FLA_Error FLA_Gemv_t_blk_var6( FLA_Obj alpha, FLA_Obj A, FLA_Obj x, FLA_Obj beta, FLA_Obj y, fla_gemv_t* cntl )
{
FLA_Obj AT, A0,
AB, A1,
A2;
FLA_Obj xT, x0,
xB, x1,
x2;
dim_t b;
FLA_Scal_internal( beta, y,
FLA_Cntl_sub_scal( cntl ) );
FLA_Part_2x1( A, &AT,
&AB, 0, FLA_BOTTOM );
FLA_Part_2x1( x, &xT,
&xB, 0, FLA_BOTTOM );
while ( FLA_Obj_length( AB ) < FLA_Obj_length( A ) ){
b = FLA_Determine_blocksize( AT, FLA_TOP, FLA_Cntl_blocksize( cntl ) );
FLA_Repart_2x1_to_3x1( AT, &A0,
&A1,
/* ** */ /* ** */
AB, &A2, b, FLA_TOP );
FLA_Repart_2x1_to_3x1( xT, &x0,
&x1,
/* ** */ /* ** */
xB, &x2, b, FLA_TOP );
/*------------------------------------------------------------*/
/* y = alpha * A1' * x1 + y */
FLA_Gemv_internal( FLA_TRANSPOSE,
alpha, A1, x1, FLA_ONE, y,
FLA_Cntl_sub_gemv( cntl ) );
/*------------------------------------------------------------*/
FLA_Cont_with_3x1_to_2x1( &AT, A0,
/* ** */ /* ** */
A1,
&AB, A2, FLA_BOTTOM );
FLA_Cont_with_3x1_to_2x1( &xT, x0,
/* ** */ /* ** */
x1,
&xB, x2, FLA_BOTTOM );
}
return FLA_SUCCESS;
}
FLA_Error FLA_Gemv_n_blk_var1( FLA_Obj alpha, FLA_Obj A, FLA_Obj x, FLA_Obj beta, FLA_Obj y, fla_gemv_t* cntl )
{
FLA_Obj AT, A0,
AB, A1,
A2;
FLA_Obj yT, y0,
yB, y1,
y2;
dim_t b;
FLA_Part_2x1( A, &AT,
&AB, 0, FLA_TOP );
FLA_Part_2x1( y, &yT,
&yB, 0, FLA_TOP );
while ( FLA_Obj_length( AT ) < FLA_Obj_length( A ) ){
b = FLA_Determine_blocksize( AB, FLA_BOTTOM, FLA_Cntl_blocksize( cntl ) );
FLA_Repart_2x1_to_3x1( AT, &A0,
/* ** */ /* ** */
&A1,
AB, &A2, b, FLA_BOTTOM );
FLA_Repart_2x1_to_3x1( yT, &y0,
/* ** */ /* ** */
&y1,
yB, &y2, b, FLA_BOTTOM );
/*------------------------------------------------------------*/
/* y1 = alpha * A1 * x + y1; */
FLA_Gemv_internal( FLA_NO_TRANSPOSE,
alpha, A1, x, beta, y1,
FLA_Cntl_sub_gemv( cntl ) );
/*------------------------------------------------------------*/
FLA_Cont_with_3x1_to_2x1( &AT, A0,
A1,
/* ** */ /* ** */
&AB, A2, FLA_TOP );
FLA_Cont_with_3x1_to_2x1( &yT, y0,
y1,
/* ** */ /* ** */
&yB, y2, FLA_TOP );
}
return FLA_SUCCESS;
}
FLA_Error FLA_Gemv_t_blk_var2( FLA_Obj alpha, FLA_Obj A, FLA_Obj x, FLA_Obj beta, FLA_Obj y, fla_gemv_t* cntl )
{
FLA_Obj AL, AR, A0, A1, A2;
FLA_Obj yT, y0,
yB, y1,
y2;
dim_t b;
FLA_Part_1x2( A, &AL, &AR, 0, FLA_RIGHT );
FLA_Part_2x1( y, &yT,
&yB, 0, FLA_BOTTOM );
while ( FLA_Obj_width( AR ) < FLA_Obj_width( A ) ){
b = FLA_Determine_blocksize( AL, FLA_LEFT, FLA_Cntl_blocksize( cntl ) );
FLA_Repart_1x2_to_1x3( AL, /**/ AR, &A0, &A1, /**/ &A2,
b, FLA_LEFT );
FLA_Repart_2x1_to_3x1( yT, &y0,
&y1,
/* ** */ /* ** */
yB, &y2, b, FLA_TOP );
/*------------------------------------------------------------*/
/* y1 = alpha * A1' * x + y1 */
FLA_Gemv_internal( FLA_TRANSPOSE,
alpha, A1, x, beta, y1,
FLA_Cntl_sub_gemv( cntl ) );
/*------------------------------------------------------------*/
FLA_Cont_with_1x3_to_1x2( &AL, /**/ &AR, A0, /**/ A1, A2,
FLA_RIGHT );
FLA_Cont_with_3x1_to_2x1( &yT, y0,
/* ** */ /* ** */
y1,
&yB, y2, FLA_BOTTOM );
}
return FLA_SUCCESS;
}
FLA_Error FLA_Gemv( FLA_Trans transa, FLA_Obj alpha, FLA_Obj A, FLA_Obj x, FLA_Obj beta, FLA_Obj y )
{
FLA_Error r_val;
// Check parameters.
