ensure that the underlying axpy
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for a vector.
        n_iter = 1;
        n_elem = bl1_vector_dim( m, n );
        lda    = 1; // multiplied by zero when n_iter == 1; not needed.
        inca   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, a_rs, a_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;

        // An optimization: if A is row-major, then let's access the matrix
        // by rows instead of by columns to increase spatial locality.
        if ( bl1_is_row_storage( a_rs, a_cs ) )
        {
            bl1_swap_ints( n_iter, n_elem );
            bl1_swap_ints( lda, inca );
        }
    }

    bl1_dinvert2s( conj, alpha, &alpha_inv );

    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;

        bl1_zdscal( n_elem,
                    &alpha_inv,
                    a_begin, inca );
    }
}

void bl1_zinvscalm	(	conj1_t	conj,
		int	m,
		int	n,
		dcomplex *	alpha,
		dcomplex *	a,
		int	a_rs,
		int	a_cs
	)

References bl1_is_row_storage(), bl1_is_vector(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), bl1_zinvert2s(), bl1_zscal(), and BLIS1_NO_TRANSPOSE.

Referenced by FLA_Inv_scal_external(), and FLA_Inv_scalc_external().

{
    dcomplex  alpha_inv;
    dcomplex* a_begin;
    int       lda, inca;
    int       n_iter;
    int       n_elem;
    int       j;

    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
    if ( bl1_zeq1( alpha ) ) return;

    // Handle cases where A is a vector to ensure that the underlying axpy
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for a vector.
        n_iter = 1;
        n_elem = bl1_vector_dim( m, n );
        lda    = 1; // multiplied by zero when n_iter == 1; not needed.
        inca   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, a_rs, a_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;

        // An optimization: if A is row-major, then let's access the matrix
        // by rows instead of by columns to increase spatial locality.
        if ( bl1_is_row_storage( a_rs, a_cs ) )
        {
            bl1_swap_ints( n_iter, n_elem );
            bl1_swap_ints( lda, inca );
        }
    }

    bl1_zinvert2s( conj, alpha, &alpha_inv );

    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;

        bl1_zscal( n_elem,
                   &alpha_inv,
                   a_begin, inca );
    }
}