|
libflame
revision_anchor
|
Functions | |
| FLA_Error | FLA_LU_nopiv_blk_var5 (FLA_Obj A, fla_lu_t *cntl) |
| FLA_Error FLA_LU_nopiv_blk_var5 | ( | FLA_Obj | A, |
| fla_lu_t * | cntl | ||
| ) |
References FLA_Cont_with_3x3_to_2x2(), FLA_Determine_blocksize(), FLA_Gemm_internal(), FLA_LU_nopiv_internal(), FLA_MINUS_ONE, FLA_Obj_length(), FLA_Obj_width(), FLA_ONE, FLA_Part_2x2(), FLA_Repart_2x2_to_3x3(), and FLA_Trsm_internal().
Referenced by FLA_LU_nopiv_internal().
{
FLA_Obj ATL, ATR, A00, A01, A02,
ABL, ABR, A10, A11, A12,
A20, A21, A22;
dim_t b;
FLA_Part_2x2( A, &ATL, &ATR,
&ABL, &ABR, 0, 0, FLA_TL );
while ( FLA_Obj_length( ATL ) < FLA_Obj_length( A ) &&
FLA_Obj_width( ATL ) < FLA_Obj_width( A )){
b = FLA_Determine_blocksize( ABR, FLA_BR, 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_BR );
/*------------------------------------------------------------*/
// A11 = LU_nopiv( A11 )
FLA_LU_nopiv_internal( A11, FLA_Cntl_sub_lu( cntl ) );
// A12 = trilu( A11 ) \ A12
FLA_Trsm_internal( FLA_LEFT, FLA_LOWER_TRIANGULAR,
FLA_NO_TRANSPOSE, FLA_UNIT_DIAG,
FLA_ONE, A11, A12,
FLA_Cntl_sub_trsm1( cntl ) );
// A21 = A21 / triu( A11 )
FLA_Trsm_internal( FLA_RIGHT, FLA_UPPER_TRIANGULAR,
FLA_NO_TRANSPOSE, FLA_NONUNIT_DIAG,
FLA_ONE, A11, A21,
FLA_Cntl_sub_trsm2( cntl ) );
// A22 = A22 - A21 * A12
FLA_Gemm_internal( FLA_NO_TRANSPOSE, FLA_NO_TRANSPOSE,
FLA_MINUS_ONE, A21, A12, FLA_ONE, A22,
FLA_Cntl_sub_gemm1( cntl ) );
/*------------------------------------------------------------*/
FLA_Cont_with_3x3_to_2x2( &ATL, /**/ &ATR, A00, A01, /**/ A02,
A10, A11, /**/ A12,
/* ************** */ /* ****************** */
&ABL, /**/ &ABR, A20, A21, /**/ A22,
FLA_TL );
}
return FLA_SUCCESS;
}
1.7.6.1