if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
FLA_Gemv_check( transa, alpha, A, x, beta, y );
#ifdef FLA_ENABLE_BLAS2_FRONT_END_CNTL_TREES
// Invoke FLA_Gemv_internal() with flat control tree that simply calls
// external wrapper.
r_val = FLA_Gemv_internal( transa, alpha, A, x, beta, y, fla_gemv_cntl_blas );
#else
r_val = FLA_Gemv_external( transa, alpha, A, x, beta, y );
#endif
return r_val;
}
FLA_Error FLASH_Gemv( FLA_Trans transa, FLA_Obj alpha, FLA_Obj A, FLA_Obj x, FLA_Obj beta, FLA_Obj y )
{
FLA_Error r_val;
FLA_Bool enable_supermatrix;
// Check parameters.
if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
FLA_Gemv_check( transa, alpha, A, x, beta, y );
// Find the status of SuperMatrix.
enable_supermatrix = FLASH_Queue_get_enabled();
// Temporarily disable SuperMatrix.
FLASH_Queue_disable();
// Execute tasks.
r_val = FLA_Gemv_internal( transa, alpha, A, x, beta, y, flash_gemv_cntl_fm_rp );
// Restore SuperMatrix to its previous status.
if ( enable_supermatrix )
FLASH_Queue_enable();
return r_val;
}
FLA_Error FLA_Trsv_lc_blk_var1( FLA_Diag diagA, FLA_Obj A, FLA_Obj x, fla_trsv_t* cntl )
{
FLA_Obj ATL, ATR, A00, A01, A02,
ABL, ABR, A10, A11, A12,
A20, A21, A22;
FLA_Obj xT, x0,
xB, x1,
x2;
dim_t b;
FLA_Part_2x2( A, &ATL, &ATR,
&ABL, &ABR, 0, 0, FLA_BR );
FLA_Part_2x1( x, &xT,
&xB, 0, FLA_BOTTOM );
while ( FLA_Obj_length( ABR ) < FLA_Obj_length( A ) ){
b = FLA_Determine_blocksize( ATL, FLA_TL, FLA_Cntl_blocksize( cntl ) );
FLA_Repart_2x2_to_3x3( ATL, /**/ ATR, &A00, &A01, /**/ &A02,
&A10, &A11, /**/ &A12,
/* ************* */ /* ******************** */
ABL, /**/ ABR, &A20, &A21, /**/ &A22,
b, b, FLA_TL );
FLA_Repart_2x1_to_3x1( xT, &x0,
&x1,
/* ** */ /* ** */
xB, &x2, b, FLA_TOP );
/*------------------------------------------------------------*/
/* x1 = x1 - A21' * x2; */
FLA_Gemv_internal( FLA_CONJ_TRANSPOSE,
FLA_MINUS_ONE, A21, x2, FLA_ONE, x1,
FLA_Cntl_sub_gemv( cntl ) );
/* x1 = tril( A11' ) \ x1; */
FLA_Trsv_internal( FLA_LOWER_TRIANGULAR, FLA_CONJ_TRANSPOSE, diagA,
A11, x1,
FLA_Cntl_sub_trsv( cntl ) );
/*------------------------------------------------------------*/
FLA_Cont_with_3x3_to_2x2( &ATL, /**/ &ATR, A00, /**/ A01, A02,
/* ************** */ /* ****************** */
A10, /**/ A11, A12,
&ABL, /**/ &ABR, A20, /**/ A21, A22,
FLA_BR );
FLA_Cont_with_3x1_to_2x1( &xT, x0,
/* ** */ /* ** */
x1,
&xB, x2, FLA_BOTTOM );
}
return FLA_SUCCESS;
}