|
libflame
12600
|
Functions | |
| void | bl1_saxpymt (trans1_t trans, int m, int n, float *alpha, float *a, int a_rs, int a_cs, float *b, int b_rs, int b_cs) |
| void | bl1_daxpymt (trans1_t trans, int m, int n, double *alpha, double *a, int a_rs, int a_cs, double *b, int b_rs, int b_cs) |
| void | bl1_caxpymt (trans1_t trans, int m, int n, scomplex *alpha, scomplex *a, int a_rs, int a_cs, scomplex *b, int b_rs, int b_cs) |
| void | bl1_zaxpymt (trans1_t trans, int m, int n, dcomplex *alpha, dcomplex *a, int a_rs, int a_cs, dcomplex *b, int b_rs, int b_cs) |
| void bl1_caxpymt | ( | trans1_t | trans, |
| int | m, | ||
| int | n, | ||
| scomplex * | alpha, | ||
| scomplex * | a, | ||
| int | a_rs, | ||
| int | a_cs, | ||
| scomplex * | b, | ||
| int | b_rs, | ||
| int | b_cs | ||
| ) |
References bl1_callocv(), bl1_caxpy(), bl1_ccopyv(), bl1_cfree(), bl1_does_conj(), 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 bl1_cgemm(), bl1_chemm(), bl1_csymm(), bl1_ctrmmsx(), bl1_ctrsmsx(), FLA_Axpy_external(), and FLA_Axpyt_external().
{
scomplex* a_begin;
scomplex* b_begin;
scomplex* a_temp;
int inca_temp;
int lda, inca;
int ldb, incb;
int n_iter;
int n_elem;
int j;
// Return early if possible.
if ( bl1_zero_dim2( m, n ) ) return;
// Handle cases where A and B are vectors to ensure that the underlying axpy
// 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 );
}
}
}
if ( bl1_does_conj( trans ) )
{
conj1_t conj = bl1_proj_trans1_to_conj( trans );
a_temp = bl1_callocv( n_elem );
inca_temp = 1;
for ( j = 0; j < n_iter; j++ )
{
a_begin = a + j*lda;
b_begin = b + j*ldb;
bl1_ccopyv( conj,
n_elem,
a_begin, inca,
a_temp, inca_temp );
bl1_caxpy( n_elem,
alpha,
a_temp, inca_temp,
b_begin, incb );
}
bl1_cfree( a_temp );
}
else // if ( !bl1_does_conj( trans ) )
{
for ( j = 0; j < n_iter; j++ )
{
a_begin = a + j*lda;
b_begin = b + j*ldb;
bl1_caxpy( n_elem,
alpha,
a_begin, inca,
b_begin, incb );
}
}
}
| void bl1_daxpymt | ( | trans1_t | trans, |
| int | m, | ||
| int | n, | ||
| double * | alpha, | ||
| double * | a, | ||
| int | a_rs, | ||
| int | a_cs, | ||
| double * | b, | ||
| int | b_rs, | ||
| int | b_cs | ||
| ) |
References bl1_daxpy(), bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.
Referenced by bl1_dgemm(), bl1_dsymm(), bl1_dtrmmsx(), bl1_dtrsmsx(), FLA_Axpy_external(), and FLA_Axpyt_external().
{
double* a_begin;
double* b_begin;
int lda, inca;
int ldb, incb;
int n_iter;
int n_elem;
int j;
// Return early if possible.
if ( bl1_zero_dim2( m, n ) ) return;
// Handle cases where A and B are vectors to ensure that the underlying axpy
// 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 );
}
}
}
for ( j = 0; j < n_iter; j++ )
{
a_begin = a + j*lda;
b_begin = b + j*ldb;
bl1_daxpy( n_elem,
alpha,
a_begin, inca,
b_begin, incb );
}
}
| void bl1_saxpymt | ( | trans1_t | trans, |
| int | m, | ||
| int | n, | ||
| float * | alpha, | ||
| float * | a, | ||
| int | a_rs, | ||
| int | a_cs, | ||
| float * | b, | ||
| int | b_rs, | ||
| int | b_cs | ||
| ) |
References bl1_does_notrans(), bl1_does_trans(), bl1_is_col_storage(), bl1_is_row_storage(), bl1_is_vector(), bl1_saxpy(), bl1_vector_dim(), bl1_vector_inc(), bl1_zero_dim2(), and BLIS1_NO_TRANSPOSE.
Referenced by bl1_sgemm(), bl1_ssymm(), bl1_strmmsx(), bl1_strsmsx(), FLA_Axpy_external(), and FLA_Axpyt_external().
{
float* a_begin;
float* b_begin;
int lda, inca;
int ldb, incb;
int n_iter;
int n_elem;
int j;
// Return early if possible.
if ( bl1_zero_dim2( m, n ) ) return;
// Handle cases where A and B are vectors to ensure that the underlying axpy
// 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 );
}
}
}
for ( j = 0; j < n_iter; j++ )
{
a_begin = a + j*lda;
b_begin = b + j*ldb;
bl1_saxpy( n_elem,
alpha,
a_begin, inca,
b_begin, incb );
}
}
| void bl1_zaxpymt | ( | trans1_t | trans, |
| int | m, | ||
| int | n, | ||
| dcomplex * | alpha, | ||
| dcomplex * | a, | ||
| int | a_rs, | ||
| int | a_cs, | ||
| dcomplex * | b, | ||
| int | b_rs, | ||
| int | b_cs | ||
| ) |
References bl1_does_conj(), 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_zallocv(), bl1_zaxpy(), bl1_zcopyv(), bl1_zero_dim2(), bl1_zfree(), and BLIS1_NO_TRANSPOSE.
Referenced by bl1_zgemm(), bl1_zhemm(), bl1_zsymm(), bl1_ztrmmsx(), bl1_ztrsmsx(), FLA_Axpy_external(), and FLA_Axpyt_external().
{
dcomplex* a_begin;
dcomplex* b_begin;
dcomplex* a_temp;
int inca_temp;
int lda, inca;
int ldb, incb;
int n_iter;
int n_elem;
int j;
// Return early if possible.
if ( bl1_zero_dim2( m, n ) ) return;
// Handle cases where A and B are vectors to ensure that the underlying axpy
// 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 );
}
}
}
if ( bl1_does_conj( trans ) )
{
conj1_t conj = bl1_proj_trans1_to_conj( trans );
a_temp = bl1_zallocv( n_elem );
inca_temp = 1;
for ( j = 0; j < n_iter; j++ )
{
a_begin = a + j*lda;
b_begin = b + j*ldb;
bl1_zcopyv( conj,
n_elem,
a_begin, inca,
a_temp, inca_temp );
bl1_zaxpy( n_elem,
alpha,
a_temp, inca_temp,
b_begin, incb );
}
bl1_zfree( a_temp );
}
else // if ( !bl1_does_conj( trans ) )
{
for ( j = 0; j < n_iter; j++ )
{
a_begin = a + j*lda;
b_begin = b + j*ldb;
bl1_zaxpy( n_elem,
alpha,
a_begin, inca,
b_begin, incb );
}
}
}
1.7.6.1