bl1_ewinvscalmt.c File Reference

(r)

Functions
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)

Function Documentation

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
	)

{
    scomplex* a_begin;
    scomplex* b_begin;
    int       lda, inca;
    int       ldb, incb;
    int       n_iter;
    int       n_elem;
    int       j;
    conj1_t    conj;
 
    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
 
    // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for vectors.
        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( trans,             m, n, a_rs, a_cs );
        ldb    = 1; // multiplied by zero when n_iter == 1; not needed.
        incb   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;
        ldb    = b_cs;
        incb   = b_rs;
        
        // Handle the transposition of A.
        if ( bl1_does_trans( trans ) )
        {
            bl1_swap_ints( lda, inca );
        }
 
        // An optimization: if B is row-major and if A is effectively row-major
        // after a possible transposition, then let's access the matrices by rows
        // instead of by columns for increased spatial locality.
        if ( bl1_is_row_storage( b_rs, b_cs ) )
        {
            if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
                 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
            {
                bl1_swap_ints( n_iter, n_elem );
                bl1_swap_ints( lda, inca );
                bl1_swap_ints( ldb, incb );
            }
        }
    }
 
    // Extract conj component from trans parameter.
    conj = bl1_proj_trans1_to_conj( trans );
 
    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;
        b_begin = b + j*ldb;
 
        bl1_cewinvscalv( conj,
                         n_elem,
                         a_begin, inca, 
                         b_begin, incb );
    }
}

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_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
	)

{
    float*    a_begin;
    scomplex* b_begin;
    int       lda, inca;
    int       ldb, incb;
    int       n_iter;
    int       n_elem;
    int       j;
    conj1_t    conj;
 
    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
 
    // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for vectors.
        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( trans,             m, n, a_rs, a_cs );
        ldb    = 1; // multiplied by zero when n_iter == 1; not needed.
        incb   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;
        ldb    = b_cs;
        incb   = b_rs;
        
        // Handle the transposition of A.
        if ( bl1_does_trans( trans ) )
        {
            bl1_swap_ints( lda, inca );
        }
 
        // An optimization: if B is row-major and if A is effectively row-major
        // after a possible transposition, then let's access the matrices by rows
        // instead of by columns for increased spatial locality.
        if ( bl1_is_row_storage( b_rs, b_cs ) )
        {
            if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
                 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
            {
                bl1_swap_ints( n_iter, n_elem );
                bl1_swap_ints( lda, inca );
                bl1_swap_ints( ldb, incb );
            }
        }
    }
 
    // Extract conj component from trans parameter.
    conj = bl1_proj_trans1_to_conj( trans );
 
    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;
        b_begin = b + j*ldb;
 
        bl1_csewinvscalv( conj,
                          n_elem,
                          a_begin, inca, 
                          b_begin, incb );
    }
}

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_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
	)

{
    double*   a_begin;
    double*   b_begin;
    int       lda, inca;
    int       ldb, incb;
    int       n_iter;
    int       n_elem;
    int       j;
    conj1_t    conj;
 
    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
 
    // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for vectors.
        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( trans,             m, n, a_rs, a_cs );
        ldb    = 1; // multiplied by zero when n_iter == 1; not needed.
        incb   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;
        ldb    = b_cs;
        incb   = b_rs;
        
        // Handle the transposition of A.
        if ( bl1_does_trans( trans ) )
        {
            bl1_swap_ints( lda, inca );
        }
 
        // An optimization: if B is row-major and if A is effectively row-major
        // after a possible transposition, then let's access the matrices by rows
        // instead of by columns for increased spatial locality.
        if ( bl1_is_row_storage( b_rs, b_cs ) )
        {
            if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
                 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
            {
                bl1_swap_ints( n_iter, n_elem );
                bl1_swap_ints( lda, inca );
                bl1_swap_ints( ldb, incb );
            }
        }
    }
 
    // Extract conj component from trans parameter.
    conj = bl1_proj_trans1_to_conj( trans );
 
    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;
        b_begin = b + j*ldb;
 
        bl1_dewinvscalv( conj,
                         n_elem,
                         a_begin, inca, 
                         b_begin, incb );
    }
}

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_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
	)

{
    float*    a_begin;
    float*    b_begin;
    int       lda, inca;
    int       ldb, incb;
    int       n_iter;
    int       n_elem;
    int       j;
    conj1_t    conj;
 
    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
 
    // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for vectors.
        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( trans,             m, n, a_rs, a_cs );
        ldb    = 1; // multiplied by zero when n_iter == 1; not needed.
        incb   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;
        ldb    = b_cs;
        incb   = b_rs;
        
