|
libflame
revision_anchor
|
Functions | |
| FLA_Error | FLA_Trmm_lln_blk_var2 (FLA_Diag diagA, FLA_Obj alpha, FLA_Obj A, FLA_Obj B, fla_trmm_t *cntl) |
| FLA_Error FLA_Trmm_lln_blk_var2 | ( | FLA_Diag | diagA, |
| FLA_Obj | alpha, | ||
| FLA_Obj | A, | ||
| FLA_Obj | B, | ||
| fla_trmm_t * | cntl | ||
| ) |
References FLA_Cont_with_3x1_to_2x1(), FLA_Cont_with_3x3_to_2x2(), FLA_Determine_blocksize(), FLA_Gemm_internal(), FLA_Obj_length(), FLA_ONE, FLA_Part_2x1(), FLA_Part_2x2(), FLA_Repart_2x1_to_3x1(), FLA_Repart_2x2_to_3x3(), FLA_Scal_internal(), and FLA_Trmm_internal().
Referenced by FLA_Trmm_lln().
{
FLA_Obj ATL, ATR, A00, A01, A02,
ABL, ABR, A10, A11, A12,
A20, A21, A22;
FLA_Obj BT, B0,
BB, B1,
B2;
dim_t b;
FLA_Scal_internal( alpha, B,
FLA_Cntl_sub_scal( cntl ) );
FLA_Part_2x2( A, &ATL, &ATR,
&ABL, &ABR, 0, 0, FLA_BR );
FLA_Part_2x1( B, &BT,
&BB, 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( BT, &B0,
&B1,
/* ** */ /* ** */
BB, &B2, b, FLA_TOP );
/*------------------------------------------------------------*/
/* B2 = B2 + A21 * B1; */
FLA_Gemm_internal( FLA_NO_TRANSPOSE, FLA_NO_TRANSPOSE,
FLA_ONE, A21, B1, FLA_ONE, B2,
FLA_Cntl_sub_gemm( cntl ) );
/* B1 = tril( A11 ) * B1; */
FLA_Trmm_internal( FLA_LEFT, FLA_LOWER_TRIANGULAR, FLA_NO_TRANSPOSE, diagA,
FLA_ONE, A11, B1,
FLA_Cntl_sub_trmm( 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( &BT, B0,
/* ** */ /* ** */
B1,
&BB, B2, FLA_BOTTOM );
}
return FLA_SUCCESS;
}
1.7.6.1