        // Handle the transposition of A.
        if ( bl1_does_trans( trans ) )
        {
            bl1_swap_ints( lda, inca );
        }
 
        // An optimization: if B is row-major and if A is effectively row-major
        // after a possible transposition, then let's access the matrices by rows
        // instead of by columns for increased spatial locality.
        if ( bl1_is_row_storage( b_rs, b_cs ) )
        {
            if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
                 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
            {
                bl1_swap_ints( n_iter, n_elem );
                bl1_swap_ints( lda, inca );
                bl1_swap_ints( ldb, incb );
            }
        }
    }
 
    // Extract conj component from trans parameter.
    conj = bl1_proj_trans1_to_conj( trans );
 
    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;
        b_begin = b + j*ldb;
 
        bl1_sewinvscalv( conj,
                         n_elem,
                         a_begin, inca, 
                         b_begin, incb );
    }
}

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_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
	)

{
    double*   a_begin;
    dcomplex* b_begin;
    int       lda, inca;
    int       ldb, incb;
    int       n_iter;
    int       n_elem;
    int       j;
    conj1_t    conj;
 
    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
 
    // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for vectors.
        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( trans,             m, n, a_rs, a_cs );
        ldb    = 1; // multiplied by zero when n_iter == 1; not needed.
        incb   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;
        ldb    = b_cs;
        incb   = b_rs;
        
        // Handle the transposition of A.
        if ( bl1_does_trans( trans ) )
        {
            bl1_swap_ints( lda, inca );
        }
 
        // An optimization: if B is row-major and if A is effectively row-major
        // after a possible transposition, then let's access the matrices by rows
        // instead of by columns for increased spatial locality.
        if ( bl1_is_row_storage( b_rs, b_cs ) )
        {
            if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
                 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
            {
                bl1_swap_ints( n_iter, n_elem );
                bl1_swap_ints( lda, inca );
                bl1_swap_ints( ldb, incb );
            }
        }
    }
 
    // Extract conj component from trans parameter.
    conj = bl1_proj_trans1_to_conj( trans );
 
    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;
        b_begin = b + j*ldb;
 
        bl1_zdewinvscalv( conj,
                          n_elem,
                          a_begin, inca, 
                          b_begin, incb );
    }
}

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_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
	)

{
    dcomplex* a_begin;
    dcomplex* b_begin;
    int       lda, inca;
    int       ldb, incb;
    int       n_iter;
    int       n_elem;
    int       j;
    conj1_t    conj;
 
    // Return early if possible.
    if ( bl1_zero_dim2( m, n ) ) return;
 
    // Handle cases where A and B are vectors to ensure that the underlying ewinvscal
    // gets invoked only once.
    if ( bl1_is_vector( m, n ) )
    {
        // Initialize with values appropriate for vectors.
        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( trans,             m, n, a_rs, a_cs );
        ldb    = 1; // multiplied by zero when n_iter == 1; not needed.
        incb   = bl1_vector_inc( BLIS1_NO_TRANSPOSE, m, n, b_rs, b_cs );
    }
    else // matrix case
    {
        // Initialize with optimal values for column-major storage.
        n_iter = n;
        n_elem = m;
        lda    = a_cs;
        inca   = a_rs;
        ldb    = b_cs;
        incb   = b_rs;
        
        // Handle the transposition of A.
        if ( bl1_does_trans( trans ) )
        {
            bl1_swap_ints( lda, inca );
        }
 
        // An optimization: if B is row-major and if A is effectively row-major
        // after a possible transposition, then let's access the matrices by rows
        // instead of by columns for increased spatial locality.
        if ( bl1_is_row_storage( b_rs, b_cs ) )
        {
            if ( ( bl1_is_col_storage( a_rs, a_cs ) && bl1_does_trans( trans ) ) ||
                 ( bl1_is_row_storage( a_rs, a_cs ) && bl1_does_notrans( trans ) ) )
            {
                bl1_swap_ints( n_iter, n_elem );
                bl1_swap_ints( lda, inca );
                bl1_swap_ints( ldb, incb );
            }
        }
    }
 
    // Extract conj component from trans parameter.
    conj = bl1_proj_trans1_to_conj( trans );
 
    for ( j = 0; j < n_iter; j++ )
    {
        a_begin = a + j*lda;
        b_begin = b + j*ldb;
 
        bl1_zewinvscalv( conj,
                         n_elem,
                         a_begin, inca, 
                         b_begin, incb );
    }
}

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().